Design Of Hybrid Expert System And The Application In Material Design

Material design aims to develop new material with certain properties based on accumulated experience and scientific theory. The main objects are to fix on the relation among composition, process parameters and performance of materials. As far as the development of currently material science, there are not enough inherent rules on the material, but a vast amount of experimental data about composition, process parameters and their performances. Theoretical aided material design and prediction is widely used under such conditions. By means of computer technology, material design can break away from experiment and make use of fewer experiments to obtain ideal materials. With the development of expert system and artificial neural network, effective method is provided with computer aided material design.The study field of artificial intelligence includes artificial neural network (ANN) and expert system (ES), The ANN model is determined by topology structure, activation function and learning method. The model possesses with good fault tolerance, self-adaptability and non-linear mapping, all of this have not been possessed in expert system. Reversely, the clarity, agility, knowledge definituded of expert system have not in ANN, so combining neural network (NN) and expert system (ES), building hybrid expert system can deal with a mass of data.Combining neural network (NN) and expert system (ES), hybrid expert system is built in this paper, the system extract each virtue of NN and ES, discard their shortage, and establish knowledge acquisition module of NN. During the training of network, this research utilizes changes step and appended momentum methods, adds dummy stylebook and avoids over fitting in order to accelerate learning speed and avoid acute vibration. Based on the hybrid expert system, inference mechanism of NN is built, which is applied in the material design. Material properties such as hardness, yield strength, elongation and martensite start temperature are predicted by means of hybrid expert system. It is advantageous to predict material design with the hybrid expert system.