AUTHOR=Aublet Axel , N’Guyen Franck , Proudhon Henry , Ryckelynck David 

TITLE=Multimodal data augmentation for digital twining assisted by artificial intelligence in mechanics of materials

JOURNAL=Frontiers in Materials

VOLUME=Volume 9 - 2022

YEAR=2022

URL=https://www.frontiersin.org/journals/materials/articles/10.3389/fmats.2022.971816

DOI=10.3389/fmats.2022.971816

ISSN=2296-8016

ABSTRACT=Digital twins in mechanics of materials usually involve multimodal data in the sense that an instance of a mechanical component has both  experimental data and simulated data. These simulated data aim not only to replicate observational data but also to extend these data, spatially, temporally or functionally for various possible uses of this component. Related multimodal data are scarce, high dimensional and a physics-based causality relation exists between observational data and simulated data. We propose a data augmentation scheme coupled to data pruning, in order to limit memory requirements for high-dimensional augmented data. This augmentation is desirable for digital twining assited by artificial intelligence by using nonlinear model reduction. Here data augmentation aims at preserving similarities in terms of validity domain of reduced digital twins. In this paper, we consider a specimen subjected to a mechanical test at high temperature, where the as-manufacturing geometry may impact the lifetime of the component. Hence, an instance is represented by a digital twin that includes a 3D X-Ray tomography of the specimen, the related  finite element mesh and the finite element predictions of thermomechanical variables that are modified at several time steps during the test. There is thus for each specimen, geometrical data and mechanical data. Due to the difficulty and complexity of annotating multimodal data, which couples different representation modalities together, collecting and annotating multimodal data requires much more effort. Thus the analysis of multimodal data generally suffers from the problem of data scarcity. This problem is particularly pronounced when considering digital twins of thermo-mechanical tests which are very complex to implement in large amounts.
The proposed data augmentation scheme aims to train a recommending system that recognizes a category of data available in a training set that has already been fully analyzed by using high fidelity models. Such a recommanding system enables the use of a ROM-net for fast lifetime assessment via local reduced order models.