Fourier Neural Operators for Fast Multi-Physics Sensor Response Prediction: Applications in Thermal, Acoustic, and...

Area of Science:

• Computational physics and engineering
• Artificial intelligence and machine learning
• Sensor technology and instrumentation

Background:

• Traditional numerical methods (FEM, CFD) for sensor simulation are computationally expensive, limiting real-time applications.
• Accurate and rapid sensor response prediction is crucial for digital twins, real-time systems, and design optimization.

Purpose of the Study:

• To develop and validate a novel framework using Fourier Neural Operators (FNO) as efficient surrogate models for multi-physics sensor response prediction.
• To demonstrate the computational efficiency and accuracy of FNO across thermal, acoustic, and flow measurement domains.

Main Methods:

• Utilized Fourier Neural Operators (FNO), which learn mappings between infinite-dimensional function spaces for resolution-invariant predictions.
• Developed a hybrid H-FNO architecture combining spectral operators with local convolutional layers to address spectral bias.
• Trained FNO models on simulation datasets and validated against holdout data and experimental results.

Main Results:

• Achieved significant speedups: 8300× for thermal (R2>0.98), 4000× for acoustic (MAE<0.5 dB), and 31,000× for flow (>97% accuracy).
• Reduced inference time from minutes to milliseconds, demonstrating remarkable computational efficiency.
• Validated the framework's efficacy through multiple case studies and experimental data for thermal sensors.

Conclusions:

• FNO-based surrogate models offer a powerful and efficient solution for accelerating sensor simulation across diverse physics domains.
• The proposed framework enables real-time sensor calibration, uncertainty quantification, and design optimization, supporting Industry 4.0.
• Established FNO as a key tool for AI-enhanced instrumentation and measurement, overcoming limitations of traditional numerical methods.