论文部分内容阅读
针对焊接过程严重非线性和焊材中多种成分的复杂交互作用使得对接头力学性能的准确估算十分困难的实际问题,论述了神经网络技术在焊接接头力学性能预测方面的应用。研究了神经网络建模方法,提出采用均匀设计法优化设计神经网络参数,在四类17种钢材的焊接热模拟数据基础上,建立了预测焊接接头力学性能的神经网络模型。试验表明该模型可根据钢材成分和焊接规范对焊接接头及其热影响区的冲击韧度、抗拉强度、屈服强度、断面收缩率和硬度等力学性能进行较为准确的估算。试验表明,该预测方法较之传统统计方法,预测精度有了大幅度提高,为实现焊接接头力学性能预测提供了一条有效的途径。
In view of the serious nonlinearity of the welding process and the complex interactions of many components in the welding consumables, it is very difficult to accurately estimate the mechanical properties of the joint. The application of neural network technology in the prediction of mechanical properties of welded joints is discussed. The neural network modeling method is studied. The uniform design method is used to optimize the design parameters of the neural network. Based on the welding thermal simulation data of the four types of 17 kinds of steels, a neural network model is established to predict the mechanical properties of the welded joints. The test shows that the model can accurately estimate the mechanical properties such as impact toughness, tensile strength, yield strength, reduction of area and hardness according to the composition of steel and the welding code. Experiments show that the prediction method has greatly improved the prediction accuracy compared with the traditional statistical methods, which provides an effective way to predict the mechanical properties of welded joints.