Wall adhesion quantitative evaluation method and system for spray drying tower

Abstract

The invention provides a wall adhesion quantitative evaluation method and system for a spray drying tower, and relates to the technical field of drying towers. Multi-dimensional associated data and a plurality of operation parameter sequences are acquired; respectively carrying out powder wall adhesion analysis by utilizing the powder wall adhesion simulation model and the powder wall adhesion prediction model, and outputting powder wall adhesion simulation distribution and powder wall adhesion prediction distribution; performing pulverization fluctuation analysis according to the operation parameter sequence, outputting a pulverization fluctuation coefficient, and configuring a model output weight; and according to the model output weight, fitting the powder wall adhesion simulation distribution and the powder wall adhesion prediction distribution, and outputting a powder wall adhesion evaluation result. According to the method and the device, the technical problems of low wall adhesion evaluation precision and poor reliability caused by dependence on a single physical model or a data model and lack of working condition fluctuation adaptability in the prior art are solved, and the technical effects of improving the wall adhesion evaluation precision and reliability and enhancing the dynamic adaptability to complex working conditions are achieved.

Description

Technical Field

[0001] The present application relates to the technical field of drying towers, and in particular to a method and system for quantitatively evaluating wall adhesion of spray drying towers. Background Art

[0002] Spray drying technology is widely used in the food, pharmaceutical, and chemical industries to rapidly dry liquid materials into powder. During the operation of a spray drying tower, powder particles tend to adhere to the inner walls of the tower, causing wall sticking. This phenomenon not only affects product yield and leads to unstable product quality, but also increases cleaning and maintenance costs. Therefore, accurately assessing and predicting wall sticking is a key requirement for optimizing the spray drying process.

[0003] Currently, spray drying tower wall sticking assessment primarily relies on empirical judgment, single-point temperature monitoring, or physical model calculations. For example, physical models use CFD simulations to study airflow distribution and powder motion within the tower to predict areas at risk of wall sticking. Empirical models use manual inspections and sensors (such as temperature and humidity probes) to obtain local parameters and combine historical experience to determine wall sticking trends. However, traditional physical models rely on idealized assumptions and fail to fully account for the coupled effects of multiple factors influencing wall sticking, such as airflow velocity, temperature gradients, and humidity variations. This results in significant deviations between simulation results and actual operating conditions. Furthermore, empirical models typically use fixed parameters and lack consideration for fluctuations in the production process. In actual production, the physical properties of liquid materials and environmental conditions often change dynamically, making it difficult for assessment results to adapt to different operating conditions, impacting the stability and reliability of the predictions. Summary of the Invention

[0004] The present application provides a quantitative evaluation method and system for wall sticking of spray drying towers, which solves the technical problems of low accuracy and poor reliability of wall sticking evaluation in the prior art due to reliance on a single physical model or data model and lack of adaptability to operating condition fluctuations. The application achieves the technical effect of improving the accuracy and reliability of wall sticking evaluation and enhancing dynamic adaptability to complex operating conditions.

[0005] In view of the above problems, on the one hand, the present application provides a quantitative evaluation method for wall adhesion of a spray drying tower, which includes: collecting multi-dimensional correlation factors in the process of drying and powdering liquid materials, and collecting various operating parameters in the spray drying tower at fixed points to obtain multi-dimensional correlation data and several operating parameter sequences; using a powder adhesion simulation model and a powder adhesion prediction model, respectively performing powder adhesion analysis based on the multi-dimensional correlation data and several operating parameter sequences, and outputting a powder adhesion simulation distribution and a powder adhesion prediction distribution; performing a powdering volatility analysis based on the several operating parameter sequences, outputting a powdering volatility coefficient, and configuring a model output weight; fitting the powder adhesion simulation distribution and the powder adhesion prediction distribution based on the model output weight, and outputting a powder adhesion evaluation result.

[0006] On the other hand, the present application also provides a quantitative wall adhesion evaluation system for spray drying towers, which includes: an acquisition module for collecting multi-dimensional correlation factors in the liquid material drying and powdering process, and fixed-point collection of various operating parameters in the spray drying tower to obtain multi-dimensional correlation data and several operating parameter sequences; a simulation analysis module for using a powder wall adhesion simulation model and a powder wall adhesion prediction model to perform powder wall adhesion analysis according to the multi-dimensional correlation data and several operating parameter sequences, and output a powder wall adhesion simulation distribution and a powder wall adhesion prediction distribution; a fluctuation analysis module for performing a powdering volatility analysis according to the several operating parameter sequences, outputting a powdering fluctuation coefficient, and configuring a model output weight; a fitting evaluation module for fitting the powder wall adhesion simulation distribution and the powder wall adhesion prediction distribution according to the model output weight, and outputting a powder wall adhesion evaluation result.

[0007] One or more technical solutions provided in this application have at least the following beneficial effects: By collecting and acquiring multi-dimensional correlated data and several operating parameter sequences, a complete data foundation is constructed to improve the comprehensiveness and representativeness of the data. Using the powder wall adhesion simulation model and the powder wall adhesion prediction model, powder wall adhesion analysis is performed based on the multi-dimensional correlated data and several operating parameter sequences, and the powder wall adhesion simulation distribution and powder wall adhesion prediction distribution are output. This step simulates the wall adhesion behavior of the powder in the drying tower through a physical model, and uses a machine learning model combined with historical data for prediction, achieving a dual evaluation based on mechanism analysis and data-driven, thereby improving prediction accuracy. Based on the several operating parameter sequences, a powder production volatility analysis is performed, taking into account the volatility in the production process, quantifying the dynamic changes in the working conditions, outputting the powder production volatility coefficient, and adjusting the model weight according to the fluctuation state, making the evaluation method more adaptable and enhancing its adaptability to complex working conditions. By adaptively weighting the analysis results of the physical model and the data model, accurate evaluation under different working conditions is achieved, improving the stability, reliability and generalization ability of the prediction.

[0008] In summary, this application achieves a precise assessment of powder wall adhesion behavior by comprehensively collecting multi-dimensional operating parameters during the spray drying process, combining physical simulation of microfluid mechanics with machine learning prediction models. Simultaneously, it introduces milling volatility analysis and dynamically adjusts model output weights to accommodate changes in different production conditions. Ultimately, through adaptive weighted fusion of physical simulation and data prediction results, the accuracy, reliability, and adaptability of wall adhesion assessment to complex operating conditions are improved, thereby providing technical support for spray drying process optimization and quality control.

[0009] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 A schematic flow chart of a method for quantitatively evaluating wall adhesion in a spray drying tower provided in an embodiment of the present application.

[0011] Figure 2 A schematic diagram of the process of constructing a powder wall adhesion prediction model in the quantitative evaluation method for spray drying tower wall adhesion provided in an embodiment of the present application.

[0012] Figure 3 Schematic diagram of the structure of the quantitative evaluation system for spray drying tower wall adhesion provided in an embodiment of the present application.

[0013] Description of reference numerals: acquisition module 10 , simulation analysis module 20 , fluctuation analysis module 30 , fitting evaluation module 40 . DETAILED DESCRIPTION

[0014] The embodiments of the present application provide a quantitative wall sticking assessment method and system for a spray drying tower, thereby solving the technical problems of the prior art in which the wall sticking assessment has low accuracy and poor reliability due to reliance on a single physical model or data model and lack of adaptability to operating condition fluctuations. This achieves the technical effect of improving the accuracy and reliability of wall sticking assessment and enhancing dynamic adaptability to complex operating conditions.

[0015] Example 1, as Figure 1 As shown, the embodiment of the present application provides a method for quantitatively evaluating wall adhesion of a spray drying tower, the method comprising: Step S1: Collect multi-dimensional correlation factors in the process of drying and pulverizing the liquid material, and collect various operating parameters in the spray drying tower at fixed points to obtain multi-dimensional correlation data and several operating parameter sequences.

[0016] Specifically, liquid materials, such as milk and protein solutions, are those that need to be dried into powder during the spray drying process. Multidimensional factors refer to the multiple variables that influence wall sticking, including environmental factors (humidity, temperature), material properties (viscosity, solids content), and process parameters (spray pressure, feed rate). Operating parameters include airflow velocity, temperature, humidity, and pressure within the spray drying tower.

[0017] Comprehensively collect the multi-dimensional correlation factors of the liquid material during the drying and pulverizing process, as well as various operating parameters within the spray drying tower. Specifically, specialized testing equipment is used to measure the physical and chemical properties of the liquid material, such as viscosity using a viscometer, surface tension using a surface tensiometer, and composition analysis using a chromatograph. Simultaneously, multiple sensors and data acquisition devices are deployed within the spray drying tower to capture operating parameters. Operating parameters such as airflow velocity, temperature, humidity, and pressure are collected at fixed points to form an operating parameter sequence. For example, temperature sensors (thermocouples, infrared thermometers) monitor the temperature gradient within the drying tower; humidity sensors detect air moisture content to assess the degree of drying; wind speed sensors measure airflow velocity to determine drying dynamics; and pressure sensors monitor pressure distribution within the tower to optimize spray conditions.

[0018] By collecting multi-dimensional correlation factors and operating parameter sequences, the key factors affecting wall adhesion can be comprehensively collected, providing an accurate and rich data basis for subsequent analysis and improving the accuracy of wall adhesion assessment.

[0019] Step S2: using the powder sticking simulation model and the powder sticking prediction model, respectively perform powder sticking analysis according to the multi-dimensional correlation data and a plurality of operation parameter sequences, and output a powder sticking simulation distribution and a powder sticking prediction distribution.

[0020] Specifically, the powder sticking simulation model, based on computational fluid dynamics (CFD), simulates the movement and sticking distribution of powder within a spray dryer. The powder sticking prediction model, trained using a machine learning algorithm, predicts the likelihood and extent of powder sticking based on historical data. Multidimensional correlation data and a sequence of operating parameters are input into the powder sticking simulation model to simulate the flow field, temperature field, and particle movement within the tower, resulting in a simulated distribution of powder sticking. Simultaneously, machine learning models (such as neural networks and support vector machines) are trained on historical data to learn the complex relationships between sticking and various factors. Current multidimensional correlation data and a sequence of operating parameters are then input for prediction, yielding a predicted distribution of powder sticking.

[0021] By combining two different types of models, one based on physical simulation and the other based on data-driven, the powder sticking behavior can be analyzed from different perspectives. The simulation model explains the sticking phenomenon from the physical mechanism, and the prediction model predicts the sticking situation based on the statistical laws of the data. The two models complement each other and improve the accuracy and comprehensiveness of the sticking analysis.

[0022] Step S3: performing milling volatility analysis based on the plurality of operating parameter sequences, outputting a milling volatility coefficient, and configuring a model output weight.

[0023] Specifically, a volatility analysis is performed on the operating parameter sequence. Statistical analysis methods (such as standard deviation and variance) are used to assess the degree of fluctuation in parameters such as airflow velocity, temperature, and humidity. The coefficient of fluctuation for each parameter is calculated. This milling fluctuation coefficient quantifies the degree of fluctuation in operating parameters and reflects the degree of impact of the parameter on wall sticking. Model output weights are configured based on the magnitude of the fluctuation coefficient. Parameters with larger fluctuation coefficients have a more significant impact on wall sticking and are therefore given higher weights in the model. Adaptive weighting algorithms (such as weighted averaging and Bayesian updating) are used to dynamically adjust model weights. For example, when volatility is high, the weight of the data-driven model is increased to accommodate complex operating conditions. When volatility is low, the weight of the physical model is increased to provide more reliable theoretical analysis. For example, in a milk powder factory, if the feed rate and air humidity of the production line fluctuate frequently, machine learning predictions (data models) may be more adaptable to these changes than physical simulations, so the weight of the data model is increased.

[0024] Through volatility analysis and weight configuration, the model's evaluation focus can be dynamically adjusted, which enhances the model's sensitivity to changes in key factors and improves the real-time and adaptability of the evaluation results.

[0025] Step S4: fitting the simulated distribution of powder sticking to the wall and the predicted distribution of powder sticking to the wall according to the model output weights, and outputting a powder sticking to the wall evaluation result.

[0026] Specifically, a weighted fusion algorithm (such as weighted averaging and Bayesian optimization) is used to combine the physical simulation results (simulated powder adhesion distribution) with the machine learning prediction results (and the predicted powder adhesion distribution) to obtain the final powder adhesion assessment result. By fusing the results of the two previous models, the advantages of both models are combined. The resulting powder adhesion assessment results take into account both the results of physical principle simulation and the results of data-driven prediction, making the final assessment more accurate and reliable, and better reflecting the adhesion of liquid materials during the dry powdering process.

[0027] Furthermore, step S1 of the embodiment of the present application includes: Step S11: Collect multi-dimensional correlation factors during liquid material drying and pulverizing to obtain multi-dimensional correlation data, wherein the multi-dimensional correlation data at least includes liquid material characteristics, air environment parameters, predetermined air drying parameters and predetermined atomization parameters.

[0028] Step S12: During the liquid material drying and pulverizing process, the air flow velocity, gas flow, temperature, humidity and pressure in the spray drying tower are collected at fixed points at predetermined time intervals, and are arranged in chronological order according to the monitoring time to obtain a plurality of operating parameter sequences, wherein the operating parameter sequence includes K continuous operating parameters, where K is an integer greater than 20.

[0029] Specifically, sensors, detection instruments, and online monitoring equipment are used to comprehensively collect multi-dimensional, correlated data during the drying and powdering of liquid materials. This multi-dimensional, correlated data includes at least liquid material properties, air environment parameters, predetermined air drying parameters, and predetermined atomization parameters. Liquid material properties refer to the physical and chemical properties of the liquid material, such as the properties of the concentrated emulsion (solids content, milk fat content, protein structure, etc.), the type of material, and the proportion of the material. To collect liquid material properties, a variety of analytical methods are required. Densitometers, viscometers, and refractometers are used to measure the solids content of the material (for example, the solids content of milk powder concentrate is typically between 40% and 55%). Fourier transform infrared spectroscopy (FTIR) is used to analyze protein structure and determine the denaturation of whey proteins. A milk fat analyzer is used to measure milk fat content to ensure the consistency of different batches. Air environment parameters refer to parameters such as temperature, humidity, and flow rate within the spray dryer. These parameters directly affect the drying process and powder formation. Temperature and humidity sensors can be used to measure the humidity and temperature inside and outside the spray dryer, and a dew point meter can be used to measure the moisture content of the air to ensure the stability of the drying air. Predetermined air drying parameters refer to the pre-set air drying conditions during the spray drying process, such as inlet and outlet air temperatures, and air flow rate. Predetermined atomization parameters refer to the operating parameters of the atomizer, such as atomization pressure and atomization speed, which affect droplet size and distribution. Inlet and outlet air temperatures can be measured using thermocouples and infrared thermometers, while nozzle pressure can be monitored using differential pressure sensors to ensure stable atomization. Multi-dimensional correlation factor collection provides comprehensive data on material, environmental, and process parameters, ensuring accurate input information for the spray drying process and providing data support for subsequent wall adhesion assessments.

[0030] During the liquid material drying and pulverizing process, parameters such as airflow velocity, gas flow, temperature, humidity, and pressure are collected at fixed intervals (predetermined time intervals) within the spray drying tower. Specifically, a sensor network, such as temperature, humidity, and pressure sensors, is installed within the tower to continuously collect data at K time points at set intervals (e.g., every minute or every second). The collected data is then arranged in chronological order to form an operating parameter sequence. K is an integer greater than 20, representing the number of continuous operating parameters included in the operating parameter sequence. For example, multiple sensors are installed at different locations in the spray drying tower, collecting data every minute for 30 minutes, resulting in a 30-point operating parameter sequence. By collecting operating parameters at fixed points and forming a sequence, dynamic changes in these parameters during the drying process can be captured, providing detailed data support for subsequent fluctuation analysis and wall adhesion assessment, improving the real-time and accuracy of the assessment.

[0031] Furthermore, the construction process of the powder adhesion simulation model includes: Step S01: Collect the geometric structure, size information and inner wall smoothness of the spray drying tower, and use CFD software to construct a three-dimensional geometric model of the spray drying tower.

[0032] Step S02: constructing a powder wall adhesion simulation model based on the microfluid mechanics principle and the three-dimensional geometric model.

[0033] Specifically, laser scanning or CAD drawings are used to obtain information about the drying tower's shape and dimensions, including its height, diameter, and inlet and outlet shapes. For example, a typical milk powder spray drying tower is cylindrical, 10 to 20 meters tall and 5 to 10 meters in diameter. A surface profilometer is used to measure the tower's inner wall smoothness (e.g., Ra value, which quantifies surface roughness). Inner wall smoothness refers to the degree of roughness on the tower's inner wall, which affects powder adhesion; smoother surfaces are less likely to cause powder deposition than rougher ones. Using CFD pre-processing software (such as ANSYS Fluent and COMSOL Multiphysics), a 3D geometric model of the spray drying tower is constructed based on its geometry, dimensions, and inner wall smoothness. Key features, such as the air inlet, exhaust, and spray nozzle locations, are ensured to facilitate subsequent fluid simulations.

[0034] CFD software (computational fluid dynamics) is used to simulate physical phenomena such as fluid flow, heat and mass transfer. Based on mathematical models (such as the Navier-Stokes equations), it uses numerical calculations to determine the spatial and temporal distributions of various fluid quantities (such as velocity, pressure, and temperature). Based on the principles of microfluid mechanics and a pre-established three-dimensional geometric model, a powder adhesion simulation model was constructed using CFD software. Specifically, various parameters must be set within the model, including fluid properties (such as air velocity, temperature, and humidity), spray particle information (such as particle size distribution and spray pressure), material characteristics (such as milk powder emulsion concentration, milk fat content, and protein structure), and operating conditions (such as inlet temperature, humidity, and flow rate). At the microscale, van der Waals forces between particles and the tower wall cause particle adhesion. Furthermore, particles may carry a surface charge, which attracts the tower wall, generating electrostatic forces that contribute to adhesion. Furthermore, particle elasticity and the microscopic roughness of the tower wall surface also influence adhesion. Using the CFD software's solver, the critical adhesion conditions between the powdered milk particles and the tower wall are calculated based on microfluidic principles (such as van der Waals forces, electrostatic forces, and surface tension), along with parameters such as the powdered milk powder's particle size distribution, surface smoothness, and external environment (such as humidity and airflow velocity). This allows the simulation of the powder particles' movement and wall adhesion within the tower. Based on the simulation results, the specific areas of particle adhesion are determined, and their location, area, and adhesion strength are analyzed. The deposition of each particle under specific conditions is also simulated, further analyzing the particle adhesion behavior. The CFD software's post-processing function then calculates the density and coverage of the adhered particles, generating a quantitative analysis of their wall adhesion.

[0035] The powder sticking simulation model constructed through the above steps comprehensively considers the influence of various factors on powder sticking based on the principles of microfluid mechanics. Through detailed parameter settings and numerical simulation calculations, it accurately simulates the powder sticking situation in the spray dryer, including the sticking area, sticking particle density, and coverage area, providing a theoretical basis for the analysis and evaluation of powder sticking.

[0036] Further, such as Figure 2 As shown, the construction process of the powder wall adhesion prediction model includes: Step S03: using the similarity tolerance interval, the geometric structure, size information and inner wall smoothness of the spray drying tower are expanded to generate similarity comparison conditions.

[0037] Step S04: Using the similarity comparison condition as a constraint, big data is used to retrieve the operation records of similar spray drying towers, collect sample association data sets and several sample operation parameter sequence sets, and obtain the powder wall adhesion distribution under different sample association data and several sample operation parameter sequences to obtain the sample powder wall adhesion distribution set.

[0038] Step S05: Based on the principle of ensemble learning, the sample association data set, several sample operation parameter sequence sets and sample powder wall adhesion distribution sets are used as training data to construct a powder wall adhesion prediction model.

[0039] Specifically, the similarity tolerance range refers to the allowable range of similarity in terms of geometry, dimensional information, and inner wall smoothness, allowing for the identification of similar cases. The similarity comparison condition is derived from the expanded geometry, dimensional information, and inner wall smoothness of the spray dryer. It is used to filter out operating records of similar spray dryers that meet the requirements from the big data. The tolerance ranges for geometry and dimensions are set, for example: ±15% variation in tower height; ±10% variation in diameter; ±5° variation in spray angle; and ±0.1µm variation in inner wall roughness (Ra). Based on these tolerance ranges, the geometry, dimensional information, and inner wall smoothness of the spray dryer are mathematically expanded to generate the similarity comparison condition. For example, if the original height of the spray dryer is h and the similarity tolerance range is ±ɑ, then the height range in the similarity comparison condition is [h(1-ɑ) to h(1+ɑ)]. For example, the parameters of a spray dryer A are: 8-meter diameter, 15-meter height, and an inner wall Ra value of 0.2µm. According to the set similarity tolerance range, the similarity comparison conditions are generated as follows: diameter between 7.2 meters and 8.8 meters, tower height between 12.75 meters and 17.25 meters, and Ra value between 0.1µm and 0.3µm.

[0040] Using the generated similarity comparison criteria as constraints, a database management system or big data analytics platform (such as Hadoop and Spark) is used to retrieve a large number of spray dryer operation records. For spray dryers that meet the similarity comparison criteria, relevant data is collected, including sample association datasets and several sets of sample operating parameter sequences, as well as the powder wall adhesion distribution under different sample association data and several sample operating parameter sequences, to obtain the sample powder wall adhesion distribution set. Through big data retrieval and sample dataset collection, a rich set of actual operating data can be obtained, providing a reliable foundation for model training and improving the model's generalization ability and predictive accuracy.

[0041] Based on the principles of ensemble learning, a powder adhesion prediction model is constructed using a sample association dataset, a sample operating parameter sequence set, and a sample powder adhesion distribution set as training data. Specifically, an appropriate ensemble learning algorithm (such as random forest or XGBoost) is selected and the training data is input into the algorithm for training. Taking random forest as an example, multiple sub-datasets are first extracted from the sample association dataset and the sample operating parameter sequence set. A simple decision tree model (weak learner) is then constructed on each sub-dataset. Each decision tree model is trained based on the sample powder adhesion distribution set. During training, the decision tree structure (such as node splitting rules and leaf node values) is continuously adjusted to minimize prediction error. Finally, multiple trained decision trees are combined to form a random forest model (strong learner), which serves as the powder adhesion prediction model.

[0042] The powder sticking prediction model, built based on ensemble learning principles, leverages the strengths of multiple weak learners to improve the ability to predict powder sticking. Compared to a single learning model, it offers improved accuracy, stability, and generalization, and can better adapt to different spray dryer operating conditions, providing a reliable basis for predicting powder sticking.

[0043] Furthermore, in step S04 of the embodiment of the present application, different sample association data and powder wall adhesion distribution under several sample operation parameter sequences are obtained, including: Step S04-1: collecting images of the inner wall of the spray drying tower under the first sample associated data and a plurality of first sample operating parameter sequences, and performing image analysis to determine a first wall-adhesive area.

[0044] Step S04-2: The powder adhering to the inner wall of the spray drying tower is peeled off by a physical peeling method, and then weighed. The weight is divided by the area of the corresponding peeled area to obtain a plurality of first wall adhesion densities.

[0045] Step S04 - 3 : constructing a first sample powder wall adhesion distribution according to the first wall adhesion area and a plurality of first wall adhesion densities.

[0046] Specifically, the first sample-related data can be any set of data from multiple sample-related data retrieved from the big data search. The first sample-related data corresponds to several operating parameter sequences, namely, first sample operating parameter sequences. When collecting sample data, an industrial endoscope or a high-definition camera is used to capture images of the spray drying tower's inner wall under the first sample-related data and several first sample operating parameter sequences. Images of the inner wall are captured from different positions and angles within the tower. The captured images are then analyzed using image processing software (such as ImageJ or OpenCV) to identify wall-adherent areas and calculate their areas. For example, during the milk powder production process, an industrial endoscope is used to capture images of the tower's inner wall. OpenCV software is used to segment the images to distinguish wall-adherent areas from non-wall-adherent areas. For example, segmentation can be performed based on color difference (if the wall-adherent powder is a different color from the tower wall) or texture features. The number of pixels in the wall-adherent areas is then counted, and a first wall-adherent area is calculated based on known image proportional relationships (such as the actual area corresponding to each pixel). Through image acquisition and analysis, information on the wall-adherent area of the spray drying tower's inner wall is directly obtained. This method is relatively intuitive and can quickly obtain the distribution range of wall adhesion on the tower wall, providing important area dimension information for the subsequent construction of powder wall adhesion distribution, which helps to more comprehensively understand the powder wall adhesion situation.

[0047] After the spray dryer ceases operation, a physical stripping method is used to remove powder adhered to different areas of the spray dryer's inner wall. Specifically, a scraper or specialized stripping tool is used to carefully scrape off any powder adhering to the tower wall and collect it in a weighing bottle. The mass of the stripped powder is then measured using a high-precision balance and divided by the area of the corresponding stripped area to obtain multiple first wall adhesion densities. For example, if the mass of powder stripped from a specific stripping area (e.g., 1 square meter) is 50 grams, the wall adhesion density for that area is 50 grams per square meter.

[0048] The first sample powder wall adhesion distribution is constructed based on the first wall adhesion area and multiple first wall adhesion densities. The spray dryer's inner wall can be divided into several small regions (based on coordinates or specific partitioning rules). For each small region, the corresponding first wall adhesion area and first wall adhesion density data are integrated. For example, a two-dimensional matrix or data structure can be created, where rows represent different small regions and columns represent information such as wall adhesion area and wall adhesion density, to construct the first sample powder wall adhesion distribution. By constructing the first sample powder wall adhesion distribution, a comprehensive and quantitative description of wall adhesion under specific conditions can be achieved, providing detailed data support for subsequent model training and prediction.

[0049] Furthermore, step S05 includes: Step S05-1: Configure Q ensemble learning operators, wherein the ensemble learning operators include at least a feedforward neural network, a random forest, and a gradient boosting tree, wherein Q is an integer greater than or equal to 3.

[0050] Step S05-2: Use the sample association data set, several sample operation parameter sequence sets and sample powder wall adhesion distribution sets as training data, and divide them into Q equal parts, and divide them into Q training sets and Q test sets in proportion.

[0051] Step S05-3: Using the Q training sets and the Q test sets, Q ensemble learning operators are trained and tested respectively to obtain Q powder wall adhesion prediction branches.

[0052] Step S05-4: Based on the principle of ensemble learning, a powder wall sticking prediction model is constructed according to the Q powder wall sticking prediction branches.

[0053] Specifically, ensemble learning operators refer to the specific algorithms or model structures used to build models in ensemble learning. These include at least feedforward neural networks, random forests, and gradient boosting trees. These ensemble learning operators have different characteristics and advantages. By combining these different operators, a more optimized powder adhesion prediction model can be constructed. Using a machine learning framework (such as TensorFlow, PyTorch, or scikit-learn), Q ensemble learning operators are selected and configured. Q is an integer greater than or equal to 3. This means that at least three different ensemble learning operators are selected to ensure the accuracy and generalization of the powder adhesion prediction model.

[0054] Using data processing software, the sample association data set, sample operation parameter sequence set, and sample powder wall adhesion distribution set are divided into Q parts. For each data set, it is divided into training set and test set according to a certain ratio (for example, the common 80% as training set and 20% as test set), resulting in Q training sets and Q test sets.

[0055] Each ensemble learning operator is trained using the corresponding training set. Taking a random forest as an example, the sample association dataset and sample operating parameter sequence set in the training set serve as input features, and the sample powder adhesion distribution set serves as the target label. By adjusting parameters such as the number of decision trees and tree depth in the random forest, the model learns the relationship between input and output. The trained model is then tested using the corresponding test set, and error metrics (such as mean squared error and accuracy) are calculated between the predicted and true results. Through this training and testing process, Q ensemble learning operators are trained and tested separately, resulting in Q powder adhesion prediction branches. Each powder adhesion prediction branch has the ability to predict powder adhesion. Since it is constructed based on different ensemble learning operators, each branch has different prediction characteristics and performance, providing multiple components with different prediction properties for constructing the final powder adhesion prediction model.

[0056] Based on the principle of ensemble learning, a powder adhesion prediction model is constructed based on Q powder adhesion prediction branches. First, the performance of the Q powder adhesion prediction branches on their respective test sets is evaluated. For example, the evaluation can be based on indicators such as the mean square error and accuracy between the predicted results and the true results. Then, a weight is assigned to each branch based on these performance indicators. For example, if a powder adhesion prediction branch has a higher accuracy and a lower mean square error, a higher weight can be assigned to it. Suppose the Q powder adhesion prediction branches are (M1, M2, …, M Q ), the corresponding weights are (ω1, ω2, …, ω Q ), and ω1+ω2+…+ω Q = 1. For new input data (such as new sample association data and sample operation parameter sequence), each branch M i A prediction result ri will be generated, so the final prediction result of the powder adhesion prediction model is R=∑ω i r i (i=1, 2,…, Q).

[0057] The powder adhesion prediction model constructed based on the principle of ensemble learning combines Q powder adhesion prediction branches through weighted averaging. It can integrate the advantages of each branch, improve the accuracy, stability and generalization ability of the prediction, and thus obtain a more accurate and reliable powder adhesion prediction distribution.

[0058] Furthermore, in step S3, milling fluctuation analysis is performed according to the plurality of operating parameter sequences, and a milling fluctuation coefficient is output, including: Step S31: Acquire several operating parameter sequences, including an air flow velocity sequence, a gas flow sequence, a temperature sequence, a humidity sequence, and a pressure sequence.

[0059] Step S32: performing fluctuation analysis on the plurality of operating parameter sequences respectively to obtain a plurality of fluctuation coefficients.

[0060] Step S33: Obtain the milling fluctuation coefficient according to the weighted fluctuation coefficients, wherein the weight of the operating parameters is positively correlated with the influence of the powder sticking to the wall.

[0061] Specifically, the operating parameter sequences include airflow velocity, gas flow, temperature, humidity, and pressure, which are continuous data for the corresponding parameters within a certain time interval. Sensors installed in the spray dryer collect parameters such as airflow velocity, gas flow, temperature, humidity, and pressure at predetermined time intervals to generate the corresponding sequence data.

[0062] Perform fluctuation analysis on several operating parameter sequences to obtain several fluctuation coefficients. Taking the airflow velocity sequence as an example, use statistical methods (such as standard deviation, variance, and coefficient of variation) to calculate its fluctuation coefficient. For example, first calculate the mean of the airflow velocity sequence, then calculate the sum of the squares of the differences between each data point and the mean, divide this by the number of data points to obtain the variance, and finally take the square root of the variance to obtain the standard deviation σ. This standard deviation σ can be used as the fluctuation coefficient of the airflow velocity sequence. Similar methods are used to calculate the fluctuation coefficients of the remaining operating parameter sequences, resulting in several fluctuation coefficients.

[0063] The milling fluctuation coefficient is obtained by weighting several fluctuation coefficients. Specifically, the influence of each operating parameter on powder wall adhesion is first determined, thereby determining its weight. Assume that the fluctuation coefficients of airflow velocity, gas flow rate, temperature, humidity, and pressure are k1, k2, k3, k4, and k5, respectively, and the corresponding weights are ω1, ω2, ω3, ω4, and ω5, respectively. Based on the principle that the operating parameter weights are positively correlated with the influence of powder wall adhesion, these weights are determined through experimentation, experience, or data analysis. For example, experiments have found that the influence of temperature on powder wall adhesion is twice that of airflow velocity. In this case, the weight distribution is ω3 = 2ω1. Then, the milling fluctuation coefficient K is calculated using the weighted average method, using the formula K = ω1k1 + ω2k2 + ω3k3 + ω4k4 + ω5k5.

[0064] The milling fluctuation coefficient, derived from the above steps, comprehensively considers the fluctuations in various operating parameters and their impact on powder sticking. Through a weighted calculation, the milling fluctuation coefficient more comprehensively and accurately reflects the overall fluctuations in the milling process, providing an important reference indicator for monitoring and optimizing the milling process.

[0065] Furthermore, step S32 includes: A first operating parameter sequence is randomly selected, and a first parameter standard deviation and a first parameter mean are calculated; and a ratio of the first parameter standard deviation to the first parameter mean is set as a first fluctuation coefficient.

[0066] Specifically, in one embodiment, when calculating the fluctuation coefficient, a random operating parameter sequence is selected from a number of operating parameter sequences as the first operating parameter sequence. For example, an airflow velocity sequence is randomly selected from an airflow velocity sequence, a gas flow sequence, a temperature sequence, a humidity sequence, and a pressure sequence. The standard deviation and mean of this sequence are then calculated, and the ratio of the standard deviation to the mean is set as the first fluctuation coefficient. Fluctuation analysis quantifies the degree of fluctuation of each operating parameter, providing a basis for subsequent weight assignment and enhancing the model's sensitivity to changes in key factors.

[0067] Furthermore, in step S3, the model output weights are configured, including: Step S34: Calculate the ratio of the historical milling fluctuation coefficient average to the milling fluctuation coefficient, and set it as the prediction weight adjustment coefficient.

[0068] Step S35: multiplying the prediction weight adjustment coefficient by the predetermined prediction weight to obtain the real-time prediction weight, wherein the predetermined prediction weight is the initial weight of the powder wall adhesion prediction model, and the value is 0.6.

[0069] Step S36: Subtract the real-time prediction weight from 1 to obtain the real-time simulation weight, and use the real-time prediction weight and the real-time simulation weight as the model output weight.

[0070] Specifically, the mean value of the historical milling fluctuation coefficient is the average value of the milling fluctuation coefficient in the past period of time (historical period), which reflects the average fluctuation level of the milling process in the historical stage. The historical milling fluctuation coefficient is obtained from the historical data storage and its mean value is calculated. Then, the milling fluctuation coefficient K calculated in the current step is obtained (calculated in the previous step S33). Finally, the ratio of the mean value of the historical milling fluctuation coefficient to the milling fluctuation coefficient is calculated to obtain the prediction weight adjustment coefficient. By calculating the prediction weight adjustment coefficient, the connection between the historical milling fluctuation level and the current milling fluctuation level is established. The ratio of the mean value of the historical milling fluctuation coefficient to the milling fluctuation coefficient can reflect the degree of change of the current milling fluctuation situation relative to the historical situation, and provides a basis for the subsequent adjustment of the prediction weight.

[0071] The pre-set prediction weight is the initial weight of the powder sticking prediction model, with a value of 0.6, indicating the initial importance of the prediction model in the model output. The prediction weight adjustment coefficient is multiplied by the pre-set prediction weight to obtain the real-time prediction weight, which reflects the impact of current operating parameter fluctuations on the prediction model.

[0072] Subtracting the real-time prediction weight from 1 yields the real-time simulation weight, which is used to adjust the importance of the simulation model in the final evaluation. For example, if the prediction weight adjustment factor is 0.8, the real-time prediction weight is 0.8 × 0.6 = 0.48. The real-time simulation weight = 1 - 0.48 = 0.52. The real-time prediction weight and the real-time simulation weight are then used as the model output weights for calculating the final powder adhesion evaluation results.

[0073] The above-mentioned weight distribution method can reasonably distribute the weights of the prediction and simulation parts in the powder adhesion prediction model according to the milling fluctuation situation, so that the model can predict and simulate powder adhesion more flexibly and accurately under different milling fluctuation situations.

[0074] In summary, the quantitative evaluation method for spray drying tower wall adhesion provided by the embodiments of the present application has the following beneficial effects: This embodiment of the present application achieves accurate assessment and prediction of powder wall adhesion in a spray dryer through four core steps: multi-dimensional data acquisition, physical simulation modeling, machine learning predictive modeling, and adaptive weighted fusion. First, in the data acquisition phase, liquid material properties, air environment parameters, and predetermined drying and atomization parameters are extracted. Key operating parameters such as airflow velocity, temperature, humidity, and pressure are collected at fixed points during the drying process to generate multi-dimensional correlated data and time series data, providing a comprehensive data foundation for subsequent analysis. Subsequently, in the physical simulation modeling phase, a three-dimensional simulation model is constructed using CFD (computational fluid dynamics) software based on the spray dryer's geometry, fluid properties, and particle characteristics. Incorporating the principles of microfluidics, the adhesion force, motion trajectory, and critical adhesion conditions of the particles are simulated to quantitatively predict the wall adhesion distribution under different conditions. Next, in the machine learning predictive modeling phase, similar comparison conditions for the spray dryer are expanded, operating records of similar equipment are retrieved from big data, and historical wall adhesion distribution data is obtained through image analysis and physical exfoliation experiments. The prediction model is trained using ensemble learning methods (including feedforward neural networks, random forests, and gradient boosting trees), thereby establishing a data-driven powder wall adhesion prediction model. Then, during the operating fluctuation analysis phase, milling volatility is calculated based on the real-time collection of operating parameter sequences. The prediction model weights are adjusted using the historical fluctuation mean, emphasizing greater reliance on machine learning predictions when fluctuations are small, while increasing the weight of the physical simulation model when fluctuations are large, ensuring the stability and reliability of the evaluation results. Finally, through adaptive weighted fusion, the output weights of the physical simulation and machine learning prediction models are dynamically adjusted according to the milling fluctuation coefficient, and the final powder adhesion evaluation results are obtained using weighted calculations.

[0075] Overall, the present embodiment achieves a precise assessment of powder wall adhesion behavior by comprehensively collecting multi-dimensional operating parameters during the spray drying process, combining physical simulations of microfluid mechanics with machine learning prediction models. Furthermore, a milling volatility analysis is introduced to dynamically adjust the model output weights to accommodate changes in different production conditions. Ultimately, through the adaptive weighted fusion of physical simulation and data prediction results, the accuracy, reliability, and adaptability of the wall adhesion assessment to complex operating conditions are improved, thereby providing technical support for spray drying process optimization and quality control.

[0076] Example 2, as Figure 3 As shown, based on the same inventive concept as the aforementioned embodiment 1, the embodiment of the present application provides a quantitative evaluation system for wall adhesion of a spray drying tower, the system comprising: The acquisition module 10 is used to collect multi-dimensional correlation factors in the process of drying and pulverizing the liquid material, and to collect various operating parameters in the spray drying tower at fixed points to obtain multi-dimensional correlation data and several operating parameter sequences.

[0077] The simulation analysis module 20 is used to use the powder sticking simulation model and the powder sticking prediction model to perform powder sticking analysis according to the multi-dimensional correlation data and a plurality of operation parameter sequences, and output the powder sticking simulation distribution and the powder sticking prediction distribution.

[0078] The fluctuation analysis module 30 is used to perform milling fluctuation analysis based on the plurality of operating parameter sequences, output a milling fluctuation coefficient, and configure a model output weight.

[0079] The fitting evaluation module 40 is used to fit the simulated distribution of powder sticking to the wall and the predicted distribution of powder sticking to the wall according to the weights output by the model, and output a powder sticking to the wall evaluation result.

[0080] Furthermore, the acquisition module 10 in the embodiment of the present application is further configured to perform the following steps: Multi-dimensional correlation factors during liquid material drying and pulverizing are collected to obtain multi-dimensional correlation data, wherein the multi-dimensional correlation data at least includes liquid material characteristics, air environment parameters, predetermined air drying parameters, and predetermined atomization parameters; during the liquid material drying and pulverizing process, the air flow velocity, gas flow, temperature, humidity, and pressure in the spray drying tower are collected at fixed points at predetermined time intervals, and are arranged in chronological order according to the monitoring time to obtain a plurality of operating parameter sequences, wherein the operating parameter sequence includes K continuous operating parameters, where K is an integer greater than 20.

[0081] Furthermore, the system of the embodiment of the present application further includes a first model building module, which is used to build a powder sticking to the wall simulation model, and the execution steps include: The geometric structure, size information and inner wall smoothness of the spray drying tower are collected, and a three-dimensional geometric model of the spray drying tower is constructed using CFD software. A powder wall adhesion simulation model is constructed based on the principles of microfluid mechanics and the three-dimensional geometric model.

[0082] Furthermore, the system of the embodiment of the present application further includes a second model building module, which is used to build a powder wall adhesion prediction model, and the execution steps include: The geometric structure, dimensional information and inner wall smoothness of the spray drying tower are expanded using the similarity tolerance interval to generate similar comparison conditions. Using the similarity comparison conditions as constraints, the operation records of similar spray drying towers are retrieved through big data, and a sample association data set and several sample operation parameter sequence sets are collected. The powder adhesion distribution under different sample association data and several sample operation parameter sequences is obtained to obtain a sample powder adhesion distribution set. Based on the principle of ensemble learning, the sample association data set, several sample operation parameter sequence sets and sample powder adhesion distribution sets are used as training data to construct a powder adhesion prediction model.

[0083] Furthermore, the second model building module is further configured to perform the following steps: Images of the inner wall of the spray drying tower under the first sample associated data and several first sample operating parameter sequences are collected, and image analysis is performed to determine the first wall adhesion area; the powder attached to the inner wall of the spray drying tower is peeled off by a physical peeling method and then weighed, and the weight is divided by the area of the corresponding peeling area to obtain multiple first wall adhesion densities; and the first sample powder wall adhesion distribution is constructed based on the first wall adhesion area and the multiple first wall adhesion densities.

[0084] Furthermore, the second model building module is further configured to perform the following steps: Q ensemble learning operators are configured, wherein the ensemble learning operators include at least a feedforward neural network, a random forest, and a gradient boosting tree, wherein Q is an integer greater than or equal to 3; the sample association data set, several sample operation parameter sequence sets, and a sample powder adhesion distribution set are used as training data, and are divided into Q equal parts, and then divided into Q training sets and Q test sets in proportion; the Q ensemble learning operators are trained and tested using the Q training sets and Q test sets, respectively, to obtain Q powder adhesion prediction branches; based on the ensemble learning principle, a powder adhesion prediction model is constructed according to the Q powder adhesion prediction branches.

[0085] Furthermore, the fluctuation analysis module 30 of the embodiment of the present application is further configured to perform the following steps: Acquire several operating parameter sequences, including an air flow velocity sequence, a gas flow sequence, a temperature sequence, a humidity sequence, and a pressure sequence; perform fluctuation analysis on the several operating parameter sequences respectively to obtain several fluctuation coefficients; obtain the powder making fluctuation coefficient according to weighted fluctuation coefficients, wherein the operating parameter weight is positively correlated with the degree of powder wall adhesion.

[0086] Furthermore, the fluctuation analysis module 30 of the embodiment of the present application is further configured to perform the following steps: A first operating parameter sequence is randomly selected, and a first parameter standard deviation and a first parameter mean are calculated; and a ratio of the first parameter standard deviation to the first parameter mean is set as a first fluctuation coefficient.

[0087] Furthermore, the fluctuation analysis module 30 of the embodiment of the present application is further configured to perform the following steps: Calculate the ratio of the historical mean of the milling fluctuation coefficient to the milling fluctuation coefficient, and set it as the prediction weight adjustment coefficient; multiply the prediction weight adjustment coefficient by the predetermined prediction weight to obtain the real-time prediction weight, wherein the predetermined prediction weight is the initial weight of the powder adhesion prediction model, and the value is 0.6; subtract the real-time prediction weight from 1 to obtain the real-time simulation weight, and use the real-time prediction weight and the real-time simulation weight as the model output weight.

[0088] Through the detailed description of the quantitative evaluation method for wall adhesion of a spray drying tower in the foregoing specification, those skilled in the art can clearly understand the quantitative evaluation system for wall adhesion of a spray drying tower in this embodiment. For the system disclosed in Example 2, since it corresponds to the method disclosed in Example 1 and has corresponding functional modules and beneficial effects, the relevant details can be referred to the description of the method section.

[0089] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present application. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application is not limited to the embodiments shown herein, but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A quantitative evaluation method for spray drying tower wall adhesion, characterized in that the method include: Collect multi-dimensional correlation factors in the liquid material drying and pulverizing process, and collect various operating parameters in the spray drying tower at fixed points to obtain multi-dimensional correlation data and several operating parameter sequences; Using a powder sticking simulation model and a powder sticking prediction model, respectively performing powder sticking analysis based on the multi-dimensional correlation data and a plurality of operating parameter sequences, and outputting a powder sticking simulation distribution and a powder sticking prediction distribution; Performing milling volatility analysis based on the plurality of operating parameter sequences, outputting a milling volatility coefficient, and configuring a model output weight; According to the model output weight, the powder wall sticking simulation distribution and the powder wall sticking prediction distribution are fitted, and a powder wall sticking evaluation result is output.

2. The method for quantitatively evaluating wall adhesion of a spray drying tower according to claim 1, wherein: Obtain multi-dimensional correlation data and several operating parameter sequences, including: Collecting multi-dimensional correlation factors during the drying and pulverizing of liquid materials to obtain multi-dimensional correlation data, wherein the multi-dimensional correlation data at least includes liquid material characteristics, air environment parameters, predetermined air drying parameters, and predetermined atomization parameters; During the liquid material drying and pulverizing process, the air flow velocity, gas flow, temperature, humidity and pressure in the spray drying tower are collected at fixed points at predetermined time intervals and arranged in chronological order according to the monitoring time to obtain a number of operating parameter sequences, wherein the operating parameter sequence includes K continuous operating parameters, where K is an integer greater than 20.

3. The method for quantitatively evaluating wall adhesion of a spray drying tower according to claim 1, wherein: The construction process of the powder adhesion simulation model includes: Collect the geometric structure, size and inner wall smoothness of the spray drying tower, and use CFD software to build a three-dimensional geometric model of the spray drying tower; A powder adhesion simulation model is constructed based on the principles of microfluid mechanics and the three-dimensional geometric model.

4. The method for quantitatively evaluating wall adhesion of a spray drying tower according to claim 3, wherein: The process of constructing the powder wall adhesion prediction model includes: Using the similar tolerance interval, the geometric structure, size information and inner wall smoothness of the spray drying tower are expanded to generate similar comparison conditions; Using the similarity comparison conditions as constraints, big data is used to retrieve the operation records of similar spray drying towers, collect sample association data sets and several sample operation parameter sequence sets, and obtain powder wall adhesion distributions under different sample association data and several sample operation parameter sequences to obtain a sample powder wall adhesion distribution set; Based on the principle of ensemble learning, a powder wall sticking prediction model is constructed by using the sample association data set, several sample operation parameter sequence sets and sample powder wall sticking distribution sets as training data.

5. The method for quantitatively evaluating wall adhesion of a spray drying tower according to claim 4, wherein: Obtain different sample correlation data and powder wall adhesion distribution under several sample operation parameter sequences, including: collecting images of the inner wall of the spray drying tower under the first sample associated data and a plurality of first sample operating parameter sequences, and performing image analysis to determine a first wall-adhesive area; The powder attached to the inner wall of the spray drying tower is peeled off by a physical peeling method, and then weighed, and the weight is divided by the area of the corresponding peeled area to obtain a plurality of first wall adhesion densities; A first sample powder wall adhesion distribution is constructed according to the first wall adhesion area and a plurality of first wall adhesion densities.

6. The method for quantitatively evaluating wall adhesion of a spray drying tower according to claim 4, wherein: Based on the principle of ensemble learning, the sample association data set, several sample operation parameter sequence sets and sample powder wall adhesion distribution sets are used as training data to construct a powder wall adhesion prediction model, including: Configure Q ensemble learning operators, where the ensemble learning operators include at least a feedforward neural network, a random forest, and a gradient boosting tree, where Q is an integer greater than or equal to 3; The sample association data set, several sample operation parameter sequence sets and sample powder wall adhesion distribution sets are used as training data, and are equally divided into Q parts, which are then divided into Q training sets and Q test sets in proportion; Using the Q training sets and the Q test sets, Q ensemble learning operators are trained and tested respectively to obtain Q powder wall adhesion prediction branches; Based on the principle of integrated learning, a powder sticking to the wall prediction model is constructed according to the Q powder sticking to the wall prediction branches.

7. The method for quantitatively evaluating wall adhesion of a spray drying tower according to claim 1, wherein: Perform milling fluctuation analysis based on the several operating parameter sequences and output a milling fluctuation coefficient, including: Acquire a plurality of operating parameter sequences, including an air flow velocity sequence, a gas flow sequence, a temperature sequence, a humidity sequence, and a pressure sequence; performing fluctuation analysis on the plurality of operating parameter sequences respectively to obtain a plurality of fluctuation coefficients; The milling fluctuation coefficient is obtained by weighting the plurality of fluctuation coefficients, wherein the weight of the operating parameters is positively correlated with the influence of the powder sticking to the wall.

8. The method for quantitatively evaluating wall adhesion of a spray drying tower according to claim 7, wherein: Performing fluctuation analysis on the several operating parameter sequences respectively, including: Randomly select a first operating parameter sequence, and calculate the first parameter standard deviation and the first parameter mean; The ratio of the first parameter standard deviation to the first parameter mean is set as a first fluctuation coefficient.

9. The method for quantitatively evaluating wall adhesion of a spray drying tower according to claim 1, wherein: Configure model output weights, including: Calculate the ratio of the average historical milling fluctuation coefficient to the milling fluctuation coefficient, and set it as the prediction weight adjustment coefficient; The prediction weight adjustment coefficient is multiplied by the predetermined prediction weight to obtain the real-time prediction weight, wherein the predetermined prediction weight is the initial weight of the powder wall adhesion prediction model, which is 0.6; The real-time prediction weight is subtracted from 1 to obtain the real-time simulation weight, and the real-time prediction weight and the real-time simulation weight are used as the model output weight.

10. A quantitative evaluation system for spray drying tower wall adhesion, characterized in that: The system is used to implement the method for quantitatively evaluating wall adhesion of a spray drying tower according to any one of claims 1 to 9, comprising: The acquisition module is used to collect multi-dimensional correlation factors in the liquid material drying and pulverizing process, as well as to collect various operating parameters in the spray drying tower at fixed points, to obtain multi-dimensional correlation data and several operating parameter sequences; a simulation analysis module for performing powder sticking analysis based on the multi-dimensional correlation data and a plurality of operating parameter sequences using a powder sticking simulation model and a powder sticking prediction model, and outputting a powder sticking simulation distribution and a powder sticking prediction distribution; A fluctuation analysis module is used to perform milling fluctuation analysis based on the plurality of operating parameter sequences, output a milling fluctuation coefficient, and configure a model output weight; The fitting evaluation module is used to fit the powder wall adhesion simulation distribution and the powder wall adhesion prediction distribution according to the model output weight, and output the powder wall adhesion evaluation result.

Images