A performance-centred approach to optimising maintenance of complex systems

Highlights

• Motivated by systems where performance and condition may deteriorate at different rates.

• Introduces a general theoretical framework for Performance Centred Maintenance (PCM).

• Explains PCM modelling choices for maintenance optimisation of a real world industry system.

• Develops exact and heuristic solutions for this maintenance optimisation model.

• Evaluates reinforcement learning algorithm for an industry scale optimisation.

Abstract

This paper introduces performance-centred maintenance (PCM) as a novel approach to maintain systems when dual consideration is given to operational performance and degradation condition. We consider situations where performance and condition do not necessarily deteriorate at the same rate typified by, say, an ageing system still achieving good performance or a new system performing poorly. In this problem context, competing interests may arise between different decision-makers, such as operators and maintainers, since alternative strategies may benefit either performance or condition at the expense of the other. To address this challenge we introduce a theoretical framework for the PCM approach and discuss key characteristics of the modelling problem. The general PCM approach is motivated by a real-world industrial system for which maintenance decisions required to be optimised. A specific application is shown for the industry problem which we model by a Markov decision process capable of interrogating decisions over multiple time-scales. We obtain an exact solution using dynamic programming. We also explore a less computationally challenging heuristic using a reinforcement learning algorithm and evaluate its accuracy for the large-scale industry model. We show that optimal maintenance policies from a PCM model can provide decision support to both maintainers and operators taking account of both perspectives of the problem.