Efficient Exploration of Microstructure-Property Spaces via Active Learning

Abstract

In materials design, supervised learning plays an important role for optimization and inverse modeling of microstructure-property relations. To successfully apply supervised learning models, it is essential to train them on suitable data. Here, suitable means that the data covers the microstructure and property space sufficiently and, especially for optimization and inverse modeling, that the property space is explored broadly. For virtual materials design, typically data is generated by numerical simulations, which implies that data pairs can be sampled on demand at arbitrary locations in microstructure space. However, exploring the space of properties remains challenging. To tackle this problem, interactive learning techniques known as active learning can be applied. The present work is the first that investigates the applicability of the active learning strategy query-by-committee for an efficient property space exploration. Furthermore, an extension to active learning strategies is described, which prevents from exploring regions with properties out of scope (i.e., properties that are physically not meaningful or not reachable by manufacturing processes).

Accurate distances measures and machine learning of the texture-property relation for crystallographic textures represented by one-point statistics

“…The data that support the findings of this study are openly available at the following URL/DOI: https://fordatis.fraunhofer.de/handle/fordatis/319 [64].…”

“…In addition, the simulation framework is used to calculate material properties. The properties we focus on are the orientation dependent Young's moduli and anisotropy similar to R-values, as in [61]. R-values, however, are originally developed to express plastic anisotropy in sheet metals.…”

Neural Networks for Constitutive Modeling: From Universal Function Approximators to Advanced Models and the Integration of Physics

“…Besides constitutive modeling, the potential of using machine learning models is actively being investigated across multiple fields of computational material science. Among others, these methods are used in computational design of materials [5][6][7][8][9], in design of processes [10][11][12], in development of digital twins [13] and soft sensors [14], and in multi-scale simulations and homogenization schemes. Numerical prediction in these applications often involve frequent execution of simulations with variations of model parameters [15].…”

“…To tackle this issue, intelligent sampling algorithms, e.g. from the field of active learning [200], can be used for data generation and model training, see [201,9] and [82] for application examples. • Measuring prediction quality: Another effect of the large space of possible loading conditions is that learned constitutive models cannot be fitted accurately for any imaginable loading path.…”

A generalised deep learning-based surrogate model for homogenisation utilising material property encoding and physics-based bounds

“…In this direction, physics-informed neural networks (PINNs) 5 – 7 have gained much attention for accurately solving the PDEs of the underlying physics. A knowledge-based sampling of inputs is another way of utilising the physics of the problem in model training 8 , 9 . In addition to prior knowledge infusion, the type of ANN architecture plays an essential role in effective and effortless learning.…”