Pressure-controlled method for predicting shear rheology of particle suspensions

Abstract

The present application relates to a kind of particle suspension shear rheological prediction method based on pressure control, steps: S1 obtains reference data, constructs simulation model and verifies;S2 constructs different gradation sample and carries out multi-working condition coupling simulation, obtains μ-Iv data and carries out preprocessing and uniform sampling, establishes original data set;S3 constructs and trains random forest shear rheological prediction model, then carries out hyperparameter optimization, forward prediction output μ-Iv curve;S4 establishes Iv partition rheological model and verifies, constructs target working condition requirement and constraint condition input module, then based on S3 forward prediction model, the inverse solving module of target reverse design is constructed, and gradation optimization is carried out, and output comprehensive optimal solution scheme;S5 the output result in S4 is back to CFD-DEM coupling simulation model for verification, then closed loop correction and form traceable data archives again.The method outputs fine particle / coarse particle addition amount or ratio adjustment amount gradation optimization scheme.

Description

Technical Field

[0001] This invention relates to the field of shear rheology prediction and intelligent modeling technology for particulate materials, and in particular to a method for predicting the shear rheology of particulate suspensions based on pressure control. Background Technology

[0002] Characterizing and predicting the rheological behavior of particulate suspension systems under pressure-controlled shear conditions is a fundamental technology in fields such as particulate slurry transport, sediment / clump flow dynamics, and particulate material processing and separation. With the development of computational fluid dynamics and the discrete element method (CFD), coupled CFD-DEM simulation has become an important means of obtaining the microscopic motion and macroscopic rheological response of particulate-fluid two-phase systems. It can output information such as shear stress, normal stress, and volume fraction under different normal pressures, shear rates, fluid viscosities, and gradations, and further construct μ–Iv... Isorheological relationships. Meanwhile, due to the high computational cost, high parameter dimensionality, and high sample acquisition cost of coupled simulation, machine learning surrogate models based on simulation data have gradually become an important direction for improving prediction efficiency. They are used to quickly predict rheological curves or key feature points from gradation parameters and operating condition parameters, thereby replacing some high-cost numerical calculations and serving engineering scheme evaluation.

[0003] Although the above technologies can achieve rheological calculation and rapid prediction under pressure-controlled shear conditions, they still have the following shortcomings for the engineering application requirements of "gradation design and scheme optimization based on target rheological behavior": (1) It mainly relies on forward calculation or forward prediction, lacking the ability of goal-oriented reverse design. The existing process is usually "given gradation and working conditions → obtain μ-Iv curve". When specific goals need to be met in engineering (such as specifying the μ level of the Iv interval, the target gel point / blockage point interval, the target shear state, etc.), it often relies on repeated manual calculation and parameter scanning, which is inefficient and does not easily converge to a feasible solution.

[0004] (2) Lack of a regionalized expression and verification mechanism for Iv global rheological behavior. Engineering conditions often correspond to different Iv intervals. Low, medium and high Iv regions may exhibit different dominant shear states. Existing technologies mostly directly fit curve point values ​​or regress on a single index, making it difficult to form a structured description of "region-state-constraint". This results in unclear expression of target constraints and insufficient accessibility discrimination ability.

[0005] (3) Lack of comprehensive evaluation and scheme ranking for engineering decision-making. Existing technologies for scheme optimization often focus on minimizing error or optimizing a single performance, and usually do not systematically incorporate economic factors such as material costs and gradation adjustment costs, making it difficult to output cost-controllable and comprehensively optimal feasible solutions while meeting rheological objectives.

[0006] (4) Lack of high-fidelity back-substitution verification and closed-loop correction mechanism. If the candidate solutions of the surrogate prediction or optimization output are not back-substituted to the high-fidelity coupled model for verification, the surrogate model error is prone to accumulation or extrapolation distortion, resulting in insufficient reliability of the reverse design results; at the same time, there is no process for closed-loop correction of the database, model or solution strategy based on the verification error, making it difficult to continuously improve the robustness of the system. Summary of the Invention

[0007] In view of the technical problems existing in the background technology of rheological prediction and gradation design of particulate suspension systems under pressure-controlled shear conditions, such as "high computational cost, only forward prediction, difficulty in expressing target constraints, lack of comprehensive cost optimization of schemes, and insufficient reliability of results", the technical problem to be solved by the present invention is to provide a pressure-controlled shear rheological prediction method for particulate suspensions. This method uses coupled simulation of computational fluid dynamics and discrete element method to obtain shear rheological data under pressure-controlled shear conditions, and combines machine learning to establish a forward rheological prediction model and a target inverse design solution method to achieve optimized determination of particle gradation parameters and operating condition parameters.

[0008] This invention is accomplished using the following technical solution: a method for predicting the shear rheology of particulate suspensions based on pressure control, comprising the following steps: S1. Obtain benchmark shear rheological data, construct and verify a CFD-DEM coupled simulation model, and determine particle parameters, fluid parameters, and loading conditions; S2. Construct samples with different gradations and conduct multi-condition coupled simulations to obtain μ-Iv data for multi-gradation and multi-pressure conditions. Then, perform preprocessing and unified sampling to establish the original dataset. S3. Divide the original dataset, construct and train a random forest shear rheology prediction model, and then optimize the hyperparameters using the Bayesian method to obtain a forward prediction model and forward predict the output μ-Iv curve; S4. Based on the μ–Iv data in S1 and S2, establish and verify the Iv partition rheological model, then construct the target working condition demand and constraint input module. On this basis, call the forward prediction model of S3 to construct the reverse solution module of the target reverse design, perform gradation optimization, and output the comprehensive optimal solution scheme and the candidate scheme ranking results. S5. Substitute the output results from S4 back into the CFD-DEM coupled simulation model for verification, then close the loop to correct and form a traceable data archive.

[0009] Furthermore, the steps for establishing and validating the Iv partition rheological model in S4 include: 1) Based on the μ–Iv data of the reference sample in S1 and the multi-stage multi-pressure working condition in S2, the Iv range is divided into low Iv region, medium Iv region and high Iv region, which correspond to three shear rheological states respectively. 2) Extract and verify the curve features and key feature points within each partition to form a partition rheological model and key indicators for each partition. These are used to characterize the differences in rheological behavior under different shear states and serve as the constraint basis for subsequent target reverse design.

[0010] Furthermore, a target operating condition requirement and constraint input module is constructed to receive user-defined target rheological performance requirements and parameter feasible domain constraints. The target rheological performance requirements include the target μ–Iv curve, the target volume fraction–Iv curve, the target gel point or blockage point range, and the target viscosity or shear stress response range. The target operating condition requirements include the target Iv partition range or the target shear rheological state requirements. The constraints include gradation constraints, pressure and shear loading constraints, fluid viscosity constraints, interaction parameter constraints, and engineering feasibility constraints.

[0011] Furthermore, the steps for outputting the comprehensive optimal solution and the ranking of candidate solutions include: 1) Using the forward prediction model obtained from S3 as a surrogate model, construct the reverse solution module for the target reverse design; 2) Under the premise of satisfying the constraints, perform inversion search with the objective μ-Iv and the key indicators of the partition as the objective function to generate one or more candidate solutions that meet the target working condition requirements and the target shear rheological state; 3) Establish an economic cost calculation module to calculate the material cost and comprehensive economic cost of candidate schemes, and construct a comprehensive evaluation index based on rheological target error and economic cost, and output the comprehensive optimal solution scheme and the ranking results of candidate schemes.

[0012] Furthermore, the steps for forward prediction of the output μ-Iv curve include: 1) Divide the original dataset into a training set, a validation set, and a test set; 2) Establish a random forest regression model, with input Iv as log 10 The model is obtained by using the gradation and working condition characteristic parameters, and the output is the friction coefficient μ. The hyperparameters are optimized using the Bayesian method with the validation set to obtain and save the forward prediction model with the best parameters μ. 3) Input the test set into the trained forward prediction model, obtain the comparison results between the predicted μ and the true μ, and calculate the model performance index to evaluate the model accuracy; 4) Combine the gradation parameters of the sample to be predicted with the pressure condition parameters and Iv log. 10 The sampling points are input into the prediction model, which outputs predicted μ–Iv curve data and can simultaneously output volume fraction–Iv curve data.

[0013] Furthermore, the steps for creating the original dataset in S2 are as follows: 1) Adjust the particle size distribution parameters of the sample to establish a discontinuous gradation sample model containing different fine particle contents and different particle size ratios. 2) Based on the CFD-DEM coupled simulation model, coupled simulations were performed on each sample under the set normal pressure control and shear loading conditions to obtain steady-state shear stress and normal stress, and the corresponding μ and Iv were calculated to obtain μ–Iv data for multi-graded multi-pressure conditions. 3) Preprocess and uniformly sample the obtained μ-Iv data, and pair the characteristic parameters such as fine particle content, particle size ratio, normal pressure, and fluid viscosity with the corresponding μ values ​​of the sampling points to establish the original dataset.

[0014] Furthermore, the steps in S1 for determining particle parameters, fluid parameters, and loading conditions include: 1) Obtain the shear stress, normal stress and shear rate data of the benchmark graded specimen under pressure-controlled shear conditions, and calculate the benchmark μ–Iv curve data; 2) A computational fluid dynamics-discrete element coupled pressure-controlled shear simulation model was established using the same gradation as the reference gradation sample. The model accuracy was verified by comparing it with the reference μ–Iv data, and then the particle parameters, fluid parameters and loading conditions were determined.

[0015] Furthermore, the specific steps of S5 include: 1) Substitute the comprehensive optimal solution or the top-ranked candidate solution output by S4 back into the pressure-controlled shear CFD-DEM coupled simulation model for positive verification to obtain the back-substitution verification rheological results. 2) Compare the error between the target rheological behavior and the back-substitution verification results, and make closed-loop corrections to the reverse solution module, the original database or the random forest prediction model based on the error results to improve the accuracy and reliability of the target reverse design. 3) Output forward prediction results, target reverse design candidate schemes, back-substitution verification results, error and credibility evaluation, scheme ranking and implementation suggestions, and structured storage of simulation parameters, training samples, model version, partition rheological model, reverse solution results and back-substitution verification results to form traceable data archives and standardized interfaces.

[0016] Furthermore, the candidate scheme includes at least a particle size distribution optimization scheme, which includes the amount of fine particles added, the amount of coarse particles added, the amount of adjustment of the coarse-fine ratio, or the amount of adjustment of the particle size distribution parameter.

[0017] The beneficial effects of this invention are: 1. Achieve rapid prediction of shear rheological behavior: By establishing a pressure-controlled shear CFD-DEM coupled simulation database and training a forward shear rheological prediction model, the μ–Iv curve can be quickly output under given gradation parameters and operating conditions (with optional output). (Curve), reducing repeated calls to high-cost coupled simulations and improving the efficiency of rheological evaluation.

[0018] 2. Provides Iv zone rheological state expression and verification for engineering working conditions, dividing the Iv range into low / medium / high zones and corresponding to different shear rheological states, forming key features and constraint expression methods for each zone, so that the target working condition requirements can be described in a structured way using zone shear states or zone key indicators, which facilitates constraint input, accessibility judgment and solution selection.

[0019] 3. It has the capability of target-oriented reverse design of gradation. Based on the forward prediction model, it builds a reverse solution module, which can solve for feasible gradation optimization schemes based on the target μ-Iv curve, target gel point or blockage point range, target shear state, etc., and output parameters such as fine / coarse particle addition amount or ratio adjustment amount, reducing the workload of manual trial calculation and parameter scanning.

[0020] 4. Supports the ranking and decision-making of the optimal performance-cost scheme, introduces economic cost calculation and comprehensive evaluation index, ranks candidate schemes and outputs the optimal solution scheme under the premise of meeting the target rheological performance requirements, so that the output results are more in line with the needs of engineering implementation and decision-making.

[0021] 5. Improve the reliability of reverse design results and support closed-loop correction. Perform CFD-DEM back-substitution verification on candidate solutions obtained by reverse solving, and perform closed-loop correction on the database, prediction model or solution strategy based on the verification error. This can reduce the risk of deviation caused by extrapolation of surrogate models and improve the credibility and engineering applicability of design results. Attached Figure Description

[0022] Figure 1 The flowchart shows a pressure-controlled method for predicting the shear rheology of particulate suspensions. Figure 2 The comparison chart between the reference data and the coupled simulation model shows the relationship between the friction coefficient-inertial viscosity number curve of the reference sample and the simulation results. Figure 3 This is a schematic diagram of the pressure-controlled shear CFD-DEM model simulation and boundary loading method, showing the upper boundary normal pressure control and horizontal shear velocity loading method; Figure 4 The graph shows the hyperparameter optimization results of the random forest shear rheology prediction model, illustrating the relationship between the number of decision trees, the maximum tree depth, and the mean absolute error. Figure 5 A scatter plot comparing the predicted and actual values ​​of the test set shows the fitting effect between the predicted and actual friction coefficients, and includes the coefficient of determination and error index. Figure 6 The graph shows a comparison of the predicted effects of the friction coefficient-inertial viscosity coefficient relationship curves under different gradations and pressure conditions. A, b, c, and d correspond to the comparison results of the actual curves and predicted curves under different combinations of particle size ratios and fine particle content, respectively. Figure 7 A comparison of μ–Iv curves for different candidate schemes. Detailed Implementation

[0023] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0024] refer to Figures 1-7 As shown: This invention provides a method for predicting the shear rheology of particulate suspensions based on pressure control, the steps of which are as follows: S1. Determine particle parameters, fluid parameters, and loading conditions. The steps are as follows: S11. Obtain baseline shear rheological data.

[0025] Shear stress, normal stress, and shear rate data of the benchmark graded specimen were obtained under pressure-controlled shear conditions, and benchmark μ–Iv curve data were calculated.

[0026] Specifically, a set of shear rheological data of a reference gradation sample is obtained. The reference sample is placed in a shearing device at the upper and lower boundaries and shear loading is performed under a given normal pressure control condition. By recording parameters such as shear stress, normal stress, shear rate and shear thickness in the steady state stage, the μ–Iv curve of the sample is calculated to obtain the reference shear rheological data.

[0027] μ and Iv are preferably defined and calculated as follows: Friction coefficient: μ=τ / σn, where τ is the shear stress and σn is the normal stress; Inertial viscosity number: Iv = γ˙ηf / σn, where γ is the actual shear rate and ηf is the hydrodynamic viscosity.

[0028] Preferably, to ensure statistical stability, the data in the second half of the steady-state phase are averaged over time; the shear thickness is preferably the average value of the second half of the steady-state phase, thereby obtaining the true shear rate γ=Δv / h, where Δv is the relative shear velocity between the upper and lower boundaries, and h is the distance between the upper and lower boundaries (shear thickness).

[0029] S12. Construct and validate a CFD-DEM coupled simulation model, and determine particle parameters, fluid parameters, and loading conditions.

[0030] Using the same gradation as the reference gradation sample in S11, a pressure-controlled shear rheological simulation model (CFD-DEM coupled simulation model) was established using computational fluid dynamics and discrete element method (CFD-DEM coupled simulation model) to obtain the μ-Iv curve of the sample. The model's accuracy was verified by comparing it with the reference μ-Iv data described in S11, thereby determining particle parameters, fluid parameters, and loading conditions. The comparison of the μ-Iv curves between the numerical simulation and the reference data is as follows: Figure 2 As shown.

[0031] Specifically, S12 includes, but is not limited to, the following steps: 1) To meet the steady-state shearing condition, the moving speed of the upper and lower boundaries cannot be too fast; at the same time, if the loading speed is too low, it will increase the calculation cost. Therefore, it is preferable to select a suitable shearing speed under the premise of ensuring stability. The preferred relative shearing speed Δv of the upper and lower boundaries is 0.028m / s, or a value in the range of 0.005m / s to 0.05m / s.

[0032] 2) To maintain computational stability, the time step of the coupled simulation must be smaller than the limiting time step; the preferred time step Δt is 1×10⁻⁶. -6 s, or in 1×10 -7 s~2×10 -6 It takes values ​​within the range of s.

[0033] 3) To avoid boundary effects, the ratio of the sample feature size to the maximum particle diameter is preferably not less than 10; preferably, the maximum particle diameter dmax is 1.1 mm and the simulated domain feature size L is 24 mm, so that L / dmax≥10.

[0034] 4) Optimal particle parameters include a particle density ρp of 2650 kg / m³ and a particle Young's modulus Ep of 5 × 10⁻⁶. 7 Pa ~ 5 × 10 9 Pa and preferably 1×10 8 Pa, Poisson's ratio ν is preferably 0.28, coefficient of restitution e is preferably 0.20, sliding friction coefficient μs is preferably 0.32, rolling friction coefficient μr is preferably 0.10; fluid parameters preferably include a fluid density ρf of 1000 kg / m³. 3 The dynamic viscosity ηf is 1×10 -3 Pa·s ~ 5 × 10 -2 Pa·s and preferably 1×10 -3 Pa·s; preferred boundary conditions include the target normal pressure σ target The pressure ranges from 200Pa to 600Pa. Normal pressure servo control is used to keep the normal average stress within the preset tolerance range, while shear loading is controlled by shear rate.

[0035] S2. Create the original dataset.

[0036] Adjust the particle parameters of the sample in S1, construct samples with different gradations, and conduct multi-condition coupled simulations to obtain μ-Iv data for multiple gradations and multiple pressure conditions. Then, perform preprocessing and unified sampling to establish the original dataset. The specific steps are as follows: S21. Construct sample models with different gradations and set gradation parameters.

[0037] Based on the coupled simulation model verified in S2, the particle gradation parameters in the sample were adjusted to establish a discontinuous gradation sample model containing different fine particle contents (FC) and different particle size ratios (SR). The sample particles consist of coarse and fine particles.

[0038] Specifically, samples with a specified fine particle content and particle size ratio can be randomly generated within a given simulation domain, where the position of each particle is randomly distributed. The particle size ratio and fine particle content are determined by the following formulas.

[0039] In the above formula, SR represents the particle size ratio, and d max and d min These represent the diameters of the coarse and fine particles, respectively; FC represents the fine particle content; m fine and m all These represent the mass of the fine particles and the mass of all particles, respectively.

[0040] S22. Obtain shear rheological μ-Iv data under different gradations and pressure conditions.

[0041] Based on the CFD-DEM coupled simulation model, coupled simulations were performed on samples with different gradations in S21 under set normal pressure control and shear loading conditions to obtain steady-state shear stress and normal stress, and calculate the corresponding μ and Iv, thus obtaining shear rheological μ–Iv data under multi-gradation and multi-pressure control conditions. Specific steps include: 1) After the sample is generated, normal pressure servo stabilization is performed to make the sample reach a stable state under the target normal pressure, forming a dense and macroscopically stable initial structure. 2) Under the condition of maintaining the target normal pressure, apply different shear rates (or shear speeds) and record the shear stress, normal stress, shear speed, shear thickness and volume fraction in the steady state stage. 3) The steady-state shear stress and steady-state normal stress are obtained by time averaging, and μ is calculated; the actual shear rate is calculated by shear velocity and shear thickness, and Iv is calculated by combining fluid viscosity and normal stress; 4) Repeat the above process for multiple Iv conditions to obtain a complete μ-Iv curve corresponding to the same set of gradation parameters (and simultaneously obtain...). curve).

[0042] S23. Establish the original dataset.

[0043] The μ–Iv data obtained from S21 is preprocessed and uniformly sampled. Characteristic parameters such as fine particle content, particle size ratio, normal pressure, and fluid viscosity are paired with the corresponding μ values ​​of the sampling points to establish the original dataset. Specific steps include: 1) Directly using the unprocessed raw μ-Iv curves as input and output for machine learning models significantly increases the training burden and causes inconsistencies in the number of points on different curves. Therefore, the μ-Iv data obtained in S4 is preprocessed to obtain simplified μ-Iv curves, and then the μ-Iv curves are uniformly sampled to ensure that different curves have the same number of points on the x-axis. Preferably, in log... 10 The (Iv) axis is divided into N equal parts (e.g., 100 parts), and the corresponding N fixed-interval abscissa points are used to represent a μ–Iv curve; 2) Convert the log of each sampling point obtained in 1) 10 (Iv) and gradation characteristic parameters (e.g., SR, FC, d) max d min The input feature vector is formed by concatenating the target normal pressure, fluid viscosity, initial volume fraction, etc., and the output label is the corresponding μ value, thereby obtaining samples of uniform dimension and establishing the original database.

[0044] S3. Forward prediction model and forward prediction output μ-Iv curve.

[0045] The original dataset in S2 is partitioned, and a random forest shear rheology prediction model is constructed and trained. Then, hyperparameter optimization is performed using the Bayesian method to obtain the forward prediction model and forward predict the output μ-Iv curve. Specific steps include: S31. Divide the original dataset.

[0046] The original dataset in S2 is divided into a training set, a validation set, and a test set. The preferred ratio of the training set, validation set, and test set is 8:1:1.

[0047] Specifically, the dataset can be split by importing the "train_test_split" function from the "sklearn.model_selection" module using a Python script, and setting "random_state=42" to ensure that the results are reproducible.

[0048] S32. Forward prediction model.

[0049] Construct a random forest regression model, where the input Iv to the random forest model is log. 10The gradation and operating condition characteristic parameters are used to output the corresponding friction coefficient μ. A random forest model is trained to obtain the μ prediction model. Hyperparameters are optimized using Bayesian method with a validation set to obtain and save the forward prediction model of μ with the best parameters.

[0050] Specifically, the random forest regression model is an ensemble technique that incorporates multiple decision trees. Its output is the average of the predictions from all trees, which effectively reduces overfitting and improves generalization ability. The steps for building a random forest model include: 1) Randomly select samples from the training set; 2) Generate multiple decision trees and create a model by calling the random forest regressor through a Python script; 3) The data subset for each decision tree is generated through a bootstrap sampling method; 4) Take the average of all decision tree prediction results to obtain the final prediction value.

[0051] Bayesian hyperparameter optimization preferably uses a Bayesian search approach, progressively updating the posterior distribution based on historical evaluation results to find a better combination of hyperparameters with fewer iterations. It prioritizes tuning hyperparameters such as "n_estimators" and "max_depth". The Bayesian optimization results are as follows: Figure 6 As shown in Table 1.

[0052] Table 1. Results of Bayesian Hyperparameter Optimization

[0053] S33. Evaluate the trained forward prediction model (random forest model).

[0054] Based on the optimal hyperparameters determined in S32, the test set data is input into the trained forward prediction model to obtain the comparison results between the predicted μ and the true μ, and the model performance index is calculated to evaluate the model accuracy.

[0055] Specifically, model performance evaluation metrics include goodness of fit R. 2 The mean absolute error (MAE) is calculated using the following formula:

[0056] In the formula, For predicted values, The true value is n, and the sample size is n. This is the average of the true values.

[0057] Reference Appendix Figure 5 It can be seen that by comparing the predicted values ​​of the random forest model with the actual values, Figure 5 Chinese R 2MAE and 10% tolerance represent the coefficient of determination, mean absolute error, and prediction accuracy with a 10% tolerance, respectively. The closer the scatter points are to the diagonal, the closer the predicted values ​​are to the true values, and the better the model prediction performance.

[0058] S34. Predict the pressure-controlled shear rheology μ–Iv curve and output the results.

[0059] The gradation parameters of the sample to be predicted are combined with the pressure condition parameters (including SR, FC, coarse and fine particle size, target normal pressure, fluid viscosity, etc.) and Iv log. 10 The sampling points are input into the S32-trained prediction model, which outputs predicted μ–Iv curve data, and can further output... The curve and other shear rheological response parameters can be output synchronously, along with volume fraction – Iv curve data.

[0060] refer to Figure 6 As shown, Figure 6 Figures a, b, c, and d show the comparison results of μ-Iv curves for different combinations of particle size ratios and fine particle content, respectively. The solid line represents the actual curve obtained from the coupled simulation, and the dashed line represents the curve predicted by the random forest model. It can be seen that the random forest model has good prediction performance for μ-Iv curves under different gradations and pressure control conditions, and the predicted values ​​are basically consistent with the actual values.

[0061] S4. Output the comprehensive optimal solution and the ranking results of candidate solutions.

[0062] Based on the μ–Iv data in S1 and S2, an Iv partitioned rheological model is established and validated, serving as the partitioned constraint basis for the target reverse design. A target operating condition requirement and constraint input module is then constructed to receive the target rheological performance requirements and parameter feasible region constraints. Building upon this, the forward prediction model in S3 is invoked to construct a reverse solution module for the target reverse design, enabling candidate solution search and ranking. Gradation optimization is performed, outputting the comprehensive optimal solution and the candidate solution ranking results.

[0063] S41. Establish and validate the Iv partition rheological model. Specific steps include: 1) Based on the μ-Iv data of the reference specimen shear rheology in S1 and the μ-Iv data of the multi-stage multi-pressure working condition in S2, the Iv range is divided into low Iv region, medium Iv region and high Iv region, which correspond to three shear rheological states respectively. 2) Extract and verify the curve features and key feature points within each partition to form a partition rheological model and key indicators for each partition. These are used to characterize the differences in rheological behavior under different shear states and serve as the constraint basis for subsequent target reverse design.

[0064] S42. Construct a target operating condition requirement and constraint input module to receive the target rheological performance requirements and parameter feasible domain constraints set by the user.

[0065] The target rheological performance requirements include the target μ–Iv curve, the target volume fraction–Iv curve, the target gel point or blockage point range, and the target viscosity or shear stress response range. The target operating condition requirements include the target Iv partition range or the target shear rheological state requirements. The constraints include gradation constraints, pressure and shear loading constraints, fluid viscosity constraints, interaction parameter constraints, and engineering feasibility constraints.

[0066] S43. Output the comprehensive optimal solution and the ranking results of candidate solutions. The specific steps include: 1) Using the forward prediction model obtained from S3 as a surrogate model, construct the reverse solution module for the target reverse design; 2) Under the premise of satisfying the constraints, an inversion search is performed with the target μ–Iv and the key indicators of the partition as the objective function to generate one or more candidate schemes that meet the target working condition requirements and the target shear rheological state. The candidate schemes include at least a particle size distribution optimization scheme, which includes the amount of fine particles added, the amount of coarse particles added, the amount of coarse-fine ratio adjustment, or the amount of particle size distribution parameter adjustment. 3) Establish an economic cost calculation module to calculate the material cost and comprehensive economic cost of candidate schemes, and construct a comprehensive evaluation index based on rheological target error and economic cost, and output the comprehensive optimal solution scheme and the ranking results of candidate schemes.

[0067] The project includes an economic cost calculation module to quantitatively evaluate the engineering implementation costs of each candidate scheme. Specifically, based on the amount of fine particles added, coarse particles added, the adjustment amount of the coarse-fine ratio, or the adjustment amount of particle size distribution parameters in the candidate schemes, combined with the corresponding material unit price, preparation and processing costs, and implementation and operation costs, the material cost and comprehensive economic cost of each candidate scheme are calculated. This comprehensive economic cost, along with the rheological target error, is used as a comprehensive evaluation criterion to rank the candidate schemes, outputting the comprehensive optimal solution and the ranking results of the candidate schemes.

[0068] The economic cost calculation module can be represented as follows: C(x) = Cm(x) + Cp(x) + Co(x) in, C(x) represents the comprehensive economic cost of candidate solution x; Cm(x) represents the material cost, which characterizes the cost incurred by adjusting particulate materials, adding fine particles, adding coarse particles, reorganizing particle size distribution, or other material inputs in the candidate scheme. Cp(x) is the preparation or processing cost, used to characterize the costs incurred during particle sieving, mixing, grading adjustment, sample preparation or process processing; Co(x) represents the implementation or operating cost, characterizing the costs incurred during simulation implementation, parameter loading, fluid medium adjustment, operating condition control, or engineering operation under the target operating conditions. Through the above cost breakdown calculations, the comprehensive economic cost of each candidate scheme under the condition of satisfying the rheological target can be obtained.

[0069] Furthermore, to achieve a joint evaluation of candidate solutions in terms of rheological performance and economic cost, a comprehensive evaluation function is constructed: F(x) = αJ(x) + βC(x) Wherein, F(x) is the comprehensive evaluation index of candidate scheme x, J(x) is the error function of candidate scheme x relative to the target rheological behavior, used to characterize the degree of deviation of the scheme in the target μ−Iv curve, target volume fraction-Iv curve, target gel point or blockage point range, target viscosity response range, and other key rheological indicators; C(x) is the comprehensive economic cost; α and β are weighting coefficients used to adjust the relative weight of rheological performance target and economic cost target in the comprehensive evaluation. The weighting coefficients can be preset according to actual engineering needs, application scenarios, design goals, or user preferences, and can also be adjusted according to different working conditions.

[0070] Under the premise of meeting the target operating conditions and constraints, the error function J(x), comprehensive economic cost C(x), and comprehensive evaluation index F(x) of multiple candidate solutions are calculated respectively, and the candidate solutions are ranked according to the magnitude of the comprehensive evaluation index. Preferably, the smaller F(x) is, the better the comprehensive economic performance of the corresponding candidate solution while meeting the target rheological behavior requirements. Therefore, the candidate solution with the best comprehensive evaluation index can be determined as the comprehensive optimal solution, and the remaining solutions can be output as alternative solutions, thereby realizing gradation optimization and target reverse design that takes into account both rheological performance and engineering feasibility.

[0071] S5. Substitute the output results from S4 back into the CFD-DEM coupled simulation model for verification, then perform closed-loop correction and create a traceable data archive. Specific steps include: S51. Substitute the comprehensive optimal solution or the top-ranked candidate solution output by S4 back into the pressure-controlled shear CFD-DEM coupled simulation model for positive verification and obtain the back-substitution verification rheological results. S52. Compare the error between the target rheological behavior and the back-substitution verification results, and make closed-loop corrections to the reverse solution module, the original database or the random forest prediction model based on the error results to improve the accuracy and reliability of the target reverse design. S53. Output forward prediction results, target reverse design candidate schemes, back-substitution verification results, error and credibility evaluation, scheme ranking and implementation suggestions, and structured storage of simulation parameters, training samples, model version, partition rheological model, reverse solution results and back-substitution verification results to form a traceable data archive and standardized interface.

[0072] refer to Figure 7 As shown, Figure 7 The diagram illustrates the selection of schemes S4 and S5, showing the comparison results of μ–Iv curves corresponding to different candidate schemes.

[0073] The graph shows Iv on the horizontal axis and μ on the vertical axis. Three representative μ-Iv curves—the initial curve, the high-fluidity scheme, and the high-strength scheme—are presented to characterize the differences in shear rheological response among different candidate schemes. By comparing the relationship between the curves of each candidate scheme and the target rheological requirement, the differences in fluidity, strength characteristics, and target adaptability of the candidate schemes output by the inverse solution module in S4 can be intuitively reflected. Furthermore, this graph can also be used to illustrate the matching between the rheological response of different schemes and the target requirements when back-substituting the top-ranked schemes in S5, thus providing a visual basis for the comprehensive optimal solution and the ranking results of candidate schemes.

[0074] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A pressure-controlled method for predicting the shear rheology of particulate suspensions, characterized by: The steps are as follows: S1. Obtain baseline shear rheological μ–Iv data, construct and validate a CFD-DEM coupled simulation model, and determine particle parameters, fluid parameters, and loading conditions; S2. Construct samples with different gradations and conduct multi-condition coupled simulations to obtain μ-Iv data for multi-gradation and multi-pressure conditions. Then, perform preprocessing and unified sampling to establish the original dataset. S3. Divide the original dataset, construct and train a random forest shear rheology prediction model, and then optimize the hyperparameters using the Bayesian method to obtain a forward prediction model and forward predict the output μ-Iv curve; S4. Based on the μ–Iv data in S1 and S2, establish and verify the Iv partition rheological model, then construct the target working condition demand and constraint input module. On this basis, call the forward prediction model of S3 to construct the reverse solution module of the target reverse design, perform gradation optimization, and output the comprehensive optimal solution scheme and the candidate scheme ranking results. S5. Substitute the output results from S4 back into the CFD-DEM coupled simulation model for verification, then close the loop to correct and form a traceable data archive.

2. The method for predicting the shear rheology of particulate suspensions based on pressure control according to claim 1, characterized in that: The steps for establishing and validating an Iv partitioned rheological model in S4 include: Based on the μ–Iv data of the reference sample in S1 and the multi-graded multi-pressure conditions in S2, the Iv range is divided into low Iv region, medium Iv region and high Iv region, which correspond to three shear rheological states respectively. Curve features and key feature points within each partition are extracted and verified to form partition rheological models and key indicators for each partition. These are used to characterize the differences in rheological behavior under different shear states and serve as constraints for subsequent target reverse design.

3. The method for predicting the shear rheology of particulate suspensions based on pressure control according to claim 2, characterized in that: A target operating condition requirement and constraint input module is constructed to receive user-defined target rheological performance requirements and parameter feasible domain constraints. The target rheological performance requirements include the target μ-Iv curve, the target volume fraction-Iv curve, the target gel point or blockage point range, and the target viscosity or shear stress response range. The target operating condition requirements include the target Iv partition range or the target shear rheological state requirements. The constraints include gradation constraints, pressure and shear loading constraints, fluid viscosity constraints, interaction parameter constraints, and engineering feasibility constraints.

4. The pressure-controlled shear rheology prediction method for particulate suspensions according to claim 2 or 3, characterized in that: The steps for outputting the comprehensive optimal solution and the ranking of candidate solutions include: Using the forward prediction model obtained from S3 as a surrogate model, a reverse solution module for the target reverse design is constructed. Under the premise of satisfying the constraints, an inversion search is performed with the objective μ–Iv and the key indicators of the partition as the objective functions to generate one or more candidate solutions that meet the target working condition requirements and the target shear rheological state. An economic cost calculation module is established to calculate the material cost and comprehensive economic cost of candidate schemes. Based on the rheological target error and economic cost, a comprehensive evaluation index is constructed, and the comprehensive optimal solution and the ranking results of candidate schemes are output.

5. The method for predicting the shear rheology of particulate suspensions based on pressure control according to claim 1, characterized in that: The steps for forward prediction of the output μ-Iv curve include: 1) Divide the original dataset into a training set, a validation set, and a test set; 2) Establish a random forest regression model, with input Iv as log 10 The model is obtained by using the gradation and working condition characteristic parameters, and the output is the friction coefficient μ. The hyperparameters are optimized using the Bayesian method with the validation set to obtain and save the forward prediction model with the best parameters μ. 3) Input the test set into the trained forward prediction model, obtain the comparison results between the predicted μ and the true μ, and calculate the model performance index to evaluate the model accuracy; 4) Combine the gradation parameters of the sample to be predicted with the pressure condition parameters and Iv log. 10 The sampling points are input into the prediction model, which outputs predicted μ–Iv curve data and can simultaneously output volume fraction–Iv curve data.

6. The method for predicting the shear rheology of particulate suspensions based on pressure control according to claim 1, characterized in that: S2 The steps to create the original dataset are as follows: 1) Adjust the particle size distribution parameters of the sample to establish a discontinuous gradation sample model containing different fine particle contents and different particle size ratios. 2) Based on the CFD-DEM coupled simulation model, coupled simulations were performed on each sample under the set normal pressure control and shear loading conditions to obtain steady-state shear stress and normal stress, and the corresponding μ and Iv were calculated to obtain μ–Iv data for multi-graded multi-pressure conditions. 3) Preprocess and uniformly sample the obtained μ-Iv data, and pair the fine particle content, particle size ratio, normal pressure, fluid viscosity characteristic parameters with the corresponding μ values ​​of the sampling points to establish the original dataset.

7. The method for predicting the shear rheology of particulate suspensions based on pressure control according to claim 1, characterized in that: The steps for determining particle parameters, fluid parameters, and loading conditions in S1 include: 1) Obtain the shear stress, normal stress and shear rate data of the reference gradation specimen under pressure-controlled shear conditions, and calculate the reference μ–Iv curve data; 2) A computational fluid dynamics-discrete element coupled pressure-controlled shear simulation model was established using the same gradation as the reference gradation sample. The model accuracy was verified by comparing it with the reference μ–Iv data, and then the particle parameters, fluid parameters and loading conditions were determined.

8. The method for predicting the shear rheology of particulate suspensions based on pressure control according to claim 1, characterized in that: The specific steps of S5 include: 1) Substitute the comprehensive optimal solution or the top-ranked candidate solution output by S4 back into the pressure-controlled shear CFD-DEM coupled simulation model for positive verification to obtain the back-substitution verification rheological results. 2) Compare the error between the target rheological behavior and the back-substitution verification results, and make closed-loop corrections to the reverse solution module, the original database or the random forest prediction model based on the error results to improve the accuracy and reliability of the target reverse design. 3) Output forward prediction results, target reverse design candidate schemes, back-substitution verification results, error and credibility evaluation, scheme ranking and implementation suggestions, and structured storage of simulation parameters, training samples, model version, partition rheological model, reverse solution results and back-substitution verification results to form traceable data archives and standardized interfaces.

9. The method for predicting the shear rheology of particulate suspensions based on pressure control according to claim 4, characterized in that: The candidate schemes include at least a particle size distribution optimization scheme, which includes the amount of fine particles added, the amount of coarse particles added, the amount of adjustment of the coarse-fine ratio, or the amount of adjustment of the particle size distribution parameter.

Images