PROBABILISTIC FAULT DIAGNOSIS AND PROGNOSIS IN PRECOGNITIVE MAINTENANCE FOR HIGH-PERFORMANCE MANUFACTURING INDUSTRIES

Abstract

In an era of intensive competition, where manufacturing efficiency must be maximized, unexpected downtime and breakdown failures are more expensive than before. Precognitive maintenance is actively pursued in high-performance manufacturing industries nowadays, which uses mixed time-/condition-based information for probabilistic assessment of machinery health condition under aperiodic inspection. This dissertation consists of two parts. In the first part, Gaussian mixture model under semi-supervised learning is proposed to diagnose new types of faults with balanced data. Furthermore, deep learning techniques are developed for fault diagnosis with imbalanced data. In the second part for probabilistic prognosis, the degradation PDF and remaining useful life are predicted via enhanced particle filter under periodic inspection and Gamma process under aperiodic inspection. Finally, a novel non-fixed periodic inspection strategy is formulated and optimized to reduce overall maintenance cost and operational hazard. The effectiveness of my proposed techniques is verified on various realistic engineering systems.