孔洞缺陷的红外无损检测和PNN识别与定量评估

针对红外无损检测中因特征信息缺失,致使识别与评估效果不佳这一问题,研究以铝板为对象,基于红外无损检测技术,结合主成分分析和概率神经网络对铝板正常区及三类孔洞缺陷区进行了识别与面积定量评估.研究首先采集铝板降温过程的红外时序热图,提取了正常区和各类孔洞缺陷区的时序灰度值作为初始特征.其次,采用主成分分析对初始特征进行提取,并结合概率神经网络,以像素点为单位实现孔洞缺陷的识别及面积定量评估,并采用了支持向量机进行了对比研究.实验结果表明,对于正常区和三类孔洞缺陷区测试样本的面积评估正确率分别为99.6%、97.0%、94.7%和93.0%,相比支持向量机的评估结果,所提出的研究方法具有更高的正确率.研究论证了采用主成分分析和概率神经网络,基于时序特征,以像素点为单位,实现孔洞缺陷识别和面积定量分析的有效性和准确性.

Infrared nondestructive testing of cavity defects and PNN recognition and quantitative evaluation

According to the less accessibility characteristics for the detection of defects will result in detection ineffective and quantitative inaccurate. The study focused on the subject of aluminum plate,based on infrared nondestructive testing technology, combined with principal component analysis and probabilistic neural network(PNN)on the normal area and three kinds of cavity defects area for the recognition and area of quantitative evaluation. Firstly, research during the cooling process of heating aluminum plate,the initial characteristics were obtained from the sequence grey value of normal and three kinds of cavity defects area on the basis of sequence infrared image. And the principal component analysis was used to extract initial characteristics. Finally, combined with the probabilistic neural network, the cavity defects were identified and quantitatively evaluated in pixels. And the support vector machine was used to carry on the comparative study. Experimental results show that the evaluation accuracy rates of the normal and the three kinds of cavity defects area were 99.6%, 97.0%, 94.7% and 93.0% respectively, compared with the evaluation results of support vector machine, the proposed research method has higher accuracy. Research demonstrates that using principal component analysis and PNN, based on the temporal characteristics, to achieve the effectiveness and accuracy of the cavity defects identification and quantitative analysis of the area in units of pixels.