Work

Development of a Failure Prediction System for Industrial Laser Diodes Based on Artificial Neural Networks: Online First: 18/11/2026

Year: 2026
Type: article
Abstract: Industrial laser diodes are essential devices used in semiconductor chip manufacturing, where the precision and stability of laser output strongly affect the process yield. However, under long-time operation and high thermal stress, these lasers tend to degrade slowly, sometimes leading to unexpected failure that causes production interruption and cost increasing. In this work, a data-driven fault prediction system is proposed for early detection of degradation behavior in industrial laser diodes. The experimental data were collected from a Monaco 517-40-30 femtosecond laser operating in an actual production environment. The proposed system monitors a set of electrical, thermal, and optical parameters that reflect the operational condition and degradation characteristics of the laser diode. After normalization, noise filtering and outlier removal, a feedforward neural network was trained using the Levenberg-Marquardt algorithm with Mean Squared Error (MSE) and regularization to reduce the overfitting problem. The model achieved above 98.9% for both training and testing accuracy, with minimum F1-score and Recall values of 0.9. This shows that the system can do early fault detection quite reliably even in industrial conditions. From these results, the system is considered a practical solution for predictive maintenance and can help improve uptime and reliability of laser equipment.
Language: English
Open Access status: diamond
OpenAlex ID: W7236407252
FWCI:
Cites: 8
Cited by:
Related to: 10