Modelling the Voltage Degradation of High-Power LEDs Using approach based on BayesianOptimized Bidirectional Long Short-Term Neural Network

Modelling the Voltage Degradation of High-Power LEDs Using approach based on BayesianOptimized Bidirectional Long Short-Term Neural Network

Stať ve sborníku v databázi WoS či Scopus

High-power LEDs have become integral components in modern systems, including lighting, signalling, visible communication, medical applications, and other critical fields. In addition to their practical applications, these LEDs have garnered significant attention in reliability research. Key objectives in this domain include data collection, degradation modelling, reliability prediction, and comprehensive reliability assessment. This study introduces a novel methodology based on Bayesian optimized-Bidirectional Long Short-Term Memory Neural Network to model and predict the degradation of LEDs under ageing test conditions. The proposed approach leverages the strengths of these techniques to capture the nonlinear and complex degradation behaviours of LEDs effectively. The performance of the model is rigorously verified using widely accepted metrics such as Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), and R-squared (R-2). The results indicate that the proposed method offers robust and accurate predictions, showcasing its potential as a reliable approach for modelling and predicting the reliability of high-power LEDs. This approach contributes to advancing the field of LED reliability research and supports the development of innovative solutions for predicting the performance and lifespan of these critical devices.

High-power LEDs have become integral components in modern systems, including lighting, signalling, visible communication, medical applications, and other critical fields. In addition to their practical applications, these LEDs have garnered significant attention in reliability research. Key objectives in this domain include data collection, degradation modelling, reliability prediction, and comprehensive reliability assessment. This study introduces a novel methodology based on Bayesian optimized-Bidirectional Long Short-Term Memory Neural Network to model and predict the degradation of LEDs under ageing test conditions. The proposed approach leverages the strengths of these techniques to capture the nonlinear and complex degradation behaviours of LEDs effectively. The performance of the model is rigorously verified using widely accepted metrics such as Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), and R-squared (R-2). The results indicate that the proposed method offers robust and accurate predictions, showcasing its potential as a reliable approach for modelling and predicting the reliability of high-power LEDs. This approach contributes to advancing the field of LED reliability research and supports the development of innovative solutions for predicting the performance and lifespan of these critical devices.

High-power LEDs, reliability, degradation, Bayesian optimization, LSTM, Bi-LSTM

High-power LEDs, reliability, degradation, Bayesian optimization, LSTM, Bi-LSTM

LA, Q.; VINTR, Z.; VALIS, D.; ŽÁK, L.; KOHL, Z.

27.05.2025

IEEE345 E 47TH ST, NEW YORK, NY 10017 USA

NEW YORK

979-8-3315-2339-8

INTERNATIONAL CONFERENCE ON MILITARY TECHNOLOGIES

327

333

7