Scientific Machine and Deep Learning Investigations of the Local Buckling Behaviour of Hollow Sections. Issue 4 (14th September 2022)

• Record Type: Journal Article
• Authors: Müller, Andreas; Taras, Andreas; Kraus, Michael A.
• Is Part Of: Ce/papers. Volume 5:Issue 4(2022)
• Page Start: 1034
• Page End: 1042
• Publication Date: 2022-09-14
• Subjects: Machine Learning -- Deep Neural Networks -- Scientific Machine Learning -- HSS hollow‐sections -- benchmark data sets; Civil engineering -- Periodicals; 624
• DOI: 10.1002/cepa.1848
• Languages: English
• ISSNs: 2509-7075

Abstract

Abstract: The recent advent of Machine and Deep Learning (ML/DL) increased interest among applicability of data‐driven methods to design and verification in structural civil engineering concerned, specifically with members made of steel within this paper. Scientific ML (SciML) follows the idea of combining domain knowledge with the abilities of ML and DL algorithms to make the best out of both worlds. This paper investigates the SciML algorithms Random Forrests, XGBoost and a deep neural network model for the prediction of the cross‐section dependent local buckling reduction factor χ for different hot rolled and cold‐formed SHS and RHS profiles made of mild and high strength steel. The process of data collection, cleaning, compilation and model training and evaluation is described. It is found, that especially tree‐based methods can serve as feature importance and selection methods and have similarly high predictive capabilities such as deep neural networks for the case of of the cross‐section dependent local buck‐ling reduction factor χ for different hot rolled and cold‐formed SHS and RHS profiles made of mild and high strength steel. The data of this paper furthermore serve as the initial data‐set for establishing the Github repository "SciML4StructEng_Repository". This repository which will foster the dissemination of data sets for structural civil engineering to allow the prototyping, development and benchmark testing of scientific machine and deep learning (SciML) algorithms.