A Life Prediction-Driven Maintenance Decision Method for Aircraft Engines
TL;DR
Experimental results based on the C-MAPSS dataset show that this method achieves reasonable performance in common prediction indicators and demonstrates certain advantages over traditional maintenance strategies in the comparison experiments of maintenance decisions.
Abstract
As the operating conditions of aircraft engines become more complex, the traditional maintenance strategies based on experience or fixed cycles are difficult to balance economic efficiency and reliability. This paper tentatively proposes a maintenance decision-making method driven by life prediction for aircraft engines. In the life prediction stage, considering the complex spatio-temporal coupling relationships contained in the degradation signals of multiple sensors, a spatio-temporal feature fusion transformation network is designed to jointly model the spatial dependence and temporal dynamics. At the same time, a probabilistic deep learning model is introduced to model the mean and uncertainty of the remaining life, providing risk warnings beyond point estimation. In the maintenance decision-making stage, the predicted distribution parameters are introduced into the Markov decision process state space. Based on the deep Q-network, a reward function is designed, and cost constraints are considered, thereby achieving a balance between safety and economy. Experimental results based on the C-MAPSS dataset show that this method achieves reasonable performance in common prediction indicators and demonstrates certain advantages over traditional maintenance strategies in the comparison experiments of maintenance decisions. The effectiveness of the method has been verified.