Aircraft Maintenance Scheduling Problem

• Aircraft Maintenance Scheduling Problem is a combinatorial optimization challenge ensuring timely maintenance while minimizing costs and operational disruptions.
• It encompasses diverse models from finite-horizon flight assignments to real-time heuristics, addressing regulatory compliance and multi-resource constraints.
• Recent methodologies integrate predictive maintenance and metaheuristics, achieving notable cost reductions and improved fleet readiness.

The Aircraft Maintenance Scheduling Problem (AMSP) is a multifaceted, combinatorial optimization problem critical to the management of aircraft fleets across civil and military domains. Its core objective is to ensure that aircraft receive required maintenance within prescribed safety, regulatory, and operational constraints, while simultaneously minimizing costs and service disruptions. AMSP encompasses a heterogeneous landscape of models, spanning periodic and non-periodic flight schedules, preventive and corrective maintenance actions, multi-resource assignment and allocation, and varying time-scales from real-time dispatching to long-term strategic planning.

1. Formal Models and Core Problem Variants

AMSP admits multiple mathematical formulations, each reflecting operational realities and regulatory regimes.

1.1 Finite-Horizon Flight and Maintenance Assignment

The classical one-shot AMSP formulation assigns a unique set of flight legs $\mathcal{Q}$ to a fleet $\mathcal{A}$ over a discrete planning horizon $H$, subject to maintenance base visits and interval $\gamma$ (maximum number of consecutive days between required maintenance). Typical decision variables are $x_{a\ell} \in \{0,1\}$, indicating assignment of aircraft $a$ to flight $\ell$, and $y_{a,d} \in \{0,1\}$, representing maintenance at the end of day $d$. The model incorporates flow conservation, coverage, and maintenance window constraints:

$$\sum_{k=0}^{\gamma-1}y_{a,d+k}\;\ge\;1 \quad \forall\,a\in\mathcal{A},\;d=1,\dots,H-\gamma+1$$

This ensures that each aircraft visits a maintenance base at least once within any consecutive $\mathcal{A}$0-day window (Meunier et al., 7 Aug 2025).

1.2 Periodic Assignment and Maintenance Circuit Decomposition

In periodic settings (identical daily flight sets), the problem reduces to decomposing daily line-of-flight arcs into periodic walks $\mathcal{A}$1, each visiting maintenance bases within $\mathcal{A}$2 arcs. The main existence theorem (Meunier et al., 7 Aug 2025) establishes that for $\mathcal{A}$3, absolutely periodic solutions (“Maintenance Circuit Decompositions”) always exist and can be found in polynomial time, whereas for $\mathcal{A}$4, optimal periodicity is not guaranteed.

1.3 Maintenance Task and Workforce Scheduling

Beyond routing, AMSP extends to multi-stage scheduling of maintenance work-packages (WPs) and work-orders (WOs), considering eligibility, sequence, and resource (technician) assignment. The combinatorics grow with concurrent qualification constraints and non-preemptive task dependencies, as in the AERO-EA evolutionary framework (Urquhart et al., 19 Dec 2025):

$\mathcal{A}$5

where $\mathcal{A}$6 counts uncovered staff slots (resource infeasibility) and $\mathcal{A}$7 counts tardy aircraft.

1.4 Degradation-Aware Predictive Maintenance

Recent approaches couple survival analysis with cost-optimization, exploiting real-time condition monitoring. The decision is a hazard-threshold $\mathcal{A}$8 defining proactive replacement, with total cost comprising preventive ($\mathcal{A}$9) and failure ($H$0) events based on a Cox proportional hazards model (Ziyad et al., 2022):

$H$1

2. Solution Methodologies

AMSP is NP-complete in its general forms even for moderate parameter settings, necessitating sophisticated algorithmic strategies.

2.1 Exact Methods: MIP and Continuous-Time MILP

Mixed Integer Programming (MIP) remains the default for small to medium-scale, fully-specified AMSP instances (Peschiera et al., 2020, Pazhooh et al., 4 Aug 2025, Urquhart et al., 19 Dec 2025). The continuous-time MILP paradigm exploits event-based formulations, where time is modeled explicitly as continuous decision variables rather than discretized intervals (Pazhooh et al., 4 Aug 2025), yielding tighter formulations and superior scalability for integrated scheduling and spatial layout of hangar operations.

2.2 Decomposition and Metaheuristics

Logic-based Benders decomposition (LBBD) separates assignment (MIP) and sequencing/resource-feasibility (CP or local search) subproblems, accelerating convergence, especially in dynamic repair shop settings (Bajestani et al., 2014). Evolutionary algorithms (EAs), notably the AERO-EA, employ permutation representations, order-preserving crossover, and steady-state replacement to maintain both feasibility and diversity across rugged search spaces (Urquhart et al., 19 Dec 2025). Genetic algorithms leverage precedence repair, resource-based decoding, and dispatching rule-driven initializations for resource-constrained project scheduling (Pimapunsri et al., 2022).

2.3 Real-Time and Heuristic Adaptations

Real-time insertion heuristics dynamically adapt the schedule to late-arriving corrective tasks, seeking to minimize idle slots through local window fitting and competence-weighted assignments (Aboussalim et al., 2010). Such methods deliver substantial cost reductions in lost-time, with performance largely independent of the number of corrective insertions given moderate schedule density.

3. Key Constraints and Operational Dimensions

A variety of operational and regulatory constraints shape AMSP formulations:

• Maintenance intervals: Enforced via sliding-window or circuit constraints to comply with airworthiness directives.
• Resource assignments: Technician/crew matching with qualification matrices; non-overlapping shifts; shift-length limits (Urquhart et al., 19 Dec 2025).
• Temporal sequencing: Non-preemptive, intra-package precedence; hangar entry/exit event ordering; blocking and buffer windows (Pazhooh et al., 4 Aug 2025).
• Capacity and sustainability: Maintenance bay/shop capacity, fleet cluster serviceability/sustainability to prevent localized deficits (Peschiera et al., 2020).
• Dynamic arrivals and uncertainty: Stochastic failure and repair-shop job creation (Bajestani et al., 2014), modeled via expectation or incorporated as dynamic reoptimization steps (receding horizon, rolling horizon).

4. Computational Complexity and Tractability

NP-completeness is established for the finite-horizon AMSP for $H$2 (Meunier et al., 7 Aug 2025), as well as for long-term block planning problems that generalize fixed-interval scheduling (Peschiera et al., 2020). However, certain periodic or “quiet-night” special cases are solvable in polynomial time for fixed fleet size, utilizing dynamic programming over depot-layered path partitions (Meunier et al., 7 Aug 2025). Decomposition-based and metaheuristic approaches provide practical tractability for large-scale, real-world scheduling (Bajestani et al., 2014, Pimapunsri et al., 2022, Urquhart et al., 19 Dec 2025).

5. Numerical Results and Practical Insights

Benchmarking studies confirm major sources of difficulty: tight resource windows, dynamic arrivals, and high workload-to-capacity ratios. Key findings include:

• Continuous-time MILP achieves proven optimality for hangar scheduling instances up to 25–40 aircraft—exceeding discrete-time formulations in efficiency and cost savings, particularly as temporal/spatial density increases (Pazhooh et al., 4 Aug 2025).
• Predictive maintenance using hazard-threshold optimization delivers 37–50% cost reductions versus naïve strategies, with per-cluster policy tailoring yielding an additional 13% savings (Ziyad et al., 2022).
• Evolutionary algorithms robustly deliver fully feasible schedules under both relaxed and tight conditions, with solution time distribution reflecting the multi-modal nature of the search space (Urquhart et al., 19 Dec 2025).
• Logic-based Benders decompositions boost long-term fleet readiness by 10% over static or myopic reschedule policies under dynamic failure, with expected daily operational run-times (Bajestani et al., 2014).
• Real-time scheduling heuristics, utilizing competence-weighted window insertion, reduce lost-time (idle slot) costs by up to 54% with negligible computational overhead (Aboussalim et al., 2010).

6. Extensions, Open Questions, and Research Directions

Robust optimization under parameter uncertainty, stochastic arrivals, and multi-level (long/medium/short-term) check integration remain open research arenas (Peschiera et al., 2020, Bajestani et al., 2014). For periodic routing, polynomial-time existence of absolutely periodic decompositions for $H$3 is unresolved, as are refined boundaries of tractability for various practical constraints (Meunier et al., 7 Aug 2025). Emerging lines of work also prioritize decision support and dashboard visualization for operational transparency and real-time rescheduling (Pazhooh et al., 4 Aug 2025, Aboussalim et al., 2010). Hybrid EA/local search and scenario-based stochastic programming are identified as promising development threads.

7. Summary Table of AMSP Models

AMSP Variant	Core Decision Variables	Main Constraint Classes	Representative Methods
Finite-horizon Routing	$H$4, $H$5	Coverage, flow, maintenance windows	MIP, DP, heuristics
Periodic Assignment	$H$6 (walks)	Arc coverage, $H$7-bounded circuits	Matching, flow, DP
Maintenance-Workforce	$H$8, $H$9	Precedence, technician, shift, covering	EA, MIP, repair
Predictive Maintenance	$\gamma$0, $\gamma$1	Hazard-threshold, cost, failure-prob	Cox+sim-opt, bootstraps
RCPSP-heavy maint.	$\gamma$2, $\gamma$3, $\gamma$4	Precedence, resource, duration	GA (EST/WEST), repair
Real-time scheduling	$\gamma$5, $\gamma$6 (crew)	Windows, earliness/tardiness, competence	Greedy, window-insertion
Dynamic repair shop	$\gamma$7, $\gamma$8	Assign/repair, capacity, availability	LBBD, rolling horizon

These models collectively constitute a rich theoretical and practical foundation underlying the aircraft maintenance scheduling literature, each contributing distinct analytical and computational tools for the resolution of large-scale, safety- and cost-critical fleet management problems.