Research On Fault Diagnosis Of Sensor And Actuator Of Fuel Cell Engine Based On Generation Countermeasure Network

Proton Exchange Membrane Fuel Cell(PEMFC)engines convert the chemical energy of hydrogen and oxygen into electricity to produce water,which has the advantages of fast start-up speed and zero carbon emissions.The sensor is the "eye" of the PEMFC engine,and the actuator is the "hand and foot",both of which play a pivotal role in hydrogen fuel cell engines.PEMFC engine has a complex reaction mechanism,sensor and actuator long-term work in high temperature,high pressure,external environmental impact,electromagnetic environment and other harsh environments,easy to lead to failure,light lead to the performance of the stack,heavy lead to complete damage to the stack.Aiming at the problem that most of the PEMFC engines are in normal operation and the fault data sets of sensors and actuators are difficult to obtain,an auxiliary classifier generative adversarial network based on one-dimensional convolutional neural network is proposed,and the fault characteristics of data are mined through the mutual game of generators and discriminators,simulation data is generated,fault samples are expanded,and fault diagnosis of sensors and air compressors based on a small number of fault samples is realized,and experimental verification is carried out.The main research contents are as follows:The fault characteristics of PEMFC engine sensors and actuators were analyzed,and the corresponding mathematical models were established for the typical faults of sensors and air supply systems.Study of fault characteristics of engine sensors and actuators based on the PEMFC public model of Michigan State University.In the model,the air compressor jamming,net power sensor drift,voltage sensor bias and peroxide ratio sensor short-circuit faults are set to extract the corresponding fault characteristics.Aiming at the problem that the actual fault data set is difficult to obtain,the fault data set is generated based on model simulation,and Gaussian white noise is added to simulate the actual fault.An auxiliary classifier generative adversarial network based on one-dimensional convolutional neural network is proposed for data enhancement of unbalanced classes.The generated fault data is evaluated by using Euclidean distance,PCC,and KL divergence,and the fault data set is expanded and diagnosed.Using the measured healthy voltage data set,the bias fault is added and the voltage sensor bias fault is simulated,and the proposed generative adversarial network model is verified.Comparative analysis was carried out with CGAN and DCGAN to verify the accuracy of the model.