Physics-informed machine learning for matched field source-range estimationa).

Area of Science:

• Ocean acoustics
• Machine learning
• Signal processing

Background:

• Ocean acoustic source localization is crucial for underwater surveillance and research.
• Traditional matched field processing (MFP) methods often require extensive environmental data and are sensitive to mismatches.
• Purely data-driven machine learning (ML) approaches may lack physical consistency.

Purpose of the Study:

• To develop a physics-informed machine learning (ML) framework for ocean acoustic source localization.
• To integrate physics-informed neural networks (PINNs) into the matched field processing (MFP) scheme.
• To enable accurate source-receiver range estimation with sparse measurements and reduced environmental characterization.

Main Methods:

• A physics-informed neural network (PINN) was employed to predict acoustic pressure fields from sparse measurements and a known sound speed profile (SSP).
• The PINN-predicted replica fields were integrated into the MFP algorithm.
• The framework was validated using experimental data from the Shallow Water Evaluation Cell Experiment 1996 (SWellEx-96).

Main Results:

• The proposed method achieved accurate source-receiver range estimation, even in challenging scenarios like the closest point of approach.
• The framework demonstrated robustness against sparse-array configurations and moderate sound speed profile (SSP) mismatches.
• Performance was maintained for array element depths excluded during training, showing good interpolation/extrapolation capabilities.

Conclusions:

• Physics-informed ML offers a powerful approach for ocean acoustic localization in realistic, data-limited environments.
• This method overcomes limitations of conventional model-based MFP by reducing environmental dependency and mitigating mismatch effects.
• The integration of physics into ML models yields physically consistent predictions, enhancing localization accuracy and generalizability.