Using image processing techniques in computational mechanics. (15th April 2023)

• Record Type: Journal Article
• Authors: Trent, Stephen; Renno, Jamil; Sassi, Sadok; Mohamed, M. Shadi
• Is Part Of: Computers & mathematics with applications. Volume 136(2023)
• Page Start: 1
• Page End: 24
• Publication Date: 2023-04-15
• Subjects: Conditional Generative Adversarial Network -- Applied mechanics -- Image processing -- Computational mechanics -- Finite element method -- Real-time predictions; Electronic data processing -- Periodicals; Mathematics -- Data processing -- Periodicals; 510.28541
• DOI: 10.1016/j.camwa.2022.11.024
• Languages: English
• ISSNs: 0898-1221

Abstract

Abstract: The implementation methods of finite element analysis (FEA) have remained essentially unchanged since the inception of FEA in the 1960s. Alterations of any of the input or design parameters to the FEA model can potentially nullify the previous results and subsequent additional simulations will be required. This is particularly relevant for situations that require active monitoring where telemetry is to be passed to remote systems capable of carrying out FEA computations. In this paper, we train an artificial neural network that was originally developed for image processing to emulate FEA. Conventionally generated FEA results are transformed into image pairs where the load, material and geometric properties are assigned different colour channels. These images are used to train a conditional Generative Adversarial Network (cGAN). The subsequent "trained" cGAN can generate predictions for arbitrary inputs which correspond to the domain of input on which the developed cGAN was trained. Three numerical experiments were conducted resulting in three separate cGANs trained to infer (a) deflections from forces, (b) stresses from deflections and (c) stresses from forces. After a moderate training regime of 200 epochs each, the outputs of the trained networks are shown to be in reasonable agreement to the ground truth with mean errors in the range of 5-10%. The contribution of this work lies in transforming FEA results into images which enables the usage of cGANs to solve a computational mechanics problem. The implementation herein allows for near real-time computations which highlights the potential of the proposed methodology in applications where simulation results are required in a timely manner such as predictive control, interactive virtual environment, etc. All the codes used in this research will be openly available at Qatar University's Institutional Repository 1 ; the data used in this work will be available upon request from the corresponding author.