Machinery time to failure prediction - Case study and lesson learned for a spindle bearing application

Abstract

One of the important roles of prognostics health management (PHM) is to predict the time to failure of a system in order to avoid unexpected downtime and optimize maintenance activities. Although many attempts to predict time to failure have been reported in the literature, there are still challenges related to data availability and methodology. In addition, there is significant variation from case to case due to complexity of system usage and failure modes. This paper reveals various aspects related to such challenges experienced while applying a novel predictive technology to a spindle test-bed. The goal was to evaluate the ability of the technology to predict the remaining useful life of a bearing with seeded faults. Testing has been conducted to reveal the effectiveness of signal processing, health modeling and prediction techniques. While conducting the evaluation tests, besides some well-known bearing failure modes, an unusual case was recorded. This atypical bearing failure mode created a new challenge for the predictive technology being investigated, which prompted the development of an advanced feature discovering methodology using genetic programming. This new methodology and the technology evaluation results obtained for both the well-known and the atypical failure modes will be discussed in the paper. In addition, the paper will describe the test-bed and instrumentation approach, the data acquisition system and the experimental design for testing and validation of the technology.