超声探伤信号的人工神经网络识别

Flaw signature recognition in ultrasonic testing using artificial neural network

粗晶奥氏体不锈钢的超声探伤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来提高缺陷回波比例．本文则介绍一种用傅里叶变换作特征提取、用前馈网络自动识别奥氏体钢中缺陷信号的方法．在作者的实验中．这种方法的正确识别率达到90％．

The effectiveness of ultrasonic detection in coarse-grained austenitic stainless steel is limited by whether it can seperate useful signals and background noise effectively.Presently, people mostly incline to improve the defect echo with the technique of frequen cy diversity. This paper introduces a method of signal processing which makes character istics extraction by Fourier Transform and use feedforward networks to identify the defect signal automatically. Experiments on austenitic steel samples are presented in which the correct identification ratio reaches 90 percent.