The research on real-time fault diagnostic method of engine embedded by GA optimization and BP neural networks

Embedded system real time measurement and representation method is widely used in industrial CPS control system application, such as engine fault diagnosis. The malfunction diagnostic methods, which filter the changes in measurable parameters through Kalman filter and then map them to changes in performance parameters of the engine via neural network, have the weaknesses in that the neural network has a slow convergence rate and the optimum is prone to local optimum. Therefore, proposes an approach using genetic algorithm, which stands out in finding global optimum, to refine BP neural network. Our approach uses global optimum to refine the initial and threshold values of the neural network, decreasing the times of neural network trainings, and thus reducing the training error and lowering the time of diagnosis. The simulation shows that the approach is feasible, improves the real-timing and error precision of malfunction diagnosis, and therefore is highly valuable in applications.