摘要
针对焊接过程中因保护气体和外部气流干扰等情况所产生的异常焊接状态,提出了一种基于电弧声音信号特征结合长短期记忆(LSTM)神经网络的监测方法,基于梅尔频谱(Mel⁃spectrogram)下的电弧声音特征对焊接过程进行分析。采用高速脉冲全位置熔化极气体保护焊(GMAW)下行焊焊接技术,搭建了短弧高速脉冲X80管线钢GMAW实验平台,记录了全工况条件下的电弧声数据。根据电弧声音信号特点,采用daubechies小波作为小波基函数,利用启发式阈值对电弧声信号进行分解和重构,有效提高了电弧声音信号的信噪比。通过正常和异常焊接状态下声音信号的梅尔频谱特征训练LSTM神经网络,建立了电弧声音信号的预测模型来检测异常焊接状态,该方法的灵敏度可以达到94.57%。
In allusion to the abnormal welding states due to shielding gas,external airflow interference and other situations,a monitoring method on the basis of the arc acoustic signal features and in combination with long and short⁃term memory(LSTM)neural network is proposed,with which the welding process is analyzed according to the arc acoustic features in the Mel⁃spectrogram.The downward welding technology of high⁃speed pulsed all⁃position gas metal arc welding(GMAW)is adopted.An experimental platform for steel GMAW of short⁃arc high⁃speed pulsed X80 pipeline is built.The arc acoustic data under full working conditions is recorded.According to the features of the arc acoustic signals,the Daubechies wavelet is adopted as the wavelet basis function,and the heuristic threshold is utilized to decompose and reconstruct the arc acoustic signals,which improve the signal⁃to⁃noise ratio(SNR)of the arc acoustic signals effectively.The Mel⁃spectrum features of the sound signals in normal and abnormal welding states are adopted to train the LSTM neural network,and a prediction model of the arc acoustic signals is established to detect abnormal welding states.The sensitivity of the proposed method can reach 94.57%.
作者
倪畅
薛瑞雷
刘宏胜
周建平
李晓娟
NI Chang;XUR Ruilei;LIU Hongsheng;ZHOU Jianping;LI Xiaojuan(School of Mechanical Engineering,Xinjiang University,Urumqi 830000,China)
出处
《现代电子技术》
2022年第23期115-120,共6页
Modern Electronics Technique
基金
大型油气输送管道全位置自动焊接技术研究(2017D01C038)。
