Deep reinforcement learning for predictive aircraft maintenance using probabilistic Remaining-Useful-Life prognostics

• DOI: 10.1016/j.ress.2022.108908
• Publication Date: 2022-10-22T06:53:24Z
• Authors: Juseong Lee; Mihaela Mitici
• Keywords: Prognostics; Turbofan; Aircraft Maintenance; Condition-Based Maintenance

The increasing availability of sensor monitoring data has stimulated the development Remaining-Useful-Life (RUL) prognostics and maintenance planning models. However, existing studies focus either on RUL only, or propose based simple assumptions about degradation trends. We a framework to integrate data-driven probabilistic into predictive planning. estimate distribution using Convolutional Neural Networks with Monte Carlo dropout. These are updated over time, as more measurements become available. further pose problem Deep Reinforcement Learning (DRL) where actions triggered estimates distribution. illustrate our for aircraft turbofan engines. Using DRL approach, total cost is reduced by 29.3% compared case when engines replaced at mean-estimated-RUL. In addition, 95.6% unscheduled prevented, wasted life limited only 12.81 cycles. Overall, we roadmap from prognostics,