Navigating the Prognostics Landscape: Deep Reinforcement Learning-Enabled Remaining Useful Life Estimation with Novel Methodology
Ramgopal Kashyap, Zainab Failh Al, Zamen Latef Naser, Maki Mahdi Abdulhasan
Abstract
This study shows how deep reinforcement learning in a mixed training system might determine how long something can be used. It uses both supervised learning and reinforcement learning to make RUL predictions more accurate and reliable. This makes proactive and better repair methods possible, which improves the reliability and usefulness of assets in many industry settings. The suggested approach does better than current ways in many different success rating measures, as shown by many tests and studies. In a different study, Ablation looks at how neural network structures, hyperparameters, mixed training, and transfer learning all work together. The results help us plan more studies and figure out what the main factors are that make the suggested strategy work. This paper provides a new and complete way to estimate RUL in prognostics. This helps with managing manufacturing assets and planning for upkeep.