engleski

Improvement of the mathematical model of steel carburizing using neural network and genetic algorithm

In the paper, using of neural network and genetic algorithm for calculating the laws of the complex processes (among which are diffusion processes at steel carburizing) is presented. For determining of technological parameters of carburizing which are necessary for obtaining the required flow of carbon curve in the carburized layer, simulation of mathematical model for Carbomaag carburizing process is presented. For training of the neural network, the results of the empirical carburizing model that was proven in practice were used and compared to the results of computer simulation of the mathematical model. Comparing the experimental data and simulation data, it was proven that neural network shows good generalization properties for estimating of time and carbon potential required for carburizing. Based on results of the neural network (NN), using genetic algorithm (GA), the experimental equation, which is a part of the mathematical model, showing influence of alloying elements to the flow of carburizing curve was improved. Introducing of the improved equation into the existing mathematical model enables achieving of the empirically required 0.6-0.8%C in the surface layer, at the required effective carburizing depth (Edp) about 0.35%C, which enables achieving of 550 HV1 hardness after quenching.

carburizing; neural networks; genetic algorithm

nije evidentirano