Modern mechanical and physical systems, such as robots, vehicles, industrial machines, fluids, and materials, are becoming increasingly complex and data-intensive. Despite major advances in sensing, computation, and automation, accurately predicting and controlling their behavior remains a significant challenge. Traditional computational models, based on physical laws and numerical methods, can provide interpretable and accurate results but are computationally expensive and limited by idealized assumptions. In contrast, AI models learn directly from data and can capture highly nonlinear behaviors efficiently, yet they often ignore the governing physical principles, leading to unreliable or nonphysical predictions.

This contrast between physics-based and data-driven approaches highlights a fundamental gap: the absence of a unified framework that integrates physical knowledge with data-driven intelligence. To address this challenge, the project proposes a Scientific Machine Learning (SciML) framework that combines the strengths of both approaches. By embedding physical constraints within AI architectures, the proposed framework will produce models that are physically consistent, data-efficient, and computationally scalable. The central innovation question guiding this research is how such hybrid models can accurately and efficiently capture, predict, and control the behavior of complex mechanical and physical systems while maintaining physical interpretability and generalization.

The project’s expected outcomes include generalizable SciML models capable of fast and accurate prediction of dynamic behavior across diverse mechanical and physical systems. These models will lead to tangible benefits such as improved performance, safety, and energy efficiency in intelligent machines and engineered systems operating under uncertainty. In the long term, this research will contribute to the global transition toward sustainable, intelligent technologies while supporting Vietnam’s Industry 4.0 agenda. It will also strengthen VinUni’s research capacity by training students and early-career researchers in physics-guided artificial intelligence, fostering both scientific excellence and societal impact.