Application of Two-level Information Fusion Technology in the Diesel Engine Malfunctions Diagnosis
Wong Jian-hua · Journal of Academy of Armored Force Engineering · 2010
Aiming at the disadvantages of the net being huge in the result of too much testing point information and converging being difficult of the traditional neural network, this paper proposes the composite neural network to enhance the efficiency of fusion diagnosis. At the same time, this paper proposes the D-S evidence theory which is a decision-making fusion method to solve the problem of inconsistent diagnosis result from each subnet. In the application of diesel engine malfunction diagnosis, firstly the wavelet packet AR model spectrum is applied to pick up the characteristic frequency strip energy of normal and fault samples, and these characteristic parameters are put into the subnet of the composite neural network to diagnosis the faults. If the neural network cannot confirm the diagnosis result, the evidence theory is applied to take decision-making fusion and output the final diagnosis result. The experiment indicates that the two-level synthesis model of integrating the composite neural network and the D-S evidence theory can enhance the accuracy and the stability of diagnosis effectively.