Abstract: In order to solve the problem of engine misfire fault, a diagnosis method based on cylinder pressure identified by Fourier Transform (FT) and L-M optimized BP neural network is proposed. The BP neural network is trained by the steady state simulation data from AMESim to obtain the relationship between open value of valve and frequency, and FT is used to identify the cylinder pressure. The misfire failure can be diagnosed by comparison of the FT identified pressure with the pressure mapped from crank speed. By offsetting the phase and frequency of the identified model, the accuracy and generalization of the identified model are improved. When misfire fault occurs, high tracking ability can be obtained by re-offsetting the phase and frequency of the identified model. Two random and independent open values of the valve are used to verify the proposed method, Results show that no matter engine at high speed with light load condition or at low speed with heavy load condition, this method can precisely diagnose the signal-cylinder continuous misfire fault and multi-cylinder random misfire fault.

Key words: automatic control technology, Fourier series, BP neural network, engine misfire fault, engine cylinder pressure, AMESim