LED junction temperature prediction using machine learning techniques

Light Emitting Diodes (LEDs) are the longest lasting source of artificial illumination whose duration can exceed 50.000 continuous working hours. Nevertheless, they show a gradual reduction of the luminous flux due to the increase of the device temperature. In this work, a Machine Learning algorithm will be introduced and discussed, able to predict the junction temperature value of a LED in real-time while connected in the end-user circuit, taking into account current and voltage flowing in the device and, further, the actual model and aging of the LED. The algorithm was implemented on a microcontroller, showing the feasibility of performing edge machine learning on tiny yet powerful devices.