A deep learning-based yield stress viscous fluid rheological parameter calculation method

Abstract

A deep learning-based method for calculating the rheological parameters of viscous fluids with yield stress is disclosed. The method involves instrument preparation and parameter setting; pre-shearing the sample, allowing it to stand for a set time before starting the formal rheological test and simultaneously capturing flow images; within each acquisition interval, a deep learning model automatically marks and tracks the positional changes of feature points on the rotor's periphery, and eliminates errors through perspective transformation; based on the polar coordinate changes of the feature points, the shear rate of each pair of points is calculated to obtain the average shear rate; the shear stress values ​​recorded by the rheometer within the same acquisition interval are extracted, and their arithmetic mean is calculated as the average shear stress; tests and calculations are performed sequentially at different preset rotor speeds to obtain the corresponding average shear rate and average shear stress data at each speed; the data from multiple speeds are substituted into the rheological model for nonlinear regression to obtain the rheological parameters. This invention improves the accuracy of rheological parameter fitting results and the reliability of rheological performance characterization results.

Description

Technical Field

[0001] This invention belongs to the field of materials engineering testing technology, specifically relating to a method for calculating yield stress rheological parameters of viscous fluids based on deep learning. Background Technology

[0002] Fluid materials involve multiple disciplines such as fluid mechanics, materials science, materials chemistry, and biology, and their development directly impacts various application fields including civil engineering, biomedicine, materials chemistry, petroleum engineering, and mining engineering. Rheological properties are one of the core properties of fluid materials, and accurately measuring their rheological parameters and characterizing their rheological properties is of great significance for the production, processing, and practical application of fluid materials.

[0003] The rise of deep learning technology has brought new challenges and opportunities to industry and academia, and has also provided a completely new approach to characterizing the rheological behavior of fluid materials and fitting the rheological parameters of viscous fluids at yield stress. Combining deep learning with image processing technology has solved the problem of calculating the rheological parameters of viscous fluids at yield stress, which is of great significance for promoting the intelligent development of traditional engineering disciplines.

[0004] A coaxial rotating rheometer is a commonly used device for measuring the rheological properties of fluids. Based on theoretical formulas, when a fluid is subjected to complete shearing, the shear deformation rate of the fluid can be calculated using the rotor speed. Simultaneously, based on the torque applied to the rotor, the shear stress of the fluid at a certain location from the rotor center can be determined. By combining the relationship between shear stress and shear deformation rate and substituting it into a suitable rheological model, key rheological parameters such as yield stress and viscosity can be fitted.

[0005] However, viscous fluids with yield stress will not undergo shear deformation when the internal shear stress is less than their yield stress. Since the shear stress within the fluid is inversely proportional to the square of the distance from the rotor center, the shear stress is higher near the rotor and lower farther away. Therefore, when the shear stress at a certain location far from the rotor is lower than the fluid's yield stress, the fluid in that region remains stagnant, forming a slug layer (such as...). Figure 2 (As shown). The presence of a slug layer renders the theoretical assumption of complete fluid shearing invalid, leading to a certain range of deviation in fluid shear rate calculations, and this deviation increases with the expansion of the slug layer. The influence of the slug layer on shear rate calculations cannot be ignored. Therefore, accurately measuring the actual shear rate of fluid samples is crucial for improving the accuracy of rheological parameter calculations and ensuring the reliability of rheological performance characterization.

[0006] To address the shortcomings of the existing technologies, there is an urgent need to provide a deep learning-based method for calculating the rheological parameters of viscous fluids with yield stress, so as to effectively improve the accuracy of rheological parameter calculation and the reliability of rheological performance characterization results. Summary of the Invention

[0007] To address the problems existing in the prior art, this invention provides a deep learning-based method for calculating rheological parameters of viscous fluids with yield stress. This method can accurately calculate the actual shear deformation rate of the fluid, significantly improve the accuracy of rheological parameter fitting results and the reliability of rheological performance characterization results, and is of great significance for accurately characterizing and evaluating the rheological properties of viscous fluids with yield stress.

[0008] To achieve the above objectives, the present invention provides a method for calculating the yield stress rheological parameters of viscous fluids based on deep learning, comprising the following steps; S1: Instrument preparation and parameter setting; Configure a coaxial rotational rheometer, camera and sample collection device, set test parameters and prepare the fluid sample to be tested; S2: Pre-shear and rheological testing; The sample was pre-sheared to eliminate structural errors, and after being left to stand for a set time, the formal rheological test was started and images of the flow process were captured simultaneously. S3: Feature point tracking and localization based on deep learning; Within each acquisition interval, a deep learning model is used to automatically mark and track the positional changes of feature points on the rotor periphery, and perspective transformation is used to eliminate shooting angle errors. S4: Calculate the average shear rate at the rotor edge; Based on the polar coordinate changes of feature points, the shear rate of each pair of points is calculated by pairing them, and then the average shear rate is obtained. S5: Calculate the average shear stress; The shear stress values ​​recorded by the rheometer within the same acquisition interval are extracted, and their arithmetic mean is calculated as the average shear stress at the rotor edge. ; S6: Multi-speed testing and data processing; Repeat steps S2 to S5 and perform tests and calculations at different preset rotor speeds to obtain the average shear rate at each speed. With mean shear stress data; S7: Rheological parameter fitting; Data from various rotational speeds were substituted into a rheological model for nonlinear regression to obtain rheological parameters.

[0009] As a preferred option, the instrument preparation and parameter setting process in S1 is as follows: S11: Set the operating parameters of the coaxial rotary rheometer; S12: To prepare fluid materials, take a set amount of sample and pour it into a measuring cylinder, making the surface of the sample flush with the sampling mark line on the measuring cylinder; S13: Adjust the rotor embedment depth, adjust the camera angle and white balance to ensure that the outer area of ​​the rotor can be clearly photographed.

[0010] As a preferred option, the pre-shear and rheological testing process in S2 is as follows: S21: Pre-shear the sample according to the set parameters to destroy the microstructure of the freshly mixed fluid material and eliminate experimental errors caused by structural recovery. S22: After the pre-shearing is completed, the sample is allowed to stand for a set time, and then the rheological test is started; during the test, the flow behavior images around the rotor are captured continuously using a camera.

[0011] As a preferred option, in S22, the set time is no more than 30 seconds.

[0012] As a preferred option, the deep learning-based feature point tracking and localization process in S3 is as follows: S31: Perform perspective transformation on the original image captured by the camera to restore the elliptical sample outline obtained by the camera to the actual circular outline, and convert the image coordinate system to a circular coordinate system with the rotor center as the origin to eliminate the coordinate error caused by the shooting angle. S32: Based on the grayscale differences in the cropped image, automatically label the set of feature points around the rotor and record the start time of the acquisition interval. and the initial coordinates of each feature point ; S33: Run a deep learning model to detect and identify feature points, retain the feature points that persist within the acquisition interval, and update their coordinates to the current time. S34: When the data acquisition interval ends, record the termination time. The updated feature point coordinates are used as the termination coordinates. .

[0013] As a preferred embodiment, in S4, the process for calculating the average shear rate at the rotor edge is as follows: S41: Sort and number all tracked feature points according to their average coordinate radius within the acquisition interval from smallest to largest; S42: Pair the remaining feature points according to their sorting order, using a fixed difference value; S43: For each pair of feature points A and B, calculate the shear rate corresponding to that pair of points according to formula (1). ; (1); In the formula, , These are the average coordinate radii of feature points A and B within the acquisition interval, respectively. , These are the polar angles of feature points A and B at the beginning of the interval, respectively; , These are the polar angles of feature points A and B at the end of the interval, respectively; The data collection interval, ; S44: Calculate the average shear rate of all effective feature points to obtain the average shear rate of the rotor edge within this acquisition interval. ; (2); In the formula, This represents the number of valid feature point pairs.

[0014] As a preferred method, in S7, shear rate and shear stress data at different rotational speeds are processed and substituted into a selected rheological model for nonlinear regression fitting to obtain the yield stress of the fluid. Consistency coefficient rheological index The rheological parameters.

[0015] This invention provides a deep learning-based method for calculating the yield stress rheological parameters of viscous fluids, aiming to overcome the shortcomings of existing rheological shear rate calculation methods. First, by rationally setting test parameters, a unified test benchmark is provided, ensuring the consistency of the initial state and the comparability of data in subsequent rheological tests. Second, the pre-shear operation can, to a certain extent, disrupt the fluid's microstructure, effectively eliminating the initial structural errors caused by the recovery of freshly mixed fluid materials after settling, further ensuring the consistency of the initial state and the comparability of data in subsequent rheological tests. Starting the test after a set set settling time allows the fluid to recover moderately from the pre-shear disturbance, avoiding test distortion due to excessive recovery. Furthermore, a deep learning model is used to automatically mark and track feature points on the rotor's periphery, effectively replacing traditional manual marking or theoretical formula calculations, significantly improving the automation and robustness of feature point recognition. Simultaneously, the introduction of perspective transformation completely eliminates coordinate distortion caused by the camera's shooting angle, ensuring the accuracy of feature point position measurement. Subsequently, a direct shear rate calculation method based on the polar coordinate variation of feature points was established, which does not rely on the theoretical formulas of traditional coaxial cylindrical rheometers (such as the Newtonian fluid assumption or narrow gap approximation), thus avoiding the failure of traditional formulas due to fluid choking, wall slippage, or non-Newtonian characteristics. Next, the shear stress data recorded by the rheometer was directly extracted and its average value was calculated, maintaining strict time synchronization (same acquisition interval) with the shear rate obtained from deep learning. This ensured the correspondence between the two sets of physical quantities, providing accurate input data pairs for subsequent rheological parameter fitting. Finally, by sequentially executing different preset rotor speeds, multiple data points covering a wide range of shear rates were obtained, fully reflecting the non-Newtonian flow characteristics of the fluid. Further regression fitting using a rheological model can simultaneously obtain three key parameters: yield stress, consistency coefficient, and rheological index, applicable to viscous fluids with yield stress (such as high-concentration slurry, concrete, drilling fluid, etc.). The entire process achieves full automation from image acquisition, feature tracking, rate calculation to parameter fitting, significantly improving testing efficiency and result repeatability.

[0016] This invention measures the rheological properties of fluids using a coaxial rotating rheometer. During the rheometer's operation and the fluid shearing process, images of the fluid shearing are captured by a camera. Deep learning technology is then used to calculate the rheological shear rate. Subsequently, the shear stress obtained from the coaxial rotating rheometer, along with the actual shear rate calculated by deep learning, is substituted into the rheological model to obtain the rheological parameters of the viscous fluid at yield stress. This invention fully considers the influence of slug layers induced by yield stress on the shear flow within the fluid. It innovatively uses deep learning technology to first obtain the rheological shear rate at different rotor speeds from the images, effectively eliminating the influence of factors such as slug flow, slip, or wall effects when calculating the rheological shear rate using theoretical formulas. After obtaining the actual rheological shear rate, it is substituted into the rheological model to fit high-precision rheological parameters, ultimately enabling accurate calculation of the actual shear deformation rate of the fluid. This significantly improves the accuracy of rheological parameter calculation and the reliability of rheological performance characterization results.

[0017] This invention breaks through the limitations of traditional coaxial rheometers that rely on theoretical formulas to calculate shear rates. It innovatively applies deep learning technology to obtain shear rates from actual fluid flow images, fundamentally avoiding the accuracy degradation problems caused by factors such as choke flow, slippage, and wall effects. Simultaneously, through a series of techniques including perspective transformation, marker range optimization, and feature point pairing, a high-precision, robust, and fully automated rheological parameter analysis scheme is constructed. This scheme is particularly suitable for characterizing the rheological properties of complex non-Newtonian fluids containing solid particles (such as filling slurry, concrete, and mud), and has significant engineering application value.

[0018] This method is simple to implement, low in cost, highly operable, and efficient in testing. It can accurately calculate the actual shear deformation rate of the fluid, significantly improve the accuracy of rheological parameter fitting results and the reliability of rheological performance characterization results, and is of great significance for accurately characterizing and evaluating the rheological properties of viscous fluids with yield stress. Attached Figure Description

[0019] Figure 1 This is a flowchart of the present invention; Figure 2 A schematic diagram of the plug flow layer during the shearing process of fluid in a coaxial rotating rheometer; Figure 3 This is a schematic diagram of the rotor speed setting during the rheological testing process of the present invention; Figure 4 This is a schematic diagram illustrating the changes in the position of feature points in this invention; Figure 5 This is a schematic diagram illustrating the calculation of the shear rate of cement mortar in this invention.

[0020] Figure 6 This is a comparison chart of ideal conditions, real conditions, and the rheological curves calculated by the present invention in the embodiments of the present invention. Detailed Implementation

[0021] like Figure 1 As shown, the present invention provides a method for calculating the yield stress rheological parameters of viscous fluids based on deep learning, including the following steps; S1: Instrument preparation and parameter setting; Configure a coaxial rotational rheometer, camera and sample collection device, set test parameters and prepare the fluid sample to be tested; S2: Pre-shear and rheological testing; The sample was pre-sheared to eliminate structural errors, and after being left to stand for a set time, the formal rheological test was started and images of the flow process were captured simultaneously. S3: Feature point tracking and localization based on deep learning; Within each acquisition interval, a deep learning model is used to automatically mark and track the positional changes of feature points on the rotor periphery, and perspective transformation is used to eliminate shooting angle errors. As a preferred embodiment, the deep learning model in this invention uses a pre-trained model. That is, before implementing this method, the model (such as convolutional neural network, U-Net, key point detection network, etc.) has been trained using a large number of image samples with labeled feature points, so that it has the ability to recognize rotor edge feature points.

[0022] S4: Calculate the average shear rate at the rotor edge; Based on the polar coordinate changes of feature points, the shear rate of each pair of points is calculated by pairing them, and then the average shear rate is obtained. S5: Calculate the average shear stress; The shear stress values ​​recorded by the rheometer within the same acquisition interval are extracted, and their arithmetic mean is calculated as the average shear stress at the rotor edge. ; S6: Multi-speed testing and data processing; Repeat steps S2 to S5 and perform tests and calculations at different preset rotor speeds to obtain the average shear rate at each speed. With mean shear stress data; As a preferred option, the rotor speed has at least five different operating conditions, such as... Figure 3 As shown; S7: Rheological parameter fitting; Data from various rotational speeds were substituted into a rheological model for nonlinear regression to obtain rheological parameters.

[0023] As a preferred option, the instrument preparation and parameter setting process in S1 is as follows: S11: Set the operating parameters of the coaxial rotary rheometer, including rotor speed sequence, shearing time, data acquisition interval, etc. S12: To prepare fluid materials, take a set amount of sample and pour it into a measuring cylinder, making the surface of the sample flush with the sampling mark line on the measuring cylinder; S13: Adjust the rotor embedment depth, adjust the camera angle and white balance to ensure that the outer area of ​​the rotor can be clearly photographed.

[0024] In this technical solution, the standardization and repeatability of test conditions are ensured by reasonably presetting the operating parameters (rotation speed, time, acquisition interval, etc.) of the coaxial rotary rheometer; the quantitative sampling method and the alignment of the liquid level with the mark line are adopted to eliminate the influence of sample volume differences on the test results; by adjusting the rotor burial depth and camera parameters, the clarity and consistency of the captured images are ensured, providing high-quality, distortion-free raw image data for subsequent deep learning feature extraction.

[0025] As a preferred option, the pre-shear and rheological testing process in S2 is as follows: S21: Pre-shear the sample according to the set parameters to destroy the microstructure of the freshly mixed fluid material and eliminate experimental errors caused by structural recovery. S22: After the pre-shearing is completed, the sample is allowed to stand for a set time, and then the rheological test is started; during the test, the flow behavior images around the rotor are captured continuously using a camera.

[0026] In this technical solution, the microstructure of the freshly mixed fluid material is destroyed by pre-shearing, which effectively eliminates the initial experimental error caused by structural recovery and makes the starting state of subsequent rheological tests more consistent. After pre-shearing, the fluid is allowed to recover moderately from the disturbance, avoiding test distortion caused by excessive recovery. During the test, the flow behavior images around the rotor are continuously captured by a camera, providing a complete dynamic image sequence for subsequent feature point tracking and shear rate calculation based on deep learning, ensuring the synchronization and integrity of data acquisition.

[0027] To avoid test distortion due to excessive recovery, the time in S22 is set to no more than 30 seconds, preferably 10 seconds.

[0028] As a preferred option, the deep learning-based feature point tracking and localization process in S3 is as follows: S31: Perform perspective transformation on the original image captured by the camera to restore the elliptical sample outline to the actual circular outline, and convert the image coordinate system to a circular coordinate system with the rotor center as the origin to eliminate coordinate errors caused by the shooting angle; when using deep learning technology to mark feature points on the rotor periphery, the selection of the marking range is affected by the maximum particle size of solid particles in the fluid sample. It is recommended that the marking range be 3 times the maximum particle size and not less than 3 mm to ensure that the feature points are located in a uniform fluid area and avoid interference from particles; S32: Based on the grayscale differences in the cropped image, automatically label the set of feature points around the rotor and record the start time of the acquisition interval. and the initial coordinates of each feature point ; S33: Run a deep learning model to detect and identify feature points, retain the feature points that persist within the acquisition interval, and update their coordinates to the current time. S34: When the data acquisition interval ends, record the termination time. The updated feature point coordinates are used as the termination coordinates. .

[0029] In this technical solution, the elliptical sample outline is restored to a circle through perspective transformation and converted into a circular coordinate system centered on the rotor, effectively eliminating coordinate distortion caused by the shooting angle and ensuring the geometric accuracy of feature point position measurement. The marking range is set according to the maximum particle size of solid particles (3 times the particle size and not less than 3 mm), so that the feature points are located in a uniform fluid region, effectively avoiding the influence of local particle interference on the recognition results. The feature points on the outer periphery of the rotor are automatically marked using grayscale differences, and the feature points are continuously detected, identified and tracked within the acquisition interval by combining a deep learning model, recording the coordinates at the start and end times. This achieves high-precision and high-robust automated tracking of feature point positions, providing reliable displacement data for subsequent shear rate calculation.

[0030] As a preferred embodiment, in S4, the process for calculating the average shear rate at the rotor edge is as follows: S41: Sort and number all tracked feature points according to their average coordinate radius within the acquisition interval, from smallest to largest (the smallest radius is number 1, and the largest is number 2). Number); As a preferred option, to ensure that the number of feature points is even, if the total number of points... If the number of features is odd, delete the feature point with the largest radius. If the original number is even, no deletion is needed. This effectively ensures that the number of remaining feature points is even. S42: Pair the remaining feature points according to their sorting order, using a fixed difference value; the fixed difference value is half of the largest index of the remaining feature points, i.e., the [number]th [feature point]. Number and Number pairing, where This represents the total number of remaining feature points. ; S43: For each pair of feature points A and B (where, ), calculate the shear rate corresponding to that point using formula (1). The unit is s -1 ; (1); In the formula, , These are the average coordinate radii of feature points A and B within the acquisition interval, respectively, in mm; , These are the polar angles of feature points A and B at the beginning of the interval, respectively, in rad; , These are the polar angles of feature points A and B at the end of the interval, respectively, in rad; The data collection interval, The unit is seconds (s). S44: Calculate the average shear rate of all effective feature points to obtain the average shear rate of the rotor edge within this acquisition interval. ; (2); In the formula, This represents the number of valid feature point pairs.

[0031] In this technical solution, the symmetry and rationality of subsequent pairings are guaranteed by sorting and numbering feature points according to their average coordinate radius and ensuring that the number is even (deleting outliers with the largest radius when necessary); a fixed difference is used to assign pairs (the i-th and the i-th...). The feature points involved in the calculation are evenly spaced in the radial direction, effectively representing the flow state at different radii. By sorting the feature points by average radius, removing abnormally large radius points, and pairing them with fixed differences, it is ensured that each pair of points involved in the calculation has a reasonable radial spacing and angular displacement, making the shear rate calculation more stable and more consistent with the actual fluid flow state. The formula also utilizes the angular displacement information at the beginning and end times to eliminate the rigid body rotation error caused by the overall rotation of the rotor, thus improving the calculation accuracy. The shear rate is calculated directly based on the polar angle change and radius difference of the feature points within the acquisition interval using formula (1), avoiding the dependence of traditional theoretical formulas on the fluid constitutive model. At the same time, the error caused by the overall rigid body rotation of the rotor is eliminated by the angular displacement difference at the beginning and end times. Finally, the average shear rate is obtained by averaging the shear rates of multiple pairs of feature points, which significantly improves the stability and representativeness of the measurement results.

[0032] As a preferred option, in S7, the shear rate and shear stress data at different rotational speeds are processed and substituted into a selected rheological model (e.g., the Herschel-Bulkley model). Nonlinear regression fitting is performed to obtain the yield stress of the fluid. Consistency coefficient rheological index Rheological parameters such as...

[0033] In this technical solution, multi-rotation test data covering a wide range of shear rates can fully reflect the non-Newtonian flow characteristics of the fluid. By using nonlinear rheological models such as Herschel-Bulkley for regression fitting, three key parameters—yield stress, consistency coefficient, and rheological index—can be obtained simultaneously, making it suitable for viscous fluids with yield stress. Nonlinear regression fitting can minimize the deviation between data points and the model, improving the accuracy and reliability of rheological parameter solutions, and providing a scientific basis for accurately characterizing and evaluating the rheological properties of viscous fluids with yield stress.

[0034] The working principle of the deep learning-based method for calculating yield stress viscous fluid rheological parameters of the present invention will be described in conjunction with the following specific embodiments. Example 1:

[0035] In this embodiment, the Herschel-Bulkley model is used to fit the rheological parameters.

[0036] Freshly mixed cement mortar was prepared as a viscous fluid to be tested, exhibiting yield stress.

[0037] The prepared fresh cement mortar was thoroughly mixed and placed into the sample measuring cylinder of the rheometer. Specifically, the sample measuring cylinder was 10cm in diameter and 11cm in height, and a cross-shaped paddle rotor was selected, with a rotor diameter of 3cm and an effective shearing cylinder height of 6cm.

[0038] Set the rheometer's rotational speed to a speed control mode, where the rotational speed decreases gradually during the shearing phase. Figure 3 The process is divided into three stages: the pre-shearing stage, the settling stage, and the shearing stage.

[0039] Specifically, during the pre-shearing stage, the rotor accelerates from a standstill to 50 rpm in 10 seconds, maintains this speed for 10 seconds, and then decelerates from 50 rpm to a stop in another 10 seconds. The cement mortar is left to stand for 10 seconds before entering the shearing stage.

[0040] Five rotational speeds were set at 50 rpm, 40 rpm, 30 rpm, 20 rpm, and 10 rpm during the shearing phase. Each phase was maintained for 10 seconds and 10 measurement points were set.

[0041] The torque and shear stress corresponding to different rotational speeds during the shearing stage were recorded using a rheometer. The values ​​were obtained according to formula (3). Shear stress at a rotational speed , These correspond to 50rpm, 40rpm, 30rpm, 20rpm, and 10rpm respectively; (3); In the formula, The shear stress at the rotor edge; The torque on the rotor; This refers to the effective shear height of the rotor cylinder. The radius of the rotor is denoted as .

[0042] Within the data collection interval, fine sand particles were located and tracked. A schematic diagram illustrating the changes in the position of the fine sand particles is shown below. Figure 4 As shown, deep learning technology is used to automatically label and locate the positional changes of fine sand particles within the acquisition interval. Based on the positional changes of feature points within the acquisition interval, the shear rate of freshly mixed cement mortar is calculated non-contactly. A schematic diagram of cement mortar shear rate calculation using a set of fine sand A and B as an example is shown below. Figure 5 As shown.

[0043] The rheological parameters can be fitted by deriving the shear stress provided by the rheometer and the rheological shear rate calculated using deep learning technology, and then substituted into the rheological model.

[0044] The experimental data, including shear stress, shear rate, rotational speed, and torque, measured by the rheometer are shown in Table 1 below.

[0045] The shear rate output by the rheometer is calculated according to formula (4); (4); In the formula, For shear angular velocity, The rotor radius is... Let be the radius of the sample-holding cylinder.

[0046] The above shear rates are calculated directly from theoretical formulas and are proportional to angular velocity. Due to the presence of the slug layer, This refers to the radius of the sample-holding cylinder, not the actual shear radius, which leads to errors in the shear rate derived from the theoretical formula, and the error increases with the expansion of the slug layer.

[0047] Corrected shear rate in the prior art The calculation is performed using theoretical formulas, as shown in formula (5); (5); In the formula, This is the actual shear radius.

[0048] Table 1: Comparison of Torque, Shear Stress, and Shear Rate Measured by Rheometer at Different Rotation Speeds and Shear Angles The comparison between the shear rate calculated using the modified formula in the prior art and the shear rate calculated using deep learning technology in this invention is shown in Table 2 below.

[0049] Table 2: Comparison of Shear Rates at Different Shear Angles in, The shear rate is obtained by modifying existing technology. The shear rate is calculated using this invention.

[0050] The shear rate calculated using the modified formula in the prior art is used as the true shear rate. This is compared with the rheological shear rate calculated by deep learning in this invention. Unlike existing methods that use modified dyes to obtain the actual shear radius to obtain the shear rate, this invention uses deep learning technology to directly calculate the shear rate in the image, which is more efficient.

[0051] Based on the Herschel-Bulkley model, the rheological curves of cement mortar were fitted using ideal conditions, real conditions, and the calculation method proposed in this invention, respectively. The fitting parameters are shown in Table 3 below.

[0052] Table 3: Comparison of Fitting Results for Cement Mortar under Ideal Conditions, Real Conditions, and the Calculation Method Extracted in This Invention Among them, error 1 refers to the error between the ideal fitting parameters and the true fitting parameters, and error 2 refers to the error between the fitting parameters of the present invention and the true fitting parameters. The fitting parameters of the present invention are closer to the true fitting parameters. The fitting error of the rheological parameters using the present invention is low and the maximum error is 11.4%, which significantly improves the fitting accuracy of the rheological parameters.

[0053] In this embodiment, the results of fitting the ideal conditions, the actual conditions, and the calculations of this invention are as follows: Figure 6 As shown.

[0054] This case study uses the Herschel-Bulkley model, which is suitable for freshly mixed cement mortar. The method of this invention is also applicable to other rheological models, such as the Bingham model and the modified Bingham model, which can significantly reduce the calculation error of rheological parameters.

[0055] This invention measures the rheological properties of fluids using a coaxial rotating rheometer. During the rheometer's operation and the fluid shearing process, images of the fluid shearing are captured by a camera. Deep learning technology is then used to calculate the rheological shear rate. Subsequently, the shear stress obtained from the coaxial rotating rheometer, along with the actual shear rate calculated by deep learning, is substituted into the rheological model to obtain the rheological parameters of the viscous fluid at yield stress. This invention fully considers the influence of slug layers induced by yield stress on the shear flow within the fluid. It innovatively uses deep learning technology to first obtain the rheological shear rate at different rotor speeds from the images, effectively eliminating the influence of factors such as slug flow, slip, or wall effects when calculating the rheological shear rate using theoretical formulas. After obtaining the actual rheological shear rate, it is substituted into the rheological model to fit high-precision rheological parameters, ultimately enabling accurate calculation of the actual shear deformation rate of the fluid. This significantly improves the accuracy of rheological parameter calculation and the reliability of rheological performance characterization results.

[0056] This invention breaks through the limitations of traditional coaxial rheometers that rely on theoretical formulas to calculate shear rates. It innovatively applies deep learning technology to obtain shear rates from actual fluid flow images, fundamentally avoiding the accuracy degradation problems caused by factors such as choke flow, slippage, and wall effects. Simultaneously, through a series of techniques including perspective transformation, marker range optimization, and feature point pairing, a high-precision, robust, and fully automated rheological parameter analysis scheme is constructed. This scheme is particularly suitable for characterizing the rheological properties of complex non-Newtonian fluids containing solid particles (such as filling slurry, concrete, and mud), and has significant engineering application value.

[0057] This method is simple to implement, low in cost, highly operable, and efficient in testing. It can accurately calculate the actual shear deformation rate of the fluid, significantly improve the accuracy of rheological parameter fitting results and the reliability of rheological performance characterization results, and is of great significance for accurately characterizing and evaluating the rheological properties of viscous fluids with yield stress.

Claims

1. A deep learning-based yield stress viscous fluid rheological parameter calculation method, characterized in that, Includes the following steps; S1: Instrument preparation and parameter setting; Configure the coaxial rotational rheometer, camera and sample holding device, set the test parameters, and prepare the fluid sample to be tested; S2: Pre-shear and rheological test; The sample is pre-sheared to eliminate structural errors, and after a set time of set time, the formal rheological test begins and images of the flow process are captured simultaneously. S3: Feature point tracking and localization based on deep learning; within each acquisition interval, the deep learning model is used to automatically mark and track the positional changes of feature points on the rotor periphery, and perspective transformation is used to eliminate shooting angle errors; S4: Calculate the average shear rate at the rotor edge; based on the polar coordinate changes of feature points, calculate the shear rate of each pair of points by pairing, and then obtain the average shear rate; S5: Extracting the shear stress values recorded by the rheometer within the same collection interval, calculating the arithmetic mean thereof as the average shear stress at the rotor edge ; S6: Multi-speed testing and data processing; Repeat S2 to S5 and perform the test and calculation in sequence at different preset rotor speeds to obtain the average shear rate corresponding to each speed with the average shear stress data; S7: Rheological parameter fitting; Substitute data at various rotational speeds into the rheological model for nonlinear regression to obtain rheological parameters.

2. The method of claim 1, wherein the method is based on deep learning. In S1, the instrument preparation and parameter setting process is as follows: S11: Set the operating parameters of the coaxial rotary rheometer; S12: To prepare fluid materials, take a set amount of sample and pour it into a measuring cylinder, making the surface of the sample flush with the sampling mark line on the measuring cylinder; S13: Adjust the rotor embedment depth, adjust the camera angle and white balance to ensure that the outer area of ​​the rotor can be clearly photographed.

3. The deep learning-based yield stress viscous fluid rheological parameter calculation method according to claim 1 or 2, characterized in that, In S2, the pre-shear and rheological testing process is as follows: S21: Pre-shear the sample according to the set parameters to destroy the microstructure of the freshly mixed fluid material and eliminate experimental errors caused by structural recovery. S22: After the pre-shearing is completed, let the sample stand for a set time, and then start the rheological test. During the test, a camera was used to continuously capture images of the flow behavior around the rotor.

4. The method for calculating the yield stress rheological parameters of viscous fluids based on deep learning according to claim 3, characterized in that, In S22, the time is set to no more than 30 seconds.

5. The method for calculating the yield stress rheological parameters of viscous fluids based on deep learning according to claim 1, characterized in that, In S3, the deep learning-based feature point tracking and localization process is as follows: S31: Perform perspective transformation on the original image captured by the camera to restore the elliptical sample outline obtained by the camera to the actual circular outline, and convert the image coordinate system to a circular coordinate system with the rotor center as the origin to eliminate the coordinate error caused by the shooting angle. S32: Based on the grayscale differences in the cropped image, automatically label the set of feature points around the rotor and record the start time of the acquisition interval. and the initial coordinates of each feature point ; S33: Run a deep learning model to detect and identify feature points, retain the feature points that persist within the acquisition interval, and update their coordinates to the current time. S34: When the data acquisition interval ends, record the termination time. The updated feature point coordinates are used as the termination coordinates. .

6. The method for calculating the yield stress rheological parameters of viscous fluids based on deep learning according to claim 1, characterized in that, In S4, the process of calculating the average shear rate at the rotor edge is as follows: S41: Sort and number all tracked feature points according to their average coordinate radius within the acquisition interval from smallest to largest; S42: Pair the remaining feature points according to their sorting order, using a fixed difference value; S43: For each pair of feature points A and B, calculate the shear rate corresponding to that pair of points according to formula (1). ; （1）； In the formula, , These are the average coordinate radii of feature points A and B within the acquisition interval, respectively. , These are the polar angles of feature points A and B at the beginning of the interval, respectively; , These are the polar angles of feature points A and B at the end of the interval, respectively; The data collection interval, ; S44: Calculate the average shear rate of all effective feature points to obtain the average shear rate of the rotor edge within this acquisition interval. ; （2）； In the formula, This represents the number of valid feature point pairs.

7. The method for calculating the yield stress rheological parameters of viscous fluids based on deep learning according to claim 1, characterized in that, In S7, shear rate and shear stress data at different rotational speeds are processed and substituted into a selected rheological model for nonlinear regression fitting to obtain the yield stress of the fluid. Consistency coefficient rheological index The rheological parameters.

Images