基于EMD和GA-SVM的超聲檢測缺陷信號識別
中國測試李大中， 趙 杰
摘 要：為提高金屬探傷時對缺陷的識別能力，提出一種遺傳優化支持向量機，結合經驗模態分解（EMD），對超聲波缺陷信號進行自動識別。首先進行經驗模態分解法分解，提取出原始信號特征，構建特征向量。鑒于常用的神經網絡模型識別率不高及支持向量機參數難確定的問題，利用遺傳算法優化支持向量機模型（GA-SVM）的懲罰因子和核參數，提高支持向量機建模精度。分別采用神經網絡模型、SVM模型和GA-SVM模型對特征向量進行訓練與測試，GA-SVM模型識別率達到98.437 5%，優于神經網絡方法和未改進的交叉驗證法SVM模型。試驗結果表明：遺傳算法能有效提高支持向量機的性能，在小樣本條件下能夠提高超聲缺陷的識別率。
關鍵詞：缺陷信號識別；遺傳算法；支持向量機；經驗模態分解
文獻標志碼：A 文章編號：1674-5124（2016）01-0102-05
Flaw signal identification in ultrasonic testing based on EMD and GA-SVM
LI Dazhong， ZHAO Jie
（Dept of Automation，North China Electric Power University，Baoding 071003，China）
Abstract: In order to improve the flaw-recognizing ability in crack detection， a genetic algorithm optimization support vector machine （GA-SVM） has been proposed to identify automatically the ultrasonic defect signals in combination with the empirical model decomposition （EMD）. First， the EMD is applied to extract the features of original ultrasonic signals and create feature vectors. Considering that common neural network models are low in recognition rate the SVM parameters are difficult to determine， the penalty factor and kernel parameter of the GA-SVM were employed to enhance the modeling precision of the GA-SVM. The feature vectors are trained and tested with the neural network model， SVM model and GA-SVM model. The recognition rate of the GA-SVM model is up to 98.437 5%， higher than the neural network model and the unimproved cross validation SVM model. Experimental results show that genetic algorithm can improve SVM performance. This machine can increase the recognition rate of ultrasonic defects in small samples.
Keywords: flaw signal recognition； genetic algorithm； SVM； EMD
