High‐Speed Train Bearing Health Assessment Based on Degradation Stages Through Diagnosis and Prognosis by Using Dual‐Task LSTM With Attention Mechanism

ABSTRACT Health assessment of bearings in high‐speed trains is critical for smooth operation and maintenance. Bearings are subject to high mechanical stresses and harsh environmental conditions, making their health assessment essential throughout their working life. Health assessments based on degradation thresholds can predict more accurate fault classification and remaining useful life, which is suitable for deciding whether to consider end of life or apply proactive maintenance strategies. Traditional approaches predict singular outcomes using diagnosis or prognosis. This study proposes a dual‐task long‐short‐term memory model with an attention mechanism using kernel principal component analysis for feature extraction, and the Weibull failure function enables a strong framework for fault diagnosis and residual life prediction. The performance of the proposed model has an edge over existing traditional methods. The combined outcomes with the high accuracy of predictions and improved maintenance planning ultimately lead to better operational reliability, allowing more informed decisions to support the system that avoids unnecessary replacements and minimizes maintenance.