engleski

Determining nitriding parameters with neural networks

The choice of correct plasma nitriding parameters is usually experience based. There are no successful mathematical models for the nitriding process simulation. An attempt has been made to accurately determine required nitriding time for the specified effective nitriding layer thickness, sum of weight contents of nitride forming elements in steel, and nitriding temperature. Two methods were used to solve this problem: the statistical multiple regression, and the artificial neural network. It is not possible to find a regression model that would relate these three variables to nitriding time, whereas good results were achieved with neural networks. The second problem that was investigated was the determination of post-nitriding surface hardness on the basis of three known parameters: nitriding time and temperature, and the sum of weight contents of nitride forming elements in steel. Besides regression models and neural networks, genetic algorithms (GA) and genetic programming (GP) were applied. Again, a general regression model was not found, and the neural networks produced very good results. By combining of genetic algorithms and genetic programming a mathematical model was determined. This model uses the following independent variables: nitriding-alloying level, duration, and temperature to determine the dependent variable: surface hardness of nitriding process. Relative errors in determining hardness with GA-GP were much greater than the ones obtained with neural network.

nitriding parameters; regression method; neural network; genetic algorithm; genetic programming

nije evidentirano