NOVEL APPROACH FOR FAULT DIAGNOSIS OF HIGH-SPEED TRAIN BEARINGS USING TRANSFORMER MODELS OPTIMIZED BY AQUILA OPTIMIZER AND LIGHT GRADIENT BOOSTING MACHINE

Abstract

High-speed train bearing fault diagnosis is crucial for rail system safety and reliability. Traditional methods like vibration analysis often suffer from low accuracy and long processing times. This paper introduces an advanced diagnostic approach using Improved Aquila Optimizer (IAO) and Light Gradient Boosting Machine (Light GBM) combined with Laser-Induced Fluorescence (LIF) technology. The Aquila Optimizer was enhanced with opposition-based learning and Nelder-Mead search methods to optimize Light-GBM parameters. Various detailed bearing fault types, including Inner and outer race scratch and wear, roller ring failures, cage and compound faults, were analyzed. LIF technology captured spectral images, and multivariate scattering correction (MSC) and normalization prepared the fluorescence curves. Kernel-based principal component analysis (KPCA) reduced data dimensionality, which was then used to train the Light-GBM model. The IAO optimized the model, demonstrating superior convergence and prediction accuracy compared to particle swarm optimization and the original Aquila Optimizer. The MSC-IAO-Light GBM model achieved the best fault prediction results, with an MSE of 7.0521 × 10-9, a MAE of 5.451 × 10-2, and R² approaching 1. This method offers a novel approach for high-speed train bearing fault detection, enhancing rail system stability and safety.