CNN network discrimination mechanism-based gas-liquid two-phase flow image reconstruction method and system

Abstract

The invention discloses a gas-liquid two-phase flow image reconstruction method and system based on a CNN network discrimination mechanism, and the method comprises the steps: S1, building a three-dimensional model of a capacitive sensor, simulating the potential distribution of the cross section of a pipeline, and obtaining a capacitive sensitive field matrix; s2, capacitance values of the capacitive sensors in the capacitive sensor physical model are obtained, and a capacitance matrix is formed; s3, based on the capacitance sensitive field matrix and the capacitance matrix, performing image reconstruction by using an LBP algorithm; s4, judging the flow pattern of the reconstructed image; if the two circular streams exist, the LBP algorithm is switched to the Landweber algorithm for image reconstruction; otherwise, directly outputting an image result reconstructed by the LBP algorithm; and S5, segmenting the final image result, and calculating the volume content ratio of the gas phase to the liquid phase. The method has the advantages of improving image reconstruction precision, facilitating accurate calculation of the liquid-gas ratio and the like.

Description

Technical Field

[0001] The present invention mainly relates to the technical field of two-phase flow detection, and particularly relates to a gas-liquid two-phase flow image reconstruction method and system based on a CNN network discrimination mechanism. Background Art

[0002] Two-phase flow is widely used in fields such as aerospace, aviation, petrochemical, etc. The real-time detection of two-phase flow fluids is of great significance for the safe operation and maintenance of equipment and the inversion of environmental data.

[0003] The existing technical solutions are as follows: capacitance sensors, resistance sensors or other sensors are arranged in the measured pipeline to measure the physical parameters of the fluid, and then based on these parameters, the flow pattern image of the pipeline cross-section is reconstructed. The liquid-gas ratio of the cross-section image is calculated through an image segmentation algorithm to complete the monitoring of fluid parameters. Among them, capacitance sensors are the most widely used, which have the advantages of convenient measurement, strong anti-interference ability, and no radioactivity. The capacitance imaging principle is that the spatial distribution of two fluids in the pipeline cross-section (i.e., the spatial distribution of dielectric constants) will be reflected in the capacitance measurement values of the capacitance sensor. Image reconstruction is to inversely solve the image pixel matrix according to the measured capacitance values and considering the influence of the dielectric constant spatial distribution on the capacitance values (i.e., finding the sensitive field distribution). Image reconstruction algorithms mainly include direct methods and iterative methods. The direct methods include the Linear Back Projection (LBP) method and the Tikhonov regularization algorithm. The iterative algorithms include the New-Raphson algorithm and the Landweber algorithm.

[0004] The existing LBP algorithm has a small amount of calculation, and the imaging time is usually at the ms level. However, for complex flow patterns, such as two-circle flow, the imaging accuracy is poor. The Landweber algorithm is an iterative algorithm that approximates the true result through local linearization and multiple iterations. Therefore, the imaging accuracy is high, but the calculation amount is large, the imaging speed is slow, and the imaging time is at the s level.

[0005] In practical engineering applications, the online detection of two-phase flow patterns is a dynamic process, which requires high real-time and synchronization of imaging. It is necessary to use sensors to collect fluid data in real time, and then reconstruct the image of the two-phase flow fluid morphology to complete the detection of the flow pattern. The existing technology uses a single imaging algorithm to reconstruct the flow pattern. Since each algorithm cannot balance the imaging accuracy and calculation time, a good imaging quality cannot be obtained. Summary of the Invention

[0006] Aiming at the technical problems existing in the prior art, the present invention provides a gas-liquid two-phase flow image reconstruction method and system based on a CNN network discrimination mechanism, which can improve the image reconstruction accuracy and facilitate the accurate calculation of the liquid-gas ratio.

[0007] To solve the above technical problems, the technical solution proposed by the present invention is as follows:

[0008] A method for reconstructing gas-liquid two-phase flow images based on a CNN network discrimination mechanism, comprising the steps of:

[0009] S1. Calculate the sensitive field matrix of computational imaging: Establish a three-dimensional model of a capacitance sensor, evenly distribute multiple electrodes on the outer wall of the pipeline, any two electrodes form a capacitance sensor, simulate the potential distribution of the pipeline cross-section, and obtain the capacitance sensitive field matrix;

[0010] S2. Collect the capacitance matrix: Prepare a physical model of the capacitance sensor, obtain the capacitance values of the capacitance sensors in the physical model of the capacitance sensor, and form a capacitance matrix;

[0011] S3. Preliminary image reconstruction: Based on the capacitance sensitive field matrix and the capacitance matrix, use the LBP algorithm for image reconstruction to obtain a reconstructed image;

[0012] S4. Flow pattern discrimination: Use the trained CNN network to discriminate the flow pattern of the reconstructed image; if the discrimination result is two-circle flow, switch the LBP algorithm to the Landweber algorithm to re-perform image reconstruction to obtain the image result reconstructed by the Landweber algorithm; otherwise, directly output the image result reconstructed by the LBP algorithm;

[0013] S5. Image segmentation and volume ratio calculation: Segment the final image result and calculate the volume content ratio of the gas phase to the liquid phase.

[0014] Preferably, in step S1, the potential distribution of the pipeline cross-section is simulated by COMSOL, and the capacitance sensitive field matrix S is calculated ij , and the calculation formula is:

[0015]

[0016] In the formula is the spatial distribution of the potential when the spatial medium is ε, the i-th electrode applies an excitation voltage V, and the remaining electrodes are grounded; is the spatial distribution of the potential when the spatial medium is ε, the j-th electrode applies an excitation voltage V, and the remaining electrodes are grounded.

[0017] Preferably, the specific process of obtaining the capacitance sensitive field matrix is as follows:

[0018] Apply a 1v voltage excitation to the i-th electrode, and ground the remaining electrodes to obtain the spatial distribution of the potential when the i-th electrode acts alone; similarly, obtain the spatial distribution of the potential when the j-th electrode acts alone; then multiply the spatial potential distribution obtained by the i-electrode by the spatial potential distribution obtained by the j-electrode to obtain the sensitive field matrix S between the i-j electrodes ij .

[0019] Preferably, in step S4, the CNN network structure includes 7 layers: an input layer, a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a fully connected layer, and an output layer; the convolutional kernel size of the first convolutional layer is 5x5, and the number of convolutional kernels is 6; the first pooling layer uses max pooling with a window size of 2x2; the convolutional kernel size of the second convolutional layer is 5x5, and the number of convolutional kernels is 16; the second pooling layer uses max pooling with a window size of 2x2.

[0020] Preferably, two-circle flow image data is obtained through experimental testing, finite element simulation, or data augmentation methods, and training samples are constructed therefrom to train the CNN network to obtain a trained CNN network.

[0021] Preferably, in step S5, the region growing method is used to segment the final reconstructed image.

[0022] Preferably, 8 electrodes are evenly distributed on the outer wall of the pipeline, and any two electrodes form a capacitance sensor, with a total of 28 capacitance sensors.

[0023] The present invention also discloses a computer program product, including a computer program, and the steps of the above-mentioned method are executed when the computer program is run by a processor.

[0024] The present invention further discloses a computer-readable storage medium, on which a computer program is stored, and the steps of the above-mentioned method are executed when the computer program is run by a processor.

[0025] The present invention also discloses a gas-liquid two-phase flow image reconstruction system based on a CNN network discrimination mechanism, including a memory and a processor connected to each other, a computer program is stored on the memory, and the steps of the above-mentioned method are executed when the computer program is run by a processor.

[0026] Compared with the prior art, the advantages of the present invention are as follows:

[0027] The present invention can automatically match the image reconstruction algorithm according to the flow pattern, improve the image reconstruction accuracy, facilitate the accurate calculation of the liquid-gas ratio, improve the imaging efficiency, reduce the computational amount of image reconstruction to shorten the imaging time, and finally ensure the real-time performance and accuracy of the two-phase flow fluid morphology detection, so as to meet the two-phase flow detection requirements in complex industrial scenarios; the capacitance sensitive field is obtained through the calculation of the electric potential distribution, which simplifies the calculation process.

[0028] In the aspect of two-phase flow pattern image reconstruction, the present invention combines the advantages of the LBP algorithm and the Landweber algorithm, improves the accuracy of flow pattern imaging, facilitates image segmentation, and enables accurate calculation of the liquid-gas ratio. For the two-circle flow pattern, the Landweber algorithm is adopted, and for other flow patterns, the LBP algorithm is used. Overall, the computational amount is reduced, the imaging time is shortened, and the imaging quality and efficiency are improved. In addition, a flow pattern discrimination mechanism is added to enhance the automation level of flow pattern image reconstruction. Description of the Drawings

[0029] Figure 1 It is a schematic diagram of common flow patterns in the present invention.

[0030] Figure 2 It is a schematic diagram of the two-phase flow pattern detection process in the present invention.

[0031] Figure 3 It is a flowchart of the gas-liquid two-phase flow image reconstruction method based on the CNN network discrimination mechanism of the present invention.

[0032] Figure 4 It is a three-dimensional model diagram of the capacitance sensor in the present invention.

[0033] Figure 5 It is a spatial distribution diagram of the electric potential when the i-th electrode acts alone in the present invention. Detailed Embodiments

[0034] The present invention will be further described below in conjunction with the accompanying drawings of the specification and specific embodiments.

[0035] As Figure 3 shown, the gas-liquid two-phase flow image reconstruction method based on the CNN network discrimination mechanism provided by the embodiment of the present invention includes the steps of:

[0036] S1. Calculate the sensitive field matrix of imaging: Establish a three-dimensional model of the capacitance sensor, evenly distribute multiple electrodes on the outer wall of the pipeline, and any two electrodes form a capacitance sensor. Through simulation software, simulate the electric potential distribution of the pipeline cross-section to obtain the sensitive field matrix;

[0037] Specifically, establish a three-dimensional model of the capacitance sensor, evenly arrange n electrodes on the outer wall of the pipeline, and each two electrodes form a capacitance sensor, with a total of capacitance sensors; as Figure 4 shown, evenly distribute 8 electrodes on the outer wall of the pipeline, and any two electrodes form a capacitance sensor, with a total of 28 capacitance sensors. Of course, in other embodiments, the number of electrodes can also be one of 16, 32, etc.

[0038] As Figure 5As shown, the potential distribution of the pipeline cross-section is simulated by COMSOL, and the capacitance sensitive field matrix S is calculated ij , and the calculation formula is:

[0039]

[0040] In the formula is the spatial distribution of the potential when the spatial medium is ε, the excitation voltage V is applied to the i-th electrode plate, and the remaining electrode plates are grounded as above

[0041] Specifically, apply a 1V voltage excitation to the i-th electrode, and ground the remaining electrodes to obtain the spatial distribution of the potential when the i-th electrode acts alone; similarly, the spatial distribution of the potential when the j-th electrode acts alone can be obtained; then multiply the spatial potential distribution obtained from the i-th electrode by the spatial potential distribution obtained from the j-th electrode to obtain the sensitive field matrix between the i-j electrodes ij .

[0042] In addition, the working principle of the capacitance sensor is closely related to the potential distribution. By calculating the potential distribution, the interaction between the capacitance sensor and different substances (such as gas phase and liquid phase) can be simply described. Compared with directly using complex hydrodynamic models or electromagnetic field models to calculate the sensitive field, the potential distribution method transforms the problem into relatively simple mathematical calculations by solving the static distribution problem of the electric field. This method avoids directly modeling the detailed distribution of fluids or substances, simplifying the system modeling and calculation process

[0043] In the potential distribution method, through finite element simulation or approximate models, the potential distribution when a voltage is applied to each electrode plate is calculated, and the sensitive field matrix is obtained by superimposing the potential distributions of different electrode plates. This method effectively reduces the amount of calculation required to solve each physical phenomenon one by one in traditional complex calculations. The potential distribution method utilizes the superposition principle, effectively integrating the interactions between electrode plates and avoiding complex simulations that require considering each tiny variable

[0044] Using the potential distribution method, the potential distribution and sensitive field matrix of the capacitance sensor can be calculated in advance through simulation tools (such as finite element software like COMSOL). In this way, most of the required sensitive field data can be obtained through simulation during the design stage, reducing a large amount of physical experiment verification work. Simulation calculations can quickly predict the potential distribution of sensors with different configurations without the need for cumbersome experimental measurements

[0045] The electric potential distribution method is particularly suitable for the calculation of large-scale capacitance sensor arrays, such as the multiple electrode sensors (28 in total) used in the present invention. Through the electric potential distribution method, the sensitive field matrix between each pair of electrodes can be efficiently obtained without the need to separately calculate the response or configuration of each sensor, significantly improving the calculation efficiency. This method can reduce manual calculation and point-by-point analysis through matrix calculation and vectorization operations, thus enhancing the processing efficiency.

[0046] The mathematical model of the electric potential distribution method has high repeatability and consistency. Under different experimental conditions and system configurations, the electric potential distribution method can stably generate an accurate sensitive field matrix. This consistency enables the system to quickly and reliably calculate the sensitive field matrix when dealing with different flow patterns or different sensor configurations, reducing human errors and uncertainties.

[0047] Summary: The electric potential distribution method simplifies the complex physical model, reduces the amount of calculation, optimizes the simulation process, and improves the calculation efficiency, making the process of calculating the sensitive field matrix more efficient and accurate. This enables the system to respond quickly, provide high-quality capacitance sensor data processing, and greatly reduce the calculation complexity when applied to a large-scale sensor array.

[0048] Of course, in other embodiments, the capacitance sensitive field matrix S can also be obtained through the numerical calculation method. ij .

[0049] S2. Collect the capacitance matrix: As Figure 2 shown, prepare the physical model of the capacitance sensor in this embodiment, and obtain the capacitance values of the capacitance sensor through a capacitance acquisition device to form a capacitance matrix C. ij ;

[0050] Specifically, the capacitance values of capacitance sensors are collected through the acquisition device to obtain the capacitance matrix C. ij ;

[0051] Since the capacitance tomography process is:

[0052] SG = C;

[0053] where S is the sensitive field matrix, G is the normalized medium distribution image matrix, and C is the normalized capacitance matrix.

[0054] Therefore, image reconstruction is an inverse problem of the imaging process, that is, the image matrix G can be obtained by inversely solving through the capacitance matrix C and the sensitive field matrix S.

[0055] S3. Preliminary image reconstruction: Based on the LBP algorithm, perform image reconstruction on the capacitance matrix to obtain a reconstructed image.

[0056] The principle of the LBP algorithm is as follows:

[0057] G = S'C

[0058] Where S' is the transpose of the sensitive field matrix S.

[0059] S4. Flow pattern discrimination: Use the trained CNN network to discriminate the reconstructed image. If the discrimination result is two-circle flow, switch the reconstruction algorithm to the Landweber algorithm to re-perform image reconstruction to obtain the final higher imaging accuracy and quality; otherwise, output the image result reconstructed by the LBP algorithm. Specifically, as Figure 1 shown, the common two-phase flow patterns in industrial practice mainly include: two-circle flow, core flow, laminar flow, circulation flow, etc.

[0060] Since the LBP has a fast imaging speed, but has a poor imaging accuracy for the two-circle flow pattern. Although the Landweber algorithm has a slow imaging speed, it has a high imaging accuracy for the two-circle flow pattern. Therefore, integrating the advantages of the LBP algorithm and the Landweber algorithm can make the final imaging effect of the system reach the optimal.

[0061] The principle of the Landweber algorithm is as follows:

[0062] G k+1 = P[G k + αS T (C - S k ), k ≥ 0

[0063]

[0064] Where k is the number of iterations, G k is the image after k iterations, P is the threshold function, S T is the transpose matrix of the sensitive field matrix, α is the iteration factor, α = 2 / λ max , λ max is the maximum eigenvalue of the matrix S T S.

[0065] The CNN network structure includes 7 layers: input layer, the first convolutional layer (the convolutional kernel size is 5x5, and the number of convolutional kernels is 6), the first pooling layer (using max pooling, the window size is 2x2), the second convolutional layer (the convolutional kernel size is 5x5, and the number of convolutional kernels is 16), the second pooling layer (using max pooling, the window size is 2x2), the fully connected layer and the output layer. Of course, in addition to the above structure, the CNN network can also adopt other network structures, and the CNN network can also be replaced by other networks with image recognition functions, such as one of ResNet (Residual Network), VGGNet (Visual GeomrtryGroup), etc.

[0066] A large amount of two - circle flow image data is obtained through methods such as experimental testing, finite - element simulation, and data augmentation, and training samples are constructed therefrom to train the CNN network, enabling the CNN network to have high - precision recognition performance for various two - circle flow patterns.

[0067] S5. Image segmentation and volume ratio calculation: The region - growing method is used to segment the final reconstructed image, and the volume content ratio of the gas phase and the liquid phase is calculated. Of course, in other embodiments, one of the edge - detection algorithm, threshold method, watershed algorithm, clustering method, or Graph Cut method can also be used to replace the region - growing method.

[0068] The method of the present invention can automatically match the image reconstruction algorithm according to the flow pattern, improve the image reconstruction accuracy, facilitate the accurate calculation of the liquid - gas ratio, improve the imaging efficiency, reduce the computational amount of image reconstruction to shorten the imaging time, and ultimately ensure the real - time and accurate detection of the two - phase flow fluid morphology, thereby meeting the two - phase flow detection requirements in complex industrial scenarios; the capacitance - sensitive field is obtained through the calculation of the electric - potential distribution, simplifying the calculation process.

[0069] In the aspect of two - phase flow pattern image reconstruction, the present invention makes the advantages of the LBP algorithm and the Landweber algorithm complementary, improves the flow - pattern imaging precision, facilitates image segmentation and the accurate calculation of the liquid - gas ratio; for the two - circle flow pattern, the Landweber algorithm is adopted, and for other flow patterns, the LBP algorithm is adopted, globally reducing the computational amount, shortening the imaging time, and improving the imaging quality and efficiency; a flow - pattern discrimination mechanism is added, improving the automation level of flow - pattern image reconstruction.

[0070] The present invention also discloses a computer program product, including a computer program, and the steps of the above - described method are executed when the computer program is run by a processor. The present invention further discloses a computer - readable storage medium, on which a computer program is stored, and the steps of the above - described method are executed when the computer program is run by a processor. The present invention also discloses a gas - liquid two - phase flow image reconstruction system based on the CNN network discrimination mechanism, including a memory and a processor connected to each other, a computer program is stored on the memory, and the steps of the above - described method are executed when the computer program is run by the processor. The products, media, and systems of the present invention, corresponding to the above - described method, have the same advantages as those of the above - described method.

[0071] The implementation of all or part of the processes in the above-described embodiment methods of the present invention can also be completed by hardware related to computer program instructions. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, the steps of the above-described method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file, or some intermediate form, etc. The computer-readable storage medium includes: any entity or device capable of carrying computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disc, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc. The memory is used to store computer programs and / or modules. The processor realizes various functions by running or executing the computer programs and / or modules stored in the memory, and by invoking the data stored in the memory. The memory can include high-speed random access memory, and can also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one magnetic disk storage device, flash device, or other volatile solid-state storage devices, etc.

[0072] The above are only the preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions falling within the idea of the present invention belong to the protection scope of the present invention. It should be noted that for those of ordinary skill in the art in this technical field, several improvements and refinements made without departing from the principle of the present invention should be regarded as within the protection scope of the present invention.

Claims

1. A gas-liquid two-phase flow image reconstruction method based on CNN network discrimination mechanism, characterized in that: Includes steps: S1. Calculate the sensitive field matrix of imaging: Establish a three-dimensional model of the capacitive sensor, evenly distribute multiple electrodes on the outer wall of the pipeline, any two electrodes constitute a capacitive sensor, simulate the potential distribution of the pipeline cross section, and obtain the capacitive sensitive field matrix; S2, collecting capacitance matrix: preparing a physical model of the capacitance sensor, obtaining the capacitance values ​​of the capacitance sensors in the physical model of the capacitance sensor, and forming a capacitance matrix; S3, preliminary image reconstruction: based on the capacitance sensitive field matrix and the capacitance matrix, the LBP algorithm is used to reconstruct the image to obtain a reconstructed image; S4, flow type discrimination: Use the trained CNN network to discriminate the flow type of the reconstructed image; if the discrimination result is two-circle flow, switch the LBP algorithm to the Landweber algorithm to reconstruct the image and obtain the image result reconstructed by the Landweber algorithm; otherwise, directly output the image result reconstructed by the LBP algorithm; S5. Image segmentation and volume ratio calculation: Segment the final image result and calculate the volume content ratio of the gas phase and the liquid phase.

2. The gas-liquid two-phase flow image reconstruction method based on the CNN network discrimination mechanism according to claim 1 is characterized in that: In step S1, the potential distribution of the pipeline cross section is simulated by COMSOL, and the capacitance sensitive field matrix S is calculated. ij , the calculation formula is: In the formula is the spatial distribution of the potential when the space medium is ε, the excitation voltage V is applied to the i-th electrode, and the other electrodes are grounded; It is the spatial distribution of the electric potential when the spatial medium is ε, the excitation voltage V is applied to the jth electrode, and the other electrodes are grounded.

3. The gas-liquid two-phase flow image reconstruction method based on the CNN network discrimination mechanism according to claim 2 is characterized in that: The specific process of obtaining the capacitive sensitive field matrix is: Apply 1V voltage excitation to the i-th electrode, and ground the other electrodes to obtain the spatial distribution of the potential when the i-th electrode acts alone; similarly, obtain the spatial distribution of the potential when the j-th electrode acts alone; then multiply the spatial potential distribution obtained by the i-th electrode by the spatial potential distribution obtained by the j-th electrode to obtain the sensitive field matrix S between the ij electrodes ij .

4. The gas-liquid two-phase flow image reconstruction method based on the CNN network discrimination mechanism according to claim 1, 2 or 3, characterized in that: In step S4, the CNN network structure includes 7 layers: an input layer, a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a fully connected layer and an output layer; the convolution kernel size of the first convolutional layer is 5x5, and the number of convolution kernels is 6; the first pooling layer adopts maximum pooling, and the window size is 2x2; the convolution kernel size of the second convolutional layer is 5x5, and the number of convolution kernels is 16; the second pooling layer adopts maximum pooling, and the window size is 2x2.

5. The gas-liquid two-phase flow image reconstruction method based on the CNN network discrimination mechanism according to claim 4 is characterized in that: The two-circle flow image data are obtained through experimental testing, finite element simulation or data enhancement method, and the training samples are constructed to train the CNN network to obtain the trained CNN network.

6. The gas-liquid two-phase flow image reconstruction method based on the CNN network discrimination mechanism according to claim 1, 2 or 3, characterized in that: In step S5, the final reconstructed image is segmented using a region growing method.

7. The gas-liquid two-phase flow image reconstruction method based on the CNN network discrimination mechanism according to claim 1, 2 or 3, characterized in that: Eight electrodes are evenly distributed on the outer wall of the pipe. Any two electrodes constitute a capacitive sensor, and there are 28 capacitive sensors in total.

8. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are performed.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the computer program performs the steps of the method according to any one of claims 1 to 7.

10. A gas-liquid two-phase flow image reconstruction system based on a CNN network discrimination mechanism, comprising a memory and a processor connected to each other, wherein a computer program is stored in the memory, characterized in that: When the computer program is executed by a processor, the computer program performs the steps of the method according to any one of claims 1 to 7.

Images