Challenges in the Practical Application of Data-Driven Fault Detection and Diagnosis

Abstract

Machine health and condition monitoring have become a billion-dollar industry, an area where fault detection and diagnosis is no longer just a subject of academic research, but are now increasingly embedded into commercial tools and products. This thesis addresses several practical challenges in the implementation of machine learning data-driven fault detection and diagnosis systems, from hardware design to testing methodology. This research introduces novel methods in the areas of vibration based ball bearing damage detection and optimal classification accuracy estimation. It also reveals how individual ball bearing parts contain their own unique signatures and recommendations on proper testing procedures to mitigate the impact of this effect. Lastly, it covers how advances in micro-electromechanical technology may be leveraged in order to reduce the cost of hardware while maintaining high sampling rates.