Recorded as part of a hybrid seminar co-hosted by the International Centre for Advanced Materials (ICAM) and the Henry Royce Institute, this presentation by Professor John Lambros explores how microscale strain accumulation influences damage and failure in metallic materials.
Professor Lambros is Donald B. Willett Professor of Engineering in the Department of Aerospace Engineering at the University of Illinois Urbana-Champaign and a Distinguished International Associate of the Royal Academy of Engineering.
Professor Lambros examines how the complex microstructure of polycrystalline metals creates localised strain variations during deformation and how these variations can contribute to material degradation over time. He discusses experimental approaches for measuring strain at the microscale using high-resolution digital image correlation (HR-DIC), with examples drawn from traditionally manufactured and additively manufactured metals.
Professor Lambros also highlights how machine learning and crystal plasticity finite element simulations can be used to predict microstructural strain behaviour. By combining advanced experimental methods with data-driven modelling, the research provides insights into deformation, fatigue, creep, and failure mechanisms in engineering materials.
The session was moderated by Professor Phil Withers, Regius Professor of Materials at the Henry Royce Institute and Professor of Materials at The University of Manchester.
This recording will be of interest to researchers, students and professionals working in materials science, materials engineering, mechanics, manufacturing, structural integrity, fatigue, fracture mechanics, additive manufacturing and computational modelling.
Note: Captions are auto-generated and may contain errors.
Under deformation the heterogeneous microstructure of polycrystalline metals generates locally complex strain variations at the microscale. The details of the strain variations may depend upon microstructural characteristics (e.g., grain size and morphology, grain boundary types, grain orientation, etc.) and loading type (e.g., tensile/compressive/shear, creep, fatigue, fracture etc.).
These local strain variations often ultimately control failure mechanisms since plastic strain tends to accumulate persistently once formed. To measure in detail such local strain variations we employ a high-resolution digital image correlation technique (HR-DIC) in which rastered digital images are stitched together to provide an ultra-high-resolution image for use in DIC codes. The method is applied to metallic materials, both traditionally and additively manufactured, under conditions of plastic cycling, elastic fatigue, and elastic or plastic creep. For modelling this response, we use machine learning (ML) methods to develop a capability for predicting microstructural-level strains. To obtain much larger datasets than HR-DIC can provide, we use numerical datasets generated by crystal plasticity finite element simulations (CPFEMs) to train a fully convolutional network (FCN). Two-dimensional patterns of plastic strain field variations (output) are predicted from grain orientation information (input) at the microscale across a large subset of grain morphologies. Previously applied FCN architectures have predicted the general patterns of plastic strain distributions, but with accuracies that saturate quickly with increasing size of the training dataset. We overcome this limitation by augmenting the traditional convolution architecture with modern architectural elements such as skip connections, depthwise separable convolutions, residual functions, and inverted bottleneck convolution modules, reducing the number of trainable parameters and floating-point operations by 88% and 77%, respectively. Our FCN architecture, trained on predominantly equiaxed grains with a fixed (lognormal) distribution of grain sizes under a small subset of macroscopic strain states, is capable of interpolation and limited extrapolation to other strain states. Its ability to predict the microscale strain patterns across a wide range of grain sizes, grain distributions, and grain shapes without retraining, further suggests its generalizability to different grain architectures. Finally, we discuss the utility of transfer learning to reduce the amount of training data required to adapt the FCN to materials with different stress–strain response.
Professor Lambros is the Donald B. Willett Professor of Engineering in the Aerospace Engineering Department of the University of Illinois Urbana-Champaign. He received a BEng degree in Aeronautics from the Imperial College of Science and Technology, London, in 1988. He then spent several years at the Graduate Aeronautical Laboratories at Caltech, Pasadena CA, obtaining an MS and a PhD in Aeronautics in 1989 and 1994, respectively. In 1995 he joined the Mechanical Engineering department of the University of Delaware as an Assistant Professor and subsequently the Aerospace Engineering department of the University of Illinois at Urbana-Champaign in 2000 as an Associate Professor, where he became a full professor in 2007. He has served as Director of the Grainger Engineering College’s Advanced Materials Testing and Evaluation Laboratory (AMTEL) during 2018-2022 and as Associate Head for Graduate Studies in Aerospace Engineering during 2011-2016. He has also served as Vice-President, President Elect, and President of the Society for Experimental Mechanics between 2017-2020. He is a Fellow of the American Society of Mechanical Engineers (ASME), the Society for Experimental Mechanics (SEM) and the American Academy of Mechanics (AAM), and an Associate Fellow of the American Institute of Aeronautics and Astronautics (AIAA). He has served as Associate Technical Editor for the ASME “Journal of Applied Mechanics”, the SEM journal “Experimental Mechanics”, and currently serves as Editor-in-Chief of “Engineering Fracture Mechanics”.
