Non-destructive inspection method and system for in-line inspection of an object; and using a method

专利摘要:
The present invention relates to a non-destructive inspection method (10) for in-line inspection of an object and a related system. The method comprises moving (11) an object between a radiation source and an image detector and through a field of view of the three-dimensional digitizer, imaging (12) the object using the image detector to obtain a projection radiograph of an internal structure of the object, scan (13) an external surface of the object using the 3D digitizer, fit (14) an object shape model to a point cloud provided by the 3D digitizer to obtain a surface model of the external surface , create (15) a solid model of the surface model considering a gray value distribution of a reference object, simulate (16) a reference radiograph from the solid model and compare (17) the reference radiographs and projection to detect internal deviations of the object.
公开号:BR112017019834B1
申请号:R112017019834-7
申请日:2016-03-16
公开日:2022-02-01
发明作者:Mattias VAN DAEL；Pieter Verboven；Bart Nicolaï；Jelle DHAENE；Luc Van Hoorebeke；Jan Sijbers
申请人:Katholieke Universiteit Leuven；Universiteit Antwerpen；
IPC主号:

专利说明:

FIELD OF THE INVENTION
[001] The invention relates to the field of non-destructive testing of a product or produce, for example, by using ionizing radiation imaging. More specifically, it refers to a method and system for in-line product inspection, for example for quality control and/or automated selection of an object, for example a product or produce such as a vegetable or fruit. , which is at least partially transparent to ionizing radiation, for example an object transparent to X-rays. The invention further relates to the use of such a method and/or system for quality control and/or selection of a quality object. BACKGROUND OF THE INVENTION
[002] Line detection of internal defects in produce or produce, eg food items, with the use of X-rays is known in the art to detect defects that are easily discernible on radiographs. In particular, X-ray imaging has become a valuable tool in many industrial sectors to perform non-destructive testing to ensure the quality of a product. Since most materials are X-ray translucent, internal defects can be visualized without opening and damaging the product. For example, the use of two-dimensional X-ray radiographic imaging for non-destructive testing of product quality and/or product defect detection is known in the art, for example for in-line inspection of food products in the food industry. Such a simple 2D radiographic projection, in which all features on the inside and outside of the object are superimposed into a single image, can provide a quick way to visualize the interior of an object in line.
[003] X-ray systems are commercially used for the detection of foreign bodies and, with even more limitation, for the detection of certain internal defects and unwanted properties common in food systems, such as the presence of insects in fruits, for example, as revealed by Haff et al in “X-ray detection of defects and contaminants in the food industry”, Sensing and Instrumentation for Food Quality and Safety, 2(4), pages 262 to 273, and by Jiang et al. in “An adaptive image segmentation algorithm for X-ray quarantine inspection of selected fruits”, Computers and Electronics in Agriculture, 60(2), pages 190 to 200. Also known in the art is the use of X-ray imaging for the detection of automatic fishbone, for example, as revealed by Mery et al. in “Automated fish bone detection using X-ray imaging”, Journal of Food Engineering, 105(3), pages 485 to 492.
[004] This approach, as known in the art, may, however, have several disadvantages. For example, density differences need to be large enough that defects and/or undesirable properties or objects are visible on projection radiographs. This implies that this approach may not be usable in particular applications. Additionally, custom algorithms may need to be developed for each type of defect or undesirable property that must be detected. This can prove to be very time consuming, certainly considering that when imaged at different hardware settings, the appearance of these defects can differ substantially.
[005] To detect subtle features, a full three-dimensional CT reconstruction of the object may be necessary, as particular internal defects cannot be discerned in projection images captured from a single point of view, or even by simultaneously evaluating a plurality of images corresponding to a plurality of complementary projection views, for example, images corresponding to two or more projection views along mutually orthogonal projection geometric axes. For example, in the food industry, some defects, such as fruit rot disorders, inherently show low contrast to their surroundings and can be very small.
[006] CT imaging methods imply that projections are taken from many angular positions around the sample, by rotating the source-detector pair, for example, in an arrangement commonly used for medical digitizers, or by rotating the object sample during object imaging, as may be known by industrial definitions. This approach can have several implications when applied in a line inspection system to inspect an object transported by a line conveyor system. For example, rotating the source-detector pair around a conveyor belt is impractical because of the high speeds that would be required to maintain an acceptable throughput conveyor belt speed. A high-speed swivel gantry would require very expensive hardware, cause massive forces, entail additional security restrictions, and make the hardware large and bulky. Additionally, rotating the sample object by an angular range large enough for CT imaging, while moving on a conveyor belt, may also be undesirable due to the fact that rotation would also require high speed and precise control, which is practically hard to reach. Even though these problems can be prevented, an image processing algorithm may need to be developed for each type of defect or undesirable property that must be detected.
[007] Due to cost, time and hardware constraints, a complete 3D tomographic reconstruction is therefore difficult to achieve, or even impractical, in an in-line application, for example in an in-line sorting system for sorting objects. X-ray transparent objects that move in a stream of objects, for example produce or produce, such as a vegetable or fruit, that move on a conveyor belt or similar transport system. Furthermore, the complexity of 3D CT imaging techniques, as known in the art, can have the disadvantages of being expensive and complex and can substantially compromise desired production line throughput while providing sufficient image quality to ensure detectability of acceptable defect. For example, the trade-off between high acquisition speed and high contrast and resolution imaging may be one of the main reasons 3D X-ray CT has not yet established itself as an inspection tool in the food industry. In other industrial sectors, the use of in-line or on-line CT may be known, for example as a metrology tool, for example with the use of high throughput batch delivery systems or a throughput conveyor belt system continuous using a helical scanning approach. Regardless, such approaches remain expensive and complex.
[008] Attempts were made to prevent the above mentioned problems. For example, Rapiscan Systems developed a CT line scanner for baggage inspection by combining a large number of source-detector pairs into one definition, eg Rapiscan RTT™ 110 from Rapiscan Systems, Torrance, CA 90503 USA. This functional but expensive solution can reach speeds of 1,500 to 1,800 bags per hour, corresponding to a throughput speed of around 0.5 m/s, which may not be fast enough for a low-value, low-volume application. high, such as the food industry.
[009] In “Automated knot detection for high speed computed tomography on Pinus sylvestris L. and Picea abies (L.) Karst. using ellipse fitting in concentric surfaces” Computers and Electronics in Agriculture, 96 (2013), pages 238 to 245, by Johansson et al., a method has been revealed that combines three-dimensional scanning and X-ray radiographs. method is that the data processing proposed in this prior art article is limited to the estimation of durame diameter and density in trunks.
[0010] Another approach is to use the translation of an object on the conveyor belt to obtain projections in a limited angular range. However, a three-dimensional reconstruction from projection data in a limited angular range is not straightforward and can introduce large image artifacts. Research on this subject was reported, for example, by lovea et al. in “Pure Translational Tomography - a Non-Rotational Approach for Tomographic Reconstruction”, Proceedings of the 9th European Conference on NDT - ECNDT, Tu.1.4.1. BRIEF DESCRIPTION OF THE INVENTION
[0011] It is an object of embodiments of the present invention to provide satisfactory and effective means and methods for inspecting, sorting and/or sorting a moving object, for example, an article of produce or produce that moves along a predetermined path. on a production, processing or packaging line.
[0012] The above objective is fulfilled by a method and a device according to the present invention.
[0013] It is an advantage of embodiments of the present invention that a quality object can be controlled and/or selected based on model-based classification of internal defects or deficiencies of that object.
[0014] It is an advantage of embodiments of the present invention that a quality of an object can be classified, for example, the quality of produce or produce, by combining 3D scanning and X-ray radiographs.
[0015] It is an advantage of embodiments of the present invention that a method or system according to embodiments of the present invention can be applied to identify anomalies in a wide variety of objects by applying a model of the shape of the object of interest as knowledge previous.
[0016] It is an additional advantage of embodiments of the present invention that it is not necessary to develop custom algorithms specifically adapted for each type of defect to be detected.
[0017] It is an advantage of embodiments of the present invention that the amount of information that needs to be collected from an object, e.g. scanning and/or imaging data, can be relatively low, e.g. when comparing to methods prior art of similar performance in defect detection, due to the incorporation of extensive prior knowledge about the objects under investigation in accordance with embodiments of the present invention, e.g. prior knowledge represented by a shape model and an average gray value distribution .
[0018] It is an advantage of embodiments of the present invention that a simple inspection and/or separation system can be implemented using simple and inexpensive hardware, e.g. relatively simple and inexpensive compared to prior art systems that have similar performance. Additionally, such deployment may not require any moving parts of the imaging and/or detection hardware, thereby reducing costs compared to a complete CT scanner.
[0019] It is an advantage of embodiments of the present invention that no tomographic reconstruction may be necessary, as classification can be advantageously performed by comparing two 2D images. It is a further advantage of embodiments of the present invention that a high throughput rate can be achieved, for example, due to relatively simple processing requirements. It is still an additional advantage that in-line applications such as automated quality control and/or selection of objects that are transported in-line in a conveyor system are feasible due to the high throughput speeds achievable.
[0020] It is an advantage of embodiments of the present invention that multiple defects in an object need not be identified one by one, as such defects are detected simultaneously as a large deviation from a reference object.
[0021] Embodiments of the present invention may combine model-based classification, 3D scanning, and radiographic imaging to detect discernible internal defects on radiographs and to advantageously detect more subtle features using a simulated radiograph of a three-dimensional reconstruction image. completeness of a reference object.
[0022] In a first aspect, the embodiments of the present invention refer to a non-destructive inspection method for in-line inspection of an object, the non-destructive inspection method comprising moving, with the use of a transport system in line, an object that is at least partially transparent to radiation of a predetermined radiation quality, for example, an object at least partially transparent by X-ray, along a predetermined path between a radiation source to emit radiation of radiation quality predetermined, for example, a source of ionizing radiation, and an image detector, and through a field of view of a three-dimensional digitizer. The method further comprises imaging the object using the image detector by detecting the radiation emitted by the radiation source and transmitted through the object to obtain a projection radiograph of an internal structure of the object. The method further comprises digitizing an outer surface of the object using the three-dimensional digitizer to obtain three-dimensional digitizing data of the object in the form of a point cloud representative of at least part of the outer surface. The method further comprises fitting, with the use of a processor, an object shape model to the point cloud to obtain a surface model of the external surface. The method also comprises creating, using the processor, a solid model of the surface model considering a gray value distribution of a reference object. The method comprises additionally simulating, with the use of the processor, a reference radiography from the solid model, and comparing, with the use of the processor, the reference radiography with the projection radiography to detect and/or measure the internal deviations of the object relative to the reference object. The step of creating the solid model and/or the step of simulating the reference radiograph considers a predetermined relative spatial configuration of the image detector, the radiation source and the three-dimensional digitizer.
[0023] In a method according to embodiments of the present invention, digitizing the outer surface may comprise generating a partial point cloud of the object, wherein the fit comprises estimating the complete outer surface and position of the object by fitting the model shape, where the creation of the solid model comprises filling in a volume defined by the surface model with the gray value distribution, where the volume corresponds to the space coordinates of the object when imaged by the image detector, and where the simulation of reference radiography comprises simulating an imaging process of the solid model by direct projection with the use of predetermined spatial configuration of the image detector and radiation source in the space coordinates.
[0024] In a method according to embodiments of the present invention, the format model and gray value distribution can be created by acquiring CT scans of a plurality of off-line reference object samples and determining the model of shape and gray value distribution from the CT scans to be used as background knowledge during the steps of tuning the shape model and creating the solid model at runtime.
[0025] In a method according to embodiments of the present invention, determining the format model may comprise extracting a plurality of external surfaces, corresponding to the plurality of reference object samples, from the CT scans using image processing techniques , determining a plurality of corresponding spatial features on the plurality of external surfaces, determining an average position and/or a variation in position of each of the plurality of corresponding spatial features, and determining the shape model by considering the positions. The format model may have free parameters that can be adjusted to account for object position, object orientation and/or object variation modes representative of variations in position of the plurality of corresponding spatial features.
[0026] In a method according to embodiments of the present invention, the shape model may comprise a linear model that parameterizes an object shape as a linear combination of an average shape and a plurality of variation modes.
[0027] In a method according to embodiments of the present invention, the gray value distribution can be determined as a normalized reference volume image from the plurality of CT scans.
[0028] In a method according to embodiments of the present invention, determining the gray value distribution may further comprise applying a surface normalization scheme to derive the normalized reference volume image from a population of objects represented by the object samples. independent of the shape of any individual object.
[0029] In a method according to embodiments of the present invention, determining the gray value distribution may comprise applying a normalized spherical sampling scheme to obtain the normalized reference volume image and creating the solid model may comprise applying a normalized spherical sampling scheme to obtain the normalized reference volume image and creating the solid model. inverted normalized spherical sampling to fit the reference volume image normalized to the surface model.
[0030] In a method according to embodiments of the present invention, the movement of the object may comprise moving the object in line on a conveyor belt.
[0031] In a method according to embodiments of the present invention, moving the object may move the object at a speed in the range of conveyor belt speeds of commercial facilities, for example, in the range of 0.1 m/s to 0.7 m/s
[0032] In a method according to embodiments of the present invention, during the movement of the object along the predetermined path, the object may first pass through the three-dimensional digitizer, and then, consequently, may pass through the field of view of the detector. Image.
[0033] In a method according to embodiments of the present invention, the object may be moved in-line through a radiation field of each of a plurality of radiographic imaging systems, each comprising a radiation source, for example, a source of ionizing radiation, and an image detector, and through a digitization stage of at least one three-dimensional digitizer.
[0034] In a method according to embodiments of the present invention, the radiation source, e.g. the ionizing radiation source, and the image detector may be statically arranged with respect to the in-line transport system, e.g. mechanically. fixed in relation to the on-line transport system.
[0035] In a method according to embodiments of the present invention, the radiation source, e.g., the ionizing radiation source, and the image detector may be fixed above a conveyor belt on which the object is transported.
[0036] In a second aspect, the embodiments of the present invention refer to a non-destructive inspection system for in-line inspection of an object. The non-destructive inspection system comprises a radiation source for emitting radiation of a predetermined radiation quality, for example a source of ionizing radiation, and an image detector. The radiation source and the image detector form a radiographic imaging system for detecting radiation emitted by the radiation source and transmitted through an object that is at least partially transparent to radiation of the predetermined radiation quality, for example, an object at less partially transparent by X-ray, to provide a projection radiograph of an object's internal structure. The system further comprises a three-dimensional digitizer for digitizing an outer surface of the object to obtain three-dimensional digitizing data of the object in the form of a point cloud representative of at least part of the outer surface. The system also comprises an in-line conveyor system for moving the object along a predetermined path between the radiation source, e.g., the ionizing radiation source, and the image detector, and through a field of view of the digitizer. three-dimensional.
[0037] The system also comprises a processor adapted to: obtain the projection radiograph of the image detector; get the point cloud from the three-dimensional digitizer; fit an object shape model to the point cloud to get a surface model of the outer surface; create a solid model of the surface model considering a gray value distribution of a reference object; simulate a reference radiograph from the solid model; and comparing the reference radiograph with the projection radiograph to detect and/or measure internal deviations of the object from the reference object. The processor is further adapted to create the solid model and/or simulate the reference radiography considering a predetermined relative spatial configuration of the image detector, the radiation source and the three-dimensional digitizer.
[0038] In a system according to embodiments of the present invention, the radiation source, for example the ionizing radiation source, may be a stationary radiation source and the image detector may be a stationary image detector.
[0039] A system in accordance with embodiments of the present invention may further comprise a plurality of stationary radiation sources, for example, a plurality of stationary ionizing radiation sources, and stationary image detectors which form a plurality of radiographic imaging systems.
[0040] In a system according to embodiments of the present invention, the three-dimensional digitizer may comprise a laser or stationary light source and a stationary light detector.
[0041] In a system according to embodiments of the present invention, the radiation source may comprise a source of ionizing radiation adapted to deliver a pulse of X-ray exposure to the object, and the image detector may comprise a digital image detector adapted to provide object image data corresponding to the X-ray exposure pulse as an input to the processor. The three-dimensional digitizer may be adapted to provide light ray exposure to an object and may comprise a digital detector for providing data regarding the object obtained by light ray exposure as an input to the processor.
[0042] In a further aspect, the present invention may also refer to the use of a method according to embodiments of the first aspect of the present invention to identify anomalous objects transported on a conveyor line in an industrial process.
[0043] Particular and preferred aspects of the invention are defined in the appended independent and dependent claims. Dependent claims features may be combined with independent claims features and other dependent claims features as appropriate and not merely as explicitly defined in the claims.
[0044] These and other aspects of the invention will be apparent and elucidated with reference to the embodiment (or embodiments) described later herein. BRIEF DESCRIPTION OF THE FIGURES
[0045] Figure 1 illustrates an exemplary method according to embodiments of the present invention.
[0046] Figure 2 illustrates a method according to embodiments of the present invention.
[0047] Figure 3 illustrates an exemplary system in accordance with embodiments of the present invention.
[0048] Figure 4 illustrates a surface normalization approach using spherical sampling that relates to embodiments of the present invention.
[0049] Figure 5 shows an example of basic image processing techniques that can be applied in a calibration step of a method according to embodiments of the present invention.
[0050] Figure 6 shows a parameterized format according to a format model that can be applied in embodiments of the present invention.
[0051] Figure 7 illustrates a runtime process in an example that refers to embodiments of the present invention.
[0052] Figure 8 further illustrates the process flow of said example in relation to embodiments of the present invention.
[0053] Figure 9 illustrates the exemplary format - defect combinations with respect to an example to demonstrate embodiments of the present invention.
[0054] Figures are schematic only and are non-limiting. In the figures, the size of some of the elements may be exaggerated and not drawn to scale for illustrative purposes.
[0055] Any numerical references in the claims should not be interpreted as limiting the scope.
[0056] In the different figures, the same numerical references refer to the same or similar elements. DETAILED DESCRIPTION OF THE INVENTION
[0057] The present invention will be described with respect to particular embodiments and with reference to certain figures, but the invention is not limited thereto, but only by the claims. The figures depicted are schematic only and are non-limiting. In drawings, the size of some of the elements may be exaggerated and not drawn to scale for illustrative purposes. Dimensions and relative dimensions do not correspond to actual reductions for the practice of the invention.
[0058] Additionally, the terms first, second and similar in the description and claims are used to distinguish between similar elements and not necessarily to describe a sequence, temporally, spatially, in classification or in any other way. It is to be understood that the terms so used are interchangeable under appropriate circumstances and that the embodiments of the invention described herein are capable of operation in sequences other than those described or illustrated herein.
[0059] Furthermore, the terms top, under and the like in the description and claims are used for descriptive purposes and not necessarily to describe relative positions. It is to be understood that the terms so used are interchangeable under appropriate circumstances and that the embodiments of the invention described herein are capable of operation in orientations other than those described or illustrated herein.
[0060] It should be noted that the term "comprising", used in the claims, is not to be interpreted as restricted to the meanings listed after this; they do not exclude other elements or steps. It should therefore be interpreted as specifying the presence of the declared features, integers, steps or components as named, but not excluding the presence or addition of one or more other features, integers, steps or components, or groups thereof. Thus, the scope of the expression "a device comprising means A and B" should not be limited to devices consisting only of components A and B. It means that, in relation to the present invention, the only relevant components of the device are A and B.
[0061] Reference throughout this specification to "one (01) embodiment" or "one embodiment" means that a particular feature, structure or feature described in connection with the embodiment is included in at least one embodiment of the present invention. Thus, the appearances of the expressions “in one (01) modality” or “in one modality” in various places throughout this descriptive report are not necessarily all referring to the same modality, but they may be. Additionally, the particular features, structures or features may be combined in any suitable manner, as would be apparent to a person of ordinary skill in the art from this disclosure, in one or more embodiments.
[0062] Similarly, it should be noted that, in describing exemplary embodiments of the invention, various features of the invention are sometimes grouped into a single embodiment, figure, or description thereof for the purpose of simplifying disclosure and aiding understanding. of one or more of the various inventive aspects. This method of disclosure, however, should not be interpreted as reflecting an intention that the claimed invention requires more resources than is expressly stated in each claim. Rather, as the following claims reflect, the inventive aspects reside in less than all the features of a single disclosed modality mentioned above. Accordingly, the claims which follow the detailed description are expressly incorporated into that detailed description, each claim being held by itself as a separate embodiment of this invention.
[0063] Additionally, while some embodiments described herein include some, but not other features included in other embodiments, combinations of features from different embodiments are intended to be within the scope of the invention, and form different embodiments, as would be understood by those in the technique. For example, in the claims below, any of the claimed modalities can be used in any combination.
[0064] In the description provided in this document, several specific details are presented. However, it is to be understood that embodiments of the invention may be practiced without these specific details. In other situations, well-known techniques, structures, and methods have not been shown in detail in order not to obscure an understanding of this description.
[0065] When reference is made in embodiments of the present invention to "gray value", reference is made to a digital pixel or voxel value. In particular, it may refer to a scalar location-dependent value defined by a pixel or voxel coordinate system. The gray value of a pixel may be indicative of an amount of radiation received by a corresponding image detector, for example, which is proportional or has a monotonic functional relationship with a radiation intensity or magnitude. For example, in volumetric images, the voxel gray value may be proportional to a local measure of attenuation of the radiation quality used for imaging by the material present at the voxel location. For example, for X-ray imaging, this voxel gray value can be proportional to the linear attenuation coefficient corresponding to the attenuation of that X-ray radiation in the voxel volume. For example, the voxel gray value can be normalized to Hounsfield units.
[0066] In a first aspect, the present invention relates to a non-destructive inspection method for in-line inspection of an object. This non-destructive inspection method comprises moving, using an in-line conveyor system, an object that is at least partially transparent to radiation of a predetermined quality, for example an X-ray transparent object, along a predetermined path. between the radiation source for emitting radiation of that predetermined quality, e.g. a source of ionizing radiation, and an image detector, e.g. a stationary radiation source and a stationary image detector, and across a field of view of a three-dimensional scanner. The method further comprises imaging the object using the image detector by detecting the radiation emitted by the radiation source and transmitted through the object to obtain a projection radiograph of an internal structure of the object. The method also comprises digitizing an outer surface of the object using the three-dimensional digitizer to obtain three-dimensional digitizing data of the object in the form of a point cloud representative of at least part of the outer surface.
[0067] The method further comprises fitting, using a processor, an object shape model to the point cloud to obtain a surface model of the external surface, and creating, using the processor, a solid model of the model. surface considering a gray value distribution of a reference object, for example, a defect-free reference object. The method also comprises simulating, with the use of the processor, a reference radiography from the solid model, and comparing, using the processor, the reference radiography with the projection radiography to detect and/or measure the internal deviations of the object, for example defects, in relation to the reference object.
[0068] The step of creating the solid model and/or the step of simulating the reference radiograph considers a predetermined relative spatial configuration of the image detector, the radiation source and the three-dimensional digitizer, for example, in order to generate the radiograph reference in a coordinate system compatible with the projection radiography coordinate system obtained by imaging the object.
[0069] Embodiments of the present invention may refer to a method, and related system, of using radiographic simulation, which may involve the use of ionizing radiation to detect a radiographic image for the inspection of internal defects in an object. , in which additional features can be detected in the object, e.g. more subtle features, by comparing the acquired radiograph image with a simulated radiograph based on a full three-dimensional reconstruction image of a reference object, adjusted to geometric constraints of an object shape detected by a three-dimensional scanner.
[0070] A method according to embodiments of the present invention may combine a three-dimensional digitizer, for example, comprising a laser line array and/or multiple RGB cameras, with projection imaging, for example, optical projection imaging of a object that is (at least partially) optically transparent or X-ray radiography of an object that is (at least partially) transparent to X-rays. Referring to Figure 2, the working principles of a method according to the modalities are schematically illustrated. According to embodiments, a 3D digitizer can produce a partial point cloud 31 by digitizing a physical object 37. The partial point cloud 31 can then be used to estimate the complete outer surface 32 of the object by adjusting a format model, for example, a statistical format model. This surface model, e.g. the "hollow" model of the object, can then be filled with a density distribution, e.g. a representative gray value distribution of an average object without defects, to produce a volume complete 33 from which the radiograph 34 can be simulated 35, for example using a forward projection. This simulated radiograph 34 can then be compared with a measured radiograph 36, for example a measured radiograph 36 obtained by projection imaging 40 of the physical object 37 using an X-ray system. Because the simulated radiograph represents an object perfect without defects, the differences observed 41 may have contributed to internal defects.
[0071] With reference to Figure 1, an exemplary non-destructive inspection method 10 for in-line inspection of an object in accordance with embodiments of the present invention is shown. For example, such a method 10 may be a method for automatic real-time non-destructive inspection control, for example, to detect damage or a defect of an object, for example, a produce or fruit and vegetable item, such as a vegetable or a fruit. The method can be adapted for non-destructive testing of a produce or produce, for example by using ionizing radiation imaging. The method can be adapted for in-line product inspection, e.g. for automated quality control and/or selection of an object, e.g. a produce or produce, such as a vegetable or fruit, that is at least partially transparent to the ionizing radiation, for example, an X-ray transparent object. However, embodiments of the present invention are not necessarily limited thereto, for example, a method according to embodiments of the present invention may also refer to on-line product inspection of an optically transparent object, for example a glass or transparent polymeric object. Additionally, one skilled in the art will understand that embodiments of the present invention may also refer to other qualities of radiation, such as electron beams, infrared radiation, ultraviolet radiation, hadron radiation, or acoustic waves.
[0072] This non-destructive inspection method 10 comprises moving 11 an object that is at least partially transparent to a predetermined radiation quality, for example an object at least partially transparent by X-ray, along a predetermined path between a source of radiation, for example, to emit radiation having said radiation quality, and an image detector and through a field of view of a three-dimensional digitizer. The 3D digitizer's field of view refers to a spatial volume on which the digitizer can operate, when an object is positioned in that spatial volume, to acquire the digitization data for that object.
[0073] In particular, the object can be moved along this predetermined path by an in-line conveyor system, for example on a conveyor belt. In accordance with embodiments of the present invention, moving the object may comprise moving the object in line on a conveyor belt. According to embodiments of the present invention, the movement of the object may comprise moving the object at a speed in the range of 0.1 m/s to 0.7 m/s, for example on a conveyor belt.
[0074] The radiation source and the image detector may form a projection imaging system, eg a radiography projection imaging system, eg an X-ray projection imaging system. The radiation source it may comprise, for example, a Rontgen tube, a gamma radiation source or a linear particle accelerator for generating X-ray radiation from a suitable target.
[0075] The ionizing radiation source and the image detector can be statically arranged in relation to the in-line transport system. For example, the radiation source and image detector may comprise a stationary radiation source and a stationary image detector. The source of ionizing radiation and the image detector can be, for example, fixed above a conveyor belt on which the object is transported.
[0076] The radiation source and the image detector can form a radiographic imaging system. In accordance with embodiments of the present invention, the object may be moved in-line through a radiation field of each of a plurality of radiographic imaging systems, each comprising a source of ionizing radiation and an image detector. and through a digitization stage of at least one three-dimensional digitizer. For example, the method may comprise combining the point clouds that characterize the outer surface of the object obtained by multiple three-dimensional scanners at different locations along a transport line in order to improve a model fit of the format model to the cloud data. of aggregated and/or filtered points. For example, the method may comprise imaging the object along different projection angles by multiple radiographic imaging systems, and performing the steps of simulating a reference radiograph and comparing the reference radiograph to each of the projection radiographs obtained. for different projection angles. In this way, a subtle defect that may be obscured on a first projection radiograph can be detected on another projection radiograph.
[0077] The method 10 additionally comprises imaging the object using the image detector by detecting the radiation emitted by the radiation source. This radiation is additionally transmitted through the object during propagation from the source to the detector, for example, in order to encode internal information regarding the object in spatial variations in the intensity of the radiation field on the detector's sensing surface. In this way, a projection radiograph is obtained by an internal structure of the object.
[0078] Method 10 further comprises scanning 13 an external surface of the object using a three-dimensional digitizer to obtain three-dimensional digitizing data of the object in the form of a point cloud representative of at least part of the external surface. For example, such a three-dimensional scanner may comprise a laser line scanner or multiple RGB cameras. Such a 3D digitizer can be a device adapted to analyze a physical object to collect the data in its format, so as to collect data that can be used to build a partial or complete digital three-dimensional model of the object. The 3D digitizer can be an optical 3D digitizer. The 3D digitizer may comprise an active non-contact digitizer, for example with the use of light emissions or ultrasound. For example, the 3D scanner may comprise a time-of-flight 3D laser scanner, a triangulation-based 3D laser scanner, or a conoscopic holographic laser scanner. The 3D scanner may comprise a structured 3D light scanner or a modulated 3D light scanner. The 3D scanner may also comprise a non-contact passive 3D scanner, such as a stereoscopic optical imaging system, a photometric imaging system, or a silhouette imaging system.
[0079] In a method according to embodiments of the present invention, during movement 11 of the object along the predetermined path, the object may first pass through the three-dimensional digitizer. A processed point cloud can be produced by the 3D digitizer when the object passes through the digitizer. This point cloud can be incomplete, for example, due to scan artifacts on a lower side of the object where it is supported by a conveyor belt. Such a shortcoming can be removed by estimating the complete external shape of the object using a shape model, for example a statistical format model, as described further in this document below. The object can therefore pass through the field of view of the image detector in which it is imaged 12.
[0080] This scan 13 of the outer surface may comprise generating a partial point cloud of the object, for example, a partial point cloud representative of at least one segment of the outer surface of the object that is positioned and oriented in a direct line of sight of the three-dimensional digitizer. For example, the three-dimensional digitizer can be adapted to generate a three-dimensional point cloud of points on the outer surface of the object. For example, the three-dimensional digitizer may comprise a laser line and an RGB camera system.
[0081] Method 10 additionally comprises fitting 14 an object shape model to the point cloud, with the use of a processor, to obtain a surface model of the external surface. For example, this fit may comprise estimating a complete outer surface and position, eg a position and orientation, of the object by fitting the shape model to the point cloud, eg the partial point cloud. By fitting the shape model, a complete surface, eg a hollow shape descriptor, can be obtained. Such adjustment may comprise determining a plurality of format model parameters, for example, determining a linear combination of components, a translation vector, a rotation vector and/or a similar transformation matrix. This fit may comprise a search algorithm to find a combination of parameters corresponding to a maximum or a minimum of an objective function, for example, in order to maximize the overlap between the shape model and the measured point cloud or to minimize a deviation between the format model and the measured point cloud.
[0082] Method 10 also comprises creating a solid model of the surface model, using the processor, considering a gray value distribution, for example, a normalized gray value distribution, of a reference object , for example, a defect-free reference object. Creating the solid model can comprise filling in a volume defined by the surface model with the gray value distribution, where that volume corresponds to the space coordinates of the object when imaged by the image detector. This volume may, for example, correspond to the space coordinates of the object due to the estimation of the position and/or orientation of the object in the adjustment mentioned above in this document.
[0083] For example, a normalized reference volume, e.g. derived from a CT dataset previously obtained from reference object samples, can be used to produce a volumetric image approximation of a conforming reference object with the format template. The solid model may comprise, for example, a volumetric image, for example an approximation of a CT scan of the reference object, obtained by reverse sampling to a reference volume normalized using a surface normalization scheme. The reference object can be representative of a perfect situation of the object that passed under the three-dimensional digitizer. This perfect situation can refer to a perfect situation that has a format corresponding to the scanned object, achievable to some extent by parameterization of the format model, yet not having defects or abnormal deviations in internal structure for the object samples used to build the volume. standard reference.
[0084] The method 10 also comprises simulating 16 a reference radiograph of the solid model using the processor. This step of simulating the reference radiography may comprise simulating an imaging process of the solid model by forward projection using the predetermined spatial configuration of the image detector and radiation source in the space coordinates of the object when imaged by the image detector. This reference radiograph can be, for example, simulated from the solid model, for example the volumetric image, using a forward projection method.
[0085] The step of creating 15 the solid model and/or the step of simulating 16 the reference radiograph considers a predetermined relative spatial configuration of the image detector, the radiation source and the three-dimensional digitizer, for example, in order to generate the reference radiograph in a coordinate system compatible with the projection radiography coordinate system obtained by imaging the object.
[0086] The method 10 additionally comprises comparing 17 the reference radiography with the projection radiography, using the processor, to detect and/or measure the internal deviations, for example, defects, of the object in relation to the reference object. For example, the measured projection radiograph can be compared to the simulated reference radiograph, in which any differences between the modeled and measured projection may be indicative of the presence of defects. Thus, if a substantial difference is detected, the object can be classified as defective. The result of this comparison may be to catch an operator's attention, or it may be fed as a signal to an automatic separator to automatically remove the object from the in-line conveyor line.
[0087] A method according to embodiments of the present invention may comprise obtaining the format model and the gray value distribution in the form of a predetermined format model and a predetermined gray value distribution as an input to be received by the processor .
[0088] A method according to embodiments of the present invention may comprise determining the shape pattern and gray value distribution in a calibration phase, as further described below herein.
[0089] In a method according to embodiments of the present invention, the format model and gray value distribution can be created 21 by acquiring CT scans 22 of a plurality of off-line reference object samples, for example , in an initial calibration phase prior to in-line application of the method, and determine the format model 23, for example by building a CAD model or applying a surface modeling method, and the gray value distribution 24 from these CT scans in order to be used as background knowledge during the steps of adjusting the format model and creating the solid model at runtime, for example during in-line application of the method. A method according to embodiments of the present invention may comprise such an initial step of creating the shape model and the gray value distribution.
[0090] In a method according to embodiments of the present invention, determining the shape model may comprise determining a surface model by extracting 25 a plurality of external surfaces, corresponding to the plurality of reference object samples, from the CT scans using image processing techniques. For example, image processing techniques as known in the art can be applied to construct such an external surface, e.g. image segmentation, edge detection, morphological filtering, image pre-processing and/or image post-processing. Image. Such image processing may also comprise, for example, image recording the plurality of CT scans, for example, aligning the images in position and orientation, to facilitate detection of corresponding spatial features in the step of determining corresponding spatial features discussed. additionally below in this document.
[0091] Determining 23 the shape model may further comprise determining 26 a plurality of corresponding spatial features on the plurality of external surfaces, for example detecting each spatial feature on each of the plurality of external surfaces, so that each spatial feature of the outer surface of a sample object corresponds to a corresponding feature of the outer surface of each sample object. For example, such corresponding spatial features may be called landmark features.
[0092] Determining the format model may further comprise determining 27 an average position and/or a variation in position of each among the plurality of corresponding spatial features.
[0093] Determining the format model may also comprise determining the format model that considers the average positions of each among the plurality of corresponding spatial features, for example, that considers the average positions and variations in positions.
[0094] Thus, according to embodiments of the present invention, the format model can have free parameters that can be adjusted to explain object position, object orientation and/or object variation modes representative of variations in position of the object. plurality of corresponding spatial resources.
[0095] In a method according to embodiments of the present invention, the shape model may comprise a linear model that parameterizes an object shape as a linear combination of an average shape and a plurality of variation modes. For example, this medium format may correspond to a surface constructed using the average positions of each of the plurality of corresponding spatial features, while the variation modes may correspond to different surfaces with respect to this average format, so that the medium format linear distance and variation modes span the plurality of external surfaces determined from the CT scans.
[0096] In a method according to embodiments of the present invention, the gray value distribution can be determined as a normalized reference volume image from the plurality of CT scans. For example, determining the gray value distribution may comprise applying a surface normalization scheme to derive the normalized reference volume image from a population of objects represented by the object samples, so that the gray value distribution is defined independently of the shape of any individual object. In accordance with embodiments of the present invention, the gray value distribution can be determined by applying a normalized spherical sampling scheme 28 to obtain the normalized reference volume image. For example, the gray value distribution can be derived from the CT dataset by performing a spherical sampling, e.g. from the center of each scan, normalized to the surface of the object sample, e.g. by normalizing the radial distance to the surface to one for each radial sampling line. For example, a normalized spherical sampling scheme can be applied to each of the plurality of CT scans individually, and the resulting normalized reference volume images can be aggregated by applying a summary statistic over the set of CT volume images. normalized reference, for example by estimating the average of the plurality of reference volume images.
[0097] In accordance with embodiments of the present invention, creating 15 the solid model may comprise applying 29 an inverted normalized spherical sampling scheme to fit the normalized reference volume image to the surface model. For example, the normalized reference volume image can define common internal structure of several scanned reference object samples, regardless of their format. In this way, any shape defined by the surface model of a particular object manually can be filled with this normalized reference volume image by applying an inverted normalized spherical sampling scheme.
[0098] In a second aspect, the present invention also relates to a non-destructive inspection system for in-line inspection of an object. The non-destructive inspection system comprises a radiation source, for example a radiation source for emitting radiation of a predetermined radiation quality, for example a source of ionizing radiation, and an image detector, the source and detector being form a radiographic imaging system for detecting radiation emitted by the radiation source and transmitted through an object that is at least partially transparent to said predetermined radiation quality, for example an object at least partially X-ray transparent, to provide a projection radiography of an object's internal structure. The system further comprises a three-dimensional digitizer for digitizing an outer surface of the object to obtain three-dimensional digitizing data of the object in the form of a point cloud representative of at least part of the outer surface. The system also comprises an in-line conveyor system for moving the object along a predetermined path between the source of ionizing radiation and the image detector and through a field of view of the three-dimensional digitizer.
[0099] A system in accordance with embodiments of this second aspect of the present invention may implement a method in accordance with embodiments of the first aspect of the present invention. Therefore, the features of a system in accordance with embodiments of the second aspect of the present invention may be clear to the person skilled in the art in view of the description provided above herein regarding embodiments of the first aspect of the present invention. Similarly, features of a method in accordance with embodiments of the first aspect of the present invention may be clear to the person skilled in the art in view of the description provided below herein with respect to embodiments of the second aspect of the present invention.
[00100] The non-destructive inspection system additionally comprises a processor adapted to obtain the projection radiograph from the image detector and to obtain the point cloud from the three-dimensional digitizer. This processor is further adapted to fit an object shape model to the point cloud to obtain a surface model of the object's outer surface. The processor is also adapted to create a solid model of this surface model by considering a gray value distribution of a reference object. The processor is further adapted to simulate a reference radiograph from this solid model and to compare the reference radiograph with the projection radiograph to detect and/or measure the internal deviations of the object in relation to the reference object. The processor is also adapted to create the solid model and/or simulate the reference radiography considering a predetermined relative spatial configuration of the image detector, the radiation source and the three-dimensional digitizer.
[00101] Referring to Figure 3, a non-destructive inspection system 50 for in-line inspection of an object in accordance with embodiments of the present invention is shown. This non-destructive inspection system 50 can be a multimodal imaging device, for example, combining a 3D scanner and an X-ray imaging device, for in-line product inspection.
[00102] The non-destructive inspection system 50 comprises a source of ionizing radiation 2 and an image detector 3, the source and detector forming a radiographic imaging system to detect the radiation emitted by the radiation source and transmitted through an object at least partially X-ray transparent to provide a projection radiograph of an internal structure of the object. The ionizing radiation source 2 may be a stationary radiation source for emitting the radiation, for example a stationary X-ray source for emitting the X-ray radiation. The image detector 3 may be a stationary image detector, for example , a digital X-ray image detector, for capturing a projection or radiography image of the object, for example, a produce or produce, by detecting the radiation when transmitted through the object. The radiation source 2 can emit the ionizing radiation towards a central point of a radiation sensitive area of the image detector 3, e.g. it can emit radiation in an average direction, e.g. along a geometric axis direction beam 7, which may correspond to a mathematical line connecting the source focus to a central detector pixel location in detector 3. The image detector 3 may consist of a row of detector pixels, or a two-dimensional array of detector pixels.
[00103] The radiation source 2 can emit radiation 7 at a predetermined cone angle so as to preferentially and substantially cover the entire radiation sensitive area of the image detector 3. The source 2 can also provide a sufficiently high radiation flux. in order to obtain a satisfactory signal-to-noise ratio when imaging object 6 using image detector 3.
[00104] The radiation source 2 may comprise a light source, for example a light source for emitting light in the infrared, near infrared, optical and/or ultraviolet spectral range. The radiation source 2 may comprise an acoustic wave source.
[00105] The radiation source 2 may comprise a radioactive source, for example a gamma emitter, or an X-ray tube. The X-ray source may, for example, emit photons having energies in the range of 100 eV to 400 keV, for example, in the range of 1 keV to 200 keV, for example, in the range of 10 keV to 100 keV. However, embodiments of the present invention may also refer to other types of radiation, for example, particularly to types of radiation that can be transmitted through the object being tested along a substantially path, for example, without reflection, refraction or diffraction. of the radiation wave, while the absorption of radiation in the object to be tested is neither too high nor too low, so that an acceptable contrast range can be achieved in the radiographic image. It should be noted that the person skilled in the art is aware of suitable radiation types, as known in the art, for imaging radiation of a specific object, given its material properties and spatial dimensions, and is aware of known corresponding image sources and detectors. in the art for such type of radiation, which can thus be used as the radiation source 2 and image detector 3 in accordance with embodiments of the present invention.
[00106] The image detector 3 is adapted to capture a projection or radiograph image of the object 6 by detecting the radiation when transmitted through the object 6. The projection image can be obtained by means of an image detector 3 which is adapted to capture, e.g. acquire or detect, parts of the projection image at different points in time, e.g. acquire the projection image not simultaneously and/or in time frames corresponding to mutually disjunctive exposure time frames. For example, the image detector may comprise a one-dimensional array of pixels, e.g. a line array, and a two-dimensional image may be collected as the object moves through the field of view of the radiographic imaging system.
[00107] The system 50 further comprises a three-dimensional digitizer 1 for digitizing an outer surface of the object to obtain three-dimensional digitizing data of the object in the form of a point cloud representative of at least part of the outer surface. This three-dimensional digitizer can be a 3D digitizing device, for example, comprising a laser line or multiple RGB cameras. The 3D scanner 1 may comprise one or more scanner sources 8 for emitting radiation 9 and one or more detectors 51 for capturing reflected radiation 9 from object 6 as it moves in the transport system 5. Digitizer source 8 may comprise one or more multiple radiation sources, for example comprising a laser and/or a light source. Digitizer 1 may use laser triangulation in which detector 51 captures laser light that is reflected from the object. Using trigonometric triangulation, using a precisely predetermined distance between the laser source and the detector, as well as a precisely predetermined angle between the laser and the sensor, the system can calculate the distance from the point on the object surface to the digitizer. . The 3D digitizer 1 can also use a pulse-based technique, also known as time-of-flight scanning, based on a constant speed of light and a period of time in which light from a laser source reaches the object and is reflected to the detector. The 3D digitizer 1 can operate in a phase shift mode, in which the power of a laser beam is modulated, and the digitizer is adapted to compare the phase of the sent laser light and the laser light at the detector after reflection. of the object. The 3D digitizer 1 may also comprise a conoscopic system, in which a laser beam is projected onto the surface and the immediate reflection along the same ray-path is transmitted through a conoscopic crystal and projected onto a detector. Digitizer 1 can also use a structured light scanning method that projects a series of linear patterns onto the object and detects the edge of the projected pattern with a camera, and similarly calculates the distance.
[00108] The 3D digitizer 1 can produce point cloud data of the three-dimensional surface topology of the object 6, which can be processed, e.g. stored and processed, in the processor 4 described further below in the present document, e.g. stored and processed in a machine vision device. The point cloud can consist of several points, e.g. coordinates that identify such points, on the surface of the object 6, which allows a geometric shape model to be adapted to the cloud in order to describe the surface of the object in a complete way. geometrically. To limit the number of points needed to adjust the geometric shape, the processor 4, for example the machine vision device, can include a database of reference shapes that have been trained externally from the object category.
[00109] The system also comprises an in-line transport system 5 for moving the object along a predetermined path between the source of ionizing radiation and the image detector and through a field of view of the three-dimensional digitizer. The predetermined path thereby traverses a field of view of the radiographic imaging system, for example, so as to enable imaging of the object as it moves along the predetermined path. The predetermined path also traverses a field of view of the three-dimensional digitizer, for example, in order to enable scanning of the object as it moves along the predetermined path. Such an in-line conveyor system may comprise a conveyor line for moving the object, for example the produce or produce, between the radiation source and the image detector and, for example, under the 3D scanner.
[00110] The in-line transport system 5 may comprise a transport line for moving the object 6 along a predetermined path through the 3D scanner 1 and between the radiation source 2 and the image detector 3. For example, the conveyor line 5 can be adapted to move a chain of objects 6 along the predetermined path, for example to move each object 6 in the chain along the path.
[00111] The object 6 can thus be moved along a path in the 3D digitizer 1 and between the detector 3 and the radiation source 2, for example through a field of view observed by the detector 3, for example , a digital X-ray detector, in relation to a radiation field, e.g. an X-ray beam, emitted by radiation source 2, e.g. a divergent X-ray beam emitted from a substantially similar focal point to a point on an X-ray tube and substantially directed to a digital X-ray image detector.
[00112] Such a conveyor line may comprise, for example, a conveyor belt, for example, a moving conveyor belt on which the item of product is supported as it is moved along the path, or an overhead conveyor from which the item of product is suspended as it is moved along the path.
[00113] In operation, the object 6 to be inspected can travel along a path formed by the predetermined path, in the volume of space between the digitizer source 8 and the image detector 51, for example, in such a way that the images acquired by the image detector 51 can be used by the processor 4 to provide a sufficiently accurate 3D format of the object, for example, sufficiently accurate in view of the product inspection needs imposed by a specific application.
[00114] In operation, the object 6 to be inspected can then travel along a trajectory formed by the predetermined path, in the volume of space between the radiation source 2 and the radiation detector 3, for example, in such a way that the images acquired by the image detector 3 can be used by the processor 4 to provide a sufficiently accurate projection image of the object positioned in the same orientation on the transport line as when traversing through the 3D digitizer, for example, sufficiently accurate in view of needs inspection requirements imposed by a specific application. Alternatively, a predetermined change of orientation and/or position of the object when scanned by the 3D scanner 1 and when imaged by the radiographic imaging system can be considered by the processor 4.
[00115] The non-destructive inspection system 50 additionally comprises a processor 4. This processor may comprise a machine vision device. Such a machine vision device may comprise a 3D geometry reconstruction unit to determine the three-dimensional external shape of the object, for example the produce or produce item, based on the data provided by the three-dimensional digitizer data. The machine vision device may comprise a 3D volume rendering unit for generating an internal representation of the product, and a radiography generating unit for determining a radiographic image of the internal representation of the product.
[00116] Processor 4 may have associated memory for storing executable program code and/or data. The processor may comprise, or may be integrated into, a computer or digital processing workstation. The processor may comprise general purpose hardware for executing software code to instruct the processor to perform some or all of the tasks as further described below herein. The processor may also comprise such software code, for example stored in a memory or on a data carrier readable by a data storage reader connected to or as part of the processor. The processor may comprise hardware specifically adapted to perform some or all of the tasks further described below herein, for example, a specific integrated circuit, a field programmable gate array device or a similar configurable hardware device known in the art. The processor may be provided with data representing an image of the object, for example, via a connection to the radiographic imaging system, and with data representing a three-dimensional digitization of the object, for example, via a connection to the three-dimensional digitizer. . The processor may additionally store this data in memory associated with the processor. Optionally, a user interface having at least one input may be operatively connected to the processor so that a user can receive processor output and/or input commands to the processor.
[00117] Processor 4 is adapted to obtain the projection radiograph from the image detector and to obtain the point cloud from the three-dimensional digitizer. This processor 4 can form a machine vision unit to complement the imaging chain after 3D scanning and radiographic image collection.
[00118] This processor is further adapted to fit an object shape model to the point cloud to obtain a surface model of the object's outer surface. In this way, the point cloud can be used to compute a complete outer surface of the object by fitting a shape model (SM) of the object. Shape models can comprise any technical method for describing the three-dimensional shape of an object. The shape model can be rigid or deformable. A deformable model can be interpreted as a generalization of a rigid representation. Thus, methods and systems according to embodiments of the present invention may refer to deformable or rigid models, for example, when applied to rigid models, the degree of freedom related to shape deformation will be absent. An example of rigid models are those that result from conventional computer-aided design (CAD). Such CAD models may be widely available for products in production environments where substantially identical products need to be inspected. In this case, the 3D digitizer will serve to assess the exact translational and rotational position of the product, for example on a conveyor, based on the measured point cloud. An example of deformable format models are statistical format models (SSM), such as those described by Heimann et al. in “Statistical shape models for 3D medical image segmentation: a review”, Medical Image Analysis, 13(4), pages 543 to 563. The contents of this referenced document are incorporated herein by reference. In embodiments according to the present invention, the format model may comprise an SSM obtained by principal component decomposition, for example, which results in an average format for the object population and possible deviations therefrom. However, in other embodiments according to the present invention, other methods for defining the shape model can be applied, for example, parametric methods, such as parametric methods based on spherical harmonics. In the case of a deformable shape model, the 3D digitizer 1 can assist in the assessment of the translational and rotational position of the object, and can additionally serve to determine a situation of best fit of a model population to the object shape of the object 6.
[00119] Processor 4 is also adapted to create a solid model of this surface model considering a gray value distribution of a reference object. For example, the processor 4, for example the machine vision unit, can fill the 3D format model of the object 6 to obtain an internal representation or volumetric image of the object. In accordance with embodiments of the present invention, an internal representation of the object can be provided by generating a distribution within the object representation that provides values of a physical property of the object that influences the interaction of radiation applied by the ionizing radiation source 2 with the object. In one embodiment, this property may be a density that determines the energy absorption of X-ray radiation. In one example, this value distribution may be spatially uniform within the object, for example, the entire 3D model of the object is filled with the same value. This might, for example, be a suitable option for objects composed of a material that has a uniform density. In another example, this distribution is normalized to the surface by applying spherical sampling from the center of the shape where the radius is normalized to the distance to the surface, as illustrated in Figure 4. In yet another example, this distribution can normalized by applying a reversible non-rigid record to a reference sample.
[00120] Value distribution may not be additionally uniform, but position dependent. In embodiments according to the present invention, the object representation may be subdivided into discrete volume elements, called voxels, to which different values may be assigned. This value distribution can represent internal structures in the object, for example, in a case where the object consists of parts made of different materials that have a different density, for example, possibly including the presence of internal cavities.
[00121] Such a value distribution can be obtained from a prior knowledge database 52 that contains normalized object reference descriptions, e.g. descriptors in a machine-readable and machine-parsable form, including, for example, the 3D format and the internal representation and possibly other statistics. This database can be acquired using 3D volumetric scans of various sample object samples, for example, defect-free product items. Such volumetric scans can, for example, be obtained using a tomography imaging and reconstruction system, and can be used to create a model of the object under consideration, to be used as background knowledge during runtime.
[00122] The value distribution in the database can be normalized to the object format population, for example, as represented in Figure 4, and can be reverse sampled in the measured situation of the described format, for example, the model fitted conformation that forms a surface model of the digitized object, thereby resulting in a volumetric representation, for example a volumetric image, of the object under investigation with internal reference properties. Methods for obtaining surface normalized volume data may include surface normalized spherical sampling schemes and non-rigid fluid based recordings for a reference volume. The normalized sampling scheme can allow the derivation of an average spatial density distribution of a population of samples, regardless of the format of the individual samples.
[00123] Processor 4 is further adapted to simulate a reference radiograph from this solid model. The processor is also adapted to create the solid model and/or simulate the reference radiography considering a predetermined relative spatial configuration of the image detector, the radiation source and the three-dimensional digitizer. Processor 4, e.g. the machine vision unit, can produce a reference projection or radiograph image of the filled reference object or product item using the measured and filled 3D format model and predetermined characteristics of the data source. radiation 2 and detector 3, using appropriate projection algorithms. Default characteristics may include polychromatic behavior of font and detector, position of font and detector, detector size and pixel size, font focal point size, etc. Suitable projection algorithms may include line tracing and integration algorithms to determine the amount of radiation energy that passes through the product item.
[00124] Processor 4 is also adapted to compare reference radiography, for example, obtained by simulation, with projection radiography, for example, obtained by measurement performed by the radiographic imaging system, to detect and/or measure internal deviations of the object in relation to the reference object.
[00125] This processor 4, for example the machine vision unit 4, can be further programmed to apply machine vision algorithms for the detection of defects in the imaged object. For example, processor 4 may comprise a computer or group of computers programmed to apply such machine vision algorithms. Images can be analyzed to extract information about possible defects by combining, such as by subtraction and/or analysis, the reference projection image with the actual image captured by the detector 3. Computer vision reconstruction and analysis can be addressed with appropriately scaled computing power and well-developed algorithms, as they are available in the state of the art and can be readily applied by the person skilled in the art. However, it should be noted that a convenient selection of suitable projection and defect detection algorithms can have an impact on signal-to-noise tolerances in the imaging chain. Therefore, it is an additional advantage of embodiments of the present invention that improved quality tolerances in reference radiography and/or detection can be used to increase the throughput of the system 50 without requiring extensive reconfiguration of the system 50.
[00126] For example, reference radiography can be compared to projection radiography, for example, with the use of an image difference metric, to detect an object that deviates from the sample population represented by the shape model and the distribution of gray value, for example, by applying a threshold criterion on the image difference metric. However, if an abnormality is then detected, additional machine vision algorithms can be engaged to further classify or determine the nature of the abnormality. It is an advantage that for an in-line conveyor system that feeds a stream of objects through imaging components that have a low defect rate, complex machine vision algorithms for classifying or determining the nature of defects need only be activated. infrequently. Thus, simple scanning can be provided in accordance with embodiments of the present invention which may not unduly and substantially impede the productivity of such an online system.
[00127] For example, the non-destructive inspection system 50 may be an object selection system or apparatus comprising a processor, for example, which has an associated memory for storing programs and data, from a computer or workstation. Such an object selection system or apparatus may further comprise a device that combines a 3D digitizer, for example using a laser line and/or multiple RGB cameras, and an X-ray radiography system. can be coupled to a processor input on the computer or workstation. In such an object selection system or apparatus, which combines X-ray radiographs, 3D scanning and processing, the X-ray radiography system can deliver pulses of exposure to the X-ray source to the object and can provide image data of the object. corresponding to such an exposure pulse, for example, by means of an energized (digital) detector, at the processor input. The processor may be powerable to acquire such a set of image data. The 3D digitizer can provide the light ray exposure to the object and can provide object digitization data corresponding to such exposure pulse, for example, via a power-up capable detector (digital), at the input of the processor, the which is powerable to acquire scan dataset. The object selection system or apparatus can be further characterized by the fact that the processor, when powered, adapts a shape model and a density model to the partial point clouds when captured by 3D detection systems. In one embodiment of the present invention, the processor of the system described above can process 3D scan data into a point cloud, estimate the external surface thereof by fitting a shape model (SM) to create a surface model of the object. , use a surface normalized gray value distribution to fill the surface model to produce a full volume image of a reference object conforming to the object's shape, for example, a perfect reference object that has no defects or abnormalities of the same or similar shape, simulate a radiograph from this solid model obtained by filling in the surface model, and analyze the differences between the radiographic image, for example, the measured radiograph of the object, and the simulated radiograph of the full-volume image perfect for calculating or displaying the object's internal defects.
[00128] It is an advantage of embodiments of the present invention that anomalous objects can be identified using a shape model and a gray value distribution as prior knowledge, so that a system according to embodiments of the present invention can be easily adapted to different object types by providing a suitable format model and gray value distribution as prior knowledge.
[00129] Additionally below in the present document, examples referring to embodiments of the present invention are provided. These examples are not to be construed as limiting the scope of the present invention in any way, but are merely provided for informational purposes, for example, to assist the person skilled in the art in understanding working principles of embodiments of the present invention and to assist in person skilled in the art in reducing the invention to practice.
[00130] In a first example, a training dataset is used to train the models used in modalities according to the present invention. This training dataset was acquired by taking CT scans of multiple defect-free samples in an industrial CT system. In this example, the model comprises two parts: an outer shape and an inner density distribution.
[00131] After the reconstruction of the CT scans, the external format of the scanned samples was extracted using basic image processing techniques such as segmentation, edge detection and pre- and post-processing, illustrated in Figure 5. The corresponding points , for example, reference points, on all these surfaces were extracted, where the variation in position of each point was determined thereafter. The result of this process is an average format with several modes of variation, which represents the variability of the format. Any shape in the population represented by the acquired samples can then be reconstructed through a linear combination of the mean shape and its modes of variation, for example, using a method as disclosed by Heimann et al., previously referred to above in the present document, and incorporated herein by reference. Figure 6 shows several random apple shapes generated with the method described above.
[00132] The inner side of the object was modeled with an average spatial density distribution derived from the CT dataset. This was done by performing a spherical sampling from the center of each scan, normalized to the surface of the sample, as illustrated in Figure 4. Such a normalized sampling scheme allows the derivation of the mean spatial density distribution of a population of samples, regardless of the format of the individual samples. This means that the average interior of all scanned samples is described, regardless of their format. This also means that any format obtainable by varying the parameters of the format model, as described above in this document, can be populated with this reference distribution by applying an inverted normalized spherical sampling scheme.
[00133] During runtime, the object may first pass through a 3D digitizer. This 3D digitizer can be based on a laser line, RGB cameras or any other system that produces a three-dimensional point cloud. The resulting point cloud may be incomplete, for example, due to the fact that such systems cannot commonly detect the underside of an object. In combination with the format model that was built during training, as described above in this document, it can, however, be used to approximate the complete external format of the object. Once this - hollow - shape is known, it can be combined with the mean spatial density distribution by reverse sampling this distribution, for example using an inverted normalized sampling scheme. The resulting "filled" shape may be an approximation of a CT scan of a perfect situation of the object that passed under the 3D scanner, for example, which has the same or very similar shape, but under the assumption that no defects are present in it. . An X-ray radiograph can be simulated by calculating a forward projection from this model. In a final step, the object passes through an X-ray radiography digitizer. The resulting "measured" radiograph is compared with the "simulated" one. Because the modeled object is perfect, for example without any defects, any differences between the modeled and measured projection indicate the presence of defects, thus allowing it to be classified as "defective" or "unsuitable". This runtime process is schematically illustrated in Figure 7. Additionally, the processing flow is schematically illustrated in Figure 8.
[00134] Figure 2 shows an optimization and a simulation in relation to the embodiments of the present invention. This illustration exemplifies a definition applied to a rigid cubic sample with a spherical abnormality with a density difference of 10%. The measured point cloud and the provided 3D model are used to assess the rotational and translational position of the sample. When combined with the prior knowledge density distribution, a perfect object situation is obtained, from which a radiograph is simulated. This is then compared to a measured radiograph to identify the abnormality.
[00135] Figure 8 shows a similar optimization and simulation with respect to embodiments of the present invention. This illustration shows a definition applied to a sample with a complex variable shape: an apple with a spherical abnormality with a 10% density difference from the surrounding tissue. The sample that passes through the definitions results in a measured point cloud and a measured radiograph. The deformable shape model is combined with the point cloud and average density distribution to result in a perfect sample situation. From this perfect situation, a radiograph is simulated, which is compared with the measured radiograph to identify the abnormality.
[00136] Another example to illustrate features and advantages of a method according to embodiments of the present invention is shown in Figure 9. Figure 9 shows two different shapes, respectively, an ellipsoid 91 and a toroid 92, combined with different types of defects. , for example, ellipsoidal, spherical and toroidal defects, with contrasting density. In the second row 93 of the Figure, the X-ray radiographs are shown from a random orientation of the objects. In the third row 94, the difference between the measured radiograph, for example shown in the second row 93, and the simulated radiograph obtained using the flawless 3D format, for example shown in the first row 95, is represented. Although it may be very difficult to distinguish defects on the measured radiographs, these defects may be readily visible on the comparison images, for example illustrated in the third row 94, for example, in such a way that image processing can be easily performed.
[00137] In this example, 1,250 samples with varying defect intensities and 1,250 samples without defects were processed by a method according to embodiments of the present invention. A simple 10-fold cross Bayesian classification using the simple sum of all pixel values in the comparison images can correctly identify 97.4% of the samples as having a defect or not. However, to achieve the same result according to methods known in the art, separate image processing algorithms, e.g. including thresholds, can be developed for each format-defect combination to detect defects in the measured radiographs.

权利要求:
Claims (19)
[0001]
1. NON-DESTRUCTIVE INSPECTION METHOD (10) FOR IN-LINE INSPECTION OF AN OBJECT, characterized by comprising: - moving (11), with the use of an in-line transport system, an object, which is at least partially transparent to radiation of a predetermined quality of radiation, along a predetermined path between a stationary radiation source, for emitting radiation of said predetermined radiation quality, and a stationary image detector, and through a field of view of a three-dimensional scanner; - imaging (12) said object using the image detector, detecting said radiation emitted by the radiation source and transmitted through said object to obtain a projection radiograph of an internal structure of said object; - scanning (13) an external surface of said object using the three-dimensional scanner to obtain three-dimensional digitizing data of said object in the form of a point cloud representative of at least part of said external surface; - adjusting (14), using a processor, a deformable shape model of said object to said point cloud to obtain a surface model of said external surface; - creating (15), using said processor, a solid model of said surface model considering a gray value distribution of a reference object; - simulating (16), using said processor, a reference radiograph from said solid model; and - comparing (17), with the use of said processor, said reference radiography with said projection radiography to detect and/or measure the internal deviations of said object in relation to the reference object, wherein said three-dimensional digitizer is a 3D digitizer selected from the group consisting of a laser line digitizer, multiple RGB cameras, an optical 3D digitizer, an active non-contact digitizer using light or ultrasound emissions, a time-of-flight 3D laser digitizer, a triangulation-based 3D laser scanner, a conoscopic holographic laser scanner, a structured light 3D scanner, a modulated light 3D scanner, a stereoscopic optical imaging system, a photometric imaging system, a laser line in combination with a RGB camera system and a 3D silhouette imaging scanner, wherein said step of creating (15) the solid model and/or said step of simulating (16) said radiograph reference considers a predetermined relative spatial configuration of said image detector, said radiation source and said three-dimensional digitizer.
[0002]
2. METHOD, according to claim 1, characterized in that said scanning (13) of said external surface comprises generating a partial point cloud of said object, wherein said adjustment (14) comprises estimating the complete external surface and the position of the object by fitting said format model, wherein said creation (15) of the solid model comprises filling a volume defined by said surface model with said gray value distribution, wherein said volume corresponds to the spatial coordinates of the object when imaged by the image detector, and wherein said simulation (16) of said reference radiograph comprises simulating an imaging process of the solid model by forward projection using said predetermined spatial configuration of the image detector and the radiation source in said spatial coordinates.
[0003]
3. METHOD according to any one of claims 1 or 2, characterized in that said format model and said gray value distribution are created (21) by acquiring (22) CT scans of a plurality of object samples from off-line reference and determining the format model (23) and the gray value distribution (24) of said CT scans to be used as background knowledge during said steps of adjusting (14) the format model and creating (15) the solid model at runtime.
[0004]
4. METHOD, according to claim 3, characterized in that the determination (23) of said format model comprises: - extracting (25) a plurality of external surfaces, corresponding to said plurality of reference object samples, from said CT scans using image processing techniques, - determining (26) a plurality of corresponding spatial features on said plurality of external surfaces, - determining (27) an average position and/or a variation in the position of each of said plurality of corresponding spatial characteristics, and - determining (23) the shape model considering said average positions, wherein said shape model has free parameters that can be adjusted to explain the object's position, the object's orientation and/or the modes of object variation representative of said variations in position of said plurality of corresponding spatial features.
[0005]
5. METHOD, according to claim 4, characterized in that said format model comprises a linear model that parameterizes an object format as a linear combination of an average format and a plurality of variation modes.
[0006]
METHOD according to any one of claims 3 to 5, characterized in that said gray value distribution is determined (24) as a normalized reference volume image of said plurality of CT scans.
[0007]
7. METHOD, according to claim 6, characterized by determining (24) said gray value distribution, further comprising applying a surface normalization scheme to derive the normalized reference volume image from a represented population of objects. by said object samples irrespective of the shape of any individual object.
[0008]
METHOD, according to claim 7, characterized in that the determination (24) of said gray value distribution comprises applying (28) a normalized spherical sampling scheme to obtain the normalized reference volume image and in which to create (15) ) said solid model comprises applying (29) an inverted normalized spherical sampling scheme to fit the normalized reference volume image to said surface model.
[0009]
9. METHOD according to any one of claims 1 to 8, characterized in that said movement (11) of said object comprises moving said object in line on a conveyor belt.
[0010]
10. METHOD, according to any one of claims 1 to 9, characterized in that said movement (11) of said object moves said object at a speed in the range of 0.1 m/s to 0.7 m/s.
[0011]
11. METHOD according to any one of claims 1 to 10, characterized in that, during said movement (11) of said object along said predetermined path, said object first passes through said three-dimensional digitizer, and then, consequently pass through the field of view of the image detector.
[0012]
METHOD according to any one of claims 1 to 11, characterized in that said object is moving (11) in line through a radiation field of each of a plurality of radiographic imaging systems, each comprising an ionizing radiation source and an image detector, and through a digitization stage of at least one three-dimensional digitizer.
[0013]
13. METHOD according to any one of claims 1 to 12, characterized in that said radiation source and said image detector are statically arranged with respect to said in-line transport system.
[0014]
14. METHOD, according to claim 13, characterized in that the radiation source and the image detector are fixed above a conveyor belt on which said object is transported.
[0015]
15. NON-DESTRUCTIVE INSPECTION SYSTEM (50) FOR ONLINE INSPECTION OF AN OBJECT (6), characterized by comprising: - A radiation source (2) and an image detector (3) that form a radiographic imaging system to detect radiation emitted by the radiation source and transmitted through an object (6) which is at least partially transparent to said radiation, to provide a projection radiograph of an internal structure of said object; - a three-dimensional scanner (1) for scanning an external surface of said object (6) to obtain three-dimensional scanning data of said object in the form of a point cloud representative of at least part of said external surface; - an in-line transport system (5) for moving the object (6) along a predetermined path between the radiation source (2) and the image detector (3) and through a field of view of the three-dimensional digitizer ( 1); and - a processor (4) adapted to: - obtain said projection radiograph from said image detector, - obtain said point cloud from said three-dimensional digitizer, - adjust a deformable format model of said object to the said point cloud to obtain a surface model of said external surface, - create a solid model of said surface model considering a gray value distribution of a reference object, - simulate a reference radiograph from said solid model; and - comparing said reference radiography with said projection radiography to detect and/or measure the internal deviations of said object in relation to the reference object, wherein the processor is further adapted to create said solid model and/or or simulating said reference radiography considering a predetermined relative spatial configuration of said image detector, said radiation source and said three-dimensional digitizer, wherein said three-dimensional digitizer is a 3D digitizer selected from the group consisting of a digitizer line laser scanner, multiple RGB cameras, an optical 3D scanner, an active non-contact scanner using light or ultrasound emissions, a time-of-flight 3D laser scanner, a triangulation-based 3D laser scanner, a conoscopic holographic laser, a structured 3D light scanner, a modulated 3D light scanner, a stereoscopic optical imaging system, a photometric imaging, a laser line in combination with an RGB camera system and a silhouette imaging 3D digitizer, wherein said radiation source (2) is a stationary radiation source and said image detector (3) ) is a stationary image detector.
[0016]
16. SYSTEM, according to claim 15, characterized in that it further comprises a plurality of stationary radiation sources (2) and stationary image detectors (3) that form a plurality of radiographic imaging systems.
[0017]
17. SYSTEM, according to any one of claims 15 or 16, characterized in that said three-dimensional digitizer (1) comprises a laser or stationary light source and a stationary light detector.
[0018]
18. SYSTEM according to any one of claims 15 to 17, characterized in that said radiation source (2) is adapted to provide an X-ray exposure pulse to said object (6) and by said image detector (3) being a digital image detector adapted to provide image data of said object corresponding to said X-ray exposure pulse as an input to said processor, and wherein said three-dimensional digitizer (1) is adapted to provide X-ray exposure. light to said object and comprises a digital detector for providing data related to said object obtained by exposing said light rays as an input to said processor.
[0019]
19. USE OF A METHOD, as defined in any one of claims 1 to 14, characterized in that it is to identify anomalous objects transported on a conveyor line in an industrial process.

类似技术:

---

公开号 | 公开日 | 专利标题

---

BR112017019834B1|2022-02-01|Non-destructive inspection method and system for in-line inspection of an object; and using a method

---

CN105784731B|2019-02-22|Mesh calibration method and safe examination system in a kind of positioning three-dimensional CT image

---

US7424141B2|2008-09-09|System and method for performing auto-focused tomosynthesis

---

US20050078861A1|2005-04-14|Tomographic system and method for iteratively processing two-dimensional image data for reconstructing three-dimensional image data

---

US10825165B2|2020-11-03|Inspection method for a manufactured article and system for performing same

---

US20140010437A1|2014-01-09|Compound object separation

---

JP2015520846A|2015-07-23|Method and System for Determining Z-Effective Value of Voxel Set Using CT Density Image and Sparse Multi-Energy Data

---

US8670522B2|2014-03-11|Stereo X-ray inspection apparatus and method for forming three-dimensional image through volume reconstruction of image acquired from the same

---

EP3290912A1|2018-03-07|Examination system for inspection and quarantine and method thereof

---

CN105849772B|2019-06-11|Check system and method

---

EP3187904A1|2017-07-05|Inspection devices for quarantine

---

US8411814B2|2013-04-02|Method of and apparatus for measuring center detector index on CT scanner with limited field of view access

---

US10586324B2|2020-03-10|Inspection devices and methods for inspecting a container

---

KR102045079B1|2019-11-14|Inspection apparatus using terahertz wave

---

JP2017510364A|2017-04-13|Patient table with integrated X-ray volume imaging device

---

US10009593B2|2018-06-26|Device and method for obtaining densitometric images of objects by a combination of radiological systems and depth-sensing cameras

---

Kehoe1990|Detection and evaluation of defects in industrial images

---

KR20210049086A|2021-05-04|Article inspection by dynamic selection of projection angle

---

CN113167746A|2021-07-23|Dynamic radiation collimation for non-destructive analysis of test objects

---

CN107771058A|2018-03-06|The gap resolution ratio of linear detector arrays

同族专利:

---

公开号 | 公开日

---

WO2016146703A1|2016-09-22|

---

EP3271717B1|2019-11-13|

---

CA2979932A1|2016-09-22|

---

NZ735513A|2020-07-31|

---

US20180113083A1|2018-04-26|

---

PL3271717T3|2020-06-01|

---

PT3271717T|2020-02-20|

---

EP3271717A1|2018-01-24|

---

ZA201706920B|2019-02-27|

---

GB201504360D0|2015-04-29|

---

ES2772698T3|2020-07-08|

---

BR112017019834A2|2018-05-29|

---

US10520452B2|2019-12-31|

---

CL2017002353A1|2018-07-13|

---

CA2979932C|2021-01-05|

引用文献:

---

公开号 | 申请日 | 公开日 | 申请人 | 专利标题

---

---

US5848115A|1997-05-02|1998-12-08|General Electric Company|Computed tomography metrology|

---

US7602963B2|2006-01-10|2009-10-13|General Electric Company|Method and apparatus for finding anomalies in finished parts and/or assemblies|

---

EP2396646B1|2009-02-10|2016-02-10|Optosecurity Inc.|Method and system for performing x-ray inspection of a product at a security checkpoint using simulation|US10282499B2|2015-12-18|2019-05-07|Triple Ring Technologies, Inc.|Method and apparatus for x-ray ionizing radiation control|

---

GB2550117A|2016-05-04|2017-11-15|E M & ILtd|Inspection method|

---

EP3519764B1|2016-09-29|2020-07-29|Marel Iceland EHF|A method of generating a three dimensional surface profile of a food object|

---

CN107085001B|2017-04-19|2019-10-25|天津博迈科海洋工程有限公司|Extensive process pipeline defect detection method|

---

US10529082B2|2017-06-20|2020-01-07|Mitutoyo Corporation|Three-dimensional geometry measurement apparatus and three-dimensional geometry measurement method|

---

US10460512B2|2017-11-07|2019-10-29|Microsoft Technology Licensing, Llc|3D skeletonization using truncated epipolar lines|

---

US11049236B2|2017-11-17|2021-06-29|Kodak Alaris Inc.|Automated in-line object inspection|

---

CN109377487A|2018-10-16|2019-02-22|浙江大学|A kind of fruit surface defect detection method based on deep learning segmentation|

---

ES2847236R1|2020-01-11|2021-08-03|Quandum Aerospace S L|CAMERA AND LIGHTS POSITIONING SYSTEM FOR INSPECTION OF HOSES USED IN AIR REFUELING AND INSPECTION PROCEDURE|

---

EP3954985A1|2020-08-04|2022-02-16|Biometic S.r.l.|Industrial tomography apparatus and method for checking the composition of industrial products which comprise a plurality of primary parts|

---

US11232554B1|2021-06-07|2022-01-25|Elementary Robotics, Inc.|Machine-learning based camera image triggering for quality assurance inspection processes|

法律状态:
2019-12-17| B06U| Preliminary requirement: requests with searches performed by other patent offices: procedure suspended [chapter 6.21 patent gazette]|

---

2020-06-02| B25A| Requested transfer of rights approved|Owner name: KATHOLIEKE UNIVERSITEIT LEUVEN (BE) ; UNIVERSITEIT ANTWERPEN (BE) |

---

2021-12-07| B09A| Decision: intention to grant [chapter 9.1 patent gazette]|

---

2022-02-01| 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 16/03/2016, OBSERVADAS AS CONDICOES LEGAIS. |

优先权:

---

申请号 | 申请日 | 专利标题

---

GBGB1504360.7A|GB201504360D0|2015-03-16|2015-03-16|Automated quality control and selection system|

---

GB1504360.7|2015-03-16|

---

PCT/EP2016/055718|WO2016146703A1|2015-03-16|2016-03-16|Automated quality control and selection|

---