Fault detection and prediction according to the location of vibration occurrence in motors

Abstract

Torsional vibration occurring in a motor can cause not only simple mechanical damage but also severe problems in the entire connected operating system. To prevent this and maintain operational efficiency, preemptive fault detection and prediction technology using motor vibration data is emerging as an important field in automation. However, existing studies primarily focus on fault detection techniques based on data collected in a specific RPM range and have limitations in sufficiently reflecting the environmental variability and data collection constraints of the actual industry. To solve these problems, in this paper, we propose a new classification model that can operate effectively even under various RPM conditions and limited data environments. The proposed model complements the limitations of existing approaches and improves the accuracy and reliability of fault detection by combining similarity-based analysis and machine-learning techniques based on vibration data collected in the time and frequency bands. In addition, various algorithms that are widely used in multi-class classification problems were applied to compare the performance of each technique and select an optimal fault detection model.