基于逆传感网络模型辨识的激波管动态压力重构  

Dynamic pressure reconstruction of shock tube based on inverse sensing network model identification

作　　者：李永生 姚贞建 刘臣 丁义凡 LI Yongsheng;YAO Zhenjian;LIU Chen;DING Yifan(School of Electrical and Information Engineering,Wuhan Institute of Technology,Wuhan 430205,China)

机构地区：[1]武汉工程大学电气信息学院,武汉430205

出　　处：《振动与冲击》2023年第23期223-230,324,共9页Journal of Vibration and Shock

基　　金：国家自然科学基金(52005202);武汉工程大学研究生教育创新基金项目(CX2022120);武汉工程大学校内科学基金研究项目(K2021025);湖北省教育厅科学技术研究项目(Q20221512)。

摘　　要：提出一种基于逆传感网络模型辨识的激波管动态压力重构方法。首先,基于经验模态分解,将压力传感器动态响应信号分解为一系列不同频带的分量;然后,采用相关系数和振铃幅值占比指标,实现振铃分量和趋势分量的识别,构建逆传感网络模型训练集和测试集;最后,基于双向长短期记忆神经网络训练及测试,建立压力传感器逆传感网络模型,实现激波管动态压力重构。分别通过仿真试验和激波管测量试验验证该方法对于动态压力重构的性能。仿真试验结果显示,重构的动态压力信号的RMSE(root mean square error)和MAPE(mean absolute percentage error)远小于传统趋势估计法,其值比长短时记忆(long short-term memory, LSTM)方法得到的结果分别减小了2倍和5倍,并且在不同阶数的压力传感器仿真试验中验证了该方法的鲁棒性,通过对比不同压力传感器系统下该方法的动态压力重构精度,验证了该方法的适用性;激波管测量试验结果显示,模型训练和测试输出的RMSE和MAPE分别为0.001 6 V、0.003 6%和0.002 5 V、0.062%,重构得到的激波管动态压力在平稳区间内的平均相对误差约为2.14%。Here,a dynamic pressure reconstruction method for shock tube was proposed based on inverse sensing network model identification.Firstly,based on empirical mode decomposition(EMD),dynamic response signal of pressure sensor was decomposed into a series of components in different frequency bands.Then,correlation coefficients and ringing amplitude proportion indexes were used to realize identification of ringing components and trend components,and construct training and testing sets of inverse sensing network model.Finally,based on training and testing of bidirectional long short-term memory(LSTM)neural network,inverse sensing network model of pressure sensor was established to realize dynamic pressure reconstruction of shock tube.The performance of the proposed method for dynamic pressure reconstruction was verified through simulation experiments and shock tube measurement experiments,respectively.The results showed that root mean square error(RMSE)and mean absolute perccntage error(MAPE)of the reconstructed dynamic pressure signal are much smaller than those obtained with the traditional trend estimation method,and their values are reduced by 2 and 5 times compared to those obtained by LSTM method,respectively;the robustness of this method is verified through simulation experiments of pressure sensors with different orders;by comparing different pressure sensor systems’dynamic pressure reconstruction accuracies obtained with this method,the applicability of this method is verified;in shock tube measurement experiments,RMSE and MAPE output by model training and testing are 0.0016 V,0.0036%and 0.0025 V,0.062%,respectively,average relative error of the reconstructed dynamic pressure of shock tube within stationary intervals is about 2.14%.

关 键 词：激波管 压力传感器 经验模态分解(EMD) 双向长短期记忆神经网络 动态压力 

分 类 号：TH71[机械工程—测试计量技术及仪器]