【摘要】 对于非均匀材料，超声无损检测技术受到能否有效区分有用信号与背景噪声的限制，目前人们大多倾向使用频率分隔与统计算法来提高粗晶材料（一种非均匀材料件相对颗粒散射的缺陷回波比例．文中介绍一种用Ｗｉｇｎｅｒ分布作特征提取、用前馈网络自动识别超声散射回波中的缺陷信号．由于普通的人工神经网络要求输入信号的特征与时间起始点无关，因此采取了一种数学变换方式来实现这一要求，这样训练好的网络就有很强的识别能力．在实验中，正确识别率达到９０％．所述方法对其他非均匀介质的超声检测与评价工作也有益处．更多还原

【Abstract】 For inhomogeneous materials, conventional ultrasonic nondestructive testing techniques are limited in their ability to differentiate the signal of interest from the background noise. The current trend in improving ultrasonic flaw-to-grain echo ratio in large-grain materials (inhomogneous materials) is to use frequency diversity and statistical approach. This paper describes the use of artificial neural network utilizing the joint time-frequency distribution-Wingner distribution for feature extraction and feedforward network for classification of flaw signal in ultrasonic backscattered grain echoes. For normal artificial neural network, it is required that the input signal be independent of any possible shifting of the Origin in time. So we make a mathematical transformation such that the transformed Wigner distribution is time-origin independent and hence may serve as input to the neural network. Then the trained network has significant classincation ability. The ratio of correct classification in our experiment reached 90%. We believe that ultrasonic t6sting and evaluation for other inhomogeneous media may also benefit from the present method.更多还原