A Combination of Fourier Transform and Machine Learning for Fault Detection and Diagnosis of Induction Motors

Abstract

Induction motors are widely used in different indus-try areas and can experience various kinds of faults in stators and rotors. In general, fault detection and diagnosis techniques for induction motors can be supervised by measuring quantities such as noise, vibration, and temperature. The installation of mechanical sensors in order to assess the health conditions of a machine is typically only done for expensive or load-critical machines, where the high cost of a continuous monitoring system can be justified. Nevertheless, induced current monitoring can be implemented inexpensively on machines with arbitrary sizes by using current transformers. In this regard, effective and low-cost fault detection techniques can be implemented, hence reducing the maintenance and downtime costs of motors. In this work, machine learning techniques have been combined with traditional Fast Fourier Transform (FFT) to proposes a simple but yet efficient method for fault detection and diagnosis of induction motors. Raw signals are converted from time domain to frequency domain using FFT. The FFT spectrum are then divided into frequency segments and energy coefficients of each segment are calculated as the sum over the FFT amplitudes. These energy coefficients are used as features for machine learning in our platform. The proposed method is validated on real-world data and achieves a precision of 99.7% for fault detection and 100% for fault classification with minimal expert knowledge requirement. In addition, this approach allows users to be able to optimize/balance risks and maintenance costs to achieve the highest benefit based on their requirements. These are the key requirements of a robust prognostics and health management system.