An intelligent engine condition monitoring system

Abstract

The main focus of the work reported here is in the design of an intelligent condition monitoring system for diesel engines. Mechanical systems in general and diesel engines in particular can develop faults if operated for any length of time. Condition monitoring is a method by which the performance of a diesel engine can be maintained at a high level, ensuring both continuous availability and design-level efficiency. A key element in a condition monitoring program is to acquire sensor information from the engine, and use this information to assess the condition of the engine, with an emphasis on monitoring causes of engine failure or reduced efficiency. A Ford 70PS 4-stroke diesel engine has been instrumented with a range of sensors and interfaced to a PC in order to facilitate computer controlled data acquisition and data storage. Data was analyzed to evaluate the optimum use of sensors to identify faults and to develop an intelligent algorithm for the engine condition monitoring and fault detection, and in particular faults affecting the combustion process in the engine. In order to investigate the fault-symptom relationships, two synthetic faults were introduced to the engine. Fuel and inlet air shortage were selected as the faults for their direct relationship to the combustion process quality. As a subtask the manually operated hydraulic brake was adapted to allow automatic control to improve its performance. Two modes of controlling were designed for the developed automatic control of the hydraulic brake system. A robust mathematical diesel engine model has been developed which can be used to predict the engine parameters related to the combustion process in the diesel engine, was constructed from the basic relationships of the diesel engine using the minimum number of empirical equations. The system equations of a single cylinder engine were initially developed, from which the multi-cylinder diesel engine model was validated against experimental test data. The model was then tuned to improve the predicted engine parameters for better matching with the available engine type. The final four-cylinder diesel engine model was verified and the results show an accurate match with the experimental results. Neural networks and fuzzification were used to develop and validate the intelligent condition monitoring and fault diagnosis algorithm, in order to satisfy the requirements of on-line operation, i. e. reliability, easily trained, minimum hardware and software requirements. The development process used a number of different neural network architecture and training techniques. To increase the number of the parameters used for the engine condition evaluation, the Multi-Net technique was used to satisfy accurate and fast decision making. Two neural networks are designed to operate in parallel to accommodate the different sampling rate of the key parameters without interference and with reduced data processing time. The two neural networks were trained and validated using part of the measured data set that represents the engine operating range. Another set of data, not utilized within the training stage, has been applied for validation. The results of validation process indicate the successful prediction of the faults using the key measured parameters, as well as a fast data processing algorithm. One of the main outcomes of this study is the development of a new technique to measure cylinder pressure and fuel pressure through the measurement of the strain in the injector body. The main advantage of this technique is that, it does not require any intrusive modification to the engine which might affect the engine actual performance. The developed sensor was tested and used to measure the cylinder and fuel pressure to verify the fuel fault effect on the combustion process quality. Due to high sampling rate required, the developed condition monitoring and fault diagnosis algorithm does not utilize this signal to reduce the required computational resources for practical applications.