Vibration-Based Fault Detection in Belt Drive Systems Using FFT and LSTM Models

Abstract

Fault detection in rotating machinery is essential for preventing costly maintenance, minimizing downtime, and ensuring operational safety. In this study, a belt drive system was analyzed for fault detection using vibration analysis with Fast Fourier Transform (FFT) and machine learning (ML). A custom test rig, consisting of a drive and driven pulley, was designed, and various fault conditions, including a loosened belt, misalignment, and mass imbalance, were intentionally introduced, along with a healthy baseline scenario. Vibration data were collected using a high-sampling- rate EnDAQ sensor, a high-quality data acquisition device. The signals were analyzed in both the time and frequency domains, revealing significant changes in vibration patterns and FFT spectra across different conditions. FFT analysis provided insights into the characteristics of each fault type. Additionally, a Long Short-Term Memory (LSTM) neural network was developed to classify the system state. The model achieved high accuracy (98%) in detecting and distinguishing between fault types. Overall, the results demonstrate the strong potential of FFT and ML techniques for effective fault detection in belt drive systems.