Fault isolation using Self-Organizing Map (SOM) ANNs

This paper presents a Self-Organizing Map (SOM) Artificial Neural Networks method for fault isolation based on the condition of manufacturing components, equipments and processes. The signals reflecting the conditions of equipment is collected from a set of sensors and processed by signal processing methods, such as filter and de-noising. The features that are extracted in time domain, wavelet domain and wavelet domain are used to train SOM ANNs. After training, the faults are able to be isolated according to the features extracted from the real time information. This approach is very helpful to the maintenance decision- making. A case study shows that SOM ANN can isolate faults correctively and clearly.