Simulation Methods for LED Multi-Domain Models Parameter Extraction

This paper presents a new approach to estimating the parameters of the Delphi-like multi-domain light-emitting diode (LED) model. The increased demand for digital twins in the LED industry requires an efficient and robust method to fit the multi-domain (electric, optical, and thermal) model parameters. The accurate non-linear model contains 24 parameters, making the fitting process a computational challenge. For parameter extraction, we used two methods; the brute force, and the minimax algorithms. We assume that the brute force method finds the correct values of the parameters, but the computational cost is high. We expect that the minimax method will find the solution in good agreement with the brute force, but with much lower computational demand. Therefore, this research uses both methods on a few measurement data recorded previously in the Delphi4LED project. We can benchmark the accuracy of the parameter estimation, by recalculating the IVL data and comparing that to the measured ones by estimating the root mean square error (RMSE) and maximum relative error values. We found that the brute force resulted in 2% accuracy for the electrical model and 5% accuracy for the optical model with a maximum run time approximately of 76h, while the minimax resulted in 3% accuracy for the electrical model and 7% accuracy for the optical model with a maximum run time 54 sec. is still acceptable.