A New Machine Learning Method for Ball Bearing Condition Monitoring Based on Vibration Analysis

In recent years, with the advent of the Fourth Industrial Revolution concepts and the development of artificial intelligence technologies, new approaches such as the digital twin have been introduced. In a digital twin, a virtual counterpart of the physical system during its whole life is created, with abilities such as analyzing, evaluating, optimizing, and predicting. The first step in creating a digital twin model is to construct a (multi) digital health indicator that describes different aspects of the physical component state during the whole life of the component. In this research, a new method for constructing health indicators based on vibration measurement and a deep learning model has been introduced. For this purpose, the Continuous Wavelet Transform was used to convert the raw vibration signals into two-dimension images; Then, the deep learning model was used to extract features from the images and the health indicator is constructed based on the differences of the images in normal and failure stages. In this article, various Autoencoder architectures are discussed, and it is demonstrated that the Convolutional Autoencoder has better performance in terms of detecting incipient faults. The performance of the proposed model is evaluated by the vibration data of the bearing, and the constructed health indicator exhibited a monotonically increasing degradation trend and had good performance in terms of detecting incipient faults.