H. Li
info
Please Note
<p>This page displays the records of the person named above and is not linked to a unique person identifier. This record may need to be merged to a profile.</p>
3 records found
1
Industrial aircraft maintenance policies still rely on rigid, static slot structures that limit operational efficiency. Existing research has mainly addressed task (re-)allocation within a pre-defined number and duration of slots, while direct slot planning (jointly determining slot duration and allocation) remains difficult due to its combinatorial complexity. We propose a graph-based, reinforcement-learning framework that sequentially constructs maintenance task packages within a fixed planning horizon. A Graph Neural Network (GNN) encodes relations among tasks to inform a deep reinforcement learning (DRL) policy, which learns to group tasks into slots efficiently. This framework is trained and evaluated on operational data from a major European airline and benchmarked against a fixed-time Mixed-Integer Linear Programming (MILP) benchmark and the current industry practice. The results demonstrate that the model rapidly generates feasible schedules, and quantify trade-offs among interval spillage, panel reuse, and drop-out tasks relative to these benchmarks. Our model is able to decrease task spillage and increase panel reuse consistently. Finally, as the reward weights can be tuned to reflect an airline’s operational preferences, this framework provides a flexible decision-support tool for dynamic aircraft maintenance task packaging.
...
Industrial aircraft maintenance policies still rely on rigid, static slot structures that limit operational efficiency. Existing research has mainly addressed task (re-)allocation within a pre-defined number and duration of slots, while direct slot planning (jointly determining slot duration and allocation) remains difficult due to its combinatorial complexity. We propose a graph-based, reinforcement-learning framework that sequentially constructs maintenance task packages within a fixed planning horizon. A Graph Neural Network (GNN) encodes relations among tasks to inform a deep reinforcement learning (DRL) policy, which learns to group tasks into slots efficiently. This framework is trained and evaluated on operational data from a major European airline and benchmarked against a fixed-time Mixed-Integer Linear Programming (MILP) benchmark and the current industry practice. The results demonstrate that the model rapidly generates feasible schedules, and quantify trade-offs among interval spillage, panel reuse, and drop-out tasks relative to these benchmarks. Our model is able to decrease task spillage and increase panel reuse consistently. Finally, as the reward weights can be tuned to reflect an airline’s operational preferences, this framework provides a flexible decision-support tool for dynamic aircraft maintenance task packaging.
Aircraft maintenance scheduling is a focus point for airlines. Maintenance is essential to ensure the airworthiness of aircraft, but it comes at the cost of rendering them unavailable for operations. In current operations, aircraft maintenance scheduling must often be updated to include time for non-routine (i.e., non-schedule) tasks. Non-routine maintenance tasks (NRTs) introduce significant uncertainty into aircraft maintenance scheduling, often leading to delays and increased operational costs. Despite their importance, limited research has been conducted on predicting NRT maintenance workload using data-driven methods. This study compares Random Forest and LGBM with a Weibull survival model for NRT workload prediction at the individual routine task (RT) level. Results indicate that model performance depends on data availability and the evaluation metric: using data from all tasks can help in sparse settings, while using an independent occurrence model to correct labor predictions may improve aggregate error at the cost of other outcomes. The historical-average baseline performed comparably to Independent LGBM on MAE, ROC-AUC, and average precision. Finally, results show that summing task-level predictions does not result into a good forecast for the work packages that contains all the tasks. Further data and prospective validation are needed before these estimates can guide work-package planning.
...
Aircraft maintenance scheduling is a focus point for airlines. Maintenance is essential to ensure the airworthiness of aircraft, but it comes at the cost of rendering them unavailable for operations. In current operations, aircraft maintenance scheduling must often be updated to include time for non-routine (i.e., non-schedule) tasks. Non-routine maintenance tasks (NRTs) introduce significant uncertainty into aircraft maintenance scheduling, often leading to delays and increased operational costs. Despite their importance, limited research has been conducted on predicting NRT maintenance workload using data-driven methods. This study compares Random Forest and LGBM with a Weibull survival model for NRT workload prediction at the individual routine task (RT) level. Results indicate that model performance depends on data availability and the evaluation metric: using data from all tasks can help in sparse settings, while using an independent occurrence model to correct labor predictions may improve aggregate error at the cost of other outcomes. The historical-average baseline performed comparably to Independent LGBM on MAE, ROC-AUC, and average precision. Finally, results show that summing task-level predictions does not result into a good forecast for the work packages that contains all the tasks. Further data and prospective validation are needed before these estimates can guide work-package planning.
Assigning aircraft to gates is one of the most important daily decision problems that airport professionals face. The solution to this problem has raised a significant effort, with many researchers tackling many different variants of this problem. However, most existing studies on gate assignment contain only a static perspective without considering possible future disruptions and uncertainties. We bridge this gap by looking at gate assignments as a dynamic decision-making process. This paper presents the Real-time Gate Assignment Problem Solution (REGAPS) algorithm, an innovative method adept at resolving pre-assignment issues and dynamically optimizing gate assignments in real-time at airports through the integration of Deep Reinforcement Learning (DRL). This work represents the first time that DRL is used with real airport data and a configuration containing a large number of flights and gates. The methodology combines a tailored Markov Decision Process (MDP) formulation with the Asynchronous Advantage Actor–Critic (A3C) architecture. Multiple factors, such as flight schedules, gate availability, and passenger walking time, are considered. An empirical case study demonstrates that the REGAPS outperforms two classic deep Q-learning algorithms and a traditional Genetic Algorithm in terms of reducing passenger walking time and apron gate assignment. Finally, supplementary experiments highlight REGAPS’s adaptability under various gate assignment rules for international and domestic flights. The finding demonstrates that not only did REGAPS outperform COVID restrictions, but it can also produce considerable benefits under other policies.
...
Assigning aircraft to gates is one of the most important daily decision problems that airport professionals face. The solution to this problem has raised a significant effort, with many researchers tackling many different variants of this problem. However, most existing studies on gate assignment contain only a static perspective without considering possible future disruptions and uncertainties. We bridge this gap by looking at gate assignments as a dynamic decision-making process. This paper presents the Real-time Gate Assignment Problem Solution (REGAPS) algorithm, an innovative method adept at resolving pre-assignment issues and dynamically optimizing gate assignments in real-time at airports through the integration of Deep Reinforcement Learning (DRL). This work represents the first time that DRL is used with real airport data and a configuration containing a large number of flights and gates. The methodology combines a tailored Markov Decision Process (MDP) formulation with the Asynchronous Advantage Actor–Critic (A3C) architecture. Multiple factors, such as flight schedules, gate availability, and passenger walking time, are considered. An empirical case study demonstrates that the REGAPS outperforms two classic deep Q-learning algorithms and a traditional Genetic Algorithm in terms of reducing passenger walking time and apron gate assignment. Finally, supplementary experiments highlight REGAPS’s adaptability under various gate assignment rules for international and domestic flights. The finding demonstrates that not only did REGAPS outperform COVID restrictions, but it can also produce considerable benefits under other policies.