Research On Random Error Suppression Technology Of Micromachined Silicon Resonant Accelerometer

In recent years,the rapid development of MEMS technology has led to the gradual development of various sensors towards miniaturization.Micromachined silicon resonant accelerometer has the advantages of miniaturization that traditional accelerometer do not have,and its potential high precision characteristics make it a key research object of various research institutions.In this paper,random errors existing in the output data of micromachined silicon resonant accelerometer are analyzed and error suppression studies are carried out.The main contents include the following aspects:(1)The operating principle of micromachined silicon resonant accelerometer and their basic mechanical model are analyzed,and the functional relationship between the acceleration and the resonant frequency of the resonator is deduced;Allan variance analysis was used to identify the random errors in the micromachined silicon resonant accelerometer,and it was concluded that velocity random walk,rate random walk and zero bias instability noise were the three kinds of random errors in the accelerometer.(2)Based on the ARMA(2,1)model and the adaptive Kalman filtering technique(SHAKF),a static model for random error suppression of micromachined silicon resonant accelerometer was developed.The purpose is to attenuate the random error in the accelerometer output data in static environment.The online ambient ±0g stability,±1g stability and multi-temperature point 0g stability experiments were designed to test the adaptability of the model.The experimental results show that the model works as expected,and the velocity random walk is significantly suppressed in these tests.(3)Aiming at the random errors of accelerometer data in complex environment,a suppression model is established by using wavelet threshold denoising algorithm.The wavelet threshold function is improved,and the new threshold function has better noise reduction performance than the traditional soft and hard threshold functions.It was used for noise reduction of accelerometer data at temperatures ranging from-40°C to 60°C,and the results show that the short-time velocity random walk are better suppressed.At various fixed temperature points,the accelerometer output data are used for noise reduction to attenuate the rate random walk when the accelerometer is in motion.(4)A random error suppression model is build by using the wavelet threshold noise reduction results and the CFPSO-ELM algorithm.It is used to predict the true value of the accelerometer output.In complex environments,the problem of random errors in the output data of micromachined silicon resonant accelerometer is solved,making the model output results more close to the true value.The Extreme Learning Machine(ELM)reduces its stochasticity and easy access to optimal solutions with the help of the Compression Factor Particle Swarm Optimization(CFPSO)algorithm.The ELM and CFPSO-ELM were trained with the data denoised by wavelet threshold,and the data prediction model of accelerometer in full temperature environment was established,which had high prediction accuracy.At 20°C,for example,ELM and CFPSO-ELM were trained with the wavelet threshold denoised data and the calibration data of turntable.A data prediction model of accelerometer in motion was established,which had a high goodness of fit value.