传感器标定的神经网络杂交建模方法

Neural Network Hybrid Modeling Method For Transducer Calibration

传感器标定是工程测试中的一个重要环节,直接影响测试结果的精度及可靠性。当被测物理量与传感器输出信号间的关系包含复杂、未知的非线性特性时,传统的标定方法难以达到满意的精度。引入神经网络杂交建模的思想,提出传感器标定的神经网络杂交建模方法,阐明建模过程和步骤。分别以单输入单输出和多输入多输出传感器为例,进行杂交建模标定的仿真研究,并对一个6维力传感器样机完成了神经网络杂交建模试验标定。仿真与试验结果表明,与传统的标定方法相比,神经网络杂交建模方法能够显著提高传感器的标定精度,同时比神经网络黑箱建模方法具有更小的网络规模和更快的收敛速度,且精度高,泛化推广能力强。

The transducer calibration is a key link in engineering test tasks and has significant influence on accuracy and reliability of test results.When input-output relationships of transducers contain unknown and complex nonlinear characteristics,the conventional calibration method can hardly achieve satisfactory accuracy.A neural network（NN） hybrid modeling approach is proposed and applied to transducer calibration.The modeling procedures are also presented in detail.The simulated studies on the calibration of single output and multiple output transducers are conducted respectively by use of the developed hybrid modeling scheme.The NN hybrid modeling approach is utilized to calibrate a six-dimensional force sensor prototype based on the measured data obtained from calibration tests.The simulated and experimental results show that the NN hybrid modeling approach can improve significantly calibration precision in comparison with traditional calibration methods.In addition,the NN hybrid modeling is superior to NN black box modeling because the former possesses smaller network scale,higher convergence speed,higher calibration precision and better generalization performance.