一种冲击波压力传感器的准静态校准神经网络模型  

A neural network model of quasi-static calibration for shock wave pressure sensors

作　　者：赵传荣[1] 孔德仁[2] 王胜强[3] 商飞[2] 

机构地区：[1]安徽工业大学电气与信息工程学院,安徽马鞍山243032 [2]南京理工大学机械工程学院,南京210094 [3]西安近代化学研究所,西安710065

出　　处：《振动与冲击》2017年第13期92-95,139,共5页Journal of Vibration and Shock

基　　金：国家自然科学基金(11372143)

摘　　要：由冲击波压力传感器准静态校准原理,间接比对式校准的精度取决于重锤的落高与压力拟合模型的精度,本文采用RBF神经网络建立了以落高为输入量、冲击波压力峰值为输出量的神经网络模型。选用典型标准压力传感器,在7~30 MPa量程范围开展校准实验;通过对测试样本进行分析,结果表明:该神经网络模型预测的最大相对误差不超过0.04%,比多项式拟合模型和指数拟合模型高一个数量级。落高与压力拟合模型引入的不确定度是构成冲击波压力传感器动态测量不确定度的一个重要分量,通过建立高精度的重锤落高与冲击波压力峰值神经网络拟合模型,为进一步提高冲击波压力传感器的测量精度奠定了基础。; According to the principle of quasi-static calibration of shock wave pressure sensors, the indirect comparative calibration accuracy depends on the accuracy of the fitting model for pressure versus height of drop-hammer. Here, the RBF neural network was used to set up a neural network model taking height of drop-hammer as input and shock wave pressure peak as output. Choosing a typical standard pressure sensor, the calibration test was performed in the range of 7-30 MPa. Through analyzing test samples, the results indicated that the maximum relative error of the neural network model,s prediction is not more than 0. 04% , the accuracy of this model is one order of magnitude higher than those of the polynomial fitting model and the exponential fitting model ; the uncertainty induced by the fitting model for pressure versus height of drop-hammer is an important component of the dynamic measurement uncertainty of shock wave pressure sensors; to build a neural network fitting model for shock wave pressure peak versus height of drop-hammer with a higher accuracy can lay a foundation for improving the measurement accuracy of shock wave pressure sensors.

关 键 词：冲击波压力传感器 RBF神经网络 准静态校准 拟合模型 

分 类 号：O385[理学—流体力学]