Surrogate model for predicting stresses of concrete slabs subjected to real vehicle loading using multilayer perceptron neural network

Abstract

Listen

Translate article

This work proposes an innovative method that uses machine learning with computational finite element simulations to optimize the design of concrete slabs for rigid pavements subjected to moving loads of different speeds. The objective is to create a surrogate model that takes into account the uncertainties of weight, shape and speed of vehicle loading on the concrete slabs, together with the uncertainties of the pavement temperatures, to predict the displacements and stresses in the concrete slabs. Realistic finite element models consider the three-dimensionality of the multilayer problem as well as the inertial effects due to moving loading at different speeds. Based on the results of finite element analyses, machine learning techniques are used to train and validate a surrogate model. This model allows the analysis and quantification of the uncertainties of displacements and stresses in concrete slabs under different conditions of vehicle speed, vehicle load and pavement temperatures. Based on these analyses, it is possible to optimize the shape and thickness of concrete slabs to cope with the effects of uncertainties, thus ensuring adequate performance of the structure under a wide range of operating conditions. This approach allows for a more precise and efficient optimization of concrete slabs, taking into account the dynamic and stochastic variables involved in the process.