Design Optimization of an Innovative Instrumental Single-Sided Formwork Supporting System for Retaining Walls Using Physics-Constrained Generative Adversarial Network

Abstract

Single-sided formwork supporting systems (SFSSs) play a crucial role in the urban construction of retaining walls using cast-in-place concrete. By supporting the formwork from one side, an SFSS can minimize its spatial footprint, enabling its closer placement to boundary lines without compromising structural integrity. However, existing SFSS designs struggle to achieve a balance between mechanical performance and lightweight construction. To address these limitations, an innovative instrumented SFSS was proposed. It is composed of a panel structure made of a panel, vertical braces, and cross braces and a supporting structure comprising an L-shaped frame, steel tubes, and anchor bolts. These components are conducive to modular manufacturing, lightweight installation, and convenient connections. To facilitate the optimal design of this instrumented SFSS, a physics-constrained generative adversarial network (PC-GAN) approach was proposed. This approach incorporates three objective functions: minimizing material usage, adhering to deformation criteria, and ensuring structural safety. An example application is presented to demonstrate the superiority of the instrumented SFSS and validate the proposed PC-GAN approach. The instrumented SFSS enables individual components to be easily and rapidly prefabricated, assembled, and disassembled, requiring only two workers for installation or removal without the need for additional hoisting equipment. The optimized instrumented SFSS, designed using the PC-GAN approach, achieves comparable deformation performance (from 2.49 mm to 2.48 mm in maxima) and slightly improved component stress levels (from 97 MPa to 115 MPa in maxima) while reducing the total weight by 20.85%, through optimizing panel thickness, the dimensions and spacings of vertical and lateral braces, and the spacings of steel tubes. This optimized design of the instrumented SFSS using PC-GAN shows better performance than the current scheme, combining significant weight reduction with enhanced mechanical efficiency.