Prediction and Analysis of High Power pcLED’s Useful Life Using Machine Learning Model: A ML Predictions with Physical Degradation Result

To address the rapid expansion of Solid-State Lighting applications, cost-effective deployment is essential. This requires an accurate estimation of the“useful life” of white phosphor-converted LEDs (pcLEDs). In this research article, high-power pcLEDs from four different manufacturers were tested for their “useful life” under tropical conditions. These tests were conducted using an Integrating Sphere (which follows LM-79/LM-80 standards) and a customized TeSUL chamber for approximately 10,000 hours. Under tropical conditions, the “useful life” of pcLEDs was measured with respect to their lumen maintenance change over time, temperature, and humidity. After recording the lumen maintenance, a machine learning regression model was fed with the collected data and was trained to find out the lumen maintenance of a pcLED sample when hours passed, temperature, and humidity were given as a query. The regression model achieved an accuracy score ranging from 92.36% to 94.19% for different manufacturers. Upon testing we found out that results derived from machine learning and the actual data had a deviation of $$\sim 1.5\%$$ to 1.9%, which proves that physical degradation data are closely allied with ML predictions. The results prove that combining ML results with internationally standardized testing methods can provide a robust and accelerated framework for reliable “useful life” predictions.