A genetic training algorithm of wavelet neural networks for fault prognostics in condition based maintenance

The main idea of condition based maintenance (CBM) is to monitor the health of critical machine components and system almost continuously during operation and maintenance actions based on the assessed condition. If done correctly, CBM has the benefits such as reducing catastrophic failures, minimizing maintenance and logistical cost, maximizing system security and availability and improving platform reliability. A CB! system usually has four major functional modules, namely feature extraction, diagnostics, prognostics and decision support. Among them, fault prognostics is the most important enabling technology. It is the most challenging research area which is so called crystal ball of CBM. But it has the potential to be the most beneficial. This paper presents a fault prognostic algorithm based on a generic wavelet neural networks (WNN) architecture. Its training process based on genetic algorithm is described in detail. Finally, the fault prognostic algorithm has been verified using a simulation experiment, and the results are very satisfactory.