A New ICA-SOM Based Method for Compound Faults Diagnosis

Compound faults diagnosis is an important but difficult task in diagnostics. When several faults arise at the same time, vibration measurements by sensors show themselves as a complex integrated symptom, not as a simple and linear combination of several single faults, which makes it difficult to detect compound faults correctly. We proposed a new ICA-SOM based method for compound faults diagnosis by combining independent component analysis (ICA) based feature extraction with the self-organizing map (SOM) based classifier. ICA is a powerful tool for redundancy reduction, with which the multichannel measurements were fused. By the use of a novel-processing unit for further feature extraction, features about underlying sources embedded into the measurements were captured subsequently. At last, a medium-scale SOM network was constructed for symptom decomposition and topological display of these features. Experiment results of single and compound faults diagnosis show that our new method is effective and of great potential in health condition monitoring of machines