Mahdi Nasiri Abarbekouh

Research Problem and Question:
This research project focuses on physics-informed machine learning (PIML), where the idea is to model physical systems by leveraging both the available data and known or partially known physical laws. Combining domain knowledge with data can improve the performance of machine learning (ML) models by enabling them to utilize data for making precise predictions while producing physically interpretable solutions. However, PIML models encounter several challenges from both theoretical aspects, such as issues of convergence and stability, and practical implementation perspectives, such as optimization strategies.

This research aims to address challenges that limit the applicability of physics-informed machine learning in complex industrial and scientific settings, particularly in distributed control systems.

Method and Benefit:
Physics-informed machine learning has been shown to overcome the limitations faced in first-principles and data-driven ML methods, by embedding differential equations into the learning process. PIML models leverage both data and prior physical knowledge to make accurate and physically interpretable predictions, even in cases where the data may be noisy, sparse, or involve high-dimensional spaces. These models have shown improvements in interpretability, extrapolation capabilities, and efficiency in computation while offering additional advantages such as dimensionality reduction and superior generalization capabilities.