A Review on Data-Driven and Deep Learning Methods in Computational Solid Mechanics

Artificial intelligence (AI) has transformed computational modeling across diverse industrial sectors. Among AI approaches, Deep Neural Networks (DNNs), Physics-Informed Neural Networks (PINNs), Graph Neural Networks (GNNs), and generative models demonstrate distinct strengths in solving complex engineering and scientific problems. This review systematically examines these methodologies, comparing their architectures, accuracy, computational efficiency, and applicability in real-world scenarios. Key findings reveal that while DNNs excel in general pattern recognition, PINNs effectively incorporate physical constraints, GNNs efficiently handle structured relational data, and generative models enable novel design and prediction tasks. The study further identifies current limitations, including scalability, interpretability, and integration challenges, and discusses strategies to overcome them. By providing a comprehensive evaluation, this work offers critical insights for researchers and practitioners seeking to select or develop AI methods tailored to industrial applications. The review also outlines promising future directions, emphasizing hybrid models and explainable AI to enhance performance and adoption.

کلیدواژه ها:

Deep Neural Networks ، Physics-Informed Neural Networks ، Graph Neural Networks ، Generative Models ، Industrial Applications