中文摘要：

超声检测中，需要根据不同的情况，如试样厚度、分辨率、缺陷深度及方向等而使用不同中心频率的探头。这成为信号自动分类中的一个主要问题。因为大多数模式分类算法与信号的形状密切相关．而信号的形状很大程度上随检测频率的变化而变化。为使分类系统不受检测频率的影响而能识别不同频率的同一类缺陷信号．文中采用了基于时间尺度化的频率一致性分析方法把不同频率的信号映射到同一参考频率。采集了两个焊接缺陷样本库也即尺度化处理样本库和没有尺度化处理样本库，并用类别可分性判据做定量对比。实验结果验证了频率一致性处理方法的有效性，能够消除换能器频率对分类的影响。

英文摘要：

In ultrasonic inspection, transducers with different center frequencies are used depending on a number of factors such as thickness of the sample, resolution desired, and depth and orientation of the flaw. This poses a major problem when the data is analyzed using automated signal classification systems. Most pattern classification algorithms rely heavily on the shape of the signal, which can vary considerably with frequency. In order to recognize the same class of signals with different frequencies and classify them in such a manner that the system performance is independent of frequency, a frequency invariance technique using time scaling is described in this paper to transform different signal frequencies to a reference frequency. Two different databases containing weld inspection flaw signals were acquired, namely, the scaled database and the unscaled database, compared in quantity by sort separability criterion. Experimental results demonstrate effectiveness of the time scaling for invariance to the effect of transducer frequency.