BeMALGS: A Multi-Stage Benchmark for Aircraft Landing Gear Systems Predictive Maintenance

Abstract

Predictive Maintenance (PdM) is increasingly adopted in safety-critical aviation systems, including landing gear systems, where unexpected failures can compromise mission completion and increase operational costs. Recent studies have tackled this issue by developing separate solutions for Anomaly Detection, Failure Diagnosis, and Degradation Prognosis. As a result, there is a significant gap in the literature for a unified framework that facilitates sequential benchmarking within a cohesive PdM pipeline. Thus, this work introduces BeMALGS, the first comprehensive benchmark that explicitly covers the entire PdM pipeline for a Landing Gear System (LGS). By providing a unified framework, BeMALGS enables the evaluation of the transition from Anomaly Detection to Failure Diagnosis and, finally, Degradation Prognosis, also known as Remaining Useful Life (RUL) estimation. To establish a robust and reproducible baseline, we leverage the AutoGluon AutoML framework, which automates model selection, feature engineering, and hyperparameter optimization across multiple learning paradigms. The resulting benchmark provides an open and modular framework for accelerating the development of integrated predictive maintenance systems that cover the sequential stages of the maintenance lifecycle for aerospace systems. The dataset is available at https://github.com/esd-univr/BeMALGS.