连续油管直缝高频焊热影响区最薄弱区硬度的神经网络预测  

The Neural Network Prediction of Hardness in the Weakest Area of Coiled Tubing in HFW HAZ

作　　者：李继红[1] 李琳[1] 赵鹏康[1] 余晗[2,3] 毕宗岳[2,3] 张敏[1] 

机构地区：[1]西安理工大学材料科学与工程学院,西安710048 [2]宝鸡石油钢管有限责任公司钢管研究院,陕西宝鸡721008 [3]国家石油天然气管材工程技术研究中心,陕西宝鸡721008

出　　处：《焊管》2012年第7期5-8,共4页Welded Pipe and Tube

基　　金：陕西省教育厅自然科学基金资助项目(00k904);陕西省重点学科建设专项资金资助项目(00X901)

摘　　要：通过试验得出了连续油管HFW焊接接头最薄弱区域的力学性能,采用BP神经网络对该区域工艺性能进行仿真预测,研究了不同训练函数对网络性能的影响。对比分析不同训练函数下的网络性能,得出连续油管HFW焊接接头最薄弱区线能量-硬度预测模型,最终选取LM算法、SCG算法和动量BP算法对网络进行训练,采用这3种算法建立起的线能量-硬度模型精度较高,测试数据预测值与实测值平均相对误差分别为0.12%,0.095%和0.11%,表明神经网络模型能够很好地对"未知"硬度进行预测。The mechanical properties of the weakest area in HFW joint of coiled tubing were obtained by experiment. The simulation and prediction to process performance in the said area were conducted by adopting BP neural network, the effect on network performance of different training function was studied, and the network performance under different training function were compared and analyzed. In the end, the line energy-hardness prediction model of the weakest area in HFW joint of coiled tubing was received. The LM, SCG, and BP algorithm were selected to train the network, the precision of line energy-hardness prediction model which was built by the said three algorithm is higher, and the average relative error of predicted and measured values in test data is 0.12%, 0.095% and 0.11% respectively, which shows that the neural network networkmodel can well predict the unknown hardness.

关 键 词：连续油管 HFW BP神经网络 硬度 

分 类 号：TE973[石油与天然气工程—石油机械设备]