Parametrizable Model for Thermal Behaviour Estimation and Control of LED Automotive Headlamps using Reinforcement Learning

Light Emitting Diodes (LEDs) are currently widely applied in automotive headlamps. Moreover, the manufacturing of small-scaled LED packages has encouraged the development of headlamps using a matrix of LEDs, providing highly adaptive light beams. However, the use of LEDs produces heat, which affects the lighting quality as well as lifetime of the headlamp. This issue becomes particularly challenging for LED matrix models, wherein the heat generated by each accumulates and increases the headlamp temperature drastically, thereby augmenting the risk of failure due to overheating. For these modules. The present paper uses a modeling approach to investigate the effect of input power on headlamp temperature and subsequently designs a control law to determine the optimal input power. Our paper focuses on devising a rapid estimator that can be deployed for various designs of headlamps, so that it can serve as the basis of an assessment design for the thermal derating control algorithm. Thus, a generic parameterizable mathematical model for the thermal behaviour is proposed and validated, based on the thermal-electrical equivalent circuit model. As well, a control policy for operating the headlamp is developed based on the proximal policy optimization Reinforcement learning based algorithm.