A method for determining the occurrence boundary of liquid column separation and risk level
By combining dimensionless analysis and machine learning, the problem of accurately determining the liquid column separation boundary and risk level was solved, enabling rapid and accurate liquid column separation determination and risk assessment, thus improving system safety.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- ZHEJIANG SCI-TECH UNIV
- Filing Date
- 2026-05-13
- Publication Date
- 2026-07-21
AI Technical Summary
Existing technologies are insufficient to accurately determine the boundaries and risk levels of liquid column separation. Traditional methods are costly, time-consuming, and lack quantitative assessment criteria, making it impossible to precisely assess the severity of cavitation.
A method combining dimensionless analysis and machine learning was adopted. Through a visualized liquid column separation experiment, a set of dimensionless parameters was constructed. The importance of the parameters was trained using machine learning algorithms, regression analysis was performed to fit the limits, and a composite judgment parameter was constructed for risk level assessment.
It enables quantitative determination of liquid column separation, provides a quantitative reference for system safety margin design, improves determination speed and accuracy, and allows for refined evaluation based on the interaction of system parameters.
Smart Images
Figure CN122171165B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of fluid mechanics and transient process analysis technology of hydraulic systems, specifically a method for determining the boundary and risk level of liquid column separation, as well as a system and computer-readable storage medium for applying this method. Background Technology
[0002] In engineering fields involving liquid transportation, such as rocket engine turbopumps, long-distance water pipelines, and various hydraulic systems, liquid column separation caused by transient flow (such as water hammer) is a highly destructive and complex physical process. When the local pressure within the pipeline suddenly drops below the liquid's saturated vapor pressure, the liquid vaporizes, forming large cavitation bubbles that interrupt the continuous liquid column. Subsequently, these cavitation bubbles collapse under high pressure, generating extremely high local pressure and shock waves, leading to pipeline vibration, noise, cavitation erosion, and even catastrophic damage to the system structure. Therefore, accurately predicting and determining the boundaries and states of liquid column separation is crucial for the safe design and stable operation of hydraulic systems, especially rocket engine turbopumps under extremely harsh conditions.
[0003] Currently, the industry's prediction and determination of liquid column separation phenomena mainly rely on complex computational fluid dynamics simulations or traditional empirical threshold methods. However, these existing technologies have significant limitations in practical engineering applications: First, traditional fluid dynamics simulations are costly and time-consuming, making it difficult to meet the needs of rapid assessment of complex pipeline conditions in engineering projects. Second, conventional judgment methods based on empirical thresholds lack universality. Because liquid column separation is influenced by the complex coupling of numerous physical parameters such as pipe diameter, flow velocity, pressure, and acceleration, existing methods often fail to extract the essential characteristics affecting the separation phenomenon from the complex surface variables, resulting in insufficient analysis of the degree of influence of each parameter and overly simplistic or vague judgment conditions. Furthermore, existing judgment techniques typically only reach a superficial qualitative stage of roughly predicting whether separation has occurred, lacking a set of quantitative evaluation criteria with clear physical meaning. In practical applications, completely eliminating cavitation is often unrealistic; engineering requires finding a reasonable balance. Once liquid column separation occurs beyond the safety limit, current assessment methods cannot further refine the assessment of the severity of cavitation (e.g., slight or severe) based on the interaction of system parameters. This creates a significant blind spot in the design of system safety margins and subsequent risk management, thus requiring urgent solutions. Summary of the Invention
[0004] To address the technical problem of accurately defining the critical equilibrium point of cavitation and assessing its risk severity in existing technologies, this invention provides a method for determining the boundary and risk level of liquid column separation.
[0005] To achieve the above objectives, the present invention provides the following technical solution: This invention discloses a method for determining the threshold and risk level of liquid column separation, comprising: S1. Using a visual liquid column separation test bench, conduct multiple tests under different combinations of test parameters, collect test parameters, and simultaneously record the corresponding liquid column separation status; wherein, the liquid column separation status includes separation occurring and no separation occurring; S2. Based on Buckingham's π theorem, select the repeated and residual variables in the experimental parameters, and combine the dimensional experimental parameters through dimensional analysis to construct multiple independent dimensionless numbers, forming a set of dimensionless parameters; S3. Using the dimensionless parameter set as input features and the liquid column separation state as the target variable, a machine learning algorithm is introduced for multiple batches of independent training, and the importance score of each dimensionless parameter is calculated to quantify the sensitivity. S4. Select the dimensionless number with the highest importance score and the most stable ranking from the dimensionless parameter group as the dominant dimensionless number; S5. Near the critical region of the dominant dimensionless number, the regression analysis method is used to fit the function of the boundary between the data points that have separated and those that have not separated, and the intersection of the boundary fitting function and the preset judgment threshold line is solved. The dimensionless critical equilibrium point is output as the quantitative benchmark value of the judgment boundary. S6. After confirming that liquid column separation has occurred, select multiple dimensionless parameters with importance scores higher than the set threshold and combine them to construct a composite judgment parameter for assessing the severity. Based on the relationship between the composite judgment parameter and the preset level threshold, the liquid column separation is divided into different risk levels.
[0006] As a further improvement to the above scheme, in step S3, the machine learning algorithm is the random forest regression algorithm; the method for performing multiple batches of independent training is as follows: randomly select 80% of the total data as training samples according to a preset ratio, repeat the model training multiple times using different data combinations in different training processes, and extract the importance score corresponding to each dimensionless parameter in each training.
[0007] As a further improvement to the above scheme, in step S5, the regression analysis method is logistic regression or linear regression; the preset judgment threshold line is a horizontal or vertical line determined according to the preset occurrence probability; the boundary fitting function is a curve or a straight line; the intersection of the preset judgment threshold line and the boundary fitting function is the dimensionless critical equilibrium point.
[0008] As a further improvement to the above scheme, in step S6, the method for constructing the composite determination parameters is as follows: Set the preset sensitivity threshold to 0.5. Based on the importance score calculated in step S3, remove dimensionless parameters whose importance score is less than or equal to the threshold, and select the remaining dimensionless parameters as candidate parameters. Multiple sets of correction ratios are constructed by combining the candidate parameters in pairs or in groups of three. The importance score of each group of correction ratios is calculated according to the training method in step S3. The correction ratio with the highest importance score and the most stable ranking is selected as the composite judgment parameter.
[0009] As a further improvement to the above scheme, in step S6, regression analysis is used to fit the data point boundaries between severe and slight separation, and the corresponding critical intersection point is solved as the preset level threshold range; the composite judgment parameter is compared with the preset level threshold range, and the risk level of liquid column separation is output according to the range in which the composite judgment parameter falls.
[0010] As a further improvement to the above scheme, the following steps are included before step S1: Identify the target research object, which includes liquid delivery pipelines; Obtain pipeline data of the target research object under actual operation or simulated working conditions, and statistically derive typical values of the fluid at a specific pipeline location. The typical values include at least the specific pipe diameter of the actual equipment, and the average velocity and acceleration of the fluid when it flows through the location. Based on the typical values, working conditions are mapped, and multiple sets of test parameter combinations that can cover the actual working condition variation range of the target research object are configured on the visual liquid column separation test bench.
[0011] As a further improvement to the above scheme, in step S1, the test parameters include pipe diameter, initial velocity, acceleration, bubble characteristic height, fluid density, and pressure; in step S2, the initial velocity, pipe diameter, and fluid density are selected as repeating variables, and the acceleration, bubble characteristic height, and pressure are selected as residual variables to construct at least 6 independent dimensionless numbers as the dimensionless parameter set.
[0012] As a further improvement to the above scheme, the specific addition of multiple sets of test parameter combinations on the visual liquid column separation test bench that can cover the actual operating condition variation range of the target research object includes: The pipe diameter, fluid density, and bubble characteristic height in the experimental parameters are used as the static intrinsic parameters of the target research object. The initial velocity, acceleration, and pressure in the test parameters are used as dynamic operating parameters; When configuring a single-batch test matrix, the static intrinsic parameters are kept constant, and the typical values obtained from statistics are used as a benchmark. The dynamic operating parameters are adjusted up and down according to the preset distribution gradient, and several different combinations of test parameters are generated by cross-combining variables.
[0013] This invention also discloses a system for determining the occurrence limit and risk level of liquid column separation, used to perform the method for determining the occurrence limit and risk level of liquid column separation as described above. The system includes: The data acquisition module is used to collect test parameters and simultaneously record the corresponding liquid column separation status when conducting tests using the visual liquid column separation test bench. The data processing module, based on Buckingham's π theorem, selects repeating and residual variables from the experimental parameters. Through dimensional analysis, it combines the dimensional experimental parameters to construct multiple independent dimensionless numbers, forming a dimensionless parameter set. Using this dimensionless parameter set as input features and the liquid column separation state as the target variable, a machine learning algorithm is introduced for multiple batches of independent training to calculate the importance score of each dimensionless parameter to quantify sensitivity. The dimensionless number with the highest importance score and most stable ranking is selected from the dimensionless parameter set as the dominant dimensionless number. Near the critical region of the dominant dimensionless number, regression analysis is used to fit a function to the boundary between data points where separation has occurred and those where separation has not occurred. The intersection of this boundary fitting function and a preset judgment threshold line is solved, and the dimensionless critical equilibrium point is output as a quantitative benchmark value for determining the occurrence boundary. After confirming liquid column separation, multiple dimensionless parameters with importance scores higher than the set threshold are selected and combined to construct a composite judgment parameter for assessing severity. Based on the relationship between the composite judgment parameter and the preset level threshold, liquid column separation is classified into different risk levels.
[0014] The present invention also discloses a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method for determining the liquid column separation occurrence limit and risk level as described above.
[0015] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention integrates dimensionless analysis with machine learning. On one hand, by selecting dominant dimensionless numbers and performing regression fitting, it outputs a dimensionless critical equilibrium point for determining cavitation, overcoming the limitations of traditional empirical thresholds which have weak universality and ambiguous critical points. On the other hand, after confirming separation, it selects highly sensitive parameters to construct composite judgment parameters for secondary classification. This mechanism can output a quantitative risk level of severe or slight separation based on the interaction between system parameters, providing a quantitative reference for the design of pipeline safety margins in various hydraulic systems.
[0016] 2. In constructing composite decision parameters, this invention uses importance scores derived from multiple independent training sessions to eliminate weakly correlated variables by setting preset thresholds and applying restrictive combinations of parameters (such as pairwise or triadic combinations). This mechanism effectively avoids the problem of a surge in computational load caused by unrestricted variable combinations, improving the computation and decision-making speed of the system program. Simultaneously, the generated correction ratios retain the physical meaning of nonlinear coupling between parameters (such as inertial force, pressure, and gravity), resulting in a final output decision equation with good anti-interference capability and fluid dynamics rationality.
[0017] 3. Before formal testing, this invention designs a condition mapping step based on the target research object (such as a rocket engine turbopump, a long-distance water pipeline, etc.). By extracting the static intrinsic parameters (such as specific pipe diameter, fluid density) and dynamic operating parameters (such as velocity, average acceleration) of the actual pipeline, and using these as a benchmark for gradient adjustment and variable cross-combination, the generated test data matrix can cover the physical fields that the actual engineering equipment may encounter during transient processes. This design enables the final judgment rules to be directly correlated with and applied to the operating conditions of real equipment. Attached Figure Description
[0018] Figure 1 This is a flowchart of the method for determining the boundary and risk level of liquid column separation in Embodiment 1 of the present invention.
[0019] Figure 2 This is a sensitivity analysis diagram of the dimensionless array in Embodiment 2 of the present invention.
[0020] Figure 3 This is a diagram showing the separation effect of the liquid column separation state in Embodiment 2 of the present invention. Detailed Implementation
[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0022] Example 1 Please see Figure 1 This embodiment provides a method for determining the boundary and risk level of liquid column separation, including steps S1 to S5.
[0023] Before step S1, the following steps are also included: Identify the target research object, which includes liquid delivery pipelines; Obtain pipeline data of the target research object under actual operation or simulated working conditions, and statistically derive typical values of the fluid at a specific pipeline location. The typical values include at least the specific pipe diameter of the actual equipment, and the average velocity and acceleration of the fluid when it flows through the location. Based on the typical values, working conditions are mapped, and multiple sets of test parameter combinations that can cover the actual working condition variation range of the target research object are configured on the visual liquid column separation test bench.
[0024] Transient flow conditions in real-world engineering environments (such as rocket engine turbopumps or long-distance water pipelines) are extremely complex. The steps described above aim to reduce the dimensionality of actual, large-scale, and complex engineering pipeline conditions and map them onto a laboratory test bench. By extracting typical values such as specific pipe diameters, characteristic velocities, and accelerations as benchmarks, it ensures that the data generated by the subsequent test bench can realistically encompass (cover) the extreme transient physical fields that engineering equipment may encounter, thereby greatly improving the effectiveness and reliability of the judgment model in practical engineering applications.
[0025] The test parameters include pipe diameter, initial velocity, acceleration, bubble characteristic height, fluid density, and pressure; The visualization liquid column separation test bench is equipped with multiple sets of test parameter combinations that can cover the actual operating condition variation range of the target research object, specifically including: The pipe diameter, fluid density, and bubble characteristic height in the experimental parameters are used as the static intrinsic parameters of the target research object. The initial velocity, acceleration, and pressure in the test parameters are used as dynamic operating parameters; When configuring a single-batch test matrix, the static intrinsic parameters are kept constant, and the typical values obtained from statistics are used as a benchmark. The dynamic operating parameters are adjusted up and down according to the preset distribution gradient, and several different combinations of test parameters are generated by cross-combining variables.
[0026] S1. Using a visual liquid column separation test bench, conduct multiple tests under different combinations of test parameters, collect test parameters, and simultaneously record the corresponding liquid column separation status; wherein, the liquid column separation status includes separation occurring and no separation occurring.
[0027] Traditional methods of judging cavitation by monitoring pressure drop using pressure sensors have limitations in terms of latency and location. This embodiment utilizes a visualization test bench (such as one combined with a high-speed camera) to intuitively and instantaneously capture the microscopic physical processes of cavitation formation, evolution, and convergence within the liquid column, as well as the cavitation process that cuts off the pipeline. This provides the machine learning model with absolutely accurate binary state labels of "separation occurred (label 1)" and "no separation occurred (label 0)".
[0028] S2. Based on Buckingham's Π theorem, select the repeated and residual variables in the experimental parameters, and combine the dimensional experimental parameters through dimensional analysis to construct multiple independent dimensionless numbers, forming a dimensionless parameter set.
[0029] In step S2, initial velocity, pipe diameter, and fluid density are selected as repeating variables, and acceleration, bubble characteristic height, and pressure are selected as residual variables to construct at least 6 independent dimensionless numbers as the dimensionless parameter set.
[0030] The physical appearance of fluid systems is subject to size effects (e.g., large-diameter and small-diameter pipes behave differently at the same flow velocity). Buckingham's π theorem is a core mathematical tool in fluid mechanics for eliminating size effects. By selecting initial velocity, pipe diameter, and density as repeating variables (covering the fundamental dimensions of kinematics, geometry, and dynamics), the physical parameters that originally had units (e.g., m / s, Pa) are transformed into dimensionless numbers reflecting the relative ratios of physical quantities such as inertial force, pressure, and gravity. This treatment reduces the degrees of freedom of the system, making the laws obtained from experiments universal across scales.
[0031] S3. Using the dimensionless parameter set as input features and the liquid column separation state as the target variable, a machine learning algorithm is introduced for multiple batches of independent training. The importance score of each dimensionless parameter is calculated to quantify the sensitivity.
[0032] In step S3, the machine learning algorithm is the random forest regression algorithm; the method for performing multiple batches of independent training is as follows: randomly select 80% of the total data as training samples according to a preset ratio, repeat the model training multiple times using different data combinations in different training processes, and extract the importance score corresponding to each dimensionless parameter in each training.
[0033] The Random Forest algorithm, by constructing a large number of decision trees and averaging them, can effectively handle nonlinear relationships and has a built-in feature importance output function. Extracting 80% of the data as the training set (leaving 20% for validation) and performing multiple independent repeated sampling training sessions is to eliminate random errors caused by single data partitioning. This utilizes statistical principles to objectively expose the true weight contribution of each dimensionless number to inducing liquid column separation, i.e., sensitivity.
[0034] S4. Select the dimensionless number with the highest importance score and the most stable ranking from the dimensionless parameter group as the dominant dimensionless number.
[0035] The dominant dimensionless number represents the main contradiction that causes the current system's cavitation (e.g., it may be a variant characterizing the ratio of local pressure drop to absolute pressure). It is the parameter that exhibits the highest correlation across all independent training batches and has the strongest physical determinant.
[0036] S5. Near the critical region of the dominant dimensionless number, regression analysis is used to fit a function to the boundary between the data points that have separated and those that have not, and the intersection of the boundary fitting function and the preset judgment threshold line is solved. The dimensionless critical equilibrium point is output as the quantitative benchmark value of the judgment boundary.
[0037] In step S5, the regression analysis method is logistic regression or linear regression; the preset judgment threshold line is a horizontal or vertical line determined according to the preset occurrence probability; the boundary fitting function is a curve or a straight line; the intersection of the preset judgment threshold line and the boundary fitting function is the dimensionless critical equilibrium point.
[0038] Because fluid experimental data often overlap or transition zones at the boundary between where separation occurs and where it doesn't, methods like logistic regression can be used to output a probability function indicating system separation. Setting the location of a preset probability as a threshold line yields a universal "dimensionless critical equilibrium point" as the corresponding mathematical solution. In engineering monitoring, if the system's real-time dominant dimensionless number exceeds this equilibrium point, it can be determined that the pipeline has breached safety limits.
[0039] S6. After confirming that liquid column separation has occurred, select multiple dimensionless parameters with importance scores higher than the set threshold and combine them to construct a composite judgment parameter for assessing the severity. Based on the relationship between the composite judgment parameter and the preset level threshold, the liquid column separation is divided into different risk levels.
[0040] In step S6, the method for constructing the composite determination parameters is as follows: Set the preset sensitivity threshold to 0.5. Based on the importance score calculated in step S3, remove dimensionless parameters whose importance score is less than or equal to the threshold, and select the remaining dimensionless parameters as candidate parameters. Multiple sets of correction ratios are constructed by combining the candidate parameters in pairs or in groups of three. The importance score of each group of correction ratios is calculated according to the training method in step S3. The correction ratio with the highest importance score and the most stable ranking is selected as the composite judgment parameter.
[0041] Regression analysis is used to fit the data point boundaries between severe and slight separation, and the corresponding critical intersection points are solved as preset level threshold intervals. The composite judgment parameters are compared with the preset level threshold intervals, and the risk level of liquid column separation is output according to the range in which the composite judgment parameters fall.
[0042] Whether liquid column separation occurs is usually determined by the dominant dimensionless number, but its destructive extent (i.e., the maximum volume of cavitation bubbles or the collapse impact force) is influenced by the coupling effects of multiple secondary physical mechanisms such as pipeline length and liquid elasticity. Therefore, by eliminating weakly correlated parameters (sensitivity ≤ 0.5) and cross-combining strongly correlated parameters to generate a correction ratio (i.e., a combined dimension in a physical sense), the nonlinear coupling effect between various parameters can be fully captured. Using multi-dimensional composite judgment parameters for secondary classification solves the technical problem of the difficulty in accurately assessing the degree of cavitation risk with a single parameter, providing a quantitative classification basis for the redundancy design of engineering cavitation defense.
[0043] Example 2 This embodiment uses the six test parameters mentioned in Example 1 as examples to elaborate on the method for determining the boundary and risk level of liquid column separation.
[0044] 1. Data acquisition and status recording: Based on a visualized liquid column separation test bench, relevant parameters of the liquid column are obtained: pipe diameter. Initial velocity acceleration Bubble characteristic height The density of water and pressure The experiment recorded the liquid column separation characteristics under more than 60 different parameter conditions, and simultaneously collected and recorded the experimental parameters and liquid column separation status, i.e. whether the liquid column separated and to what extent.
[0045] 2. Dimensionless process: Considering a comprehensive flow system, the six variables selected are: , , , , , .in, The dimensions are ; The dimensions are ; The dimensions are ; The dimensions are ; The dimensions are ; The dimensions are .
[0046] These variables encompass three fundamental dimensions: mass (M), length (L), and time (T). According to Buckingham's π theorem, the number of variables n=6, the number of fundamental dimensions k=3, therefore the number of independent dimensionless π terms should be nk=3. , and As a repeating variable , , As a remaining variable.
[0047] For variables :set up ; Dimensional equation: ; Solving ; therefore .
[0048] Similarly, six independent dimensionless numbers are constructed: ; in Units are , Units are , and The unit is mm. Units are , The unit is Pa.
[0049] 3. Intelligent analysis and determination of dominant dimensionless number: The collected experimental data were converted into a structured data frame, with each row representing a set of experimental observations, including the original parameters ( The six dimensionless numbers obtained by calculation ( The system uses the liquid column separation state label (0 or 1) as the target variable, selects a random forest regressor, and treats the liquid column separation state as a continuous regression target.
[0050] Based on the above data, a random forest regression method was used to analyze the relationship between the liquid column separation state and various dimensionless parameters. The number of decision trees in the model was set to 100 to ensure good stability of the calculation results. A fixed random seed was used to ensure consistency of results under different experimental conditions. Other model parameters were not subject to special restrictions, allowing each decision tree to grow sufficiently during training, thereby reflecting the influence of each dimensionless parameter on the liquid column separation state. Considering that the results of a single modeling iteration may be affected by differences in sample selection, this implementation scheme introduced multiple random sampling methods for repeated training during the modeling process. Specifically, a portion of the data was randomly selected from all experimental data according to a preset proportion as training samples. The number of training samples accounted for 80% of the total data volume, and different data combinations were used in different training processes, repeating the model training four times. After each model training iteration, the importance coefficients corresponding to each dimensionless parameter in the model were extracted to characterize the degree of influence of that parameter on the liquid column separation state under the current training conditions—sensitivity.
[0051] After the above training, the importance results of each dimensionless parameter under different training conditions can be obtained. The importance coefficients of the same dimensionless parameter obtained in each training iteration are statistically processed to obtain and plot a bar chart. The bar length represents the average importance score; the longer the bar, the more important the feature. To illustrate the fluctuations in each training result, an error bar is superimposed on each bar, with data derived from the changes in the feature's score over four iterations. This visually demonstrates the stability of its importance score. Figure 2 As shown in the line graph.
[0052] Analyzing the generated bar chart, the criteria for determining the dominant dimensionless number are twofold: (1) High importance: Its average importance score ranks first among all features.
[0053] (2) Strong stability: Its importance score ranked first in each training session.
[0054] The characteristic sensitivity analysis of the six dimensionless numbers is shown in the table below: Table 1: Characteristic Sensitivity of Six Dimensionless Numbers
[0055] In this implementation scheme, dimensionless numbers In four independent training sessions, its importance score not only had the highest average, but it also ranked first after each individual training session, demonstrating undisputed discriminative ability. Therefore, it will... It was determined to be the dominant dimensionless number used to establish the final judgment criterion.
[0056] 4. Calculation of critical equilibrium point: Drawing A scatter plot with "separation occurred" (1) on the horizontal axis and "no separation occurred" (0) on the vertical axis. It can be observed that when... When the value exceeds a certain critical value R, the data points are mainly concentrated in the "separation" region. Therefore, a judgment criterion is established: if... If this occurs, liquid column separation is highly likely to occur in the system. Therefore, the first equilibrium point is determined. To determine more precisely Near the critical region, the state of liquid column separation (0 or 1) is taken as the dependent variable. Using this as the independent variable, perform logistic or linear regression to fit a curve (or straight line). The intersection of this curve and the preset threshold line is the theoretical critical equilibrium point, which the program automatically calculates and outputs. The value equals 1, which serves as the final dimensionless criterion, i.e., the first criterion.
[0057] 5. Determination of the degree of separation: The process begins after the initial determination (confirming that liquid column separation will occur). The core aspect lies in constructing a composite determination parameter. By integrating information from six dimensionless parameters into a comprehensive index Using existing experimental data, the optimal combination of coefficients is calculated through inversion. This ensures the final parameters... To determine the composite dimensional values of the degree of separation, a refined and graded assessment of the severity of liquid column separation is achieved. For six dimensionless numbers... The sensitivities obtained earlier, sorted from largest to smallest, are as follows: This sorting result serves as the basis for subsequent priority determination of dimensionless parameter combinations, because Parameters with a sensitivity less than 0.5 have little value in combination judgment and are discarded. A performance judgment baseline model is established, using four dimensionless numbers with a sensitivity greater than 0.5 as input features. The same random forest algorithm is used to train the model, and its performance metrics are recorded. Based on the sensitivity ranking, high-priority parameter combinations to be tested are generated. The generation of combinations follows the principles of "high-sensitivity parameters first" and "combination efficiency."
[0058] This embodiment employs a combination method of dominant parameters and ratios, combining dimensionless numbers in pairs and groups of three to form correction ratios with clear physical meaning to correct for separation, thereby constructing a comprehensive judgment equation: The pairwise combination can be represented as: ; ; The combination of three groups can be represented as: ; ; Following the method for determining the dominant dimensionless number in Section 3 of this embodiment, the same random forest algorithm was used to train the model, and its performance indicators were recorded to obtain the sensitivity values of the combined dimensionless functions (as shown in Table 2). It was determined to be the dominant dimensionless number used to establish the final judgment criterion.
[0059] Table 2: Sensitivity values of combined dimensionless functions
[0060] Based on the degree of risk posed to the system when liquid column separation occurs, it is divided into two levels: severe separation and slight separation. Based on the numerical analysis results and experimental database generated in Section 1, the observed liquid column separation state level is labeled for each sample, and threshold calibration is performed. In the program, when the system receives a "liquid column separation has occurred" signal, it will automatically read or calculate the parameters required to construct the second judgment parameter under the current operating condition. , calculate The value is compared with a preset threshold range, and the separation status is output. The value is obtained after the program runs multiple times. The accuracy rate is between 95% and 95% on average.
[0061] Example 3 This embodiment provides a system for determining the occurrence limit and risk level of liquid column separation, used to implement the method for determining the occurrence limit and risk level of liquid column separation in Embodiment 1. The system includes: The data acquisition module is used to collect test parameters and simultaneously record the corresponding liquid column separation status when conducting tests using the visual liquid column separation test bench. The data processing module, based on Buckingham's π theorem, selects repeating and residual variables from the experimental parameters. Through dimensional analysis, it combines the dimensional experimental parameters to construct multiple independent dimensionless numbers, forming a dimensionless parameter set. Using this dimensionless parameter set as input features and the liquid column separation state as the target variable, a machine learning algorithm is introduced for multiple batches of independent training to calculate the importance score of each dimensionless parameter to quantify sensitivity. The dimensionless number with the highest importance score and most stable ranking is selected from the dimensionless parameter set as the dominant dimensionless number. Near the critical region of the dominant dimensionless number, regression analysis is used to fit a function to the boundary between data points where separation has occurred and those where separation has not occurred. The intersection of this boundary fitting function and a preset judgment threshold line is solved, and the dimensionless critical equilibrium point is output as a quantitative benchmark value for determining the occurrence boundary. After confirming liquid column separation, multiple dimensionless parameters with importance scores higher than the set threshold are selected and combined to construct a composite judgment parameter for assessing severity. Based on the relationship between the composite judgment parameter and the preset level threshold, liquid column separation is classified into different risk levels.
[0062] Example 4 This embodiment provides a computer-readable storage medium storing a computer program thereon. When the program is executed by a processor, it implements the steps of the method for determining the liquid column separation boundary and risk level as described in Embodiment 1.
[0063] The computer-readable storage medium may include flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the storage medium may be an internal storage unit of a computer device, such as the hard disk or memory of the computer device. In other embodiments, the storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, smart memory card, secure digital card, flash memory card, etc., provided on the computer device. Of course, the storage medium may include both internal storage units and external storage devices of the computer device. In this embodiment, the memory is typically used to store the operating system and various application software installed on the computer device. In addition, the memory can also be used to temporarily store various types of data that have been output or will be output.
[0064] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.
Claims
1. A method for determining the boundary and risk level of liquid column separation, characterized in that, include: S1. Using a visual liquid column separation test bench, conduct multiple tests under different combinations of test parameters, collect test parameters, and simultaneously record the corresponding liquid column separation status; wherein, the liquid column separation status includes separation occurring and no separation occurring; S2. Based on Buckingham's π theorem, select the repeated and residual variables in the experimental parameters, and combine the dimensional experimental parameters through dimensional analysis to construct multiple independent dimensionless numbers, forming a set of dimensionless parameters; S3. Using the dimensionless parameter set as input features and the liquid column separation state as the target variable, a machine learning algorithm is introduced for multiple batches of independent training, and the importance score of each dimensionless parameter is calculated to quantify the sensitivity. S4. Select the dimensionless number with the highest importance score and the most stable ranking from the dimensionless parameter group as the dominant dimensionless number; S5. Near the critical region of the dominant dimensionless number, the regression analysis method is used to fit the function of the boundary between the data points that have separated and those that have not separated, and the intersection of the boundary fitting function and the preset judgment threshold line is solved. The dimensionless critical equilibrium point is output as the quantitative benchmark value of the judgment boundary. S6. After confirming that liquid column separation has occurred, select multiple dimensionless parameters with importance scores higher than the set threshold and combine them to construct a composite judgment parameter for assessing the severity. Based on the relationship between the composite judgment parameter and the preset level threshold, the liquid column separation is divided into different risk levels.
2. The method for determining the boundary and risk level of liquid column separation according to claim 1, characterized in that, In step S3, the machine learning algorithm is the random forest regression algorithm; the method for performing multiple batches of independent training is as follows: randomly select 80% of the total data as training samples according to a preset ratio, repeat the model training multiple times using different data combinations in different training processes, and extract the importance score corresponding to each dimensionless parameter in each training.
3. The method for determining the boundary and risk level of liquid column separation according to claim 1, characterized in that, In step S5, the regression analysis method is logistic regression or linear regression; the preset judgment threshold line is a horizontal or vertical line determined according to the preset occurrence probability; the boundary fitting function is a curve or a straight line; the intersection of the preset judgment threshold line and the boundary fitting function is the dimensionless critical equilibrium point.
4. The method for determining the boundary and risk level of liquid column separation according to claim 1, characterized in that, In step S6, the method for constructing the composite determination parameters is as follows: Set the preset sensitivity threshold to 0.
5. Based on the importance score calculated in step S3, remove dimensionless parameters whose importance score is less than or equal to the threshold, and select the remaining dimensionless parameters as candidate parameters. Multiple sets of correction ratios are constructed by combining the candidate parameters in pairs or in groups of three. The importance score of each group of correction ratios is calculated according to the training method in step S3. The correction ratio with the highest importance score and the most stable ranking is selected as the composite judgment parameter.
5. The method for determining the boundary and risk level of liquid column separation according to claim 4, characterized in that, In step S6, regression analysis is used to fit the data point boundaries between severe and slight separation, and the corresponding critical intersection points are solved as preset level threshold intervals. The composite judgment parameters are compared with the preset level threshold intervals, and the risk level of liquid column separation is output according to the range in which the composite judgment parameters fall.
6. The method for determining the boundary and risk level of liquid column separation according to claim 1, characterized in that, Before step S1, the following steps are also included: Identify the target research object, which includes liquid delivery pipelines; Obtain pipeline data of the target research object under actual operation or simulated working conditions, and statistically derive typical values of the fluid at a specific pipeline location. The typical values include at least the specific pipe diameter of the actual equipment, and the average velocity and acceleration of the fluid when it flows through the location. Based on the typical values, working conditions are mapped, and multiple sets of test parameter combinations that can cover the actual working condition variation range of the target research object are configured on the visual liquid column separation test bench.
7. The method for determining the boundary and risk level of liquid column separation according to claim 6, characterized in that, In step S1, the test parameters include pipe diameter, initial velocity, acceleration, bubble characteristic height, fluid density, and pressure; in step S2, the initial velocity, pipe diameter, and fluid density are selected as repeating variables, and the acceleration, bubble characteristic height, and pressure are selected as residual variables to construct at least 6 independent dimensionless numbers as the dimensionless parameter set.
8. The method for determining the boundary and risk level of liquid column separation according to claim 7, characterized in that, The specific configuration of multiple sets of test parameter combinations on the visualized liquid column separation test bench, capable of covering the actual operating condition variation range of the target research object, includes: The pipe diameter, fluid density, and bubble characteristic height in the experimental parameters are used as the static intrinsic parameters of the target research object. The initial velocity, acceleration, and pressure in the test parameters are used as dynamic operating parameters; When configuring a single-batch test matrix, the static intrinsic parameters are kept constant, and the typical values obtained from statistics are used as a benchmark. The dynamic operating parameters are adjusted up and down according to the preset distribution gradient, and several different combinations of test parameters are generated by cross-combining variables.
9. A system for determining the boundary and risk level of liquid column separation, characterized in that, For performing the method for determining the liquid column separation occurrence limit and risk level as described in any one of claims 1 to 8, the system comprises: The data acquisition module is used to collect test parameters and simultaneously record the corresponding liquid column separation status when conducting tests using the visual liquid column separation test bench. The data processing module, based on Buckingham's π theorem, selects repeating and residual variables from the experimental parameters. Through dimensional analysis, it combines the dimensional experimental parameters to construct multiple independent dimensionless numbers, forming a dimensionless parameter set. Using this dimensionless parameter set as input features and the liquid column separation state as the target variable, a machine learning algorithm is introduced for multiple batches of independent training to calculate the importance score of each dimensionless parameter to quantify sensitivity. The dimensionless number with the highest importance score and most stable ranking is selected from the dimensionless parameter set as the dominant dimensionless number. Near the critical region of the dominant dimensionless number, regression analysis is used to fit a function to the boundary between data points where separation has occurred and those where separation has not occurred. The intersection of this boundary fitting function and a preset judgment threshold line is solved, and the dimensionless critical equilibrium point is output as a quantitative benchmark value for determining the occurrence boundary. After confirming liquid column separation, multiple dimensionless parameters with importance scores higher than the set threshold are selected and combined to construct a composite judgment parameter for assessing severity. Based on the relationship between the composite judgment parameter and the preset level threshold, liquid column separation is classified into different risk levels.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the steps of the method for determining the liquid column separation boundary and risk level as described in any one of claims 1 to 8.