A high-level technique for diesel engine combustion system condition monitoring and fault diagnosis

Abstract

This paper presents a technique for diagnosis of a class of engine faults, which adversely affect the combustion efficiency of a diesel generator set. The diagnosis is made by combining the evidence of two separate estimations of engine load with the outputs from a predictive fuzzy model of engine speed. Certain faults affect the periodicity of the engine speed signal. The variation in periodicity means that the load estimation from power spectral density of the speed signal is not robust. This problem is countered by implementing a reference model to predict speed fluctuations with respect to crank angle. This model has the additional benefit that its output may be used to detect periodic fault symptoms in the speed signal, thereby leading to identification of the individual cylinder affected by the fault. A fuzzy rule based system has been developed to diagnose faults based on the load estimations and the residuals obtained from the reference model. Testing the diagnostic system with data from the normal, and two other separate engine fault conditions, resulted in a classification success rate greater than 90 per cent in each case. A further benefit is reported where the combination of evidence from the three sources effectively validates the load estimation, which may then be used to infer faults in other subsystems.