基于BP神经网络的细晶Ti₂AlNb基合金粉末 球磨工艺研究

An artificial-neural-network (ANN) model which is used for the prediction of properties of the as-milled powder was developed for the analysis and prediction of correlations between processing (high-energy planetary ball milling) parameters and the morphological characteristics of Ti₂AlNb-based alloy powder by applying the back-propagation (BP) neural network technique. In the BP model, the input parameters of the neural network model were milling speed, milling time and ball-to-powder weight ratio. The output of the model was the properties of the as-milled powder (specifically crystallite size). The number of node in the hidden layer was 9. Input and output functions were tansig and purelin, respectively. The accuracy of the established artificial neural network model was tested by the test data sample. It is shown that the predicted values coincide well with the test results owing to the advantages in fault-tolerance and commonality. Not only can the trained neural network model be used to predict the crystallite size of the as-milled Ti₂AlNb-based alloy powder, but also can make up for deficiency of all kinds of physical model for ball milling process in application and expression, which has application value and far-reaching significance for the research work of the actual powder metallurgy process.