论文部分内容阅读
声发射信号普遍存在于管道、锅炉泄漏事故中,是否能在设备出现裂纹及微小泄漏时有效检测到相关信号是防止事故发生的关键。支持向量机(SVM)有比神经网络更强的泛化能力,其找到的极值解就是全局最优解,同时它还较好地解决了小样本学习分类问题。针对声发射信号的非线性非平稳特征和现实中难以获得大量故障样本的实际情况,提出了一种基于经验模态分解(EMD)、自回归(AR)模型和SVM相结合的泄漏声发射信号分类方法。首先将采集到的信号进行EMD分解,得到若干个固有模态函数(IMF)之和,然后分别对每一阶IMF分量进行AR建模,提取每一个AR模型的参数组成特征向量,最后利用SVM对3种模拟声发射信号进行分类识别。实验表明,该方法比BP神经网络的识别率更高。
Acoustic emission signal is ubiquitous in pipelines and boiler leaks, and it is the key to prevent accidental detection of relevant signals when there are cracks and tiny leaks in equipment. Support Vector Machine (SVM) has more generalization ability than neural network. The extreme solution it finds is the global optimal solution, and it solves the small sample learning classification problem well. Aiming at the nonlinear non-stationary characteristics of acoustic emission signals and the fact that it is difficult to obtain a large number of fault samples in reality, a leakage acoustic emission signal based on empirical mode decomposition (EMD), autoregressive (AR) model and SVM Classification. First, the collected signals are decomposed by EMD to obtain the sum of several intrinsic mode functions (IMFs), and then the AR components of each IMF component are modeled separately, the parameters of each AR model are extracted to form the eigenvectors, and finally the SVM Three kinds of analog acoustic emission signal classification and identification. Experiments show that this method has higher recognition rate than BP neural network.