Enhanced LED Lamp Fault Diagnosis using Optimized XGBoost with Hilbert and Wavelet Feature Extraction

This paper presents the machine learning method for LED lamp fault diagnosis, applied using advanced signal processing techniques. Conventional fault detection methods depend solely on optical parameter measurements, which may not find subtle defects. In contrast, here, a high-frequency light output signals recorded by a photodiode, followed by feature extraction using Hilbert and wavelet transforms is presented. It uses Greedy Forward Selection for feature selection. The fault classification is executed by using a Bayesian Optimization approach, with an Optuna-based tuning of the used XGBoost classifier. The result shows improvement in diagnostic performance, the accuracy going up from 92–94%, precision from 0.91 to 0.92, recall from 0.92 to 0.94, F1 score from 0.9 to 0.93 and Area Under the Curve(AUC) from 0.92 to 0.96 after tuning. The results confirm the ability of the proposed methodology to provide improved robustness and reliability of LED lamp fault detection.