A deep learning-aided analytical model for the whole process analysis of RC slabs under compressive arching action

Abstract

This paper proposes a deep learning (DL)-aided analytical model for the whole process analysis of laterally restrained RC slabs, based on a displacement-loading pattern. In the given model, the eccentricity of lateral restraint, slab depth variation, and connection gap are taken into account. Due to the difficulty in solving this analytical model directly, DL models of multi-layer perceptron (MLP), recurrent neural network (RNN), and long short-term memory (LSTM) networks are trained to predict the curvature at midspan, compression force, and support bending moment. According to the statistical indices, the MLP is selected for the former two and the LSTM is chosen for the third one to determine their initial values in the solution. Subsequently, the solution procedure for the analytical model with the aid of trained DL models is proposed. The numerical tests indicate that the DL models can significantly improve the convergence, stability, and efficiency of the solution procedure, especially in cases with big step lengths. Moreover, the proper number of layers and sections for the numerical solution is investigated. Finally, the proposed model is validated through several test results by comparing the load-displacement curve, the displacement-compression force curve, the strain distribution of concrete. It is demonstrated that the proposed model can effectively predict the whole process behavior of laterally restrained RC slabs with variable depth, eccentric lateral restraints, and connection gaps.