Performance Evaluation of Support Vector Machine for System Level Multi-fault Diagnosis

Rotating elements are the essential part of various industries. Progressive degradation of rotating parts leads to system failure and economic losses. Several studies have been carried out to diagnose incipient faults in rotating components using the knowledge-based self-diagnosis Machine Learning (ML) models. But in real scenarios expecting the occurrence of one fault at a time is very unlikely. Multiple components and subcomponent faults take place simultaneously in a system. In most industries, machine parts are replaced directly to avoid downtime. Hence detection of multi-faults at a system level is very much important. In this paper, two major rotating components (motor and bearing) were considered, and all possible multi-fault conditions were simulated under different speed and load conditions. The raw vibration signals were acquired from three different locations and used directly for the training of the Support Vector Machine (SVM) model. The highest classification accuracy of 100% was achieved for the multi-fault diagnosis. Performance evaluation of the SVM model was done using eleven different performance matrixes. The model showed a greater potential to identify different multi-faults using the raw signal without using any further data processing or feature engineering techniques.

Automated summary

This study considers two major rotating components, motor and bearing, and simulates all possible multi-fault conditions under different speed and load conditions. The Support Vector Machine (SVM) model is trained directly using raw vibration signals, achieving a highest classification accuracy of 100% for multi-fault diagnosis. The model shows great potential for detecting multi-faults at a system level.