Multi-classification ERT flow pattern identification method

Abstract

The invention discloses a multi-classification ERT flow pattern identification method. The method comprises the following steps: 1, constructing a multi-classification ERT flow pattern simulation model; 2, acquiring a measurement voltage; 3, processing the data of the measured voltage; 4, constructing a database; 5, training and testing a CNN; and 6, performing multi-classification ERT flow pattern identification. According to the method, the electrode system is used for collecting and measuring voltage, one-dimensional voltage data information is converted into two-dimensional dot matrix information, time domain-to-frequency domain conversion is carried out on the basis, data features are enhanced while original data integrity is guaranteed, time domain and frequency domain dual-channel samples are adopted, and the robustness of an algorithm for recognizing samples of different scales is improved. A multi-flow pattern classification ERT voltage image database is constructed, multiple groups of easily-confused flow pattern categories are added, a high-difficulty identification task is completed, the convolutional neural network is selected to perform ERT flow pattern identification, and the method has superiority in easily-confused flow pattern identification.

Description

technical field [0001] The invention belongs to the technical field of multi-classification ERT flow pattern recognition, and in particular relates to a multi-classification ERT flow pattern recognition method. Background technique [0002] As a cutting-edge technology of modern industrial detection, Electrical Resistance Tomography has the characteristics of low cost, no radioactivity, visualization and non-invasiveness, and is widely used in the measurement of two-phase flow in petroleum, chemical, electric power and metallurgy industries . Flow pattern is an important index of two-phase flow, and the measurement of other parameters often depends on the accurate identification of convection pattern. However, due to the "soft field" characteristics of ERT, the number of subdivided units in the imaging area, noise, transmission, and imaging algorithms, there are many problems in the existing ERT flow pattern recognition methods: [0003] First, most of the existing flow pa...

AI Technical Summary

Problems solved by technology

[0003] First, most of the existing flow pattern identification methods directly use the measured voltage without preprocessing as samples, or use inappropriate preprocessing methods, resulting in the loss of data characteristics, and it is difficult to accurately characterize different flow patterns;

[0004] Second, the ERT flow pattern reconstruction algorithm has problems such as unclear imaging and poor imaging quality, and it is difficult to accurately identify the flow pattern category based on the reconstructed image;

[0005] Third, the existing flow pattern identification categories are mainly divided into core flow, bubbly flow, circular flow, and laminar flow, and no more detailed classification is made for the characteristics of different flow patterns;

[0006] Fourth, the accuracy of existing flow pattern identification algorithms is not ideal, especially when identifying some confusing flow patterns, it is prone to misjudgment

Images