Prediction system of CTOD for high strength pipeline steel welded joint based on back propagation artificial neural network

Aiming at limitation of selecting the main technical parameters for high strength pipeline steel welding in practical operation, a back propagation artificial neural network (ANN)was established with Visual C++ 6.0 for predicting the welding performance parameter-crack tip opening displacement (CTOD)-of high strength pipeline steel joint.Based on the experiment data, the average heat input, wall thick, preheat temperature and joint region were used as the input parameters of ANN, which includes one input layer with 4 nodes, one hidden layer with 14 nodes, and Sigmoid activation function.The average absolute error of prediction result is 15.4%.The number of the sample whose error is less than ±20% is about 93.3% in total 15 experimental data.The result showed that ANN method is a kind of effective method to predict the welding performance parameter CTOD of high strength pipeline steel welded joint. The ANN system can be used as selecting and optimizing the key welding parameters.