Event Based Robot Prognostics Using Principal Component Analysis

Abstract

As industrial systems are getting complicated, challenges in coming up with efficient maintenance strategies which include predicting failures in the system become important industry specific research topic. Traditionally, research focuses on developing failure prediction models based on physical understanding of the system. But, development of such models are often time consuming and labour intensive for complex systems. In recent past, due to advent of cheaper data collection mechanisms and efficient algorithms, data driven approaches for predicting failures are gaining significant interest in industrial research community. In this paper, we provide a Principal component Analysis (PCA) based approach of failure prediction in industrial robots using event log information. The event logs are collected through remote service set-up from a robot controller. The proposed method will reduce the dimensionality of the original data which consist of interrelated events while retaining the variation present in the data. Using PCA and multivariate statistics such as Hotelling T2, Q Residuals and Q contributions charts, we are able to detect abnormal behavior of event pattern within 30 days before failure.