Hydrogen fuel cell fault detection device and method adopting data driving method
By combining data-driven deep learning models with sensor data, the shortcomings of fuel cell modeling methods in generalization ability and fault diagnosis are addressed, enabling accurate detection and health management of hydrogen fuel cell faults, and improving battery reliability and lifespan.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-11
- Publication Date
- 2026-04-03
AI Technical Summary
Existing fuel cell modeling methods are insufficient in terms of application scope and generalization ability for complex hydrogen fuel cell systems, especially due to the complexity of their structure and working mechanism, which makes it difficult to accurately predict performance degradation and diagnose faults.
A data-driven approach is adopted, and a hydrogen fuel cell fault detection device is established by combining deep learning models with sensor data. The device includes a central control module, a computing module, a storage module, a sensor module, and an input/output module. Deep learning models such as LSTM and GRU are used to perform real-time detection and diagnosis of fault conditions.
It enables precise detection and health management of hydrogen fuel cell faults, improves the accuracy and real-time nature of detection, allows for timely problem identification and maintenance, and extends battery life.
Smart Images
Figure CN121790448A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of hydrogen fuel cell fault detection, and in particular to a hydrogen fuel cell fault detection device and method using a data-driven approach. Background Technology
[0002] Hydrogen fuel cells using proton exchange membrane (PEMFC) as the electrolyte offer advantages such as high energy conversion efficiency, lightweight design, and high power density, making them the ideal power source for fuel cell vehicles. However, during long-term operation, the performance of fuel cells gradually degrades, causing a drop in output voltage. Besides total operating time, complex operating conditions such as frequent starts and stops, and load variations can accelerate battery degradation. Furthermore, various faults such as flooding, lack of air, and membrane drying can also shorten the remaining lifespan of the battery.
[0003] Establishing battery performance degradation models to predict battery performance decline trends is crucial for improving fuel cell reliability and extending battery lifespan. Accurate models not only help predict remaining battery lifespan but also assist in battery fault diagnosis, thereby optimizing battery management strategies, achieving more reliable battery system operation, and extending battery service life.
[0004] Existing fuel cell modeling methods are mainly divided into data-driven methods and model-driven methods. Model-driven methods offer the advantage of strong interpretability, but due to the complexity of fuel cell system structure and operating mechanisms, the incomplete understanding of cell aging mechanisms, and the numerous factors influencing the magnitude and trend of degradation, the application scope and generalization ability of model-driven methods are often difficult to guarantee. Data-driven methods predict the degradation trend and remaining lifespan of PEMFCs based on historical data, without needing to consider their aging mechanisms, making them more suitable for modeling systems like hydrogen fuel cells with numerous input and output parameters and complex characteristics. Summary of the Invention
[0005] Based on the above-mentioned technical problems, this invention proposes a hydrogen fuel cell fault detection device and method using a data-driven approach.
[0006] The technical solution adopted in this invention is as follows: A hydrogen fuel cell fault detection device using a data-driven approach includes a central control module, a computing module, a storage module, a sensor module, and an input / output module. The central control module is connected to the input / output module, the storage module, and the sensor module. The storage module and the input / output module are connected to the computing module, and the sensor module is connected to the hydrogen fuel cell. The sensor module is used to collect the load current, output voltage, inlet pressure of hydrogen and air, dew point temperature, flow rate, temperature, and anode plate temperature of the hydrogen fuel cell. The calculation module is used to train a deep learning model for fault diagnosis of hydrogen fuel cells and to calculate the fault state of the battery based on the data collected by the sensor module. The storage module is used to store the deep learning model for hydrogen fuel cell fault diagnosis, the data collected by the sensor module, and the results calculated by the deep learning model for hydrogen fuel cell fault diagnosis. The input / output module is used to display the real-time operating status and fault diagnosis status of the hydrogen fuel cell, and to input the parameters for training the deep learning model for fault diagnosis of the hydrogen fuel cell. The central control module is used to control the coordinated operation between the input / output module, storage module, and sensor module.
[0007] Furthermore, the deep learning model for hydrogen fuel cell fault diagnosis includes a sequence input layer, a feature extraction layer, an RNN layer, a loss function layer, a fully connected layer, and an output layer; the feature extraction layer includes several CNN layers, several TCN layers, or a combination of several CNN and TCN layers; the RNN layer uses one of LSTM, GRU, BiLTSM, and BiGRU; or variations of LSTM, GRU, BiLTSM, and BiGRU.
[0008] Furthermore, the number of neurons in the RNN layer is 16-512, the number of feature extraction layers is 1-6, the kernel size of the feature extraction layer is 2-9, and the stride is 2-5.
[0009] The present invention also provides a hydrogen fuel cell fault detection method using a data-driven approach, which is based on the above-mentioned hydrogen fuel cell fault detection device using a data-driven approach, and includes the following steps: Step 1: Obtain training data for the deep learning model for hydrogen fuel cell fault diagnosis through the sensor module; Step 2: Based on the training data obtained in Step 1, train the deep learning model for hydrogen fuel cell fault diagnosis through the calculation module, and store the trained deep learning model for hydrogen fuel cell fault diagnosis into the storage module. Step 3: Collect test data through the sensor module and store the test data in the storage module; Step 4: The calculation module processes the test data stored in the storage module and calls the trained deep learning model for hydrogen fuel cell fault diagnosis to calculate the output voltage of the hydrogen fuel cell. Step 5: Compare the output voltage obtained from the sensor test with the output voltage predicted by the deep learning model for hydrogen fuel cell fault diagnosis. If the deviation exceeds the user-preset threshold, the input / output module displays the fault status of the hydrogen fuel cell.
[0010] Furthermore, in step 4, the steps for processing the test data are as follows: Step 41, the test data includes the total running time of the hydrogen fuel cell and the running time after each restart collected by the sensor module, load current, inlet pressure of hydrogen and air, dew point temperature, flow rate, temperature, and anode plate temperature. Step 42: Arrange the test data into a data sequence according to time order; Step 43: Sample data according to the sequence length input by the input / output module; Step 44: Produce the final data sequence as input data for the deep learning model for hydrogen fuel cell fault diagnosis.
[0011] Furthermore, during sensor sampling, the interval between each data point is 0.1s-10s, and the length of the data sequence is 1-50.
[0012] Furthermore, in step 2, the following steps are used to train the deep learning model for hydrogen fuel cell fault diagnosis: Step 21: Collect test data from the hydrogen fuel cell sensor module; Step 22: Arrange the test data into a data sequence according to the time order, sample the data according to the sequence length input by the input / output module, and produce the final data sequence; Step 23: Use the data sequence obtained in step 22 as input and the output voltage as output to train the deep learning model for hydrogen fuel cell fault diagnosis. Step 24: Calculate the predicted value of the hydrogen fuel cell fault diagnosis deep learning model obtained through training on the training input data, calculate the error between the predicted value and the measured value, and optimize the minimum value of the error between the predicted value and the measured value to obtain the optimal value of the network parameters. Step 25: Use the optimal values of the network parameters obtained through optimization to continue training, obtain the trained deep learning model for hydrogen fuel cell fault diagnosis, and store it in the storage module.
[0013] Furthermore, the deep learning model for hydrogen fuel cell fault diagnosis includes a sequence input layer, a feature extraction layer, an RNN layer, a loss function layer, a fully connected layer, and an output layer; the feature extraction layer includes several CNN layers, several TCN layers, or a combination of several CNN and TCN layers; the RNN layer uses one of LSTM, GRU, BiLTSM, and BiGRU; or variations of LSTM, GRU, BiLTSM, and BiGRU.
[0014] Furthermore, the number of neurons in the RNN layer is 16-512, the number of feature extraction layers is 1-6, the kernel size of the feature extraction layer is 2-9, and the stride is 2-5.
[0015] Furthermore, in step 5, the user-preset threshold range is 10%-30%.
[0016] The beneficial technical effects of the present invention are as follows: This invention processes a series of sensor data that significantly affect the output characteristics of hydrogen fuel cells to establish a predictive model for the output voltage. This model then determines whether the battery characteristics deviate from normal operating conditions, ultimately obtaining a detection result indicating whether the battery is in a faulty state. This invention offers advantages such as accurate detection results and precise health management of hydrogen fuel cells.
[0017] Specifically, this invention leverages the powerful data processing capabilities of deep learning to select a series of test parameters that may affect the characteristics of hydrogen fuel cells, thereby extracting in-depth information related to battery faults. Compared to other methods, this invention can use a pre-built model to analyze whether the battery's operating state is abnormal from real-time, complex detection data, providing excellent real-time diagnostics that allow users to promptly identify problems and perform maintenance. Attached Figure Description
[0018] Figure 1 This is a structural principle block diagram of the hydrogen fuel cell fault detection device using the data-driven method of the present invention; Figure 2 This is a schematic flowchart of the hydrogen fuel cell fault detection method using a data-driven approach according to the present invention. Detailed Implementation
[0019] like Figure 1 As shown, a hydrogen fuel cell fault detection device using a data-driven method includes a central control module, a computing module, a storage module, a sensor module, and an input / output module. The central control module is connected to the input / output module, the storage module, and the sensor module. The storage module and the input / output module are connected to the computing module, and the sensor module is connected to the hydrogen fuel cell.
[0020] The sensor module collects data on the hydrogen fuel cell's load current, output voltage, inlet pressure of hydrogen and air, dew point temperature, flow rate, temperature, and anode plate temperature. The calculation module trains a deep learning model for hydrogen fuel cell fault diagnosis and calculates the cell's fault state based on the data collected by the sensor module. The storage module stores the deep learning model for hydrogen fuel cell fault diagnosis, the data collected by the sensor module, and the results calculated by the model. The input / output module displays the real-time operating status and fault diagnosis status of the hydrogen fuel cell and allows input of parameters used to train the deep learning model. The central control module coordinates the operation of the input / output module, storage module, and sensor module, effectively managing the coordination between these modules.
[0021] The deep learning model for hydrogen fuel cell fault diagnosis used in this invention comprises a sequentially connected input layer, a feature extraction layer, an RNN layer, a loss function layer, a fully connected layer, and an output layer. The feature extraction layer includes several CNN layers, several TCN layers, or a combination of CNN and TCN layers. The RNN layer is one of LSTM, GRU, BiLTSM, and BiGRU, or a variant of LSTM, GRU, BiLTSM, or BiGRU. The RNN layer has 16-512 neurons, the feature extraction layer has 1-6 layers, the kernel size of the feature extraction layer is 2-9, the stride is 2-5, and the number of convolutional kernels is 2. 2 -2 8 .
[0022] The deep learning model for hydrogen fuel cell fault diagnosis used in this invention comprises a sequentially connected input layer, a feature extraction layer, an RNN layer, a loss function layer, a fully connected layer, and an output layer. The feature extraction layer includes several CNN layers, several TCN layers, or a combination of CNN and TCN layers. The RNN layer has 16-512 neurons, and the feature extraction layer has 1-6 layers.
[0023] like Figure 2 As shown, the present invention also provides a hydrogen fuel cell fault detection method using a data-driven approach, which employs the hydrogen fuel cell fault detection device using a data-driven approach as described above, and includes the following steps: Step 1: Obtain training data for the deep learning model for hydrogen fuel cell fault diagnosis through the sensor module.
[0024] Step 2: Based on the training data obtained in Step 1, train the deep learning model for hydrogen fuel cell fault diagnosis through the calculation module, and store the trained deep learning model for hydrogen fuel cell fault diagnosis into the storage module.
[0025] Step 3: After the model training is completed, test data is collected through the sensor module and stored in the storage module.
[0026] Step 4: The calculation module processes the test data stored in the storage module and calls the trained deep learning model for hydrogen fuel cell fault diagnosis to calculate the output voltage of the hydrogen fuel cell.
[0027] Step 5: Compare the output voltage obtained from the sensor test with the output voltage predicted by the deep learning model for hydrogen fuel cell fault diagnosis. If the deviation exceeds the user-preset threshold, such as 10%-30%, the input / output module displays the fault status of the hydrogen fuel cell.
[0028] In both model training and prediction, the obtained training or test data are first processed, and the data processing steps are identical. For example, in step 4, the steps for processing the test data are as follows: Step 41, the test data includes the total running time of the hydrogen fuel cell and the running time after each restart collected by the sensor module, load current, inlet pressure of hydrogen and air, dew point temperature, flow rate, temperature, and anode plate temperature.
[0029] Step 42: Arrange the test data into a data sequence in chronological order.
[0030] Step 43: Perform data sampling based on the sequence length input from the input / output module. During sampling, the interval between each data point is 0.1s-10s, and the length of the data sequence is 1-50.
[0031] Step 44: Produce the final data sequence as input data for the deep learning model for hydrogen fuel cell fault diagnosis.
[0032] In step 2 above, the following steps are used to train the deep learning model for hydrogen fuel cell fault diagnosis: Step 21: Collect test data from the hydrogen fuel cell sensor module.
[0033] Step 22, the test data processing steps are the same as above. That is, the test data is arranged into a data sequence according to the time order, and data sampling is performed according to the sequence length input by the input / output module to produce the final data sequence.
[0034] Step 23: Use the data sequence obtained in step 22 as input and the output voltage as output to train the deep learning model for hydrogen fuel cell fault diagnosis.
[0035] Step 24: Calculate the predicted value of the hydrogen fuel cell fault diagnosis deep learning model obtained through training on the training input data, and calculate the error between the predicted value and the measured value. Optimize the minimum value of the error between the predicted value and the measured value (e.g., by using Bayesian optimization, genetic algorithm, etc.) to obtain the optimal value of the network parameters.
[0036] Step 25: Use the optimal values of the network parameters obtained through optimization to continue training, obtain the trained deep learning model for hydrogen fuel cell fault diagnosis, and store it in the storage module.
[0037] The aforementioned deep learning model for hydrogen fuel cell fault diagnosis includes a sequence input layer, a feature extraction layer, an RNN layer, a loss function layer, a fully connected layer, and an output layer. The feature extraction layer comprises several CNN layers, several TCN layers, or a combination of CNN and TCN layers. The RNN layer is one of LSTM, GRU, BiLTSM, or BiGRU, or a variant of LSTM, GRU, BiLTSM, or BiGRU.
[0038] The RNN layers mentioned above have 16-512 neurons, 1-6 feature extraction layers, 2-9 kernels, 2-5 strides, and 2^30 kernels. 2 -2 8 .
[0039] This invention processes sensor data that most significantly affects the output characteristics of hydrogen fuel cells to establish a predictive model for the output voltage. This model then determines whether the battery characteristics deviate from normal operating conditions, ultimately obtaining a detection result indicating whether the battery is in a faulty state. This invention offers advantages such as accurate detection results and precise health management of hydrogen fuel cells.
[0040] For any parts not mentioned in the above embodiments, existing technologies can be adopted or referenced.
[0041] Of course, the above description is not intended to limit the present invention, and the present invention is not limited to the above embodiments. Any changes, modifications, additions or substitutions made by those skilled in the art within the scope of the present invention should be protected by the present invention.
Claims
1. A hydrogen fuel cell fault detection device employing a data-driven method, characterized in that: It includes a central control module, a computing module, a storage module, a sensor module, and an input / output module. The central control module is connected to the input / output module, the storage module, and the sensor module. The storage module and the input / output module are connected to the computing module, and the sensor module is connected to the hydrogen fuel cell. The sensor module is used to collect the load current, output voltage, inlet pressure of hydrogen and air, dew point temperature, flow rate, temperature, and anode plate temperature of the hydrogen fuel cell. The calculation module is used to train a deep learning model for fault diagnosis of hydrogen fuel cells and to calculate the fault state of the battery based on the data collected by the sensor module. The storage module is used to store the deep learning model for hydrogen fuel cell fault diagnosis, the data collected by the sensor module, and the results calculated by the deep learning model for hydrogen fuel cell fault diagnosis. The input / output module is used to display the real-time operating status and fault diagnosis status of the hydrogen fuel cell, and to input the parameters for training the deep learning model for fault diagnosis of the hydrogen fuel cell. The central control module is used to control the coordinated operation between the input / output module, storage module, and sensor module.
2. The hydrogen fuel cell fault detection device using a data-driven method according to claim 1, characterized in that: The deep learning model for hydrogen fuel cell fault diagnosis includes a sequence input layer, a feature extraction layer, an RNN layer, a loss function layer, a fully connected layer, and an output layer; the feature extraction layer includes several CNN layers, several TCN layers, or a combination of several CNN and TCN layers; the RNN layer uses one of LSTM, GRU, BiLTSM, and BiGRU.
3. A hydrogen fuel cell fault detection device using a data-driven method according to claim 2, characterized in that: The number of neurons in the RNN layer is 16-512, the number of feature extraction layers is 1-6, the kernel size of the feature extraction layer is 2-9, and the stride is 2-5.
4. A fault detection method for hydrogen fuel cells using a data-driven approach, characterized in that, The hydrogen fuel cell fault detection device based on any one of claims 1-3 using a data-driven method includes the following steps: Step 1: Obtain training data for the deep learning model for hydrogen fuel cell fault diagnosis through the sensor module; Step 2: Based on the training data obtained in Step 1, train the deep learning model for hydrogen fuel cell fault diagnosis through the calculation module, and store the trained deep learning model for hydrogen fuel cell fault diagnosis into the storage module. Step 3: Collect test data through the sensor module and store the test data in the storage module; Step 4: The calculation module processes the test data stored in the storage module and calls the trained deep learning model for hydrogen fuel cell fault diagnosis to calculate the output voltage of the hydrogen fuel cell. Step 5: Compare the output voltage obtained from the sensor test with the output voltage predicted by the deep learning model for hydrogen fuel cell fault diagnosis. If the deviation exceeds the user-preset threshold, the input / output module displays the fault status of the hydrogen fuel cell.
5. A hydrogen fuel cell fault detection method using a data-driven approach according to claim 4, characterized in that, In step 4, the steps for processing the test data are as follows: Step 41, the test data includes the total running time of the hydrogen fuel cell and the running time after each restart collected by the sensor module, load current, inlet pressure of hydrogen and air, dew point temperature, flow rate, temperature, and anode plate temperature. Step 42: Arrange the test data into a data sequence according to time order; Step 43: Sample data according to the sequence length input by the input / output module; Step 44: Produce the final data sequence as input data for the deep learning model for hydrogen fuel cell fault diagnosis.
6. A hydrogen fuel cell fault detection method using a data-driven approach according to claim 5, characterized in that, When the sensor samples, the interval between each data point is 0.1s-10s, and the length of the data sequence is 1-50.
7. A hydrogen fuel cell fault detection method using a data-driven approach according to claim 4, characterized in that, In step 2, the following steps are used to train the deep learning model for hydrogen fuel cell fault diagnosis: Step 21: Collect test data from the hydrogen fuel cell sensor module; Step 22: Arrange the test data into a data sequence according to the time order, sample the data according to the sequence length input by the input / output module, and produce the final data sequence; Step 23: Use the data sequence obtained in step 22 as input and the output voltage as output to train the deep learning model for hydrogen fuel cell fault diagnosis. Step 24: Calculate the predicted value of the hydrogen fuel cell fault diagnosis deep learning model obtained through training on the training input data, calculate the error between the predicted value and the measured value, and optimize the minimum value of the error between the predicted value and the measured value to obtain the optimal value of the network parameters. Step 25: Use the optimal values of the network parameters obtained through optimization to continue training, obtain the trained deep learning model for hydrogen fuel cell fault diagnosis, and store it in the storage module.
8. A hydrogen fuel cell fault detection method using a data-driven approach according to claim 7, characterized in that, The deep learning model for hydrogen fuel cell fault diagnosis includes a sequence input layer, a feature extraction layer, an RNN layer, a loss function layer, a fully connected layer, and an output layer; the feature extraction layer includes several CNN layers, several TCN layers, or a combination of several CNN and TCN layers; the RNN layer uses one of LSTM, GRU, BiLTSM, and BiGRU.
9. A hydrogen fuel cell fault detection method using a data-driven approach according to claim 8, characterized in that, The number of neurons in the RNN layer is 16-512, the number of feature extraction layers is 1-6, the kernel size of the feature extraction layer is 2-9, and the stride is 2-5.
10. A hydrogen fuel cell fault detection method using a data-driven approach according to claim 4, characterized in that, In step 5, the threshold range is set to 10%-30%.