Neural network algorithm for accuracy control in modelling of structures with changing characteristics

Abstract

The paper is devoted to the creation of neural network algorithm for control of the accuracy of durability calculation in problems of modelling the behavior of structures with changing characteristics, in particular, corroding trusses. In contrast to known methods, the approach suggested in this paper takes into account the change of forces in elements of the truss over time. First, the analysis of corrosive wear models is given to choose the most suitable one for further research. Then, the mathematical statement of durability determination problem is given. The paper describes the approach to the approximation of the numerical solution error for the durability problem. To build an approximating function, artificial neural networks are used. The architecture of these networks and the procedure of their training are described further in the paper. At the end, the results of numerical experiments, which prove the correctness of the chosen approach, are given. The developed algorithm can be especially effective in solving optimization problems with constraints on the durability of a structure. The same approach can also be generalized to other classes of structures.