AI in Materials Mechanics

Abstract

AI and ML are increasingly transforming materials mechanics by enabling accelerated material design, property prediction, constitutive modelling, and insight into complex material behaviors. We combine theoretical foundations with extensive practical exercises, focusing on civil engineering materials (e.g., concrete, alloys, composites, wood) and the modeling of their mechanical behavior.

Objective

By the end of this course, you will demonstrate proficiency in several core competencies: You will be able to evaluate the role of AI in transforming materials mechanics for design, property prediction, and data-driven constitutive modeling while mastering the mathematical foundations of supervised learning, including regression techniques and multi-layer perceptrons. You will learn to differentiate among network architectures to analyze path- and time-dependent material behaviors and to select appropriate models for specific mechanical problems. Advanced model selection, validation, and classification strategies will further refine your analytical toolkit. You will gain technical and methodological proficiency by implementing data-driven models using industry-standard tools like scikit-learn, TensorFlow, XGBoost and PyTorch within Jupyter Notebook environments. You will execute end-to-end computational workflows and develop Physics-Informed Neural Networks (PINNs) that integrate physical, kinematical, and thermodynamic constraints directly into the model training process. Applying these techniques to construction materials (concrete, composites, metals, wood), you will utilize neural networks as approximators for complex constitutive models. This includes predicting non-linear phenomena such as hyper-elasticity, plasticity, and creep. Furthermore, you will model path-dependent responses using recurrent networks and design AI-driven systems for structural health monitoring. In your final real-world micro-project, you will solve a specific materials mechanics problem using AI. Proficiency will be demonstrated through a comprehensive Jupyter Notebook (including documentation and code) and an in-class presentation. This culminates in a practical portfolio that merges machine learning expertise with materials mechanics knowledge.

Content

The weekly 2-hour sessions will typically be divided between interactive lectures and lab-style hands-on tutorials in which you apply concepts to real-world data. We segment the course into the following blocks: ML Foundations: • Introduction to ML in material mechanics and design (1 week) • Recap on Regression (polynomial, radial basis, logistic, ridge regression, regression trees, random forest, cross-validation…) (2 weeks, supervised learning) • Working with Multi-layer Perceptron (MLP) (mathematical foundation, backpropagation, implementation, data preparation, network types like FF, CNN, RNN) (2 weeks, supervised learning) ML Applications: • Regression Models for Material Properties (data-driven predictive models for materials) (1 week) • Neural Networks for Material Modeling: NNs as approximators for complex material behavior. (data-driven constitutive modeling.) (1 week) • Modeling path-dependent material responses with Recurrent NNs. (1 week) • Physics-Informed NN (PINNs) adding kinematical, thermodynamic or symmetry constraints to the training. (1 week) ML Case studies: • Hyper-elastic model identification, plasticity modeling, Creep prediction, and health monitoring (1 week) ML Student project presentations Micro-project: Students select or propose a project topic by mid-semester. Ideally, a real-world materials mechanics problem using AI techniques from the course – for instance, building a predictive model for a material property, or structural health monitoring with malfunction detection, model identification, microstructure optimization, or others…. The deliverables include a Jupyter Notebook (documentation and code) and an in-class presentation (Last semester's week).

Resources