SYSTEM FOR MONITORING A HEAT TREATMENT

专利摘要:
system for monitoring a heat treatment the present invention relates to a system for monitoring a heat treatment comprising a sensor unit having at least one sensor to determine the current sensor data of a food being heated ; a processing unit to determine the current characteristic data from the current sensor data; and a monitoring unit adapted to determine a current heating process state in a current heating process of the monitored food by comparing the current characteristics data with the characteristics reference data of a heating process. reference.
公开号:BR112015012978B1
申请号:R112015012978-1
申请日:2013-12-04
公开日:2021-03-30
发明作者:Ingo Stork (Genannt) Wersborg
申请人:Ingo Stork (Genannt) Wersborg；
IPC主号:

专利说明:

[001] The present invention relates to a system for monitoring a heat treatment, in particular a monitoring system for heating, cooking or manufacturing foods to be heated such as bread, pasta or the like.
[002] Most likely, the heat treatment of food has been carried out by humans since the invention of fire. However, until now, this function is still controlled by a human operator. The aim of the present invention is to automate the treatment of food and in particular the baking or making of bread in such a way that no human interaction is necessary.
[003] Many inventions are known, which come close to this goal. For example, German patent No. DE 10 2005 030483 describes an oven for heat treatment with an opening device that can be opened or closed automatically. In German patent No DE 20 2011 002570 an apparatus for the heat treatment of food products and for receiving them in a food conveyor is revealed. The latter is equipped with a control system to control a treatment process to detect the type and quantity of the products. The controller selects the performance of an automatic product identification with predetermined data. A camera external to the treatment chamber can be used as a sensor.
[004] European patent No. EP 250 169 A1 describes a cooking oven door which incorporates a camera for viewing the internal lining of the heating or cooking chamber. Visualization is advantageous to save energy losses created by visual windows.
[005] The US patent application published under US No. 2011/0123689 describes an oven comprising a camera and a distance sensor in order to extract product characteristics for the heating processes.
[006] German patent No. DE 20 2011 002 570 U1 describes a system with sensor acquisition in ovens.
[007] However, the thermal treatment of food, in particular with regard to baking bread in an oven, still follows manual adjustments and occurs under human supervision. When a human operator places bread in an oven, important properties such as temperature, time and circulation must be adjusted and established. The settings are usually stored within a database of oven control programs. A human operator has to choose the appropriate program and this factor is still a source of errors and creates a human workforce with a certain degree of knowledge. In addition, many process parameters can result in an undesired end result with respect to the food product. Bread can be undercooked or overcooked, even if a correct program has been chosen. This can be caused by differences in the preheating of the oven, in the preparation of the dough, in the temperature outside, in the humidity outside, in the load distribution, in the opening times of the oven door and many other factors. Specialized human labor is still required to supervise cooking or heat treatment of food.
[008] Additionally, when processing food, for example, in the manufacture of raw or pre-cooked pasta, the objects being processed undergo several variations of processes. Due to the nature of various food products, the objects being processed can vary in: shape, color, size, and a variety of other parameters. This is one of the main threats in industrial food processing because processing devices often have to be adjusted to compensate for these variations. In this way, it is desirable to automate the industrial processing steps ideally making manual adjustments unnecessary. When baking, for example, the change in the characteristics of the flour, can result in severe procedural variations of industrial dough processing devices. For example, it may be necessary to adapt the parameters of a blender, a dough divider, dough forming devices, manufacturing, cutting, packaging, an oven cooking program or a vacuum cooking unit .
[009] For such an objective of cooking or automated food processing to be achieved, it is necessary to provide the system for monitoring corresponding to the data from appropriate monitoring devices. Therefore, there is a need for monitoring systems with monitoring devices to collect adequate data.
[010] For goods cooked in an oven, a system for monitoring with a camera can be used to monitor the cooking process through a window in an oven. However, for the prevention of thermal losses caused by heat dissipation through the window, in conventional ovens such check windows are made with double glazing, for example, they have an internal glass slide and an external glass slide. Therefore, light from outside the oven can penetrate the outer glass sheet and be reflected in the camera through the inner glass sheet, resulting in distorted images of the cooked goods.
[011] Therefore, it is desirable to provide a system for monitoring a heat treatment which reduces disturbances in the images of cooked goods captured through a double-glazed window.
[012] The data from food processing systems with regard to the structure of the processed food should be obtained without stopping the processing of the foodstuff itself, with the aim of not reducing output production. Therefore, it is desirable to adjust the parameters of the aforementioned devices of a food processing system or any other device in the area of food processing, based on non-contact measurement techniques.
[013] In order to generate data captured by means of monitoring devices that are useful for automatic processing of cooking or food processing, it is desirable to provide a method for classifying a multiplicity of images recorded by means of monitoring devices observing a processed food processing area and provide a machine using the same. Once the data is properly classified, it is desirable to take advantage of the cognitive ability in order to increase the flexibility, quality and efficiency of the heat treatment machine. This can be further separated into the following objectives:
[014] It is desirable to provide a system that is capable of gaining knowledge through learning from a human expert on how to summarize relevant information within food processing and how to operate an oven, in which the system should demonstrate reasonable behavior in unfamiliar situations and should be able to learn without supervision.
[015] It is desirable to provide a system that increases efficiency by means of a feedback control of the energy supply adapting to changes in processing time and maintaining a desired state of food processing.
[016] It is desirable to provide a system that has the flexibility for individually different food processing functions by adapting to different types of food or processing functions.
[017] These objectives are achieved through a system for monitoring a heat treatment in accordance with the attached claims.
[018] In particular, for capturing images from a heating chamber (oven), it is advantageous to use lighting in combination with a darkened or shaded window on the outside. This provides less impact caused by the light on the outside to image processing the photo on the inside of the oven. It is recommended that the window be shaded by at least 40%.
[019] With regard to industrial food processing, it is advantageous to use a laser line generator, or other light source, and a camera sensor, or any optical sensor, to acquire information about the food being processed . With a procedure also known as laser triangulation, a laser line can be projected over a measuring object in order to obtain its characteristics.
[020] In addition, it is advantageous that the heat treatment of food is automated in such a way that no additional human interaction is necessary other than loading and unloading the oven or heat treatment machine. However, even this step can be automated, if desired. In order to do this, the heat treatment machine needs a treatment chamber that is monitored by a camera and equipped with a temperature sensor on the inside of the treatment chamber such as a thermometer. Instead of using a camera, an array of elements from at least two photodiodes can also be used. It is advantageous to use more sensors by acquiring signals related to humidity, time, ventilation, heat distribution, load volume, load distribution, load weight, surface temperature of the food and interior temperature of the treated food, inside the storage chamber. treatment. The following sensors can thus be applied: hygrometer, laser triangulation, insertion temperature sensors, acoustic sensors, scales, timers and several others. In addition, cooling systems attached to any applied heat sensor can be applied. For example, this can be an electric air or water cooling system such as a Peltier cooler or fan, a thermoelectric heat pump or a vapor - compression refrigeration, and several others.
[021] Additionally, it is advantageous that in a process of heat treatment of food and in particular of cooking food with a heat treatment machine, such as an oven with a heating chamber, the temperature inside and the interior image camera or other sensors, can be used to control power supply or treatment parameters. According to the invention, the camera image is suitable for detecting parameters related to the change in volume and / or color of the food during its heating. According to a previously taught or fixed model machine, it can be determined with this method for the heat treatment machine, whether the treated food is in a predefined state of the desired process, and with a closed loop control of the process energy. heat treatment, the process can be individually adjusted. The desired result of the process can be achieved with several heat treatment machines distributed locally, through the distribution of parameters defined by the desired processing conditions of the treated food. In addition, the sensors used and the derived processing data, in particular the image from the camera, can be used to determine the type and quantity of the food based on the characteristics of the data and, therefore, initiate the appropriate processing variants automatically .
[022] In accordance with an embodiment of the present invention, a system for monitoring a heat treatment comprises: a heat treatment machine comprising a heating chamber, a double-glazed window comprising an inward (internal) window and a window to the outside (outside), and a lighting device to illuminate the inside of the heating chamber, and a monitoring device mounted to the heat treatment machine and comprising a camera to observe the inside. of the heating chamber through the inside window, where the visible transmittance of the outside window is less than the visible transmittance of the inside window to reduce reflections inside the double-glazed window frame and the lighting effects on the outside on the images when processing the images recorded by the camera. Preferably, the outside window is darkened by means of a coating. Preferably a metal foil or a shading foil is applied to the outside window. Preferably, the outside window comprises a shaded glass. Preferably the outside window has a maximum visible transmittance of 60%. Preferably, the double glazed window is a heat treatment machine portal window from a heat treatment machine heat treatment machine portal. Preferably, the monitoring apparatus is adapted to generate images processed with a high dynamic range (HDR) of the food to be treated inside the heating chamber. Preferably, the monitoring apparatus additionally comprises an assembly for a housing and a sensor camera, on which the camera is mounted. Preferably the enclosure is equipped with penetrations and fans to provide cooling for the camera. Preferably, the heat treatment machine is a convection or deck oven having at least two trays arranged in a stocked manner. Preferably, the camera is tilted in such a way in a horizontal and / or vertical direction with respect to the double glazed window to be adapted to observe at least two trays at the same time in the form of convection or deck. Preferably, the heat treatment monitoring system comprises at least two cameras to observe each of the trays separately. Preferably, the system for monitoring heat treatment additionally comprises a control unit adapted to process and classify the food images observed by the camera based on the training data to determine a time to complete the heating process for the food. Preferably, the control unit is adapted to stop the heating of the heat treatment machine when the heating process has come to an end. Preferably, the control unit is adapted to automatically open the heat treatment machine door when the cooking process has ended, or where the control unit is adapted to ventilate the heating chamber with cold air or with air when the heating process has come to an end.
[023] According to another embodiment of the present invention, a system for monitoring heat treatment comprises a sensor unit having at least one sensor for determining current sensor data of the food being treated; a processing unit for determining the current characteristic data from the current sensor data; and a monitoring unit adapted to determine a current state of the heating process in a current heating process of the monitored food by comparing the current characteristic data with the reference characteristic data of a reference heating process. Preferably, the system for monitoring heat treatment additionally comprises a learning unit adapted to create a mapping of current sensor data to current characteristic data and / or to determine reference characteristic data of a reference heating process based on in the characteristic data of at least one preliminary heating process.
[024] Preferably, the learning unit is adapted to create a mapping of the current sensor data to the current characteristic data by analyzing the variation of at least one preliminary heating process to reduce the dimensionality of the current sensor data.
[025] Preferably, the analysis of variation comprises at least a principal component analysis (PCA), a mapping of isometric characteristics (ISOMAP) or a discriminative linear analysis (LDA) or a dimensionality reduction technique.
[026] Preferably the learning unit is adapted to determine the reference characteristic data from a reference heating process by combining predetermined characteristic data from a heating program with a preliminary set of characteristic data from at least one process preliminary heating system classified as part of a preliminary set by a user's preference.
[027] Preferably, the system for monitoring thermal heating additionally comprises a recording unit for recording current characteristic data from a current heating process, in which the learning unit is adapted to receive the recorded characteristic data from the recording unit to be used as characteristic data of a preliminary heating process.
[028] Preferably, the sensor unit comprises a camera recording a pixel image of the food being heated, wherein the current sensor data from the camera corresponds to the current pixel data of a current pixel image. Preferably, the current pixel data comprises a first pixel data corresponding to a first color, a second pixel data corresponding to a second color, and a third pixel data corresponding to a third color. Preferably, the first, second and third colors correspond to R, G and B, respectively. Preferably, the camera is adapted to generate processed HDR pixel images as current pixel data.
[029] Preferably, the system for monitoring heat treatment additionally comprises a classification unit adapted to classify the type of food to be heated and to choose a reference heating process corresponding to the type of food determined.
[030] Preferably, the system for monitoring heat treatment additionally comprises a control unit adapted to change a heating process from a preliminary heating process to a cooking process based on a comparison of the current state of the heating process. heating determined by the monitoring unit with a predetermined state of the heating process. Preferably, the system for monitoring heat treatment additionally comprises a control unit adapted to control a display unit, being adapted to indicate a time remaining in the heating process based on a comparison of the current state of the heating process determined by means of of the monitoring unit with a predetermined heating process state corresponding to a heating end point and / or display images inside the heating chamber.
[031] Preferably, the system for monitoring heat treatment additionally comprises a control unit adapted to alert a user when the heating process must be completed.
[032] Preferably, the system for monitoring heat treatment additionally comprises a control unit adapted to control a temperature control of a heating chamber, the means for adapting humidity in the heating chamber through the addition of water or steam , a control of the ventilation mechanism, the means to adapt the fan speed, the means to adapt the pressure differential between the heating chamber and the respective environment, the means to establish a time dependent on the temperature curve inside the chamber heating means, the means for performing and adapting different heat treatment procedures such as checking or cooking, the means for adapting internal gas flow profiles inside the heating chamber, the means for adapting the intensity of the electromagnetic and sound emission of the respective electromagnetic and sound emitters to probe and observe the properties of the food to be heated of.
[033] Preferably, at least one sensor of the sensor unit comprises at least one of the following: hygrometer, insertion temperature sensor, treatment chamber temperature sensor, acoustic sensors, scales, timer, camera, image sensor, arrangement of photodiodes elements, a gas gas analyzer inside the treatment chamber, the means for determining temperature profiles of the temperature insertion sensors, the means for determining emissions from electromagnetic or acoustic processes of the food being treated, such as light or sound being reflected or emitted in response to a light or sound emitter or source, the means to determine the results from 3D measurements of the food to be heated including 3D or stereo camera systems or radars, or the means to determine the type or constitution or pattern or optical characteristics or the volume or mass of the food to be treated.
[034] In accordance with another embodiment of the present invention, a system for monitoring heat treatment is provided and comprises: a heat treatment or cooking unit for cooking or checking goods or food to be heated or a line of food processing; a laser light distribution unit for generating a first laser beam and a second laser beam for directing the first laser beam and the second laser beam to a position for cooking the goods within the cooking unit; a first light detection unit for detecting the reflection of the first laser beam spread from the cooking goods; a second light detection unit for detecting the reflection of the second laser beam spread from the cooking goods; a measuring unit for determining a height profile of the goods being cooked according to the detections of the first light detection unit and the second detection unit; and a mobile unit for changing a distance between the laser light distribution unit and the goods being cooked. Here, the laser light distribution unit preferably comprises: a first laser light generating unit for generating the first laser beam; and a second laser light generating unit for generating the second laser beam. In addition, the laser light distribution unit preferably comprises: a primary laser light generating unit for generating a primary laser beam; an optical unit to generate the first laser beam and the second laser beam from the primary laser beam. The optical unit preferably comprises: a movable and rotating mirror towards which the primary laser beam is directed to generate the first laser beam and the second laser beam; and a mirror, in the direction of which the secondary laser beam is directed to generate the second laser beam. The first laser beam is preferably directed in a direction to a first position; the second laser beam is preferably directed to a second position; a piece of the cooking goods is preferably moved from the first position to the second position by means of the mobile unit; and a change in the height profile of the piece of goods being cooked is preferably determined by means of the measuring unit. Preferably, the first laser beam is directed at a first end of a piece of cooking goods and has an inclination of less than 45 ° with respect to a support of the cooking piece of goods; the second laser beam is directed at a second end of the piece of cooking goods opposite the first end and has an inclination of less than 45 ° with respect to the support; and the minimum angle between the first laser beam and the second laser beam is greater than 90 °. Preferably the mobile unit is a mechanical conveyor that moves the cooking goods through the cooking unit. Preferably, the laser light distribution unit is located inside the cooking unit; and the first laser beam and the second laser beam are directed directly from the laser light distribution unit in one direction to the goods being cooked. Preferably, the laser light generating units are located outside the cooking unit; and the laser beams are directed in one direction to the cooking goods by means of reflecting mirrors. Preferably, the light sensing units are located outside the cooking unit; and the reflection of the laser beams is guided to the light detection units by means of the handle mirrors. Preferably, the mirrors are heated. Preferably, the first laser beam and the second laser beam are formed as a fan; and the reflection of the first laser beam and the second laser beam is focused on the first and the second light detection unit by means of lenses. Preferably, the optical system consisting of the laser light distribution unit, cooking goods and light detection units meets the Scheimpflug Principle. A method for monitoring cooking of the present invention comprises the steps of: processing the cooking goods in a cooking unit; moving the cooking goods through the cooking unit; generating a first laser beam and a second laser beam and directing the first laser beam and the second laser beam to a position of the goods being cooked inside the cooking unit; detecting the reflection of the first laser beam spread from the cooking goods; detecting the reflection of the second laser beam spread from the cooking goods; and determining a height profile of the goods being cooked according to the detection of the first and second scattered laser beams. Brief Description of Drawings
[035] The accompanying drawings, which are included herein to provide an additional understanding of the invention and are incorporated into it, form a part of the present patent application, illustrate realizations of the invention and together with the description serve to explain the principle of the invention. In the drawings:
[036] Figures 1A and 1B show a schematic view of the cross section and a schematic side view of an embodiment of a system for monitoring heat treatment;
[037] Figures 2A and 2B illustrate the reflection properties of a conventional double-glazed window and a double-glazed window of an embodiment of a system for monitoring heat treatment;
[038] Figure 3 shows schematic views of another system for monitoring heat treatment;
[039] Figure 4 shows a schematic view of an embodiment of an image sensor;
[040] Figure 5 shows a schematic view of another embodiment of an image sensor;
[041] Figures 6A and 6B show a schematic front and side view of another embodiment of a system for monitoring heat treatment;
[042] Figure 7 shows a schematic view of an embodiment of a heating chamber;
[043] Figure 8 shows a schematic view of an embodiment of a food production system;
[044] Figure 9 shows a schematic view of an implementation of a food production system using a triangulation of lasers;
[045] Figure 10 shows a schematic view of another embodiment of a food production system using a triangulation of lasers;
[046] Figure 11 shows a top view of an embodiment of a tray with an indication for the dough arrangement;
[047] Figure 12 shows a schematic view of an embodiment of a sensor system integrated in an oven shelf;
[048] Figure 13 shows a flow of schematic data processing of an implementation of a system for monitoring heat treatment;
[049] Figure 14 shows a cognitive perception-action loop for food production machines with sensors and actuators in accordance with the present invention;
[050] Figure 15 shows categories of linear and non-linear dimensionality reduction techniques;
[051] Figure 16 shows a mapping of two-dimensional test data to a three-dimensional space with an optimized linear separator;
[052] Figure 17 shows an architecture according to the present invention and groups of components to design agents for the closed loop monitoring or control process in a foodstuff production system using a black box model with sensors and actuators ;
[053] Figure 18A shows a schematic cross-sectional view of an implementation of a system for monitoring heat treatment; and
[054] Figure 18B shows a block diagram of an implementation of a system for monitoring heat treatment. Detailed Description of the Preferred Realization
[055] Figures 1A and 1B illustrate a system for monitoring heat treatment 100 in accordance with an embodiment of the present invention. Figure 1A illustrates a schematic top cross-sectional view of the system for monitoring heat treatment 100, while Figure 1B illustrates a schematic front view of the same.
[056] As shown in Figures 1A and 1B, the system for monitoring heat treatment or system for monitoring cooking and / or system for monitoring checking and / or cooking 100 has an oven 110 with a heating chamber. oven or heat treatment 120, at least a double glazed window 130 on a side wall of the oven 110 and a lighting apparatus 140 on the inside of the oven chamber 120.
[057] The heat treatment machine or oven 110 can be any oven that can be conventionally used for cooking food, in particular for baking and / or checking bread. The oven can cook food using different techniques. The oven can be a convection type oven or a radiation type oven.
[058] The heating chamber or oven 120 captures most of the interior of oven 110. Inside the chamber of oven 120, the food is cooked. The food can be placed on a different number of trays, which can be supported on the walls of the oven chamber. The food can also be positioned on mobile carts with several trays, which can be moved inside the oven chamber 120. Inside the oven chamber 120, a heat source is provided, which is used to cooking food. In addition, a ventilation system can also be understood inside the oven chamber to distribute the heat produced through the heat source more evenly.
[059] The inside of the oven or heating chamber is lit by means of a lighting fixture 140. The lighting fixture 140 can be arranged inside the oven or heating chamber as shown in Figure 1A . The lighting apparatus 140 can also be located outside the oven chamber 120 and illuminate the oven chamber 120 through a window. The lighting apparatus 140 may be any conventional light-emitting device, for example, a lamp, a halogen lamp, a photodiode or a combination of several such devices. The lighting apparatus 140 can be focused on the food to be cooked inside the oven chamber 120. In particular, the lighting apparatus 140 can be adjusted or focused in such a way that there is a high contrast between the food to be cooked. cooked and the interior of the oven chamber 120 or between the food and the tray and / or the carts on which the food is located. Such high contrast can also be supported or generated only through the use of special colors with respect to the light emitted by the lighting apparatus 140.
[060] In a wall of the oven chamber 120, a window is provided. In order to prevent heat loss outside the oven chamber 120, the window is preferably a double glazed window 130 having an outer glass sheet or an outside window 135 in an inner glass sheet or inside window 136. The double glazed window 130 can prevent heat dissipation between the inside window 136 and the outside window 135 by providing a special gas or vacuum between the window and the inside 136 and outside window 135. Double-glazed window 130 can also be cooled by means of air ventilation between inside window 136 and outside window 135 to prevent heating from the outside window 135, in which no special gas or vacuum is provided between the inside window 136 and the outside window 135. The lighting apparatus 140 can also be provided between the inside window 136 and the window outside 135. The outside glass surface of the outside window 135 is less warm and is therefore suitable for mounting a camera 160. It may be additionally beneficial to use an optical tunnel between the window on the inside 136 and the window outside 135 because again this reduces reflections and the impact of heat.
[061] Through the double glazed window 130 a cooking or baking procedure inside the oven chamber 120 can be observed from the outside of the treatment machine or oven.
[062] As shown in Figure 1B, a monitoring device 150 is mounted over the heat treatment machine or oven 110. The monitoring device 150 is mounted across the entire window 135 outside the double glazed window 130 and it comprises a camera 160 arranged near the outside window 135, which is used to observe the food inside the oven chamber 120 during cooking or cooking. Camera 160 can be any conventional camera, which is capable of providing image data in a form accessible to a computer. The camera 160 can, for example, be a charged charged camera device (CCD) or a complementary metal - oxide - semiconductor (CMOS) camera. Camera 160 takes images of the food cooked during the cooking procedures. As will be described below, these images can be used to automatically control the cooking or cooking procedure. Although the camera 160 is preferably mounted on the outside of the outside window 135 to be easily integrated into the monitoring apparatus 150, in which the camera 160 then looks at the inside of the heating chamber 120 through the window double-glazed 130, the camera 160 can also be provided between the inside window 136 and the outside window 135 to look inside the heating chamber through the inside window 136.
[063] However, a problem occurs if an external light source is present outside the oven chamber 120 in front of the double glazed window 130.
[064] As shown in Figure 2A, an irritating light 272 emitted through an external light source 270 can pass through a window outside 235 'of a double glazed window, but it can be reflected through the inside window 236 on camera 260 watching food 180 being cooked. Therefore, camera 260 not only obtains light 282 emitted or reflected from food 280, but also irritating light 272 reflected on the inside wall 236. This results in deterioration of the image data provided through camera 260 and can thus adversely affecting an automatic cooking process.
[065] In the present embodiment, this adverse effect is prevented by obstructing the irritating light from passing through the outside window 235. This can be done by shading or darkening the outside window 235. Then , the irritating light 272 is reflected or absorbed through the outside window 235 and does not reach the inside window 236. Therefore, no irritating light 272 is reflected in the camera 260 through the inside window 236 and camera 260 only captures the correct information about food 280. Therefore, according to the present embodiment, a deterioration of the automatic food processing procedure is prevented by shading or darkening the outside window 235.
[066] Therefore, to capture images from the heating chamber 120 of the oven 110, it is advantageous to use a lighting device 140 in combination with the shading or darkening of the outside window 235. This provides less impact of light on the outside when processing images from photographs generated on the inside of the oven.
[067] According to the present invention, the visible transmittance of the outside window 135 is less than the visible transmittance of the inside window 136. Here, the transmittance of the outside window 135 is less than 95%, more preferably less than 80% and in particular less than 60% of the visible transmittance of the inside window 136. Additionally, the outside window 235 of the double glazed window 130 may preferably have a maximum transmittance 75%. Visible transmittance is the transmittance of light, the surface of the glass window being incident and normal within a visible wavelength range, for example, 380 nm to 780 nm. It is additionally preferable to shade the window by at least 40%, so the maximum visible transmittance is 60%. In other words, at least 40% of the incoming light is absorbed or reflected through the outside window 235 and 60% of the light is transmitted through the outside window 235. The inside window 236 may have a visible transmittance of usual glass. It is additionally preferable to shade the window by at least 60%, resulting in a transmittance of 40%. A darkening coating or a blade can be advantageously applied to the window outside a double glazed oven door to prevent the coating from deteriorating due to thermal effects. Due to the darkening of the outside window, reflections of light coming from outside the oven can be significantly reduced. The oven window door can be darkened by means of a metal blade or a cladding (mirrored window) or by a shading blade. The oven window door can be a shaded window comprising, for example, a shaded glass on the outside and / or on the inside. If the camera is mounted on the outside window 135, the darkening or reflectivity of the outside window 135 at the camera location can be relegated, for example, by having a hole in the coating to ensure observation of the camera through the hole in the outside window covering 135, where the hole area is not included for determining the transmittance of the outside window 135.
[068] The heat treatment machine or oven 110 can additionally comprise an oven door or a heat treatment machine door, by means of which the oven chamber 120 can be opened and closed. The oven door may comprise a window, through which the oven chamber 120 can be viewed. Preferably, the window comprises the double-glazed window 130 to prevent thermal loss of heating energy to the oven chamber 120. Therefore, the heat treatment monitoring system 100 may comprise the monitoring apparatus 150 and the oven 110 comprising the monitoring apparatus 150, or an oven 110 having the monitoring apparatus 150 mounted on its oven door.
[069] Therefore, reflections within the double glazed window frame of the oven window door can also be reduced. Consequently, the lighting effects on the outside on image processing are negligible. Therefore, with the respective lighting intensity of the oven chamber 120, the inside of the oven chamber 120 can be seen through the camera 160 of the monitoring apparatus 150.
[070] Figure 3 shows different views of an implementation of the system for monitoring heat treatment illustrated in Figures 1A and 1B.
[071] As shown in Figure 3, a monitoring device 350 is mounted on the front side of a deck oven 310 of a system for monitoring heat treatment 300. The monitoring device 350 comprises an enclosure, a sensor camera assembly , and a camera mounted on the sensor camera mount to observe an inside of an oven chamber through an oven window door 330. The camera is tilted in such a way in a horizontal and / or vertical direction with respect to I refer to oven window door 330 to be adapted to observe at least two cooking trays at the same time in deck oven 310.
[072] According to another embodiment, the sensor assembly and the housing are cooled with fans on the inside. Additionally, as can be seen from Figures 4 and 5, the sensor camera assembly of the monitoring device 350 can be equipped with penetrations and heat fans to provide cooling. The sensor assembly and the housing can be optimized to have an optimized angle of view to see two cooking trays at the same time inside the oven.
[073] Figures 6A and 6B show a top view and a side view of another embodiment of the system for monitoring heat treatment illustrated in Figures 1A and 1B, respectively.
[074] As shown in Figure 6A, a monitoring device 650 is mounted on an oven 610 of a system for monitoring heat treatment 600. The monitoring device 650 partially overlaps with the double glazed window 630 of a door. oven 632. The monitoring device 650 comprises a camera inside an enclosure. In addition, the monitoring device 650 comprises a display 6755, which allows information to be displayed to a user and allows user interaction.
[075] Oven 610 can have a convection oven on top and ovens with two decks on the bottom as shown in Figs 6A and 6B.
[076] Additionally, according to one embodiment, the monitoring device 150 may comprise an alert device to inform the user when the cooking process has come to an end. In addition, the monitoring apparatus 150 may comprise a control output to stop, for example, the heat treatment of the oven 110 and / or to automatically open the oven door and / or to ventilate the oven chamber 120 with cold air or with air. Furnace 110 and monitoring apparatus 150 together form the system for monitoring heat treatment 100.
[077] According to an additional embodiment, the monitoring device 150 is adapted to generate processed images with a high dynamic range (HDR) of the goods cooked inside the oven chamber 120. This is particularly advantageous in combination with the window on the outside shaded 135, since the light intensity of the light coming from inside the cooking chamber 120 is reduced by means of the metal blade and the HDR processing allows for better segmentation. Additionally, using HDR processing, a contrast between the cooking goods and their surroundings such as the walls and the oven trays, can be intensified. This allows the heat treatment monitoring system 100 to determine a contour or shape of the goods being cooked even more precisely.
[078] Figure 7 demonstrates a possible sensor adjustment for a treatment chamber 720 according to an additional embodiment. As previously, the treatment chamber 720 is monitored with at least one camera 760. The camera 760 can also comprise an image sensor or an array of photodiode elements with at least two photodiodes. It is advantageous to use more than one camera in order to monitor multiple trays that can be loaded differently. At least one camera 760 can be positioned inside the treatment chamber 720, but it is advantageous to apply a window that reduces the influence of heat with respect to the camera 760, in particular a double-glazed window 730. The double-glazed window 730 can be on any of the walls of the treatment chamber.
[079] As described above, it is advantageous to apply lighting to the treatment chamber 720 by integrating at least one lighting device, for example, a lamp or a light-emitting diode (LED). A defined treatment chamber lighting supports taking robust / good camera images. It is additionally advantageous to apply lighting for at least a specific wavelength and to apply a filter of appropriate wavelength for the camera or the image sensor or for the arrangement of 760 photodiodes elements. This additionally increases the robustness of the system quality for visual monitoring. If the wavelength is chosen to be infrared or close to infrared and the 760 image sensor and optional filters are chosen accordingly, the system for visual monitoring can collect information related to the temperature distribution that may be critical for certain food treatment processes.
[080] The camera or the visual system 760 can Sr equipped with a specific lens system that optimizes the visualization of the food. It is not necessary to capture images related to all the loaded food, since the state of processing of a load is very similar between the loads themselves. Additionally, it can be equipped with an autofocus system and brightness optimization techniques. It is advantageous to use several 760 image sensors for specific wavelengths in order to collect information about color changes related to food treatment. It is advantageous to position the camera or the 760 image sensors to collect food volume change information during heat treatment. It can be particularly advantageous to establish a top or top view of food products.
[081] It may also be advantageous to attach a second oven door or treatment chamber opening to a pre-existing opening system. The sensor system, or in particular the camera, and the lighting unit can then be positioned at the height of the oven window door. This door on top of a door or a double door system could be applied if the sensor system is attached to an oven.
[082] Each of the monitoring devices described above can be mounted on the front side of an oven, as can be seen, for example, in Figures 1A, 1B, 3, 4A, and 4B. The monitoring apparatus comprises a housing, a camera sensor assembly, and a camera mounted on the camera sensor assembly to observe an inside of an oven chamber through an oven window door. The camera is tilted in such a way in a horizontal and / or vertical direction with respect to the oven window door to be adapted to observe at least two cooking trays in the deck oven. The monitoring device can additionally comprise an alert device to inform the user when the cooking process is finished. In addition, the monitoring apparatus may comprise a control output to stop, for example, the heating of the oven and / or to automatically open the oven door and / or to ventilate the oven chamber with cold or air. The oven and the monitoring device together form a system for monitoring heat treatment.
[083] As discussed above, a camera sensor is used to observe cooking processes. According to another embodiment, it is beneficial to use multiple camera sensors. If a tray inside a heating chamber has at least one camera sensor aligned, monitoring and control software can acquire information for all the trays individually. Therefore, it is possible to calculate a remaining cooking time for each of the trays.
[084] The remaining cooking time can be used to alert the oven user to open the door and to remove at least one of the trays, in case the cooking time has ended before the other trays. According to the invention, it is possible to alert the user through a remote information technology system. The alert can be given on a website display, on a smartphone, or on a flashlight near the counter. This has and provides the advantage of the factor that the user is being alerted at his usual place of work which may not be in front of the oven.
[085] According to another embodiment of the system for monitoring the present invention, the system for monitoring can be used in industrial food production systems, for example, cooking or pre-cooking lines or in food preparation systems. dough that form and break dough. However, the system for monitoring can also be used in any other area of food production and processing.
[086] Figure 8 illustrates a system for monitoring 800 with at least one established sensor system 850, for heat treatment machines or ovens 810 (cooking units), with mechanical conveyors 815 (mobile unit). These 810 ovens are usually used in industrial food production systems.
[087] The 850 sensor system can have at least one sensor of the following: hygrometer, temperature insertion sensor, temperature sensor in the treatment chamber, acoustic sensors, laser triangulation, scales, timers, camera, image sensors, arrangement of photodiodes elements. Also included in this sensor system are 850 support devices such as lighting or cooling or movement algorithms.
[088] According to one embodiment, the triangulation of lasers can be used to acquire information regarding the volume of food. Then the established sensor system 850 comprises a laser light distribution unit, which generates and directs laser rays in a direction to the cooking goods inside the oven or cooking unit 810. The laser light distribution unit can direct the laser beams on a single piece of cooking goods at the same time or, according to another embodiment, at least twice within the food treatment process to acquire information regarding the change in volume in relation to the time.
[089] The volume information and / or the height profile of the goods being cooked is then acquired by means of a measurement unit, which analyzes the detection results of the light detection units, which detect the reflection of the rays lasers from cooking goods. There can be a single, or several light detection units for all of the laser rays or a light detection unit for each of the laser rays.
[090] According to another embodiment, at least one additional sensor system 852 can be positioned in different positions on the inside or outside of the heat treatment machine. Alternatively, the 850 sensor system can be applied in a position where the mechanical conveyor passes the food twice at different processing times. Alternatively, the 850 sensor system can move at the same speed as the 815 mechanical conveyor.
[091] According to other realizations, more than one sensor camera or optical sensor and more than one laser line generator for the triangulation of lasers can be used.
[092] According to an embodiment illustrated in Figure 9, a system for monitoring 900 comprises at least two monitoring devices, each with a 955 laser line generator and a 969 light receiving device, for example, a camera or an array of photodiode elements. Accordingly, a laser light distribution unit according to this embodiment comprises a first laser light generating unit and a second laser light generating unit.
[093] From laser light generators 955, laser rays 956 are sent in one direction to food 980, such as raw or pre-cooked pasta on a 915 mechanical conveyor. The laser rays are reflected from the food 980 in one direction to the light receiving devices 960. According to the position of the laser light generators 955 and the light receiving devices 960 with respect to each other and with respect to the mechanical conveyor 915 is known, the distance from the laser light generators 955 for food 980 can be obtained by means of triangulation from the exact position in which the laser rays 956 are observed inside the light receiving devices 960. Therefore, using such laser triangulation, the profile surface area of processed food 980 can be determined.
[094] As shown in Figure 9, the 856 laser beams are directed directly towards the food or cooking goods 980 and are spread directly in one direction to the light receiving device or light detection units 960. According to another realization, the light path of the laser beams can be changed using deflection or guide mirrors. Then, the 955 laser light generators or 960 light detection units can also be located on the outside of the heating chamber or cooking unit. This allows for a more flexible design of the system for monitoring heat treatment. Furthermore, in order to prevent the mirrors from becoming foggy, they can be heated to a temperature that is high enough to prevent the mirrors from fogging, but a temperature low enough not to damage the mirrors.
[095] As shown in Figure 9, the 956 laser beams from the 955 laser beam generators are focused in such a way that food 980, in the various stages of production, is monitored. Note well that although in Figure 9 it is shown that the 955 laser light generators focus on two neighboring pieces of food 980, they can focus in any way on the pieces of food 980 that are at a greater distance from each other. For example, the two pieces of food can be separated by several meters or the 955 laser light generators can be located at an entrance and an exit of a cooking chamber through which the 915 conveyor belt travels and observes the surface profile of food 980 during the entry and exit of the cooking chamber. For this purpose, laser light generators or 955 generating units can also be arranged in such a way that they emit light approximately perpendicularly from the top in one direction to food 980.
[096] Also note that the 955 laser light generators do not need to be located on top of the 915 mechanical conveyor, but can also be located on one side of the 915 mechanical conveyor. Of course, at least two 955 laser light generators also can be located on different sides of the 915 mechanical conveyor.
[097] Hence, therefore, through the use of two or more laser light generators 955 that focus on different food pieces 980 and observe the structure of the respective surfaces of the food pieces 980, a difference in this surface structure caused by through cooking or the food production process can be observed as the conveyor belt or mobile unit 915 moves food 980 through the cooking unit from a focal point of a first laser beam in a direction to a focal point of a second laser beam. This information on the difference in surface structure at various stages of the cooking or food production process can be used to automatically control the process and thus allow automated cooking or food production.
[098] The 956 laser beams can be stippled or in the shape of a fan and extend across the width of the mat or the mechanical mat 915. Using laser beams in the form of 956 fans, a three-dimensional profile of the food 980 running over the mat mechanics 915 can be obtained which can serve even better for the automatic control of the cooking process or food production. Then, the reflection of the laser rays in the fan shape from the food can be collimated or concentrated in lenses on the 960 light detecting units, with the aim of allowing small 960 light detecting units, which can be easily integrated in the system for the monitoring of heat treatment.
[099] As shown in Figure 10, in addition to the observation of different pieces of food, it is especially beneficial to align at least two sensor systems of one system for monitoring 1000 per piece of food at a 45 degree tilt angle observing the measurement of objects 1080 from the top left and top right. This is advantageous because when observing objects, 1055 laser light generators and their respectively aligned 1060 light receiving devices can measure the surface structure of rounded objects in areas that may have been hidden when using only one sensor from a top view it would have been used. According to another embodiment, the laser beams can be tilted even less than 45 ° with respect to the mechanical mat or the tray, which supports the food 1080. So, the surface structure close to the food support can still be observed best.
[0100] In the event that rays formatted as a fan are used, the inclination of the planes covered by the fans should be less than 45 ° with respect to the food support 1080. This also means that the angle between the laser rays must be greater than 90 °.
[0101] Note that although Figure 10 shows that 1055 laser light generators are focused on the same piece of food 1080, they can also focus on two different pieces of food 1080, which are separated from each other . For example, the two pieces of food can be separated by several meters or the 1055 laser light generators can be located at an entrance and at the exit of a cooking chamber through which the mechanical belt travels and observes the surface profile of the food. 1080 food during the entry and exit of the cooking chamber.
[0102] Also note that 1055 laser light generators do not have to be located on top of the mechanical belt, but can also be located on one side of the mechanical belt. Of course, at least two 1055 laser light generators can also be located on different sides of the 915 mechanical conveyor.
[0103] Additionally and according to another embodiment, there may be a display of laser triangulation inside the oven. Then, at least two laser triangulation sensors and two laser lines, looking at the baked products from an angle of approximately 45 degrees (top left and top right), can be used. This provides the advantage that one can also measure the surroundings of the cooked product on its bottom, while through the use of a laser line and a camera from a top view, the bottom half of the environment is hidden and not accounted for in the measurements.
[0104] Thus, according to the additional information these achievements on the food production or cooking process can be provided on the basis of which food production or automated cooking can be carried out more efficiently and more reliably.
[0105] According to another embodiment, a laser line generator, or other light source, and a camera sensor, or any other optical sensor, can be used to collect information regarding the food being processed. With the procedure described above, also known as laser triangulation, a laser line can be projected over a measurement object. An optical sensor, an array of sensor elements or typically a camera, can be directed in one direction to the measurement object. If the perspective of the camera or the point of view and the respective plane and plane of the laser line generator formed by means of the light source and the end of the projected laser line are not parallel or are at an angle, the information Detected optics can be used to perform measurements providing information about size and shapes including a three-dimensional structure or volume.
[0106] In these achievements described above, two laser light generating units were used in order to generate and direct laser beams. According to another embodiment, a primary laser light generating unit can be used to generate a primary laser beam, which is then distributed via an optical unit within the cooking unit. Using such a structure inside the system for monitoring heat treatment, it is possible to save energy and space costs by reducing the number of laser light generating units.
[0107] In addition, the laser light generating unit can be located on the outside of the cooking unit and only the primary laser beam can be inserted into the cooking unit. This makes it possible to choose a system structure for more flexible heat treatment monitoring, especially if light detection units are also provided on the outside of the cooking unit.
[0108] The optical unit can be any type of optical system that allows the division of a primary laser beam into two or more lasers. For example, the optical system may comprise a semitransparent mirror, which reflects a part of the primary laser beam in one direction to a first position to be observed and transmits a part of the primary laser beam in one direction to a mirror, which reflects the light in a sense to a second position of interest. The primary laser beam can also be separated in such a way that its parts are directly directed in one direction to the positions to be observed. According to another embodiment, there may be more mirrors and / or lenses within the light path of the primary laser beam.
[0109] According to another embodiment, the optical unit may comprise a movable and rotating mirror, which generates lasers alternately. For this purpose, the rotating and movable mirror can be provided above the food or cooking goods and can be moved and rotated in such a way that the primary laser beam is directed at different pieces of food or different positions on a single piece of food at different times. In this way, the volume information collected through the measuring unit will refer to different positions inside the cooking unit according to the time.
[0110] Using such mirrors, the space requirements inside the cooking unit are reduced and allows a flexible system design for monitoring heat treatment. Additionally, a user can easily change operations from a mode, in which two different pieces of food are observed in order to obtain information about the change in height profile and / or volume profile of the food, and a mode in the which a single piece of food is observed from different directions in order to obtain the complete and general three-dimensional shape of the food piece also close to the support of the food piece. The mobile and rotating mirror can also perform these different tasks in parallel.
[0111] It is clear that the mirrors used with the primary laser beam can be heated in order to prevent them from fogging up.
[0112] According to another embodiment, the optical system constituted by means of the laser light distribution unit, the cooking food or merchandise and the laser light detection unit can satisfy the Scheimpflug principle. This ensures that the image of the cooking goods sampled by the lasers is always focused on the light detection unit, and therefore allows an accurate measurement of the height profile of the cooking goods.
[0113] According to another embodiment, laser triangulation can be combined with gray image processing to obtain simultaneous information about the shape and size as well as information about texture, color and other optical characteristics. The resulting processing data can be used to generate unique characteristics for the measurement of the object, in this case, food. This can be shape, size, volume, color, toasting process, texture, pore size and density of the food being processed such as dough or baked bread, which can be sliced. Some or all of the information given here can be used to interpret the sensor data in order to allow automated cooking or food processing.
[0114] In the achievements described above, data capture is performed mainly through image sensors such as cameras or arrangement of photodiodes elements. However, according to additional realizations, the data obtained through the image sensors can be supplemented with data from a variety of other sensors such as, for example, hygrometers, temperature insertion sensors, camera temperature sensors. treatment, acoustic sensors, lasers, scales and timers. In addition, a gas gas analyzer inside the treatment chamber, the means for determining temperature profiles of temperature insertion sensors, the means for determining emissions from an electromagnetic or acoustic process of the food to be treated as light or sound that is reflected or emitted in response to light or sound sources, the means to determine the results from 3D measurements of the food to be heated including 3D or stereo camera systems or radars, the means to determine the type of construction or standard or optical characteristics or volume or mass of the food to be treated, can also be used as sensors for the sensor unit 1810 as described here below. Automated food processing or cooking can then be controlled based on all data from all sensors.
[0115] For example, with reference to Figure 7, the treatment chamber 720 can be additionally equipped with at least one temperature sensor or a thermometer 762. Although this is only illustrated in Figure 7, any other embodiment described here can also comprise such a temperature sensor 762. When treating food with heat, the temperature information is related to processing characteristics. This may contain information regarding the development of heat over a period of time and its distribution inside the treatment chamber. This can also gather information about the state of the oven, its heat treatment system and optional preheating.
[0116] It may also be advantageous to integrate insertion thermometers. Insertion thermometers allow you to collect information about the temperature inside the food, which is critical to determine the state of food processing. It is advantageous in the case of baking bread, to acquire information related to the temperature inside and crumbs.
[0117] Additionally, the progress in the color change over time, of the food to be heated can be used to determine a real temperature inside the oven chamber and can be additionally used for a respective temperature control in the cooking process. . The treatment chamber 720 or any other embodiment described herein can be equipped with at least one humidity-related sensor in the treatment chamber such as a hygrometer 764. In particular, the information obtained related to moisture when baking bread is advantageous. When the mass is heated, the water contained therein evaporates resulting in a difference in humidity inside the treatment chamber. For example, with air circulating, the humidity in the treatment chamber during a cooking process may first rise and then drop indicating the state of processing of the food.
[0118] The treatment chamber 720 or any other embodiment described here can additionally be equipped with at least one sensor acquiring information on the weight of the loaded food and eventually its distribution. This can be achieved by means of integrated scales 766 in a tray assembly system of the heating chamber 720. The tray assembly or the stack of storage trays can be supported by wheels or rotating discs facilitating the loading of the oven. The 766 scales could be integrated with the wheels or discs and load them as a transducer. It is advantageous to acquire weight information for each of the used trays or set of trays individually in order to have related information about the total weight of food and its relative distribution according to the desired energy supply and its direction during treatment thermal power can vary significantly. In addition, it is advantageous to acquire information about the weight differences of the food during periods of time while treating the food with heat. For example, in the case of baking bread, the dough loses approximately 10% of its initial weight. Additionally, it is possible to acquire information regarding the state of the pasta or food through the emission and capture of sound signals, for example, a loudspeaker and a microphone 768.
[0119] Additionally, in the realizations described here, cameras or image sensors or arrangements of elements of alternative photodiodes and possibly alternative lighting sets can be used. Instead of placing a camera behind a window on any wall of a treatment chamber, the same or a second camera can also be integrated with the oven door or with an opening in the treatment chamber.
[0120] Instead of integrating lighting into any treatment chamber wall, it can also be integrated into the oven door or the opening of the treatment chamber. Commonly, oven doors have windows to allow human operators to view the treated food and supervise processing. According to another embodiment, at least one camera or an image sensor or an array of photodiodes elements or any other type of imaging device can be integrated into an oven door or the opening of a treatment chamber. A windowless oven door for human operators can be designated as more efficient with regard to the energy factor since the heat insulation may be better. In addition, the differences in lighting on the outside do not influence the images from the treatment chamber's monitoring camera, which would then depend only on the lighting defined in the treatment chamber. However, an individual with expertise in the technique should note that such an adjustment might not be easily installed later in an existing oven.
[0121] Additionally, it may be advantageous to integrate a screen or a digital visual display on a wall outside the oven door or in any other suitable location outside the treatment chamber. This screen can show images captured from the treatment chamber's monitoring camera. This allows a human operator to visually supervise the cooking process, although it is an object of the invention to make this unnecessary.
[0122] Additionally, it may be advantageous to use trays or a stack of storage trays that indicate the distribution of food. For example, when baking bread, when loading the oven, the positioning of the dough may vary for each of the baking cycles. These differences can be managed through image processing with comparison and recognition techniques. It is advantageous to have a similar load or similar food placement for each of the production cycles as shown in Figure 11. An automated positioning system can be applied when adjusting the 1100 trays. For manual positioning, at least some of used trays may have an indication of 1110 where to place the dough. As an indication, reliefs, flutes, pans, molds, food icons, food designs, or lines can be used.
[0123] Additionally, when integrating a camera sensor in an oven environment or a food processing system, the integration of cooling devices can be advantageous. There can be at least one cooling plate, at least one fan and / or at least one water cooling system.
[0124] Additionally, a shutter can be used that only exposes the camera sensor when necessary. It may often be necessary to take several photos and it may often be appropriate to take photos every 5 seconds or less. If the shutter only opens every 5 seconds, the heat impact on the camera chip is significantly lower, something that reduces the possibility of error due to a heat impact and therefore increases the system's reliability for monitoring heat treatment.
[0125] It can be additionally advantageous to take at least two photos or more or to make an exposure with several non-destructive readings and to combine the pixel values. The combination can be to take an average or to calculate a photo of at least two of the High Dynamic Image Range. In combination with a shutter or alone, it is possible to apply wavelength filters that only allow the passage of relevant wavelengths, for example, visible light or infrared radiation. This can additionally reduce the heat impact on the camera chip and, in this way, increase the reliability of the system for heat treatment monitoring even further.
[0126] In another embodiment illustrated in Figure 12, the integration of a sensor system for oven shelves or mobile carts used in some oven designs, can be used. For the rotation of the oven shelves, the sensor system can be integrated into the oven shelf as demonstrated in 1200. The sensor system is integrated above at least one of the trays carrying food. The sensor system in the trailer can have at least one sensor of the following: hygrometer, temperature insertion sensor, temperature sensor in the treatment chamber, acoustic sensors, scales, timers, cameras, image sensor, arrangement of photodiodes elements. Part of the sensor system integrated with the shelf also includes support devices such as lighting or cooling as demonstrated in this invention. It is an additional object of the invention, to have an electrical connection such as a wire or electrical plugs in the shelf assembly as shown in 1210. It is additionally advantageous to integrate at least part of the sensor system into the rotating oven wall shelves as shown in 1220. This is advantageous to reduce the effects of heat on the sensor system. For the camera, the image sensor or the arrangement of elements of photodiodes it is advantageous to apply an image rotation algorithm or motion correction in order to correct the rotation of the shelf or the movement of the food. This algorithm can be supported by means of a pre-established parameter or measured from the kiln control with respect to the speed of rotation or movement.
[0127] In another embodiment, a graphical user interface (GUI) can show pictures of each of the trays and the deck inside an oven. In a convection oven the final time for each of the trays can be determined separately. This means that if one tray is finished before another, the user can receive a signal to empty that tray and leave the others inside. This is advantageous because many ovens may not have the same results for different trays. Additionally, it is possible to cook different products on each of the trays, if they have the same cooking temperature. In this way, it is possible to operate a single furnace that is more flexible and efficient.
[0128] In another embodiment, the oven can also determine the distribution of the cooked goods on a tray. An oven can also reject trays that have been loaded incorrectly.
[0129] Using one or more of the sensors described above, data on the cooking or food processing procedure can be collected. In order to enable reliably automated food processing or cooking, processing machines such as ovens or conveyor belts need to learn how to extract relevant data from all data, how to classify processed food and the stage of food processing based on this data, and how to automatically control processing based on the data and classification. This can be achieved through a heat treatment monitoring system that is able to control a cooking process based on machine learning techniques.
[0130] Figure 13 demonstrates a control unit and a data processing diagram according to which, the data of any of the above achievements can be managed.
[0131] Here, the control unit or the heat treatment monitoring system 1300, for the heat treatment machine 1310, recognizes the food to be processed with any of the sensor systems described here. Recognition of the food to be processed can be achieved with the single sensor data entry matrix {EMBED Microsoft Formel-Editor 3.0}. This sensor data entry matrix or a reduced representation of it can be used to identify a food treatment process with its characteristic data or its fingerprint data.
[0132] The 1300 control unit has access to a database that allows to compare the sensor data input matrix with previously stored information indicated with 1301. This allows the 1300 control unit to choose a program or a processing procedure for control for the present food treatment. Part of this procedure is according to an X c mapping of the X c sensor data input matrix to a data control actuator matrix X c .D aXc = D b. (Formula 1.00)
[0133] With the data control actuator matrix D aXc = D b, the heat treatment machine 1310 controls food processing, for example, by controlling oven control parameters such as power supply or start and end of processing time. The heat treatment machine then operates in a closed loop control mode. Typically, the sensor input matrix D aXc = D b is significantly higher in dimensions when compared to the control data actuator matrix D aXc = D b.
[0134] According to one realization, it is advantageous to find a DX = D mapping, as well as a reduced representation of the DaXc = D control data input matrix with the methods from machine learning. This is because the type of food to be processed and the procedures accordingly, are usually individually different.
[0135] From a data processing point of view, the relationship between the input of sensor data and the appropriate output of the actuator can be: highly non-linear and time-dependent. Today these parameters are chosen through human operators, commonly with significant knowledge regarding the configuration of the time consuming machine of the food treatment machine. According to an embodiment of the present invention with initial data sets from a human operator, machine learning methods can perform the configuration of a future system and present the configuration times as well as increase and intensify the processing efficiency as well like quality.
[0136] All applied data can be stored in databases. According to the invention, it is beneficial to connect the heat treatment machine with a chain network. With the means of this chained network, any database data can be exchanged. This allows a human operator to interact with various machines and heat treatment, distributed locally. In order to do this, the heat treatment machine has equipment to interact with a chain network and to use certain protocols such as Transmission Control Protocol (TCP) and Internet Protocol (IP). According to the invention, the heat treatment machine can be equipped with daisy chaining devices for a local area daisy chain (LAN), a wireless area daisy chain (WLAN) or a daisy chain access mobile phone used in mobile telecommunication.
[0137] In any of the previously described achievements, a cooking or food processing procedure can contain a learning phase and a production phase. In the learning phase, a human operator places the food in a heat treatment machine. The food is heat treated as desired through the human operator. This can be done with or without preheating the heating chamber. After thermal processing, the human operator can specify the type of food and when the desired processing state is reached. The human operator can also provide information when the product is to be cooked, over cooked and in what state of processing is desired.
[0138] Using the machine learning methods described above, the machine calculates the processing parameters for future food production. Then the heat treatment machine or heat treatment machines in a connected chain network can be used to have additional learning phases or to enter automated production. When in automated production, the human operator only places the food in the heat treatment machine with an optional preheat. The machine then detects the food in the treatment chamber and performs the previously learned heat treatment procedure.
[0139] When the desired food processing state is reached or reached, or simply when the bread is ready, the machine ends the heat treatment process. The machine can also do this by opening the door or by terminating the power supply or by venting hot air out of the treatment chamber. The machine can also give the human operator a visual or an audible signal. In addition, the heat treatment machine can request a return from the human operator. The machine can ask for the choice of a category such as undercooked (rare), good (to the point) or over cooked (well done). An automated loading system that loads and unloads the treatment chamber can fully automate the procedure. For this purpose a robotic arm or a convection mat can be used.
[0140] Recent techniques in machine learning and food processing control have been examined to create adaptive monitoring. The classifications of: Artificial Neuron Chain Networks (ANN), Support Vector Machines (SVM) and K-Near-Neighbor Neighbors (KNN) have been investigated since they are applicable to special applications with regard to processing of food. A target of the present invention is to assess what machine learning can achieve without a defined model process through a human operator.
[0141] The following is a brief overview of the theories supporting the present invention. This includes techniques to reduce data from dimensionally reduced sensors such as Principal Component Analysis, Linear Discriminant Analysis and Isometric Characteristics Mapping. It can also include an introduction to supervised as well as unsupervised classification and learning methods such as K-Nearly Neighbors (KNN), Artificial Neuron Chain Networks (ANN), Support Vector Machines (SVM) and enhanced learning. For number format, the thousand separator is a comma "," and the decimal separator is a period "."; therefore, one thousand is represented by the number 1,000.00.
[0142] Extraction and dimensionality reduction feature
[0143] The present invention does not seek, nor does it wish to achieve human behavior on the part of machines. However, the investigation of something like cognitive ability inside and through food processing machines and artificial agents production machines capable of managing food processing functions can provide an application scenario for some of the most sophisticated approaches in what it says. respect to cognitive architectures. The approaches to production machines can be structured within a cognitive perception architecture - action loop, as shown here in Figure 14, which also defines cognitive technical systems. Cognitive abilities such as perception, learning and knowledge gain allow the machine to interact with an environment autonomously through sensors and actuators. Therefore, in what follows, some methods known from machine learning that will be suitable for different parts of a cognitive perception - action loop working in a production system will be discussed here.
[0144] If a technical cognitive system simply has a characteristic representation of its sensor input, it may be able to manage a high volume of data. In addition, the extraction characteristics emphasize or increase the signal-to-noise ratio by focusing on more relevant information in a data set. However, there are several ways to extract relevant characteristics from a data set, the theoretical aspects of which are summarized below below.
[0145] In order to select or learn characteristics in a cognitive way, we need to have a method that can be applied completely automatically, without the need for human supervision. One way to achieve this is to use dimensionality reduction (DR), where a data set - with a size tx n is mapped over a data set - is smaller than size tx n. In this context, it refers to - 'as an observation space and refers to as a characteristic space.
[0146] The idea is to identify or learn a higher dimensional distribution in a specific data set by creating a representation with a smaller dimension.
[0147] The methods used to find characteristics in a dataset can be subdivided into two groups: linear and non-linear, as shown in Figure 15. The techniques of linear dimensional reduction appear to be over performed by the reduction of non-linear dimensionality , when the data set has a non-linear structure. This is due to the factor that nonlinear techniques generally have longer run times than linear techniques. Additionally, in contrast to non-linear methods, linear techniques allow for a more direct approach of forward and backward mapping (return mapping). The question is whether a linear dimensionality reduction technique is sufficient for food processing, or whether linear techniques produce more advantages than costs. The following nonlinear techniques are very advantageous for artificial data sets: Hessian LLE, Laplacian Eigenmaps, Local Linear Inlay (LLE), Multiple Layer Autocoders (ANN Aut), PCA Kernel, Multidimensional Scaling (MDS), Isometric Characteristics Mapping ( Isomap), and others. As a result, ISOMAP proved to be one of the best tested algorithms for artificial data sets. We think that the ISOMAP algorithm is apparently the most applicable nonlinear dimensionality reduction technique when it comes to food processing. Therefore, ISOMAP and two linear dimensionality reduction techniques are presented below. Core Component Analysis
[0148] Principal Component Analysis (PCA) allows the discovery of characteristics that separate a data set by variance. It identifies a set of independent characteristics that represent as many variances as possible from a set of data, but are smaller in size. PCA is known in other disciplines such as the Karhunen-Loèvre transformation and the part referred to as Singular Value Decomposition (SVD) is also a well-known name. It is often used in statistical patterns or facial recognition. In summary, it computes the Eigen vectors and the Eigen dominant covariance values of a data set.
[0149] We want to find a representation - of lower dimension with elements of tx p of a data set of higher dimension tx p of an adjusted average matrix, keeping as much of variance as possible and with columns not correlated with the purpose of computing a V representation of data for the x data set. Therefore, the PCA seeks an MPCA linear mapping of MPCA size that maximizes the term MPCA with MPCA and MPCA as the covariance of the '' - matrix. Solving the Eigen problem with MPCA (Formula 2.3)
[0150] We obtain the main Eigen values ordered with. The desired projection is given by Y = XM PC (Formula 2.4)
[0151] Which gives us the desired projection over the MPCA linear base. It can also be shown that the Eigen vectors or the main components (PCs) that represent the variance within the representation of high dimension data are given by the first columns of the selected matrix by means of variance. The value is determined through the analysis of the residual variance reflecting the loss of information due to the reduction of dimensionality.
[0152] Finding a linear orthogonal combination of the variables with the greatest variance, the PCD reduces the size of the data. PCA is a very powerful tool for analyzing data sets. However, you may sometimes not find the best lower dimensional representation, especially if the original data set has a linear structure. Linear Discriminant Analysis
[0153] Regardless of the utility of the PCA, the Linear Discriminant Analysis (LDA) can be seen as a supervised dimensionality reduction technique. It can also be categorized as something that uses a linear method because it also provides a linear MLDA mapping for a data set - for a smaller dimension matrix -, as stated through MPCA in equation 2.4. The necessary supervision is a disadvantage if the desire behind everything is to create a completely autonomous system. However, the LDA supports a misunderstanding of the nature of the sensor data because it can create characteristics that represent a desired test data set.
[0154] Because of the fact that the details of the LDA and Fisher's discriminant are known, the following is a short and simplified survival. We assume that we have zero mean data X. A supervisory process provides the class information to divide X into C classes with the zero mean data for the class - '. We can compute this with

[0155] The dispersion within the class, a measurement for the variance of the class c data for its own mean. Dispersion between classes follows

[0156] The dispersion between classes is a measure of the decade-by-one variance of the classes relative to the mean of the other classes. We obtain the MLDA linear mapping by optimizing the dispersion ratio between class and within the class in the lowest dimensional representation using Fisher's criterion

[0157] Maximizing Fisher's criterion by solving the Eigen problem for S-S provides Eigen values - _ 1 that are not zero. Therefore, this procedure looks for the optimized characteristics to separate the classes given in a subspace with linear projections.
[0158] Therefore, the LDA separates a lower dimensional representation with a maximized ratio of variance between classes for variance within classes. Isometric Characteristics Mapping
[0159] The PCA and LDA methods produce a linear mapping from high-dimensional data sets to a lower dimensional representation. This can be expressed as a distribution of learning in an observation space and finding a representation for it in a space of lower dimension characteristics. For data sets with a non-linear structure, such as Swiss-roll data sets, linear projections will lose the non-linear character of the original distribution. Linear projections are not able to reduce the dimension in a concise way: data points in the characteristic space may appear close even though they are not or are in the observation space. In order to face and solve this problem, non-linear dimensionality reduction techniques have recently been proposed in relation to linear techniques. However, it is an unclear priority whether, in fact, nonlinear techniques will perform established linear techniques such as PCA and LDA for data from food processing sensor systems.
[0160] Isometric features mapping or ISOMAP algorithms attempt to preserve geodesic parity or curvilinear distances between data points in the observation space. In contrast to an Euclidean distance, which is the ordinary or direct distance between two points that can be measured with a ruler or with the Pythagorean theorem; the geodetic distance is the distance between two points measured over the distribution in an observation space. In other words, we don't take the shortest path, but we have to use neighboring data points as means to jump between data points. The geodesic distance of data points V in the observation space can be estimated by constructing a graph - of neighborhood that connects the data point with its neighbor - '■ - nearest in the data set •' -. A matrix of geodesic distance of parity can be constructed with the shortest path Dijkstra algorithm. In order to reduce the dimensions and obtain a data set - the multidimensional scaling (MDS) can be applied to the geodetic distance parity matrix. MDS seeks to retain parity distances between data points as much as possible. The first step is the application of a stress function, such as a gross stress function given through

[0161] In order to obtain a measurement of the quality or the error between the parity distances in the characteristics and in the observation spaces. Here, lxi xjl is a Euclidean distance from the data points xi and xj in the observation space with yi and yj being the same for the characteristic space. The stress function can be minimized by solving the Eigen problem of the parity distance matrix.
[0162] Therefore, the ISOMAP algorithm reduces the dimension by retaining a geodesic parity distance between the data points as much as possible. Classification for machine learning
[0163] With regard to machine learning, it is not only the extraction of characteristics that is of great scientific interest, but also the need to make decisions and judge situations. Classification techniques can help a machine to differentiate between complicated situations such as those encountered in food processing. Therefore, classifiers use so-called classes that segment existing data. These classes can be learned from certain training data sets. With regard to current research on AI and cognitive machines, Artificial Neuron Chain Networks have been developed relatively recently in the process. In comparison, the concepts of Kernel Machines and enhanced learning only appeared recently, but showed increased cognitive abilities. Artificial Neuron Chain Networks
[0164] Artificial Neuron Chain Networks (ANN) have been discussed extensively for decades. ANNs were one of the first successes in the history of Artificial Intelligence (AI). Using natural brains as models, several artificial neurons are connected in a chain-network topology in such a way that an ANN can learn to approach functions such as pattern recognition. The model allows a neuron to activate its output if a certain threshold is reached or exceeded. This can be modeled using a threshold function. Natural neurons appear to “fire” with a binary threshold. However, it is also possible to use a sigmoid function,

[0165] With ■ as a transition parameter. For all incoming connections, a weight factor '. adjustable is defined, which allows ANN to carry out the so-called learning paradigm. A threshold function p can be expressed using the weighting factors «ie the outputs from the preceding neurons - ■ ', = 1 - ■', with a matrix vector annotation. Neurons can be layered in a forward, Multi-Layer Perceptron (MLP) food structure or, for example, with an infinite input response achieved using loopbacks with a delay element in the so-called Chain Chain of Recurrent Neurons. An MLP is a forward chain feeder with a layered structure; several hidden layers can be added if necessary, to solve non-linear problems. MLP can be used with continuous threshold functions such as a sigmoid function in order to support the rear propagation algorithm stated below for supervised learning. This attempt is to minimize the error - in

[0166] From the output of the designated output -., Where the particular weights are recursively adjusted. For an MLP with a hidden layer, if they are hidden layer values, they are input values, - ■ - it is the learning rate and 7. - -;., Then the weights of the hidden layer 1 2 wj and the input layer wj are adjusted according to,

[0167] The layers are enumerated starting from the input to the output. For rear propagation, the weights are adjusted for the corresponding output vectors until the general error can no longer be reduced. Finally, for class classification - the output layer can consist of either the output neurons - representing the probability of the respective class, or a single output neuron that has defined ranges for each of the classes.
[0168] Therefore, ANN can learn from or adapt to a set of training data and can find a linear or non-linear function from the input neurons - to the output neurons -. This can be used for classification to differentiate a set of classes in a data set. Kernel machines
[0169] In general, a classification technique should serve the purpose of determining the probability of classes learned occurring based on the measured data. The classification can be mathematically formulated as a set of classes = - '. : and in -, with a data set represented by x. = -, and a probability of pi,

[0170] The parameter it can then be chosen separately for all classifications, or it can be learned from a set of training data.
[0171] In order to achieve learning, it is desirable to facilitate efficient training algorithms and represent complicated nonlinear functions. Kernel machines or Support Vector Machines (SVM) can assist with both purposes. A simple explanation of SVM, or in this context in particular Support Vector Classification (SVC), is as follows: in order to differentiate between two classes, good or bad, we need to draw a line and indicate which is which; since an item cannot be both, a binary decision is required, = -1 ^. If we can only find a nonlinear separator for the two classes in a low dimensional space, we can find a linear representation for the same in a higher dimensional space, a hyperplane. In other words, if a linear separator is not possible in real space, an increase in size allows for linear separation. For example, we can map with the function - ■ a two-dimensional space - = <, • "= <with a circular separator to a three-dimensional space fTXf = XfizI2x using a linear separator as shown in Figure 16.
[0172] SVC seeks for this case, an optimized linear separator, a hyperplane,

[0173] In high dimensional space for a set of classes' .. In three-dimensional space these can be separated with a hyperplane, where ° is a normal vector of -, a distance perpendicular to the origin | b | / | o | , and ° with a Euclidean standard of | o |. In order to find the hyperplane that serves as an optimized linear separator, SVC maximizes the given margin through,

[0174] Between the hyperplane and the nearest data points xi. this can be achieved by minimizing the ratio | o | / 2 and solving with the Lagrange optimized multiplier parameter αi. With, the objective of achieving this, the expression,

[0175] It has to be maximized under the restrictions ai> 0 and ∑iac = 0. The linear separator optimized for a non-deviated hyperplane is then given using,
Allowing the classification of two classes.
[0176] SVM has two important properties: it is efficient with regard to time of use in computers and can be demonstrated with equations 2.16 and 2.17. First, with the so-called support vectors or set of parameters ai associated with each of the data points, it is zero, except for the points closest to the separator. The effective number of parameters defining the hyperplane is usually much less than, increasing the performance per computer. Second, the data enter the expression 2.16 only in the form of indicative products of pairs of points. This allows the opportunity to apply the so-called Kernel trick with

[0177] Which often allows us to compute F (xi) • F (xj) without the need to explicitly know F. The Kernel function K (xi, xj) allows the calculation of the indicative product of the pairs of input sets in space corresponding characteristics, directly. However, the Kernel function applied throughout the present invention is the Gaussian Radial Base Function and has to fulfill certain conditions, such as in,

[0178] With Y as the adjustable kernel parameter.
[0179] Because of the factor that we have discussed so far, only binary decisions between two classes, we note here that it is also possible to allow for smoother, multiple-class decisions. The latter can be achieved in stages by coupling the parities of each class against the remaining classes _ 1.
[0180] Therefore, SVC can be used to learn complicated data. The same structures such data in a set of classes in a temporal manner. Mapping in a higher dimensional space and finding optimized linear separators allows SVM to use efficient computer techniques such as support vectors and the Kernel trick. Near-K Cotanosos Neighbors
[0181] Unlike the Support Vector Machines previously discussed, a less complicated but highly efficient algorithm called the Near K-neighbor Cotanosos (KNN) classifier can also separate classes within the data. The algorithm can categorize unknown data by calculating the distance to a set of closest neighbors.
[0182] Assume that we have a set of samples identified with ■ being part, as a member, of a known class group. If a new sample * is added, it is possible to calculate the probability of society x for a certain class with the distance of the vector to the members of the existing classes. The probability of membership as a member in class A is 90% compared to class B with 6% and C with only 4%, the best results appear to be apparent. In contrast, if the probability of membership as a member in class A is 45% and 43% in class B, this is no longer obvious. Therefore, KNN provides membership information as a function for K's closest neighbors and their membership as a member in possible classes. This can be summarized with

[0183] Where pij is the probability of membership as a member in the ith class of the jth vector within the labeled sample set. The variable is a weight for the distance and its influence with regard to the contribution to the membership value as a calculated member.
[0184] When applied, we often adjust '■ = - and the number of closest neighbors
[0185] Enhanced learning
[0186] In contrast to previous learning methods, which learn models of functions or probabilities from training data, enhanced learning (RL) can facilitate learning using environmental returns from an agent's own long-term actions , without the need for a teacher. This involves the difference between supervised and unsupervised learning. If a long-term goal is pursued, a positive environmental return, also known as a reward or reinforcement, can support improvement. An agent can learn from rewards how to optimize his policy or strategy to interact with the real world, the best policy being one that optimizes the total expected reward. RL does not require a complete previous model of the environment or a full reward function. Therefore, artificial agents indicate a cognitive ability and act in a similar way to animals, which can learn from negative results like pain and hunger and from positive rewards like pleasure and food. In this case we can choose that the agent has a value function approach, in which he tries to maximize his environmental return.
[0187] In RL an agent takes actions, even, in an environment that he perceives to be his current state, ts, with the objective of maximizing long-term rewards, rt, learning certain policies, ", However, before so that we can start learning with reinforcement we have to find the answers with regard to the appropriate design of the agent. The agent could try to maximize the expected return through an estimate of the return for a policy ", This behavior of the agent is also referred to a as a value function estimate. The agent can evaluate the action by estimating the value of the state using a value function - state 5, considering certain policies • ■ q that are continuously differentiable, as in

[0188] Using this function, the agent can estimate the expected return for a given state and following a policy. The expected return for an action could also be estimated following state and policy data. Therefore, the agent chooses an action considering the given state from the state - action or Q function, as in,

[0189] Therefore, the next action depends on the reward function 7 and in order to allow the agent to grant a concession for future expected rewards on current rewards, the discount factor £ • £ 1 can be selected. It is possible to establish how much the agent should discount for future rewards, for example, future rewards are irrelevant to = ':.
[0190] In RL, methods can be subdivided into groups such as a method-based value function or direct policy search. Many different actor - critical algorithms are value - based methods estimating and optimizing the expected return for a policy. In order to carry out a method based on value function, the behavior for an artificial agent and the correlated control problem can be declared as a Markov Decision Process (MDP). The system perceives its environment on the set of continuous state, where 5: = e are the initial state. It can choose from a set of possible actions of -v = - 'with respect to a stochastic and parametric policy defined as π (at | st) = p (at | st, wt), with the parameters policy - •• - ••. With a learning policy, it can be mapped from states to actions with respect to expected rewards> = -. The reward after each action depends on><-v. If no environmental model is available, the aforementioned actor - critical methods can potentially develop policy search algorithms. The name is derived from the theater where an actor adapts his actions in response to feedback from a critic. This can be achieved using a given evaluation function as a weight function of a set of characteristics or a so-called base function ^ (s), which then provides the approximation of the state-value function with the value function parameters. v, as in

[0191] The improvement and intensification of the policy is a matter of optimization that can be addressed with a policy gradient. The choice of the policy gradient method is critical to convergence and efficiency. Both appear to be achieved through the natural Actor - Critical algorithm (NAC) as described by J. Peters and S. Schaal, “Natural actor-critic”, Neurocomputing, Vol. 71, no 7-9, pp. 1180-1190, 2008, where an actor intensifies and improves by using a policy derivative from a critic g according to equation 2.24,

[0192] The steps for intensifying and improving policy parameters of the NAC algorithm are then calculated using, wt + 1 = wt + α g, (Formula 2.25)
[0193] where α is a learning rate, and g is the natural gradient calculated using Fisher's metric or is derived from the policy as shown in the publication of the mentioned NAC algorithm. The NAC algorithm with the LSTD-Q is fully documented in table 1 on page 1183 by J. Peters and S. Schaal, “Natural actor-critic”, Neurocomputing, vol. 71, no. 7-9, pp. 1180-1190, 2008. The same is applied with a parametric policy with the initial parameters w = w0 comprising the following steps in the pseudocode:
[0194] 1: START: initial state of draw s0 ~ p (st) and select parameters At + 1 = 0; bt + 1 = zt + 1 = 0
[0195] 2: For t = 0,1,2, ... of
[0196] 3: Execute: Draw action at ~ π (atIst), observe the next state
[0197] st + 1 ~ p (st + 11 st, at), and reward rt = r (st, at).
[0198] 4: Critical Evaluation (LSTD-Q (À)): Updates
[0199] 4.1: Basic functions:


[0200] 4.2: statistics: zt + 1 = ÀZt + Φt; At + 1 = At + zt + 1 (^ t Y ^ t) T .... ); bt + 1 = bt + zt + 1 rt,
[0201] 4.3: Critical Parameters: [vt + 1, gt + 1] = At + 1 1 bt + 1,
[0202] 5: Actor: if an estimated gradient is accurate, it updates the policy parameters,
[0203] 5.1: wt + 1 = wt + α g t + 1 and forget (reset) the statistics. TERMINATION.
[0204] The base functions ^ (s) can be represented by mapping the input of sensor data to a feature space as we discussed elsewhere in this document. In this case, the basic functions are the same as the characteristic values. The base functions can also be chosen differently or the agent can use raw sensor data. The base functions can also incorporate adaptive methods or an own learning step that is maximized with the results of the reward function.
[0205] It is important to note that other RL agents may also be applicable. Many other concepts of policy learning agents can be applied. It is additionally inventive to use other sources as a rt reward signal in addition to the rating output or quality indicator. For example, it is possible to apply a later process or a pre-process sensor as a source of reward signal. The reward function could be the probability value between 0 and 1 or -1 to 1 of a post process sensor measurement data to be part of a good or bad class, something that is determined through a classifier as it is. described above. In case a pre-process sensor is used for a given reward ": an RL agent could find a set of parameters to achieve this goal. Therefore, reinforcement learning can be a step towards a goal in the long term as it encompasses learning a policy from given rewards using policy search algorithms such as that of Actor - Natural Critic. Architecture of Cognitive Technique
[0206] An artificial agent is anything that perceives its environment through sensors and that acts in consequence and through these actuators. An agent is defined as an architecture with a program. The role model of inspiration for this is natural cognition, and we want to perform a cognition of similar performance for technical systems. Therefore, the agent will be equipped with cognitive skills, such as summarizing information, learning and decision making for a manufacturing workstation. As part of the process, this section introduces an architecture that creates and allows agents to manage production functions. In order to do this, the agents follow a cognitive perception - action bond, through the reading of data from sensors and defining actions for the actuators.
[0207] A natural cognitive ability is the ability to summarize relevant information from a larger data set and to differentiate between categories within this information. The transfer of this concept from natural cognition to the world of mathematical data analysis, a combination of data reduction techniques and classification methods is used in accordance with the present invention to achieve and achieve something that exhibits similar behavior. In industrial production, many manufacturing processes can be performed using a black box model that focuses on the pros and cons of the box, its exits and entrances, rather than what actually happens inside the box. The black box connections that can be used in production systems are generally sensors and actuators. Sensors such as cameras, microphones, tactical sensors and others, monitor the production process. These systems also need actuators such as linear operators or robotic positioning operators, in order to interact with the environment. For each of the production processes, these actuators have to be parameterized. In order to learn how an agent can adaptively control at least one parameter of these production systems, many combinations of self-learning algorithms, classification techniques, knowledge repositories, feature extraction methods, dimensionality reduction techniques and learning techniques distribution systems could be used. The present invention also provides different control techniques, both open-loop and closed-loop, using multiple different sensors and actuators. After many simulations and experiments, a simple architecture that demonstrates how these techniques can be combined, proved to be something successful and reliable, at least when it comes to food processing. However, food processing can be interpreted as a form of black box and can therefore be applied to other types of production processes.
[0208] Figure 17 illustrates a cognitive architecture that may be suitable for design agents that can provide monitoring or adaptive process control for production functions. The diagram describes the steps of the communication and information processing unit. Natural cognition apparently summarizes information, first, through the identification of representative symbolism, such as structured signs. A similar process can be achieved using dimensionality reduction (DR), in which the agent uses a low dimensional representation of the input sensor data. So, natural cognition recognizes whether or not knowledge about sensational input events is already present. This step can be achieved and achieved through the use of classification techniques that categorize “sensory” events or characteristics. A natural subject can decide to learn or plan new actions. In order to replicate this factor, the architecture of the present invention offers self-learning techniques that feed a processing logic. In the quest to achieve quick reactions without the need to initiate a complex decision-making process, we can also “directly connect” a sensor input that can directly initiate an actuator when using a closed loop control design. Therefore, the architecture of the present invention can be designed with respect to four modes of use, which will be discussed individually below: first, summarizing relevant information; second, receiving feedback from a human expert on how to monitor and control processes, or supervised learning; third, act with respect to knowledge learned; and fourth, to automatically control processes in previously unknown situations.
[0209] As with other cognitive architectures, the goal here is to create agents with some type of artificial intelligence or cognitive abilities related to human beings.
[0210] The agents can be composed of several components based on dimensionality reduction and different classification techniques, which allow us to compare the performance of the composite agents and the modules in terms of the quality of food processing in general. Many different reductions in dimensionality and classification techniques can be applied and some of these have already been evaluated in the research project. The cognitive architecture of the present invention offers the following modules for the composition of agents: Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA), Isometric Characteristics Mapping (ISOMAP), Support Vector Machines (SVM), Nearby Neighbors -K Cotanosos (KNN), Artificial Neuron Chain Networks (ANN), and Reinforced Learning (RL), in conjunction with other methods. Three realizations of the present invention of control agents within this architecture would be the agent A ISOMAP, SVM ,. ANN and PID power supply control, or Agent B connecting ISOMAP, SVM and PID power supply control or Agent C connecting ANN with Cotanosos KNN, for control. Summary of relevant information
[0211] In a natural condition of cognition, we summarize or absorb information from everything we hear, feel and see. So, in general, we just remember what is most interesting. Inspired by this, a technical cognitive system should similarly summarize relevant information from a production process. Working with summarized characteristics instead of working with raw sensor data has certain advantages. Many weak sensor signals can be scaled down to less, but better signals, resulting in a more reliable feature. Additionally, in order to perform real-time processing control, it is necessary to reduce the volume of input data from the sensors because a larger amount of data can have a significant influence in terms of causing longer run times across the system.
[0212] The architecture of the present invention requires a function test in order to summarize the initial information. During this agent training period, the parameter rate of the actuator that will be controlled is changed. In order to determine which information is the most relevant, the agent should explore its own share rate. After the initial benchmark test, the system analyzes the recorded sensor data in order to discover representative characteristics. The agent can solve characteristic calculations separately for different types of sensors, but the sensory units should ideally be trained to map the sensory input into the learned characteristic space. Finding a useful representation of the characteristic space is critical because the system will only be able to recognize or react to changes in characteristic values. The purpose of the cognitive processing of the present invention is to provide as much information as possible for subsequent processing steps. However, raw sensor data contains repetitions, correlations and interdependencies that can be ignored. Therefore, in order to summarize the relevant information, the most significant characteristics, or those that contain most of the information, should be identified. In order to accomplish this “cognitively”, an agent should perform this function without the necessary supervision of a human specialist. Therefore, an information extraction method is chosen, which can be applied to all different types of processing functions and to the corresponding sensor data without the need to change the parameterization or reconfiguration. The distribution of learning or dimensionality reduction techniques satisfies this need. They can reduce a data set of sensors ■ '■ of dimension: in the observation space in a data set - of the dimension in the characteristic space. Often, the new amount is much less than ■. However, many linear and non-linear dimensionality reduction techniques have been tested and tested for different purposes. The present invention provides a feature extraction technique suitable for workstation / operational production in accordance with the following requirements of the feature extraction method that works transparently and is capable of displaying the processing steps to the user. The method of extracting characteristics is capable of being carried out without supervision. The feature extraction method is executable within a reasonable time frame for configuration, especially during processing. The extracted features contain sufficient process information for reliable classification within various food loads.
[0213] In essence, the PCA seeks linear orthogonal combinations that represent a larger data set. These combinations can be calculated for the input vectors of the sensor data. These Eifgen vectors can serve as the characteristics for classification up to a threshold of d. Characteristic extraction combined with classification can be achieved using Linear Discriminant Analysis. The analysis of the same data set using LDA and three classes of learned qualities defined as “good”, “average”, and “bad / bad”, provides another set of characteristics. Feature extraction can also be performed using the ISOMAP algorithm. Unfortunately, nonlinear characteristics cannot be displayed in the same way as the linear extraction of LDA and PCA characteristics. The characteristics extracted from the methods mentioned above are compared below. The characteristics of LDA appear to contain more details than any other characteristics of PCA. Using this method of calculating, the characteristics of LDA appear to contain more processing information in fewer characteristics than PCA because they are specially designed to separate the desired classes. Additionally, it is possible to display the characteristics calculated using PCA and LDA in such a way as to make two methods more transparent than ISOMAP. The user has an idea of what a process looks like, if a feature is identified in a procedural video simply by looking at it. PCA and ISOMAP have the advantage that they can work without supervision, something that is not possible with LDA. Therefore, LDA merely serves as a comparison to PCA, but it is not considered as an alternative to the desired architecture. In addition, the characteristics of LDA appear to be very individualized for a particular process. ISOMAP has considerably higher execution times for analysis and for an out-of-sample extent. Therefore, if the classification with PCA achieves sufficient results, then it is more applicable to the system under research. Therefore, the method of choice would be PCA, unless ISOMAP shows significantly better performance in a sense to the first objective of the present invention. We have to postpone the final choice of dimensionality reduction techniques because the most important quality measures are experimental results, which are the basis of the present invention.
[0214] In essence, the reduction of dimensionality can allow agents to summarize relevant information in terms of detecting variances and similarities during a training period. This helps the agent to process only a few values compared to the significantly higher volume of raw sensor data. Additionally, dimensionality reduction can support the perception of similarities in unknown situations, for example, similar characteristics of food processing such as food size and shape of the food, even if they are not part of the training. This can improve and intensify the adaptability of agents with respect to unknown but similar situations. Supervised learning from human experts
[0215] In natural human cognition, for example, during childhood, we often learn from others how to manage complex functions. Similarly, a machine should be able to learn this function initially from a human specialist. Supervised learning appears to be the most efficient way to establish a cognitive agent for production. In industrial production, a qualified human supervisor is usually present when the production system is being installed or configured. The architecture we are examining uses human - machine communication in order to receive feedback from an expert, for example, through an intuitive user graphic on a "tablet" computer with a touch screen. As mentioned above, at least one action test per actuator or function test is necessary in this architecture as an initial learning phase. During these tests, the agent executes an actuator from within the actions with a desired range, and the input of the sensor data is stored. After this action, a specialist provides feedback regarding the factor whether the robot performed the actuator correctly or whether the action was unsuccessful and unwanted. The feedback can occur in several different categories in such a way that different types of failures and exit strategies can be defined. A classification technique can then collect the characteristics in conjunction with the corresponding supervisory feedback. Combined with comparison tables, the classifier module will serve as knowledge and as a planning repository for a classification of the current state of the system. How an agent can perform its own actions and provide a return itself will be of importance for the next section; this section mainly covers the cognitive capacity of learning from a human specialist and the application of this knowledge for monitoring purposes.
[0216] Support vector machine, K-Near-Neighbor Neighbors, and Artificial Neuron Chain Networks as classification techniques were discussed. The more a human expert teaches the machine, the more likely it is that the system will achieve the desired goal. In order to save costs, the necessary human supervisor time should be minimized to just one or two benchmark tests, if possible. Semi-supervised learning
[0217] The previous discussion shows how agents in the investigated cognitive architecture perceive and assimilate their surroundings and learn from a human specialist, as well as exhibiting their knowledge in terms of monitoring. The monitoring signal provided based on selected characteristics is obtained from different sensors that are interpreted using a trained classifier. This monitoring signal appears to have an enhanced quality and may be applicable with regard to the control of processing parameters. The agent would then change his position from observing the processing in order to really act with respect to the acquired knowledge. However, if an agent is also applicable to process industrial processing control, the agent has to fulfill several requirements with near-perfect performance. The following are some of the requirements for the cognitive architecture in question: the processing control module should be able to complete at least one control cycle from the sensor input to the actuator output. The controller parameter should have an effect on the processing output when changed, while simultaneously responding in a timed manner. The processing control module should be optimized in terms of providing a balance of reliable stability and required dynamics.
[0218] In order to achieve a robust and strong processing control that is suitable for the industrial production process, fast or real-time closed-loop control is often required. The advantage of the architecture under investigation is that the use of features instead of data from raw sensors allows a faster complementation of the control loops with a minimum of information loss. In this architecture, any type of controller design can be implemented as long as it fits with the rating output. A simple version would have three possible classification output values: undercooked, class I; correct, class II; and over stew, class III. This can be expressed using

[0219] Where are the class probabilities and the quality indicator.
[0220] A PID controller could adjust a parameter of the system's actuators according to the monitoring signal discussed above covering supervised learning from human experts. Combining PID - control with the classification results allows agents to perform controlled processing of power supply. This can be done as shown in

[0221] With - for proportional, - for integral, and - for derivative behavior. The objective is to minimize the error =: between the quality indicator. '' = -, the output of the classification module, and the desired value of 0.0. In this context, the inventive applicability of the desired value depending on a class probability related to the quality indicator provides the opportunity to vary this value to optimize the results of the desired process. One approach describes a PID control with an ANN and corresponding experiments. Another approach investigates the use of an SVM classification module to control food processing. Unsupervised learning
[0222] As suggested, a self-learning mechanism is integrated into the system of the present invention. A check for news based on training characteristics can detect new or previously unknown situations. In these cases, the system performs another test action and classifies the new food using the characteristics previously trained. This time you don't need to consult a human specialist; the mechanism can map the knowledge acquired over the new food automatically and can adjust the control process accordingly.
[0223] In order to achieve a process return control, the monitoring signal is used as a control variable. An active variable, which could possibly be any changeable process parameter with an interrelation with '=, the power supply appears to be adequate for its low inertia and its strong relationship with. ’■■ = ■. Its magnitude is calculated using a PID algorithm as shown in equation 3.2. In order to achieve processing control, the agent closes the loop by connecting the monitoring signal to a PID controller, as shown in equation 3.2. The return controller is designed as a simple input and simple output (SISO) control system, which receives the monitoring signal. = ■ from the rating unit with ::: s 1 for very low power supply and -1 J for very high power supply, and use this as reference values to minimize controller error.
[0224] The previous description indicated how cognitive agents learned from feedback through a human specialist. It should be possible for the cognitive system to learn from its own actions or to give a return to itself. This type of cognitive ability can be achieved with reinforced learning (RL). A classifier can assume the role of providing return and provide an RL agent with rewards for his own actions. The agent then learns a policy on how to act or how to bake based on the return or rewards received from his previous performance. In order to test this, the learning function is therefore for the agent to learn how to process food based on knowledge acquired at different speeds without additional supervision from a human specialist.
[0225] In order to achieve the given learning function using reinforced learning, a reward function is necessary. As the system has multiple sensor data inputs, a classifier identifying characteristics of good cooking, such as a Support Vector Machine, can serve as an rt reward function, as shown in Figure 23 here. These rewards can fill the role of a critic in the Natural Actor - Critical Method, which was previously described. Therefore, the next action that the agent chooses is that of absolute power supply. The action chosen depends on the policy learned, as shown here,

[0226] The policy parameters depend on the gradient £ ew: -i, as in equation 2.25. However, for a complete review of the applied algorithm, please refer to the Natural Actor - Critical Algorithm with Learning Less Squares of Time Difference LSTD-Q (À). The policy should allow the agent to map from st states, to at actions, through learning from rt rewards. Rewards naturally influence policy parameters. The best policy of the RL agent of the present invention under investigation of the present invention has been discovered with a sigma function.

[0227] where Lm is the maximum allowed power and n is the scanning noise determined by the product of a random number from -1 to 1 and the scanning parameter ε.
[0228] The present invention investigated modules that are suitable for a cognitive architecture for food production machines within a cognitive perception - action loop connecting sensors and actuators. Cognitive skills are: to summarize relevant information; to learn from a human expert; to use the knowledge acquired for decision-making; and to learn how to manage situations that the agent has not previously been trained on.
[0229] As previously mentioned, the machine learning techniques previously discussed can be implemented in any of the realizations described here of a system for monitoring heat treatment.
[0230] In the following, an embodiment of a heat treatment system 100 illustrated in Figure 18A and 18B will be described. The system for monitoring heat treatment comprises an oven 100 and a monitoring apparatus 150 as described here above with respect to Figures 1A and 1B. However, the realization as described with respect to Figure 18A should not be restricted to the use of window 130 as described above, so any type of window 1800 adapted to allow the camera 160 to observe the food to be heated, can to be used. The realization of the monitoring apparatus 150 should not, additionally, be restricted to its use in the realization of Figures 1A and 1B, but it can additionally be used in cooking or in the pre-cooking lines of in the food heating lines as described in the with respect to Figures 8 and 10 or any other embodiment as described above.
[0231] A block diagram of an embodiment of the monitoring apparatus 150 is shown in Figure 18B. Accordingly, the monitoring apparatus 150 and the monitoring system 100 comprise a sensor unit 1810 having at least one sensor 1815 to determine the current sensor data of the food being heated, a processing unit 1820 to determine the sensor data currents from the current sensor data, and a monitoring unit 1830 adapted to determine a state of the current heating process in a current heating process of the monitored food by comparing the current data of characteristics with reference data of a reference reference heating process. In addition, the heat treatment monitoring system comprises an 1840 learning unit adapted to determine a mapping of current sensor data to current characteristic data, and to determine reference characteristic data from a reference heating process based on the characteristic data from at least one preliminary heating process. The monitoring apparatus 150, additionally, comprises a classification unit 1850 adapted to classify the type of food to be heated and to choose a reference heating process corresponding to the given type of food. It should be emphasized here that the respective units 1820, 1830, 1840 and 1850 can be provided separately or can also be implemented as software being executed via a CPU of the monitoring device 150.
[0232] The sensor unit 1810 comprises at least one sensor 1812, in which the sensor 1812 can be any sensor as described in the above description, in particular a camera 160 as described with respect to Figures 1A and 1B, any sensor of the sensor system 850 described with respect to Figures 7 or 8 or the sensor system described with respect to Figure 12. In particular, the at least one sensor 1812 of the sensor unit 1810 comprises at least one hygrometer, a temperature insertion sensor, a temperature sensor in the treatment chamber, acoustic sensors, scales, timer, camera image sensor, the means to determine the temperature profiles of temperature insertion sensors, the means to determine electromagnetic emissions or acoustics from the processing of the food to be treated as light or sound being reflected or emitted in response to light or sound emitters or sources, the means to determine using 3D measurements of the food to be heated including 3D or stereo or radar camera systems, or the means to determine the type of constitution or standard or optical characteristics or volume or mass of the food to be treated. According to this realization, it is beneficial to use as many sensor data inputs as possible. Which sensor signal provides the best information is difficult to predict. As the algorithms detect the variance of a cooking reference, the 1840 learning unit used to implement machine learning can choose different sensor data for individually different cooking products. Sometimes the color and volume variance can be the most significant data, sometimes it can be humidity, temperature and weight.
[0233] In one embodiment, the sensor unit 1810 comprises camera 160 as the only sensor 1812, which has the advantage that no additional sensor has to be integrated in the monitoring device 150. Therefore, the monitoring device 150 can be formed as a simple and compact housing being mounted on an oven door of oven 110. However, it is also possible to provide an 1814 sensor data input interface on the monitoring device 150, through which the current sensor data of the above-mentioned sensors, can be read via sensor unit 1810 and transferred to processing unit 1820. Current sensor data from sensors 1812 is not necessarily raw data, but can be pre-processed as pre-processed pixel data HDR from camera 160 or pre-processed sensor data from laser triangulation sensors, which may contain, for example, a calculated part volume value observed food.
[0234] The processing unit 1820, the monitoring unit 1830, the learning unit 1840 and the sorting unit 1850 cooperate to provide a user with an optimized food heating result based on the machine learning techniques as it is here described above.
[0235] Here, processing unit 1820 and learning unit 1840 are provided to reduce the amount of current sensor data from the above mentioned at least one 1812 sensor. In particular, learning unit 1840 is adapted to determine a mapping from current sensor data to current characteristic data through an analysis of variance of at least one preliminary heating process, to reduce the dimensionality of current sensor data. The 1840 learning unit can be integrated into the monitoring device 150 or it can be an external unit located in another location, in which a data connection can be provided, for example, via the Internet (as described here below with respect to the use of PCA loops). The at least one preliminary heating process can therefore be based on the current sensor data from the sensor unit 1810 of the local monitoring device 150, but also be based on the current sensor data from the sensor units of the local monitoring device. additional monitoring in different locations (in the world), as long as the case of the sensor data type is comparable, with each other. Through the training heating processes, the sensor data are reduced in dimensionality, in which the data from sensors with the highest variance in time are the ones that have more weight.
[0236] The analysis of variance performed using the 1840 learning unit comprises at least one principal component analysis (PCA), a mapping of isometric characteristics (ISOMAP) or discriminating linear analysis (LDA), or a dimensionality reduction technique , which have been described in detail here above.
[0237] Therefore, an interpretation and selection of the dominant characteristics can be performed through the application of PCA or principal component analysis to a sequence of food processing data. As described above, in this way the characteristics can be selected by variance and the most prominent one can be very beneficial for monitoring. Through the performance of the analyzes described above, a mapping can be derived to map the sensor data to characterize data being reduced in dimensionality and being characteristic for the heating process unit being performed and monitored by the monitoring device 150. The mapping , which can also be received from an external server, or can be stored in a memory on the monitoring device 150, is then applied via processing unit 1820 to map current input sensor data from the unit of sensors 1810 for the data of current characteristics, which are then transmitted to the monitoring unit 1830. It is emphasized here that in some cases, the “mapping” can be for some sensor data, an identification mapping, thus some sensor data can be the same as the respective characteristic data, in particular with regard to right to pre-processed sensor data already containing characteristic values such as absolute temperature inside the heating chamber, a volume value of the food to be heated, a humidity value of the humidity inside the heating chamber. However, the mapping is preferably a mapping in which the dimensionality of the data is reduced. The learning unit can additionally be adapted to determine a mapping of the current characteristic data to characterize data through an analysis of variance of at least one preliminary heating process to reduce the dimensionality of the current characteristic data.
[0238] The 1830 monitoring unit is then adapted to determine a current heating process in a monitored food heating process by comparing the current characteristic data with reference to the reference characteristic data of a heating process. reference.
[0239] During this monitoring, one of the desired interests is to interpret the current data of characteristics and reach a decision on regular and irregular processing. With the indicated method, it is possible to collect characteristics of regular behavior and then assume irregular behavior, since the values of characteristics differ from the previously learned regular behavior. This can be supported through the inclusion of classifiers such as Support Vector Machines or K-Close Neighbor Cotanosos as described above. The monitoring unit 1830 can be adapted to determine at least one action of at least one actuator based on the data of determined current characteristics or state of the current heating process, in which the control unit 1300 as described here above, can be implemented in the 1830 monitoring unit. Therefore, the 1830 monitoring unit can be adapted to perform all machine learning techniques as described above.
[0240] According to one embodiment, the reference characteristic data of a reference heating process is compared with the current characteristic data to determine a current heating process state. The reference characteristic data can be predetermined data received from an external server or stored in a memory of the monitoring device 150. In another embodiment, the learning unit 1840 (external or internal in relation to the monitoring device 150), can be adapted to determine reference characteristic data from a reference heating process by combining predetermined characteristic data from a heating program with a preliminary set of reference data from at least one preliminary heating process being classified as part of the preliminary set through a user. The heating program can be understood as a time-dependent sequence of characteristic data being characteristic for certain types of food to be heated.
[0241] For example, a reference heating process or a predetermined heating program can be a sequence of time characteristics data of a certain type of food to be heated like a Croissant, which results in a heating or heating result. optimized cooking. In other words, if the current characteristic data exactly follow the time-dependent trajectory of the reference characteristic data points in the reference space having the dimensionality of the number of relevant characteristics chosen, the food will be heated in an optimized way after a predetermined optimized time, for example, the Croissant will be cooked perfectly. The optimized time can be dependent on the temperature inside the heating or cooking chamber.
[0242] Combining the predetermined characteristic data from a heating program with a preliminary set of characteristic data from at least one preliminary heating process being classified as part of the preliminary set by a user, means that a data point cloud of characteristics in the characteristic space of the preliminary set (for example, of at least one preliminary heating process being considered to be “good” by a user), is measured for each time point (a central point of the point cloud is determined within the characteristic space) and then used to adapt the predetermined heating program. This can be done by additionally mediating the characteristics of the heating program and the characteristics of the preliminary set equally or in a heavy manner for point in time. For example, the weighing of the preliminary set can be 25%, the weighing for the predetermined heating program can be 75%.
[0243] Therefore, at least one reference cooking (preliminary heating process) can be considered to optimize subsequent cooking. Additional returns from subsequent cookings can optimize individual cooking programs accordingly. Accordingly, it is possible to achieve a more consistent cooking quality if current cooking is being adapted using current sensor data and its calculated changes taken from the difference in current cooking and the so-called “complete truth” (cooking process). reference heating), which is the cooking program (predetermined heating program) combined with the characteristic data of at least one reference cooking (preliminary set), as well as the characteristic data from the most recent return (set preliminary) for the cooking program and its sensor data accordingly.
[0244] Therefore, it is possible to calculate significant characteristics with corresponding characteristic values from the sensor data of a reference cooking combined with the time that has passed from the cooking program. Here it is possible to use several variations of calculations of different characteristics and not select them by means of variance. One possible mechanism for selecting the variance is Principal Component Analysis (PCA), described above. When various characteristics and characteristic values over time are calculated from a reference cook, it is possible to select these sets of characteristics and characteristic values over time with the PCA.
[0245] It is possible to automatically design a control algorithm for repetitive cooking by considering at least the characteristics and data sets of the most significant characteristic values, preferably those with the most significant variances. If several reference cooks are present, it is preferable to consider the one with the highest variance and the highest repetition of reference values.
[0246] To implement the aforementioned possibility of adapting the predetermined heating program to form a "complete truth", for example, the reference heating process, the monitoring apparatus 150 may additionally comprise a recording unit 1822 for recording data current characteristics of a current heating process, in which the 1840 learning unit is adapted to receive the current characteristics data recorded from the 1822 recording units to be used as characteristics data of a preliminary heating process.
[0247] The 1850 classification unit can be provided to classify the type of food to be heated. This can be accomplished through image processing of a pixel image of the food to be heated, for example, through facial recognition techniques. After determining the type of food to be heated (hamburger bread, muffin, croissant or regular bread), the classification can be used to select a predetermined heating program or a stored reference heating process corresponding to the respective type of food to be heated. be heated. In addition, subcategories can be provided, for example, small croissants. Medium croissant or large croissant. Different reference heating process can also be stored with respect to categories other than food types. For example, there may be a reference heating program corresponding to different time-dependent environments or oven parameters.
[0248] For example, climatic data can be implemented in the cooking procedure of the present invention. Through the knowledge of the geographical altitude of the geometric position of the cooking oven, the boiling point can be determined, thus resulting in an adaptation of the cooking program. In addition, local pressure, temperature and humidity data of an oven environment can be used to further adapt the cooking program. Therefore, these data can be recorded and used as index data for certain reference heating programs, which can then be consulted in memory.
[0249] Additionally, the statistics of loads, units and corrections can also be used as data for an inventive self-learning cooking procedure. Accordingly, a history of cooking data can assist in improving the cooking procedure of the present invention. Through the distributed return being accounted for through a function definition, the cooking process of the present invention can be improved. The heat treatment monitoring systems in use can be additionally displayed on a zoomed world map.
[0250] In addition, the cooking data history can also consider and include the amount of cooking products produced in a given time. The system for monitoring heat treatment can search the cooking data history for the minimum production and the maximum production occurring periodically and estimate the occurrence of the next minimum and the next maximum. The system for monitoring heat treatment can then inform a user of the system whether too many or too few foods are produced for that period of the expected minimum and maximum.
[0251] The current heating process status is determined by comparing the current characteristic data with reference characteristic data. The comparison can be the determination of the distances from the current characteristic data and the reference characteristic data for each time point of the reference heating program. Therefore, by determining the shortest distance from the determined distances, the time point of the shortest distance can be consulted in the reference heating program and, for example, a remaining cooking time can be determined.
[0252] As described above, the sensor unit 1810 can comprise a camera like camera 160 recording a pixel image of the food being heated, where the current sensor data from the camera corresponds to the current pixel data of an image of current pixel.
[0253] The detection of characteristics for image processing can comprise the following steps: detection of edges, corners, spots, regions of interest, points of interest, color processing or images of gray levels, shapes, ridges, spots or regions of interest or points of interest. The characteristics from the sensor data can also comprise selection of target amplitude or selection of characteristics based on frequency.
[0254] Here, the edges are the points where there is a limit (or an edge) between two regions of images. In general, a border can be of any arbitrary shape, and can include joints. In practice, the edges are usually defined as sets of points in the image, which have a very strong gradient magnitude. In addition, some common algorithms will chain the high gradient points together to form a more complete description of an edge. These algorithms usually introduce some restriction with respect to the properties of an edge, such as shape, smoothness and gradient value. Locally, the edges have a dimensional structure.
[0255] The terms corners and points of interest are used interchangeably and refer to features similar to a point in an image, which have a two-dimensional structure. The name “quina” came about since previous algorithms first performed edge detection and then analyzed the edges to find rapid changes in direction (corners). These algorithms were then developed in such a way that the explicit detection of an edge was no longer required, for example, when looking at high levels of curvature in the image gradient. It was then observed that so-called corners were also detected on parts of the image that were not corners in the traditional sense (for example, a small shiny mark on a dark backdrop can be detected). These points are often known as points of interest, but the term “quina” is used for the sake of tradition.
[0256] Spots provide a complementary description of image structures in terms of region, as opposed to corners that are more like dots themselves. Either way, spot descriptors often contain a preferred spot (a maximum location for an operator response or center of gravity), which means that many spot detectors can also be considered as point of interest operators. Spot detectors can detect areas in an image that are too smooth to be detected by a corner detector. Consider shrinking an image and then performing corner detection. The detector will respond to points which are well defined in the shrunk image, but which can be smooth in the original image. It is at this point that the difference between a corner detector and a spot detector becomes somewhat vague. To a large extent, this distinction can be remedied by including an appropriate notion scale. However, due to the response properties of different types of image structures at different scales, the LoG and DoH spot detectors are also mentioned in the corner detection article.
[0257] For elongated objects, the notion of ridges is a natural tool. A peak descriptor computed from a gray level image can be seen as a generalization of a medial axis. From a practical point of view, a ridge can be considered as a one-dimensional curve that represents an axis of symmetry and, in addition, has a local ridge width attribute associated with each of the ridge points. Unfortunately, however, it is algorithmically more difficult to extract peak features from general classes of gray level images than from edge, edge or stain features. In any case, peak descriptors are often used for the extraction of highways in aerial images and for the extraction of blood vessels in medical images.
[0258] The current pixel data may comprise the first pixel data corresponding to a first color, the second pixel data corresponding to a second color, and the third pixel data corresponding to a third color, in which the first, the second and third colors correspond to R, G and B, respectively. Here, the source of illumination to illuminate the food with white light is advantageous. However, it is also possible to provide a monochromatic light source in a preferred wavelength area in the optical region, for example, 600 nm, to observe a gray pixel image at the respective wavelength.
[0259] Due to the provision of a separate analysis of pixel values of R, G and B, it is possible to implement an algorithm which can learn bread colors. Here, it is essential to segment the bread pixels from the oven pixels, which can be done using colors. It is advantageous to use pre-processed photographs with a high dynamic range (HDR) to have more intense information to have the best segmentation. Therefore, the camera is preferably adapted to generate processed HDR pixel images as current pixel data. Here, too, logarithmic scaling can be implemented, in which the camera is adapted to record linear logarithmic images or combined logarithmic linear and pixel images. To learn the bread pixels, a Chain of Artificial Neurons with return propagation or a class of SVM as described above, can be used, which are trained with photographs, where an oven is manually masked.
[0260] As an example, it may be that for cooking hamburger buns, the most significant variance during cooking is a change in color (change in pixel intensity), and a change in volume (change in number of pixels with certain intensity). This can be the two most significant characteristics during the reference cooking or the reference heating process and the change of reference values over the corresponding time. For example, the characteristic value representing the volume change can have a maximum after 10 minutes of 20 minutes and the color change after 15 minutes of 20 minutes of cooking. It is then possible to detect in repeated cooks by means of a classifier such as the aforementioned Support Vector Machine in the repeat cook input sensor data, that the highest probabilities match the reference cook or reference heating program. It may be that, for example, the color change in repeated cooking has a maximum after 5 minutes for the volume change. Therefore, the time difference between repeated cooking and reference cooking would be 50%. This would result in an adaptation of the remaining cooking time by at least 50%. Here, a passage of time of 5 minutes instead of 15 minutes.
[0261] In addition, it may be possible to integrate an impact factor that can influence the impact of the control algorithm with respect to the repeated cooking program. This can be done automatically, in such a way that the number of reference cooks influences the confidence factor, or in such a way that it is manually established in a certain factor. This can also be optimized by means of a remote system using information technology previously described here.
[0262] Additionally, it may be especially possible to change the temperature within this system through a change in a characteristic representing the change in color. As described, it may be possible to calculate the characteristics representing the color change (change in pixel intensity). It is possible to normalize the pixel intensity. After normalization, it is possible to adjust the temperature according to the color change. If, for example, after the remaining 75% of time the expected color change has not occurred, the temperature may be elevated, or if there has been more color change than that expected from the reference cooking, the temperature may be reduced.
[0263] The monitoring device 150 may additionally comprise a control unit 1860 adapted to change a heating process from a cooking process to a cooking process based on a comparison of the current heating processing state determined by means of of the monitoring unit with a predetermined heating process state. The current heating process status is calculated as mentioned above, by determining the "shortest distance" time point. By comparing these time points of the predetermined heating process state and the calculated time point, the heating process is changed, if the calculated time point is later than the heating process state time point. predetermined. For example, as a basic rule, an ordeal will end after a 100% volume change in the food to be heated, so if the hamburger bread or Croissant is twice the volume, the ordeal will end and the procedure cooking will begin. The change in the volume of the bread or food to be baked can be detected by the pixel characteristics in the camera in a very efficient way. The heat treatment machine to be controlled can be an integrated cooking / tasting machine, however, other different tasting or cooking machines can also be controlled.
[0264] To simplify calculations and to ensure repeatable results, it is preferable that the heating temperature is kept constant in a current heating process.
[0265] The 1860 control unit is additionally adapted to stop the heating process based on a comparison of the current heating process state determined via the monitoring unit with a predetermined heating state corresponding to a heating end point . The 1860 control unit can be adapted to alert the user when the heating process has been completed. Therefore, the monitoring apparatus may comprise an 1879 alert unit and an 1880 display unit. The 1880 display unit is provided to indicate the current heating process status, for example, the remaining heating or cooking time. The 1880 display unit can additionally display a pixel image inside the heating chamber for visual monitoring of the food to be heated through a user. The 1960 control unit can be adapted to control the 1880 display unit and is adapted to indicate a remaining heating process time based on a comparison of the current heating process state determined via the monitoring unit with a termination point. and / or display images inside the heating chamber.
[0266] The 1860 control unit is additionally connected to an 1890 output interface to control actuators as described above or below, as a temperature control of a heating chamber, means for adapting the humidity in the heating chamber by means of adding water, or a control of the ventilation mechanism (ventilation shutter). The actuators can additionally include the means for adapting the fan speed, the means for adapting the differential pressure between the heating chamber and the respective environment, the means for establishing a time dependent on the temperature curve inside the treatment chamber steam, the means to perform and to adapt different heat treatment procedures such as tasting and cooking, the means to adapt internal gas flow profiles inside the heating chamber, the means to adapt the intensity of electromagnetic and acoustic emission of the respective electromagnetic and sound emitters to probe or observe the properties of the food to be heated.
[0267] In particular, the 1860 control unit is adapted to control a temperature control of a heating chamber, the means to adapt the humidity in the heating chamber by means of adding water, a control of the ventilation mechanism, the means to adapt the fan speed, the means to adapt the differential pressure between the heating chamber and the respective environment, the means to establish a time dependent on the temperature curve inside the steam treatment chamber, the means to perform and to adapt different heat treatment procedures such as tasting and cooking, the means to adapt internal gas flow profiles inside the heating chamber, the means to adapt the intensity of the electromagnetic and acoustic emission of the respective electromagnetic and sound emitters to probe or observe the properties of the food to be heated.
[0268] A heat treatment monitoring method of the present invention comprises determining data from current sensors of the food being treated; determine current characteristic data from current sensor data; and determining a state of the current heating process in a current heating process of the monitored food by comparing the data of current characteristics with the data of the reference characteristics of a reference heating process. The method preferably further comprises determining a mapping of the current sensor data to the current characteristic data and / or determining the reference characteristic data of a reference heating process based on the characteristic data of at least one heating process. training. In addition, the method comprises determining a mapping of the current sensor data to the current characteristic data through an analysis of variance of at least one preliminary heating process to reduce the dimensionality of the current sensor data. The analysis of variance preferably comprises at least one of: principal component analysis (PCA), Isometric Characteristics Mapping (ISOMAP) or Linear Discriminant Analysis (LDA), or a dimensionality reduction technique. The method additionally preferably comprises determining the reference characteristic data from a reference heating process by combining predetermined characteristic data from a heating program with the characteristic data from a preliminary set of at least one heating process. preliminary heating being classified as part of the preliminary set by a user. In addition, by means of the method of the present invention, the data of current characteristics of a current heating process can be recorded, in which the data of recorded characteristics are used as data of characteristics of a preliminary heating process. In addition, the method can comprise the classification of the type of food to be heated and choose a reference heating process corresponding to the particular type of food. Preferably, a heating process is changed from a testing process to a cooking process based on a comparison of the current heating process state with a predetermined heating process state. The heating temperature is preferably kept constant in a standard heating process. Preferably the heating process is stopped based on a comparison of the current heating process state determined via the monitoring unit with a predetermined heating process state corresponding to the heating end point. In an advantageous embodiment, a user is alerted when the heating process has to be completed.
[0269] According to another realization, the monitoring device 150, the learning machine can be used for a system of multiple inputs and multiple outputs (MIMO). In particular, an adjustment system for added water, remaining cooking time and / or temperature can be implemented through a system for monitoring heat treatment using machine learning techniques.
[0270] The system is collecting all sensor data during reference cooking. In the case of humidity, at least one hygrometer detects a reference value for the humidity over the cooking time during the reference cooking. When repeating a cooking of the same product, the amount of water to be added may be different. The quantity of cooked products may be different, the volume inside the oven may be different, or there may be more or less ice or water on the cooked product when loading the oven.
[0271] Next to other adaptations, the control system according to the invention adds as much water as necessary to achieve similar conditions compared to the reference cooking. Since the remaining cooking time can be adapted through the control system, the time in which the water will be added changes as well. Instead of using a fixed time, such as adding 1 liter of water after 10 minutes of a 20 minute cooking program, according to this realization, the system will add as much water as necessary to reach the humidity level of reference cooking after 50% of the time passed.
[0272] Once irregular behavior is recognized in an implementation of this invention, this signal or irregularity and its corresponding amplitude can be used to adjust processing devices such as mixers (mass-induced energy), mass dividers ( cutting frequency) or industrial ovens (cooking program time or temperature) within a food production process.
[0273] According to another embodiment, the observation of food inside the cooking chamber can be performed “live”, so a live view from the inside of the oven allows remote access to the cooking process. Also, a remote oven setting can be made possible to improve the cooking behavior of a system for monitoring self-learned heat treatment.
[0274] In one realization, "perception," cognition "and" action "(P - C - A), cognitive agents and machine learning techniques suitable for an industrial process with intelligent actuators and sensors can be used. The capacities of cognitive transfer, of knowledge and of skill / expertise, as well as the creation of many P - C - A interaction bonds will be advantageous in a cognitive factory.
[0275] Only very few production processes are unique. Most food production processes take place at different facilities or at different times, performing identical functions in similar environments. Still, there is often no or limited exchange of information between these processes. The same food processing stations often require individual configuration of all entities managing similar process functions. In order to increase the capacity of the auxiliary machine, one to the other, it is advantageous to combine the distributed P - C - A loops in space in time. Certain topics emerge for an approach of this objective: in order to allow the transfer of skills between different entities, it is advantageous to establish a reliable and adaptable topology of Multi-PCA-loops. This system target should be able to identify similar processes, translate data from sensors, acquire characteristics and analyze results from different entities. Dimensionality reduction, grouping and classification techniques can allow machines to communicate at higher levels. The models of machine - machine trust, collective learning and knowledge representation are essential for this purpose. In addition, some industrial processes can be redefined to optimize overall performance in terms of cognitive terms. Both: data processing and hardware configuration should result in a safe, reliable and powerful procedure for sharing information and transferring skills.
[0276] Using self-optimization algorithms to control or to parameterize industrial applications offers the possibility to continuously improve the individual knowledge base. Reinforced learning, for example, provides a set of methods that provide this possibility. These algorithms depend on the exploration in the state - space process in order to learn the optimized state - action combinations. A reinforced learning agent can also be described through a simple P - C - A loop, where the process of assessing the information state of the environment is the "perception" element of the loop, the change in current control laws represents the “action” part and the estimated mapping process from the information state to new control laws provides the “cognition” section of the simple P - C - A loop. In industrial applications exploring a large state - space is not always possible or feasible for various reasons such as safety, speed, or costs. Using a P - C - A Multiple Loop approach for distributing the learning function over multiple agents, it can reduce the amount of exploration for individual agents, while the amount of learning experiences still remains high. In addition, this allows teaching between different P - C - A loops. A possible designation for the multiple P - C - A loops approach is the combination of multiple agents in a system or an assembly line, for example, a closed-loop control unit. Two different agents could be trained to optimize different processing parameters. The combination of both at a Multi - P - C - A level could be used to find a route for all parameters.
[0277] Amos the indicated multi - P - C - A loops can improve and intensify the manufacturing performance with regard to the time of establishment and configuration, processing flexibility, as well as quality. An approach combines and jointly intensifies workstations / benches with a shared knowledge and transfer of skills. The other allows different units to self-improve with the return of the others. Next, a chain network system for cognitive processing devices according to the present invention should be described. It is an advantage of the present invention that once the collaborative system acquires sufficient machine knowledge, they avoid repetitive configuration steps and can significantly reduce downtime as well as increase productivity flexibility.
[0278] In accordance with an embodiment of the present invention, in order to facilitate the integration of several heat treatment monitoring systems 100, all distributed systems are connected to each other via the Internet. The knowledge acquired through these systems is shared, therefore, allowing a global database of processing configurations, sensor adjustments and the quality of work benches.
[0279] In order to share information between machines, all machines have to use a similar method of acquiring characteristics, as a first scenario to achieve this goal using cognitive data processing approaches to combine the input data, from multiple sensors of the respective sensor units 1810, from the system for monitoring 100, in order to receive a good estimate of the current processing status.
[0280] Using cognitive dimensionality reduction techniques, unnecessary and redundant data from these sensors can be removed. The reduced sensor data is used to classify the process status. The grouping allows the identification of specific processing states, even between different settings. If a significant difference from the references and, therefore, an unknown process condition is detected, the supervisor will be alerted. The specialist can then teach the new state and countermeasures (if possible) to the system in order to improve its performance.
[0281] The cognitive system to be developed should be able to learn from separate, acceptable and unacceptable results and additionally, be able to avoid unacceptable results whenever possible. The use of technical cognition eliminates the need for a complete physical model of the cooking and food production process. The system is able to stabilize the process by improving at least one directional variable. Distributed cognition allows for a central database between different manufacturing locations. Information collected from one process can be transferred to a similar process at a different location.

权利要求:
Claims (14)
[0001]
1. System for monitoring a heat treatment applied to a food, characterized by the fact that it comprises: a detection unit (1810) having at least one sensor (1812) to determine current sensor data of a food being heated; a processing unit (1820) for mapping the current sensor data linearly or nonlinearly being fed by the detection unit (1810) as a mapping of data of current characteristics with reduced dimensions, in which a learning unit (1840) creates the mapping by means of an analysis of variance of at least one preliminary warm-up process; wherein the sensor data from the at least one preliminary heating process with the greatest variation over time in the at least one preliminary heating process has the greatest weight; and wherein the mapping is stored in memory on a monitoring device (150) or received from an external server; and a monitoring unit (1830) adapted to determine a state of the current heating process in a current heating process of the monitored food by comparing the current characteristic data with the reference characteristic data of a reference heating process.
[0002]
2. System according to claim 1, characterized by the fact that the analysis of variance comprises at least one of a principal component analysis (PCA), a mapping of isometric characteristics (ISOMAP) or an analysis of linear discriminants (LDA) or a dimensionality reduction technique.
[0003]
3. System according to claim 1 or 2, characterized by the fact that it additionally comprises determining, by the learning unit (1840), reference characteristic data from a reference heating process by combining predetermined characteristic data from a heating program. heating with a data set of characteristics of a preliminary heating process classified as part of a user's preferred preliminary heating set.
[0004]
System according to claim 2 or 3, characterized in that it additionally comprises a recording unit (1822) for recording the current characteristics data of a current heating process and using the recorded characteristics data of the recording unit (1822) as data of characteristics of a preliminary heating process.
[0005]
System according to any one of the preceding claims, characterized by the fact that the detection unit (1810) comprises a camera (160) for recording, in pixels, an image of the food being heated, in which the current sensor data of the camera correspond to the current pixel data of a current pixel image.
[0006]
System according to claim 5, characterized in that the current pixel data comprises first pixel data corresponding to a first color, second pixel data corresponding to a second color and third pixel data corresponding to a third color.
[0007]
7. System according to claim 6, characterized by the fact that the first, second and third colors correspond to R (red, red), G (green, green) and B (blue, blue) respectively.
[0008]
System according to any one of claims 5 to 7, characterized by the fact that the camera (160) is adapted to generate pixel images processed by high definition resolution (High Resolution Definition = HDR) as current pixel data.
[0009]
9. System according to any one of the preceding claims, characterized by the fact that it additionally comprises a classification unit (1850) for classifying the type of food to be heated and choosing a reference heating process corresponding to the type of food determined.
[0010]
System according to any one of the preceding claims, characterized by the fact that it additionally comprises a control unit (1860) for changing a heating process from a preliminary heating process to a cooking process by comparing a state of the heating process current determined by the monitoring unit (1830) with a predetermined heating process state.
[0011]
11. System according to any one of the preceding claims, characterized by the fact that it additionally comprises a control unit (1860) to control a display unit (1880) configured to indicate a time remaining in the heating process based on a comparison of the state of the current heating process determined by the monitoring unit (1830) with a predetermined heating process state corresponding to an end point of the heating and / or to display images of the interior of the heating chamber.
[0012]
12. System according to any of the preceding claims, characterized by the fact that the control unit (1860) additionally alerts a user when the heating process has to be interrupted.
[0013]
13. System according to any one of the preceding claims, characterized by the fact that the control unit (1860) additionally controls a temperature control of the heating chamber, means for adapting the humidity in the heating chamber by adding water or steam, a control of the ventilation mechanism, means for adapting the speed of fans, means for adapting the differential pressure between the heating chamber and the respective environment, means for defining a time-dependent temperature curve inside the heating chamber, means for carrying out and adapt different heating procedures, such as fermentation or cooking, means to adapt internal gas flow profiles inside the heating chamber, means to adapt the intensity of electromagnetic and acoustic emissions of the respective electromagnetic or acoustic emitters to probe or observe the properties of the food to be heated.
[0014]
System according to any one of the preceding claims, characterized in that the at least one sensor (1812) of the detection unit (1810) comprises at least one of: a hygrometer, an insertion temperature sensor, a temperature sensor temperature of the heating chamber, acoustic sensors, scales, timers, a camera, an image sensor, a set of photodiodes, a gas gas analyzer inside the heating chamber, means for determining temperature profiles of the temperature sensors of insertion, means for determining electromagnetic or acoustic process emissions from food being heated, such as reflected light or sound or emitted in response to emitters or sources of light or sound, means for determining 3D measurement results of the food being heated, including 3D or stereo camera or radar, or means to determine the type or constitution or pattern or optical characteristics or volume or mass of the food being heated gone.

类似技术:

---

公开号 | 公开日 | 专利标题

---

BR112015012978B1|2021-03-30|SYSTEM FOR MONITORING A HEAT TREATMENT

---

US20200267996A1|2020-08-27|Monitoring System for a Food Processing System

---

JP5136965B2|2013-02-06|Image processing apparatus, image processing method, and image processing program

---

CN108027881A|2018-05-11|Use the user authentication of a variety of capturing technologies

---

US20170135177A1|2017-05-11|Occupancy sensing smart lighting system

---

JP2021522591A|2021-08-30|How to distinguish a 3D real object from a 2D spoof of a real object

---

US20200253415A1|2020-08-13|Monitoring system and food preparation system

---

Chen et al.2019|Visual modeling of laser-induced dough browning

---

US20210068582A1|2021-03-11|Automatic cooking device and method

---

CN113569726A|2021-10-29|Pedestrian detection method combining automatic data amplification and loss function search

---

Ponnaiah et al.2013|GA based automatic optic disc detection from fundus image using blue channel and green channel information

同族专利:

---

公开号 | 公开日

---

PT2928305T|2019-03-14|

---

WO2014086487A1|2014-06-12|

---

EP3521705B1|2021-09-01|

---

US20150330640A1|2015-11-19|

---

EP2928305A2|2015-10-14|

---

WO2014086486A3|2014-09-12|

---

EP2929252B1|2018-10-24|

---

JP6525884B2|2019-06-05|

---

WO2014086486A2|2014-06-12|

---

EP2928305B1|2018-11-28|

---

US20210259256A1|2021-08-26|

---

CA2893601A1|2014-12-12|

---

EP2929252A1|2015-10-14|

---

MX2019008045A|2019-10-14|

---

CN105142408B|2019-06-11|

---

EP3521705A1|2019-08-07|

---

JP2016502061A|2016-01-21|

---

AU2013354500A1|2015-07-16|

---

MX366270B|2019-07-04|

---

AU2013354500B2|2018-02-15|

---

RU2653733C2|2018-05-14|

---

US11013237B2|2021-05-25|

---

MX2015006979A|2016-02-18|

---

RU2015124383A|2017-01-12|

---

KR102099726B1|2020-04-13|

---

CN110235906A|2019-09-17|

---

ES2713984T3|2019-05-24|

---

US20200178543A1|2020-06-11|

---

CN105142408A|2015-12-09|

---

KR20150130262A|2015-11-23|

---

JP2019124464A|2019-07-25|

---

US20150366219A1|2015-12-24|

---

JP2021139619A|2021-09-16|

引用文献:

---

公开号 | 申请日 | 公开日 | 申请人 | 专利标题

---

---

US1774433A|1927-03-24|1930-08-26|Gen Electric|Regulation of the color of cooked products|

---

NL6405384A|1963-06-04|1964-12-07|

---

GB1058421A|1964-03-02|1967-02-08|Baker Perkins Ltd|Oven control systems|

---

US3536405A|1968-06-12|1970-10-27|Singer General Precision|Optical thickness gauge|

---

US3507267A|1968-07-12|1970-04-21|Gen Electric Canada|Windowed oven door|

---

US3895870A|1971-06-01|1975-07-22|Autech Corp|Laser dimension comparator|

---

US4048978A|1972-03-02|1977-09-20|Glaverbel-Mecaniver|Heat insulating screen|

---

US3828763A|1973-05-14|1974-08-13|Gen Electric|Oven door construction with front glass panel|

---

US4084571A|1976-09-20|1978-04-18|The Tappan Company|See-through oven door with reticulated heat shield|

---

FR2599991B1|1986-06-16|1993-04-02|Bp Chimie Sa|EVACUATION OF PRODUCTS PRESENT IN AN ALPHA-OLEFIN POLYMERIZATION REACTOR IN A FLUIDIZED BED|

---

US4744407A|1986-10-20|1988-05-17|Inductotherm Corp.|Apparatus and method for controlling the pour of molten metal into molds|

---

US5726423A|1988-05-19|1998-03-10|Quadlux, Inc.|Apparatus and method for regulating cooking time in a radiant energy oven|

---

JPH02157526A|1988-12-09|1990-06-18|Matsushita Electric Ind Co Ltd|Utensil of cooking by automatic heating|

---

JPH0420311A|1990-04-26|1992-01-23|Toshiba Corp|Kneading device|

---

US5157254A|1991-02-19|1992-10-20|Anderson John D|Reflective electro-luminescent measurement adapter for quality control of baked and fried goods|

---

DE4206820A1|1991-08-28|1993-03-04|Bosch Siemens Hausgeraete|OVEN DOOR|

---

IT1258067B|1992-04-02|1996-02-20|Zeltron Spa|AUTOMATIC CONTROL COOKING SYSTEM|

---

GB2293027A|1994-09-07|1996-03-13|Sharp Kk|Apparatus for and method of controlling a microwave oven|

---

US5786568A|1995-02-16|1998-07-28|Leonhard May|Programmable oven for cooking holding and proofing comestibles|

---

US6080972A|1995-02-16|2000-06-27|May; Leonhard|Remotely operated universal programmable oven controller|

---

DE19831635A1|1998-07-15|2000-01-20|Thomas Atzinger|Detector controlling rising of leavened dough in proofing chamber, uses ultrasonic proximity sensor or split laser beams to determine and indicate when dough rises to predetermined height|

---

JP2000275409A|1999-01-18|2000-10-06|Sumitomo Metal Mining Co Ltd|Low-transmission transparent base material, its manufacture, and display device employing the same|

---

US6559882B1|1999-09-02|2003-05-06|Ncr Corporation|Domestic appliance|

---

FR2800998B1|1999-11-17|2002-04-26|Saint Gobain Vitrage|TRANSPARENT SUBSTRATE HAVING AN ANTI-REFLECTIVE COATING|

---

JP2001330257A|2000-05-25|2001-11-30|Matsushita Electric Ind Co Ltd|Control device|

---

ES2625808T3|2002-04-25|2017-07-20|Vitro, S.A.B. De C.V.|Articles coated with a protective coating|

---

US7033679B2|2001-01-25|2006-04-25|Kyocera Optec Co., Ltd.|Metal film and metal film-coated member, metal oxide film and metal oxide film-coated member, thin film forming apparatus and thin film forming method for producing metal film and metal oxide film|

---

RU2002113288A|2001-05-21|2003-11-10|Консорцио Аграрио Провинчале Ди Пьяченца|System for operational control of the "HISTORY OF EXISTENCE" of the food product at the time of purchase|

---

AUPR675801A0|2001-08-01|2001-08-23|Lawson, Anthony Charles|Improved browning control|

---

DE10141453A1|2001-08-23|2003-03-13|Schott Glas|Glazing for heating and cooling devices|

---

US6638553B2|2001-10-09|2003-10-28|Recot, Inc.|System and method for monolayer alignment snack chip transfer|

---

US7265900B2|2002-09-30|2007-09-04|Applied Materials, Inc.|Inspection system with oblique viewing angle|

---

US7660440B2|2002-11-07|2010-02-09|Frito-Lay North America, Inc.|Method for on-line machine vision measurement, monitoring and control of organoleptic properties of products for on-line manufacturing processes|

---

DE10300465A1|2003-01-09|2004-07-29|Rational Ag|Cooking using a cluster analysis and cooking devices for this|

---

DE10318860A1|2003-04-25|2004-11-11|BSH Bosch und Siemens Hausgeräte GmbH|Cooking appliance|

---

CN100528784C|2003-05-30|2009-08-19|Ppg工业俄亥俄公司|Appliance with coated transparency|

---

US7998602B2|2003-05-30|2011-08-16|Ppg Industries Ohio, Inc.|Appliance with coated transparency|

---

CN100482346C|2003-06-09|2009-04-29|日本板硝子株式会社|Photocatalyst member|

---

US7517583B2|2003-09-17|2009-04-14|Central Glass Company, Limited|Laminated glass|

---

DE102004003406A1|2004-01-23|2005-08-18|Electrolux Schwanden Ag|A method for constructing a door for microwave ovens has an inner frame forming a quarter wavelength pocket around the opening and a smaller gap to the housing|

---

WO2006101531A1|2005-03-23|2006-09-28|Middleby Corporation|Conveyor oven apparatus and method|

---

EP1595453A1|2004-05-10|2005-11-16|SMEG S.p.A.|Automatic control method of baking food products in an oven, and automatically controlled oven|

---

US20060081135A1|2004-08-16|2006-04-20|Britton Douglas F|Industrial overline imaging system and method|

---

JP2006149593A|2004-11-29|2006-06-15|Matsushita Electric Ind Co Ltd|Rice cooker with bread making function|

---

US10687391B2|2004-12-03|2020-06-16|Pressco Ip Llc|Method and system for digital narrowband, wavelength specific cooking, curing, food preparation, and processing|

---

DE102005030483B4|2005-06-28|2008-12-04|Werner & Pfleiderer Lebensmitteltechnik Sachsen Gmbh|shop oven|

---

KR101232612B1|2005-07-20|2013-02-13|삼성전자주식회사|Cooking apparatus, cooking system and cooking control method thereof|

---

AT414243T|2005-12-16|2008-11-15|Convotherm Elektrogeraete|Commercial appliance and method for its operation|

---

GB2439324B|2006-06-22|2011-12-28|Panasonic Mfg Uk Ltd|Domestic appliance with controlled door opening|

---

WO2008016309A1|2006-08-04|2008-02-07|Sinvent As|Multi-modal machine-vision quality inspection of food products|

---

EP2139341B1|2007-03-10|2016-11-16|TurboChef Technologies, Inc.|Compact conveyor oven|

---

US20090017173A1|2007-07-13|2009-01-15|Kozman Austin J|Automatic process control of conveyorized ovens for improved baking efficiency and product uniformity|

---

US9670369B2|2007-09-05|2017-06-06|Kabushiki Kaisha Toshiba|Visible-light-responsive photocatalyst powder, and visible-light-responsive photocatalytic material, photocatalytic coating material and photocatalytic product each using the same|

---

US8215137B2|2007-11-21|2012-07-10|Electrolux Home Products, Inc.|Laundry appliance over-molded metal porthole door frame|

---

US20090142458A1|2007-11-30|2009-06-04|Mccann Kevin|Portable electric cooler|

---

GB0725094D0|2007-12-21|2008-01-30|Univ Liverpool|Image processing|

---

JP2009175141A|2007-12-27|2009-08-06|Ono Pharmaceut Co Ltd|Method for screening compound|

---

DE102007063041A1|2007-12-28|2009-07-02|Fraunhofer-Gesellschaft zur Förderung der angewandten Forschung e.V.|Laser light section arrangement for determining e.g. elevation profile of object, has image processing device for identifying and separating laser sectional lines from each other in recorded image, and assigning lines to respective lasers|

---

DE102008009660A1|2008-02-18|2009-08-27|Rational Ag|Cooking appliance for cooking e.g. meat, has sensor for determining condition of product to be cooked, where parameter calculated by central control unit/processing unit correlates with condition of product to be cooked|

---

US8891955B2|2008-04-04|2014-11-18|Whitham Holdings, Llc|Digital camera with high dynamic range mode of operation|

---

EP2149755B1|2008-07-30|2012-12-05|Electrolux Home Products Corporation N.V.|Oven and method of operating the same|

---

US8158175B2|2008-08-28|2012-04-17|Frito-Lay North America, Inc.|Method for real time measurement of acrylamide in a food product|

---

JP2010127545A|2008-11-28|2010-06-10|Hitachi Appliances Inc|Cooking device|

---

US20120060822A1|2009-05-14|2012-03-15|Schott Gemtron Corporation|Full-view oven doors having locking mechanisms|

---

EP2494307B1|2009-10-27|2018-12-05|Formax, Inc.|Automated product profiling apparatus and product slicing system using the same|

---

MX2009014046A|2009-12-18|2011-06-21|Mabe Sa De Cv|Illumination system for cavities.|

---

JP2011149606A|2010-01-21|2011-08-04|Panasonic Corp|Heating apparatus|

---

EP2362150B1|2010-02-26|2017-05-10|Electrolux Home Products Corporation N.V.|An oven door for a domestic cooking oven|

---

DE102010016651A1|2010-04-27|2011-10-27|Rational Ag|Method for determination of degree of browning of cooked food of cooking appliance, and for determining contamination degree of cooking chamber, involves calibrating gas sensor by recording of measured values of pure substances|

---

JP2011251028A|2010-06-03|2011-12-15|Panasonic Corp|Bread maker|

---

JP5857503B2|2010-07-29|2016-02-10|小野薬品工業株式会社|Biomarkers for stress-related diseases|

---

EP2638325B1|2010-11-12|2015-03-11|Convotherm Elektrogeräte GmbH|A cooking device, and a procedure for cooking food|

---

US8712739B2|2010-11-19|2014-04-29|General Electric Company|System and method for hybrid risk modeling of turbomachinery|

---

GB2486165A|2010-11-30|2012-06-13|St Microelectronics Res & Dev|Oven using a Single Photon Avalanche Diode array|

---

JP5762756B2|2011-01-20|2015-08-12|オリンパス株式会社|Image processing apparatus, image processing method, image processing program, and photographing apparatus|

---

DE202011002570U1|2011-01-31|2012-05-07|Wiesheu Gmbh|Device for the automatic heat treatment of food|

---

JP5791336B2|2011-04-01|2015-10-07|キヤノン株式会社|Image processing apparatus and control method thereof|

---

CN102141260B|2011-04-20|2013-03-20|申家群|Internal recycle radiating type nonradiative microwave oven|

---

DE102011002187A1|2011-04-20|2012-10-25|Miele & Cie. Kg|household appliance|

---

EP2520169B1|2011-04-29|2019-12-04|Electrolux Home Products Corporation N.V.|Baking oven door and baking oven|

---

US20130277353A1|2012-04-23|2013-10-24|Dacor, Inc.|Android controlled oven|

---

KR101476500B1|2012-05-10|2014-12-24|엘지전자 주식회사|Composition for enamel, preparation method of composition for enamel, and cooking appliance|EP2520169B1|2011-04-29|2019-12-04|Electrolux Home Products Corporation N.V.|Baking oven door and baking oven|

---

DE102013114227A1|2013-12-17|2015-06-18|Rational Aktiengesellschaft|Cooking appliance with charge detection|

---

FI125053B|2014-01-10|2015-05-15|Innohome Oy|Procedure and apparatus for monitoring safe use of stove|

---

DK178416B1|2014-07-25|2016-02-15|Spx Flow Technology Danmark As|Infusion plant|

---

DE102014110559A1|2014-07-25|2016-01-28|Rational Aktiengesellschaft|Method for controlling a cooking appliance|

---

US10517313B2|2014-08-05|2019-12-31|Koninklijke Philips N.V.|Coffee roasting apparatus, coffee brewing apparatus and coffee roasting method|

---

EP3183503B1|2014-08-18|2019-05-01|Stork genannt Wersborg, Ingo|Heat treatment system|

---

EP3194853B1|2014-09-03|2018-07-11|Electrolux Appliances Aktiebolag|Domestic appliance, in particular cooking oven, with a camera|

---

EP3256786B1|2015-02-10|2020-04-08|Electrolux Appliances Aktiebolag|Oven door and oven comprising an oven door|

---

US10555633B2|2015-03-27|2020-02-11|Prashant Choudhary|Autonomous cooking device to prepare food from a recipe file and method for creating recipe files|

---

JP6464013B2|2015-04-01|2019-02-06|富士フイルム株式会社|Imaging apparatus and method|

---

US9644847B2|2015-05-05|2017-05-09|June Life, Inc.|Connected food preparation system and method of use|

---

US9927129B2|2015-06-01|2018-03-27|June Life, Inc.|Thermal management system and method for a connected oven|

---

US10739013B2|2015-05-05|2020-08-11|June Life, Inc.|Tailored food preparation with an oven|

---

US9615007B2|2015-06-12|2017-04-04|Haier Us Appliance Solutions, Inc.|Oven camera assembly|

---

KR102362654B1|2015-07-03|2022-02-15|삼성전자주식회사|Oven|

---

US11156366B2|2015-09-10|2021-10-26|Brava Home, Inc.|Dynamic heat adjustment of a spectral power distribution configurable cooking instrument|

---

US20180324908A1|2015-09-10|2018-11-08|Brava Home, Inc.|In-oven camera and computer vision systems and methods|

---

JP2018534514A|2015-09-10|2018-11-22|ブラバ・ホーム・インコーポレイテッド|Camera in oven|

---

US10064244B2|2015-09-10|2018-08-28|Brava Home, Inc.|Variable peak wavelength cooking instrument with support tray|

---

USD787041S1|2015-09-17|2017-05-16|Whirlpool Corporation|Gas burner|

---

US10837651B2|2015-09-24|2020-11-17|Whirlpool Corporation|Oven cavity connector for operating power accessory trays for cooking appliance|

---

DE102015117310A1|2015-10-12|2017-04-13|Rational Aktiengesellschaft|Method for controlling an interior lighting of a cooking appliance and cooking appliance|

---

US10785363B2|2015-10-23|2020-09-22|Traeger Pellet Grills, Llc|Cloud system for controlling outdoor grill with mobile application|

---

US10791208B2|2015-10-23|2020-09-29|Traeger Pellet Grills, Llc|Mobile application for controlling outdoor grill|

---

US10701199B2|2015-10-23|2020-06-30|Traeger Pellet Grills, Llc|Cloud system for controlling outdoor grill with mobile application|

---

US10455022B2|2015-10-23|2019-10-22|Traeger Pellet Grills, Llc|Cloud system for controlling outdoor grill with mobile application|

---

US10757244B2|2015-10-23|2020-08-25|Traeger Pellet Grills, Llc|Cloud system for controlling outdoor grill with mobile application|

---

KR20180074615A|2015-10-23|2018-07-03|트래거 펠레트 그릴스, 엘엘씨|Cloud system for controlling outdoor grill with mobile application|

---

US10708409B2|2015-10-23|2020-07-07|Traeger Pellet Grills, Llc|Mobile application for controlling outdoor grill|

---

US10735575B2|2015-10-23|2020-08-04|Traeger Pellet Grills, Llc|Mobile application for controlling outdoor grill|

---

US10491738B2|2015-10-23|2019-11-26|Traeger Pellet Grills, Llc|Cloud system for controlling outdoor grill with mobile application|

---

GB2558089B|2015-10-23|2022-03-02|Traeger Pellet Grills Llc|Mobile application for controlling outdoor grill|

---

US20170130967A1|2015-11-05|2017-05-11|General Electric Company|Oven Appliance|

---

JP6830184B2|2015-11-27|2021-02-17|パナソニックＩｐマネジメント株式会社|Cooker, control method of cooker, and cooker system|

---

US20170188741A1|2015-12-31|2017-07-06|C. Douglass Thomas|Method and System for Acquiring Cooking Information|

---

CN109073231A|2016-03-29|2018-12-21|松下知识产权经营株式会社|Heating device|

---

CN108885009B|2016-03-29|2020-06-30|松下知识产权经营株式会社|Heating cooker|

---

JP6554439B2|2016-04-18|2019-07-31|日立グローバルライフソリューションズ株式会社|High frequency heating cooker|

---

CN105996796B|2016-05-30|2020-04-17|北京小焙科技有限公司|Oven and camera thereof|

---

CN105974832B|2016-06-20|2018-09-04|无锡虹业自动化工程有限公司|Carry the SVC device of robot manipulation|

---

US10817778B2|2016-06-27|2020-10-27|International Business Machines Corporation|Customized cooking utilizing deep learning neuromorphic computing of hyperspectral input|

---

US10145568B2|2016-06-27|2018-12-04|Whirlpool Corporation|High efficiency high power inner flame burner|

---

WO2018044067A1|2016-09-01|2018-03-08|Samsung Electronics Co., Ltd.|Oven|

---

US11205103B2|2016-12-09|2021-12-21|The Research Foundation for the State University|Semisupervised autoencoder for sentiment analysis|

---

EP3554322A4|2016-12-15|2020-06-03|Breville Pty Limited|Toaster sensing device|

---

DE102017202773A1|2017-02-21|2018-08-23|BSH Hausgeräte GmbH|Cooking appliance with a removably designed sensor module|

---

DE102017202778A1|2017-02-21|2018-08-23|BSH Hausgeräte GmbH|Cooking appliance with a receptacle for a removably designed sensor module|

---

US10551056B2|2017-02-23|2020-02-04|Whirlpool Corporation|Burner base|

---

US10451290B2|2017-03-07|2019-10-22|Whirlpool Corporation|Forced convection steam assembly|

---

US10660162B2|2017-03-16|2020-05-19|Whirlpool Corporation|Power delivery system for an induction cooktop with multi-output inverters|

---

EP3604931A4|2017-03-23|2020-03-18|Panasonic Intellectual Property Management Co., Ltd.|Heating cooking device|

---

US10548398B2|2017-05-18|2020-02-04|Edible Education Llc|Wheeled kitchen apparatus with collapsible work surface|

---

US10223933B1|2017-08-09|2019-03-05|Brava Home, Inc.|Multizone cooking utilizing a spectral-configurable cooking instrument|

---

WO2019038011A1|2017-08-25|2019-02-28|Arcelik Anonim Sirketi|A cooking device comprising a cooling system|

---

EP3685633A1|2017-09-18|2020-07-29|InterProducTec Consulting GmbH & Co. KG|Monitoring system and food preparation system|

---

CN107752794B|2017-09-27|2020-10-09|珠海格力电器股份有限公司|Baking method and device|

---

JP6921709B2|2017-10-24|2021-08-18|リンナイ株式会社|Cooker|

---

WO2019092249A1|2017-11-13|2019-05-16|InterProducTec Consulting GmbH & Co. KG|Monitoring system and heat treatment system comprising the same|

---

US11206949B1|2017-11-15|2021-12-28|Brava Home, Inc.|High power density toaster|

---

KR102069675B1|2017-12-13|2020-01-28|동국대학교 산학협력단|Apparatus for predicting roasting completion time and operating method thereof|

---

KR102329592B1|2017-12-30|2021-11-19|미디어 그룹 코 엘티디|Food preparation methods and systems based on ingredient recognition|

---

US11116050B1|2018-02-08|2021-09-07|June Life, Inc.|High heat in-situ camera systems and operation methods|

---

CN108521691A|2018-03-19|2018-09-11|上海点为智能科技有限责任公司|Radio frequency defrosting heating equipment|

---

JP6429138B1|2018-04-10|2018-11-28|コネクテッドロボティクス株式会社|Food manufacturing apparatus, food manufacturing system and control program|

---

RU2694625C1|2018-04-12|2019-07-16|Владимир Ильич Маклюков|Device for studying and research of bakery products baking process|

---

US10440245B1|2018-05-02|2019-10-08|Haier Us Appliance Solutions, Inc.|Oven appliance camera assembly including a light shield|

---

US10627116B2|2018-06-26|2020-04-21|Whirlpool Corporation|Ventilation system for cooking appliance|

---

US10619862B2|2018-06-28|2020-04-14|Whirlpool Corporation|Frontal cooling towers for a ventilation system of a cooking appliance|

---

US10837652B2|2018-07-18|2020-11-17|Whirlpool Corporation|Appliance secondary door|

---

WO2020055370A2|2018-09-13|2020-03-19|Cms Jant Ve Maki̇na Sanayi̇i̇ Anoni̇m Şi̇rketi̇|Fault detection system for a heat treatment process|

---

CN110941203A|2018-09-25|2020-03-31|珠海格力电器股份有限公司|Control method and device for cooking food and cooker|

---

US10502430B1|2018-10-10|2019-12-10|Brava Home, Inc.|Particulates detection in a cooking instrument|

---

EP3864346A1|2018-10-10|2021-08-18|BSH Hausgeräte GmbH|Method for preparing a cooking product, cooking device, and cooking device system|

---

CN109287021B|2018-10-15|2021-01-12|南京航空航天大学|Online learning-based intelligent monitoring method for microwave heating temperature field|

---

USD920995S1|2018-11-07|2021-06-01|Samsung Electronics Co., Ltd.|Display screen or portion thereof with graphical user interface|

---

CN109717722B|2018-11-14|2021-01-26|中画高新技术产业发展有限公司|Multifunctional monitoring electric cooker|

---

US20220072595A1|2018-12-10|2022-03-10|Net Boru Sanayi Ve Dis Ticaret Kollektif Sirketi Bora Saman Ve Ortag|Heating system for production of a double-layer tube|

---

KR102183170B1|2018-12-13|2020-11-25|정태우|Alcohol fermentation system with complex neural network algorithm|

---

KR102177053B1|2018-12-27|2020-11-10|재단법인대구경북과학기술원|Roasting Levels Predicted Method and Roaster Machine For Performing The Method|

---

US11047578B2|2019-01-04|2021-06-29|Whirlpool Corporation|Automatic oven|

---

WO2020171087A1|2019-02-18|2020-08-27|コネクテッドロボティクス株式会社|Food cooking device and control program|

---

US11009238B2|2019-03-01|2021-05-18|Midea Group Co., Ltd.|Staged oven self-clean preheat temperature control|

---

US10865999B2|2019-03-01|2020-12-15|Midea Group Co., Ltd.|Targeted oven self-clean preheat temperature control|

---

DE102019107812A1|2019-03-27|2020-10-01|Miele & Cie. Kg|Method for operating a cooking appliance and cooking appliance|

---

DE102019204531A1|2019-04-01|2020-10-01|BSH Hausgeräte GmbH|Household appliance and method for determining contour information of goods|

---

US20200353178A1|2019-05-10|2020-11-12|Enthermics Medical Systems, Inc.|Intravenous Fluid Warmer Using Uniform Thermal Field|

---

DE102019120008B4|2019-07-24|2021-04-01|Miele & Cie. Kg|Method for operating a cooking appliance and cooking appliance|

---

CN110631711B|2019-10-25|2020-12-15|北京华通兴远供热节能技术有限公司|Indoor temperature networked analysis method and system|

---

KR20210056173A|2019-11-08|2021-05-18|엘지전자 주식회사|Artificial intelligence cooking device|

---

US11058132B2|2019-11-20|2021-07-13|June Life, Inc.|System and method for estimating foodstuff completion time|

---

RU2726769C1|2019-12-13|2020-07-15|федеральное государственное автономное образовательное учреждение высшего образования «Национальный исследовательский Томский политехнический университет»|System for monitoring parameters of robotic installations|

---

US20210186032A1|2019-12-19|2021-06-24|Whirlpool Corporation|Monitoring system|

---

US10798332B1|2020-02-20|2020-10-06|Varjo Technologies Oy|Dual pass-through imaging system and method|

---

CN111420854A|2020-03-31|2020-07-17|高博半导体有限公司|L ED lamp production oven capable of monitoring temperature|

---

KR102223927B1|2020-08-28|2021-03-05|화신주방산업|Electric heating type pancake cooker|

---

KR102212008B1|2020-12-30|2021-02-04|화신주방산업 주식회사|Electric heating type pancake cooker|

法律状态:
2018-03-06| B06F| Objections, documents and/or translations needed after an examination request according [chapter 6.6 patent gazette]|

---

2018-03-13| B06F| Objections, documents and/or translations needed after an examination request according [chapter 6.6 patent gazette]|

---

2018-03-20| B06I| Publication of requirement cancelled [chapter 6.9 patent gazette]|Free format text: ANULADA A PUBLICACAO CODIGO 6.6.1 NA RPI NO 2462 DE 13/03/2018 POR TER SIDO INDEVIDA. |

---

2019-08-13| B06U| Preliminary requirement: requests with searches performed by other patent offices: procedure suspended [chapter 6.21 patent gazette]|

---

2020-06-30| B07A| Application suspended after technical examination (opinion) [chapter 7.1 patent gazette]|

---

2021-02-09| B09A| Decision: intention to grant [chapter 9.1 patent gazette]|

---

2021-03-30| B16A| Patent or certificate of addition of invention granted [chapter 16.1 patent gazette]|Free format text: PRAZO DE VALIDADE: 20 (VINTE) ANOS CONTADOS A PARTIR DE 04/12/2013, OBSERVADAS AS CONDICOES LEGAIS. |

优先权:

---

申请号 | 申请日 | 专利标题

---

EP12008113.8|2012-12-04|

---

EP12008113|2012-12-04|

---

EP13004786|2013-10-04|

---

EP13004786.3|2013-10-04|

---

PCT/EP2013/003662|WO2014086486A2|2012-12-04|2013-12-04|Heat treatment monitoring system|

---