End-to-end predict-and-optimize dynamic predictive maintenance planning integrating prognostics - the case of short-range electric aircraft with lithium-ion batteries

Abstract

Modern assets are continuously monitored by sensors. As a result, large datasets on the health condition of these systems are often available. Using supervised machine learning, recent studies have leveraged such data to generate remaining useful life (RUL) prognostics. Here, the focus of the machine learning regressors is on achieving prognostics of high accuracy. Once obtained, in a second stage, these prognostics are usually integrated into maintenance planning optimisation models. However, aiming for high accuracy prognostics in a first stage does not guarantee that the maintenance costs are also minimized in a second, maintenance planning stage. To address this, we propose an end-to-end, dynamic framework for the predictive maintenance problem that integrates the planning stage into the prediction stage. For this, the maintenance costs are directly estimated from sensor data, instead of being derived based on RUL prognostics. We apply our end-to-end framework for the maintenance planning of a fleet of electric Vertical Take-Off and Landing (eVTOL) aircraft equipped with Lithium-ion batteries. We show that, when compared to state-of-the-art prognostics-based maintenance planning, the proposed framework reduces the number of battery failures by 24% and the total maintenance costs by 9.4%. Overall, our framework proposes an effective, data-driven paradigm for an end-to-end predictive maintenance planning.