Artificial Intelligence Recognition Method Of Two - Phase Flow Pattern In Small And Medium Diameter

Multiphase flow is a complex fluid flow phenomena that is broadly exist in the industrial production. The detecting of its relevant parameters has been the focus of the research at home and abroad. In order to study the gas-liquid two-phase flow pattern recognition in vertical riser pipe, this paper designed a kind of conductance sensor array and its peripheral acquisition circuit, optimized the sensor structure by using the finite element simulation software, optimized the excitation electrode by means of axial and radial sensitivity. In accordance with the principle of economic efficiency, he use of a simplified method for docking pressure ring embedded electrode production technology requirements saving the research cost effectively. Designed and made current source and conductance fluctuation signal acquisition circuit. Constant current source could obtain a stable sine wave generated AD9850and output constant current value a voltage-controlled current source. The signal acquisition circuit achieve a differential signal amplification, phase-sensitive demodulation and low-pass filter the data are collected and stored by PCI8735data acquisition card and data acquisition program which using Labview to collect and store the data. Designed a simple experimental gas-liquid two-phase flow generating device is designed, conducted gas-liquid two-phase flow patterns characteristic signal acquisition in the experimental pipe whose inner diameter is18mm.Through the use of relevant methods of measurement on the flow rate of the transit time measurement principle, calculate the different points between the data acquisition points from the two signals. Eliminated the error of the collected dual conductance fluctuation signal by applying the averaging method. Subdivided of the extracted conductance fluctuation signal which shows more detailed information of the signals and further refined the collected signal conductance fluctuations. Extracted the stream pattern feature of the conductance signals decomposed from wavelet packet is extracted by information entropy. A specific calculation method for computing discrete one-dimensional signal information entropy by using MATLAB was given. The extracted information entropy of each band are regarded as training samples to build the neural networks, and the flow pattern intelligently recognize. Experimental results show that:the system can basically achieve the desired design goals and has some practical significance.