基于BP人工神经网络的高强度管线钢焊接接头性能参数CTOD预测系统  

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

作　　者：童莉葛[1] 白世武[2] 刘方明[2] 

机构地区：[1]北京科技大学机械工程学院,北京100083 [2]中国石油天然气管道科学研究院,河北廊坊065000

出　　处：《焊接学报》2007年第8期96-98,共3页Transactions of The China Welding Institution

摘　　要：针对实际中高强度管线钢焊接工艺参数的选择主要依据试验和经验的局限性,使用VC++6.0建立了预测高强度管线钢焊接接头性能参数裂纹尖端张开位移(CTOD)的BP神经网络模型。该模型输入层节点数为4,1个隐层,节点数为14,激活函数为Sigmoid型。根据试验数据提取平均热输入、壁厚、预热温度和接头区域作为预测模型的输入量,预测结果的平均绝对误差为0.154,预测值误差在±20%以内的样本数占总样本数的93.3%。结果表明,人工神经网络方法是预测管线钢焊接接头性能参数CTOD的一种有效途径,可为管线钢焊接过程中主要工艺参数的选择和优化提供有效的手段。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.

关 键 词：神经网络 焊接 裂纹尖端张开位移 预测 性能参数 

分 类 号：TG456[金属学及工艺—焊接]