Predictive maintenance under the conditions of limited data availability
With the scale and need for industrial operations increasing, industrial systems are becoming far more complex than they had ever been. As a consequence, traditional maintenance strategies are not only failing to prevent downtime, they introduce additional costs due to over-maintenance. Predictive maintenance (PdM) aims to address this problem by delivering on the promise of an optimal maintenance policy that allows maintenance to be delayed until the risk of failure is not critical and the performance of the equipment remains unaffected. In order to effectively implement PdM, one must first perform prognostics with minimal error and uncertainty. Unfortunately, due to the limited data availability for prognostics in industrial scenarios, effective implementation of PdM remains a vision rather than a reality for industrial organisations.
Improve the ability to effectively plan and schedule PdM tasks under the conditions of limited data availability for prognostics, whilst maximising the value of the equipment. More specifically, the research focuses on addressing the following research questions: (i) how to predict equipment failure under the conditions of limited data availability; (ii) how can such improved prediction of equipment failure be exploited to develop an optimal maintenance policy.
A methodology to effectively implement PdM in industrial scenarios that suffer from the problem of limited data availability for data-driven prognostics. This methodology will consist of a technique to generate real-valued (i.e. new and realistic) failure data using generative modelling and industrial information engineering, and a maintenance policy optimisation model.
Engineering and Physical Sciences Research Council (EPSRC). Grant number: EP/M508007/1