Microcosmic visual photoetching model manufacturing method based on image processing

Abstract

The invention discloses a microcosmic visual photoetching model manufacturing method based on image processing. The design scheme comprises the following steps: identifying pores and a rock skeleton in a rock core casting body slice image by utilizing machine learning so as to extract the pores; generating a throat by using a watershed algorithm; extracting a throat by using a contour recognition algorithm; assigning the throat radius to the throat size distribution obtained by high-pressure mercury injection by using an expansion algorithm; combining pore and throat images, extracting pore and throat contours, converting the pore and throat contours into a geometric model, importing the geometric model into COMSOL to calculate porosity and permeability, and verifying physical properties of the model; and finally, connecting a photoetching machine, and manufacturing a microscopic visual photoetching model with a real pore throat structure. The model manufacturing method is programmed, simple and practical, the problem that the throat cannot be extracted from a slice image due to the compact rock core is avoided, meanwhile, the physical properties of the model and the real rock core are compared and verified by using a simulation means, the physical properties of the reservoir can be better represented, and a reliable basis is provided for a microscopic visualization experiment.

Description

Technical Field

[0001] The invention relates to a method for manufacturing a microscopic visualized photolithography model based on image processing, and belongs to the technical field of oil and gas field development. Background Art

[0002] At present, most of my country's oil fields have entered the secondary and tertiary oil recovery stages. Further research on the seepage and storage characteristics of multiphase fluids is of great significance for improving crude oil recovery. Microscopic visualization technology can intuitively see the fluid flow and distribution in the model and effectively characterize the fluid flow mechanism. It is an important means of studying the fluid flow mechanism. The main existing microscopic visualization models are: 1. Bead model, 2. Real core model, 3. Sand filling micro model, 4. Etching model

[0003] The bead-sandwich model uses two pieces of glass or plexiglass to hold glass beads to form a layered porous medium model. The disadvantage is that it is difficult to accurately control the change in pore size. The real core model is made by grinding the full-diameter core into thin slices, clamping it with glass, and bonding it with rubber. The disadvantages are long production time, short service life, unclear phenomena, and poor comparability. The sand-filled micromodel uses quartz sand particles glued to a glass sheet and embedded in a transparent rubber sleeve. The disadvantages are one-time use, large differences in pore structure, and large influence of human factors. The etching model is made into a pore network structure through photolithography mask and dry etching, and the glass is bonded to form a model. The disadvantage is that the process is complicated, and most pore network structure diagrams are extracted from cast thin slices, which makes it difficult to accurately reflect the structure and size of small throats, especially for unconventional reservoirs such as dense sandstone.

[0004] Existing microscopic visualization models are insufficient in representing the complex changes of real core characteristics, such as pore throat geometry and morphology, which limits the effectiveness and efficiency of experimental research. Therefore, a new method for making microscopic visualization models is urgently needed, so that the constructed model can be closer to the actual pore throat structure characteristics of the reservoir, and at the same time realize the programmatic construction process, which has simple, fast and practical characteristics to meet the needs of microscopic visualization experimental research. Summary of the invention

[0005] In view of the above problems, the present invention provides a method for making a microscopic visual lithography model based on image processing. The design method can restore the pore and throat structure of real rocks and has the characteristics of simplicity, repeatability and programming.

[0006] The present invention adopts the following technical solutions:

[0007] A method for making a microscopic visual lithography model based on image processing comprises the following steps:

[0008] S1. Image selection

[0009] Select a high-resolution microscope to take images of representative core cast thin sections to ensure that pores and rock skeletons can be clearly distinguished;

[0010] S2. Pore extraction

[0011] The image size is calibrated, and contrast enhancement and denoising are performed to improve the image quality and make it easier to distinguish between pores and rock skeletons. Then, the pores and rock skeleton parts in the image are manually annotated with different colors to generate a training data set. A suitable machine learning model for classification problems (Random Forest algorithm) is selected for training. The training data set is continuously increased to improve the generalization ability and accuracy of the model. Finally, the trained model is used to complete the segmentation of pores and rock skeletons in the casting thin section image, extract intergranular pores, dissolution pores and other pore spaces, and use the threshold segmentation method to obtain a binary image. The outlier removal algorithm is used to set a suitable pixel radius and threshold to remove isolated outliers in the image, and the pore radius distribution is calculated.

[0012] S3. Pore throat connectivity classification

[0013] The pore-throat space image after segmentation by watershed algorithm is connected to each pore space, and the connecting line is the throat;

[0014] S4. Extraction of throat position

[0015] Subtract the extracted pore space image from the image containing pore and throat information processed by the watershed algorithm to obtain an image containing only the throat;

[0016] S5. Throat size assignment

[0017] The throat size distribution probability function obtained by high-pressure mercury injection experiment is fitted to obtain the functional relationship between radius and distribution probability. Then, all throat contours in the throat binary image are detected by contour extraction algorithm. The fitted throat size distribution probability function is used for each detected throat contour. The width of the throat contour is randomly expanded by using the expansion algorithm to complete the random assignment of each throat radius and calculate the average pore-throat ratio.

[0018] S6. Image combination

[0019] By adding the gray values ​​of the throat image and the pore image after the value assignment, a binary image with the characteristics of the real core pore structure can be obtained.

[0020] S7. Model property calculation

[0021] The image obtained in S6 was converted into a DXF file using the ezdxf library in Python, and then imported into C0MSOL to build the model. The porosity was determined by dividing the pore throat volume by the total geometric volume, and the permeability was determined according to Darcy's law:

[0022]

[0023] where u is the Darcy or superficial velocity, k is the permeability, μ is the dynamic viscosity, is the pressure gradient;

[0024] With the help of the creep flow interface in COMSOL, the inlet, outlet, wall boundary conditions and fluid properties are set, and the flow rate at the outlet boundary is obtained through the variable spf.out1.Mflow. The model permeability is determined by Darcy's law, compared with the core physical properties, and the model is adjusted;

[0025] S8. Model making

[0026] Connect a photolithography machine, etch the pore throat network structure on the carrier glass through the photolithography machine, and then bond the side of the carrier glass with the pore throat network structure to the cover glass to form a microscopic visual photolithography model.

[0027] Preferably, in step S2, the image processing operations such as image size calibration, contrast enhancement, denoising, threshold segmentation method, etc. are implemented by ImageJ software, and the random forest algorithm is a machine learning algorithm for processing classification tasks, which is implemented by ImageJ software.

[0028] Preferably, in step S3, the outlier removal algorithm is a filtering method for removing outliers; the watershed algorithm is an image segmentation method for connecting pores to form throats, and these two algorithms are implemented by ImageJ software.

[0029] Preferably, in step S5, the throat size assignment method is implemented by Python programming, comprising the following steps:

[0030] S51, loading the laryngeal grayscale image extracted in S4 and performing binarization processing;

[0031] S52. Input throat size distribution data based on high-pressure mercury injection data;

[0032] S53, using the RETRTREE mode of the findContours algorithm to find all contours, including internal contours, and create a blank image for storing the results of laryngeal expansion;

[0033] S54. Each time a throat contour is extracted, it is drawn on a temporary image, a value is randomly selected from the input throat scale distribution data to obtain the throat expansion amount, and an expansion kernel is created. The lines are expanded using the expansion kernel, and finally the expanded lines are superimposed on the result image. The above processing is performed on each throat, and the radius of each throat is counted.

[0034] Preferably, in step S7, converting the image extraction contour into a DXF file is implemented by Python programming, including the following steps:

[0035] S71. Import the combined image and the ezdxf library;

[0036] S72, smoothing the image using GaussianBlur algorithm;

[0037] S73, setting the length of the line segment used to extract the contour, and then obtaining the contour points, extracting and connecting them into vector lines;

[0038] S74, inverting the Y-axis coordinate to ensure that the contour is consistent with the image direction;

[0039] S75. Check whether the contour is closed.

[0040] Compared with the prior art, the invention has the following beneficial effects: first, the pores and rock skeletons in the casting thin section image are trained and identified using a machine learning model to extract the pore space. Furthermore, the watershed algorithm is used to segment the rock skeleton to connect the pore space, and the pore position distribution, size and throat position distribution are obtained; then, each throat is extracted using a contour recognition algorithm, and the width of each throat is randomly assigned using the distribution of throat size obtained by high-pressure mercury injection using an expansion algorithm; finally, the pore and throat images are combined and the contours are extracted using the zedxf library and converted into DXF files, which are then imported into COMSOL to calculate the porosity and permeability of the model, and the physical properties of the model are verified. Finally, the model is made by importing a photolithography machine to obtain a microscopic visualization photolithography model with a real pore throat structure.

[0041] The model making method of the present invention is programmed, simple and practical, and avoids the problem that the throat cannot be extracted from the thin section data due to the density of the core. At the same time, the physical properties of the model and the real core are compared and verified by simulation means, which can better characterize the reservoir properties and provide a reliable basis for microscopic visualization experiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some examples of the present invention, rather than limitations of the present invention.

[0043] Figure 1 It is a schematic diagram of the casting thin slice of the present invention;

[0044] Figure 2 Schematic diagram of the pore space extracted by the present invention;

[0045] Figure 3 A schematic diagram of the connection relationship between pores and throats using the watershed segmentation algorithm of the present invention;

[0046] Figure 4 This is a schematic diagram of the throat extracted by the algorithm of the present invention;

[0047] Figure 5 This is a schematic diagram of the distribution of the high-pressure mercury injection throat of the present invention;

[0048] Figure 6 A schematic diagram of the invention showing the expansion of the posterior throat and combining it with the pore image;

[0049] Figure 7 It is a schematic diagram of the results of the COMSOL simulation of the present invention. DETAILED DESCRIPTION

[0050] The present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0051] like Figures 1 to 7 As shown, a method for making a microscopic visualized lithography model and its experimental process include the following steps:

[0052] S1. Image selection

[0053] Select high-resolution microscope to take images of representative core casting thin sections ( Figure 1 ) to ensure that pores and rock skeleton can be clearly distinguished;

[0054] S2. Pore extraction

[0055] The selected casting thin section images were imported into ImageJ, the image size was calibrated, and contrast enhancement and denoising were performed to improve the image quality and make it easier to distinguish the pore throats and rock skeletons. Then, the pores and rock skeleton parts in the image were manually annotated with different colors to generate a training data set. A suitable machine learning model for classification problems (Random Forest algorithm) was selected for training. The training data set was continuously increased to improve the generalization ability and accuracy of the model. Finally, the trained model was used to complete the segmentation of pores and rock skeletons in the casting thin section images, and the result map was generated. The threshold segmentation method was used to obtain a binary map. The outlier removal algorithm was used to set the radius to 1 pixel and the threshold to 50 to remove isolated outliers in the image to obtain the pore space distribution map ( Figure 2 );

[0056] S3. Pore throat connectivity classification

[0057] The watershed algorithm is used to process the pore distribution map ( Figure 2 ), which is used to segment the image and connect the various pore spaces. The connecting line is the throat ( Figure 3 );

[0058] S4. Laryngeal extraction

[0059] Use Image Calculator to convert the extracted pore image ( Figure 2 ) and the image containing pore and throat information after watershed algorithm processing ( Figure 3 ) to obtain an image containing only the larynx ( Figure 4 );

[0060] S5. Throat size assignment

[0061] Load the extracted laryngeal grayscale image in S4 ( Figure 4 ) and perform binarization processing, input the distribution data of throat size according to the high-pressure mercury injection data, import it into Python, and use the RETR of the contour extraction algorithm (findContours) _ TREE mode finds all contours, including internal contours, and creates a blank image to store the expanded throat. Each time a throat contour is extracted, it is drawn on a temporary image, a random value is taken from the input throat scale distribution data, the throat expansion amount is obtained, and an expansion kernel is created. The line is expanded using the dilation algorithm (dilate), and finally the expanded line is superimposed on the result image. The above processing is performed for each throat.

[0062] S6. Image combination

[0063] By adding the assigned throat image to the pore image, a binary image with the characteristics of the real core pore structure can be obtained ( Figure 6 );

[0064] S7. Model property calculation

[0065] The image obtained in S6 was smoothed using the GaussianBlur algorithm. The ezdxf library in Python was used to set the length of the line segment used to extract the contour, and then the contour points were obtained, extracted and connected into lines, the Y-axis coordinates were reversed, and the contour was ensured to be consistent with the image direction. The contour was checked to be closed, and then imported into COMSOL to build a geometric model and generate a triangular mesh. The porosity was determined to be 0.26 by dividing the pore throat volume by the total geometric volume, and the permeability was calculated according to Darcy's law:

[0066]

[0067] where u is the Darcy or superficial velocity, k is the permeability, μ is the dynamic viscosity, is the pressure gradient.

[0068] With the help of the creep flow interface in COMSOL, the pressure inlet boundary is set to 5Pa, the pressure outlet boundary is set to 0Pa, the wall is set to no-slip boundary conditions, and the fluid properties (density is 1000kg / m 3; viscosity is 0.001Pa·s), the model and simulation results are as follows Figure 7 As shown in the figure, the mass flow rate at the outlet boundary is obtained by using the variable spf.out1.Mflow, which is 6.1859×10 -6 kg / s, and the model permeability was determined to be about 1780mD by Darcy's law. Compared with the core physical properties, it was found to be consistent with the reservoir characteristics of the study block;

[0069] S8. Model making

[0070] The model was imported into EzCad2 software, connected to a photolithography machine, and the pore throat structure was etched on the slide glass by the photolithography machine. Then, the side of the slide glass with the pore throat network structure was bonded to the cover glass to form a microscopic visual photolithography model.

[0071] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technician familiar with this profession can make some changes or modifications to equivalent embodiments of equivalent changes using the technical contents disclosed above without departing from the scope of the technical solution of the present invention. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention still falls within the scope of the technical solution of the present invention.

Claims

1. A method for making a microscopic visual lithography model based on image processing, characterized in that: The following steps are involved: S1. Image selection Representative thin-section images of core casts taken with a high-resolution microscope were selected to ensure that pores and rock skeletons could be clearly distinguished; S2. Pore extraction The image size is calibrated, and contrast enhancement and denoising are performed to improve the image quality and make it easier to distinguish between pores and rock skeletons. Then, the pores and rock skeleton parts in the image are manually annotated with different colors to generate a training data set. A suitable machine learning model for classification problems (Random Forest algorithm) is selected for training. The training data set is continuously increased to improve the generalization ability and accuracy of the model. Finally, the trained model is used to complete the segmentation of pores and rock skeletons in the cast thin section image, and the pore space such as intergranular pores and dissolution pores is extracted. The threshold segmentation method is used to obtain a binary image. The outlier removal algorithm is used to set a suitable pixel radius and threshold to remove isolated outliers in the image, and the pore radius distribution is calculated. S3. Pore throat connectivity classification The pore-throat space image after segmentation by watershed algorithm is connected to each pore space, and the connecting line is the throat; S4. Extraction of throat position Subtract the extracted pore space image from the image containing pore and throat information processed by the watershed algorithm to obtain an image containing only the throat; S5. Throat size assignment The throat size distribution probability function obtained by high-pressure mercury injection experiment is fitted to obtain the functional relationship between radius and distribution probability. Then, all throat contours in the throat binary image are detected by contour extraction algorithm. The fitted throat size distribution probability function is used for each detected throat contour. The width of the throat contour is randomly expanded by using the expansion algorithm to complete the random assignment of each throat radius and calculate the average pore-throat ratio. S6. Image combination By adding the gray values ​​of the throat image and the pore image after the value assignment, a binary image with the characteristics of the real core pore structure can be obtained. S7. Model property calculation The image obtained in S6 was converted into a DXF file using the ezdxf library in Python, and then imported into COMSOL to build the model. The porosity was determined by dividing the pore throat volume by the total geometric volume, and the permeability was determined according to Darcy's law: where u is the Darcy or superficial velocity, k is the permeability, μ is the dynamic viscosity, is the pressure gradient; With the help of the creep flow interface in COMSOL, the inlet, outlet, wall boundary conditions and fluid properties are set, and the flow rate at the outlet boundary is obtained through the variable spf.out1.Mflow. The model permeability is determined by Darcy's law, compared with the core physical properties, and the model is adjusted; S8. Model making Connect a photolithography machine, etch the pore throat network structure on the carrier glass through the photolithography machine, and then bond the side of the carrier glass with the pore throat network structure to the cover glass to form a microscopic visual photolithography model.

2. The method for making a microscopic visual lithography model based on image processing according to claim 1, characterized in that: In step S2, the image processing operations such as image size calibration, contrast enhancement, denoising, and threshold segmentation are implemented by ImageJ software, and the random forest algorithm is a machine learning algorithm for processing classification tasks, which is implemented by ImageJ software.

3. The method for making a microscopic visualized lithography model based on image processing according to claim 1, characterized in that: In step S3, the outlier removal algorithm is a filtering method used to remove outliers; the watershed algorithm is an image segmentation method used to connect pores to form throats. These two algorithms are implemented using ImageJ software.

4. The method for making a microscopic visualized lithography model based on image processing according to claim 1, characterized in that: In step S5, the throat size assignment method is implemented by Python programming, including the following steps: S51, loading the laryngeal grayscale image extracted in S4 and performing binarization processing; S52. Input throat size distribution data based on high-pressure mercury injection data; S53, using the RETR_TREE mode of the findContours algorithm to find all contours, including internal contours, and create a blank image for storing the results of laryngeal expansion; S54. Each time a throat contour is extracted, it is drawn on a temporary image, a value is randomly selected from the input throat scale distribution data to obtain the throat expansion amount, and an expansion kernel is created. The lines are expanded using the expansion kernel, and finally the expanded lines are superimposed on the result image. The above processing is performed on each throat, and the radius of each throat is counted.

5. The method for making a microscopic visualized lithography model based on image processing according to claim 1, characterized in that: In step S7, the image extraction contour is converted into a DXF file by Python programming, including the following steps: S71. Import the combined image and the ezdxf library; S72, smoothing the image using GaussianBlur algorithm; S73, setting the length of the line segment used to extract the contour, and then obtaining the contour points, extracting and connecting them into lines; S74, inverting the Y-axis coordinate to ensure that the contour is consistent with the image direction; S75. Check whether the contour is closed.

Images