基于FOA-SVM的超声信号端点检测  

Ultrasonic signal endpoint detection based on FOA optimized SVM

作　　者：李大中[1] 赵杰[1] 

机构地区：[1]华北电力大学自动化系,河北保定071003

出　　处：《中国测试》2016年第5期103-106,123,共5页China Measurement & Test

摘　　要：在超声缺陷识别系统中,端点检测是确保缺陷准确识别的重要环节。为提高在实际探伤过程中端点检测的准确率,提出一种以果蝇算法优化支持向量机的端点检测方法。针对超声检测信号的特点,采用小波包变换提取反映该信号性质的特征向量。鉴于传统方法检出率不高及支持向量机(SVM)参数难确定的问题,利用果蝇算法(FOA)优化SVM的惩罚子和核参数,提高支持向量机建模准确度。试验结果表明:FOA-SVM模型的平均检出率达到97.5%,端点检测效果明显优于传统的双门限法、普通SVM模型和GA-SVM模型。Endpoint detection is an important step to ensure accurate identification in the ultrasonic defect recognition system. In order to improve the accuracy of endpoint detection during actual flaw detection,an endpoint detection method using fruit fly optimization Algorithm（FOA）-support vector machine（SVM） has been proposed. Based on the characteristics of ultrasonic detection signals, wavelet transform was applied to extract the feature vector that reflects the nature of these signals. As the common double-threshold method is low in detection rate and the parameters of the SVM are difficult to determine, the FOA was used to optimize the penalty factor and kernel parameter of the SVM to improve the precision of the SVM. The experimental results show that the average detection rate of FOA-SVM is 97.5%. The endpoint detection effect significantly outperforms that of the traditional double threshold method and common SVM and GA-SVM models.

关 键 词：端点检测 果蝇算法 支持向量机 小波变换 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]