Machine Learning-Based Method for Identifying CO2-Water Reactive Dissolution and Porosity Assessment in Rock Slices

Abstract

This application relates to the field of carbon sequestration and geological storage technology, specifically disclosing a machine learning-based method for identifying CO2-water reaction dissolution and assessing porosity in rock slices. First, multi-focal images of rock slices before and after the reaction are acquired and registered on a scanning electron microscope platform. A trained U-Net instance segmentation model is used to automatically identify dissolution regions and extract their contours. A relative height map is generated using depth-of-field reconstruction and calibration techniques, combining the multi-focal sequence to obtain the effective dissolution depth of each region. Then, volume normalization calculation is performed based on area-depth information, directly outputting the porosity increment and its spatial distribution. This application eliminates the need for destructive testing, achieving quantitative volumetric characterization of dissolution features, significantly improving analytical efficiency and accuracy. It supports batch processing and mineral phase stratification statistics, providing reliable data support for research on CO2 geological storage reaction mechanisms and reservoir simulation.

Description

Technical Field

[0001] This application relates to the field of carbon sequestration and geological storage technology, and in particular to a machine learning-based method for identifying CO2-water-rock reaction dissolution and assessing porosity in rock slices. Background Technology

[0002] Carbon dioxide geological sequestration involves injecting carbon dioxide into deep reservoirs, causing a water-rock reaction under thermo-baric conditions to achieve long-term stable sequestration. During this process, carbonation dissolution, secondary precipitation, and ion exchange collectively reshape the microstructure and mineral interfaces, leading to spatiotemporal evolution of porosity and pore throat topology. These changes not only affect the mineralization fixation rate and final capacity but also alter the formation flow field and pressure propagation patterns, thus impacting sequestration safety and operational windows. In laboratory and pilot project assessments, high-resolution characterization at the thin section / rock slice scale directly captures subtle differences in micromorphology and pore geometry before and after the reaction, crucial for understanding reaction kinetics, establishing a petrology-fluidometry mapping, and providing constraints for numerical simulations. Therefore, developing a high-throughput, objective, traceable, and engineering parameter-compatible quantitative method for porosity changes is both practically necessary and engineeringally valuable for sequestration stability assessment, parameter inversion, and optimization of field monitoring strategies. At the same time, this method should take into account the compatibility of multiple material systems (basalt, sandstone, carbonate rocks, etc.), be compatible with the differences brought about by different polishing quality, conductive coatings and imaging conditions, and ensure consistency and comparability across samples and batches.

[0003] Existing technologies typically characterize microscopic evolution along two paths: one is two-dimensional morphology + composition analysis. This involves using SEM (Scanning Electron Microscope) to acquire secondary electron / backscattered images, supplemented by EDS (Energy Dispersive Spectrometer) surface scanning to identify mineral phases and elemental enrichment / depletion. Based on this, dissolution contours are extracted and area distribution is statistically analyzed through threshold segmentation, morphological manipulation, or deep learning segmentation. This method is relatively mature, but it mainly focuses on the two-dimensional boundary level, making it difficult to provide volumetric porosity increments. It is also sensitive to image contrast, sample charging, polishing texture, and illumination uniformity, and has limited domain migration capabilities. The other path is three-dimensional / depth acquisition, such as FIB-SEM (Focused Ion Beam-Scanning Electron Microscope) 3D reconstruction, confocal microscopy, or white light interferometry, which can provide voxel-level volumetric information. However, it has high equipment / time costs, limited field of view, and stringent requirements for sample preparation and platform, making it difficult to support routine high-throughput detection. In addition, the registration of images before and after still faces drift and local deformation errors, which affect the stability of batch processing. It is also strongly affected by the connectivity assumption and boundary condition setting, resulting in unstable indicators and insufficient cross-sample comparability. Summary of the Invention

[0004] To address the limitations of traditional two-dimensional detection in terms of accuracy and the high cost and low throughput of three-dimensional detection, this application provides a machine learning-based method for identifying CO2-water-rock reaction dissolution and assessing porosity in rock slices.

[0005] The method for identifying CO2-water reaction dissolution and assessing porosity in rock slices based on machine learning provided in this application adopts the following technical solution: A machine learning-based method for identifying CO2-water-rock reaction dissolution and assessing porosity in rock slices includes the following steps: Sample preparation and image acquisition: The rock slice samples were placed in a CO2-water environment for reaction; multi-focal images of the rock slice samples before and after the reaction were acquired on the scanning electron microscope platform, and the image sequence was obtained and the imaging parameters were recorded. Image preprocessing and front-to-back registration: noise suppression, illumination equalization and distortion correction are performed on the image sequence, and geometric registration of the front and back fields of view is completed based on image markers or features; Automatic identification of dissolved regions: The contours and numbers of dissolved regions are automatically extracted using a trained instance segmentation model, and the area of ​​each dissolved region is output. Depth of field / relative height reconstruction and calibration: Based on the multi-focal plane image, a relative height map of the field of view is generated, and the focal length change is converted into the actual height through a calibration method to obtain the effective dissolution depth of each dissolution area; Porosity change calculation: Based on the area and effective dissolution depth of each dissolution region, the porosity increment of the field of view is obtained by normalization according to the field volume. Results output and quality control: Output the porosity increment and zonal statistical reports, and perform repeatability and consistency verification.

[0006] Furthermore, in the automatic identification step of the dissolved region, the instance segmentation model is a deep learning model based on the U-Net architecture, which is trained using a dataset of dissolved region images containing manual annotations.

[0007] Furthermore, in the depth of field / relative height reconstruction and calibration step, the calibration method is the step sample method, which uses step samples with known height differences to perform multi-focal plane scanning and establish the conversion relationship between focal length changes and actual height.

[0008] Furthermore, in the depth of field / relative height reconstruction and calibration steps, the calibration method is the platform stepping method, which involves precisely controlling the platform to move at equal step lengths and perform multi-focal plane scanning, recording focal length changes and relating them to the stepping distance to reconstruct the relative height map.

[0009] Furthermore, in the step of calculating the change in porosity, the formula for calculating the porosity increment Δφ is:

[0010] in, A i Let i be the area of ​​the i-th etched region. d i Let be the effective dissolution depth of the i-th dissolution region. A FOV For the field of view area, h ref The reference thickness is the thickness of a petrographic standard thin section, or the actual thickness value after calibration with a standard sample.

[0011] Furthermore, in the results output and quality control steps, the repeatability and consistency verification includes repeatedly measuring the same field of view and calculating the relative standard deviation of the porosity increment.

[0012] Furthermore, in the result output and quality control steps, the repeatability and consistency verification also includes batch processing and stitching of multiple adjacent fields of view to verify the stability and consistency of the results.

[0013] Furthermore, the results output and quality control steps also include: combining the elemental distribution information obtained from energy dispersive spectroscopy surface scanning to help identify dissolution regions in different mineral phases and perform stratified statistical analysis.

[0014] This application also provides a system for implementing a machine learning-based method for identifying CO2-water-rock reaction dissolution and assessing porosity in rock slices, including: The image acquisition and calibration unit is used to control the scanning electron microscope platform to complete multi-focal plane image acquisition and parameter recording. The image preprocessing and registration unit is used to preprocess the image and complete the registration of the image before and after the reaction. The erosion region identification unit is used to load and run the instance segmentation model to identify and extract the erosion region; Depth of field / relative height reconstruction and calibration unit, used to generate relative height maps and perform depth calibration; The porosity calculation and result output unit is used to calculate the porosity increment and generate a report. The quality control and batch processing unit is used to manage batch processing workflows and perform quality verification.

[0015] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a machine learning-based method for identifying CO2-water-rock reaction dissolution in rock slices and assessing porosity.

[0016] In summary, this application includes at least one of the following beneficial technical effects: This application combines instance segmentation technology with multi-focal depth estimation to achieve automatic and accurate identification and three-dimensional volumetric quantitative assessment of CO2-water-rock reaction dissolution areas. This method overcomes the limitations of traditional two-dimensional morphology analysis by unifying the area and depth information of the dissolution area into a volumetric representation through depth reconstruction and calibration techniques, directly outputting the porosity increment, significantly improving the physical meaning and accuracy of the assessment results. Its workflow integrates image registration, deep learning segmentation, height reconstruction, and statistical calculation, possessing excellent automation and repeatability, effectively addressing the limitations of traditional two-dimensional detection accuracy and the high cost and low throughput of three-dimensional detection.

[0017] This application demonstrates excellent engineering applicability and scalability. The system supports depth calibration based on step-by-step standards or platform steps, ensuring traceability and cross-sample comparability of measurements. Combined with EDS data, it can also perform dissolution statistics for different mineral facies, enhancing adaptability to various lithologies and imaging conditions. This method can be performed on conventional SEM / EDS platforms without damaging samples or requiring special hardware, making it easy to deploy and promote. It also supports batch processing and multi-field stitching, meeting high-throughput detection requirements. The final output results can be directly used for reservoir numerical simulation and storage efficiency assessment, providing a reliable data foundation for reaction mechanism research, parameter inversion, and safety monitoring in carbon dioxide geological storage projects. Attached Figure Description

[0018] Figure 1 This is a flowchart of a machine learning-based method for identifying CO2-water-rock reaction dissolution and assessing porosity in rock slices according to an embodiment of this application. Figure 2 This is a schematic diagram of the CO2 water-rock reactor in the embodiments of this application; Figure 3 This is an image of a rock slice sample after reaction, as described in the embodiments of this application. Figure 4 This is a schematic diagram of the identification of the dissolved area in the embodiments of this application; Figure 5 This is a zoning statistical report of dissolution regions 1-10 in the embodiments of this application; Figure 6 This is the EDS map in the embodiments of this application. Detailed Implementation

[0019] The following is in conjunction with the appendix Figure 1-6 This application will be described in further detail.

[0020] This application discloses a machine learning-based method for identifying CO2-water reaction dissolution of rock fragments and assessing porosity. (Refer to...) Figure 1 The machine learning-based method for identifying CO2-water-rock reaction dissolution and assessing porosity in rock slices includes the following steps: Step 1, Sample Preparation and Image Acquisition, specifically includes: Step 1.1, Sample Preparation: like Figure 2As shown, the CO2 water-rock reaction apparatus includes a reactor, a temperature-controlled heating module, and a CO2 injection system. The reactor is a high-pressure reactor. The temperature-controlled heating module includes a heater for heating the reactor. The CO2 injection system includes a carbon dioxide cylinder and a carbon dioxide pump connected to the reactor. The reactor is also connected to a vacuum pump for evacuation. Temperature and pressure sensors are installed inside the reactor to monitor the temperature and pressure, respectively. Rock samples react with CO2 and water inside the reactor to simulate the water-rock reaction of CO2 with the surrounding rock under temperature and pressure conditions during the geological sequestration of carbon dioxide.

[0021] The rock sample used in this embodiment was a basalt sheet with a diameter of 25 mm and a thickness of 5 mm, and the surface was polished. Deionized water was added to the reactor, and pure CO2 gas was injected to a preset pressure of 10 MPa. The reaction temperature was set at 60°C, and the reaction time was 72 hours. After the reaction was completed, the rock sheet was removed (e.g., Figure 3 As shown in the figure, it undergoes pretreatment such as drying and conductive coating, and is ready for SEM imaging and analysis.

[0022] Step 1.2, Image Acquisition: SEM imaging was performed in SE2 mode, with the electron beam acceleration voltage set to 3.0 kV and the working distance to 7.2 mm. Multi-focal plane image sequences were acquired for specific fields of view (physical size approximately 3810 μm × 2650 μm) of rock slice samples before and after the reaction at a magnification of 30x.

[0023] Record imaging parameters, including magnification, working distance, stage coordinates, and scale information, to ensure dimensional consistency in subsequent calculations.

[0024] Step 2: Image preprocessing and registration (before and after registration): Noise suppression, illumination equalization, and distortion correction are performed on the image sequence. The pre- and post-reaction fields of view are registered under a global affine-local non-rigid hybrid model to form a standardized dataset, providing input for subsequent analysis.

[0025] Step 3: Automatic identification of the eroded area, including: Step 3.1, Model Training: Using an image dataset of manually labeled dissolution areas containing samples of this type of rock fragments, the instance segmentation model based on the U-Net architecture is trained. During the training process, data augmentation strategies including random rotation, translation, scaling, and contrast adjustment are adopted to improve the generalization ability of the model.

[0026] Step 3.2, Extraction of Dissolution Regions: Input the post-reaction SEM image registered in Step 2 into the trained model to automatically identify and extract 46 dissolution regions (numbered 1-46, their outlines and numbers are shown in the diagram). Figure 4Output the area of ​​each dissolved region. And its position in the field of view.

[0027] Step 4, Depth of field / relative height reconstruction and calibration: Based on a multi-focal plane image sequence, a relative height map of the field of view is reconstructed using a focal length evaluation algorithm. A platform stepping method (50nm step size) is employed for dimensional calibration; that is, by precisely controlling the platform to move at equal step sizes and recording focal length changes, the relationship between focal length and true height is established, thereby obtaining the effective dissolution depth of each dissolution region. Based on this, the depth-of-field coefficient is defined. (Value range 0-1), where d i The effective dissolution depth of the i-th dissolution region, with reference thickness. h ref The standard petrographic section is 30 μm thick.

[0028] Step 5: Calculation of porosity change: Based on the area and effective dissolution depth of each dissolution region, the porosity increment Δφ of the field of view is obtained by normalizing the calculation according to the field volume:

[0029] in, A i Let i be the area of ​​the i-th etched region. d i Let be the effective dissolution depth of the i-th dissolution region. A FOV For the field of view area, h ref The reference thickness is the thickness of a petrographic standard thin section, or the actual thickness value after calibration with a standard sample. In this embodiment, The total area of ​​the dissolved region accounts for approximately 25.3% of the field of view (i.e., The average effective dissolution depth of the region was approximately obtained through depth-of-field calibration. Substituting into the formula, we get:

[0030] That is, the average porosity increment of this field of view is approximately 2.1%.

[0031] Step 6, Results Output and Quality Control: Output porosity increment and zonal statistical reports. The zonal statistical reports include the location, area, and depth of each dissolution region (e.g., ...). Figure 5 (As shown).

[0032] To verify the reliability of the results, three repeated measurements were performed on this field of view, and the relative standard deviation of the porosity increment was less than 10%. At the same time, four adjacent fields of view were batch-processed and stitched together to verify the spatial consistency of the results.

[0033] Optionally, mineral phase identification and regional stratification can be performed using EDS surface scan data. The specific process is as follows: In step 1.2, EDS surface scans are performed on the same field of view of the post-reaction sample to obtain elemental distribution maps (such as Si, Ca, Fe, Mg, etc.). Then, based on the combination relationship of the main controlling elements, mineral phase classification is performed on the image region using preset mineral phase discrimination rules (such as high Ca, low Si corresponding to calcite, high Si, Al corresponding to feldspar, etc.). Subsequently, the identified mineral phase information is spatially matched with the segmented dissolution region to achieve mineralogical stratification within the dissolution region (such as...). Figure 6 (As shown).

[0034] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A machine learning-based method for identifying CO2-water-rock reaction dissolution and assessing porosity in rock slices, characterized in that: Includes the following steps: Sample preparation and image acquisition: The rock slice samples were placed in a CO2-water environment for reaction; multi-focal images of the rock slice samples before and after the reaction were acquired on the scanning electron microscope platform, and the image sequence was obtained and the imaging parameters were recorded. Image preprocessing and front-to-back registration: noise suppression, illumination equalization and distortion correction are performed on the image sequence, and geometric registration of the front and back fields of view is completed based on image markers or features; Automatic identification of dissolved regions: The trained instance segmentation model is used to automatically identify and extract dissolved regions, obtain the outline of the dissolved regions and number them, and output the area of ​​each dissolved region. Relative height reconstruction and calibration: A relative height map of the field of view is generated based on the multi-focal plane image, and the focal length change is converted into the actual height through a calibration method to obtain the effective dissolution depth of each dissolution area; Porosity change calculation: Based on the area and effective dissolution depth of each dissolution region, the porosity increment of the field of view is obtained by normalization according to the field volume; the formula for calculating the porosity increment Δφ is as follows: in, A i Let i be the area of ​​the i-th dissolution region. d i Let be the effective dissolution depth of the i-th dissolution region. A FOV For the field of view area, h ref The reference thickness is the thickness of a petrographic standard thin section, or the actual thickness value of a petrographic standard thin section after calibration with standard samples. Results output and quality control: Output the porosity increment and zonal statistical reports, and perform repeatability and consistency verification.

2. The method for identifying CO2-water-rock reaction dissolution and assessing porosity of rock slices based on machine learning according to claim 1, characterized in that: In the automatic identification step of the dissolved region, the instance segmentation model is a deep learning model based on the U-Net architecture, which is trained using a dataset of dissolved region images containing manual annotations.

3. The method for identifying CO2-water-rock reaction dissolution and assessing porosity of rock slices based on machine learning according to claim 1, characterized in that: In the relative height reconstruction and calibration steps, the calibration method is the step standard method, which uses step standard with known height differences to perform multi-focal plane scanning and establish the conversion relationship between focal length change and actual height.

4. The method for identifying CO2-water-rock reaction dissolution and assessing porosity of rock slices based on machine learning according to claim 1, characterized in that: In the relative height reconstruction and calibration steps, the calibration method is the platform stepping method, which uses precise control of the platform to move at equal step lengths and perform multi-focal plane scanning, records focal length changes and correlates them with the stepping distance to reconstruct the relative height map.

5. The method for identifying CO2-water-rock reaction dissolution and assessing porosity of rock slices based on machine learning according to claim 1, characterized in that: In the results output and quality control steps, the repeatability and consistency verification includes repeatedly measuring the same field of view and calculating the relative standard deviation of the porosity increment.

6. The method for identifying CO2-water-rock reaction dissolution and assessing porosity of rock slices based on machine learning according to claim 5, characterized in that: In the results output and quality control steps, the repeatability and consistency verification also includes batch processing and stitching of multiple adjacent fields of view to verify the stability and consistency of the results.

7. The method for identifying and assessing porosity of rock fragments based on CO2-water-rock reaction dissolution according to claim 1, characterized in that: The results output and quality control steps also include: combining the elemental distribution information obtained from scanning the same field of view of the rock slice sample with energy dispersive spectroscopy to help identify the dissolution regions in different mineral phases and perform stratified statistics.

8. A system for implementing the machine learning-based method for identifying CO2-water-rock reaction dissolution and assessing porosity of rock slices according to any one of claims 1-7, characterized in that: include: The image acquisition and calibration unit is used to control the scanning electron microscope platform to complete the acquisition of multi-focal plane images and the recording of parameters. An image preprocessing and registration unit is used to preprocess the image sequence and complete the registration of the images before and after the reaction. The erosion region identification unit is used to load and run the instance segmentation model to identify and extract the erosion region; The relative height reconstruction and calibration unit is used to generate a relative height map of the field of view and convert the focal length change into the actual height through a calibration method to obtain the effective dissolution depth of each dissolution area; The porosity calculation and result output unit is used to calculate the porosity increment and generate a report; the formula for calculating the porosity increment Δφ is: in, A i Let i be the area of ​​the i-th dissolution region. d i Let be the effective dissolution depth of the i-th dissolution region. A FOV For the field of view area, h ref The reference thickness is the thickness of a petrographic standard thin section, or the actual thickness value of a petrographic standard thin section after calibration with standard samples. The quality control and batch processing unit is used to manage batch processing workflows and perform quality verification.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements a machine learning-based method for identifying CO2-water-rock reaction dissolution and assessing porosity in rock slices, as described in any one of claims 1-7.

Images