⚡ electrical engineering

Integrated Optimization of Aircraft Line Maintenance: Scheduling under Uncertainty and Predictive Part Failure Modelling

This study proposes an integrated framework that combines a stochastic mixed-integer linear programming model for optimal line maintenance scheduling under uncertainty with a probabilistic failure prediction model, thereby enhancing cost efficiency, resource utilization, and operational resilience through proactive maintenance planning.

Original authors: Dionysios Peteinatos, Konstantia Kontodimou, Vassilios Andronis, Konstantinos Kepaptsoglou

Published 2026-09-21

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Dionysios Peteinatos, Konstantia Kontodimou, Vassilios Andronis, Konstantinos Kepaptsoglou

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The sky above is a complex web of metal and motion, where the safety of every flight depends on a hidden rhythm of care. Before an aircraft takes off, it must pass through a rigorous cycle of checks, a process where mechanics inspect the plane's many components to ensure nothing is worn, loose, or failing. This is not merely about fixing broken parts; it is about predicting when a part might break before it ever does. For decades, the aviation industry has relied on established patterns of failure, knowing that some parts wear out slowly over time while others might fail suddenly or even shortly after being installed. As the number of flights continues to rise and the average age of commercial fleets increases, the pressure on maintenance teams has grown immense. They face a difficult balancing act: they must keep planes flying on tight schedules while ensuring every safety check is completed, often with limited time and resources. When a task cannot be finished in the short window available between flights, it is deferred, or postponed, which can lead to a backlog of work that grows heavier and more dangerous over time.

Researchers at the National Technical University of Athens have developed a new way to navigate this challenge, creating a system that acts like a highly organized planner for aircraft maintenance. Instead of treating every day as a fresh start, their approach uses a computer model that learns from the past and looks ahead to the future. The core of their work is a scheduling tool that decides which maintenance tasks get done first, which can wait, and exactly when they should be performed. This tool is designed to work within the strict reality of a typical overnight maintenance window, which usually lasts only six hours. During this brief period, a team of mechanics must complete a list of critical safety checks and as many non-critical inspections as possible. The researchers built a mathematical engine that takes a list of tasks, knows which ones are absolutely essential for safety, and figures out the most efficient way to assign them to the available workstations and mechanics. If the team cannot finish a non-essential task in the six hours, the system does not simply ignore it; instead, it records the delay and automatically schedules that task for the next available opportunity, ensuring it is not forgotten.

To make this system even smarter, the team added a second layer of intelligence that predicts how likely a specific part is to fail. They analyzed the descriptions of hundreds of aircraft wing components and grouped them based on how they tend to break. Some parts fail early in their life, some fail randomly, and others fail only after they have been used for a long time. By simulating the life of these parts, the system can estimate when a component is most likely to stop working. If a part is predicted to fail within a short timeframe, the system flags the associated maintenance task as critical, ensuring it gets priority over other jobs. This predictive step allows the scheduler to make better decisions, focusing on the parts that truly need attention right now rather than treating all tasks as if they were equally urgent. The result is a schedule that is not just a list of jobs, but a dynamic plan that adapts to the specific risks of the aircraft being serviced.

When the researchers tested their system with real-world data from a typical maintenance facility, the results showed a clear improvement in how work is organized. The model successfully created weekly schedules that minimized the total cost of maintenance while keeping the number of deferred tasks to a very low level. In their tests, the system managed to complete almost all critical tasks on time, with the few non-critical tasks that had to be postponed being rescheduled efficiently for the following days. The cost of running these schedules was extremely close to the theoretical best possible outcome, with the difference being less than one percent in most cases. This level of precision suggests that the system can handle the complexity of real-world operations without needing supercomputers, running effectively on standard office equipment. The study also explored how the system would react if the costs of being early or late with a task changed, finding that the model remained robust and adaptable, always finding a way to keep the most important work on track.

The true power of this approach lies in its ability to turn a reactive process into a proactive one. In the past, maintenance planners often found themselves putting out fires, dealing with unexpected delays and urgent repairs that disrupted the entire schedule. This new framework allows them to see the big picture, understanding which tasks can wait and which cannot, and planning accordingly. By integrating the prediction of part failures directly into the scheduling process, the system helps prevent small issues from becoming major problems. It ensures that the limited time available each night is used in the most effective way possible, keeping aircraft safe and ready for their next journey. While the study focused on the specific challenges of line maintenance for a common type of commercial aircraft, the logic behind the system is broad enough to be applied to other industries where keeping complex machinery running safely is essential. The work demonstrates that with the right tools, the intricate puzzle of aircraft maintenance can be solved with greater clarity, efficiency, and safety.

Technical Summary: Integrated Optimization of Aircraft Line Maintenance

Problem Statement

The aviation industry faces increasing pressure to maintain aging fleets amidst rising passenger demand, leading to a sharp increase in maintenance task deferrals and a decline in technical dispatch reliability. Existing literature has largely focused on aircraft routing, fleet assignment, and staffing, leaving a significant gap in the optimization of Line Maintenance Scheduling. Current planning often operates reactively, addressing immediate issues rather than following a structured long-term strategy. This reactive approach accelerates system degradation and increases future restoration costs. Furthermore, the uncertainty surrounding part failures and the lack of integrated predictive models in scheduling frameworks hinder the ability to perform proactive maintenance, resulting in inefficient resource allocation and compromised safety margins.

Methodology

The study proposes an integrated framework comprising two complementary models: a primary Mixed-Integer Linear Programming (MILP) model for scheduling and an auxiliary Probabilistic Part Failure Prediction model.

1. Optimization Model (LMSP)

The primary model addresses the Line Maintenance Scheduling Problem (LMSP) within a typical six-hour overnight maintenance window. It utilizes a stochastic MILP approach to generate optimal weekly schedules.

Objective: Minimize total maintenance costs, which include earliness and tardiness penalties for critical tasks, deferral costs for non-critical tasks, resource utilization costs, and penalties for workstation assignment.
Key Features:
- Task Classification: Tasks are categorized as Critical ( $J_c$ ) or Non-Critical ( $J_n$ ). Critical tasks must be completed within the window, while non-critical tasks can be deferred if capacity is exceeded.
- Deferral Mechanism: The model explicitly calculates deferral times for non-critical tasks and incorporates them into subsequent days' schedules, creating a "rolling horizon" effect that prevents backlog accumulation.
- Constraints: The model enforces strict constraints regarding workstation capacity, resource availability (tools, personnel), task duration, and the prohibition of partial task execution (tasks must start and finish within the same 6-hour window).
- Implementation: The model was formulated using Python (PuLP interface) and solved using the Xpress Optimizer. It runs efficiently on standard hardware (8 GB RAM, Intel Core i5).

2. Part Failure Prediction Model (Auxiliary)

To support the scheduler with predictive data, a rule-based classification and simulation framework was developed to estimate the time to failure for aircraft wing components.

Classification: Component descriptions are analyzed using keyword pattern recognition to assign them to one of six Reliability-Centered Maintenance (RCM) failure patterns (e.g., infant mortality, wear-out).
Simulation: Time-to-failure is simulated using Weibull distributions (for five patterns) and an Exponential distribution (for constant hazard rates) via Monte Carlo simulation.
Criticality Assessment: A binary criticality indicator is generated based on a 500-flight-hour threshold (aligned with A-check intervals). A Logistic Regression model predicts the probability of a part being critical based on its failure pattern.
Integration: The output classifies tasks into critical ( $J_c$ ) and non-critical ( $J_n$ ) sets for the MILP model, enabling preemptive scheduling of high-risk components.

Key Results

The framework was validated using data from the Hellenic Civil Aviation Authority (HCAA) and EASA's Master Minimum Equipment List (MMEL) for an Airbus A320 wing section.

Scheduling Performance: The model successfully generated optimal weekly schedules for a representative facility.
- Optimality Gap: The solver achieved optimal solutions with a negligible gap from the theoretical lower bound (maximum deviation of 0.81% over the week).
- Deferral Management: The system effectively managed task deferrals. For instance, on Day 4, a 10.71% deferral rate was managed without disrupting the overall schedule, with deferred tasks successfully rescheduled in subsequent days.
- Cost Efficiency: Daily costs fluctuated based on accumulated deferrals, but the model consistently minimized costs while ensuring all critical tasks were completed.
Failure Prediction: Analysis of 100 wing components revealed significant variance in criticality probabilities across failure patterns.
- High Risk: Categories F (63.93% probability of being critical) and D (57.14%) showed the highest likelihood of early failure.
- Low Risk: Category B showed the lowest probability (16.67%).
Sensitivity Analysis: The model demonstrated robustness but showed sensitivity to cost coefficients. Increasing resource costs ( $e_k$ ) or assignment penalties ( $\lambda_m$ ) led to a significant rise in task deferrals, as the model restricted resource usage to minimize costs.

Significance and Claims

The paper claims to address a critical gap in aviation literature by providing a targeted solution for Line Maintenance Scheduling that integrates uncertainty management and predictive failure modeling.

Operational Resilience: The primary contribution is the embedding of a "memory" mechanism for deferred jobs directly into the optimization framework. Unlike static or heuristic approaches, this rolling-horizon method ensures that the schedule remains stable and predictable even as unexpected tasks arise.
Proactive Safety: By integrating the probabilistic failure model, the framework shifts maintenance from a reactive to a proactive stance. It allows operators to identify components at risk of early failure (specifically those in RCM categories D and F) and prioritize them, thereby enhancing operational safety and reducing emergency inspections.
Scalability and Adaptability: The framework is designed to be scalable across different fleet sizes and adaptable to various airline policies. The authors assert that the core principles—optimized scheduling, deferral tracking, and risk-based prioritization—are transferable to other industries requiring reliable scheduling under uncertainty, such as rail transport, maritime operations, and manufacturing.
Practical Implementation: The study demonstrates that complex stochastic optimization can be executed on standard computing hardware without high-performance resources, making the solution accessible for practical deployment in airline maintenance operations.