• 专利附录
  • 专利说明
  • 权利要求
  • 类似技术
  • 同族专利
  • 引用文献
  • 法律状态
  • 优先权
    • 专利摘要:
    A control device (10) for an electrochemical system (20). The device includes means for measuring electrical parameters of an electrochemical cell (7) using the associated electrochemical noise signal; It also includes processing means to compare the electrochemical noise signal obtained with a plurality of noise patterns. Each noise pattern characterizes a phenomenon that affects the operation of the electrochemical cell (7). The processing means can determine, in real time, if a phenomenon occurs that affects the electrochemical cell (7) according to comparisons made with noise patterns. (Machine-translation by Google Translate, not legally binding)
    公开号:ES2736012A1
    申请号:ES201830574
    申请日:2018-06-13
    公开日:2019-12-23
    发明作者:Gonzalez Miguel Angel Rubio；Moraleda Alfonso Urquia
    申请人:Universidad Nacional de Educacion a Distancia UNED；
    IPC主号:
    专利说明:

    [0001]
    [0002] Technical Field of the Invention
    [0003]
    [0004] The present invention belongs to the field of electrochemical systems. In particular, it refers to a device that allows monitoring the operation of said systems.
    [0005]
    [0006] Background of the invention
    [0007]
    [0008] The systems of analysis of active electrochemical cells (galvanic batteries, fuel cells, flow batteries and electrolysers among others) developed to date consist of laboratory systems, heavy, expensive and with a limitation in the set of phenomena they detect. Its use requires that the electrochemical cell stop its operation to perform the necessary measurements. In many of the systems, the phenomena detected are characterized by a significant drop in performance, when the deterioration produced in the cell is irreversible.
    [0009] The known developed systems have operating flexibility deficits, due to the limitation in the set of identifiable phenomena.
    [0010] The most common and versatile of the systems used for identification is the Electric Impedance Spectroscopy (EIE) analyzer. Through this system it is possible to identify phenomena associated with the variation in the electrical impedance of the cell. The EIE analysis system injects electrical signals of a single frequency into the cell at a time and sweeps the entire range to be analyzed. This process can be especially long if you want to study the impedance at low frequencies.
    [0011] Another device commonly used for the study of electrochemical cells is cyclic voltammetry (VC) equipment. Like EIE devices, VC devices are heavy, expensive and force the operation of the cells to stop.
    [0012] Therefore, it would be desirable to have a device to analyze the operation of an electrochemical system in real time without stopping it.
    [0013] Brief Description of the Invention
    [0014]
    [0015] The present invention relates to a device based on the analysis of electrochemical noise (ARE) as defined in claim 1. The proposed device allows to identify phenomena that affect the electrochemical cell through the variation in the amplitude of electrochemical noise in different bands of frequency. It is possible to determine even if a harmful phenomenon occurs for the cell while it is in operation.
    [0016] In this document, an electrochemical system is any system in which one or more electrochemical cells intervene. Electrochemical cell means any device that produces a conversion between chemical energy and electrical energy.
    [0017] It is an object of the invention, to identify early phenomena that may affect the operation of the electrochemical system before irreversible damage occurs or at least minimizing its effect.
    [0018] It is also an object of the invention to identify system operating points without the need for additional instrumentation.
    [0019] It is another object of the invention to implement the invention in a portable and low cost device.
    [0020] Electrochemical noise is characterized by being a very small signal (of the order of pV) superimposed on the operating voltage of the electrochemical cell (if the cell operates in a galvanostatic mode, constant current). Similarly, electrochemical noise is manifested with a signal of small current amplitude (if the cell operates in potentiostatic mode, constant voltage).
    [0021] In both cases, it has been proven that electrochemical noise is sensitive to the state of the cell and shows particular characteristics depending on different phenomena that affect the operation of the electrochemical cell.
    [0022] Thus, the invention states that it is possible to identify various phenomena that affect the electrochemical cell by analyzing the different frequency components of electrochemical noise.
    [0023] Non-operational phenomena:
    [0024] - Catalyst and electrolyte poisoning by exposure to external or internal cell agents.
    [0025] - Loss of conductivity due to deformations of the mechanical components, the partial or total breakage of the electrolyte or the corrosion of conductive metal parts. - Change of the physical-chemical properties of the components by high pressure and flow, and extreme temperature (high or very low).
    [0026] - Degradation of cell elements due to its use, such as conductivity of polymers or loss of catalyst mass by dragging.
    [0027] - Transfer of species between anode and cathode, without mediating the electrochemical reaction. Operational phenomena:
    [0028] - Pressure, flow, inlet humidity, temperature, current (or voltage), among others: - Lack of fuel, inadequate humidification or any operation with an operating configuration that causes a cell behavior away from its optimal operation (causing the high overpotential).
    [0029] - Cell operation parameters that modify the operation of the cell, but produce a cell behavior close to the optimal operating range (causing the small overpotential), not causing irreversible damage to its components.
    [0030] The invention, in order to identify those phenomena that may affect the electrochemical cell, associates each phenomenon with a specific pattern that makes it differentiable from other causes.
    [0031] The technique preferably used to classify these patterns is artificial neural networks, however, the classification task must be performed by other supervised classification algorithms, such as support vector machines, naive Bayes classifier and linear discriminant analysis. , among others. The advantages of neural networks are high flexibility and robustness, and their ability to find solutions to nonlinear problems. Also, the simple implementation of the network trained in a decision-making system and its execution with small computational cost.
    [0032] As mentioned above, the invention can be advantageously implemented in a portable device, since it does not require the use of power electronics or electrical charges. It is also achieved that the invention can be used in operation of the cell, since the electrochemical noise is superimposed on the cell voltage, unlike other proposals of the prior art.
    [0033] Advantageously, embodiments of the device object of the invention can be manufactured with lower economic costs.
    [0034]
    [0035] Brief description of the figures
    [0036]
    [0037] FIG. 1 shows several components of an embodiment of the control device.
    [0038] FIG. 2 shows an example of a training stage for the identification of phenomena that affect the operation of the cell.
    [0039] FIG. 3 shows an example of an operation stage for the identification of phenomena that affect the operation of the cell.
    [0040] FIG. 4 is a schematic of an embodiment of the multi-channel control device.
    [0041] FIG. 5 shows the signal acquired by the data acquisition card for three different phenomena.
    [0042] FIG. 6 shows the frequency decomposition by FFT, performed by the signal analyzer of an experiment of each of the analyzed phenomena. FIG. 7 shows the signal acquired by the acquisition card when phenomenon 2 occurs and the instant of time when it is identified.
    [0043] FIG. 8 is an example of a neural network.
    [0044]
    [0045] Detailed description of the invention
    [0046]
    [0047] With reference to the preceding figures, various embodiments of the device object of the invention are described.
    [0048]
    [0049] In FIG. 1, several components of an embodiment are schematically illustrated. Specifically, a high precision data acquisition card 1 is shown (to measure the amplitude of the noise). The fundamental element of the data acquisition card 1 is the analog-digital converter, which allows the analog signal to be converted into a digital signal. A digital signal analyzer 2 is included to calculate the frequency noise component, for example, by real-time Fourier or Wavelets analysis. A data processor 3 to classify the phenomenon produced (for example, classification through an already trained neural network), monitor and send the control and alert signals.
    [0050] Preferably, the data acquisition card 1 should convert the analog signal of the cell voltage into a digital signal with a resolution of <10pV over a 2.5V signal of maximum amplitude (this voltage could be higher if the electrochemical cell 7 to be studied) It has a higher operating voltage). This implies the use of a resolution equal to or greater than 18 bits.
    [0051] It is possible, with the aim of reducing costs, acquisition cards using fewer bits, using the technique of oversampling or sobresampleo (oversampling). The cost of the cards increases with the number of bits of signal resolution. The sampling frequency required to analyze electrochemical noise is very low compared to the sampling frequency that can be used in low-cost acquisition cards (usually of the order 100KHz). To perform over-employment it is necessary to use a sampling frequency higher than the highest frequency that you want to analyze. In this way, it is possible to obtain additional resolution bits according to the following expression:
    [0052] N fs = 22n
    [0053] That is, N times the sampling frequency is equal to 2 times the number n of additional bits.
    [0054] For example, in the case of using an original sampling frequency of 10KHz and 16 bits of resolution, to obtain an additional 3 bits of resolution, a sampling frequency of 2A (2 * 3) = 32 times the sampling frequency can be used original, this is 320KHz. Subsequently the signal frequency is reduced, averaging the intermediate values. In the case of the example, the electrochemical noise signal is reduced in size again, reducing its frequency to 10 KHz, where each point of the signal is averaged with the 32 adjacent points of the signal sampled at 320KHz. In the case of over-employment, this task will be performed by the signal analyzer 2 prior to the decomposition of the signal into frequencies.
    [0055] The acquisition frequency of the data acquisition card 1 may be less than 10KHz, since the phenomena of interest in cells 7 manifest themselves at frequencies below 5KHz (half of 10KHz, Nyquist sampling frequency). The large amount of information acquired by the acquisition card 1 must be processed in real time by a processor 3. The processor executes a series of steps including an algorithm for calculating FFT (Fast Fourier Transform) with time window or wavelets Real time. The digital signal analyzer 2 processes the signal and can be implemented in particular by means of a DSP (Digital Signal Processor) or an FPGA (Field Programmable Gate Array), since they are dedicated devices, suitable for real-time signal processing. The use of ARM (Advanced RISC Machine) processors is limited to the processing capacity of these devices.
    [0056] The frequency decomposition of the noise signal performed by the signal analyzer 2 must be processed by a data processor 3 that classifies the phenomena produced and performs control tasks of the entire process. The process of classification and generation of control signal does not have a high computational cost. For this reason, it is sufficient to use a low-cost ARM device.
    [0057]
    [0058] In FIG. 2 show the training mode on the device, this mode that includes two differentiated stages. A first stage of acquisition 31 and another stage of training 32.
    [0059] In the first acquisition stage 31, the phenomenon 9 (eg fault) that is desired to be classified is reproduced, for this purpose the electrochemical noise signal is obtained with a data acquisition card 1, the subsequent decomposition into frequency components (frequencies and amplitudes) with the signal analyzer 2 and its storage in a register 3 containing the information generated after reproducing said phenomenon 9. These steps must be repeated for each of the phenomena to be identified.
    [0060] A training stage 32 of a neural network 5 implemented with a processor 2. In this stage all the registers 3 are grouped in a memory 4 with the previously stored data of the various phenomena 9 reproduced during the data acquisition stage 31. The Second training stage 32 is therefore carried out with the neural network 5 and with the data stored in memory 4 obtained in the previous phase 31. From this, a previously trained neural network 5 is finally obtained that allows different types of phenomena to be properly classified 9 that occur in the electrochemical cell 7.
    [0061] Optionally, other signals such as the voltage and current of the cell can be incorporated into the data stored in the memory 4, to improve the training of the neural network 5.
    [0062]
    [0063] In FIG. 3 the device is shown in operation mode for fault diagnosis. Once the neural network 5 is properly trained, the signal from cell 7 is acquired, decomposes it into frequencies and classifies the phenomenon produced. The result of the classification can be used in a decision-making system that performs the appropriate operations depending on the identified phenomenon 9. For example, the decision-making system may order the electrochemical system to shut down if it finds a dangerous failure or notify with a warning that any of the components of the electrochemical system require maintenance.
    [0064]
    [0065] In FIG. 4 shows a multichannel scheme. Since many commercial electrochemical systems are formed by several electrochemical cells 7 in series to increase the operating voltage, some embodiments of the device object of this invention may include several equal channels 8 as number of cells 7 are in the electrochemical system to be controlled. The control device works with the signals in parallel through first channels 8 of the data acquisition card 1 and in a few second channels 9 of the signal analyzer 2. The data processor 3 since the signal is already compressed to the make a Frequency decomposition does not require such a large signal bandwidth and can use serial communication.
    [0066] Optionally, in order to reduce costs, it is possible to use the time division multiplexing technique, TDM (Time Division Multiplexing), using a single data acquisition card 1 of a single channel 8, for several electrochemical cells 7.
    [0067]
    [0068] Operating example
    [0069]
    [0070] To exemplify the use of the device, the complete procedure to identify three different phenomena is shown, using the FFT algorithm to perform frequency decomposition.
    [0071]
    [0072] Training mode
    [0073]
    [0074] The first phenomenon 9 to study is provoked in the cell. The data acquisition card 1 obtains the voltage signal. This task is performed for each of the phenomena 9 (phenomena A, B, C), thus the corresponding noise signals are obtained, as shown in FIG. 5. The training noise signals in this example of a duration of 300 seconds, obtained by the acquisition card 1, of each of the phenomena, is cut into windows of 3 seconds, called experiments in this document, obtaining a total of 100 experiments for each different phenomenon. All these noise signals are sent to signal analyzer 2.
    [0075]
    [0076] In FIG. 5 experiment 51 of the first phenomenon is shown. The signal analyzer 2 breaks down the noise signals of each experiment into its frequency components.
    [0077]
    [0078] In FIG. 6 shows the decomposition in frequencies of a noise signal of an experiment for each type of phenomenon. In this example, FFT is used to perform frequency signal decomposition.
    [0079]
    [0080] In FIG. 8 shows the inputs 11 to the hidden layer 12 of the neural network 5, which in this case are the amplitudes of the FFT of the noise signal at frequencies 0.33, 0.5, 0.83, 1.42, 2, 3.33, 5, 10 , 25, 33.33, 50, 66.66, 125 Hz of each experiment 9. The outputs 13 of the neural network 5, as shown in this FIG. 8, correspond in this case to the binary output of the three states to classify. The value "1" in one of the outputs 13 of the neural network 5, indicates that the inputs 11 correspond to the frequency pattern of the noise when said phenomenon has occurred. The value "0" at the output of a phenomenon 9 indicates that the inputs of the network do not correspond to the noise signal associated with said phenomenon 9. Optionally, 5 averages values of the neural network can be used as inputs 11 to the neural network. amplitude of the FFT in frequency bands.
    [0081] The training of the neural network 5 employs a set of noise signals from all experiments of all phenomena 9 to perform the adjustment of the neural network 5 and a different set, to perform the validation of the adjustment. In this case, 50% of randomly selected experiments out of the total 300 are selected to perform the adjustment process of the neural network 5 (150 experiments) and the other 50% to perform the validation process (150 experiments).
    [0082] If the validation shows a success of the classification that is considered adequate, the neural network is properly trained. The neural network 5 in training mode modifies the internal parameters of the network based on inputs 11 and outputs 13. Once the neural network 5 is trained, the parameters of network 5 remain fixed to be used in the operation mode
    [0083] Mention that the neural network 5, as well as any other classification procedure, improves its classification capacity the greater the number of experiments used for its training.
    [0084]
    [0085] Operation mode
    [0086]
    [0087] Once the neural network training has been carried out, the data acquisition card obtains data for 3 seconds that is sent to the signal analyzer 2 for decomposition into its frequency components. The analyzed frequencies obtained in the previous step must be the same as the frequency components that have been used as inputs in the training mode of the neural network 5.
    [0088] If any of the three phenomena occurs, the output of the neural network 5 trained in training mode will indicate that the phenomenon 9 indicating the output of the network has occurred. The structure of the neural network 5 in training mode and in the operating mode is the same, as can be seen in FIG. 8.
    [0089] If, for example, phenomenon B occurs in an instant of time, as can be seen in FIG. 7, the data acquired by the acquisition card 1 of the following experiment after the occurrence of phenomenon B, produces a variation in the frequency components of the signal (performed by the signal analyzer 2), which after being sent to the neural network 5, will indicate that the phenomenon B has occurred.
    权利要求:
    Claims (13)
    [1]
    1. Control device (10) for an electrochemical system (20) comprising: - means for measuring electrical parameters of an electrochemical cell (7) in real time configured to obtain an associated electrochemical noise signal; - processing means configured to compare the electrochemical noise signal with a plurality of noise patterns, with each pattern representative of a phenomenon (9) affecting the operation of the electrochemical cell (7), the processing means being configured to determine , with the electrochemical cell in operation, the occurrence of a specific phenomenon (9), according to the degree of correspondence between the electrochemical noise signal and each noise pattern.
    [2]
    2. Control device (10) according to claim 1, wherein the measuring means comprises:
    - a data acquisition card (1) configured to convert the electrochemical noise signal into a digital noise signal, and;
    - a signal analyzer (2) configured to frequently decompose the digital noise signal.
    [3]
    3. Control device (10) according to claim 2, wherein the signal analyzer (2) is configured to perform a spectral decomposition of the noise signal to obtain a plurality of associated frequencies and amplitudes.
    [4]
    4. Control device (10) according to claim 3, wherein the spectral decomposition of the noise signal is performed by implementing the Fourier fast transform, FFT.
    [5]
    5. Control device (10) according to claim 3, wherein the spectral decomposition of the noise signal is performed by implementing wavelets.
    [6]
    6. Control device (10) according to claim 4 or 5, wherein the processing means comprise a processor (3) that implements a neural network (5) to classify the phenomenon (9) associated with the electrochemical noise signal.
    [7]
    7. Control device (10) according to claim 6, wherein the processing means further comprises a memory (4) that stores a plurality of records (3), each record (3) containing data from experiments associated with a phenomenon ( 9) concrete.
    [8]
    8. Control device (10) according to claim 7, wherein the neural network (5) is trained with inputs (11) comprising values of frequency and amplitude of the noise signal.
    [9]
    9. Control device (10) according to claim 8, wherein the neural network (5) additionally receives as inputs (11) for training, the averages of the amplitude of the FFT of the noise signal in frequency bands.
    [10]
    10. Control device (10) according to claim 8 or 9, wherein the neural network (5) additionally receives as inputs (11) for training, the standard deviation of the wavelet coefficients associated with each frequency.
    [11]
    11. Control device (10) according to any one of claims 6 to 10, wherein the processor (3) is further configured to send a control signal to modify the operation of the electrochemical cell (7) according to the classification of the phenomenon (9) concrete.
    [12]
    12. Control device (10) according to any one of the preceding claims, wherein said device is portable.
    [13]
    13. Control device (10) according to any one of the preceding claims, wherein the data acquisition card (1) implements the time division multiplexing technique, TDM, to acquire data from a plurality of electrochemical cells (7) .
    类似技术:
    公开号 | 公开日 | 专利标题
    Rubio et al.2016|Proton exchange membrane fuel cell failure mode early diagnosis with wavelet analysis of electrochemical noise
    Steiner et al.2011|Non intrusive diagnosis of polymer electrolyte fuel cells by wavelet packet transform
    Damour et al.2015|Polymer electrolyte membrane fuel cell fault diagnosis based on empirical mode decomposition
    JP5632572B2|2014-11-26|Measuring instrument and method for obtaining power statistics
    Baker et al.2010|A frequency-based RF partial discharge detector for low-power wireless sensing
    Husar et al.2012|Experimental characterization methodology for the identification of voltage losses of PEMFC: Applied to an open cathode stack
    Astafev et al.2018|Measurement of electrochemical noise of a Li/MnO2 primary lithium battery
    US20050218915A1|2005-10-06|Method and apparatus for battery testing
    WO2018120633A1|2018-07-05|Method for detecting sensor, and apparatus for detecting sensor
    GB2483108A|2012-02-29|Monitoring neurological electrical signals to detect the onset of a neurological episode
    CN105205736A|2015-12-30|Rapid detection method for power load abnormal data based on empirical mode decomposition
    Lee et al.2008|Chaos in air pollutant concentration | time series
    KR101376910B1|2014-03-26|Method for detecting trouble of solar cell module and apparatus thereof
    CN110573869A|2019-12-13|System and method for evaluating toxic gas sensors using electrochemical impedance spectroscopy
    ES2736012B2|2020-09-22|CONTROL DEVICE FOR AN ELECTROCHEMICAL SYSTEM
    Manganiello et al.2017|FPGA implementation of the EIS technique for the on-line diagnosis of fuel-cell systems
    Boškoski et al.2014|Optimal selection of proton exchange membrane fuel cell condition monitoring thresholds
    CN105021888B|2018-09-18|Harmonic data monitoring method based on data clusters
    Astafev2020|The measurement of electrochemical noise of a Li-ion battery during charge-discharge cycling
    Fouquet2010|Real time model-based monitoring of a pem fuel cell flooding and drying out
    Kim et al.2015|Autocorrelation standard deviation and root mean square frequency analysis of polymer electrolyte membrane fuel cell to monitor for hydrogen and air undersupply
    Klyuev et al.2019|Analysis of discrete spectra of electrochemical noise of lithium power sources
    Chen et al.2014|Performance measurement in wireless sensor networks using time-frequency analysis and neural networks
    CN108333443B|2020-08-25|Method for alarming intermittent defects of power equipment
    CN110441584A|2019-11-12|Equipment running status monitoring method, device, storage medium and system
    同族专利:
    公开号 | 公开日
    ES2736012B2|2020-09-22|
    引用文献:
    公开号 | 申请日 | 公开日 | 申请人 | 专利标题
    法律状态:
    2019-12-23| BA2A| Patent application published|Ref document number: 2736012 Country of ref document: ES Kind code of ref document: A1 Effective date: 20191223 |
    2020-09-22| FG2A| Definitive protection|Ref document number: 2736012 Country of ref document: ES Kind code of ref document: B2 Effective date: 20200922 |
    优先权:
    申请号 | 申请日 | 专利标题
    ES201830574A|ES2736012B2|2018-06-13|2018-06-13|CONTROL DEVICE FOR AN ELECTROCHEMICAL SYSTEM|ES201830574A| ES2736012B2|2018-06-13|2018-06-13|CONTROL DEVICE FOR AN ELECTROCHEMICAL SYSTEM|
    [返回顶部]