Recovering Implicit Physics Model under Real World Constraints

Background

Model recovery involves learning the underlying physics-driven governing equations of a system from data. The goals of model recovery include accurately reconstructing the data, and deriving the fewest terms required to represent the underlying non-linear dynamics. Current methods of model recovery involve significant performance degradation on the data from real-world systems. There has been some research on improving the performance of model recovery on systems with limited data and noise, but there are still many issues that remain including low sampling rate, perturbed system, sparsity structure in high-dimensional non-linear function space, implicit dynamics, and input uncertainty.

Invention Description

Researchers at Arizona State University have developed a Liquid Time Constant Neural Network (LTC-NN)-based architecture designed to accurately recover the underlying models of physical dynamics from real-world data. This architecture addresses the limitations of existing methods by effectively handling low sampling rates, input-dependent time constraints, and perturbation timing errors. This approach is validated through experiments on benchmark dynamical systems and real-life medical examples, demonstrating superior accuracy in recovering implicit physics model coefficients.

Potential Applications:

• Development of digital twins (e.g., predictive maintenance, process optimization, resource management)
• Enhanced safety analysis & anomaly detection systems
• Advancements in explainable AI and prediction models
• Real-world system monitoring and diagnostics
• Medical diagnostics and treatment planning

Benefits and Advantages:

• Proven effectiveness – experiments with both simulation and real-world data
• Efficient – overcomes low sampling rate issues
• Increased accuracy – incorporates input uncertainty and perturbation timing errors
• Simplifies processing – handles implicit dynamics and sparse external human inputs through automatic differentiation and dense layer dropout

Related Publication: Recovering Implicit Physics Model under Real-World Constraints