Neural network-based model for predicting particle size of AZ61 powder during high-energy mechanical milling

Many industrial and metallurgical processes require powders within a specific size range which may not be readily available from conventional processes. Thus, mechanical alloying (MA) has been exploited as an important attrition process for obtaining the desirable particle sizes of powders. However, MA is highly stochastic, depends on several parameters, and often relies on expensive and time-intensive experimentations. Therefore, developing a good model that can accurately predict the process can eliminate these challenges. This study, therefore, proposed an artificial neural network (ANN)-based mathematical equation for predicting the particle size of AZ61 magnesium alloy after MA. Three input parameters comprising rotation speed, charge ratio, and milling time were used to develop the model. Its s for training, validation, and testing datasets was greater than 93%, an indication of its high prediction ability. Furthermore, the proposed model was compared with a multilinear regression (MLR) model by means of root-mean-square error (RMSE) and mean absolute error (MAE) analyses. Results showed that the RMSE and MAE of the ANN model were considerably lower than that of the MLR model. This further established the accuracy and high predictability of the ANN model. Additionally, sensitivity analysis revealed that rotation speed was the most significant parameter influencing particle size during MA. The developed model is useful for predicting the particle size of AZ61 powder, optimizing the MA process, and eliminating expensive and time-intensive experimentations.