CN-121980959-A - Parallel cylinder inter-particle liquid bridge capillary force prediction method based on machine learning

CN121980959ACN 121980959 ACN121980959 ACN 121980959ACN-121980959-A

Abstract

The invention discloses a machine learning-based parallel cylinder inter-particle liquid bridge capillary force prediction method, which comprises the steps of establishing a capillary force prediction model according to an original data set of parallel cylinder inter-particle liquid bridge capillary force, wherein the original data set is determined according to liquid bridge capillary force simulation results under various working condition parameters, the capillary force prediction model is used for training and determining a supervised learning model according to the original data set, and obtaining a capillary force prediction value under a working condition to be predicted according to the capillary force prediction model. According to the invention, the liquid bridge capillary force prediction model between parallel cylinders is obtained by using a machine learning method, so that accurate and efficient prediction of capillary force can be realized.

Inventors

YANG LEI
LIAO XUDONG
GONG XIANGYANG
ZHANG KUNYA
YANG CHAO

Assignees

中国科学院过程工程研究所

Dates

Publication Date: 20260505
Application Date: 20260225

Claims (7)

1. The machine learning-based parallel cylindrical inter-particle liquid bridge capillary force prediction method is characterized by comprising the following steps of: Establishing a capillary force prediction model according to an original data set of liquid bridge capillary forces among parallel cylindrical particles, wherein the original data set is determined according to liquid bridge capillary force simulation results under various working condition parameters, and the capillary force prediction model carries out training determination on a supervised learning model according to the original data set; and obtaining a capillary force predicted value under the working condition to be predicted according to the capillary force predicted model.
2. The machine learning based parallel cylindrical inter-particle liquid bridge capillary force prediction method of claim 1, wherein obtaining the raw dataset comprises: writing a parallel inter-cylinder liquid bridge simulation code according to a user manual of open source software Surface Evolver; according to the simulation codes, obtaining liquid bridge capillary force simulation results under the various working condition parameters; and constructing the original data set according to the liquid bridge capillary force simulation result.
3. The machine learning based parallel cylinder inter-particle liquid bridge capillary force prediction method of claim 2, wherein constructing the original dataset according to the liquid bridge capillary force simulation result comprises extracting liquid bridge capillary force data according to output results of batch simulation operation and integrating the liquid bridge capillary force data into a data file in a unified format to form the original dataset.
4. The machine learning based parallel cylindrical inter-particle liquid bridge capillary force prediction method of claim 1, wherein the training determination of the supervised learning model by the capillary force prediction model from the raw dataset comprises: calculating a critical fracture distance according to the liquid bridge fracture distance; according to the critical fracture distance, eliminating zero value data of which the particle spacing exceeds the critical fracture distance in the original data set and abnormal values generated by simulated divergence to obtain a cleaned target data set; and training the supervised learning model according to the target data set.
5. The machine learning based parallel cylindrical inter-particle liquid bridge capillary force prediction method of claim 4, wherein calculating the liquid bridge break distance comprises: ; wherein S rup is the liquid bridge breaking distance, D is the diameter of the cylindrical particles, Is the liquid-solid contact angle, V is the liquid bridge volume.
6. The machine learning based parallel cylinder inter-particle liquid bridge capillary force prediction method according to claim 4, wherein model parameter optimization training is performed on an artificial neural network according to the target data set; And packaging according to the trained artificial neural network to form the capillary force prediction model.
7. The machine learning-based parallel cylinder inter-particle liquid bridge capillary force prediction method according to claim 1, wherein obtaining a capillary force prediction value under a working condition to be predicted according to the capillary force prediction model comprises: Inputting parameters of the contact angle, the liquid quantity, the particle spacing and the particle diameter under experimental working conditions into the capillary force prediction model; And according to the output result of the capillary force prediction model, obtaining a prediction error compared with an experimental measurement value to verify the prediction accuracy of the capillary force prediction model.

Description

Parallel cylinder inter-particle liquid bridge capillary force prediction method based on machine learning Technical Field The invention belongs to the technical field of wet particle technology and process engineering, and particularly relates to a machine learning-based liquid bridge capillary force prediction method between parallel cylindrical particles. Background Wet particle systems play a key role in industrial processes (such as catalytic cracking, pelleting, coating and filtering) in chemical, energy and petroleum industries, and the flow behavior of the wet particle systems is directly related to the production efficiency, product quality and energy consumption of the industrial processes. When wet particles come into contact with each other, the liquid films at the contact points coalesce to form liquid bridges and create capillary bridge forces that are typically much greater than the gravity and van der Waals forces of the particles themselves, thereby affecting the macroscopic flow characteristics of the particles, such as adhesion, agglomeration, etc. Among the numerous particle morphologies, cylindrical particles are a common non-spherical particle, and find wide application in the preparation of biomass fuels, pharmaceutical tablets, catalyst supports, and fiber composites. Therefore, the quantitative understanding and accurate prediction of the liquid bridge force are of great significance for realizing the model construction and accurate regulation of the cylindrical wet particle process. The capillary force of the liquid bridge between the parallel cylinders can be described as a function of the volume of the liquid bridge, the contact angle, the inter-particle distance and the particle size of the particles, and can be obtained by adopting experimental measurement, analytical solution and a numerical simulation method. Compared with the other two methods, the liquid bridge capillary force under any working condition parameters can be obtained through numerical simulation, and along with the rapid development of computing hardware and a numerical method, the computing efficiency and the accuracy of the numerical simulation are continuously improved. However, when the liquid bridge force under a certain working condition needs to be obtained, the numerical simulation method needs to be calculated after the simulation parameter setting is modified, the process is long in time consumption, and the timeliness is extremely low. Machine learning (MACHINE LEARNING, ML) has progressed from the middle of the 20 th century to date, through a stage of review from symbol learning to statistical learning to deep learning. The method has the core advantages that the method can automatically learn rules from mass data, and can realize rapid prediction of target values without explicit programming and simulation. Therefore, the invention organically combines numerical simulation with machine learning to construct a model capable of rapidly predicting the capillary force of the liquid bridge between parallel cylinders. Disclosure of Invention In order to solve the technical problems, the invention provides a machine learning-based liquid bridge capillary force prediction method between parallel cylindrical particles, which avoids a complex numerical simulation or analytic solving process and can be used for liquid bridge force description of a cylindrical wet particle system in a fluidized bed. In order to achieve the above object, the present invention provides a machine learning-based method for predicting capillary force of liquid bridges between parallel cylindrical particles, comprising: Establishing a capillary force prediction model according to an original data set of liquid bridge capillary forces among parallel cylindrical particles, wherein the original data set is determined according to liquid bridge capillary force simulation results under various working condition parameters, and the capillary force prediction model carries out training determination on a supervised learning model according to the original data set; and obtaining a capillary force predicted value under the working condition to be predicted according to the capillary force predicted model. Optionally, acquiring the original dataset includes: writing a parallel inter-cylinder liquid bridge simulation code according to a user manual of open source software Surface Evolver; according to the simulation codes, obtaining liquid bridge capillary force simulation results under the various working condition parameters; and constructing the original data set according to the liquid bridge capillary force simulation result. Optionally, constructing the original data set according to the liquid bridge capillary force simulation result includes extracting liquid bridge capillary force data according to the output result of the batch simulation operation and integrating the liquid bridge capillary force data into a data file in a unifie