Flow Measurement Intelligent Calibration Technology Based On Coriolis Effect

Coriolis mass flowmeter,a new type of direct mass flowmeter,is widely applied in various fields because of its excellent performance and characteristic that the mass flow of fluid is not affected by other physical parameters.Due to the research on Coriolis mass flowmeter is late in our country,and the core technology is also monopolized by foreign companies,the products on the market generally have the problems of low measurement accuracy and poor stability.At the same time,the performance of the flowmeter is greatly reduced under the two-phase flow.For this reason,this paper focuses on the signal processing of Coriolis mass flowmeter.Based on the traditional digital signal processing method,the extraction of phase difference is studied,and the effect of non-integer periodic sampling on Fourier transform and Hilbert transform was analyzed.With the combination of theory and simulation,the problems of correlation analysis method were explained,therefore two improved methods were proposed.It was verified by simulation that the improved correlation analysis methods not only broke through the limitation of the integer periodic sampling,but also had higher measurement accuracy and good stability.In addition,a noise reduction method for sinusoidal signal based on Kalman filtering was proposed.The original signal with a signal-to-noise ratio of 30 d B was increased to 43 d B,and various phase difference extraction algorithms were improved to varying degrees.A data-driven approach was applied to signal processing in Coriolis mass flowmeter.Firstly,set up an experimental platform to collect data,taking the vibration signal as the input feature,the mass flow form standard meter as the sample label.Driven by a large amount of experimental data,the network parameters were trained by gradient descent.The results showed that the performance of the Long-Short-Term Memory network was better than other network structures;batch normalization during training process can improve convergence speed;Bayesian model fusion and moving average processing can further optimize the model.On the test set,the mean square error of the deep learning model reached 0.0047,which was superior to the calibrated meter,and the measured values were close to that of standard meter.Therefore,the feasibility of this method was verified.For gas-liquid two-phase flow,a random walk model was established to describe the vibration displacement of the measuring tube in this paper.When the flow was in a relatively stable state,the probability distribution of the mass flow from the calibrated meter was analyzed by statistical method.Then,an autoregressive moving average model was constructed based on the autocorrelation coefficient of the mass flow series.Experimental data proved that the ARMA model effectively implemented static correction of the mass flow value.Besides,for the dynamic mass flow,the vibration signal and measured value of the calibrated meter were applied as input features of deep learning model,combined with early stopping can effectively avoid overfitting.The moving average was used to further improve the stability and generalization of the algorithm,so as to realize the dynamic correction method finally.