Intelligent cooperative high-power proton exchange membrane fuel cell life prediction and fault diagnosis method and system

Abstract

The invention relates to the technical field of fuel cell monitoring and health management, and discloses an intelligent cooperative high-power proton exchange membrane fuel cell life prediction and fault diagnosis method and system, and the method comprises the following steps: S1, collecting multi-dimensional data of a stack key region in real time through a distributed sensing array; s2, removing noise and abnormal values of the collected data, and carrying out feature extraction; s3, constructing a dual-model collaborative prediction framework; s4, inputting the real-time features into the double models; s5, generating a maintenance strategy and early warning; according to the invention, special hardware accelerators such as a graphic processing unit or a field-programmable gate array are integrated on the edge computing module, so that the processing speed and efficiency of high-frequency and multi-source sensor data at a local end are remarkably improved; a distributed computing framework is adopted in a cloud analysis module to process mass feature data transmitted by an edge end, so that the capability and scale of the system for performing deep analysis on the health state of the fuel cell, accurately predicting the remaining service life and quickly tracing the fault are enhanced.

Description

Technical Field

[0001] This invention relates to the technical field of fuel cell monitoring and health management, specifically to a method and system for predicting the lifespan and diagnosing faults of intelligent collaborative high-power proton exchange membrane fuel cells. Background Technology

[0002] With the rapid development of new energy technologies, proton exchange membrane fuel cells (PEMFCs) have been widely used in electric vehicles and distributed power generation due to their advantages such as high energy density, fast start-up and zero emissions. However, in actual operation, high-power PEMFC systems will gradually degrade in performance and shorten in life due to complex operating conditions and multi-physics coupling effects. They may also experience failures such as flooding, gas starvation and cooling failure. These problems seriously affect the reliability and safety of fuel cell systems.

[0003] Currently, PEMFC lifetime prediction and fault diagnosis mainly rely on the following methods: In terms of data acquisition, a limited number of sensors are deployed on the surface of the fuel cell stack to collect basic operating parameters such as voltage, current, and temperature at a limited number of points; in the data processing stage, simple filtering and denoising methods and statistical feature extraction methods are used to perform preliminary processing on the collected data; in terms of lifetime prediction, either data-driven machine learning methods are relied upon alone, or predictions are made based solely on simplified physical mechanism models; in terms of fault diagnosis, common faults are mainly identified by monitoring voltage changes and temperature anomalies, lacking in-depth analysis of fault propagation relationships and effective methods for separating multiple coupled faults.

[0004] However, it has the following drawbacks:

[0005] First, insufficient edge computing capabilities result in high data processing latency and poor real-time performance. Traditional methods are unable to extract and respond to minute voltage changes, fast impedance spectrum signals, and temperature and humidity distributions in fuel cell cells at the millisecond level, making it difficult to meet the real-time requirements for early fault warning and accurate diagnosis.

[0006] Secondly, cloud computing resources are inefficient. When training life prediction models and inferring fault diagnosis models from large-scale historical and real-time data, the system suffers from low efficiency and long processing time. Model training usually takes several hours or even longer, and it is difficult to achieve high throughput and low latency batch fault diagnosis analysis. This limits the system's ability and scale to conduct in-depth analysis of fuel cell health status, accurately predict remaining lifespan, and quickly trace faults. Summary of the Invention

[0007] The purpose of this invention is to provide a method and system for predicting the lifespan and diagnosing faults of intelligent collaborative high-power proton exchange membrane fuel cells, so as to solve the problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention provides the following technical solution:

[0009] A method for predicting the lifespan and diagnosing faults of intelligent collaborative high-power proton exchange membrane fuel cells includes the following steps:

[0010] S1. Real-time acquisition of multi-dimensional data of key areas of the fuel cell stack using a distributed sensor array. Core data include single cell voltage, electrochemical impedance spectroscopy, temperature, humidity, pressure and tailwater quality parameters.

[0011] S2. Remove noise and outliers from the collected data and extract features to obtain degradation features such as voltage inconsistency factor, impedance relaxation time, and temperature field entropy value.

[0012] S3. Construct a dual-model collaborative prediction framework: the lifetime prediction model integrates a time-series neural network and a physical mechanism model, and the fault diagnosis model is based on a graph neural network to construct fault propagation relationships and separate coupled faults such as flooding, gas starvation, and cooling failure.

[0013] S4. Input the real-time features into the dual model and output the remaining service life and fault type;

[0014] S5 generates maintenance strategies and early warnings, and uses transfer learning combined with online learning to transform diagnosed fault cases into adversarial examples to optimize neural network weights and dynamically update model parameters.

[0015] As a further aspect of the present invention: the sensor array in S1 is deployed in the inlet and outlet manifolds of the fuel cell stack, the active area of ​​the membrane electrode, and key nodes of the cooling channel to simultaneously collect spatial heterogeneous data.

[0016] As a further aspect of the present invention: in S2, feature extraction employs wavelet transform to analyze the phase angle abrupt change points of the impedance spectrum, combined with principal component analysis for dimensionality reduction.

[0017] As a further aspect of the present invention, the physical mechanism model in S3 includes: a catalyst activity decay model, a proton exchange membrane swelling-crack coupling model, and a gas diffusion layer micropore blockage evolution model.

[0018] As a further aspect of the present invention: the fault diagnosis model in S3 matches failure modes through a fault knowledge base, which contains the evolution path of water-air-heat coupling faults.

[0019] As a further aspect of the present invention: in S5: the dynamic update adopts transfer learning, which transforms the diagnosed and confirmed fault cases into adversarial examples to optimize the neural network weights.

[0020] This invention also discloses an intelligent collaborative high-power proton exchange membrane fuel cell lifetime prediction and fault diagnosis system, comprising:

[0021] Multi-source sensing module: includes a distributed voltage acquisition unit, a high-frequency impedance spectrometer, and a temperature and humidity scanning array, used to acquire data related to the fuel cell stack;

[0022] Edge computing module: integrates hardware accelerators for real-time calculation of electrochemical impedance spectroscopy features, and deploys data preprocessing and feature extraction algorithms;

[0023] Cloud analytics module: Provides lifespan prediction and fault diagnosis models, and processes data transmitted from the edge computing module;

[0024] Decision output module: Generates RUL reports, fault source trees, and maintenance priority lists.

[0025] As a further aspect of the present invention, the sensor array of the multi-source sensing module also has self-testing and calibration functions to ensure the accuracy and reliability of the collected data.

[0026] As a further aspect of the present invention: the decision output module includes a three-dimensional visualization interface that dynamically displays the membrane water content distribution, catalyst activity hotspots, and fault propagation paths.

[0027] Compared with the prior art, the beneficial effects of the present invention are:

[0028] 1. This invention significantly improves the processing speed and efficiency of high-frequency, multi-source sensor data locally by integrating dedicated hardware accelerators such as graphics processing units or field-programmable gate arrays into the edge computing module and deploying efficient data preprocessing and feature extraction algorithms. This effectively solves the problems of high data processing latency and poor real-time performance caused by insufficient edge computing capabilities in existing technologies. It enables millisecond-level feature extraction and response for minute voltage changes, rapid impedance spectrum signals, and temperature and humidity distributions in fuel cell single cells, providing a timely and reliable data foundation for early warning and accurate diagnosis of faults.

[0029] 2. This invention utilizes a distributed computing framework in the cloud analysis module to process massive feature data transmitted from the edge, and establishes a robust data storage management mechanism. It fully leverages the powerful parallel computing and elastic resources of cloud computing, effectively solving the problems of low efficiency and long processing time in existing technologies when training lifetime prediction models and inferring fault diagnosis models with large-scale historical and real-time data. It achieves a leapfrog improvement in model training time from hours to minutes, as well as high throughput and low latency batch fault diagnosis analysis, significantly enhancing the system's ability and scale to conduct in-depth analysis of fuel cell health status, accurately predict remaining service life, and quickly trace faults. Attached Figure Description

[0030] Figure 1 A schematic diagram of a method and system for predicting the lifespan and diagnosing faults of intelligent collaborative high-power proton exchange membrane fuel cells; Detailed Implementation

[0032] Please see Figure 1 In this embodiment of the invention, the intelligent collaborative high-power proton exchange membrane fuel cell lifetime prediction and fault diagnosis method includes the following steps:

[0033] S1. Real-time acquisition of multi-dimensional data of key areas of the fuel cell stack using a distributed sensor array. Core data include single cell voltage, electrochemical impedance spectroscopy, temperature, humidity, pressure and tailwater quality parameters.

[0034] In the deployment of distributed sensor arrays, it is necessary to accurately select key nodes of the fuel cell stack to ensure that the collected data can fully reflect the operating status of the fuel cell. For example, at the inlet and outlet manifolds, the sensors need to be closely attached to the pipe wall to capture minute changes in the gas flow process; in the active area of ​​the membrane electrode, the sensors should be evenly distributed to fully cover the active catalytic reaction area and accurately monitor key electrochemical parameters such as voltage and impedance.

[0035] S2. Remove noise and outliers from the collected data, use wavelet transform to analyze the phase angle mutation points of the impedance spectrum to extract features, and combine principal component analysis to reduce dimensions to obtain degradation features such as voltage inconsistency factor, impedance relaxation time, and temperature field entropy value.

[0036] In the data preprocessing stage, in addition to removing noise and outliers, the temporal sequence and correlation of the data also need to be considered. Since the fuel cell operation data is continuous and there is a strong correlation between adjacent data points, methods such as sliding windows can be used to smooth the data. At the same time, statistical analysis and other means can be used to identify and remove outliers caused by sensor failure or external interference to ensure the purity and reliability of the input model data.

[0037] When using wavelet transform to analyze phase angle abrupt changes in impedance spectrum, it is necessary to select appropriate wavelet basis functions and decomposition scales. Different wavelet basis functions have different adaptability to signals. Based on the characteristics of the impedance spectrum signal, such as frequency range and amplitude variation, a wavelet basis that can effectively capture phase angle abrupt change features, such as the Daubechies wavelet, should be selected. The decomposition scale should balance feature extraction accuracy and computational complexity. Under the premise of being able to distinguish key abrupt changes, the number of decomposition levels should be reduced as much as possible to avoid information redundancy.

[0038] During the dimensionality reduction process of principal component analysis, it is necessary to determine the number of principal components to be retained. This can be achieved by calculating the cumulative contribution rate, which is usually required to reach more than 90% to ensure that enough original data information is retained. At the same time, the physical meaning of the principal components should be interpreted to better correspond to the degradation characteristics of fuel cells and provide more interpretable features for subsequent model construction.

[0039] S3. Construct a dual-model collaborative prediction framework: the lifetime prediction model integrates a time-series neural network and a physical mechanism model, and the fault diagnosis model is based on a graph neural network to construct fault propagation relationships and separate coupled faults such as flooding, gas starvation, and cooling failure.

[0040] The physical mechanism models include: catalyst activity decay model, proton exchange membrane swelling-crack coupling model, and gas diffusion layer micropore blockage evolution model;

[0041] The fault diagnosis model matches failure modes through a fault knowledge base, which contains the evolution path of water-air-heat coupling faults.

[0042] When applying temporal neural networks in lifespan prediction models, the choice of network structure and training strategy must be considered. For example, network structures such as Long Short-Term Memory (LSTM) or Gated Recurrent Unit (GRU) have good memory and learning capabilities for sequential data and can effectively handle the mapping relationship between historical operating data and lifespan of fuel cells. During training, hyperparameters such as the number of network layers, number of neurons, and learning rate should be set appropriately, and methods such as cross-validation should be used to avoid overfitting and improve the generalization ability of the model.

[0043] Regarding physical mechanism models, catalyst activity decay models need to be based on detailed chemical reaction kinetics, considering the influence of various factors such as catalyst surface area, number of active sites, and reactant concentration, to establish accurate mathematical expressions that describe the change in catalyst activity over time. Proton exchange membrane swelling-crack coupling models must combine material mechanical properties and electrochemical environment to analyze the interaction mechanism between mechanical stress caused by swelling and crack initiation and propagation, constructing a model that reflects membrane structural integrity and performance degradation. Micropore blockage evolution models for gas diffusion layers should consider the deposition process of liquid water, reaction products, and other substances within the micropores, as well as their blockage effect on gas diffusion channels, establishing a model of the relationship between changes in micropore structure and decrease in mass transfer performance, providing strong support for the accuracy and reliability of the models.

[0044] When constructing fault propagation relationships based on graph neural networks, the fault diagnosis model first needs to build a fault propagation graph, clarifying the definitions of nodes and edges. Nodes can represent different components or systems of the fuel cell, such as the stack, gas supply system, and cooling system; edges represent the fault propagation paths and coupling relationships between components. By analyzing historical fault data and expert knowledge, the probability and weight of fault propagation are determined. The learning capability of graph neural networks is then utilized to uncover potential fault propagation patterns, achieving accurate separation of coupled faults.

[0045] S4. Input the real-time features into the dual model and output the remaining service life and fault type;

[0046] When inputting real-time features into dual models, it is necessary to ensure the synchronization and consistency of the data. Since the models have certain requirements on the format and range of the input data, it is necessary to perform preprocessing operations such as normalization on the real-time feature data to make it conform to the input specifications of the model, thereby ensuring the accuracy and reliability of the model output results.

[0047] S5. Generate maintenance strategies and early warnings, and use transfer learning combined with online learning to transform diagnosed fault cases into adversarial examples to optimize neural network weights and dynamically update model parameters.

[0048] Dynamic updates employ transfer learning, transforming diagnosed and confirmed fault cases into adversarial examples to optimize neural network weights;

[0049] In the dynamic update mechanism that combines transfer learning and online learning, transfer learning needs to address the differences between the source and target domains. When converting diagnosed and confirmed fault cases into adversarial examples, the diversity and representativeness of the fault samples must be fully considered. Through adversarial training, the model can be optimized and adjusted in a targeted manner based on the original weights to enhance the model's ability to identify new fault modes. At the same time, online learning needs to track the changes in the operating status of the fuel cell in real time and update the model parameters in a timely manner to ensure that the model can always accurately predict lifespan and diagnose faults, adapting to the operating requirements of the fuel cell under different operating conditions and environments.

[0050] A smart collaborative high-power proton exchange membrane fuel cell lifetime prediction and fault diagnosis system includes:

[0051] Multi-source sensing module: includes a distributed voltage acquisition unit, a high-frequency impedance spectrometer, and a temperature and humidity scanning array, used to acquire data related to the fuel cell stack;

[0052] The sensor array of the multi-source sensing module also has self-testing and calibration functions to ensure the accuracy and reliability of the collected data;

[0053] When designing and selecting units within the multi-source sensing module, such as the distributed voltage acquisition unit, high-frequency impedance spectrometer, and temperature and humidity scanning array, the measurement accuracy, sampling frequency, and range must be determined based on the size, power rating, and operational requirements of the fuel cell. For example, the distributed voltage acquisition unit should have high-precision analog-to-digital conversion capability to accurately measure the minute voltage changes of each individual cell; the high-frequency impedance spectrometer should be able to quickly and stably apply and measure high-frequency AC signals to obtain accurate impedance spectrum data; and the temperature and humidity scanning array should have good temperature and humidity resolution, as well as fast response capability, to monitor the temperature and humidity distribution within the fuel cell stack in real time.

[0054] Edge computing module: integrates hardware accelerators for real-time calculation of electrochemical impedance spectroscopy features, and deploys data preprocessing and feature extraction algorithms;

[0055] The hardware accelerators integrated into the edge computing module should select appropriate chip architectures and processing capabilities based on the computational requirements of the deployed data preprocessing and feature extraction algorithms. For example, for complex wavelet transform and principal component analysis algorithms, hardware accelerators such as graphics processing units (GPUs) or field-programmable gate arrays (FPGAs) can be used to improve the running speed and efficiency of the algorithms, ensuring that data can be processed and analyzed in a timely manner at the edge, and providing effective feature data for subsequent cloud computing.

[0056] Cloud analytics module: Provides lifespan prediction and fault diagnosis models, and processes data transmitted from the edge computing module;

[0057] When the cloud analytics module runs the lifespan prediction model and fault diagnosis model, it should utilize the powerful computing resources of cloud computing to perform in-depth analysis and processing of data transmitted from the edge. This module can adopt distributed computing frameworks, such as Hadoop or Spark, to meet the parallel computing needs of large-scale data and improve the training and prediction performance of the model. At the same time, it is necessary to establish a sound data storage and management mechanism to classify, store and index historical and real-time data to facilitate model training and query.

[0058] Decision output module: Generates RUL report, fault source tree and maintenance priority list; The decision output module includes a 3D visualization interface that dynamically displays membrane water content distribution, catalyst activity hotspots and fault propagation paths;

[0059] The RUL report, fault tracing tree, and maintenance priority list generated by the decision output module should be presented to users in an intuitive and easy-to-understand way. When the 3D visualization interface dynamically displays the distribution of membrane water content, catalyst activity hotspots, and fault propagation paths, it can use advanced graphics rendering technology and interactive design to enable users to clearly observe the physicochemical processes and fault evolution inside the fuel cell, providing strong support for the formulation of reasonable maintenance strategies.

[0060] To further illustrate the technical effects of the present invention, the following experiments were conducted for verification:

[0061] Experiment 1

[0062] Experimental objective: To verify the significant speed and efficiency improvements of edge computing modules integrating GPU / FPGA hardware accelerators in processing high-frequency impedance spectrum feature extraction tasks (such as wavelet transform + principal component analysis) compared to traditional edge processors (such as general-purpose CPUs).

[0063] Experimental setup:

[0064] Control group: Edge devices use only high-performance multi-core CPUs (such as Intel Xeon E series).

[0065] Experimental group: The same edge device integrates NVIDIA Jetson AGX Orin (GPU) or Xilinx Zynq UltraScale+ (FPGA).

[0066] Software: Deploy the same feature extraction algorithm (wavelet transform + PCA).

[0067] Data: High-frequency impedance spectrum data samples (complex arrays) of different lengths (representing different sampling durations / frequencys) were generated by simulation. Each set of data was tested 10 times and the average was taken.

[0068] Metric: Execution time of a single feature extraction task (ms).

[0069] Experimental Table:

[0070]

[0071] Conclusion: Experimental data clearly demonstrate that integrating GPU or FPGA hardware accelerators significantly improves the speed of feature extraction from high-frequency impedance spectroscopy data (especially when processing large datasets, GPU / FPGA is tens of times faster than CPU). This effectively addresses the real-time bottleneck of processing complex algorithms at the edge, ensuring timely capture and analysis of minute changes and rapid signals, thus providing a foundation for subsequent rapid diagnosis and early warning.

[0072] Experiment 2:

[0073] Experimental objective: To verify the significant speed and throughput advantages of the cloud-based analytics module using the Spark distributed computing framework compared to the traditional single-machine server architecture when performing large-scale fuel cell life prediction model training and batch fault diagnosis tasks.

[0074] Experimental setup:

[0075] Control group: High-performance single-machine server (e.g., dual-socket Intel Xeon Platinum, 512GB RAM, 2x NVIDIA A100 GPU).

[0076] Experimental group: Spark cluster (e.g., 1 Master + 4 Workers, with each Worker configured as a single machine as the control group).

[0077] Task A (Model Training): Train the same deep learning lifetime prediction model (such as LSTM or Transformer network) using the same historical feature dataset (containing voltage, EIS features, temperature and humidity features, and RUL labels). The dataset size is variable.

[0078] Task B (Batch Diagnosis): Using the same trained model, perform fault diagnosis and prediction on a batch of new real-time / historical feature data.

[0079] Data: Use real or simulated fuel cell aging / failure datasets with approximately 100 feature dimensions. Adjust the dataset size.

[0080] index:

[0081] Task A: Time (minutes) required to complete model training.

[0082] Task B: Average latency (ms) and throughput (samples / second) for processing 1000 samples.

[0083] Experimental Table:

[0084] Table 1: Comparison of Model Training Time (Task A)

[0085]

[0086] Table 2: Performance Comparison of Batch Fault Diagnosis (Task B)

[0087] Computing architecture Average latency (ms) for processing 1000 samples Throughput (samples / second) Control group (single machine) 1850 540 Experimental group (Spark cluster) 420 2380

[0088] in conclusion:

[0089] Model training: Spark distributed clusters demonstrate powerful parallel capabilities when processing large-scale training data, with training speeds several times faster than high-performance single machines, and this advantage becomes more pronounced as the amount of data increases. This allows models to be updated and iterated more frequently and quickly on massive amounts of new data.

[0090] Batch diagnostics: When processing diagnostic tasks in batches, the Spark cluster significantly reduces latency (by over 77%) and greatly increases throughput (by over 340%). This means the system can perform fault diagnosis and analysis on large amounts of battery cells or historical data more quickly, meeting the needs of near real-time monitoring and efficient backtracking. This significantly improves the efficiency and scalability of the entire system in lifespan prediction and fault diagnosis.

[0091] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0092] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for predicting the lifespan and diagnosing faults of intelligent collaborative high-power proton exchange membrane fuel cells, characterized in that: Includes the following steps: S1. Real-time acquisition of multi-dimensional data of key areas of the fuel cell stack using a distributed sensor array. Core data include single cell voltage, electrochemical impedance spectroscopy, temperature, humidity, pressure and tailwater quality parameters. S2. Remove noise and outliers from the collected data and extract features to obtain degradation features such as voltage inconsistency factor, impedance relaxation time, and temperature field entropy value. S3. Construct a dual-model collaborative prediction framework: the lifetime prediction model integrates a time-series neural network and a physical mechanism model, and the fault diagnosis model is based on a graph neural network to construct fault propagation relationships and separate coupled faults such as flooding, gas starvation, and cooling failure. S4. Input the real-time features into the dual model and output the remaining service life and fault type; S5 generates maintenance strategies and early warnings, and uses transfer learning combined with online learning to transform diagnosed fault cases into adversarial examples to optimize neural network weights and dynamically update model parameters.

2. The intelligent collaborative high-power proton exchange membrane fuel cell lifetime prediction and fault diagnosis method according to claim 1, characterized in that, The sensor array in S1 is deployed at key nodes of the fuel cell stack's inlet and outlet manifolds, membrane electrode active area, and cooling channel to simultaneously collect spatial heterogeneous data.

3. The intelligent collaborative high-power proton exchange membrane fuel cell lifetime prediction and fault diagnosis method according to claim 1, characterized in that, In S2: Feature extraction uses wavelet transform to analyze the phase angle abrupt change points of the impedance spectrum, and combines it with principal component analysis for dimensionality reduction.

4. The intelligent collaborative high-power proton exchange membrane fuel cell lifetime prediction and fault diagnosis method according to claim 1, characterized in that, The physical mechanism models in S3 include: a catalyst activity decay model, a proton exchange membrane swelling-crack coupling model, and a gas diffusion layer micropore blockage evolution model.

5. The intelligent collaborative high-power proton exchange membrane fuel cell lifetime prediction and fault diagnosis method according to claim 1, characterized in that, The fault diagnosis model in S3 matches failure modes through a fault knowledge base, which contains the evolution path of water-air-heat coupling faults.

6. The intelligent collaborative high-power proton exchange membrane fuel cell lifetime prediction and fault diagnosis method according to claim 1, characterized in that, In S5: Dynamic updates employ transfer learning, transforming diagnosed and confirmed fault cases into adversarial examples to optimize neural network weights.

7. A smart collaborative high-power proton exchange membrane fuel cell life prediction and fault diagnosis system, used to implement the method described in any one of claims 1-6, characterized in that, The system includes: Multi-source sensing module: includes a distributed voltage acquisition unit, a high-frequency impedance spectrometer, and a temperature and humidity scanning array, used to acquire data related to the fuel cell stack; Edge computing module: integrates hardware accelerators for real-time calculation of electrochemical impedance spectroscopy features, and deploys data preprocessing and feature extraction algorithms; Cloud analytics module: Provides lifespan prediction and fault diagnosis models, and processes data transmitted from the edge computing module; Decision output module: Generates RUL reports, fault source trees, and maintenance priority lists.

8. The intelligent collaborative high-power proton exchange membrane fuel cell life prediction and fault diagnosis system according to claim 7, characterized in that, The sensor array of the multi-source sensing module also has self-testing and calibration functions to ensure the accuracy and reliability of the collected data.

9. The intelligent collaborative high-power proton exchange membrane fuel cell life prediction and fault diagnosis system according to claim 7, characterized in that, The decision output module includes a three-dimensional visualization interface that dynamically displays the membrane water content distribution, catalyst activity hotspots, and fault propagation paths.

Images