• CN:11-2187/TH
  • ISSN:0577-6686

机械工程学报 ›› 2016, Vol. 52 ›› Issue (16): 1-7.doi: 10.3901/JME.2016.16.001

• 仪器科学与技术 •    下一篇

基于声发射信号的铝合金点焊裂纹神经网络监测*

张勇1,2, 周昀芸1,2, 王博1,2, 谢红霞1,2, 叶武1,2, 滕辉1,2   

  1. 1. 西北工业大学凝固技术国家重点实验室 西安 710072;
    2. 西北工业大学陕西省摩擦焊接重点实验室 西安 710072
  • 出版日期:2016-08-20 发布日期:2016-08-20
  • 作者简介:张勇,男,1965年出生,博士,副教授。主要从事电阻焊理论与质量控制方面的教学和科研工作,发表论文40余篇。E-mail:hjzhyong@nwpu.edu.cn
  • 基金资助:
    * 国家自然科学基金(51275418); 陕西省重点科技创新团队(2014KCT-12)和陕西省科技统筹创新工程计划(2012HBSZS021)资助项目; 20151123收到初稿,20160613收到修改稿;

Neural Network Monitoring of Aluminum Alloy Spot Welding Crack Based on Acoustic Emission Signal

ZHANG Yong1,2, ZHOU Yunyun1,2, WANG Bo1,2, XIE Hongxia1,2, YE Wu1,2, TENG Hui1,2   

  1. 1. State Key Laboratory of Solidification Technology, Northwestern Polytechnical University, Xi’an 710072;
    2. Shaanxi Key Laboratory of Friction Welding Technologies,Northwestern Polytechnical University, Xi’an 710072
  • Online:2016-08-20 Published:2016-08-20

摘要:

铝合金热加工过程的冶金行为比较复杂,在电阻点焊快速加热和冷却条件下,极易产生裂纹缺陷。基于虚拟仪器技术,以LabVIEW为软件平台,结合Matlab数值分析软件,构建了电阻点焊过程声发射信号采集分析及铝合金点焊裂纹监测系统。以2A12铝合金电阻点焊熔核冷却结晶过程,即点焊焊接循环维持阶段的声发射信号为研究对象,提取与声发射信号强度相关的振铃计数、能量、有效电压及5层小波分解125~250 kHz频带能量系数4个特征参数作为输入矢量,裂纹作为输出矢量,建立3层BP神经网络铝合金点焊裂纹的监测模型,并利用测试样本对该模型进行验证。结果表明,裂纹监测的正确率达到89.1%,为监测铝合金电阻点焊裂纹提供了一种有效的方法。

关键词: 裂纹, 神经网络, 声发射, 电阻点焊

Abstract:

The metallurgical behavior of aluminum alloy is quite complex in hot working process, especially in resistance spot welding (RSW) with the condition of rapid heating and cooling, thus the crack is one of the major defects of the aluminum alloy RSW. Based on the virtual instrument technology, a system functioning the acquisition and analysis of the acoustic emission signal and the monitoring on the crack of aluminum alloy in RSW process is established on the software LabVIEW platform combining with Matlab. The acoustic emission signal in the cooling crystallization process of aluminum alloy 2A12 RSW, i.e., the hold time of the spot welding cycle, is chose, 4 characteristic parameters associating with the acoustic emission signal intensity, including the ring count, energy, the effective voltage and the energy coefficient from 5-layer wavelet decomposition of 125-250 kHz band, are the input vector, and the crack is the output vector. A three-layer BP neural network monitoring model for the crack of aluminum RSW is established and verified. The verification results shows that the correct rate of the monitoring system reaches 89.1%, thus this study provides an effective method which can monitors the crack of aluminum alloy RSW.

Key words: acoustic emission, crack, neural network, resistance spot welding