基于PCA的超声缺陷类别识别方法研究

A Novel Flaw Identification Approach Based on PCA

考虑超声回波信号不同参数对某一具体缺陷类别的不同重要性,在超声检测缺陷信号特征提取的基础上,提出采用主分量分析法对其分析。分析结论取代传统的直接把缺陷特征向量分类器作为输入的方法,避免冗余特征对缺陷类别识别的影响。将锻件中三类常见缺陷的特征向量分别进行主分量分析,并通过利用BP神经网络的泛化性能分别构建了分类器。利用DS证据理论对神经网络的输出进行融合识别。通过现场采集的超声探伤信号,对提出的分析方法进行验证,缺陷类别的识别率有较大提高,说明了该方法的有效性。

Each parameter of the flaw signal in ultrasonic testing has a different influence on every special kind of flaw,thus the PCA approach is adopted to analyze it on the bases of the feature of the flaw is successfully extracted.The analyzing results of PCA,instead of the eigenvector from the ultrasonic signal,are input to the flaw classifier to eliminates the disadvantages of superfluous characteristics in flaw classification.The features of the there most common flaw in forging pieces are analyzed with PCA,and the BP neural network are constituted as classifier for its parallel computing capability for those flaw respectively.The classifiers count the probability that testing flaw belongs separately.The outputs of BPs are fused with DS theory to get the classification.Using the ultrasonic signals with known defects,the feasibility and the validity of the proposed approach is proved.