Real-Time Integrated Learning and Decision Making for Cumulative Shock Degradation

Problem definition: Unexpected failures of equipment can have severe consequences and costs. Such unexpected be prevented by performing preventive replacement based on real-time degradation data. We study a component that degrades according to compound Poisson process fails when the exceeds failure threshold. An online sensor measures in real time, but interventions are only possible during planned downtime. Academic/practical relevance: characterize optimal policy integrates learning from sensor. demonstrate effectiveness practice with case interventional x-ray machines. The data set this is available companion. As such, it serve as benchmark for future studies stochastically deteriorating systems. Methodology: parameters vary one next cannot observed directly; population heterogeneous. These must therefore inferred observing signal. model situation partially observable Markov decision (POMDP) so making integrated. collapse information state space POMDP three dimensions policies analyzed computed tractably. Results: dependent control limit. limit increases age may decrease result other Numerical analyses reveal integration leads cost reductions 10.50% relative approaches do not learn signal 4.28% separate making. Managerial implications: Real-time reduce maintenance unplanned downtime considerable amount. tractably industrial systems our collapse. Finally, benefit amount initial calibration, whereas additional much less valuable ignore heterogeneity. Funding: This work was supported Fonds National de la Recherche Luxembourg [Grant AFR-12451704] Philips. C. Drent Data Science Flagship framework, cooperation between Eindhoven University Technology research M. Research Fund Luxembourg. J. Arts thanks Netherlands Foundation Scientific funding research. S. Kapodistria NWO Gravitation Project NETWORKS Dutch government. Supplemental Material: appendices at https://doi.org/10.1287/msom.2022.1149 .