Machine learning algorithms for deeper understanding and better design of composite adhesive joints. (March 2023)

• Record Type: Journal Article
• Authors: Kaiser, Isaiah; Richards, Natalie; Ogasawara, Toshio; Tan, K.T.
• Is Part Of: Materials today communications. Volume 34(2023)
• Publication Date: 2023-03
• Subjects: Composite adhesive joints -- Machine learning -- Delamination -- Cohesive failure -- Finite element analysis (FEA); Materials science -- Periodicals; 620.11
• DOI: 10.1016/j.mtcomm.2023.105428
• Languages: English
• ISSNs: 2352-4928

Abstract

Abstract: In this study, machine learning (ML), a subdivision of artificial intelligence (AI), is implemented to study the mechanical behavior of composite adhesive single-lap joints (SLJs) subjected to tensile loading. The experimental data for training and testing the ML models are compiled from peer-reviewed journal papers from various research groups to eliminate bias and increase the diversity within the dataset. The dataset consists of eight continuous SLJ input parameters, which are used to predict the SLJ damage mode and failure strength. Regression and classification models are built using deep neural networks (DNN) and random forests (RF). Finite element (FE) modeling is conducted, and the numerical performance is compared with the accuracy of the regression ML models. Results show that ML models can predict strength with high accuracy. Furthermore, both DNN and RF classify damage modes accurately without the need for complex failure criteria, which cannot be typically achieved using traditional FE methods. This study utilizes ML algorithms to gain a deeper understanding of the structure-property-performance relationships of SLJs, leading to better designs of composite adhesive joints with higher strength efficiency. Graphical Abstract: ga1 Highlights: Machine learning algorithms are used to predict SLJ strength and damage mode. ML models are trained and tested using experimental data from published journals. DNN and RF models can predict correct damage mode without complex failure criteria. Higher ratio of adherend to adhesive thickness leads to higher SLJ performance. DNN and RF models make predictions in a fraction of the time compared to FE models.