Multi-Sensor Fault Diagnosis for Misalignment and Unbalance Detection Using Machine Learning

Abstract

Rotating machines frequently undergo various faults causing increased maintenance and operation costs. To minimize these costs, effective and intelligent methods are thus required. Different sensor modalities reflecting various faults should continuously be monitored and interpreted to enable these methods. In this work, two sensor modalities, Infrared Thermography (IRT), and vibration are used complementary to form a multi-sensor fault diagnosis system. This system is used to diagnose the three most occurring faults: misalignment, unbalance, and rotor disk eccentricity, as single, dual, and multi-faults in a rotating mechanical system. The high feature processing capabilities of a Deep Convolutional Neural Network (DCNN) and the high predictive capabilities of a Support Vector Machine (SVM) are combined along with the potential dimensionality reduction using Principal Component Analysis (PCA). The results show that the proposed method is robust and signifies its reliability towards the effective diagnosis of considered faults. Further, IRT-based fault diagnosis outperforms the vibration-based classification in all working conditions.