CN110702042A - A thickness estimation method under the condition of signal aliasing in ultrasonic pulse echo thickness measurement - Google Patents

Abstract

Translated fromChinese

本发明公开了一种超声脉冲回波测厚中信号混叠条件下的厚度估算方法，包括，选取目标厚度估算区间不同厚度的标准试块作为标准厚度参考点；采用超声脉冲回波法多次采集不同厚度的标准试块的脉冲回波时域信号，加窗后进行快速傅里叶变换生成频谱图；将各标准试块的频谱图截取一定频率范围后作为训练样本集输入卷积神经网络中进行训练，形成识别模型；将待测工件的回波的频谱图输入识别模型中得到待测工件与各标准试块的频谱相似度，作为估计待测工件厚度的权重。本发明以超声脉冲回波在不同厚度下的频谱图主瓣与各次谐波旁瓣的细微频移变化为基础，采用卷积神经网络对频谱图进行识别，能够实现在回波时域信号混叠情况下进行待测工件厚度的估算。

The invention discloses a thickness estimation method under the condition of signal aliasing in ultrasonic pulse echo thickness measurement. Collect the pulse echo time-domain signals of standard test blocks of different thicknesses, and perform fast Fourier transform after windowing to generate a spectrogram; the spectrogram of each standard test block is intercepted in a certain frequency range as a training sample set and input to the convolutional neural network In order to form a recognition model, the spectrogram of the echo of the workpiece to be tested is input into the recognition model to obtain the spectral similarity between the workpiece to be tested and each standard test block, which is used as the weight for estimating the thickness of the workpiece to be tested. The invention is based on the subtle frequency shift changes of the main lobe of the ultrasonic pulse echo and the side lobes of each harmonic under different thicknesses, and uses the convolutional neural network to identify the spectrogram, which can realize the echo time domain signal. Estimate the thickness of the workpiece to be measured in the case of aliasing.

Description

Translated fromChinese

一种超声脉冲回波测厚中信号混叠条件下的厚度估算方法A thickness estimation method under the condition of signal aliasing in ultrasonic pulse echo thickness measurement

技术领域technical field

本发明公开了一种超声脉冲回波测厚中信号混叠条件下的厚度估算方法，属于超声无损检测技术领域。The invention discloses a thickness estimation method under the condition of signal aliasing in ultrasonic pulse echo thickness measurement, belonging to the technical field of ultrasonic nondestructive testing.

背景技术Background technique

在机械制造、航空航天、汽车制造等大型工业中，生产的工件需要在恶劣的环境下长期使用，损等残缺都会对生产安全造成严重的威胁，因此对工件的加工剩余厚度的要求越发严苛，超声无损测厚具有很广阔的应用场合。In large-scale industries such as machinery manufacturing, aerospace, and automobile manufacturing, the workpieces produced need to be used for a long time in harsh environments, and defects such as damage will pose a serious threat to production safety. Therefore, the requirements for the remaining thickness of workpieces are becoming more and more stringent. , Ultrasonic nondestructive thickness measurement has a wide range of applications.

传统超声测厚应用中，主要是对声波的TOF(Time of Flight)进行测定，常用的方法有阈值法、峰值法、包络法、曲线拟合法和互相关法等。其中，互相关法已能够对TOF进行较为准确的测定。但在薄件测厚中，常因探头的精度不佳以及回波经多次反射后信噪比大幅下降造成各次回波时域信号已发生严重混叠，传统的TOF测定方法将会产生较大的误差。In traditional ultrasonic thickness measurement applications, the TOF (Time of Flight) of acoustic waves is mainly measured. Commonly used methods include threshold method, peak value method, envelope method, curve fitting method and cross-correlation method. Among them, the cross-correlation method has been able to measure TOF more accurately. However, in the thickness measurement of thin parts, the time-domain signals of each echo have been seriously aliased due to the poor accuracy of the probe and the sharp drop in the signal-to-noise ratio of the echo after multiple reflections. big error.

发明内容SUMMARY OF THE INVENTION

本发明的目的在于提供一种超声脉冲回波测厚中信号混叠条件下的厚度估算方法，在超声脉冲回波在薄件测厚或在低信噪比条件下回波时域信号已产生混叠的情况下，依据脉冲回波信号在不同厚度下频谱图主瓣与各次谐波旁瓣细微的频移变化为基础采用卷积神经网络对频谱图进行识别，通过待测工件与各样块的相似度实现厚度的估算。The purpose of the present invention is to provide a thickness estimation method under the condition of signal aliasing in ultrasonic pulse echo thickness measurement, when the ultrasonic pulse echo is used for thickness measurement of thin parts or the echo time domain signal has been generated under the condition of low signal-to-noise ratio. In the case of aliasing, the convolutional neural network is used to identify the spectrogram based on the subtle frequency shift changes of the main lobe of the spectrogram and the side lobes of each harmonic under different thicknesses of the pulse echo signal. The similarity of the sample blocks enables estimation of thickness.

为实现上述目的，本发明的技术方案如下：For achieving the above object, technical scheme of the present invention is as follows:

本发明实施例提供一种超声脉冲回波测厚中信号混叠条件下的厚度估算方法，包括：The embodiment of the present invention provides a thickness estimation method under the condition of signal aliasing in ultrasonic pulse echo thickness measurement, including:

选取目标厚度估算区间不同厚度的标准试块作为标准厚度参考点；Select the standard test blocks with different thicknesses in the target thickness estimation range as the standard thickness reference point;

采用超声脉冲回波法多次采集每个不同厚度的标准试块的脉冲回波时域信号；The ultrasonic pulse echo method is used to collect the pulse echo time domain signals of each standard test block with different thicknesses for many times;

对获得的标准试块的所有脉冲回波时域信号加窗后进行快速傅里叶变换生成标准试块的频谱图；After windowing all the pulse echo time-domain signals of the obtained standard test block, fast Fourier transform is performed to generate the spectrogram of the standard test block;

将各标准试块的频谱图截取一定频率范围后作为训练样本集输入卷积神经网络中进行训练，输出识别模型；The spectrogram of each standard test block is cut into a certain frequency range and input into the convolutional neural network as a training sample set for training, and the recognition model is output;

将待测工件的超声脉冲回波的频谱图输入识别模型中得到待测工件与各标准试块的频谱相似度，以频谱相似度为权重估计待测工件的厚度。Input the spectrogram of the ultrasonic pulse echo of the workpiece to be tested into the recognition model to obtain the spectral similarity between the workpiece to be tested and each standard test block, and use the spectral similarity as the weight to estimate the thickness of the workpiece to be tested.

进一步的，所述标准试块是涵盖待测工件厚度范围内的不同厚度的试块。Further, the standard test block is a test block with different thicknesses covering the thickness range of the workpiece to be measured.

进一步的，所述超声脉冲回波法采用单探头自发自收模式，并且采用相同的回波采样率。Further, the ultrasonic pulse echo method adopts a single-probe spontaneous and self-receiving mode, and adopts the same echo sampling rate.

进一步的，所述将各标准试块的频谱图截取一定频率范围，包括：Further, the spectrogram of each standard test block is intercepted to a certain frequency range, including:

截取的上限频率为：The upper limit frequency of interception is:

截取的下限频率为：The lower limit frequency to be intercepted is:

其中，f_H为上限频率，f_m为探头的中心频率，f_w为探头的带宽。Among them, f_H is the upper limit frequency, f_m is the center frequency of the probe, and f_w is the bandwidth of the probe.

进一步的，所述训练样本集输入卷积神经网络之前，将频谱图更改为固定像素大小的图像。Further, before the training sample set is input to the convolutional neural network, the spectrogram is changed to an image with a fixed pixel size.

进一步的，所述卷积神经网络卷积层的卷积核大小为5×5。Further, the size of the convolution kernel of the convolutional layer of the convolutional neural network is 5×5.

进一步的，卷积神经网络训练中，池化层生成的特征图经全连接层处理输出识别模型，用以识别待测工件的超声脉冲回波频谱图，并得出待测工件与标准工件的频谱图相似度。Further, in the training of the convolutional neural network, the feature map generated by the pooling layer is processed by the fully connected layer to output the recognition model, which is used to identify the ultrasonic pulse echo spectrogram of the workpiece to be tested, and obtain the difference between the workpiece to be tested and the standard workpiece. Spectrogram similarity.

进一步的，所述待测工件的超声脉冲回波的频谱图的获取，包括：Further, the acquisition of the spectrogram of the ultrasonic pulse echo of the workpiece to be tested includes:

采用超声脉冲回波法采集待测工件的超声脉冲回波时域信号，The ultrasonic pulse echo method is used to collect the ultrasonic pulse echo time domain signal of the workpiece to be tested.

对所获取的超声脉冲回波时域信号加窗后进行快速傅里叶变换生成频谱图；After windowing the acquired ultrasonic pulse echo time domain signal, fast Fourier transform is performed to generate a spectrogram;

进一步的，所述将待测工件的超声脉冲回波的频谱图输入识别模型中之前，将所述频谱图截取一定频率范围后修改为固定像素大小的图像。Further, before the spectrogram of the ultrasonic pulse echo of the workpiece to be tested is input into the recognition model, the spectrogram is cut out in a certain frequency range and then modified into an image with a fixed pixel size.

进一步的，所述待测工件的厚度估计如下：Further, the thickness of the workpiece to be measured is estimated as follows:

d＝αW^TDd=αW^T D

其中，d为待测工件的厚度估计值，α为一维行修正系数矩阵，由标准试块及实际工件厚度测试标定而来，W为以频谱相似度权重为元素的对角矩阵，D为标准样块厚度值构成的一维列矩阵。Among them, d is the estimated thickness of the workpiece to be tested, α is a one-dimensional row correction coefficient matrix, which is calibrated from the standard test block and the actual workpiece thickness test, W is a diagonal matrix with spectral similarity weights as elements, D is A one-dimensional column matrix of standard block thickness values.

本发明在超声脉冲回波法薄件测厚中针对回波信号信噪比较低或回波时域信号已产生混叠的情况下，采用卷积神经网络对频谱图进行识别以实现薄件厚度的估算，改善了传统识别方法在上述情况下精度不佳或无法测量的问题。The present invention adopts the convolutional neural network to identify the spectrogram to realize the thin part when the signal-to-noise ratio of the echo signal is low or the echo time domain signal has been aliased in the thickness measurement of the thin part by the ultrasonic pulse echo method. Thickness estimation improves the problem of poor accuracy or inability to measure traditional identification methods in the above-mentioned situations.

附图说明Description of drawings

图1是产生混叠的时域信号示意图；Fig. 1 is the time domain signal schematic diagram that produces aliasing;

图2是频谱图偏移示意图；Fig. 2 is a schematic diagram of spectrogram shift;

图3是本发明实施例的厚度估算方法流程图。FIG. 3 is a flowchart of a thickness estimation method according to an embodiment of the present invention.

具体实施方式Detailed ways

下面对本发明作进一步描述。以下实施例仅用于更加清楚地说明本发明的技术方案，而不能以此来限制本发明的保护范围。The present invention is further described below. The following examples are only used to illustrate the technical solutions of the present invention more clearly, and cannot be used to limit the protection scope of the present invention.

如图1所示，超声脉冲回波在薄件测厚过程中或在低信噪比条件下，回波时域信号常常已经产生混叠，在时域上对回波波形进行处理判定已经十分困难且精确度不高。As shown in Figure 1, the ultrasonic pulse echo in the process of measuring the thickness of thin parts or under the condition of low signal-to-noise ratio, the echo time domain signal often has aliasing, and it is very difficult to process and judge the echo waveform in the time domain. Difficult and imprecise.

本发明提供一种超声脉冲回波测厚中信号混叠条件下的厚度估算方法，依据脉冲回波信号在不同厚度下频谱图主瓣与各次谐波旁瓣细微的频移变化为基础，采用卷积神经网络对频谱图进行识别。所述频移变化具体表现为在某个厚度的回波频谱图为图2中的实线，更薄厚度的频谱图为图2中的虚线，当厚度变薄时，各次谐波旁瓣与主瓣的距离存在微小收缩变化。The present invention provides a thickness estimation method under the condition of signal aliasing in ultrasonic pulse echo thickness measurement. Spectrograms are identified using convolutional neural networks. The frequency shift change is embodied in that the echo spectrogram of a certain thickness is the solid line in Fig. 2, and the spectrogram of a thinner thickness is the dotted line in Fig. 2. When the thickness becomes thinner, the side lobes of each harmonic are There is a small constriction change in the distance from the main lobe.

本发明实施例的超声脉冲回波测厚中信号混叠条件下的厚度估算方法，参见图3，包括：The thickness estimation method under the condition of signal aliasing in ultrasonic pulse echo thickness measurement according to the embodiment of the present invention, referring to FIG. 3 , includes:

(1)选取目标厚度估算区间不同厚度的标准试块作为标准厚度参考点,所述标准试块应是涵盖待测工件厚度范围内的不同厚度的标准试块，标准参考点的数量将会直接影响最终厚度估算结果的精度。(1) Select standard test blocks with different thicknesses in the target thickness estimation range as the standard thickness reference points. The standard test blocks should be standard test blocks with different thicknesses within the thickness range of the workpiece to be measured. The number of standard reference points will be directly Affects the accuracy of the final thickness estimate.

(2)采用超声脉冲回波法多次采集每个不同厚度的标准试块的脉冲回波时域信号，所述超声脉冲回波法采用单探头自发自收模式并且采用相同的回波采样率。所述每个不同厚度的标准试块的脉冲回波时域信号采集次数影响未知工件厚度估计精度。(2) Using the ultrasonic pulse echo method to collect the pulse echo time domain signals of each standard test block with different thicknesses for many times, the ultrasonic pulse echo method adopts the single-probe spontaneous and self-receiving mode and adopts the same echo sampling rate . The number of pulse echo time-domain signal acquisitions of each standard test block with different thicknesses affects the estimation accuracy of the unknown workpiece thickness.

(3)对步骤(2)中获得的所有脉冲回波时域信号加窗后进行快速傅里叶变换生成频谱图，所述加窗可以是矩形窗和/或三角窗和/或汉宁窗，其目的是提取有效的时域回波信号减少计算量。一个标准试块可对应多幅频谱图。(3) After windowing all the pulse echo time-domain signals obtained in step (2), perform fast Fourier transform to generate a spectrogram, and the windowing may be a rectangular window and/or a triangular window and/or a Hanning window , whose purpose is to extract effective time domain echo signals to reduce the amount of computation. A standard test block can correspond to multiple spectrograms.

(4)将各标准试块的频谱图截取一定频率范围后作为训练样本集输入卷积神经网络中进行训练。(4) The spectrogram of each standard test block is cut into a certain frequency range and then used as a training sample set and input into the convolutional neural network for training.

所述频率范围的上限频率为：The upper limit of the frequency range is:

所述频率范围的下限频率为：The lower limit frequency of the frequency range is:

其中，f_m为探头的中心频率，f_w为探头的带宽。Among them, f_m is the center frequency of the probe, and f_w is the bandwidth of the probe.

进一步的，所述输入卷积神经网络中进行训练的频谱图应更改为如1024*1024的固定像素大小的图像。Further, the spectrogram for training in the input convolutional neural network should be changed to an image with a fixed pixel size such as 1024*1024.

进一步的，所述卷积神经网络卷积层的卷积核大小为5×5，计算公式为：Further, the size of the convolution kernel of the convolutional layer of the convolutional neural network is 5×5, and the calculation formula is:

其中，l代表所在卷积层的层数；

代表卷积层产生的特征图的第j个像素；f()是激活函数；M_j代表输入层图像的第j个感受野；

代表输入层图像的第j个感受野的第i个像素，所述输入层图像的感受野大小i等于所述卷积核的大小；

代表所述输入层图像的感受野的卷积核的第i个参数；

代表所述输入层图像的第j个感受野的偏置。Among them, l represents the number of layers of the convolutional layer;

represents the jth pixel of the feature map generated by the convolutional layer; f() is the activation function; Mj represents the_jth receptive field of the input layer image;

represents the i-th pixel of the j-th receptive field of the input layer image, and the receptive field size i of the input layer image is equal to the size of the convolution kernel;

The ith parameter of the convolution kernel representing the receptive field of the input layer image;

represents the bias of the jth receptive field of the input layer image.

进一步的，将所述卷积层产生的特征图输入卷积神经网络的池化层，对所述卷积层特征图做下采样，以降低图像的分辨率，池化层的计算公式如下：Further, input the feature map generated by the convolutional layer into the pooling layer of the convolutional neural network, and downsample the feature map of the convolutional layer to reduce the resolution of the image. The calculation formula of the pooling layer is as follows:

其中，

代表输入池化层的所述卷积层特征图的第j个感受野，down()代表池化函数，β代表权重系数，b代表偏置。in,

represents the jth receptive field of the feature map of the convolutional layer input to the pooling layer, down() represents the pooling function, β represents the weight coefficient, and b represents the bias.

进一步的，池化层生成的特征图经全连接层处理输出识别模型，用以识别待测工件的回波频谱图，并得出所述待测工件与标准工件的频谱图相似度。Further, the feature map generated by the pooling layer is processed by the fully connected layer to output a recognition model, which is used to identify the echo spectrogram of the workpiece to be tested, and obtain the similarity between the spectrogram of the workpiece to be tested and the standard workpiece.

(5)采用步骤(2)中所述的超声脉冲回波法采集待测工件的超声脉冲时域回波信号，经步骤(3)中相同处理后得到待测工件的超声脉冲回波的频谱图，将所述待测工件的频谱图截取所述频率范围后修改为所述固定像素大小的图像输入所述识别模型中，即得到待测工件与各标准试块的频谱相似度。(5) Using the ultrasonic pulse echo method described in step (2) to collect the ultrasonic pulse time domain echo signal of the workpiece to be measured, and obtaining the frequency spectrum of the ultrasonic pulse echo of the workpiece to be measured after the same processing in step (3). Fig. 1, cut the frequency range of the spectrogram of the workpiece to be tested and then modify it into an image of the fixed pixel size and input it into the recognition model, that is, to obtain the spectral similarity between the workpiece to be tested and each standard test block.

进一步的，以频谱相似度为权重估计待测工件的厚度，待测工件的厚度估计公式为：Further, the thickness of the workpiece to be tested is estimated with the spectral similarity as the weight, and the thickness estimation formula of the workpiece to be tested is:

d＝αW^TDd=αW^T D

其中，d为待测工件的厚度估计值，α为一维行修正系数矩阵，由标准试块及实际工件厚度测试标定而来，W为以频谱相似度权重为元素的对角矩阵，D为标准样块厚度值构成的一维列矩阵。Among them, d is the estimated thickness of the workpiece to be tested, α is a one-dimensional row correction coefficient matrix, which is calibrated from the standard test block and the actual workpiece thickness test, W is a diagonal matrix with spectral similarity weights as elements, D is A one-dimensional column matrix of standard block thickness values.

以上所述仅是本发明的优选实施方式，应当指出，对于本技术领域的普通技术人员来说，在不脱离本发明技术原理的前提下，还可以做出若干改进和变形，这些改进和变形也应视为本发明的保护范围。The above are only the preferred embodiments of the present invention. It should be pointed out that for those skilled in the art, without departing from the technical principle of the present invention, several improvements and modifications can also be made. These improvements and modifications It should also be regarded as the protection scope of the present invention.

Claims (10)

Translated fromChinese

1.一种超声脉冲回波测厚中信号混叠条件下的厚度估算方法，其特征在于，包括：1. a thickness estimation method under signal aliasing condition in ultrasonic pulse echo thickness measurement, is characterized in that, comprises:选取目标厚度估算区间不同厚度的标准试块作为标准厚度参考点；Select the standard test blocks with different thicknesses in the target thickness estimation range as the standard thickness reference point;采用超声脉冲回波法多次采集每个不同厚度的标准试块的脉冲回波时域信号；The ultrasonic pulse echo method is used to collect the pulse echo time domain signals of each standard test block with different thicknesses for many times;对获得的标准试块的所有脉冲回波时域信号加窗后进行快速傅里叶变换生成标准试块的频谱图；After windowing all the pulse echo time-domain signals of the obtained standard test block, fast Fourier transform is performed to generate the spectrogram of the standard test block;将各标准试块的频谱图截取一定频率范围后作为训练样本集输入卷积神经网络中进行训练，输出识别模型；The spectrogram of each standard test block is cut into a certain frequency range and input into the convolutional neural network as a training sample set for training, and the recognition model is output;将待测工件的超声脉冲回波的频谱图输入识别模型中得到待测工件与各标准试块的频谱相似度，以频谱相似度为权重估计待测工件的厚度。Input the spectrogram of the ultrasonic pulse echo of the workpiece to be tested into the recognition model to obtain the spectral similarity between the workpiece to be tested and each standard test block, and use the spectral similarity as the weight to estimate the thickness of the workpiece to be tested.2.根据权利要求1所述的一种超声脉冲回波测厚中信号混叠条件下的厚度估算方法，其特征在于，所述标准试块是涵盖待测工件厚度范围内的不同厚度的试块。2. The thickness estimation method under the condition of signal aliasing in a kind of ultrasonic pulse echo thickness measurement according to claim 1, it is characterized in that, described standard test block is the test that covers different thicknesses in the thickness range of workpiece to be measured. piece.3.根据权利要求1所述的一种超声脉冲回波测厚中信号混叠条件下的厚度估算方法，其特征在于，所述超声脉冲回波法采用单探头自发自收模式，并且采用相同的回波采样率。3. the thickness estimation method under the condition of signal aliasing in a kind of ultrasonic pulse echo thickness measurement according to claim 1, is characterized in that, described ultrasonic pulse echo method adopts single probe spontaneous self-receiving mode, and adopts the same echo sampling rate.4.根据权利要求1所述的一种超声脉冲回波测厚中信号混叠条件下的厚度估算方法，其特征在于，所述将各标准试块的频谱图截取一定频率范围，包括：4. the thickness estimation method under the condition of signal aliasing in a kind of ultrasonic pulse echo thickness measurement according to claim 1, is characterized in that, the described spectrogram of each standard test block is intercepted to a certain frequency range, comprising:截取的上限频率为：The upper limit frequency of interception is:

截取的下限频率为：The lower limit frequency to be intercepted is:

其中，f_H为上限频率，f_m为探头的中心频率，f_w为探头的带宽。Among them, f_H is the upper limit frequency, f_m is the center frequency of the probe, and f_w is the bandwidth of the probe.5.根据权利要求1所述的一种超声脉冲回波测厚中信号混叠条件下的厚度估算方法，其特征在于，所述训练样本集输入卷积神经网络之前，将频谱图更改为固定像素大小的图像。5. The thickness estimation method under the condition of signal aliasing in ultrasonic pulse echo thickness measurement according to claim 1, wherein, before the training sample set is input to the convolutional neural network, the spectrogram is changed to fixed Pixel-sized image.6.根据权利要求1所述的一种超声脉冲回波测厚中信号混叠条件下的厚度估算方法，其特征在于，所述卷积神经网络卷积层的卷积核大小为5×5。6 . The thickness estimation method under the condition of signal aliasing in ultrasonic pulse echo thickness measurement according to claim 1 , wherein the convolution kernel size of the convolutional neural network convolution layer is 5×5. 7 . .7.根据权利要求1所述的一种超声脉冲回波测厚中信号混叠条件下的厚度估算方法，其特征在于，卷积神经网络训练中，池化层生成的特征图经全连接层处理输出识别模型，用以识别待测工件的超声脉冲回波频谱图，并得出待测工件与标准工件的频谱图相似度。7. The thickness estimation method under the condition of signal aliasing in a kind of ultrasonic pulse echo thickness measurement according to claim 1, it is characterized in that, in the training of convolutional neural network, the feature map generated by the pooling layer is processed by the fully connected layer The processing output identification model is used to identify the ultrasonic pulse echo spectrogram of the workpiece to be tested, and the similarity of the spectrogram of the workpiece to be tested and the standard workpiece is obtained.8.根据权利要求1所述的一种超声脉冲回波测厚中信号混叠条件下的厚度估算方法，其特征在于，所述待测工件的超声脉冲回波的频谱图的获取，包括：8. the thickness estimation method under the signal aliasing condition in a kind of ultrasonic pulse echo thickness measurement according to claim 1, is characterized in that, the acquisition of the spectrogram of the ultrasonic pulse echo of described workpiece to be measured, comprises:采用超声脉冲回波法采集待测工件的超声脉冲回波时域信号，The ultrasonic pulse echo method is used to collect the ultrasonic pulse echo time domain signal of the workpiece to be tested.对所获取的超声脉冲回波时域信号加窗后进行快速傅里叶变换生成频谱图。After windowing the acquired ultrasonic pulse echo time domain signal, fast Fourier transform is performed to generate a spectrogram.9.根据权利要求8所述的一种超声脉冲回波测厚中信号混叠条件下的厚度估算方法，其特征在于，所述将待测工件的超声脉冲回波的频谱图输入识别模型中之前，将所述频谱图截取一定频率范围后修改为固定像素大小的图像。9. The thickness estimation method under the condition of signal aliasing in a kind of ultrasonic pulse echo thickness measurement according to claim 8, it is characterized in that, the described spectrogram of the ultrasonic pulse echo of the workpiece to be measured is input in the identification model Previously, the spectrogram was cut into a certain frequency range and then modified into an image with a fixed pixel size.10.根据权利要求1所述的一种超声脉冲回波测厚中信号混叠条件下的厚度估算方法，其特征在于，所述待测工件的厚度估计如下：10. The thickness estimation method under the condition of signal aliasing in a kind of ultrasonic pulse echo thickness measurement according to claim 1, is characterized in that, the thickness estimation of described workpiece to be measured is as follows:d＝αW^TDd=αW^T D其中，d为待测工件的厚度估计值，α为一维行修正系数矩阵，由标准试块及实际工件厚度测试标定而来，W为以频谱相似度权重为元素的对角矩阵，D为标准样块厚度值构成的一维列矩阵。Among them, d is the estimated thickness of the workpiece to be tested, α is a one-dimensional row correction coefficient matrix, which is calibrated from the standard test block and the actual workpiece thickness test, W is a diagonal matrix with spectral similarity weights as elements, D is A one-dimensional column matrix of standard block thickness values.

Images