Physics-informed hybrid learning for predicting output behavior of 2 µm cross-level solid-state lasers.

Area of Science:

• Laser Physics and Engineering
• Materials Science
• Computational Physics

Background:

• Mid-infrared solid-state lasers are crucial for applications like medical therapy, lidar, and precision manufacturing.
• Accurate modeling of coupled gain, thermal, and optical effects in these lasers is challenging due to complex nonlinear behaviors.
• Traditional analytical models often fail to capture the intricate interactions within crystal laser systems.

Purpose of the Study:

• To develop a physics-informed, multi-model fusion machine learning framework for advanced laser modeling.
• To systematically extract crystal properties and improve dynamic process modeling in complex laser systems.
• To provide a novel tool for physical understanding and structural optimization of solid-state lasers.

Main Methods:

• Developed a physics-informed machine learning framework integrating multiple models.
• Systematically extracted physical, chemical, and thermal properties of laser crystals.
• Validated the framework on Tm: YAP, Tm: YAG, and Tm/Ho: YLF laser systems.

Main Results:

• Achieved high-precision output predictions across wide pump ranges and thermal limits, with R² = 0.952 near thermal instability.
• Demonstrated model robustness near thermal instability, with low RMSE (0.276) and MAE (0.183).
• Showcased excellent generalization in Tm/Ho: YLF lasers with minimal relative error (0.102%) under matrix variations and cross-level energy transfer.

Conclusions:

• The proposed framework significantly enhances the modeling accuracy of complex mid-infrared solid-state lasers.
• The method effectively identifies dominant performance factors and aids in structural optimization.
• This approach offers a novel and efficient tool for advancing laser technology and understanding.