Real-time numerical prediction of strain localization using dictionary-based ROM-nets for sitting-acquired deep tissue injury prevention

Rohan, Pierre-Yves, Fougeron, Nolwenn, Keenan, Bethany ORCID: https://orcid.org/0000-0001-7787-2892, Pillet, Helene, Laporte, Sebastien, Osipov, Nikolay and Ryckelynck, David 2023. Real-time numerical prediction of strain localization using dictionary-based ROM-nets for sitting-acquired deep tissue injury prevention. Chinesta, Francisco, Cueto, Elias, Payan, Yohan and Ohayon, Jacques, eds. Reduced Order Models for the Biomechanics of Living Organs, Elsevier, pp. 385-402. (10.1016/B978-0-32-389967-3.00027-5)

Full text not available from this repository.

Abstract

A dictionary-based ROM-net was developed based on the full-order models of 16 subjects (9 men and 7 women) for the assessment of subject-specific in-vivo subdermal soft tissue strains and stresses in the ischial regions during sitting. Because of the limited training data available, a data augmentation scheme was proposed combining submodeling a statistical shape model and the generation of synthetic data. A dictionary was defined as a collection of all the representative subjects identified using the K-medoid clustering of the training data set. Finally, a decision-tree classifier, acting as a model selector, was trained with a database of 707 labelled finite element models. The 202 validation data and the 6 unseen test data were used to evaluate the classification performance of the trained decision-tree classifier. The performance of the dictionary-based ROM was satisfactory (accuracy of 100% on the training data; accuracy of 50% on the unseen data). The computation of the strain field by the decision-tree classifier requires 0.1 s compared to 20 m for the submodel and about 8 h for the high-fidelity model. The projection error was 9%. As is the case with machine learning techniques, the prediction accuracy of the proposed approach is expected to improve significantly as the amount of training data increases.

Item Type:	Book Section
Date Type:	Publication
Status:	Published
Schools:	Engineering
Publisher:	Elsevier
ISBN:	9780323899673
Related URLs:	Publisher
Date of First Compliant Deposit:	4 November 2022
Date of Acceptance:	25 September 2022
Last Modified:	10 Jul 2023 14:55
URI:	https://orca.cardiff.ac.uk/id/eprint/153984

Actions (repository staff only)

Edit Item