Failure Prediction using Vibration Analysis

Background

The breakdown of the manufacturing equipment in factories or of ATMs, ticketing machines and other equipment essential to people’s lives makes it difficult to continue productions and services. In order to prevent this from happening, the idea of preventive maintenance is spreading in which the operation status of equipment is constantly monitored by sensors, and maintenance action is taken when something abnormal is detected. Therefore, to detect signs of equipment failure, OKI is developing an AI technology that automatically detects anomalies based on the change in vibration generated by the equipment in operation. With collaborators both inside and outside the company, OKI is conducting experimental demonstrations to develop methods for real-time and automatic detection of anomalies that were previously not noticed.

Features

OKI’s vibration analysis is based on a unique method that combines non-negative matrix factorization (NMF) and machine learning. This method has two features. The first is the capability for a light-weight processing and real-time detection even detecting abnormal vibrations with high frequency band that are generated at the beginning of a failure, but not usually handled in general detection methods. The second is that there is no need to identify and classify the frequency bands in advance. With these features, it is possible to automatically detect anomalies that would have been noticed only by experienced workers. In addition, maintenance can be performed before equipment fails. It is expected this technology will be applied to many devices.

vibration analysis