AUTOMATED DIESEL ENGINE CONDITION & PERFORMANCE MONITORING & THE APPLICATION OF NEURAL NETWORKS TO FAULT DIAGNOSIS

Abstract

The overall aim of this research was to design, configure and validate a system which was capable of on-line performance monitoring and fault diagnosis of a diesel engine. This thesis details the development and evaluation of a comprehensive engine test facility and automated engine performance monitoring package. Results of a diesel engine fault study were used to ascertain commonly occurring faults and their realistic severities are discussed. The research shows how computer simulation and rig testing can be applied to validate the effects of faults on engine performance and quantify fault severities. A substantial amount of engine test work has been conducted to investigate the effects of various faults on high speed diesel engine performance. A detailed analysis of the engine test data has led to the development of explicit fault-symptom relationships and the identification of key sensors that may be fitted to a diesel engine for diagnostic purposes. The application of a neural network based approach to diesel engine fault diagnosis has been investigated. This work has included an assessment of neural network performance at engine torques and speeds where it was not trained, noisy engine data, faulty sensor data, varying fault severities and novel faults which were similar to those which the network had been trained on. The work has shown that diagnosis using raw neural network outputs under operational conditions would be inadequate. To overcome these inadequacies a new technique using an on-line diagnostic database incorporating 'weight adjusting' and 'confidence factor' algorithms has been developed and validated. The results show a neural network combined with an on-line diagnostic database can be successfully used for practical diesel engine fault diagnosis to offer a realistic alternative to current fault diagnosis techniques.