To accurately identify the fault types of proton exchange membrane fuel cell
(PEMFC) systems under continuously varying operating currents, this study
develops a comprehensive PEMFC system model and proposes a robust fault
diagnosis method based on the ResNet50 convolutional neural network (CNN) and
transfer learning (TL). Initially, using Matlab/Simulink, a PEMFC model is
constructed based on the electrochemical reaction mechanisms and empirical
formulas that characterize the operation of the fuel cell. This model primarily
includes the fuel cell stack and various auxiliary systems, such as the thermal
management system, air supply system, and hydrogen supply system, each crucial
for optimal performance. By varying the model parameters, sensor data is
generated for five distinct operating conditions. After preprocessing the data,
the Gramian Angular Field (GAF) technique is utilized to convert the time series
data from each sensor into fault data images, which then serve as input for the
ResNet50 CNN. Ultimately, the implementation of transfer learning involves
utilizing the pre-trained weights of the ResNet50 model in the training process
of this model. This approach aims to improve both the convergence rate and the
generalization capacity of the classification model. A comprehensive dataset for
fault diagnosis has been established, comprising a total of 4,000 samples, with
800 image samples generated for each distinct operating state. The diagnostic
results demonstrate that the integrated PEMFC system attains an exceptional
diagnostic accuracy of 100.0% across five distinct operational scenarios:
standard operating conditions, reduced air pressure at the compressor inlet,
increased air temperature at the compressor inlet, heightened stack temperature,
and an obstructed anode gas supply line. These results demonstrate that the
proposed method not only exhibits high classification accuracy but also displays
remarkable robustness in fault diagnosis applications.