Research On State Estimation And Fault Diagnosis Of Solid Oxide Fuel Cell

Solid oxide fuel cell(SOFC)has become one of the most anticipated new energy sources due to its advantages of high energy efficiency,low emission and low noise.At present,the commercial application of SOFC mainly faces the problems of high cost and short life.Fault diagnosis can detect and isolate faults in system operation in time,which is of great significance to improve system durability and reduce maintenance costs.However,SOFC does not only have a single fault,but more often multiple faults occur at the same time.Due to the coupling effect of multiple faults and the problem of mixed characteristics,fault diagnosis becomes extremely challenging.In addition,the hightemperature and sealed operating conditions also make it impossible to directly measure important states affected by faults inside the stack,which severely restricts the research on SOFC fault diagnosis.For this reason,Thesis focuses on the state estimation and fault diagnosis of SOFC,focusing on solving the multi-fault diagnosis problem of SOFC system.The contents involved mainly include:(1)The fault model of SOFC system based on air compressor fault,stack leakage fault and electrode delamination fault is established.Firstly,the fault mechanism of air compressor fault,stack leakage fault and electrode delamination fault among the common faults of SOFC system is analyzed,and then the SOFC peripheral equipment,stack and fault sub-model are mathematically modeled based on electrochemical and thermal principles.Finally,the V-I and P-I characteristic curves of the built model are verified and the state sensitivity analysis under fault occurrence is carried out.(2)A high-order sliding mode observer and an adaptive observer are designed for the temperature and gas mole fraction inside the stack,which are greatly affected by faults and difficult to measure directly.Based on the stack node model and the Super-Twisting Algorithm,a high-order sliding film observer is designed,which can estimate the internal temperature of the stack and the distribution of gas mole fraction along the flow channel.In addition,considering the input uncertainty and the difficulty of accurate measurement in engineering practice,an adaptive observer with input parameters is designed to estimate the temperature and gas mole fraction inside the stack.The simulation results show that both observers can make the maximum relative error of the temperature estimation within 2.61% under the load step change,but the estimation effect of the adaptive observer is better than that of the high-order sliding mode observer under the input disturbance and measurement error.(3)In order to solve the problem of fault coupling and mixed features in SOFC multifault diagnosis,a fault decoupling diagnosis method based on capsule network is proposed.Firstly,the powerful feature extraction ability of convolutional neural network is used to extract features of data with high fault sensitivity,and then the fault decoupling diagnosis model of SOFC system is constructed by using the decoupling characteristics of capsule network.Finally,based on conventional convolutional neural network,Multilabel Convolutional Neural Networks and Capsule Networks are tested for fault diagnosis.The simulation results show that on a single fault,the total accuracy of the three diagnostic methods has reached 98%,and the capsule network even reached 98.8%.In compound fault diagnosis,the total accuracy rate of decoupling diagnosis based on capsule network can reach 96.7% without compound fault training.