THE USE OF A GENETIC ALGORITHM FOR OPTIMIZING THE PARAMETERS OF THE NEURAL NETWORK IN PREDICTING THE STRESS-STRAIN STATE OF A SQUARED PLATE

Abstract

In modern production, computer-aided design systems have become widespread, which allow to design technological processes with less time and money, with increased accuracy of designed processes and processing programs, which reduces material costs and processing time, due to the fact that processing modes are also calculated and optimized using a computer. The development of machine learning methods and models allows to make quick estimates of the necessary parameters of the state of the object. From a practical point of view, machine learning models for predicting the values of structural parameters can serve as interactive assistants in the design process. One of the main issues in the application of neural networks is their structural optimization, the choice of the optimal number of layers, neurons, activation functions and so on. Such optimization can be performed both manually, provided a relatively small number of parameters, and automatically. This paper considers the peculiarities of using a genetic algorithm to optimize the parameters of the neural network to predict the stress-strain state of a square plate. The test sample, which contains the possible states of the plate depending on the geometric and mechanical parameters, was constructed using analytical formulas and the finite element method. An essential stage of genetic algorithms is the definition of genetic operators: selection, crossover, mutation and selection. The choice of these operators affects the convergence and efficiency of the method as a whole. The genetic algorithm is an example of metaheuristic methods. The convergence of such methods is quite difficult to prove formally. However, the use of a genetic algorithm when setting up neural networks can minimize user intervention. A genetic algorithm is used to optimize the parameters of the neural network for predicting the stress-strain state of a square plate. Learning models based on artificial neural networks are built. The constructed models allow predicting the deflection in the center of the plate, as well as the maximum value of the stress intensity according to Mises. The main advantage of an artificial neural network is the speed of prediction. The calculation of the required characteristics in comparison with the finite element method is almost instantaneous (milliseconds). Thus, «trained» artificial neural networks can serve as interactive assistants in the design process.