Reconstruction of Cylinder Pressure of I.C. Engine Based on Neural Networks

In this paper, the characteristics of the excitations and the vibration responses of cylinder head are analyzed in detail. The methods of reconstructing cylinder pressures are investigated. To avoid some shortcomings of traditional linear methods, a new reconstructing cylinder pressure method by using ANN is presented in this paper. A standard BP neural network was trained with the measured cylinder head vibration signals as the input and with the measured cylinder pressures as the ideal output of the NN. To solve the problem of converging at slow velocity, it adopts the varying-step algorithm, which adopts a larger step at the beginning of the algorithm, and gradually decreases step as the converging of process. The comparison results are presented when the test engine operates with different loads and at different speed. The results show that the trained network can reconstruct cylinder pressures effectively when the engine operates at different operation states. It is of good repeatability and of good resolution to identify cylinder pressure with NN.