Source-linked AI summary
Mixed-integer optimization for multi-year military aircraft fleet management
Matteo Vescovi, Raffaele Giuseppe Cestari, Roberto Valdambrini, Andrea Mercurio, Valentina Breschi, Mara Tanelli
TL;DR
Defense flight and maintenance planning must address long horizons, use-dependent maintenance, and non-nominal fleet conditions. This paper proposes a multi-year mixed-integer approach with cyclic inspection and limited-dock constraints, plus strategies for computation and weight selection. The authors report advantages over myopic year-by-year planning and automatic tuning over manual weights on benchmark fleets.
Problem
Civil-sector flight and maintenance planning formulations do not fit defense settings with longer horizons, irregular use-dependent maintenance, and mission-oriented priorities.
Source: §1 · p. 2Method
The paper formulates multi-objective, multi-year mixed-integer fleet planning with cyclic inspections and limited maintenance resources, using receding horizons, aircraft-specific weights, and Bayesian Optimization.
Source: §1.1 · pp. 4–5Results
The authors report advantages over myopic year-by-year planning and automatic tuning over manual weights, alongside computational feasibility.
Source: §5.2 · p. 23, §5.3 · p. 25, §6 · p. 26Takeaways & Limitations
The approach is presented as relevant to defense fleets facing challenging initial conditions and long-term multi-objective planning.
Source: §1.1 · p. 5The current scheme assumes cyclic inspections, which may not reflect all military fleet operations.
Source: §6 · p. 27
Abstract
from arXiv · showhide
While existing strategies for Flight and Maintenance Planning for the defense sector generally address idealized conditions, real-world planning often involve non-nominal initial fleet states and complex inspection schemes. To address these challenges, we propose a multi-year planning strategy that maximizes long-term fleet availability. The formulation incorporates multiple competing objectives capturing the tight coupling between aircraft usage and maintenance, while enforcing cyclic inspection requirements and limited maintenance-dock capacity. To enhance practical applicability, we introduce three complementary strategies: a receding-horizon approach to reduce computational burden, a Bayesian Optimization routine to automatically tune cost-function coefficients, and an aircraft-specific weighting scheme that adapts model priorities based on fleet-condition statistics. Results on three realistic benchmark scenarios demonstrate the effectiveness of the approach.
1. Introduction
Defense flight and maintenance planning differs from civil planning in its longer horizons, use-dependent maintenance, and mission-oriented priorities. The paper proposes a multi-objective planning model and practical strategies for these conditions.
1. Introduction
- Civil-sector formulations built for short, route-driven operations do not fit defense planning’s longer horizons, irregular use-dependent maintenance, and non-profit priorities.The introduction contrasts civil planning, often daily or weekly, with mission-oriented defense planning.Source: §1 · p. 2
- Earlier defense-planning work includes helicopter-specific linked subproblems, use of bounds in mixed-integer optimization, and models for uncertainty or mission waves.Other studies also address computational cost through machine-learning acceleration and heuristic warm starts.Source: §1 · pp. 2–4
1.1. Contribution
- The model combines use-dependent fleet planning with cyclic, multi-type inspections and limited maintenance resources, then adds three strategies for challenging initial conditions and long-term optimization.The strategies are receding-horizon planning, aircraft-specific objective weights, and Bayesian Optimization of loss-function weights.Source: §1.1 · pp. 4–5
1.2. Outline
- The paper proceeds from planning objectives and constraints to a mathematical formulation, practical solution strategies, benchmark results, and future directions.This is the paper’s stated organization.Source: §1.2 · p. 5
2. Introducing planning objectives and constraints
The planning objectives connect aircraft use to maintenance-dock logistics and fleet-owner preferences. The formulation combines readiness goals with operational and inspection constraints.
2. Introducing planning objectives and constraints
- Objectives and constraints are enforced monthly, and the planning problem is linked to maintenance infrastructure logistics and fleet-owner procedures.The section introduces the objectives in natural language before formalizing them mathematically.Source: §2 · pp. 5–6
2.1. Fleet planning goals
- The goals balance dock saturation, minimum flying for aircraft outside maintenance, balanced flight-hour use, and faster inspection entry for aircraft nearing inspection.The framework assumes docks are sized to fleet needs when all are occupied; the goals also guard against many aircraft having low residual flight hours.Source: §2.1 · p. 6
- The planner can weight flight-hour use toward particular periods, while fleet and squadron flight-hour targets may be enforced as hard or soft constraints.Slack variables represent deviations from targets when soft enforcement is used.Source: §2.1 · p. 7
2.2. Fleet planning constraints
- All aircraft follow a repeated four-type, use-triggered inspection cycle; major inspections take longer and require specialized docks.Each dock serves one aircraft at a time, and aircraft can occupy only one dock; only a subset of docks can perform major inspections.Source: §2.2 · p. 7
- When not in inspection, aircraft must meet user-defined minimum flight hours unless that threshold is zero, and flight is capped by a maximum threshold.Initial and boundary conditions complete the operational framework.Source: §2.2 · p. 7
3. Formulating the FMP problem
The formulation translates flight and maintenance planning goals into a multi-objective optimization model whose decisions jointly govern aircraft usage, maintenance, and availability under resource constraints.
3. Formulating the FMP problem
- The objectives promote occupied inspection docks, minimum flying by aircraft outside maintenance, and a more balanced distribution of flight hours.A quadratic loss promotes homogeneous flight-hour allocation, while an added cost encourages consuming hours nearer maintenance.Source: §3 · p. 8
- The maintenance index tracks the inspection cycle, while residual flight hours decrease with use and are replenished after inspection.The index advances after inspections and resets to zero at the last, major inspection.Source: §3 · pp. 9–10
- A tunable parameter increases the cost of flight-hour consumption nearer the end of the planning horizon.The parameter γ is constrained to be nonnegative.Source: §3 · p. 10
- Monthly flight hours are constrained by residual hours to the next inspection and squadron-specific maximums.The model also uses squadron-specific slack variables to allow and penalize deviations from flight-hour targets.Source: §3 · p. 11
- Residual maintenance time tracks inspection duration, and dock-capacity constraints separately limit aircraft in major and non-major inspections.The model restores maintenance time when an inspection begins and restricts major inspections to equipped docks.Source: §3 · pp. 11–13
- Availability is constrained to remain consistent with maintenance status, so an aircraft with positive residual maintenance time cannot be flagged as available.Source: §3 · pp. 8, 13
3.1. FMP problem for defense aircraft fleets with cyclic maintenance
- The model defines aircraft, dock, month, and squadron indices for its decision variables and combines the resulting losses in the overall FMP problem.The formulation merges the losses and constraints into a single optimization problem.Source: §3 · p. 8, §3.1 · pp. 13, 15, Table 2 · p. 14
- The overall FMP cost is a weighted combination of five loss terms minimized over the planning variables.The formulation combines losses for dock use, flight-hour allocation, availability-related goals, flight-hour consumption timing, and target deviations.Source: §3.1 · pp. 13–15
4. Solution enhancing strategies
The section offers three practical strategies for making multi-year flight and maintenance planning more computationally manageable, easier to tune, and adaptable to difficult initial fleet conditions.
4.1. Coping with the computational burden: receding horizon
- Receding-horizon planning optimizes a shorter window, keeps only the first year’s decisions, and warm-starts each one-year shift with the remaining plan.This supports smoother transitions in flight-hour use and inspections, but is suboptimal to full-horizon planning; a longer window can reduce the gap at greater computational cost.Source: §4.1 · pp. 15–16
4.2. Bayesian optimization for weight tuning
- The tuning loss combines goals for squadron scaling and maintenance-dock use, matching fleet scaling to its target while avoiding idle docks.A scaling factor makes the two loss components comparable.Source: §4.2 · pp. 16–17
- Bayesian optimization automates penalty-weight tuning by iteratively selecting candidate weights from a surrogate model and evaluating them by solving the FMP problem.The search is user-bounded, and the procedure is intended to reduce manual tuning and help address inconvenient initial fleet conditions.Source: §4.2 · pp. 18–19
4.3. Hierarchical scheme for unbalanced initial fleet conditions
- For critical initial fleet states, aircraft-specific weights use distances from target scaling and upcoming maintenance to set flight and inspection priorities.Scaling position prioritizes inspection entry; flight-hour use is encouraged above target scaling and discouraged below it, while proximity to inspection further discourages use.Source: §4.3 · p. 19
- The hierarchy modifies weights w1, w2, and w3; the penalties w4 and w5 remain unchanged.The weight hierarchy shapes aircraft prioritization and flight-hour consumption.Source: §4.3 · pp. 19–20
5. Experimental results
The experiments assess the planning approach on three benchmark fleets over five years, using year-by-year and receding-horizon routines because a five-year one-shot solution was not practically computable on the employed hardware. The section also introduces performance indicators for evaluating the resulting plans.
5. Experimental results
- The benchmarks include a standard fleet near ideal scaling, a critical fleet with scarce residual flight hours, and a fleet with new aircraft forming a scaling “wall.”Each fleet has 25 aircraft in two squadrons; planning uses five docks, including two equipped for major inspections.Source: §5 · pp. 20–21
5.1. Performance indicators
- The indicators include residual flight hours to the next major inspection at each year-end; a lower yearly average means more aircraft are near major maintenance, risking queues and mission delays.The five-year horizon is evaluated with year-by-year planning and a receding-horizon routine using a two-year window; a one-shot solution was not practically computable on the employed hardware.Source: §5 · p. 20, §5.1 · p. 22
scaling loss
The section defines flight-hour surplus and deficit, alongside empty docks, as indicators of planning quality. It compares receding-horizon planning, Bayesian Optimization tuning, and hierarchical aircraft weights across benchmark fleets.
scaling loss
- Flight-hour surplus and deficit indicate excesses or shortfalls that can signal potential inspection queues; empty docks indicate maintenance-planning quality.Idle docks may also undermine fleet operational readiness.Source: scaling loss · pp. 22–23
5.2. Receding horizon vs year-by-year planning
- Receding-horizon planning remains feasible over five years when year-by-year planning becomes infeasible, while increasing average yearly RFH and reducing empty docks.The comparison uses a two-year prediction window; gains are less pronounced for the fleet with favorable initial conditions.Source: Table 5 · p. 23, §5.2 · p. 23
5.3. Focusing on the critical fleet: the impact of Bayesian Optimization
- Bayesian Optimization globally reduces empty docks and improves scaling alignment versus manual tuning on the critical fleet over a 10-year horizon.Figure 3 tracks empty docks; Figure 4 shows scaling under manual and BO-based tuning.Source: Figure 3 · p. 24, Figure 4 · p. 24, §5.3 · p. 25
5.4. Enhancing new aircraft handling with the hierarchical approach
- For the new fleet, hierarchical penalties reduce flight-hour surplus but do not reduce deficit, while spreading residual flight hours more evenly to align inspections and reduce queues.The comparison extends to 10 years; Figure 5 shows scaling with and without hierarchical weights.Source: Figure 5 · p. 25, §5.4 · p. 26
6. Conclusions and future work
The paper formulates multi-year, multi-objective flight and maintenance planning for defense fleets and reports benefits from receding horizons, automatic penalty tuning, and aircraft-specific weights. Its current cyclic-inspection assumption limits scope, and future work targets broader inspection settings and unexpected events.
6. Conclusions and future work
- The proposed approach plans defense-fleet usage and maintenance to maximize availability under operational constraints rather than economic objectives.Reported benefits include improved planning over myopic year-by-year planning, automatic tuning, and aircraft-specific weighting.Source: §6 · p. 26
- Receding horizons ease computation, Bayesian Optimization tunes cost penalties, and aircraft-specific weights address similar residual flight hours and inspection queues.The paper reports benchmark benefits for each strategy, including reduced user burden as a potential benefit of automatic tuning.Source: §6 · p. 26
- The method assumes cyclic inspections; future work targets acyclic and calendar-dependent inspections, maintenance anticipations and delays, and robustness to unexpected or stochastic events.Source: §6 · p. 27