Measuring and prognosis of remaining useful life of light-emitting diodes based on nonlinear fuzzy inference system

Measuring and prognosis of remaining useful life of light-emitting diodes based on nonlinear fuzzy inference system

Článek WoS

Measuring and predicting the remaining useful life (RUL) of products and engineering systems is crucial for effective health monitoring and maintenance planning. The key challenges in RUL prediction lie in acquiring relevant health indicators and constructing accurate predictive models based on these indicators. However, direct health indicator data that reflect product degradation are not always accessible; in some cases, only indirect informative measurements are available. This article addresses such a scenario with light-emitting diodes (LEDs). The article focuses on finding a feasible approach to RUL prediction using a non-linear fuzzy inference system (FIS). We introduce an optimization/training framework that integrates Bayesian optimization, multiobjective genetic algorithms, regression techniques, and F-Test feature selection to estimate the model's structural and operational parameters effectively. Online RUL prediction and parameter adaptation are achieved through the approaches based on the particle filter (PF), Huber likelihood, and recursive least squares (RLS) methods. The proposed methodology demonstrates promising predictive performance, enabling the prediction of RUL based on available indirect measurements.

Measuring and predicting the remaining useful life (RUL) of products and engineering systems is crucial for effective health monitoring and maintenance planning. The key challenges in RUL prediction lie in acquiring relevant health indicators and constructing accurate predictive models based on these indicators. However, direct health indicator data that reflect product degradation are not always accessible; in some cases, only indirect informative measurements are available. This article addresses such a scenario with light-emitting diodes (LEDs). The article focuses on finding a feasible approach to RUL prediction using a non-linear fuzzy inference system (FIS). We introduce an optimization/training framework that integrates Bayesian optimization, multiobjective genetic algorithms, regression techniques, and F-Test feature selection to estimate the model's structural and operational parameters effectively. Online RUL prediction and parameter adaptation are achieved through the approaches based on the particle filter (PF), Huber likelihood, and recursive least squares (RLS) methods. The proposed methodology demonstrates promising predictive performance, enabling the prediction of RUL based on available indirect measurements.

Light-emitting diode, Reliability, Remaining useful life, Non-linear fuzzy inference system, Data structural change, Rule selection, Online prediction

Light-emitting diode, Reliability, Remaining useful life, Non-linear fuzzy inference system, Data structural change, Rule selection, Online prediction

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

07.01.2026

MEASUREMENT

264

January 2026

Spojené království Velké Británie a Severního Irska

333

355

23

https://doi.org/10.1016/j.measurement.2026.120322