Dynamic Modeling, Controlling And Fault Diagnosis Of Thermal Management System In Proton Exchange Membrane Fuel Cell

Proton Exchange Membrane Fuel Cell system comprises of the following parts: PEMFC stack, Hydrogen Supply System, Air Supply System, Water and Thermal Management System and Control System, etc. Among them, Thermal Management System, as one of the key parts that determines the whole system's performance and efficiency, is to balance the heat in the stack. Base on conditions above, models, controlling strategies and fault diagnosis of thermal management in PEMFC system were studied in the paper, the main contents are shown as follows:1,Gas distribution, thermal generation and transferring are interacting with each other closely in PEMFC. According to the continuity equation and law of mass conservation, dynamic models of water and gas transmission from anode, cathode and membrane in PEMFC were established; in addition, models of stack temperature and cooling water temperature were also developed on the basis of energy conservation theory. By this method of modeling, we could avoid studying complicated processes of thermal transmission that take place in cells or cells between, so the whole models are simplified a lot and much more appropriate for real-time controlling.2,With the previous models such as multi-components two-phase transmission models, stack temperature model and cooling water temperature model, a nonlinear robust controller by the method of Lyapunov function reverse recursion was proposed to stabilize the stack temperature to some extent. And then, on the platform of Matlab/Simulink, effectiveness of the whole control strategy was verified.3,As thermal management system of PEMFC has the strong characteristics of non-linearity, unstability and uncertainty, etc. A fault diagnosis method based on BP neural network was presented by analyzing the structure and common faults of thermal management system. What's more, the pros and cons of various improved BP algorithm were discussed. Finally, the whole strategy of fault diagnosis was testified well while the training error was set at 0.01.