Minimizing Flight Hours Losses by Utilizing Deep-Q Learning Techniques in Scheduling Aircraft Maintenance

Abstract

Currently, there are insufficiently good or efficient tools or procedures for long-term scheduling of aviation maintenance activities in the world, and Vietnam specifically. Airlines are impacted by a variety of elements, including the quantity of aircraft they operate, their capacity, their personnel resources, their maintenance resources, and unforeseen and urgent occurrences that cause schedule disruptions. There are no fast, automatic solutions to the above-described problems existing in aviation today. The reinforcement learning method as described here could potentially be one answer to these problems. The idea is to plan in years running across a specified period such that the aircraft is brought as close as possible to the inspection deadline. From there, the airworthiness of the aircraft increases while the maintenance inspection decreases, reducing the cost of maintenance. Application optimization of the scheduling plan is done using the Deep Q-learning method. The results achieved by the Q-learning and Deep Q-learning algorithms are better in terms of computation times as compared to the other current techniques. The research results of the checks showed reinforcement learning potential in dealing with this problem, where the fly hours loss of planned inspections was reduced by using data from Vietnam Airlines. Computational experiments show that our methods adapt for different purposes and settings of reality. After teaching the model with these simulated conditions, they show how well a reinforcement learning application quickly arrives at lean repair plans.