Fault Detection and Diagnosis based upon Elemental Analysis of Engine Lubrication Oil

Many systems, such as aeroplanes, military projects and new cars are being built and designed around the fault diagnosis system, and in such instances the concept of fault diagnosis is put on equal parity with the system itself; either for the purposes of safety, performance or as a good selling point. On the other hand, there still remain a large number of mechanisms such as diesel engines which require an accurate and simple method for retrofitting a reduced form of a fault diagnosis system. The most fundamental problem concerns the method for inferring the state of the engine, whether that be through online methods such as vibration analysis or thermography, or offline methods such as elemental analysis. The concept of oil analysis is defined, and supported by relevant literature regarding the methods utilised by previous research. Secondly, the dissertation is concerned with the lack of sensitivity demonstrated in conventional approaches to incipient faults characterised in the time series. The dissertation presents a number of methods that have been developed including prediction and classification techniques which are evaluated against a conventional approach. Thirdly, the thesis approaches the question of inference from sparse datasets, and how this can be best addressed using a Bayesian belief network. Further, a method of learning the relationships from the distribution of the data is presented using the Expectation Maximisation algorithm. Finally, the dissertation presents a novel framework in fault detection and diagnosis for the early detection and inference of faults in a number of different diesel engines. The process to achieve the final approach is documented as well as supported by intermediate results that demonstrate the validity of the research path taken; where case studies are implemented utilising data from diesel train engines. The complete framework is applied to three datasets obtained from three train diesel engines of the same model which are evaluated for diagnosis accuracy and sensitivity to incipient faults; outperforming the conventional industrial approach that is often used in fault diagnosis.