A hybrid physics–statistics–machine learning framework for reliability optimization of repairable Load Haul Dumper equipment

In underground mining activities, the availability and reliability of mining machineries are crucial for safe underground coal mining. The availability depends upon the frequency of failure and mean time to repair the machine. In the current paper, Load Haul Dumper have been considered for the study of its failure, mean time to repair followed by the reliability optimization using physics, statistics and machine learning approach. The failure of Load Haul Dumper is affected by the degradation its various subsystems, schedule of maintenance, quality of maintenance and intensity of using the machine. These parameters cannot be used for mapping the mean time between failure (MTBF) and mean time to repair (MTTR) using static reliability metrics. In this research, a novel hybrid approach composed of physics-based, statistical and machine learning concepts is proposed for optimizing the reliability of LHD machine fleets. The Load Haul Dumpers have been considered as multi-state repairable systems with time-varying failure properties. The operational and maintenance data of Load Haul Dumpers have been analyzed over a period of 77 months using non-homogeneous Poisson process models and semi-Markov models to describe the failure and repair processes. The criticality of subsystems was assessed by means of reliability importance measures. Further, the probabilistic failure risk was estimated by applying survival analysis techniques, including Kaplan-Meier estimation and Cox proportional hazard analysis. Machine learning models with explainable AI capabilities were developed for the prediction of major reliability parameters.