Impact response prediction and optimization of SC walls using machine learning algorithms

Abstract

The dynamic impact response of steel-plate composite (SC) walls is a vital index to assess their safety. By optimizing the design parameters within the requirements of relevant specifications, the impact-induced deformation can be reduced to ensure the protection of personnel and equipment inside the structures. To process the complicated nonlinear relationship between the wall deformation and design parameters rapidly and accurately, prediction models were constructed through such machine learning algorithms as support vector regression (SVR), artificial neural network (ANN), and gaussian process regression (GPR) in this study. Then these models were further validated by the existing test results. It shows that the root mean square error (RMSE), R-Squared (R2), and the time to complete one calculation of the GPR model are 2.74, 0.9591, and 0.58 ms respectively, which has obvious advantages in terms of accuracy, stability, and efficiency among the three machine learning models. Based on the well-trained GPR model, the design of a certain SC wall in an AP1000 nuclear power plant was optimized via the genetic algorithm, with the maximum deformation and the cost of SC walls as the optimization targets. The case study shows that the optimal solution set can reduce the maximum deformation by 10.1 % while keeping the total cost constant, or reduce the total cost by 18.8 % while keeping the maximum deformation constant.