基于特征评估和概率神经网络的超声焊缝缺陷识别

为了可靠地检出并识别焊缝缺陷,提出了一种基于特征评估和概率神经网络(PNN)的超声自动识别方法.该方法分别采用小波包和经验模式分解法对缺陷信号进行分解,提取原始信号和各分解信号的时域无量纲参数组成联合特征,并计算其评估因子,根据评估因子的大小选取敏感特征作为PNN的输入,从而实现不同焊缝缺陷类型的自动识别.通过对飞机起落架焊缝进行机上原位检测,实验结果表明,上述方法能够从大量的缺陷特征中筛选出敏感特征,克服了人为选择缺陷敏感特征的盲目性,减小了PNN规模,提高了分类准确率和检测效率.该方法在飞机的外场原位测试中具有很好的应用前景.

Welding Line Flaws Identification in Ultrasonic Testing Based on Feature Evaluation and PNN

In order to check and identify welding line flaws reliably,an automatical ultrasonic flaws identification method based on feature evaluation and PNN was proposed,where flaws signals were decomposed by WPT and EMD respectively,and the dimensionless parameters in time domain were extracted from the original signals and each decomposed signal to construct the combined features.Further more,a feature evaluation method was applied to calculate evaluation factors of the combined features,and the corresponding sensitive features were selected according to the evaluation factors and input into PNN to automatically identify different welding line flaws.According to in-situ test of welding line in aircraft undercarriage,the result shows that the method mentioned above has the power of selecting sensitive ones from a large number of features,accordingly it could overcome the blindness of selecting sensitive features subjectively,reduce the networks scale and increase the classify accuracy and testing efficiency,which brings an outstanding foreground in the in-situ test of aircraft undercarriage outfield.