摘要
水下焊接的应用领域广泛,但其焊接质量难以保障。针对水下焊接处理过程中存在的非线性程度高、参数耦合性强以及检测效率低等问题,提出了一种新的基于声信号识别的水下焊接质量检测方法。该方法通过在水下构建基于声信号采集的监测系统,实时采集焊件焊接过程中的声信息,并通过对声信号进行滤波降噪处理和特征提取,构建双权值神经网络(double-weight neural network, DWNN)模型。该模型具有优秀的高维数据非线性拟合能力,可实现水下焊接多参数与声信号多特征之间的非线性映射,且在小样本情况下仍能实现高精度的模式识别。以高强度低碳合金钢——HSLA-115钢作为焊接对象,开展水下焊接质量检测实验。结果表明,DWNN模型应用于水下焊接质量检测的识别精度可达100%。研究结果可为水下焊接工艺的优化和水下焊件专家知识库的构建提供参考依据。
Underwater welding is widely used in many fields,but its welding quality is difficult to guarantee.Aiming at the problems of high nonlinearity,strong parameter coupling and low detection efficiency in underwater welding process,a new method for underwater welding quality detection based on acoustic signal recognition was proposed.The method constructed a monitoring system based on acoustic signal acquisition underwater to collect the acoustic information during the welding process of weldments in real time,and built a double-weight neural network(DWNN)model through filtering and noise reduction processing and feature extraction for the acoustic signal.The model had excellent nonlinear fitting ability of high-dimensional data and could realize nonlinear mapping between multi-parameters of underwater welding and multi-features of acoustic signals,and it could still realize high-precision pattern recognition in the case of small samples.The underwater welding quality detection experiments were carried out with high strength and low carbon alloy steel—HSLA-115 steel as welding object.The results showed that the recognition accuracy of DWNN model applied to underwater welding quality detection could reach 100%.The research results can provide reference for the optimization of underwater welding process and the construction of underwater weldparts expert knowledge base.
作者
纪晓东
程天宇
华亮
张新松
JI Xiaodong;CHENG Tianyu;HUA Liang;ZHANG Xinsong(College of Electrical Engineering,Nantong University,Nantong 226000,China)
出处
《工程设计学报》
CSCD
北大核心
2023年第5期562-570,共9页
Chinese Journal of Engineering Design
基金
国家自然科学基金资助项目(51877112)。
关键词
水下焊接
声信号
特征提取
双权值神经网络
underwater welding
acoustic signal
feature extraction
double-weight neural network