| Modern industry is concerned about extending the lifetime of its critical processes and maintaining them only when required. Significant aspects of these trends include the ability to diagnose impending failures, detect product defects, prognosticate the remaining useful lifetime of the process and schedule maintenance operations so that uptime is maximized. In this thesis, innovative fault diagnostic and prognostic algorithms were developed based on intelligent signal/image processing techniques. Novel methodologies for feature extraction, fault diagnostics, and fault prognostics were integrated into the proposed architecture. Wavelet and fractal analyses were employed to extract fault signatures from process signals. An alternative diagnostic method based on a generalized wavelet neural network was designed in order to achieve better effectiveness and efficiency. A systematic prognostic approach based on a new dynamic wavelet neural network was introduced for the purpose of predicting incipient process failures. Moreover, static and dynamic virtual sensors were constructed using wavelet neural networks to indirectly measure or predict process parameters that were difficult to measure or predict on-line and on-site. Uncertainty management and performance assessment for fault prognostication were investigated using statistical, adaptive, and fizzy techniques. The proposed diagnostic/prognostic algorithms and static/dynamic virtual sensors were applied to real processes, such as a naval shipboard chiller. |
