Key points are not available for this paper at this time.
A novel intelligent framework is presented that integrates supervised machine learning with a mathematical rule-based system to detect, localize, and diagnose electro-thermal faults in a network of multiple light-emitting diode (LED) luminaires. The proposed system employs a sensorless data acquisition system for cost-effectiveness. Based on the electrical data captured, a fault detector initially determines the presence of faulty luminaire(s) in the group. Once a fault is detected, a fault locator module pinpoints the exact faulty luminaire by referencing its serial number. Next, a fault diagnostician classifies the specific type of electro-thermal fault affecting the identified luminaire. If the diagnosed fault is open or short circuit, an additional logic block is activated that specifies the faulty LED string within that luminaire. The intelligent framework employs two machine learning models: logistic regression for fault detection and extreme gradient boosting (XGBoost) for fault diagnosis. It is found that both ML models deliver high accuracy (>99%), strong generalization (generalization errors ≤0.8%), near-perfect correlation between the predicted and the actual data (Brier scores ≤0.0029) and rapid execution (≤2.01 seconds). Multiple case studies show that the complete intelligent system demonstrates 100% effectiveness in fault detection, localization, and diagnosis which affirms its suitability for real-world deployment.
Biswadeep Gupta Bakshi (Fri,) studied this question.