A method for reconstructing a two-dimensional image of a gas-liquid two-phase flow pattern

Abstract

The application relates to a two-dimensional image reconstruction method of a gas-liquid two-phase flow pattern, and the method is as follows: (1) a three-dimensional geometric model of a capacitance tomography system for gas-liquid two-phase flow pattern detection is established, a capacitance vector between electrode pairs of the capacitance tomography system under different flow patterns is obtained through numerical simulation, and an original sample set is obtained; (2) dimension reduction processing is performed on the original sample, and the data after the dimension reduction processing is used as clustering samples; (3) a mode recognition is performed on the clustering samples through a density-based clustering algorithm, the original samples are divided, and training samples are obtained; (4) a full-connection deep learning network model is built, and original two-dimensional image data of the gas-liquid two-phase flow pattern is output; and (5) the original two-dimensional image data of the gas-liquid two-phase flow pattern is further processed, and the two-dimensional image reconstruction of the gas-liquid two-phase flow pattern is completed. The application is favorable for improving the reconstruction quality of the two-dimensional image of the gas-liquid two-phase flow pattern.

Description

Technical Field

[0001] This invention belongs to the field of gas-liquid two-phase flow detection technology, and in particular relates to a method for reconstructing two-dimensional images of gas-liquid two-phase flow patterns. Background Technology

[0002] Gas-liquid two-phase flow is widely used in petroleum, energy, chemical, and power engineering fields. In actual industrial production processes, analytically detecting and studying the distribution characteristics of various parameters (cross-sectional void fraction, flow pattern, velocity, mass flow rate, etc.) in gas-liquid two-phase flow processes is highly beneficial for managing and regulating the reliability of the production process. This directly impacts various aspects of the entire industrial production process, including socio-economic benefits and equipment safety.

[0003] Capacitance tomography (CMT) is a process imaging technique used to study the distribution of substances with different dielectric constants within a field. A CMT system for detecting gas-liquid two-phase flow patterns applies excitation to electrode plates on the outer wall of a pipe via external circuitry and measures the capacitance between each electrode plate. Based on the measured capacitance values, an image reconstruction algorithm is established to ultimately obtain a two-dimensional spatial distribution map of the gas and liquid phases within the pipe.

[0004] However, the total number of pixels in the reconstructed image of a capacitance tomography system typically reaches thousands, far exceeding the number of capacitance values ​​that the system can acquire during measurement. Furthermore, capacitance tomography systems exhibit soft-field characteristics, meaning that even small disturbances in the measured gas-liquid two-phase flow can significantly impact the reconstructed image. Therefore, traditional capacitance tomography image reconstruction algorithms suffer from low accuracy and poor reconstruction results for two-dimensional images of gas-liquid two-phase flow patterns. Summary of the Invention

[0005] The present invention aims to solve the above-mentioned technical problems and provides a method for reconstructing two-dimensional images of gas-liquid two-phase flow patterns based on a capacitance tomography system, which aims to improve the accuracy of reconstructing two-dimensional images of gas-liquid two-phase flow patterns using a capacitance tomography system.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A method for reconstructing two-dimensional images of gas-liquid two-phase flow patterns based on a capacitance tomography system includes the following steps:

[0008] Step 1: Establish a three-dimensional geometric model of the capacitance tomography system for detecting flow patterns in gas-liquid two-phase flow, and obtain the capacitance vectors between electrode pairs of the capacitance tomography system under different flow patterns as the original sample set through numerical simulation.

[0009] Furthermore, the dielectric constant serves as a key physical parameter in gas-liquid two-phase flow.

[0010] Step 2: Calculate the dielectric constant sensitivity vector between different electrode pairs of the capacitance tomography system used for gas-liquid two-phase flow pattern detection. Based on the characteristics of the dielectric constant sensitivity vector between different electrode pairs of the capacitance tomography system used for gas-liquid two-phase flow pattern detection, the original sample in Step 1 is dimensionality reduced, and the dimensionality-reduced data is used as clustering samples.

[0011] Furthermore, the gas-liquid two-phase flow region of the capacitance tomography system containing the medium is divided into n regions, and the region location information is retained. The dielectric constant sensitivity vector is calculated according to formula (1):

[0012]

[0013] In the formula S ij (e) represents the sensitivity of the e-th test unit between the i-th and j-th electrodes; Let ε represent the dielectric constant of the e-th element within the tube. h Furthermore, the dielectric constant of the other unit materials is ε. l The capacitance value between the i-th and j-th electrodes; These represent the contents of the tube filled with a dielectric constant of ε. h ,ε l The capacitance between the i-th and j-th electrodes; ζ(e) represents the correction factor related to the area of ​​the e-th unit;

[0014] Furthermore, for a capacitance tomography system with m electrodes, m(m-1) / 2 electrode combinations can be obtained, and m(m-1) / 2 capacitance values ​​can be obtained during measurement; a pair of electrode combinations can obtain a one-dimensional sensitivity vector of length n, which is rearranged according to the preserved regional location information and converted into a sensitivity RGB image in RGB format;

[0015] Furthermore, for a single sample in the original sample set in step 1, based on the sensitivity RGB map, the capacitance values ​​between electrode pairs with similar shapes in the high-sensitivity region are extracted to form a capacitance vector group, and the 1-norm value of all possible capacitance vector groups is calculated.

[0016] Furthermore, create cluster sample vectors by adding the 1-norm values ​​of all possible capacitance vector groups to the created cluster sample vectors.

[0017] Furthermore, create clustered sample vectors for all samples in the original sample set from step 1 to create the clustered sample set;

[0018] Step 3: Perform pattern recognition on the clustered samples in Step 2 using a density-based clustering algorithm, and divide the original samples in Step 1 according to the clustering results to obtain the training sample set;

[0019] Furthermore, the original sample set is divided into three training sample sets: In a gas-liquid two-phase flow, when the phase with a lower dielectric constant surrounds the phase with a higher dielectric constant, the capacitance vector between the electrode pairs of the capacitance tomography system is used as the first training sample set; in a gas-liquid two-phase flow, when the phase with a higher dielectric constant surrounds the phase with a lower dielectric constant, the capacitance vector between the electrode pairs of the capacitance tomography system is used as the second training sample set; in a gas-liquid two-phase flow, when the two phases are arranged in parallel, the capacitance vector between the electrode pairs of the capacitance tomography system is used as the third training sample set.

[0020] Step 4: Using the different training sample sets from Step 3, build different fully connected deep learning network models to output the original two-dimensional image data of the gas-liquid two-phase flow pattern.

[0021] Furthermore, the activation functions, the number of neurons in each layer, and the data augmentation methods of the deep fully connected neural network are defined separately.

[0022] Step 5: Further process the original two-dimensional image data of the gas-liquid two-phase flow pattern to complete the two-dimensional image reconstruction of the gas-liquid two-phase flow pattern;

[0023] Furthermore, an appropriate threshold is selected for binarization of the original two-dimensional image data of the gas-liquid two-phase flow pattern. The data after binarization contains only two elements: 0 and 1.

[0024] Finally, the original two-dimensional image data of the gas-liquid two-phase flow pattern is converted into an RGB image. Areas with a pixel value of 0 are represented in blue, indicating a medium with a relatively high permittivity; areas with a pixel value of 1 are represented in red, indicating a medium with a relatively low permittivity.

[0025] By adopting the above technical solution, the present invention has the following advantages:

[0026] Based on the dielectric constant sensitivity vector characteristics between electrode pairs in the capacitance tomography system used for gas-liquid two-phase flow pattern detection, the capacitance signals acquired during the measurement of the capacitance tomography system are grouped, and the information of each group is characterized by the 1-norm, which realizes data dimensionality reduction and reduces the pressure of cluster analysis. The fully connected deep learning network model is built to realize the reconstruction of two-dimensional images of gas-liquid two-phase flow patterns. Compared with traditional algorithms, this method is beneficial to improving the accuracy of two-dimensional image reconstruction of gas-liquid two-phase flow patterns. Attached Figure Description

[0027] Figure 1 This is a flowchart of a method for reconstructing two-dimensional images of gas-liquid two-phase flow patterns based on a capacitance tomography system.

[0028] Figure 2A schematic diagram showing the electrode numbering of a capacitance tomography system;

[0029] Figure 3 This is a sensitivity distribution diagram of the dielectric constant of a capacitance tomography system.

[0030] Figure 4 This is a diagram of a fully connected deep neural network structure.

[0031] Figure 5 This is a reconstruction of the two-dimensional image of the gas-liquid two-phase flow pattern. Detailed Implementation

[0032] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] A method for reconstructing a two-dimensional image of a gas-liquid two-phase flow pattern includes the following steps:

[0034] Step 1: Establish a three-dimensional geometric model of the capacitance tomography system for detecting flow patterns in gas-liquid two-phase flow, and obtain the capacitance vectors between electrode pairs of the capacitance tomography system under different flow patterns as the original sample set through numerical simulation.

[0035] In step 1, a six-electrode plate capacitance tomography system was constructed, with the electrodes numbered clockwise. The results are as follows: Figure 2 As shown; in the gas-liquid two-phase flow, the liquid phase is set to water with a relative permittivity of 78; the gas phase is set to air with a relative permittivity of 1.

[0036] Step 2: Calculate the dielectric constant sensitivity vector between different electrode pairs of the capacitance tomography system used for gas-liquid two-phase flow pattern detection. Based on the characteristics of the dielectric constant sensitivity vector between different electrode pairs of the capacitance tomography system used for gas-liquid two-phase flow pattern detection, the original sample in Step 1 is dimensionality reduced, and the dimensionality-reduced data is used as clustering samples.

[0037] Step 2.1: Divide the medium-existing area into 3043 regions, retain the region location information, and calculate the sensitivity matrix according to formula (1);

[0038]

[0039] In the formula S ij (e) represents the sensitivity of the e-th test unit between the i-th and j-th electrodes; Let ε represent the dielectric constant of the e-th element within the tube.h Furthermore, the dielectric constant of the other unit materials is ε. l The capacitance value between the i-th and j-th electrodes; These represent the contents of the tube filled with a dielectric constant of ε. h ,ε l The capacitance between the i-th and j-th electrodes; ζ(e) represents the correction factor related to the area of ​​the e-th unit;

[0040] Step 2.2: For a capacitance tomography system with 6 electrodes, 15 electrode combinations and sensitivity vectors between each electrode combination can be obtained. During measurement, 15 capacitance values ​​can be obtained. A pair of electrode combinations can obtain a one-dimensional dielectric constant sensitivity vector with a length of 3043. The one-dimensional dielectric constant sensitivity vector with a length of 3043 is rearranged according to the preserved regional position information and converted into a sensitivity RGB image in RGB format.

[0041] Step 2.3: For individual samples in the original sample set, based on the sensitivity RGB image, electrode combinations with similar high-sensitivity region shapes are grouped together to form a new capacitance vector. The 1-norm value of each group is calculated. The high-sensitivity region shapes in the 15 dielectric constant sensitivity RGB images show three representative forms, as shown in the results. Figure 3 As shown, the lowest dielectric constant sensitivity is 0, and the highest dielectric constant sensitivity is 1. The area enclosed by the dashed line and the outer contour is the high-sensitivity area, where the dielectric constant sensitivity value is greater than 0.5.

[0042] Step 2.4: Create cluster sample vectors by adding the 1-norm values ​​of all possible capacitance vector groups to the created cluster sample vectors.

[0043] Step 2.5: Repeat steps 2.3 and 2.4 for all samples in the original sample set in step 1 to create the clustered sample set;

[0044] In step 2.3, the electrode assembly can be divided into three groups, as shown below.

[0045]

[0046] Step 3: Perform pattern recognition on the clustered samples in Step 2 using a density-based clustering algorithm, and divide the original samples in Step 1 according to the clustering results to obtain the training sample set;

[0047] In step 3, the original sample set is divided into three training sample sets: in a gas-liquid two-phase flow where the phase with lower dielectric constant surrounds the phase with higher dielectric constant, the capacitance vector between the electrode pairs of the capacitance tomography system is used as the first training sample set; in a gas-liquid two-phase flow where the phase with higher dielectric constant surrounds the phase with lower dielectric constant, the capacitance vector between the electrode pairs of the capacitance tomography system is used as the second training sample set; and in a gas-liquid two-phase flow where the two phases are arranged in parallel, the capacitance vector between the electrode pairs of the capacitance tomography system is used as the third training sample set.

[0048] Step 4: Using the different training sample sets from Step 3, build different fully connected deep learning network models to output the original two-dimensional image data of the gas-liquid two-phase flow pattern.

[0049] In step 4, three fully connected deep learning network models are defined, such as... Figure 4 As shown: 1) A fully connected deep neural network contains five hidden layers, each with the activation function Tanh, and the number of neurons in each layer is 45, 135, 405, 1215, and 3043 respectively; 2) A fully connected deep neural network contains five hidden layers, the first four with the activation function Tanh, and the fifth with the activation function Sigmoid, and the number of neurons in each layer is 45, 135, 405, 1215, and 3043 respectively; 3) A fully connected deep neural network contains five hidden layers, each with the activation functions ReLU, Tanh, Tanh, Tanh, and Sigmoid, and the number of neurons in each layer is 60, 240, 960, 3840, and 3043 respectively; the mathematical expressions for ReLU, Tanh, and Sigmoid are shown below:

[0050]

[0051] According to formula (2), normal distribution noise is added to the training samples to achieve data augmentation and enhance the generalization ability of fully connected deep neural networks;

[0052]

[0053] f δ (x i f(x) represents the data after noise has been added. i ) represents the original data, and δ represents the relative error level. These are normally distributed random numbers with a mean of 0 and a variance of 1; the relative error levels for the first, second, and third fully connected deep neural networks are chosen to be 0.15, 0.1, and 0.8, respectively.

[0054] Step 5: Further process the original two-dimensional image data of the gas-liquid two-phase flow pattern to complete the two-dimensional image reconstruction of the gas-liquid two-phase flow pattern;

[0055] In step 5, the original two-dimensional image data of the gas-liquid two-phase flow pattern is binarized. The binarized data contains only 0 and 1 elements. Areas with a pixel value of 0 are represented in blue, indicating a medium with a relatively high permittivity (water in this embodiment); areas with a pixel value of 1 are represented in red, indicating a medium with a relatively low permittivity (water in this embodiment). Finally, the two-dimensional image data of the gas-liquid two-phase flow pattern is converted into an RGB format two-dimensional RGB image of the gas-liquid two-phase flow pattern. Using a high-speed camera to capture an actual gas-liquid two-phase flow pattern image as a reference, the reconstructed two-dimensional image of the gas-liquid two-phase flow pattern is shown below. Figure 5 As shown. Figure 5 The target image is a two-dimensional distribution of the actual gas-liquid two-phase flow on the AA section, where the shaded area represents the pipe, which will not be used as the target for reconstruction.

[0056] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0057] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for reconstructing a two-dimensional image of a gas-liquid two-phase flow pattern, characterized in that, Includes the following steps: Step 1: Establish a three-dimensional geometric model of the capacitance tomography system for detecting gas-liquid two-phase flow patterns, obtain the capacitance vector between electrode pairs of the capacitance tomography system under different flow patterns through numerical simulation, and obtain the original sample set; Step 2: Calculate the dielectric constant sensitivity vector between different electrode pairs in the capacitance tomography system. Based on the characteristics of the dielectric constant sensitivity vector between different electrode pairs in the capacitance tomography system, perform dimensionality reduction on the original samples in Step 1. The data after dimensionality reduction is used as clustering samples. Step 3: Perform pattern recognition on the clustered samples in Step 2 using a density-based clustering algorithm, and divide the original samples in Step 1 according to the clustering results to obtain the training sample set; Step 4: Using the different training sample sets from Step 3, build different fully connected deep learning network models to output the original two-dimensional image data of the gas-liquid two-phase flow pattern. Step 5: Further process the original two-dimensional image data of the gas-liquid two-phase flow pattern to complete the two-dimensional image reconstruction of the gas-liquid two-phase flow pattern; In step 2: The gas-liquid two-phase flow region of the capacitance tomography system is divided into n regions, and the region location information is retained; for a capacitance tomography system with m electrodes, the following can be obtained: This electrode combination method allows for measurement to obtain... Each capacitance value; each electrode pair can obtain a one-dimensional sensitivity vector of length n, which is rearranged according to the preserved regional location information and converted into a sensitivity RGB image in RGB format; For a single sample in the original sample set in step 1, based on the sensitivity RGB image, extract the capacitance values ​​between electrode pairs with similar shapes in the high-sensitivity region to form a capacitance vector group, and calculate the 1-norm value of all possible capacitance vector groups; create a cluster sample vector, and add the 1-norm value of all possible capacitance vector groups to the created cluster sample vector; create a cluster sample vector for all samples in the original sample set in step 1, thus realizing the production of the cluster sample set.

2. The method for reconstructing a two-dimensional image of a gas-liquid two-phase flow pattern according to claim 1, characterized in that, In step 3: Using the clustered sample set from step 2 as input, a density-based clustering algorithm is applied for pattern recognition, dividing the original sample set into three training sample sets: In a gas-liquid two-phase flow, when the phase with a lower dielectric constant surrounds the phase with a higher dielectric constant, the capacitance vector between the electrode pairs of the capacitance tomography system is used as the first training sample set; in a gas-liquid two-phase flow, when the phase with a higher dielectric constant surrounds the phase with a lower dielectric constant, the capacitance vector between the electrode pairs of the capacitance tomography system is used as the second training sample set; in a gas-liquid two-phase flow, when the two phases are arranged in parallel, the capacitance vector between the electrode pairs of the capacitance tomography system is used as the third training sample set.

3. The method for reconstructing a two-dimensional image of a gas-liquid two-phase flow pattern according to claim 2, characterized in that, In step 4: Based on the three training sample sets obtained in step 3, fully connected deep neural networks are built and trained respectively, and the activation functions, number of neurons in each layer, and data augmentation methods of each layer of the deep fully connected neural network are defined respectively.

Images