Study On Engine Fault Diagnosis Based On Multi-Information Fusion

As a typically reciprocating engine power machine, complex structure of the engine determines its failure shows the characteristics of complexity and diversity. The impact of uncertainties, induced by such as operating environment, system noise and sensor accuracy, results in reducing the accuracy of engine fault diagnostic. Multi-information fusion provides a new way to solve the complex structure of the engine and the problems caused by these uncertainties. In this context, this paper will study engine fault diagnosis based on multi- information fusion.At first, this paper describes the currently domestic and foreign studies of the engine fault diagnosis and the typical method, and the necessity and feasibility of applying multi- information to the engine fault diagnosis.Secondly, the paper emphasis on the three typical methods of multi-information fusion: D-S evidence theory, fuzzy set theory, artificial neural networks. The article will compare various methods improving the BP algorithm by its application to fault diagnosis of engines, to verification the superiority of the conjugate gradient method. In order to avoid the inaccuracy brought by a single sensor information, the decision level fusion of engine fault diagnosis by combining the improved BP neural network with the D-S evidence theory fusion, have greatly improved the reliability of information.Finally, as the D-S evidence theory can not resolve the evidence conflict, this paper presents the weighted evidence theory; as the determination of membership function is complicated, the paper combines the improved BP neural network and fuzzy set theory. Comparing these fusion methods, we can see the decision level fusion of engine fault diagnosis by combination of improved BP neural network and the D-S evidence is more accurate diagnosis, but the calculation is larger than fuzzy integration.