A hydrogen leakage monitoring method, monitoring device and monitoring system for a fuel cell

Abstract

The application discloses a hydrogen leakage monitoring method, a monitoring device and a monitoring system of a fuel cell, wherein the hydrogen leakage monitoring method comprises the following steps: step S100, acquiring an environmental sound sample of the fuel cell; step S200, extracting first feature data of the environmental sound sample; step S300, inputting the first feature data into a preset model to acquire one-time judgment information of whether the fuel cell has hydrogen leakage; step S400, performing time-domain to frequency-domain conversion on the environmental sound sample to acquire frequency domain data; step S500, extracting second feature data in the frequency domain data; step S600, comparing the second feature data with set data, and outputting two-time judgment information of whether the fuel cell has hydrogen leakage according to a comparison result; step S700, comparing whether the one-time judgment information and the two-time judgment information are consistent; and step S800, outputting a result. The above scheme processes the environmental sound sample to judge whether hydrogen leakage occurs, has a large coverage range and high accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fuel cells, in particular to a hydrogen leakage monitoring method, monitoring device and monitoring system of a fuel cell. BACKGROUND

[0002] Hydrogen fuel cell vehicles mainly use high-pressure gaseous / liquid hydrogen storage, and the storage pressure is mainly 35MPa and 70MPa. Because the relative molecular mass of hydrogen is small, it is easy to leak and permeate. After hydrogen leakage and accumulation, it is easy to cause safety accidents such as combustion and explosion.

[0003] At present, the monitoring of hydrogen leakage mainly monitors the concentration. Concentration sensors are usually arranged in potential leakage areas to detect the hydrogen concentration in the environment, so as to judge whether a hydrogen leakage accident has occurred. However, the connection points of the hydrogen pipeline in the vehicle are many, and the arrangement of the components is relatively dispersed. Therefore, a large number of concentration sensors need to be configured, but there is often a problem of uneven arrangement, which leads to the fact that the hydrogen leakage in some areas cannot be monitored, and the accuracy of the monitoring is low.

[0004] Therefore, how to provide a scheme to overcome or alleviate the above defects is still a technical problem to be solved by those skilled in the art. SUMMARY

[0005] The purpose of the present application is to provide a hydrogen leakage monitoring method, monitoring device and monitoring system of a fuel cell, which can improve the monitoring range, reduce omissions, and have high accuracy.

[0006] In a first aspect, the present application provides a hydrogen leakage monitoring method of a fuel cell, comprising the following steps: step S100, obtaining an environmental sound sample of the fuel cell; step S200, extracting first feature data of the environmental sound sample; step S300, inputting the first feature data into a preset model to obtain a first judgment information of whether the fuel cell has hydrogen leakage; step S400, performing time-domain to frequency-domain conversion on the environmental sound sample to obtain frequency domain data; step S500, extracting second feature data in the frequency domain data; step S600, comparing the second feature data with set data, and outputting a second judgment information of whether the fuel cell has hydrogen leakage according to the comparison result; step S700, comparing whether the first judgment information and the second judgment information are consistent; and step S800, outputting a result.

[0007] The hydrogen leakage monitoring method is to process the environmental sound sample near the fuel cell to determine whether hydrogen leakage occurs. Compared with the hydrogen concentration monitoring in the conventional technology, the sound monitoring has a large coverage, is not easy to miss, and has high accuracy. Moreover, the sound monitoring requires a relatively small number of sound monitoring components, the structure of the hardware device is relatively simple, the cost is relatively low, and the method can be easily promoted and applied.

[0008] In addition, the monitoring method adopts two judgment modes of preset model analysis and frequency domain analysis. The preset model analysis is to input the first feature data extracted from the environmental sound sample into the preset model to obtain first judgment information of whether hydrogen leakage occurs through the preset model. The frequency domain analysis is to convert the environmental sound sample from time domain to frequency domain to obtain frequency domain data, and extract second feature data of the frequency domain data. Through comparison of the second feature data and the set data, second judgment information of whether hydrogen leakage occurs can be obtained. Then, the first judgment information and the second judgment information are compared to better determine whether hydrogen leakage really occurs, so that the accuracy of the hydrogen leakage monitoring method based on sound monitoring provided by the application can be ensured.

[0009] Optionally, the step S200 comprises: a step S210 of reconstructing the environmental sound sample to obtain reconstruction data; and a step S220 of extracting features of the reconstruction data to obtain the first feature data.

[0010] Optionally, the preset model is an SVM classifier, and the step S220 specifically comprises: performing MFCC feature extraction on the reconstruction data to obtain the first feature data.

[0011] Optionally, the step S220 comprises: a step S221 of performing speech preprocessing and framing on the reconstruction data to obtain preprocessed data; a step S222 of performing Fourier transform on the preprocessed data to obtain data energy spectrum; a step S223 of processing the data energy spectrum through a frequency band-pass filter to obtain output data of M band-passes, M being a set value; a step S224 of performing discrete cosine transform on the output data of the M band-passes to obtain M cepstrum coefficients; and a step S225 of extracting the first L of the M cepstrum coefficients to obtain the first feature data.

[0012] Optionally, the step S300 further comprises a model training step before the step S300, and the model training step comprises: a step S031 of obtaining an environmental sound training sample; a step S032 of extracting first training feature data of the environmental sound training sample; and a step S033 of inputting the first training feature data into the preset model for training.

[0013] Optionally, the time domain to frequency domain conversion is one of Fourier transform, Nyquist transform, wavelet transform and Riesz transform.

[0014] Optionally, the step S500 is extracting frequency spectrum data in a set frequency range in the frequency domain data as the second feature data.

[0015] Optionally, the set frequency range is 35 kHz-50 kHz.

[0016] In a second aspect, the present application provides a hydrogen leakage monitoring device for a fuel cell, comprising: an acquisition module configured to acquire an ambient sound sample of the fuel cell; a first extraction module connected to the acquisition module and configured to receive the ambient sound sample and extract first feature data of the ambient sound sample; an information interaction module connected to the first extraction module and configured to receive the first feature data and input the first feature data into a preset model to obtain first judgment information of whether the fuel cell has hydrogen leakage; a conversion module connected to the acquisition module and configured to receive the ambient sound sample and convert the ambient sound sample from time domain to frequency domain to obtain frequency domain data; a second extraction module connected to the conversion module and configured to receive the frequency domain data and extract second feature data in the frequency domain data; a first comparison module connected to the second extraction module and configured to receive the second feature data, wherein the first comparison module stores set data and is configured to compare the second feature data with the set data and output second judgment information of whether the fuel cell has hydrogen leakage according to the comparison result; a second comparison module connected to the information interaction module and the first comparison module and configured to receive the first judgment information and the second judgment information and compare whether the first judgment information and the second judgment information are consistent; and an output module connected to the second comparison module and configured to output a result according to the comparison result of the first judgment information and the second judgment information.

[0017] Optionally, the first extraction module comprises: a reconstruction unit connected to the acquisition module and configured to receive the ambient sound sample and reconstruct the ambient sound sample to obtain reconstructed data; and a first extraction unit connected to the reconstruction unit and configured to receive the reconstructed data and extract features of the reconstructed data to obtain the first feature data.

[0018] Optionally, the preset model is an SVM classifier.

[0019] Optionally, the first extraction unit includes: a preprocessing subunit, used to perform speech preprocessing and framing on the reconstructed data to obtain preprocessed data; a first transformation subunit, signal-connected to the preprocessing subunit, used to receive the preprocessed data and to perform Fourier transform on the preprocessed data to obtain a data energy spectrum; a filtering subunit, signal-connected to the transformation subunit, used to receive the data energy spectrum and to process the data energy spectrum through a frequency bandpass filter to obtain M bandpass output data, where M is a set value; a second transformation subunit, signal-connected to the filtering subunit, used to receive the M bandpass output data and to perform discrete cosine transform on the M bandpass output data to obtain M cepstral coefficients; and an extraction subunit, signal-connected to the second transformation subunit, used to receive the M cepstral coefficients and to extract the first L cepstral coefficients from the M cepstral coefficients to obtain the first feature data.

[0020] Optionally, the system further includes a training module, comprising: an acquisition unit for acquiring environmental sound training samples; a second extraction unit, signal-connected to the acquisition unit, for receiving the environmental sound training samples and extracting first training feature data from the environmental sound training samples; and an input unit, signal-connected to the second extraction unit, for receiving the first training feature data and inputting the first training feature data into the preset model for training.

[0021] Thirdly, the present invention also provides a hydrogen leakage monitoring system for a fuel cell, including a sound monitoring component and a controller. The sound monitoring component is used to monitor and acquire environmental sound samples of the fuel cell. The controller is signal-connected to the sound monitoring component and is used to acquire the environmental sound samples. The controller is configured to perform the hydrogen leakage monitoring method for the fuel cell as described in the first aspect.

[0022] Optionally, the preset model is integrated into the controller. Attached Figure Description

[0023] Figure 1 This is a schematic flowchart of the hydrogen leakage monitoring method for fuel cells provided by the present invention.

[0024] Figure 2 for Figure 1 The detailed flowchart of step S200;

[0025] Figure 3 for Figure 2 The detailed flowchart of step S220;

[0026] Figure 4 This is a flowchart illustrating the model training steps.

[0027] Figure 5 A schematic diagram of a hydrogen leakage monitoring device for a fuel cell provided by the present application is shown in FIG. 1.

[0028] Figure 6 A schematic diagram of a first extraction module in the hydrogen leakage monitoring device for a fuel cell provided by the present application is shown in FIG. 2. Figure 5 A schematic diagram of a first extraction module in the hydrogen leakage monitoring device for a fuel cell provided by the present application is shown in FIG. 2.

[0029] Figures 5-6 Reference signs in the drawings are explained as follows:

[0030] 100 an acquisition module;

[0031] 200 a first extraction module, 210 a reconstruction unit, 220 a first extraction unit, 221 a preprocessing subunit, 222 a first transform subunit, 223 a filtering subunit, 224 a second transform subunit, and 225 an extraction subunit;

[0032] 300 an information interaction module;

[0033] 400 a transform module;

[0034] 500 a second extraction module;

[0035] 600 a first comparison module;

[0036] 700 a second comparison module;

[0037] 800 an output module;

[0038] 900 a training module, 910 an acquisition unit, 920 a second extraction unit, and 930 an input unit;

[0039] A preset model. DETAILED DESCRIPTION

[0040] In order for those skilled in the art to better understand the technical solutions of the present application, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0041] In the embodiments of the present application, the terms “first”, “second”, “third” are only for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with “first”, “second”, “third” can explicitly or implicitly include one or more of the features.

[0042] In the embodiments of the present application, the terms “S100”, “S200” and the like are only the code names of the related steps given for the convenience of step description, and do not represent the execution order of the related steps.

[0043] In the embodiments of the present application, the terms "comprising", "containing" or any other variant thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements not only includes those elements, but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without more limitations, the element defined by the phrase "comprising a" does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes the element.

[0044] Embodiment one

[0045] Please refer to Figures 1-4 , Figure 1 The flowchart of the hydrogen leakage monitoring method of the fuel cell provided by the present application is shown in Figure 2 The specific flowchart of step S200 in Figure 1 is shown in Figure 3 The specific flowchart of step S220 in Figure 2 is shown in Figure 4 The flowchart of the model training step is shown in

[0046] The present application provides a hydrogen leakage monitoring method for a fuel cell, which can be specifically configured on a vehicle such as an automobile or a rail train. Taking the configuration on an automobile as an example, the hydrogen leakage monitoring method can be executed by a vehicle-mounted controller, which can be an electronic control unit (ECU) or the like.

[0047] As shown in Figure 1 , the hydrogen leakage monitoring method comprises the following steps: S100, obtaining an environmental sound sample of the fuel cell; S200, extracting first feature data of the environmental sound sample; S300, inputting the first feature data into a preset model to obtain first judgment information of whether the fuel cell has hydrogen leakage; S400, performing time-domain to frequency-domain conversion on the environmental sound sample to obtain frequency domain data; S500, extracting second feature data in the frequency domain data; S600, comparing the second feature data with set data and outputting second judgment information of whether the fuel cell has hydrogen leakage according to the comparison result; S700, comparing whether the first judgment information and the second judgment information are consistent; and S800, outputting the result.

[0048] The hydrogen leakage monitoring method is to process the environmental sound sample near the fuel cell to determine whether hydrogen leakage occurs. Compared with the hydrogen concentration monitoring in the conventional technology, the sound monitoring has a large coverage range and is less likely to be missed, and has high accuracy. Moreover, the sound monitoring requires a relatively small number of sound monitoring components, the structure of the hardware device is relatively simple, the cost is relatively low, and the method can be easily popularized and applied.

[0049] Furthermore, the aforementioned monitoring method employs two judgment methods: preset model analysis and frequency domain analysis. Preset model analysis involves inputting the first feature data extracted from the environmental sound sample into a preset model to obtain primary judgment information on whether a hydrogen leak has occurred. Frequency domain analysis transforms the environmental sound sample from the time domain to the frequency domain to obtain frequency domain data, and extracts the second feature data from the frequency domain data. By comparing the second feature data with the preset data, secondary judgment information on whether a hydrogen leak has occurred can be obtained. Then, comparing the primary judgment information and the secondary judgment information can better determine whether a hydrogen leak has actually occurred, thus ensuring the accuracy of the hydrogen leak monitoring method based on sound monitoring provided by this invention.

[0050] When making a judgment, if the primary and secondary judgment information are consistent, the consistent judgment information can be directly output. For example, if both judgment information indicates a leak, a leak alarm can be directly output to alert relevant personnel, allowing for timely maintenance of the fuel cell and minimizing the risk of more serious accidents such as combustion or explosion. Alarm information can be one or more of the following: audible alarm (e.g., humming), indicator light alarm, or text prompts. If the primary and secondary judgment information are inconsistent, a warning can also be issued to indicate a possible hydrogen leak and to inform that the two judgment information are inconsistent. This situation may arise from a fault in either the preset model analysis or frequency domain analysis judgment method, and does not necessarily mean that a hydrogen leak has actually occurred. Relevant personnel can promptly inspect the fuel cell and the corresponding control program to eliminate the fault in a timely manner.

[0051] In the above steps, steps S200 and S400 can be executed simultaneously or sequentially, which is not limited here.

[0052] like Figure 2 As shown, step S200 may include: step S210, reconstructing the environmental sound sample to obtain reconstructed data; and step S220, extracting features from the reconstructed data to obtain first feature data. Step S210 can reconstruct the environmental sound sample. Specific reconstruction methods may include Principal Component Analysis (PCA), wavelet denoising, SOX denoising, etc., to perform dimensionality reduction and noise reduction on the original environmental sound sample. This allows for better acquisition of effective data from the environmental sound sample, and the accuracy of the first feature data obtained from subsequent feature extraction can be higher.

[0053] The extraction manner of the first feature data and the type of the preset model are correlated. In a specific implementation, a person skilled in the art can determine the type of the preset model and the like. For example, the preset model can be a binary classification model, such as a support vector machine (SVM) classifier, a perception machine, a K-Nearest Neighbor (KNN) algorithm, a logistic regression algorithm, a maximum entropy model, a clustering algorithm, and the like.

[0054] Taking the SVM classifier as an example, the step S220 can specifically be: performing Mel-Frequency Cepstral Coefficients (MFCC) feature extraction on the reconstructed data to obtain the first feature data. Figure 3 As shown in the figure, the step S220 can include steps S221-S225.

[0055] The step S221 is to perform voice preprocessing and framing on the reconstructed data to obtain preprocessed data. Specifically, the preprocessing part includes DC removal processing and pre-emphasis processing. The DC removal processing is used to remove the DC component in the reconstructed data, and the pre-emphasis processing is used to enhance the high-frequency component to reduce the dynamic range of the voice spectrum, so that the frequency spectrum of the signal becomes flat, the frequency spectrum tilt can be removed, and the high-frequency part of the voice signal suppressed by the pronunciation system can be compensated. The framing processing can set the frame length (for example, 20-30 ms) and the frame shift (for example, 10 ms), frame the audio signal (a small piece of audio signal can be regarded as short-time stationary), and add a Hanning self convolution window (HSCW) to the signal. The window length can be selected as needed to reduce the spectral energy leakage.

[0056] The step S222 is to perform Fourier transform on the preprocessed data to obtain a data energy spectrum.

[0057] The step S223 is to process the data energy spectrum through a frequency band-pass filter to obtain M band-pass output data, where M is a set value. Specifically, the data energy spectrum can be filtered through a Mel filter bank (triangular filter), and the logarithm of the sum of the energy spectrum in each passband is calculated. M is the number of passbands, which is usually 26-40. In this way, the data energy spectrum can be smoothed, which can effectively eliminate the effect of harmonics and highlight the resonance peaks, and can also reduce the computational complexity to improve the computational efficiency.

[0058] Step S224, performing a discrete cosine transform (DCT) on the M band-pass output data to obtain M cepstrum coefficients.

[0059] Step S225, extracting the first L of the M cepstrum coefficients to obtain first feature data. L is a set value, usually 12-16.

[0060] The above-mentioned preset model can be a trained model, so that the hydrogen leakage monitoring method provided by the application can be directly used. Alternatively, the hydrogen leakage monitoring method provided by the application can also include a model training step, that is, the model training step can be included before the foregoing step S300.

[0061] In combination Figure 4 The model training step can include: step S031, obtaining an environmental sound training sample; step S032, extracting first training feature data of the environmental sound training sample; and step S033, inputting the first training feature data into the preset model for training.

[0062] The environmental sound training sample in step S031 can include a sound sample (labeled) when hydrogen leakage occurs and a sound sample when hydrogen leakage does not occur, and the number of samples can be configured as needed and is not limited herein. The extraction process of the first training feature data in step S032 can refer to the foregoing detailed description of step S200, and will not be repeated here.

[0063] For the frequency domain analysis judgment mode, the time domain to frequency domain conversion in step S400 can be one of Fourier transform, Nyquist transform, wavelet transform and Riesz transform, for example, it can be Fourier transform, more specifically, it can be fast Fourier transform (FFT), which can greatly reduce the amount of calculation.

[0064] When performing FFT, in order to obtain a higher frequency resolution and ensure sufficient spectral information required for analysis, a larger number of input sample points can be set, for example, more than 200 input sample points can be set. In a specific scheme, 219 input sample points can be set, and the frequency resolution can be approximately 0.46 Hz.

[0065] Taking the application of the hydrogen leakage monitoring method provided by the present application to an automobile as an example, there are many noise sources in the fuel cell system, including but not limited to: a cathode air compressor, a tail exhaust system, and other components in the system, such as a hydrogen circulating pump, a hydrogen ejector, an anode hydrogen injection, and the like. The frequencies of the noises generated by different noise sources are different, and therefore, after the frequency domain data is obtained, the frequency domain data can be screened to obtain more effective data.

[0066] In detail, in the embodiment of the present application, the foregoing step S500 can specifically be: extracting the spectrum data in the set frequency range in the frequency domain data as the second feature data.

[0067] The set frequency range can be determined in combination with the audio information generated when hydrogen leakage occurs. According to tests, the hydrogen leakage at 35 MPa and 70 MPa often forms a supersonic under-expanded jet, and the frequency is extremely high, and therefore, in one specific scheme of the embodiment of the present application, the set frequency range can be between 35 kHz and 50 kHz.

[0068] The second feature data can be graph data or numerical data. Correspondingly, the set data used for comparison with the second feature data can also be graph data or numerical data. The set data can be used as a reference value, when the second feature data exceeds the range of the set data, the second judgment information that hydrogen leakage occurs can be given, and when the second feature data is within the range of the set data, the second judgment information that hydrogen leakage does not occur can be given.

[0069] Embodiment two

[0070] Please refer to Figure 5 and Figure 6 , Figure 5 is a schematic diagram of a hydrogen leakage monitoring device for a fuel cell provided by the present application, Figure 6 is Figure 5 a schematic diagram of a first extraction module in the device.

[0071] As Figure 5 shown, the present application also provides a hydrogen leakage monitoring device for a fuel cell, which comprises an acquisition module 100, a first extraction module 200, an information interaction module 300, a transformation module 400, a second extraction module 500, a first comparison module 600, a second comparison module 700, and an output module 800, and these modules can be integrated in a controller.

[0072] Specifically, the acquisition module 100 is configured to acquire an ambient sound sample of the fuel cell; the first extraction module 200 is signal-connected with the acquisition module 100 and configured to receive the ambient sound sample and extract first feature data of the ambient sound sample; the information interaction module 300 is signal-connected with the first extraction module 200 and configured to receive the first feature data and input the first feature data into a preset model A to obtain first judgment information about whether the fuel cell has hydrogen leakage; the transformation module 400 is signal-connected with the acquisition module 100 and configured to receive the ambient sound sample and transform the ambient sound sample from time domain to frequency domain to obtain frequency domain data; the second extraction module 500 is signal-connected with the transformation module 400 and configured to receive the frequency domain data and extract second feature data in the frequency domain data; the first comparison module 600 is signal-connected with the second extraction module 500 and configured to receive the second feature data, the first comparison module 600 stores setting data, and the first comparison module 600 is configured to compare the second feature data with the setting data and output second judgment information about whether the fuel cell has hydrogen leakage according to a comparison result; the second comparison module 700 is signal-connected with the information interaction module 300 and the first comparison module 600 and configured to receive the first judgment information and the second judgment information and compare whether the first judgment information and the second judgment information are consistent; and the output module 800 is signal-connected with the second comparison module 700 and configured to output a result according to the comparison result of the first judgment information and the second judgment information.

[0073] The hydrogen leakage monitoring device is configured to process the ambient sound sample near the fuel cell to determine whether hydrogen leakage occurs, has a large monitoring coverage, is less likely to be missed, has high accuracy, has fewer sound monitoring components required for sound monitoring, has a simple structure of hardware devices, has low cost, and can be conveniently promoted and applied.

[0074] In addition, the monitoring device compares the two judgment manners of the preset model A analysis and the frequency domain analysis, can better determine whether hydrogen leakage actually occurs, and can guarantee the accuracy of the hydrogen leakage monitoring device based on sound monitoring.

[0075] As shown in FIG. 2, the first extraction module 200 can include a reconstruction unit 210, which can be signal-connected with the acquisition module 100, configured to receive the ambient sound sample, and configured to reconstruct the ambient sound sample to obtain reconstruction data, can perform dimension reduction, denoising, and other processing on the original ambient sound sample, so that effective data in the ambient sound sample can be better acquired, and the first feature data obtained through subsequent feature extraction is more accurate; and a first extraction unit 220, which can be signal-connected with the reconstruction unit 210, configured to receive the reconstruction data, and configured to perform feature extraction on the reconstruction data to obtain the first feature data. Figure 6 The reconstruction unit 210 can be signal-connected with the acquisition module 100, configured to receive the ambient sound sample, and configured to reconstruct the ambient sound sample to obtain reconstruction data, can perform dimension reduction, denoising, and other processing on the original ambient sound sample, so that effective data in the ambient sound sample can be better acquired, and the first feature data obtained through subsequent feature extraction is more accurate; and the first extraction unit 220 can be signal-connected with the reconstruction unit 210, configured to receive the reconstruction data, and configured to perform feature extraction on the reconstruction data to obtain the first feature data.

[0076] The extraction manner of the first feature data and the type of the preset model A are correlated. In a specific implementation, a person skilled in the art can determine according to the type of the preset model A and the like. For example, the preset model A can be a binary classification model, such as a support vector machine (SVM) classifier, a perception machine, a K-Nearest Neighbor (KNN) algorithm, a logistic regression algorithm, a maximum entropy model, a clustering algorithm, and the like.

[0077] Taking the SVM classifier as an example, the first extraction unit 220 can specifically perform Mel-Frequency Cepstral Coefficients (MFCC) feature extraction on the reconstructed data to obtain the first feature data. As shown in the embodiment, in this implementation, the first extraction unit 220 can include a preprocessing subunit 221, a first transformation subunit 222, a filtering subunit 223, a second transformation subunit 224, and an extraction subunit 225. Figure 6 The preprocessing subunit 221 is configured to perform voice preprocessing and framing on the reconstructed data to obtain preprocessed data. The specific preprocessing manner can refer to Embodiment 1, and will not be repeated here. The first transformation subunit 222 and the preprocessing subunit 221 can be signal connected, configured to receive the preprocessed data, and configured to perform Fourier transform on the preprocessed data to obtain a data energy spectrum. The filtering subunit 223 can be signal connected with the transformation subunit, configured to receive the data energy spectrum, and configured to process the data energy spectrum through a frequency band-pass filter to obtain M band-pass output data, where M is a set value, and is usually 26-40. The second transformation subunit 224 and the filtering subunit 223 can be signal connected, configured to receive the M band-pass output data, and configured to perform discrete cosine transform on the M band-pass output data to obtain M cepstral coefficients. The extraction subunit 225 and the second transformation subunit 224 can be signal connected, configured to receive the M cepstral coefficients, and configured to extract the first L cepstral coefficients in the M cepstral coefficients as the first feature data; L is a set value, and is usually 12-16.

[0078] The preset model A can be a trained model, so that the hydrogen leakage monitoring device provided by the present application can be directly used. Alternatively, the hydrogen leakage monitoring device provided by the present application can also include a model training step, that is, before information interaction between the information interaction module 300 and the preset model A, the hydrogen leakage monitoring device provided by the present application can be further configured with a training module 900, and the preset model A is trained through the training module 900.

[0079] The preset model A can be a trained model, so that the hydrogen leakage monitoring device provided by the present application can be directly used. Alternatively, the hydrogen leakage monitoring device provided by the present application can also include a model training step, that is, before information interaction between the information interaction module 300 and the preset model A, the hydrogen leakage monitoring device provided by the present application can be further configured with a training module 900, and the preset model A is trained through the training module 900.

[0080] Please continue to refer to Figure 5 , the training module 900 can include: an acquisition unit 910 for acquiring an environmental sound training sample; a second extraction unit 920, which can be signal connected with the acquisition unit 910, for receiving the environmental sound training sample and for extracting first training feature data of the environmental sound training sample; an input unit 930, which can be signal connected with the second extraction unit 920, for receiving the above-mentioned first training feature data and for inputting the first training feature data into a preset model A for model training.

[0081] It should be understood that other parts not defined in this embodiment can be referred to Embodiment One, which will not be repeated here.

[0082] Embodiment Three

[0083] The application also provides a hydrogen leakage monitoring system of a fuel cell, comprising a sound monitoring component and a controller.

[0084] The sound monitoring component can specifically include a sound sensor for collecting environmental sound near the fuel cell. In actual application, the number of sound monitoring components can be one or multiple, which can be determined according to actual application scenarios. Taking application in a car as an example, the number of sound monitoring components can be one, and this sound monitoring component can be arranged near a high-pressure hydrogen pipeline interface.

[0085] The type of the above-mentioned sound sensor is not limited here, and in actual application, it can be configured by those skilled in the art according to specific needs, as long as it can meet the requirements of use. For example, the sound sensor can be an ultrasonic sensor, and its rated power can be 20KHz or above, so as to better capture the sound signal when hydrogen leaks.

[0086] The controller is signal connected with the sound monitoring component for receiving the environmental sound collected by the sound monitoring component, and the controller is configured to be capable of performing the hydrogen leakage monitoring method of the fuel cell involved in each embodiment of Embodiment One. In this way, the controller cooperates with the sound monitoring component to more accurately monitor whether hydrogen leakage occurs, and the number of sound monitoring components required by this embodiment is relatively small, the structure is relatively simple, and the cost is relatively low, so it can be conveniently popularized and applied.

[0087] In actual application, the preset model A can be integrated in the controller or independent of the controller. If the latter scheme is adopted, a wire harness or the like can also be configured to build a signal transmission relationship between the controller and the preset model A.

[0088] In some embodiments, the sound monitoring component can also be configured with a filtering and amplifying circuit for filtering and amplifying the ambient sound directly collected by the sound sensor before being transmitted to the controller. In this way, unnecessary information in the obtained ambient sound sample can be deleted and effective information can be amplified, which is more conducive to improving the accuracy of hydrogen leakage monitoring.

[0089] The above merely describes the preferred embodiments of the present application, and it should be pointed out that those skilled in the art can make several improvements and refinements without departing from the principles of the present application, and these improvements and refinements should also be considered as the protection scope of the present application.

Claims

1. A method for monitoring hydrogen leakage in a fuel cell, characterized in that, Includes the following steps: Step S100: Obtain environmental sound samples of the fuel cell; Step S200: Extract the first feature data of the environmental sound sample; Step S300: Input the first feature data into a preset model (A) to obtain a judgment information on whether the fuel cell has experienced hydrogen leakage; Step S400: Perform time-domain to frequency-domain transformation on the environmental sound sample to obtain frequency domain data; Step S500: Extract the second feature data from the frequency domain data; Step S600: Compare the second feature data with the set data, and output secondary judgment information on whether the fuel cell has experienced hydrogen leakage based on the comparison result; Step S700: Compare whether the first judgment information and the second judgment information are consistent; Step S800, output the result; Step S200 and step S400 may be executed simultaneously, or step S200 and step S400 may be executed sequentially.

2. The method for monitoring hydrogen leakage in a fuel cell according to claim 1, characterized in that, Step S200 includes: Step S210: Reconstruct the environmental sound samples to obtain reconstructed data; Step S220: Perform feature extraction on the reconstructed data to obtain the first feature data.

3. The method for monitoring hydrogen leakage in a fuel cell according to claim 2, characterized in that, The preset model (A) is an SVM classifier, and step S220 specifically involves: performing MFCC feature extraction on the reconstructed data to obtain the first feature data.

4. The method for monitoring hydrogen leakage in a fuel cell according to claim 3, characterized in that, Step S220 includes: Step S221: Perform speech preprocessing and frame segmentation on the reconstructed data to obtain preprocessed data; Step S222: Perform a Fourier transform on the preprocessed data to obtain the data energy spectrum; Step S223: Process the data energy spectrum through a frequency bandpass filter to obtain M bandpass output data, where M is a set value; Step S224: Perform discrete cosine transform on the M bandpass output data to obtain M cepstral coefficients; Step S225: Extract the first L cepstral coefficients from the M cepstral coefficients to obtain the first feature data.

5. The method for monitoring hydrogen leakage in a fuel cell according to claim 1, characterized in that, Before step S300, a model training step is included, which includes: Step S031: Obtain environmental sound training samples; Step S032: Extract the first training feature data of the environmental sound training samples; Step S033: Input the first training feature data into the preset model (A) for training.

6. The method for monitoring hydrogen leakage in a fuel cell according to any one of claims 1-5, characterized in that, The time-domain to frequency-domain transformation is specifically one of the following: Fourier transform, Nyquist transform, wavelet transform, and Rice transform.

7. The method for monitoring hydrogen leakage in a fuel cell according to any one of claims 1-5, characterized in that, Specifically, step S500 involves extracting spectral data within a set frequency range from the frequency domain data as the second feature data.

8. The method for monitoring hydrogen leakage in a fuel cell according to claim 7, characterized in that, The set frequency range is 35kHz-50kHz.

9. A hydrogen leakage monitoring device for a fuel cell, characterized in that, include: Acquisition module (100) is used to acquire environmental sound samples of the fuel cell; The first extraction module (200) is signal-connected to the acquisition module (100) and is used to receive the environmental sound sample and extract the first feature data of the environmental sound sample. The information interaction module (300) is signal-connected to the first extraction module (200) and is used to receive the first feature data and input the first feature data into a preset model (A) to obtain a judgment information on whether the fuel cell has experienced hydrogen leakage. The transformation module (400) is signal-connected to the acquisition module (100) and is used to receive the environmental sound sample and to perform time-domain to frequency-domain transformation on the environmental sound sample to obtain frequency domain data. The second extraction module (500) is signal-connected to the transformation module (400) and is used to receive the frequency domain data and extract the second feature data from the frequency domain data; The first comparison module (600) and the second extraction module (500) are connected by a signal and are used to receive the second feature data. The first comparison module (600) stores set data and is used to compare the second feature data and the set data, and to output secondary judgment information on whether the fuel cell has hydrogen leakage based on the comparison result. The second comparison module (700), the information interaction module (300), and the first comparison module (600) are all signal connected, and are used to receive the first judgment information and the second judgment information, and to compare whether the first judgment information and the second judgment information are consistent; The output module (800) is signal-connected to the second comparison module (700) and is used to output the result based on the comparison result of the first judgment information and the second judgment information.

10. The hydrogen leakage monitoring device for a fuel cell according to claim 9, characterized in that, The first extraction module (200) includes: The reconstruction unit (210) is signal-connected to the acquisition module (100) and is used to receive the environmental sound sample and to reconstruct the environmental sound sample to obtain reconstruction data. The first extraction unit (220) is signal-connected to the reconstruction unit (210) and is used to receive the reconstruction data and to perform feature extraction on the reconstruction data to obtain the first feature data.

11. The hydrogen leakage monitoring device for a fuel cell according to claim 9, characterized in that, The preset model (A) is an SVM classifier.

12. A hydrogen leakage monitoring system for a fuel cell, characterized in that, The device includes a sound monitoring component and a controller. The sound monitoring component is used to monitor and acquire ambient sound samples of the fuel cell. The controller is signal-connected to the sound monitoring component and is used to acquire the ambient sound samples. The controller is configured to perform the hydrogen leakage monitoring method for the fuel cell according to any one of claims 1-8.

Images