Abstract:

With the aviation sector coping with the unpredictability of aircraft engine problems, there is an urgent need for a transition to predictive maintenance models. This paper proposes a transformative predictive maintenance paradigm that greatly improves the forecast of aviation engine Remaining Useful Life (RUL). To process and analyze multivariate time-series data effectively, the proposed framework investigates a suite of complex machine learning models — Convolutional Neural Networks (CNN), Long Short-Term Memory networks (LSTM), Support Vector Machines (SVM), and Random Forests (RF). Importantly, this approach is enhanced with Explainable AI (XAI) principles via the use of SHapley Additive exPlanations (SHAP), which improves the predictability and transparency of the predictive characteristics. This new use of XAI enables stakeholders to obtain an intuitive knowledge of the model's outputs, laying the groundwork for trust and aiding informed decision-making in maintenance operations. The study describes the technological problems and approaches used to construct the predictive models, as well as how the combination of these models can estimate engine health with unparalleled accuracy and clarity. The findings point to a potential industry-wide shift in which powerful data and interpretable AI technologies inform maintenance methods, resulting in greater safety and efficiency in aircraft operations. This study promotes itself as a forerunner in the use of XAI in aviation, directing future research and practice towards more dependable and understandable predictive maintenance systems.