For changing electrical energy to rotational mechanical energy, the induction motors are best candidate in various industries. The reasons for that chose are four important characteristics as high efficiency, more reliability, easy installation and low cost over other kind of electrical motors. However, any kind of faults lead to failure in this device can cause irreparable damages in production lines. For example, unexpected shutting down in a plastic industries can waste the whole investment. Diagnosing bearing failures in early stage is critical to reduce maintenance costs and operational failures. Bearing failures are a major cause of machine vibrations. Unfortunately, existing methods are optimized for controlled environments. In them realistic conditions haven’t been considered such as variable load, time-varying rotational speeds and non-stationary nature of vibration. This study presents an integration of time analysis and deep learning techniques to diagnose bearing failures under time-varying speeds and varying noise levels. In this study, we present an approach to diagnosing bearing failures employing vibration signals and con- volutional neural networks (CNN) with Pre-processing of the vibration signal by using discrete wavelet transform (DWT) to remove the effect of Variable Frequency Drive (VFD) which causes odd harmonics. The experimental results demonstrate that the proposed method surpasses conventional methods in both accuracy and reducing computational costs for detecting bearing failures.

Keywords