Welding Materials Design Based On Artificial Neural Networks

In this paper, based on the method of Artificial Neural Network, a new approach is provided to predict mechanical property of E4303. A mechanical properties prediction model for welding rod was built upon the production data. In the model, the content of Mn , C of electrode (including covering and core wire) are used as the input of network, and a mechanical property (o b) of deposited metal is used as output of network. In the research, a back propagation algorithm is used as a neural network's learning rule. Based on the result of network's estimating, the prediction of electrode property can be realized. The result shows: There are good correlations between the predicted results and the experimental data. BP-network used for prediction of electrode property possesses feasibility and validity.In the same time, a software system for welding material (WMDS) has been developed. In the system, a properties prediction module was built. Besides, the system possesses a database for welding materials. Programming by Borland C^ Builder. An interface is friendly and practicability is excellent. Designing welding materials with the system is characterized by its optimization design, high efficiency, low cost, fix quantify and etc. A new scientific way has been opened up for design of welding materials.