[Truncated abstract] The development of an automated classification system of wear particles or surfaces is of great interest to the machine condition monitoring industry. The system, once developed, may also find applications in medical diagnostics. Such a tool will be able to replace human experts in the detection of the onset of early machine failure, or in the diagnosis and prognosis of, for example, joint diseases. This will improve efficiency, reliability and also reduce costs of monitoring or diagnostic systems. Current literature available on this topic has included various studies on different classification methods. However, there has been no work conducted on the development of a totally integrated automated classification system. The first part of this thesis presents a study investigating the efficiency and robustness of various pattern recognition methods currently described in literature. A special computer program was developed to test each of the classification methods against both standard image databases and tribological surface images. There are three core components of a pattern recognition system that need to be analysed: (1) feature extraction, (2) feature reduction and (3) classifier. Each of these components provides a vital link that can affect the reliability of the complete classification system. ... The optimal classifier was the Linear Support Vector Classifier. This part of research is described in Paper 2. The second part of this thesis contains work verifying the performance of the automated classification system developed using both tribological and bio-tribological surface images. Experiments were carried out to generate wear particles created under different wear mechanisms (adhesive, abrasive and fatigue wear) and various operating conditions representing different degree of wear severity. The automated classification system developed was able to successfully classify wear particles with respect to both the type of wear mechanism operating and the wear severity. The results of this classification are described in Papers 3 and 5. The success of the automated classification system was also confirmed by its ability to classify different groups of worn (osteoarthritic) cartilage surfaces (Paper 4). This could lead to potential applications of the system for early detection of the onset of osteoarthritis. In conclusion, the automated classification system developed can accurately classify both tribological and bio-tribological surface images. This system could become a vitally important tool in both machine condition monitoring and medical diagnostics.
|Qualification||Doctor of Philosophy|
|Publication status||Unpublished - 2007|