利用模糊神经网络技术预测焊接接头力学性能
Prediction of mechanical properties of welded joint using fuzzy neural network technology
-
摘要: 探讨一种自适应模糊神经网络(adaptive fuzzy neural networks,ANFIS)预测焊接接头力学性能的方法。通过测定TC4钛合金TIG焊接头的抗拉强度、抗弯强度和延伸率,结合焊接工艺参数建立了用于焊接接头力学性能预测的自适应模糊神经网络模型。利用该模型,并使用BP算法和BP算法与最小二乘相结合的混合算法,采用不同的输入变量隶属度函数、模糊子集数、迭代次数,对焊接接头力学性能进行了ANFIS仿真。结果表明,当采用混合算法,且模糊子集数为3时,网络训练和预测结果平均误差均远小于7%,能够满足实际生产的要求。使用MATLAB和Visual C++混合编程开发了基于ANFIS焊接接头力学性能预测软件,可根据焊接工艺参数对焊接接头的抗拉强度、抗弯强度和延伸率等力学性能进行较为准确地预测,为焊接接头的质量预测与控制提供了一条简捷、有效的新途径。Abstract: The method of predicting mechanical properties of welded joint based on adaptive fuzzy neural networks (ANFIS)was discussed.The tensile strength, bend strength and extensibility of TC4 titanium alloy welded joint using TIG welding were tested.ANFIS model to predict mechanical properties of welded joint was established.By BP algorithm and hybrid algorithm, the mechanical properties of welded joint were simulated using different membership functions, fuzzy subsets and training epochs.The results show that by hybrid algorithm, the average error of ANFIS training and prediction is less than 7% while the fuzzy subset is 3, which can meet the requirement of practical production.According to welding processing parameters, the mechanical properties of welded joint including tensile strength, bend strength and extensibility can be predicted accurately, which provides an effective approach to predict and control the quality of welded joint.
下载: