Correction Method Research On Nonlinear Characteristics Of Quartz Tuning Fork-Temperature Sensor

In modern measurement system, the characteristic of the whole measure-ment system is impacted by the working characteristic of sensor. because that quartz tuning fork temperature sensor is numerical frequency output sensor, whose output frequency changes with outside temperature, with good anti-jamming performance, well stability for a long time and high sensitivity and ac-curacy, so it is widely applied in high accuracy measurement field. Since quartz tuning fork temperature sensors has error in the layout of electrode and structure in the manufacturing process, so that the output of temperature frequency is nonlinear in the entire temperature measurement range. The solution of sensor non-linear is mainly used least squares polynomial fitting, or using look-up table interpolation, but more computing time is needed for the former, although the latter faster, there is interpolation nodes error .To the above issues, wavelet neural network method is presented in this pa-per for correcting the temperature frequency characteristics of quartz tuning fork temperature sensor. First algorithms and training process of wavelet neural net-work and BP neural network are introduced, the corresponding program is com-piled in Matlab, the same temperature and frequency data is used to train two kinds of networks, to select the more suitable neural network approach correction quartz tuning fork temperature sensor, through the example comparing the advan-tages and disadvantages of two kinds of networks, we find that wavelet neural network has better convergence and accuracy, finally we choose wavelet neural network method for correction the temperature frequency characteristics of quartz tuning fork temperature sensor. To remove the noise signal in frequency signal, experimental data is used for wavelet transform, obtained de-noising sig-nal is used as new experimental data, the corresponding program is compiled in Matlab. We train new data by wavelet neural network, obtained error curve close to the set goals error, the temperature frequency characteristics of quartz tuning fork temperature sensor is basic linear. And compareing with conventional cor-rection nonlinear method, experimental results shows wavelet neural network method has higher accuracy. Finally, random measurement eight groups data varifies experimental model, the experimental output value and the measured value are almost equal, temperature frequency data shows a linear relationship, the experimental results be verified.