Intelligent Hardness Prediction of Bearing Rings Using Pulsed Eddy Current Testing

Abstract

Bearings are central components of rotating machinery. Original equipment manufacturers must utilize high-quality bearings for desirable performances and safety of the equipment. Hardness, which is directly linked to product performance and reliability, is, thus, an essential mechanical property of bearings. As a result, it is necessary to evaluate the hardness during the manufacturing process. However, because of the complicated transformation of microstructures during heat treatment, there is no an established noncontact, quick, and affordable method for measuring the hardness of bearings. This article describes a pulsed eddy current testing (PECT)-based novel method utilizing deep learning for reliable nondestructive testing of bearing rings. The continuous wavelet transform (CWT) was used to combine the time-frequency information in the PECT signals, and the deep learning models were trained to predict hardness. In addition, we compared and optimized the network’s performance in feature maps, the number of convolutional kernels, and the learning rate. The results indicate that the developed novel method has an average error of 1.37% and a maximum error of 1.89%, exhibiting good hardness prediction accuracy and stability.