Multi Classification ERT Flow Pattern Recognition Method Based on Deep Learning

Abstract

Electrical resistance tomography (ERT) is the frontier technology of modern industrial detection, in which flow pattern is an important index of two-phase flow detection. Affected by many factors, ERT flow pattern recognition is difficult. In this paper, an ERT flow pattern recognition method based on deep learning is designed in order to obtain the real situation of flow pattern in pipeline in practical application. The original ERT measured voltage is transformed from one-dimensional data information to two-dimensional dot matrix information by pseudo image coding method. According to the characteristics, the flow patterns are divided into 27 categories, and then ERT voltage image databases with different scales are established in time domain and frequency domain. Convolutional neural network is used to construct ERT flow pattern recognition network model based on deep learning, and experiments are designed to verify its performance. The results show that the average accuracy of each flow pattern recognition of this algorithm can reach 98.74%, of which the accuracy of 14 types of flow pattern recognition is 100%. This method can achieve high-precision ERT flow pattern recognition task.