ANN-Based Modeling of Thermal & Multi-Phase Systems

How do you model a system whose behavior is too complex, nonlinear, and interlocked for equations to remain practical?

Thermal and multi-phase systems represent the boundary between what conventional modeling techniques can handle and what they cannot. Strong nonlinearities, coupling between phases, and dependence on the point of operation make it difficult to solve such problems analytically.

My work in ANN-based modeling aims to address the challenge of using artificial neural networks as high-accuracy engineering models rather than black-box models. This is achieved by extending physical models in cases where first-principle models become computationally expensive and analytically infeasible. The learning models are constrained and validated to preserve physical consistency.

The subject of this work is thermal diffusion and multi-phase phenomena in complex systems. Experimental results and simulations of these systems have been used as training examples for artificial neural networks able to describe complex relationships. The models thus developed enable fast and precise calculations. Results demonstrate accurate prediction across operating regimes not explicitly seen during training.

The relevance of the work is based on its applications. ANN models help in the fast assessment of system responses and hence can be applied in design optimization, real-time analysis, and control. They consume lower computational costs when compared to simulations.

One of the major contributions of this research is the focus on generalization and validation under uncertainty. The design of the models and the training processes are focused on ensuring that the results are reliable under uncertainty, because this is a major consideration in any engineering application where validation of predictions is as vital as accuracy. This directly addresses a core limitation of conventional data-driven models in engineering.

The areas of application of this technology are thermal management, energy systems, heat exchangers, and material analysis. With the trend of increased integration of systems and shrinking margins of performance, the capability of fast and accurate modeling of thermal performance assumes a key role.

This research shows that learning-based approaches can be responsibly integrated into engineering practice, not as a replacement for physics, but as a validated extension of it. The demonstrated success of this approach lies in achieving both speed and trustworthiness simultaneously.