Application of neural network and genetic algorithm to powder metallurgy of pure iron

In the present paper, soft computing techniques are applied to optimize the powder metallurgy processing of pure iron. An artificial neural network is trained to predict the stress resulting from a given trend in strain and sintering temperature. To prepare an appropriate model, pure iron powders are compacted and sintered at various temperatures. Subsequently, compression test is conducted at room temperature on the bulked samples. The sintering temperatures and the corresponding stress–strain records are used as sets of data for the training process. The performance of the network is verified by putting aside one set of data and testing the network against it. Eventually, by using a genetic algorithm, an optimization tool is created to predict the optimum sintering temperature for a desired stress–strain behavior. Comparison of the predicted and experimental data confirms the accuracy of the model.

► The testing correlation factor is 0.99, indicating that the neural network has been well trained.
► A precise match is obtained between the optimized sintering temperatures and the experiments.
► The optimum mechanical properties for iron powders can be achieved by sintering above 800 °C.