EURO 2025 Leeds
Abstract Submission

1767. Using Genetic Programming to generate Dispatching Rules for a Large-Scale Project Scheduling Problem

Invited abstract in session MC-38: Automating the Design, Generation and Control of Optimization Algorithms 2, stream Data Science meets Optimization.

Monday, 12:30-14:00
Room: Michael Sadler LG19

Authors (first author is the speaker)

1. Nuno Marques
Faculty of Engineering, University of Porto
2. Gonçalo Figueira
INESC-TEC, Faculty of Engineering of Porto University
3. Luís Guardão
INESC TEC
4. Luis Guimarães
INESC TEC, Faculadade de Engenharia, Universidade do Porto

Abstract

Scheduling in real-world settings is becoming increasingly challenging due to more complex products, pressures from customers and competitors, continuous adjustment to new technologies, and frequent unexpected events such as breakdowns, illness and supply chain disruptions. A sector in which scheduling is deeply affected by uncertainty is the aircraft maintenance and repair operations (MRO) business. Optimizing scheduling in MRO is important to reduce costs, plane unavailability and promote safety.
There are three main types of aircraft MRO: Line, Light and Heavy. Due to its complexity, this work focuses on heavy maintenance, as it encompasses thousands of tasks and can ground a plane for over two months. Moreover, uncertainty is constantly present in three main dimensions: planes' arrival dates, task durations, and a large amount of unplanned work. Lastly, aircraft MRO scheduling encompasses several subproblems: task scheduling, staff allocation and work centre allocation.
When facing such a complex, large and uncertain problem, one needs reactive solution methods such as dispatching rules (DRs) that can schedule tasks on the fly. Manually updating schedules or using approaches based on exact and meta-heuristics methods is not feasible as they would take too long. Therefore, in this work, Genetic Programming was used to generate new DRs for the aircraft MRO scheduling problem. These novel AI-generated DRs outperform existing DRs while maintaining a compact form.

Keywords

Status: accepted

Back to the list of papers