应用卷积神经网络模型的超声特征信号提取算法  

Ultrasonic feature signal extraction algorithm based on convolutional neural network model

作　　者：樊丹丹 孔明[1] 马馨玥 崔志文 徐佳奇 FAN Dandan;KONG Ming;MA Xinyue;CUI Zhiwen;XU Jiaqi(China Jiliang University,Hangzhou 310018,China;Goldcard Smart Group Co.,Ltd.,Hangzhou 310018,China)

机构地区：[1]中国计量大学,浙江杭州310018 [2]金卡智能集团股份有限公司,浙江杭州310018

出　　处：《中国测试》2024年第12期117-124,共8页China Measurement & Test

摘　　要：飞行时间差是时差法超声波流量计的关键参数,决定表具的计量精度。该文采用卷积神经网络对超声回波信号进行特征提取,提取的特征用来回归预测飞行时间差。超声回波信号作为模型的输入层,中间层为提高模型性能,加速训练,使用五层卷积层、五层池化层及RELU激活函数提取信号特征,输出层回归预测飞行时间差,提高对时间差估计的精度。仿真研究表明,模型预测的准确率高于99%,且有较好的泛化能力。搭建实验平台,进行实验研究,结果表明,卷积神经网络模型用于预测超声回波信号飞行时间差有着较高的测量准确性,其中测量误差优于±1%,重复性优于0.2%。The flight time difference is a key parameter of the time difference ultrasonic flowmeter,which determines the measurement accuracy of the meter.In this paper,convolutional neural network is used to extract features of ultrasonic echo signal,and the extracted features are used to regression predict the flight time difference.The ultrasonic echo signal is used as the input layer of the model.In order to improve the model performance and speed up the training,the middle layer uses five layers of convolution layer,five layers of pooling layer and RELU Activation function to extract the signal characteristics,and the output layer regression predicts the flight time difference,which improves the accuracy of time difference estimation.Simulation studies have shown that the accuracy of the model predict is higher than 99%and has good generalization ability.Building an experimental platform and conducting experimental research,the results show that the convolutional neural network models are used to predict flight time differences in ultrasonic echo signals which has high measurement accuracy,measurement error is better than±1%,and repeatability is better than 0.2%.

关 键 词：超声波气体流量计 飞行时间差检测 卷积神经网络模型 回波信号处理 

分 类 号：TB9[一般工业技术—计量学]