Study On Signal Processing And Correction Of Coriolis Two-phase Signal

Coriolis flowmeter is widely used in industry.Single-phase flow measurement has reached 0.01 industry precision while the bottleneck of two-phase flow measurement always exists.A signal enhancement algorithm based on a self-adaptive lattice filter can cause a delay in frequency tracking when the gas void fraction is too high.In order to speed up convergence,its bandwidth is increased,making the frequency tracking more accurate,and the mass error dropping from 2.09% to 1.49%.A signal enhancement algorithm based on Empirical Mode Decomposition(EMD)has also been tried.This algorithm could automatically extract the main trend of signal leading to its advantage in measuring highly vibrated signals.Both simulation results under improved random walk model and real-world experiments on experimental platform show that EMD algorithm out-performs adaptive filter under high gas void fraction situation.A glide smoothing on phase difference is proposed to reduce the influence of error on small phase difference.Such operation could eliminate outliers and effectively reduce the phase difference affected by the noise.Experiments under single-phase flow,multi-phase flow and down-sampling situation all show 20% and more precision improvement,proving that the proposed algorithm's efficiency.A two-phase experimental platform based on CCM2000 Coriolis flowmeter is built.Due to the change of damping ratio in two-phase flow experiment,the tube vibration would decrease in amplitude and even stop.We design a digital drive system to tackle with it.Simulation result shows that the proposed digital drive system could stabilize the amplitude effectively.The above mentioned signal enhancement and phase calculation algorithms are all integrated into DSP to build the two-phase targeted Coriolis flowmeter digital system.Regarding the two-phase flow model error,BP neural network is implemented and 3 new features are introduced.These new designed features reduce the calibration error from 2.29% to 1.93%.Convolution neural network(CNN)is used to further extract features from the original data.By adding old features into fully connected layers,the average error drops to 1.82%.To predict the presence of large bubbles in the vibrating tube,CNN is used to predict the gas void fraction of the simulated data.