基于遗传神经网络的焊缝缺陷等级磁记忆定量化研究

摘要

In order to quantify defect levels of welded joints by using the metal magnetic memory technology ( MMM) , fatigue experiments were operated to find the MMM feature law of critical damage. The experiment material is Steel Q345 that is prefabricated with incomplete penetration and slag. In the light of the X ray detection national standard and MMM testing signals, welded joints are divided into four levels:normal, stress concentration, hidden damage and macroscopic damage. BP Neural Network ( BPNN) optimized by genetic algorithm is firstly presented to quantify defect levels based on MMM parameters, which indicates that the optimized BPNN is more stable and accurate than BPNN without optimization. This research provides a new scientific tool for practical engineering.%针对金属磁记忆技术的焊缝缺陷等级定量化评定这一难题，通过对预制不同缺陷的Q345焊接试件进行疲劳试验，获得焊缝损伤演化临界状态的磁记忆信号特征规律。首次对照X射线定量检测标准和磁记忆检测结果，将焊缝损伤演化状态分为4个等级，即正常状态、应力集中、隐性损伤和宏观损伤。首次引入遗传算法优化的BP神经网络模型对焊缝等级进行磁记忆定量化评价。研究表明，遗传优化的BP网络模型与未优化的BP网络相比，预测结果更加稳定、误差更小，为工程实际中焊缝缺陷等级评定提供新的方法和依据。