论文部分内容阅读
研究电触头钎焊接头超声无损检测中的缺陷分类问题，提出了一种新的集成神经网络分类方法。该方法分四步：频率不变性预处理，多分辩分析，特征量预处理，集成 Ｂ Ｐ神经网络分类。使用不同中心频率探头检测得到的缺陷信号首先通过预处理变换到一个等效的参考频率上，然后利用离散小波变换提取特征量。特征量被预处理后，输入到集成 Ｂ Ｐ神经网络分类器中分类。本文用２１３ 个超声检测信号测试了集成神经网络的性能。实验结果表明了频率不变性技术和集成 Ｂ Ｐ神经网络分类技术的有效性。
In this paper, the defect classification of ultrasonic contactless contactless brazed joints is studied, and a new integrated neural network classification method is proposed. The method consists of four steps: frequency invariant preprocessing, multi-resolution analysis, feature preprocessing, integrated Bp neural network classification. Defective signals detected with different center frequency probes are first transformed to an equivalent reference frequency by preprocessing and then extracted using discrete wavelet transform. After the feature is preprocessed, it is input to the classifier in the integrated BP neural network classifier. In this paper, we use 213 ultrasonic testing signals to test the performance of integrated neural network. Experimental results show the effectiveness of frequency invariance techniques and integrated Bp neural network classification techniques.