Learning to reverse thermal diffusion

Area of Science:

• Thermodynamics and Heat Transfer
• Computational Physics
• Materials Science

Background:

• Thermal diffusion is irreversible, obscuring time-dependent information due to the second law of thermodynamics.
• Traditional methods struggle with reconstructing past thermal states from final-state data.
• Accurate spatiotemporal analysis is crucial for fields like electronics and energy systems.

Purpose of the Study:

• To develop a physics-informed framework for time-reversible thermal diffusion analysis.
• To introduce a novel time-reversal operator learning approach for thermal retrodiction.
• To establish a high-fidelity paradigm for spatiotemporal analysis in complex systems.

Main Methods:

• A finite-difference-based network was used to derive heterogeneous material properties.
• A novel operator learning approach was developed for thermal retrodiction, mapping function spaces.
• Synergy of analytical eigenbasis decomposition and frequency-domain operator learning reconstructed temperature fields.

Main Results:

• The time-reversal operator effectively reconstructed backward temperature field propagation.
• Validated on 3D-printed structures and chips, the method achieved retrodiction errors below 0.1%.
• Established a high-fidelity paradigm for spatiotemporal analysis.

Conclusions:

• The operator-driven time-reversal method overcomes thermal diffusion irreversibility.
• This approach offers high-fidelity spatiotemporal analysis with broad applications.
• Potential applications include non-destructive testing in energy systems and diffusion phenomena analysis.