Method for establishing electrical capacitance tomography complex flow pattern data set

Abstract

The invention discloses a method for establishing an electrical capacitance tomography complex flow pattern data set. The method comprises the following steps: generating medium distribution of a complex flow pattern by adopting a random noise filtering method; and calculating a corresponding capacitance vector by using a numerical method so as to establish a data set of the capacitance vector corresponding to the medium distribution. The method for generating the medium distribution of the complex flow pattern by adopting the random noise filtering method comprises the following steps of: performing mesh generation on a measurement section in a pipeline; generating a random number matrix; and filtering the generated random matrix by using a mean filter so as to smooth the random matrix. Compared with other existing numerical simulation generation modes, the method has the advantages that an algorithm combining random number generation and multiple filtering is adopted, a data set of acomplex flow pattern can be established, and the applicability is high.

Description

technical field [0001] The invention relates to the technical field of gas-solid two-phase flow measurement, in particular to an electrical capacitance tomography technology for measuring the flow properties of gas-solid two-phase flow. Background technique [0002] Image reconstruction is an important part of Electrical Capacitance Tomography (ECT). The deep learning algorithm uses a processing layer composed of multiple nonlinear transformations to perform high-level abstraction on the data, which can intelligently optimize the relationship between the medium distribution and the capacitance vector. Complex nonlinear mapping can effectively solve the "soft field" problem of the ECT system, and is gradually being used to solve the problem of ECT image reconstruction. Deep learning predicts unlearned samples by learning the characteristics of the data set, so the richness of the data set has a crucial impact on the training results of the deep learning model. Numerical simu...

AI Technical Summary

Problems solved by technology

An important reason why most of the tomographic reconstruction models based on deep learning are not ideal is the lack of reasonable and effective data set construction methods.

The data sets used by the neural network used for ECT image reconstruction can be generated by numerical simulation, but most of them only contain typical flow patterns such as simple laminar flow and annular flow, which makes it difficult to apply the obtained model to complex gas-solid two-phase flow process

A small number of studies use traditional algorithms to reconstruct the image of the collected fluidized bed measured capacitance data, thus establishing sample data. However, the dynamic range covered by the data set obtained through experiments is limited, and the trained machine learning model cannot analyze the capacitance data outside the sample. Flow State Forecasting

In order to ensure that the deep learning ECT image reconstruction model has a strong generalization ability, it is required that the data set used should contain flow regime data that conforms to the flow pattern of two-phase flow. Currently, research on how to establish a reasonable data set is still lacking.

Images