Upscaling complex project-level asset management systems under multiple uncertainties to a network of assets
Abstract: Probabilistic and non-linear models have been used to accurately model various phenomena in asset management systems. Using Monte Carlo simulation (MCS) to model all incidents in the life cycle of assets and heuristic algorithms to find near-optimal maintenance interventions timings, project-level asset management systems (PL-AMS) aim to maintain the performance of different assets at an acceptable level of performance. Due to its high computational costs, upscaling this framework to all assets in a network has remained a challenge. This paper puts forward a new methodology to address this gap. The conventional form of PL-AMSs that use MCS and heuristic optimization algorithms is substituted with a trained machine learning algorithm in this methodology. As a showcase, an ensemble of random forests models was trained on optimal maintenance timings of more than 1.6 million semi-synthesized bridges. The trained ensemble model could yield optimized MRR plans with more than 96% accuracy within a far shorter time. Practitioners can adopt the proposed methodology to upscale PL-AMSs in the literature of asset management to a network of assets.