A method to the design and prediction for materials and production process by using deep learning was described.The operating principle and characteristics of convolutional neural network

CNN for short were introduced briefly.Modeling method of a CNN network available to microstructural prediction was expounded

and applicable activation function was determined.Meanwhile

a loss function was established.Taking the cooling process of rolled steel plate as an example

the simulated temperature fields obtained by FEM were used as samples to train the established CNN.The trained CNN can give reliable predictions for the cooling plate.The results show that the fewer input and more output parameters can also be treated with CNN.A relatively small amount of samples was sufficient to the training of the CNN

which reached a better prediction.It is deduced that new alloys and processing parameters can be intelligible designed by using some basic CNN models trained for basic processing units.