Paper Title
REMAINING USEFUL LIFE PREDICTION USING TRANSFORMER MODELS: AN XAI-DRIVEN EVALUATION WITH LIME, SHAP, AND GRAD-CAM

Abstract
Artificial Intelligence (AI) is a cornerstone of Industry 4.0, driving advanced predictive diagnostics that play a crucial role in ensuring the reliability and efficiency of industrial machinery through precise maintenance strategies. Rotating machines, integral to manufacturing processes, are particularly prone to wear and failure, making accurate Remaining Useful Life (RUL) prediction critical for minimizing downtime and maintenance costs. Accurate Remaining Useful Life (RUL) prediction is of paramount importance in the aerospace industry due to the high stakes involved in aircraft safety and operational efficiency. However, the "black box" nature of AI models, particularly deep learning models, often obscures the reasoning behind their predictions, posing challenges for their adoption in safety-critical environments. To address this, we propose an approach that combines a Transformer-based model for RUL prediction with Explainable AI (XAI) techniques to elucidate the model's decision-making process. Our methodology unfolds in three stages: 1) Processing and filtering of sensor data; 2) RUL prediction using a Transformer model, and 3) application of XAI techniques, including Local Interpretable Model-Agnostic Explanations (LIME), Shapely Additive Explanations (SHAP), and Gradient-weighted Class Activation Mapping (Grad-CAM), to interpret the predictions. The integration of XAI methods provides human-understandable insights into the model’s predictions, enhancing trust, and facilitating informed decision-making in predictive maintenance. Our results demonstrate that the Transformer model, coupled with XAI, not only achieves high accuracy in RUL prediction but also offers transparency and interpretability, which are crucial for its deployment in industrial settings.