Method for constructing uht cleaning model of liquid milk based on dimensional analysis and multi-parameter monitoring

Abstract

The application discloses a kind of based on dimension analysis and multi-parameter monitoring liquid milk UHT cleaning model construction method, it is related to dairy processing technical field.The method first identifies 10 key cleaning influence parameters by dimension analysis, derives 6 dimensionless numbers with clear physical meaning, and constructs dimensionless equation;Then through multi-parameter monitoring, collect pH value, conductivity, turbidity and protein concentration data in cleaning process, adopt multi-index fluctuation stability determination strategy to determine cleaning end point;Finally, based on symbolic regression fitting, the alkali washing and acid washing time prediction equation is obtained.The application realizes the overall quantitative characterization of liquid milk UHT pipeline cleaning process, improves cleaning efficiency and cleaning effect, and provides technical reference for dairy industry in-situ cleaning process optimization.

Description

Technical Field

[0001] This invention relates to the field of dairy processing technology, and in particular to a method for constructing a liquid milk UHT cleaning model based on dimensional analysis and multi-parameter monitoring. Background Technology

[0002] Ultra-high temperature (UHT) sterilization is a core process in liquid milk processing, significantly extending product shelf life. During UHT production, milk residue composed of proteins, fats, and minerals forms on the inner walls of pipes. Incomplete cleaning can lead to microbial growth, affecting product quality and even clogging pipes. Clean in situ (CIP) is currently the most common pipe cleaning method in the dairy industry. However, traditional CIP processes rely heavily on experience to set cleaning parameters, often resulting in over-cleaning leading to energy and water waste and increased cleaning agent consumption, or incomplete cleaning causing food safety risks.

[0003] Cleaning models can theoretically guide CIP cleaning, monitor the cleaning process, and optimize cleaning techniques. However, existing cleaning models are mostly built based on single or a few influencing parameters, failing to comprehensively consider the coupled effects of multiple factors such as dirt characteristics, hydrodynamic properties, cleaning agent chemical properties, and temperature. This results in low model prediction accuracy and limited applicability. While some studies have attempted to introduce multi-parameter analysis, the parameters have not been effectively simplified, leading to high modeling complexity and a lack of clear physical meaning, making it difficult to promote and apply in industrial production.

[0004] Furthermore, existing methods for determining the cleaning endpoint often rely on a single indicator (such as conductivity or pH value) or human experience, which cannot accurately reflect the actual cleaning status of the dirt inside the pipeline and are prone to misjudgment. Single indicators are easily affected by interference and have poor stability, leading to cleaning times that are too long or too short, increasing production costs and failing to guarantee cleaning effectiveness. Therefore, there is an urgent need to develop a model construction method that can accurately characterize the cleaning process of liquid milk UHT pipelines and accurately determine the cleaning endpoint. Summary of the Invention

[0005] To address the aforementioned problems, this invention provides a method for constructing a liquid milk UHT cleaning model based on dimensional analysis and multi-parameter monitoring, comprising the following steps: Step S1: Simplification of Cleaning Influence Parameters and Construction of Dimensionless Equations Based on Dimensional Analysis: Identifying 10 key influencing parameters of the alkaline and acid washing processes in liquid milk UHT pipelines, with cleaning time as the primary factor. Establish a function expression for the target variable; based on the π theorem, select four repeated variables and combine them with the remaining non-repeating variables to derive six independent dimensionless numbers and clarify their physical meanings, then integrate them to obtain the comprehensive dimensionless number of dirt. ; Step S2: Cleaning evaluation and cleaning time equation modeling based on multi-parameter monitoring: A standardized model of milk residue was prepared using a small-scale UHT device to simulate the industrial CIP cleaning process. Real-time data on pH, conductivity, and turbidity of the cleaning solution were collected, and protein concentration was measured at intervals. A cleaning endpoint determination strategy based on the stability of multi-index fluctuations was constructed, and the relative fluctuation of each index per unit time was calculated. When all indicators The cleaning process was considered complete when all values ​​were below 0.01. The symbolic regression method was used to fit the cleaning time prediction equations for the alkaline and acidic washing processes based on the experimental data, and the model fitting effect was verified by the evaluation index.

[0006] Preferably, the 10 key influencing parameters mentioned in step S1 are the cleaning fluid flow rate. Cleaning fluid temperature Cleaning agent concentration Cleaning agent density Dynamic viscosity of cleaning agent Pipe inner diameter Thick layer of dirt Surface free energy of dirt dirt viscosity Dirt density .

[0007] Preferably, in step S1, based on the Π theorem, with n=10 variables and k=4 basic physical dimensions, six independent dimensionless numbers are derived; the selected repeating variable is the cleaning fluid flow rate. Pipe inner diameter Cleaning fluid temperature Cleaning agent density .

[0008] Preferably, the six independent dimensionless numbers derived in step S1 are as follows: ； ; ; ; ; ; This is the cleaning agent concentration-density ratio, reflecting the chemical concentration characteristics of the cleaning agent; It is the reciprocal of the Reynolds number Re. The Reynolds number characterizes the degree of turbulence of a fluid. The larger the Re, the stronger the turbulence and the shorter the cleaning time. It is the ratio of the thickness of the dirt layer to the inner diameter of the pipe, reflecting the physical morphological characteristics of the dirt; It is the ratio of the surface energy of the dirt to the kinetic energy of the fluid, reflecting the strength of the interfacial interaction between the dirt and the cleaning agent; This is the ratio of dirt density to cleaning agent density, reflecting the impact of the density difference between the two phases on the cleaning process; It is the ratio of the dirt's viscosity to the fluid shear force, reflecting the resistance effect of dirt viscosity on the cleaning process.

[0009] Preferably, the dimensionless number of the overall dirt in step S1 The calculation formula is: , where the Reynolds number Re is the reciprocal of Π2.

[0010] Preferably, the simulated industrial CIP cleaning process in step S2 is as follows: water washing, alkali washing, water washing, acid washing, and water washing.

[0011] Preferably, the formula for calculating the relative fluctuation Δ of each indicator per unit time in step S2 is as follows: ; Let be the detection value of a certain indicator at time t. For a certain indicator at time The detected value, For a certain indicator at time The detected value, For a certain indicator at time The detected value.

[0012] Preferably, in step S2, Eureqa software is used for symbolic regression analysis; the evaluation index for model fit includes the coefficient of determination R. 2 Correlation coefficient r, mean absolute error (MAE), mean square error (MSE), root mean square error (RMSE).

[0013] Preferably, the equation for predicting the cleaning time in the alkaline washing process in step S2 is obtained by fitting experimental data from 17 sets of different cleaning agent concentrations, cleaning flow rates, and cleaning temperatures, and its coefficient of determination R0 is [value missing]. 2 =0.9731, RMSE=0.9672.

[0014] Preferably, the equation for predicting the cleaning time in step S2 is obtained by fitting experimental data from 17 sets of different cleaning agent concentrations, cleaning flow rates, and cleaning temperatures, and its coefficient of determination R0 is [value missing]. 2 =0.9980, RMSE=0.5756.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention simplifies 10 cleaning-influencing parameters into 6 dimensionless numbers with clear physical meaning through dimensional analysis, effectively reducing modeling complexity. At the same time, it fully preserves key influencing factors such as dirt characteristics, hydrodynamic characteristics, and chemical characteristics of cleaning agents, giving the model clear physical meaning and strong interpretability.

[0016] 2. This invention constructs a cleaning endpoint determination strategy based on the stability of multiple index fluctuations. It comprehensively examines the real-time fluctuation characteristics of four detection indicators: conductivity, pH value, turbidity, and protein content. This significantly improves the accuracy of cleaning endpoint determination, avoids the problems of over-cleaning and incomplete cleaning, and reduces energy and cleaning agent consumption.

[0017] 3. This invention uses Eureqa software for symbolic regression analysis to automatically extract the optimal mathematical expression from experimental data. It comprehensively balances model fitting accuracy and structural complexity, resulting in predictive equations for alkali washing and acid washing times that demonstrate good fitting performance and high prediction accuracy. This provides reliable technical support for the intelligent optimization of CIP processes in the dairy industry. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the overall process of the method of the present invention. Detailed Implementation

[0019] The following is in conjunction with the appendix Figure 1 The present invention will be further described in detail with reference to the embodiments. The method for constructing a liquid milk UHT cleaning model based on dimensional analysis and multi-parameter monitoring provided by the present invention specifically includes the following core steps: first, simplification of cleaning influence parameters and construction of dimensionless equations based on dimensional analysis; second, cleaning evaluation and modeling of cleaning time equations based on multi-parameter monitoring. The present invention achieves accurate characterization of the liquid milk UHT pipeline cleaning process through multi-parameter coupled modeling, providing a technical reference for optimizing in-situ cleaning (CIP) processes in the dairy industry.

[0020] 1. Simplification of cleaning influence parameters and construction of dimensionless equations based on dimensional analysis: 1.1 Identification and Dimensional Analysis of Cleaning Influencing Parameters: The main influencing factors of the cleaning process can be divided into six categories: the type and quantity of dirt, the technical characteristics of the substrate, the treatment duration, the chemical properties of the cleaning agent, the mechanical parameters of the treatment equipment, and the treatment temperature. This invention focuses on alkaline and acid washing processes, selecting the following 10 key influencing parameters: cleaning fluid flow rate. Cleaning fluid temperature Cleaning agent concentration Cleaning agent density Dynamic viscosity of cleaning agent Pipe inner diameter Thick layer of dirt Surface free energy of dirt dirt viscosity Dirt density .

[0021] Cleaning time Let these be the target variable, and the above 10 parameters be the independent variables. Establish a general functional expression for the relationship between cleaning time and each influencing factor: ; In the formula: This refers to the cleaning time.

[0022] The dimensional analysis results of the above 10 variables are shown in Table 1.

[0023] Table 1. Dimensional Analysis of Cleaning Influence Parameters

[0024] 1.2 Derivation and Physical Significance of Dimensionless Number Groups: According to the π theorem, the number of dimensionless numbers equals the number of variables minus the number of basic dimensions, calculated as follows: ; In the formula: The quantity of dimensionless numbers; The number of related variables; The number of basic physical dimensions.

[0025] In this invention Therefore, six independent dimensionless numbers can be derived ( Select , , , As repeated variables, the remaining influencing parameters are combined with the repeated variables to construct the general formula for the dimensionless number group: ; In the formula: For the first One non-repeating influencing parameter; , , , There are four dimensional exponents to be determined.

[0026] Substituting the dimensions of each variable into the general formula, and solving for the exponent according to the principle of "consistent dimensions on both sides of a dimensionless formula", we finally obtain the expressions for six dimensionless numbers: ; ; ; ; ; ; This is the cleaning agent concentration-density ratio, reflecting the chemical concentration characteristics of the cleaning agent; It is the reciprocal of the Reynolds number Re. The Reynolds number characterizes the degree of turbulence of a fluid. The larger the Re, the stronger the turbulence and the shorter the cleaning time. It is the ratio of the thickness of the dirt layer to the inner diameter of the pipe, reflecting the physical morphological characteristics of the dirt; It is the ratio of the surface energy of the dirt to the kinetic energy of the fluid, reflecting the strength of the interfacial interaction between the dirt and the cleaning agent; This is the ratio of dirt density to cleaning agent density, reflecting the impact of the density difference between the two phases on the cleaning process; It is the ratio of the dirt's viscosity to the fluid shear force, reflecting the resistance effect of dirt viscosity on the cleaning process.

[0027] Temperature, as the only parameter with a dimension of temperature, has not had its influence ignored. Instead, it is indirectly incorporated into the dimensionless analysis system through temperature-related physical properties (such as density and viscosity).

[0028] 2. Cleaning evaluation and cleaning time equation modeling based on multi-parameter monitoring: 2.1 Experimental Design and Data Acquisition: 2.1.1 Experimental Procedure: A small-scale UHT (Ultra-High Temperature) device was used to simulate the liquid milk sterilization process, and standardized model milk residue was prepared under fixed sterilization conditions. Subsequently, the industrial pipeline CIP (Clean-In-Place) cleaning process was simulated, with the specific steps being: water washing, alkaline washing, water washing, acid washing, and water washing. The process parameter ranges for each cleaning step are shown in Table 2.

[0029] Table 2 Process parameters for CIP cleaning process

[0030] V1 is the total mass of clean water used for rinsing, V2 is the total mass of alkaline cleaning solution (usually sodium hydroxide solution) (consistent with the pickling volume); T1 is the control time for a single water rinse, T2 is the total duration of alkaline rinsing, and T3 is the total duration of pickling.

[0031] 2.1.2 Data Acquisition: The pH value, conductivity and turbidity data of the cleaning solution during the cleaning process are collected in real time and the change curves are drawn through the water quality analysis device supporting the small UHT equipment; the protein concentration data are measured by the interval sampling method: after the start of the alkali cleaning and acid cleaning stages, a certain volume of cleaning solution is collected from the return port of the cleaning tank at an interval of 30s, the protein concentration in the solution is measured, and the protein concentration change curve is drawn for subsequent cleaning time determination and model fitting.

[0032] 2.2. Cleaning end point determination method: Draw the change curves of pH value, conductivity, turbidity and protein concentration during the cleaning process. When the data points of each indicator reach a stable state, the corresponding time is recorded as the cleaning time of this stage.

[0033] To improve the accuracy of cleaning end point determination, the present invention constructs a determination strategy based on multi-indicator fluctuation stability, which comprehensively investigates the real-time fluctuation characteristics of four detection indicators: conductivity, pH value, turbidity and protein content. The relative fluctuation amount of each indicator per unit time is calculated by Equation (9) , when all indicators' values are all lower than 0.01, the corresponding moment is determined as the cleaning end time: ; In the equation: is the detection value of a certain indicator at time t, is the detection value of a certain indicator at moment , is the detection value of a certain indicator at moment , is the detection value of a certain indicator at moment .

[0034] The cleaning end point determination standard for the acid cleaning process is completely consistent with that of the alkali cleaning process.

[0035] 2.3. Symbolic regression-based cleaning time equation fitting: The present invention uses Eureqa software for symbolic regression analysis, based on the variable relationship to construct a cleaning time prediction model. Based on automated modeling technology, Eureqa software automatically mines the optimal mathematical expression from experimental data through intelligent algorithms, can traverse thousands of candidate equation forms, comprehensively balances the fitting accuracy and model structure complexity, and gradually optimizes to obtain the optimal mathematical model that can accurately reflect the data law and has simplicity, which effectively avoids the subjective deviation of traditional manual modeling. The model fitting effect is evaluated by the coefficient of determination R 2 , correlation coefficient r, mean absolute error (MAE), mean square error (MSE), and root mean square error (RMSE).

[0036] 2.3.1 Cleaning time equation for alkaline washing process: The experimental data for the alkaline washing process are shown in Table 3.

[0037] Table 3. Statistical Table of Cleaning Time in Alkaline Washing Process

[0038] Based on the above experimental data, the equation for predicting alkaline washing time was obtained through symbolic regression fitting:

[0039] Among them, the dimensionless number of dirt Depend on , , , The Reynolds number Re is obtained by integration. The reciprocal: ; The fitting effect of the alkaline washing time regression equation is as follows: coefficient of determination Correlation coefficient Mean square error Mean absolute error Root mean square error .

[0040] 2.3.2, Cleaning time equation for the pickling process: The experimental data for the pickling process are shown in Table 4.

[0041] Table 4. Experimental statistics of cleaning time during pickling process

[0042] Based on the above experimental data, the pickling time prediction equation was obtained through symbolic regression fitting:

[0043] The fitting effect of the pickling time regression equation is as follows: coefficient of determination Correlation coefficient Mean square error Mean absolute error Root mean square error .

Claims

1. A method for constructing a liquid milk UHT cleaning model based on dimensional analysis and multi-parameter monitoring, characterized in that, Includes the following steps: Step S1: Simplification of Cleaning Influence Parameters and Construction of Dimensionless Equations Based on Dimensional Analysis: Identifying 10 key influencing parameters of the alkaline and acid washing processes in liquid milk UHT pipelines, with cleaning time as the primary factor. Establish a function expression for the target variable; based on the π theorem, select four repeated variables and combine them with the remaining non-repeating variables to derive six independent dimensionless numbers and clarify their physical meanings, then integrate them to obtain the comprehensive dimensionless number of dirt. ; Step S2: Cleaning evaluation and cleaning time equation modeling based on multi-parameter monitoring: A standardized model of milk residue was prepared using a small-scale UHT device to simulate the industrial CIP cleaning process. Real-time data on pH, conductivity, and turbidity of the cleaning solution were collected, and protein concentration was measured at intervals. A cleaning endpoint determination strategy based on the stability of multi-index fluctuations was constructed, and the relative fluctuation of each index per unit time was calculated. When all indicators The cleaning process was considered complete when all values ​​were below 0.

01. The symbolic regression method was used to fit the cleaning time prediction equations for the alkaline and acidic washing processes based on the experimental data, and the model fitting effect was verified by the evaluation index.

2. The method for constructing a liquid milk UHT cleaning model based on dimensional analysis and multi-parameter monitoring according to claim 1, characterized in that, The 10 key influencing parameters mentioned in step S1 are the cleaning fluid flow rate. Cleaning fluid temperature Cleaning agent concentration Cleaning agent density Dynamic viscosity of cleaning agent Pipe inner diameter Thick layer of dirt Surface free energy of dirt dirt viscosity Dirt density .

3. The method for constructing a liquid milk UHT cleaning model based on dimensional analysis and multi-parameter monitoring according to claim 1, characterized in that, In step S1, based on the Π theorem, with n=10 variables and k=4 basic physical dimensions, six independent dimensionless numbers are derived; the selected repeating variable is the cleaning fluid flow rate. Pipe inner diameter Cleaning fluid temperature Cleaning agent density .

4. The method for constructing a liquid milk UHT cleaning model based on dimensional analysis and multi-parameter monitoring according to claim 3, characterized in that, The six independent dimensionless numbers derived in step S1 are as follows: ； ； ； ； ； ； This is the cleaning agent concentration-density ratio, reflecting the chemical concentration characteristics of the cleaning agent; It is the reciprocal of the Reynolds number Re. The Reynolds number characterizes the degree of turbulence of a fluid. The larger the Re, the stronger the turbulence and the shorter the cleaning time. It is the ratio of the thickness of the dirt layer to the inner diameter of the pipe, reflecting the physical morphological characteristics of the dirt; It is the ratio of the surface energy of the dirt to the kinetic energy of the fluid, reflecting the strength of the interfacial interaction between the dirt and the cleaning agent; This is the ratio of dirt density to cleaning agent density, reflecting the impact of the density difference between the two phases on the cleaning process; It is the ratio of the dirt's viscosity to the fluid shear force, reflecting the resistance effect of dirt viscosity on the cleaning process.

5. The method for constructing a liquid milk UHT cleaning model based on dimensional analysis and multi-parameter monitoring according to claim 4, characterized in that, The dimensionless number of the dirt in step S1 The calculation formula is: , where the Reynolds number Re is the reciprocal of Π2.

6. The method for constructing a liquid milk UHT cleaning model based on dimensional analysis and multi-parameter monitoring according to claim 1, characterized in that, The simulated industrial CIP cleaning process in step S2 is as follows: water washing, alkaline washing, water washing, acid washing, water washing.

7. The method for constructing a liquid milk UHT cleaning model based on dimensional analysis and multi-parameter monitoring according to claim 1, characterized in that, The formula for calculating the relative fluctuation Δ of each indicator per unit time in step S2 is as follows: ； Let be the detection value of a certain indicator at time t. For a certain indicator at time The detected value, For a certain indicator at time The detected value, For a certain indicator at time The detected value.

8. The method for constructing a liquid milk UHT cleaning model based on dimensional analysis and multi-parameter monitoring according to claim 1, characterized in that, In step S2, Eureqa software is used to perform symbolic regression analysis; the evaluation index of the model fit includes the coefficient of determination R. 2 Correlation coefficient r, mean absolute error (MAE), mean square error (MSE), root mean square error (RMSE).

9. The method for constructing a liquid milk UHT cleaning model based on dimensional analysis and multi-parameter monitoring according to claim 1, characterized in that, The equation for predicting the cleaning time in step S2 is obtained by fitting experimental data from 17 sets of different cleaning agent concentrations, cleaning flow rates, and cleaning temperatures, and its coefficient of determination R0 is [value missing]. 2 =0.9731, RMSE=0.9672.

10. The method for constructing a liquid milk UHT cleaning model based on dimensional analysis and multi-parameter monitoring according to claim 1, characterized in that, The equation for predicting the cleaning time in step S2 is obtained by fitting experimental data from 17 sets of different cleaning agent concentrations, cleaning flow rates, and cleaning temperatures, and its coefficient of determination R0 is [value missing]. 2 =0.9980, RMSE=0.5756.

Images