Localized Parameter Prediction And Profile Reconstruction Of Dilute Gas-Solid Two-Phase Flow Based On Data-Driven Modeling

In pneumatic conveying,real-time measurement of the cross-sectional velocity distribution of solid particles is beneficial to industrial safety production and quality improvement.However,the optical,microwave and acoustic sensors have inherent limitations in measuring the distribution of particle parameters in the cross-section of square pipes.In this research,a new method for measuring the particle velocity distribution in the cross-section of a square pipe by using electrostatic sensor arrays and Gaussian process regression(GPR)is proposed and verified.This method solves the problem that the non-restrictive electrostatic sensor array is difficult to measure the particle velocity distribution in the cross section of a pipe due to its limitation of sensing area and realizes the measurement of particle velocity in local areas and the cross-sectional imaging.The feasibility of the method is demonstrated and verified by modeling analysis,system design,experimental testing and other ways.The main work and innovations are as follows:(1)A novel method for measuring particle velocity distribution in the cross section of a square pipe using an electrostatic sensor array and GPR model is proposed.The sensor design,modeling process and data analysis flow related to the method are described in detail,and the relationship between the input velocity variable and the particle velocity in the local area of the pipe cross-section is established.According to the measured particle velocity in the local area of the pipe cross-section,the image of particle velocity distribution is obtained by the biharmonic spline interpolation algorithm.(2)To verify and evaluate the feasibility and effectiveness of the proposed method,an electrostatic sensor comprising restrictive and non-restrictive electrodes with a side length of 54 mm is designed and implemented.Non-restrictive sensors are used to measure the particle velocity and provide input variables for the data-driven model.Experimental tests were carried out on a tailormade pneumatic conveying platform under 16 different air velocity and particle mass flow conditions.(3)The relationship between the input velocity variable and the particle velocity in the local area of a pipeline cross-section is established based on the developed GPR model.The performance of the model is quantitatively compared with other machine learning models.Under all experimental conditions,the relative error of particle velocity prediction is within 3%.The prediction performance of the model is explored using insufficient training sets and fewer measuring electrodes,which provides guidance for further optimization of the measurement method.When the training dataset is not comprehensive enough,the performance of the model is affected,and the range of relative error is extended from-9%to+15%.With fewer measurement electrodes(input variables),the relative error of regional particle prediction speed increases slightly but remains within 5%.However,the prediction speed of the model is greatly improved.Results obtained suggest that the electrostatic sensor in conjunction with the GPR model is a feasible approach to obtain the cross-sectional velocity distribution of pneumatically conveyed particles in a square-shaped pipe.Further research will be carried out on large-size pipelines of fired plants.