Research On Intelligent Fault Diagnosis Of Proton Exchange Membrane Fuel Cell Based On Data-Driven Techniques

Green and low-carbon is the current trend of global energy development,in the transition process of energy system in China to "clean,low carbon,smart",hydrogen energy as the most potential secondary green energy occupies the vital position.As a major hydrogen energy consumer in the hydrogen industry chain,Proton Exchange Membrane Fuel Cell(PEMFC)shows a broad application prospect in many fields such as transportation,industry,construction and electric power by virtue of its advantages such as zero carbon emission,low temperature and cold start-up,high power density and low noise.However,it is difficult to promise the stability and reliability of PEMFC key parameters in the whole life cycle due to the complex and changeable operating conditions in the actual application scenario,and the change of PEMFC parameters will cause the failure modes,which will seriously affect the safety and durability of PEMFC.Therefore,in order to ensure the safe and reliable operation of PEMFC,the accurate and efficient fault diagnosis method of PEMFC in service needs to be studied urgently.Due to the close relationship between the voltage and current distribution of PEMFC and the internal PEMFC operating state,there existed many relevant studies to explore the application effect of voltage and current distribution in the fault diagnosis of PEMFC.However,there are still some difficulties that hinder the application of the two signals in PEMFC fault diagnosis.In view of the difficulties of voltage data fault characteristics,which are mainly dependent on PEMFC fault diagnosis,such as poor differentiation,weak generalization and poor transferability,as well as the difficulties in collecting current distribution data in PEMFCs,this paper takes hydrogen fuel cells as the object and conducts in-depth research on data-driven intelligent fault diagnosis method.The main research contents are as follows:1.Aiming at the problems of weak distinguishing and poor generalization when PEMFC voltage is used in fault diagnosis,a fault diagnosis method for hydrogen fuel cells based on voltage signal multi-scale feature extraction and multi-level feature fusion is proposed in this paper.By using multi-scale convolution layer to extract multi-scale features of voltage signals,and introducing dense connected network structure to realize multi-level feature fusion and relearning,this method not only improves the adaptability of the network to different scale features of voltage signals,but also makes use of the complementarity of low-level features and high-level features in the network.The precise capture of the subtle features and the comprehensive characterization of voltage signals are realized,thus improving the differentiation and generalization of voltage features.By collecting voltage data of different PEMFC under various faults based on experimental verification,the accuracy of feature representation and fault diagnosis of the proposed method is verified.2.In order to solve poor transferability of voltage fault features caused by the distribution difference between the PEMFC single cell voltage and the stack voltage,and the asymmetry of the class space,and the difficulty of mining the common characteristics of the voltage signals of a PEMFC cell and stack,this paper proposes a PEMFC fault diagnosis method of unsupervised transfer learning based on network adversarial and adaptive conditional distribution.Firstly,by combining the domain classifier with the additional source domain discriminator,the pseudo-decision boundary between the known type data and the unknown type data is determined,and the unknown fault mode data in the stack is accurately identified and excluded,so as to reduce the influence of negative migration caused by class space asymmetry.Secondly,the feature marginal distribution and conditional distribution of shared type data in single cell and stack are adaptively aligned by using network adversarial and conditional distribution adaptive,realizing accurate extraction of common diagnosis knowledge in voltage data of single cell and stack,improving the knowledge transferability from single cell to stack,and ensuring the reliability of unsupervised fault diagnosis of reactor.Based on the experimental verification,by collecting multiple PEMFC health voltage signals from a single cell and stack to construct a database,the accuracy of the proposed method in unsupervised stack fault diagnosis under the condition of open set data and its adaptability to different degrees of data openness are verified.3.Due to the difficulty of detecting the internal current distribution in PEMFC,a fault diagnosis method of PEMFC water management based on the current distribution is difficult to realize.This paper proposes a fault diagnosis method of PEMFC water management based on external magnetic field sensing.Firstly,the spatial and temporal mapping relationship between the internal current distribution and the external magnetic field is established.and then the magnetic field distribution evolution rule under water management fault is studied by simulation model and experimental verification.In addition,a PEMFC non-invasive magnetic field detection device is established to collect the information of the external magnetic field during the PEMFC running.The fault diagnosis features of PEMFC based on the magnetic field data are proposed,from which the mapping relationship between the magnetic field data features and the fault modes is established accurately.Based on experimental verification,the accuracy of the proposed method in water management fault diagnosis and fault degree tracking is verified by collecting the external magnetic field sensing data of PEMFC in health state and water management fault state.