论文部分内容阅读
提出了一种基于神经网络的缺陷表征方法.该方法采用Fischer线性判别分析对表征缺陷的时域信号的波形参数进行选择,并将这些参数作为神经网络的输入对智能缺陷表征系统进行训练,用概率神经网络和BP神经网络分别对缺陷的类型和大小进行识别.对135种人造焊接缺陷(裂纹、夹杂和气孔)的试验结果表明,文中方法对辨识缺陷表征信息和提高缺陷识别率非常有效.
This paper presents a method of neural network-based defect characterization, which uses Fischer linear discriminant analysis to select the waveform parameters of the time-domain signal that characterizes defects, and uses these parameters as the input of neural network to train the intelligent defect representation system. Probabilistic neural network and BP neural network were used to identify the type and size of defects, respectively. The test results of 135 kinds of artificial welding defects (cracks, inclusions and pores) show that the proposed method is very effective in identifying defects and improving the recognition rate of defects.