Using neural networks in solving the problem of calculating the corroding structures durability

Abstract

The algorithm for calculating the durability of beam rod elements, subjected to corrosive wear, is proposed. It is assumed that the type of constraint, that determines the limit state of the element, is unknown beforehand. Previously the type of active constraint was defined in the process of problem-solving, which significantly complicated the computational algorithm and reduced its effectiveness. The paper proposes to determine the type of active constraint before the problem-solving, using the information on geometric features of cross-section, load intensity and corrosive medium parameters. This implies formalizing a priori knowledge in the form of an artificial neural network with discrete function of the output element activation. The corroding I-beam is considered in the paper as the object of research. During the simulation of corrosion process it is taken into account that mechanical stresses cause its substantial acceleration. In this case, the limit state of the beam element can be defined under the constraints of strength and continuity of the cross-section. The neural network with a threshold activation function was used for determining the type of active constraints. Training the neural network was carried out using a genetic algorithm.