Deciphering optical coupled resonant systems with physics-data co-driven deep neural networks.

Area of Science:

• Physics
• Data Science
• Engineering

Background:

• Coupled mode theory (CMT) is widely used for resonant systems.
• Traditional fitting methods struggle with multi-solution scenarios in CMT.
• Implicit physical parameters are crucial for understanding resonant systems.

Purpose of the Study:

• To develop a novel method for predicting physical parameters of complex resonant systems.
• To address and mitigate the multi-solution challenge inherent in CMT.
• To enhance the speed and precision of parameter prediction in resonant system analysis.

Main Methods:

• Proposed a CMT physics and data co-driven deep neural network (CMT-NN).
• Incorporated physical eigenvalues and system response to ensure physics consistency.
• Validated the CMT-NN through simulations and experimental demonstrations.

Main Results:

• CMT-NN accurately predicts implicit physical parameters of complex resonant systems.
• The multi-solution problem is effectively mitigated by the CMT-NN.
• Achieved a three-order-of-magnitude reduction in computation time and a two-order-of-magnitude improvement in prediction performance compared to traditional methods.
• Demonstrated robustness through displacement sensing experiments.

Conclusions:

• CMT-NN offers a rapid, precise, and robust solution for analyzing resonant systems.
• The developed method represents a paradigm shift in applying CMT.
• Provides new insights for the design and optimization of coupled resonant systems.