基于多尺度散布熵的磁声发射信号特征识别方法

Magneto acoustic emission signal feature recognition method based on multi-scale dispersion entropy

在工程中对设备进行应力检测和微损伤检测时,采集磁声发射信号易受噪声干扰,同时其特征的提取也存在困难,为此,将变分模态分解与散布熵相结合,提出了一种基于自适应多尺度散布熵的磁声发射(MAE)信号特征识别方法。首先,设计搭建了检测实验平台,采集了Q345钢静载拉伸实验中0 MPa~400 MPa应力状态下的MAE信号;然后,采用变分模态分解方法,对磁声发射信号进行了自适应分解,生成了一系列从低频到高频分布的本征模态函数(IMF)分量;其次,计算了每个本征模态函数分量的散布熵值,构建了MAE信号的特征向量矩阵;最后,将特征向量矩阵输入到基于支持向量机建立的识别分类模型中,进行了信号的训练和识别。研究结果表明:使用基于自适应多尺度散布熵的磁声发射(MAE)信号特征识别方法,能够自适应地实现MAE信号的多尺度化目的,并且准确地识别出不同应力状态下的信号特征,分类识别准确率高达95.3704%,验证了该方法的有效性;说明基于自适应多尺度散布熵和多分类支持向量机的信号特征识别方法能够快速且有效地识别不同应力状态,在信号特征识别方面具有较好的应用潜力。

Aiming to address the challenges associated with stress and minor damage detection in engineering equipment,specifically the issues of susceptibility to noise interference and difficulties in feature extraction,a novel and academically enriched method was proposed.The principles of variational mode decomposition were combined with dispersion entropy to develop an adaptive multi-scale approach for recognizing magnetic acoustic emission(MAE)signal features.Firstly,a specialized experimental platform was designed and constructed to facilitate the capture of MAE signals under stress conditions ranging form 0 MPa to 400 MPa in a static tensile test conducted on Q345 steel.Then,the MAE signals were adaptively decomposed using variational mode decomposition,yielding a series of intrinsic mode function(IMF)components that span the frequency spectrum from low to high.Next,the dispersion entropy values of each IMF component were computed to construct a feature vector matrix representing the MAE signals.Finally,this feature vector matrix was utilized as input data for a recognition classification model based on support vector machines,enabling the model to undergo training and subsequent recognition processes.The experimental results show that the method of magneto acoustic emission signal feature recognition is able to achieve multi-scale adaptation of MAE signals and accurately identify signal features under different stress states.The classification accuracy rate is as high as 95.3704%,which validates the effectiveness of the method.The research outcomes affirm that the signal feature recognition method,based on adaptive multiscale dispersion entropy and multiclass support vector machines,exhibits promising potential for rapid and effective identification of different stress states,thereby offering significant applications in the field of signal feature recognition.