An approach for anomaly detection of motor rotating shaft alignment using deep learning

Abstract

With increasing interest in eco-friendly ships, extensive research is being conducted on electric/hybrid vessels, highlighting the growing need for real-time monitoring of motors. Motor malfunctions can occur due to structural defects or component damage, but particularly due to shaft misalignment, which leads to excessive vibration, noise, and accelerated wear on bearings and couplings. These malfunctions can pose risks of personal injury or cause significant disruptions to industrial operations, making the development of a reliable condition monitoring system essential to prevent such issues. To address these issues, this research proposes a Convolutional Neural Network (CNN)-based anomaly detection technique that uses a microphone sensor to monitor the rotating shaft alignment. The proposed CNN classification model converts sound time-series data into spectrograms using the Short-Time Fourier Transform (STFT), capturing time-frequency patterns that specifically reflect motor anomalies related to shaft misalignment. The performance of the proposed CNN model is compared with machine learning models, demonstrating superior accuracy and reliability in detecting these anomalies. By monitoring sensor data, the proposed model serves as a diagnostic system to identify and respond to motor abnormalities, particularly rotating shaft misalignment, enhancing safety and reliability.