NTN : Develops Precise Remaining Useful Life Prediction Technology for Rolling Bearings

Contributing to the improvement of productivity through appropriate maintenance of machinery and facilities by utilizing AI

NTN Corporation (hereafter, NTN) has developed a technology that precisely predicts remaining useful life by combining several AI methods. By predicting the remaining useful life with high accuracy after the flaking occurs, which causes bearing failure, to the limit of use, efficient maintenance plans for machinery and equipment can be drafted, which contributes to improvement of productivity and cost reduction.

Bearings used in machinery and equipment may cause minor flaking depending on the use conditions and it may lead to failure at the worst case. However, when it is difficult to implement maintenance of the bearing due to the equipment structure and installed location, there are some cases in which the bearings continue to be used as long as it does not affect operation. The condition of a bearing can be determined by detecting abnormalities using vibration data. However, there is no way to accurately determine how long the bearing can be used after an abnormality such as flaking occurs (remaining useful life), and it is common to replace the bearing as soon as possible or after the bearing is damaged. In addition, there are many cases in which field workers judge the timing of replacement based on years of experience, etc., and as person-hours saving and automatic production systems are progressing, there is a growing need for highly precise remaining useful life prediction technology that allows for more accurate timing of bearing replacement, in order to reduce downtime in machinery, equipment, etc. and reduce maintenance costs.

The remaining useful life forecasting technology developed by NTN has been improved by combining deep learning with Bayesian learning to improve the accuracy of estimating the remaining useful life from the occurrence of flaking of the bearing to the time it is damaged.

Among several AI methods, NTN selects deep learning, which is specialized in image processing called convolutional neural network. It can convert the vibrational data of the bearing into image data for use, enabling prediction of the damaged condition of the bearing and the remaining useful life. In addition, we established a highly reliable predictive model by combining a hierarchical Bayesian linear regression, which evaluates the reliability of predictive values by considering individual differences and variations (errors) in measurement data in the extent of damage progression of bearings. By considering the damage condition as well, the prediction accuracy of the remaining useful life is improved by approximately 30 % compared with the conventional technology.

This technology is the result of a joint research project at NTN Next Generation Research Alliance Laboratory*, established in 2017 at the Graduate School of Engineering at the National University Corporation Osaka University (headquartered in Suita City, Osaka Prefecture). It was realized by combining the technology and knowledge that NTN has cultivated over more than 100 years with the knowledge of cutting-edge AI research of the university, including Ken-ichi Fukui, associate professor of SANKEN (the Institute of Scientific and Industrial Research, Osaka University.)

NTN is pursuing initiatives in the field of services and solutions that combine sensor technology and IoT and offers a variety of products and services to help improve the maintainability of bearings, including "NTN Portable Vibroscope" that detects bearing abnormalities and a bearing diagnostic application that constantly monitors the condition of bearings.

In the future, NTN will continue to verify the feasibility of this technology. By using this technology for maintenance-related servicing, we will contribute to improving productivity and reducing environmental impact through the proper maintenance of machinery and equipment and the optimal use of bearings.