Research On Water Management Fault Diagnosis Of Vehicular Proton Exchange Membrane Fuel Cell Power System Based On Data-driven Model

Proton Exchange Membrane Fuel Cell（PEMFC）has the advantages of environmental friendliness,low temperature operation,high energy density and so on,which has a wide application prospect in the field of medium and large vehicular power systems.In operation,the water content in PEMFC is the key to its performance and service life.Due to the multi-mechanism water transport characteristics of proton exchange membrane and the complexity of dynamic phase change of water inner the cell.Water management related issues have long been the focus and difficulty of fuel cell research.The water management fault caused by improper water management strategy will lead to unstable output power of the system in the short term and irreversible attenuation or even failure of PEMFC in the long term,which will seriously affect the reliability and durability of vehicular fuel cell system.Current researches on water management fault diagnosis of PEMFC mainly focus on small and medium-sized power systems,while the research on water management fault diagnosis of high-power systems such as vehicular PEMFC power system is less.Therefore,it is of great significance to study water management fault diagnosis methods of vehicular PEMFC power system.The thesis adopts the fault diagnosis method of PEMFC water management based on data-driven model.The main work includes three parts: experimental design of water management fault,construction of data-driven model and construction of water management fault classifier,which is introduced as follows:（1）The water management fault experiment of common power points is designed and completed on PEMFC power system experimental platform.Determine that state label of the experimental data set according to the experimental phenomenon.Three key indicators: output voltage,voltage uniformity and anode pressure drop which can be used for fault diagnosis are obtained by analyzing the experimental data.The typical modes of water management fault generation and elimination of PEMFC power system for vehicles are summarized.（2）Establish a data-driven model for fault diagnosis for the key indicators which obtained in the experiment.Selecting an appropriate neural network model based on the physical characteristics of the modeling object;In order to reduce model complexity and improve model accuracy,on the one hand,selecting input features with reference to the mechanism model to reduce the dimension of input features;On the other hand,single cells with adjacent positions and similar single cell voltage performance are classified into the same group,and the dimension reduction of voltage model output is realized by grouping.It is proved that this model has good regression performance and versatility.（3）Building a water management fault classifier based on binary logic.The input of the classifier is the residual error between the model output and the actual output,and the output is the diagnosis result.The diagnosis logic of the classifier is built according to the typical mode of water management fault.The classification of five related states of water management faults is realized.Considering that the output of classifier based on binary logic is discrete value,the intermediate state and fault degree cannot be described.Therefore,fuzzy logic is further applied to the construction of classifier,which better describes the intermediate state and fault degree.It has been verified that the fault diagnosis method described in the thesis avoids diagnosis errors that may cause serious consequences,and the overall diagnosis accuracy rate is 95%.