Intelligent monitoring and regulation method for strong cyclone airflow screening stage process

Abstract

The application relates to an intelligent monitoring and regulation method for a strong-rotation airflow screen grading process, which comprises the following steps: establishing a digital twin visual model of a strong-rotation airflow screen grading machine, simulating two-phase flow through simulation calculation, and determining key process parameters affecting the airflow screen flow field characteristics; measuring the flow field data of the strong-rotation airflow screen grading machine by using a measuring device to obtain the flow field data at different positions and under different key process parameters in the grading chamber as measured data; comparing the measured data with the simulation calculation results, verifying and correcting the flow field model; performing simulation calculation based on the corrected flow field model to obtain the flow field data at different positions and under different key process parameters in the grading chamber as simulation data; based on machine learning, a flow field characteristic prediction model is established to predict the flow field characteristics; when the predicted flow field characteristics are abnormal, the flow field characteristics are regulated by adjusting the key process parameter values so that the strong-rotation airflow screen grading machine can normally operate.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of powder classification and flow field characteristic monitoring, and particularly relates to an intelligent monitoring and regulation method for strong-rotation airflow screen classification process. BACKGROUND

[0002] With the development of high-tech industries, powder has been widely used in medicine, chemical industry, building materials and other industries, and the demand is increasing year by year, and the quality of powder is also improving. The powder has the characteristics of small volume, narrow particle size distribution, uniform quality, large specific surface area, low melting point, good activity, and is developing towards ultra-fine and large-scale. In the process of classifying micro-fine powder by using traditional air classifier, after the powder enters the classification area with the airflow, the separation of coarse and fine particles is carried out by relying on inertial force, centrifugal force, gravity and other actions, but the classification accuracy and effect cannot meet the requirements of the target product, and the classification phenomenon is serious, the particle size range of the product is widened, and the overall particle size index is greatly reduced. If the qualified product is not separated from the coarse particles and the unqualified product is not crushed in time, it will cause waste of resources and excessive crushing of part of the product, which cannot meet the needs of the development of fine industry.

[0003] The existing technology has the following problems and defects: the existing airflow screen classifier mainly optimizes key components, and lacks measurement of the flow field characteristics in the classification chamber; but the flow field characteristics affect the classification efficiency and accuracy of the powder, so it is necessary to stabilize the airflow in the classification chamber within a reasonable range, and to understand the flow field state at different positions in the classification chamber in real time, a plurality of sensors need to be arranged outside the classification chamber to collect air pressure, flow rate and the like in real time, which is high in cost and prone to errors caused by manual operation. The disturbance structure and screen arrangement are key components that affect the flow field characteristics, and workers only adjust them according to experience without sufficient theoretical and technical support. Therefore, it is necessary to provide a method for intelligently monitoring and regulating the flow field characteristics in real time. SUMMARY

[0004] In view of the above analysis, the present application aims to provide an intelligent monitoring and regulation method for strong-rotation airflow screen classification process, to solve the problem that the classification accuracy and effect cannot meet the requirements of the target product in the process of classifying micro-fine powder by using traditional air classifier.

[0005] The present application provides an intelligent monitoring and regulation method for strong-rotation airflow screen classification process, comprising: establishing a digital twin visualization model of a strong-rotation airflow screen classifier, simulating two-phase flow through simulation calculation to obtain flow field data of different positions in the classification chamber under different process parameters, and then determining key process parameters affecting the airflow screen flow field characteristics;

[0006] The flow field data of the strong cyclone airflow screen classifier at different positions in the classification chamber and under different key process parameters are measured by a measuring device as measured data; the measured data and the results of simulation calculation are compared, and based on the comparison results, the flow field model is verified and corrected;

[0007] The flow field data at different positions in the classification chamber and under different key process parameters are obtained by simulation calculation based on the corrected flow field model as simulation data;

[0008] Based on machine learning, a flow field characteristic prediction model is established, and the flow field characteristics are trained by measured data and simulation data to obtain a trained flow field characteristic prediction model;

[0009] Based on the trained flow field characteristic prediction model, the flow field characteristics are predicted, and when the predicted flow field characteristics are abnormal, the flow field characteristics are adjusted by adjusting the key process parameter values to make the strong cyclone airflow screen classifier operate normally.

[0010] Further, the measuring device comprises a laser Doppler velocimeter and a wind pressure measuring device, the laser Doppler velocimeter adopts non-contact measurement, the light receiving window of the laser Doppler velocimeter is placed in the measurement port outside the classification chamber, and the flow rate and flow of the gas at different positions in the classification chamber are measured; the receiving end of the wind pressure measuring device extends into the strong cyclone airflow screen classifier, and the airflow pressure in the classification chamber is measured.

[0011] Further, the flow field characteristic prediction model is a CNN-LSTM-Attention model, which comprises an input layer, a CNN convolution layer, a pooling layer, an LSTM layer, an attention layer, a Dropout layer and a Denselayer fully connected layer in sequence; the CNN convolution layer comprises three parallel convolution layers with different expansion rates.

[0012] Further, the flow rate, flow and airflow pressure of the gas at different positions in the classification chamber are normalized and then subjected to sliding window processing, and then input into the CNN-LSTM-Attention model, and the sliding window size is set to a first threshold value.

[0013] Further, the three parallel convolution layers with different expansion rates each have one filter with a size of 3, the filter step threshold value and the expansion rates of the three convolution layers are set, the sigmoid is used as the activation function, and the padding mode is same.

[0014] Further, the strong cyclone airflow screen classifier comprises a spiral conveying device, a disturbance structure, a screen cage, a driving motor, a coarse discharge port and a fine discharge port; the left and right ends of the screen cage are a discharge end and a feeding end respectively, and the feeding end receives the powder conveyed by the spiral conveying device; the spiral conveying device and the disturbance structure are arranged on the same rotating shaft, the screen cage surrounds the disturbance structure, and the driving motor drives the rotating shaft to rotate, thereby driving the spiral conveying device and the disturbance structure to rotate; the powder is conveyed into the screen cage by the spiral conveying device, and the coarse particles are centrifuged along the screen wall to the discharge end and then enter the coarse discharge port; and the fine particles enter the fine discharge port through the side wall of the screen cage.

[0015] Further, the strong cyclone airflow screen classifier further comprises a belt device, the belt device is connected to the driving motor screw rod and the rotating shaft, and the driving motor drives the belt device to rotate, thereby driving the rotating shaft to rotate.

[0016] Further, the disturbance mechanism comprises three blades uniformly arranged on the rotating shaft, and each blade is provided with a tooth on the side facing the screen cage; and an air classification wheel is further arranged on the disturbance mechanism, and the air classification wheel is arranged on the rotating shaft and located on the side away from the spiral conveying device.

[0017] Further, the air pressure measuring device comprises a differential pressure sensor, an air pipe and a data acquisition card, the differential pressure sensor has a pagoda-shaped double nozzle; the input end of the air pipe is deeply arranged in the classification chamber, the output end of the air pipe is connected to the long nozzle of the pagoda-shaped double nozzle of the differential pressure sensor, the output end of the differential pressure sensor is connected to the data acquisition card, and the other end of the acquisition card is connected to a computer end to realize signal transmission.

[0018] Further, the laser Doppler velocity measuring device comprises a laser, an optical system and a signal processing device; the laser adopts a helium-neon laser; the optical system comprises, in sequence along the laser incidence direction, a collimating mirror, a grating, a first diaphragm, a full reflection mirror, a converging lens, a collecting light lens, a second diaphragm, a collecting light lens and a photodiode detector; and the signal processing device comprises an amplifying circuit, a filtering circuit and a signal acquisition circuit.

[0019] Compared with the prior art, the present application can achieve at least one of the following beneficial effects:

[0020] 1、The present application can facilitate numerical simulation and analysis calculation in later stage through three-dimensional modeling and mesh division of the airflow screen classifier, can provide theoretical support for analysis verification and optimization of the flow field of the strong cyclone airflow screen classifier, and is convenient for popularization and use of the air classifier.

[0021] 2. The laser Doppler velocimetry technology used is a non-contact measurement method, which reduces interference with the fluid; the emitted signal can be connected to a photodiode receiver and signal processing device for real-time monitoring; it has good directional sensitivity, can measure the velocity component, and realize multi-dimensional measurement; the measurement accuracy is as high as 0.025%.

[0022] 3. The established CNN-LSTM-Attention model can achieve intelligent prediction of flow field characteristics under different locations and process parameters.

[0023] 4. The strong cyclone airflow sieve classifier and the digital twin visualization model with predictive function can realize multi-parameter coordinated control and optimization. The design of the central disturbance structure and screen cage in the classification chamber can ensure that the airflow fully disperses and classifies the powder material, thereby improving the working efficiency and classification effect of the classifier.

[0024] 5. This method can make holistic and systematic predictions of the flow field change patterns under different states of the classifier. Through physical models, numerical models and data-driven approaches, a final prediction model is established to meet the practical needs of real-time monitoring and control of the flow field characteristics in the classifier.

[0025] In this invention, the above-described technical solutions can be combined with each other to achieve more preferred combinations. Other features and advantages of this invention will be set forth in the following description, and some advantages may become apparent from the description or be learned by practicing the invention. The objects and other advantages of this invention can be realized and obtained from the description and drawings, which are particularly pointed out. Attached Figure Description

[0026] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts.

[0027] Figure 1 A schematic diagram of a method for intelligent monitoring and control of a strong swirling airflow sieving and classification process;

[0028] Figure 2 This is a schematic diagram of an airflow sieving system in an intelligent monitoring and control method for a strong swirling airflow sieving process.

[0029] Figure 3 This is a schematic diagram of the CNN-LSTM-Attention model structure in an intelligent monitoring and control method for a strong swirling airflow sieving and classification process.

[0030] Figure label:

[0031] 1- Fan;

[0032] 2-Dust collector;

[0033] 3 - data acquisition card;

[0034] 4 - computer;

[0035] 5 - differential pressure sensor;

[0036] 6 - pitot tube;

[0037] 7 - coarse discharge;

[0038] 8 - fine discharge;

[0039] 9 - drive motor;

[0040] 10 - disturbance structure;

[0041] 11 - screen cage;

[0042] 12 - screw conveyor;

[0043] 13 - belt device;

[0044] 14 - measurement port;

[0045] 15 - Doppler velocity measurement system. DETAILED DESCRIPTION

[0046] The preferred embodiments of the present application will be described in detail below with reference to the drawings, which constitute a part of this application, and are used to explain the principles of the present application together with the embodiments of the present application, but are not used to limit the scope of the present application.

[0047] One specific embodiment of the present application discloses an intelligent monitoring and control method for strong vortex airflow screen classification process, as shown in Figure 1 , which includes steps S1-S5.

[0048] S1, a digital twin visualization model of the strong vortex airflow screen classifier is established, two-phase flow simulation is performed through simulation calculation, flow field data at different positions of the classification chamber under different process parameters are obtained, and then key process parameters affecting the flow field characteristics of the airflow screen are determined. Specifically, it includes steps S11-S16.

[0049] The schematic diagram of the airflow screen classification system is as shown in Figure 2 .

[0050] The strong cyclone airflow screen classifier comprises a screw conveying device 12, a disturbance structure 10, a screen cage 11, a driving motor 9, a coarse discharge port 7, and a fine discharge port 8. The left and right ends of the screen cage 11 are respectively a discharge end and a feeding end. The feeding end receives the powder conveyed by the screw conveying device 12. The screw conveying device 12 and the disturbance structure 10 are arranged on the same rotating shaft. The screen cage 11 surrounds the disturbance structure 10. The driving motor 9 drives the rotating shaft to rotate, thereby driving the screw conveying device 12 and the disturbance structure 10 to rotate. The powder is conveyed into the screen cage 11 by the screw conveying device 12. Under the disturbance of the disturbance structure 10, the coarse particles do centrifugal motion along the screen wall surface of the screen cage 11 to the discharge end and enter the coarse discharge port 7. The fine particles enter the fine discharge port 8 through the side wall of the screen cage 11.

[0051] The strong cyclone airflow screen classifier further comprises a belt device 13. The belt device 13 is connected to the screw rod of the driving motor 9 and the rotating shaft. The driving motor 9 drives the belt device 13 to rotate, thereby driving the rotating shaft to rotate.

[0052] Specifically, the multifunctional frequency converter controls the driving motor 9 to drive the belt device 13 to rotate, thereby driving the rotating shaft to rotate.

[0053] The disturbance mechanism comprises three blades uniformly arranged on the rotating shaft. Each blade is provided with a tooth towards one side of the screen cage 11. An air classification wheel is further arranged on the disturbance mechanism. The air classification wheel is arranged on the rotating shaft and located on the side away from the screw conveying device.

[0054] Specifically, the three blades are arranged in a curved shape on the side close to the screw conveying device 12. Each blade is fixed on the rotating shaft by screws. The part between the two screws of each blade is arranged in a special tooth shape towards one side of the screen cage 11.

[0055] The strong cyclone airflow screen classifier is provided with a circular truncated cone-shaped feeding port located directly above the screw conveying device 12 for receiving the powder. After the powder material is uniformly fed from the circular truncated cone-shaped feeding port, it is pushed by the screw conveying device 12 to the classification chamber. Due to the strong cyclone airflow generated by the curved blades and the special tooth shape, the powder material is rapidly dispersed and mixed with the airflow to do jet centrifugal motion. The coarse and fine powder particles rely on their different inertial forces. The coarse particles do centrifugal motion along the screen wall surface of the cylindrical screen cage 11 to the discharge end under the propulsion of the cyclone. After passing through the gap between the air classification wheel and the screen cage 11 port, the coarse particles freely settle into the coarse powder product collector. The fine particles are jetted through the screen under the action of high-pressure air cyclone and pass through the interference settlement into the fine powder product collector, thereby realizing fine classification of coarse and fine powder.

[0056] The cylindrical screen cage 11 is composed of a 200-mesh nylon screen and a cylindrical rigid framework.

[0057] The strong cyclone airflow screen classifier is provided with a dust removal port.

[0058] S11, determine the numerical simulation of the turbulence model.

[0059] The inside of the classifier is gas-solid two-phase flow, and the airflow is used as a continuous phase, and the flow process thereof is simulated by using a Reynolds stress model, a mathematical model is established, and the control equations used include a continuity equation, a Navier-Stokes equation and a K-epsilon model.

[0060] Continuity equation:

[0061]

[0062] Wherein, p is density, t is time, u, v, w are velocity vector components in x, y, z directions;

[0063] Navier-Stokes equation:

[0064]

[0065]

[0066]

[0067] S u , S v , S w is the external force received by unit volume of fluid, p is the external force, and constant mu is dynamic viscosity;

[0068] K-epsilon model:

[0069]

[0070]

[0071] Wherein t is time, p is density, u i is velocity, k is turbulent kinetic energy, and epsilon is dissipation rate, G k is the generation term of turbulent kinetic energy k due to average velocity gradient, G b is the turbulent kinetic energy due to buoyancy, mu t is turbulent dynamic viscosity, mu is laminar dynamic viscosity, Y m is the influence of fluctuation expansion on turbulent dissipation rate, C 1ε , C 2ε , sigma k , sigma ε are empirical constants, x i and x j represent direction vectors in the coordinate system.

[0072] S12, a 3D model of the strong cyclone airflow screen is established.

[0073] A 3D model of the airflow screen classifier is established by using SolidWorks, the model of the strong cyclone airflow screen is simplified, the established model is consistent with the actual size of the strong cyclone airflow screen classifier, saved as a.x_t file, dragged into the fluid flow fluent module in Workbench and imported into the established model, edited by using DesignModeler, clicked Generate to import the model, and clicked Tools-fill to fill the fluid in the classifier.

[0074] S13, the 3D model generated by SolidWorks is meshed based on the key components of the pre-processing software ICEM.

[0075] The strong cyclone airflow screen classifier mainly consists of a feeding port, a spiral conveying device 12, a disturbance structure 10, a cylindrical screen cage 11, a driving motor 9, a dust removal port, a coarse discharge port 7 and a fine discharge port 8. The key components of the strong cyclone airflow screen classifier that affect the flow field characteristics are the screen cage 11 and the disturbance structure 10.

[0076] Specifically, the key components of the airflow screen are meshed, and the mesh independence test is carried out. Due to the complex geometry of the model, the number of meshes and the calculation volume are large, and the calculation domain is divided by hexahedral mesh. The near-wall mesh is encrypted, the calculation results of different mesh numbers are compared, and the deviation is less than 5%, indicating that the result is reliable. The strong cyclone airflow screen geometry model established by using ICEM software is meshed, the starting surface, mesh layer and expansion rate are selected, the internal mesh of the fluid is refined, the corresponding surfaces are named as inlet, outlet and shell, and the mesh quality is evaluated by the standard determinant value in the mesh pre-quality function. The key components are specially meshed to improve the calculation accuracy.

[0077] S14, the mesh structure generated by ICEM is simulated by using the turbulent flow model based on the Fluent software package.

[0078] The material properties of micro-fine powder, the boundary conditions of simulation calculation are set, and the gas-solid two-phase flow is simulated by using Fluent. The mesh file of the strong cyclone airflow screen obtained in S13 is imported into the fluid mechanics software Fluent for two-phase flow simulation, and the material properties, simulation calculation conditions and boundary conditions are set. When the residual error of each variable reaches 10 -5 , it is considered to be converged, and the design time step is 10 -2s, and then the simulation calculation is started. The solving condition is that the pressure-velocity coupling adopts the PISO algorithm, and other physical quantities adopt the second-order upwind format. The boundary condition is that the initial feeding gas velocity of the inlet is 0 m / s, the hydraulic diameter, length scale and turbulence intensity are set to 150 mm, 650 mm and 5% respectively, the outlet flow is completely formed in the outlet part, the side wall adopts the no-slip boundary condition, the flow velocity inlet at the air inlet is taken as the boundary condition, it is assumed that the disturbance wind speed is uniformly distributed at the inlet, the direction is perpendicular to the inlet boundary, the boundary condition of the inlet is the fully developed pipe flow, and the flow velocity of the solid phase at the inlet is 1 m / s. After the initialization is completed, the iteration step number is set, and after the iteration calculation is completed, the post-processing is carried out. A series of gas flow velocity, flow and gas flow pressure distribution cloud maps and variation trend graphs at different positions in the radial direction and the axial direction under different process parameters are obtained.

[0079] S15, post-processing based on CFD.

[0080] After the convergence calculation of the solving is completed, the calculation result is imported into the post-processing tool CFD-Post to obtain the flow field distribution characteristics of the gas flow screen at different positions in the radial direction and the axial direction. In the results interface, click streamline, set the corresponding streamline details, select type as 3D Streamline, Domains as all domains, Start From as inlet, Sampling as EquallySpaced, Variable as Velocity, Direction is selected by yourself, click Apply to generate the streamline graph; draw the velocity cloud map, create a plane through surface-plane, generate a local plane in the area of the plane in the graphical interface, select the corresponding physical quantity to generate the corresponding cloud map; double-click Contours, select Pressure and StaticPressure in Contours of, select the corresponding inlet, outlet and surface in Surface, and click display to create the pressure cloud map of different cross sections; in the above created cross section, select XYPlot and set the corresponding parameters to obtain the average flow of the cross section. Select different cross sections to obtain the gas flow velocity, flow and gas flow pressure and flow field distribution in the radial direction and the axial direction in the classifier under different process parameters.

[0081] S16, based on the post-processing result, determine the key process parameters affecting the flow field characteristics.

[0082] According to the simulated gas flow velocity, flow and gas flow pressure and flow field distribution at different positions in the radial direction and the axial direction in the classifier under different process parameters, the key process parameters of the strong vortex gas flow screen are the feeding speed, the inclination angle of the classifier, the disturbance structure rotating speed and the screen mesh diameter.

[0083] It can be understood that by observing the flow field distribution under different process parameters, it can be found that the feed speed, the inclination angle of the classifier, the rotation speed of the disturbance structure and the screen aperture have a greater impact on the flow field distribution of the strong vortex airflow screen classifier, so the key process parameters are determined as the feed speed, the inclination angle of the classifier, the rotation speed of the disturbance structure and the screen aperture.

[0084] S2, the flow field data of the strong vortex airflow screen classifier is measured by the measuring equipment to obtain the flow field data at different positions and under different key process parameters in the classification chamber as measured data; the measured data and the simulation calculation results are compared, and based on the comparison results, the flow field model is verified and corrected.

[0085] The flow field data includes the flow rate, flow volume and airflow pressure of the gas.

[0086] The airflow screen classification system further comprises a measuring equipment, a fan 1 and a pipeline. The fan 1 provides the airflow required for powder classification for the strong vortex airflow screen classifier through the pipeline.

[0087] The measuring equipment comprises a laser Doppler velocimeter and a wind pressure measuring device. The laser Doppler velocimeter adopts non-contact measurement, and the light receiving window of the laser Doppler velocimeter is arranged at the measuring port 14 outside the classification chamber to measure the flow rate and flow volume of the gas at different positions in the classification chamber. The receiving end of the wind pressure measuring device extends into the strong vortex airflow screen classifier to measure the airflow pressure in the classification chamber.

[0088] The laser Doppler velocimeter comprises a laser, an optical system and a signal processing device. The laser adopts a helium-neon laser. The optical system comprises a collimating mirror, a grating, a first diaphragm, a full mirror, a converging lens, a collecting lens, a second diaphragm and a photodiode detector arranged in sequence along the laser incidence direction. The signal processing device comprises an amplifying circuit, a filtering circuit and a signal acquisition circuit.

[0089] Specifically, the laser Doppler velocimeter system 15 adopts a double-beam-double-scattering mode of a heterodyne light path mode. The helium-neon laser emits ultrasonic waves of a certain frequency, which are focused in the transparent classification chamber by the converging lens. Due to the existence of the Doppler effect, when the gas in the classification chamber flows, the frequency of the reflected back wave changes, the airflow volume in the classification chamber is controlled, and the gas particles produce a Doppler frequency shift and scattering. The photodetector is placed at the focal point of the collecting light path to receive the scattered light. Through the square law effect of the photodetector, the Doppler frequency is obtained. The recovered frequency is processed by the photodiode detector, amplification and filtering circuit, and uploaded to the computer to collect the flow rate and flow volume signals of the gas at different positions in the classification chamber, which are processed by software.

[0090] A plurality of measuring ports 14 are arranged on the top of the air flow classifier in the direction from the inlet end to the outlet end for measuring the flow velocity, flow rate and air flow pressure of the gas at different positions of the strong vortex air flow classifier, so as to more comprehensively obtain the complete flow field variation characteristics from the feed port to the discharge port in the entire classification chamber. A laser Doppler velocimeter is used to measure the flow velocity and flow cross-sectional velocity distribution of the gas at different positions in the classification chamber to determine the flow velocity and flow rate of the gas. A helium-neon laser with a wavelength of 632.8 nm, a power of 5 mW, a grating constant of 50 μm, a converging lens focal length of 100 mm, a collecting light lens focal length of 50 mm and a silicon photodiode detector as optical components are used. The installation is as follows: the guide rail is placed in the form of a T-shaped letter, the optical bench is adjusted to make each component parallel to each other, the signal processing part is placed on another guide rail; the laser is turned on, the laser beam is adjusted to be parallel to the long guide rail in the T-shaped guide rail; a collimating mirror is placed behind the laser to achieve laser collimation; the grating is placed 200 mm away from the converging lens, and the converging lens is also placed 200 mm away from the fluid; a diaphragm is placed behind the grating to filter out other diffraction orders; a full mirror is placed in front of the full mirror to reflect the scattered light of the particles; a stable and good absorbing shield is placed in front of the full mirror to block the zero-order light; the converging lens is placed behind the full mirror, the scattered light of the particles is focused to the photodiode detector through the collecting light lens, the photodiode detector is moved forward and backward, and the scattered light spot is located at the focal point of the receiving lens, so that the photodiode detector can receive the most Doppler signals; the grating is adjusted to form two parallel light beams; the height of the converging lens is adjusted so that the light passes through the center of the converging lens; the two light beams are focused at the measurement point, the laser generates interference fringes at the measurement point, and a lens is used to present the image far away. The two light spots are coincided together far away, and the interference fringes can be observed far away; the variable diaphragm is placed about 5 cm away from the measurement point and fully opened, the two light beams are blocked by the diaphragm after passing through the measurement area, the collecting light path is close to the diaphragm, and the scattered light is focused to the photodiode detector; the photodiode detector is moved forward and backward, and the scattered light spot is located at the focal point of the receiving lens, so that the photodiode detector can receive the most Doppler signals, and the optical path adjustment is completed at this time; during the experiment, the light is turned off to prevent interference with the accuracy of the measurement results. The velocity of the plurality of measuring ports 14 is measured, the Doppler signals generated by the particles are collected, FFT is realized by using Origin / MATLAB, and the Doppler frequency shift of the system is quickly obtained. Due to the difference between the refractive index of the fluid and the refractive index of the vacuum, the speed, angle and wavelength of the laser change obviously, the Doppler frequency shift is obtained according to the sampling period, and then the gas flow velocity is obtained. wherein f is the Doppler frequency shift, v is the particle velocity, λ is the wavelength, α is the angle between the two laser beams, and β is the angle between the direction of the parallel beam and the direction of motion, the Doppler frequency shift is obtained according to the sampling period, and then the gas flow velocity is obtained.

[0091] Based on the single factor method, the value of the key process parameter is changed, and the flow rate of the gas at different positions under different process parameters is measured again. The single factor method means that only the value of one key process parameter is changed, and the values of other key process parameters remain unchanged.

[0092] The flow rate is the amount of fluid flowing through the cross section of the classification chamber per unit time. The measurement point is selected in the radial cross section of the classification chamber, with the center of the cross section as the coordinate origin of the measurement point. The appropriate measurement point spacing is selected, and the remaining measurement points are positioned and measured. Based on the flow rate distribution, the flow rate is converted to flow rate by integration, and the gas flow rate at different positions in the classification chamber is obtained.

[0093] The wind pressure measuring device includes a differential pressure sensor 5, a gas pipe, and a data acquisition card 3. The differential pressure sensor 5 has a pagoda-shaped double nozzle. The input end of the gas pipe is deeply inserted into the classification chamber, and the output end of the gas pipe is connected to the long nozzle of the pagoda-shaped double nozzle of the differential pressure sensor 5. The output end of the differential pressure sensor 5 is connected to the data acquisition card 3, and the other end of the data acquisition card 3 is connected to the computer 4 to realize signal transmission.

[0094] The measurement range of the differential pressure sensor 5 is ±100 Pa, the output is 0-5 V, and the accuracy level is 1% FS. The differential pressure sensor 5 is calibrated before measurement. The other end of the data acquisition card 3 of the wind pressure measuring device is connected to the computer 4 to realize signal transmission. The input end of the gas pipe of the wind pressure measuring device is deeply inserted into the classification chamber by 2 cm, and the collected signals are analyzed and processed by Lab VIEW software. The gas pipe is inserted into different positions of the classification chamber to collect air pressure, and the trend of air pressure signal change is observed. The display value of the sensor during stable operation of the strong vortex airflow screen is recorded, and signal collection is completed after classification is completed.

[0095] The measuring device also includes a Pitot tube 6. The Pitot tube 6 is located on the pipeline of the airflow screen classification system for measuring the static pressure delivered by the fan 1.

[0096] Based on the single factor method, the process parameters are changed, and the airflow pressure at different positions of the strong vortex airflow screen is repeatedly measured.

[0097] The measurement results are compared and analyzed with the simulation results based on CFD, and the model is verified.

[0098] Specifically, the flow rate, flow rate, and airflow pressure of the gas at different positions collected by the laser Doppler velocimeter and the differential pressure sensor 5 are analyzed, and the results are compared with the post-processing results in Fluent. The distribution characteristics of the flow rate, flow rate, and airflow pressure of the gas at different positions in the radial and axial directions of the strong vortex airflow screen are observed. Based on the test results, the flow field model in Fluent is verified and corrected.

[0099] If the verification conditions do not meet the powder grading requirements, the parameters of the simulation model are optimized.

[0100] S3, simulation calculation is performed based on the corrected flow field model to obtain flow field data at different positions and different key process parameters in the classification chamber as simulation data.

[0101] Due to the limited number of actual measurement devices, the flow rate, flow and airflow pressure of the gas at different positions in the classification chamber at the same time cannot be measured, and the flow rate, flow and airflow pressure of the gas at different positions in the airflow classifier need to be calculated by simulation, and then the calculation results are input into the prediction model.

[0102] S4, based on machine learning, a flow field characteristic prediction model is established, and the flow field characteristics are trained by measured data and simulation data to obtain a trained flow field characteristic prediction model.

[0103] The flow field characteristic prediction model is a CNN-LSTM-Attention model, which sequentially includes an input layer, a CNN convolution layer, a pooling layer, an LSTM layer, an attention layer, a Dropout layer and a Dense layer full connection layer; the CNN convolution layer includes three parallel convolution layers with different expansion rates.

[0104] The structure diagram of the CNN-LSTM-Attention model is shown in Figure 3 .

[0105] The Dense layer full connection layer is used, and each node of the full connection layer is connected to all nodes of the previous layer, which is used to integrate the flow field characteristics extracted in the front, output the flow field prediction of CFD, and add a Dropout layer in the attention mechanism before the Dense layer to prevent overfitting and improve the generalization ability of the model. Compared with SVM and BP, the CNN-LSTM-Attention model can extract local features in the input data, and realize operation translation invariance through convolution operation, at the same time, LSTM can capture time sequence relationship, has the ability of memory and forgetting, and attention mechanism can adaptively allocate weights according to the importance of input data, which is suitable for processing data with time sequence and spatial structure. Therefore, the CNN-LSTM-Attention model is established to predict the flow field characteristics.

[0106] The three parallel convolution layers with different expansion rates each have one filter with a size of 3, the filter step threshold and the expansion rate of the three convolution layers are set, the sigmoid function is used as the activation function, and the padding mode is same.

[0107] The flow rate, flow and air flow pressure of the gas at different positions of the classification chamber are normalized and then subjected to sliding window processing, and then input into the CNN-LSTM-Attention model, and the sliding window size is set to the first threshold value.

[0108] The input data is set a timing sliding window, the sliding window method can increase the number of input samples, and the sliding window size of an embodiment of the present application is set to 10, that is, 10 sampling points on the time sequence are taken as a group of input data, which can quickly process large-scale data flow, and make the model more suitable for dynamic data analysis, which is simple and efficient.

[0109] The input data of the training set is standardized to map the value range of the data to the interval [0, 1], so as to eliminate the dimensional influence between different indicators and improve the performance of the prediction model, accelerate the convergence speed of the model, and improve the accuracy of the model to a certain extent. The input data of the test set is also standardized to accelerate the convergence speed of the model and improve the accuracy of the model to a certain extent. The training set enables the model to extract useful patterns and rules from the data and predict new data, and has a certain representativeness. The test set has a certain independence, ensuring that the evaluation of the model is objective and reliable. The commonly used learning rate is selected for the model, and the mean values of the root mean square error (RMSE), mean absolute error (MAE) and training time of 50 rounds under each learning rate are compared. The appropriate CNN convolution kernel is selected until the RMSE and MAE reach the minimum value.

[0110] The model uses StandardScaler for standardization, converts the data to a standard normal distribution with a mean of 0 and a variance of 1, and is converted by the following formula: Where z is the converted data, x is the original data, x includes the measured data and simulated data of the flow rate, flow and air flow pressure of the gas at different positions. u is the mean of a certain original data, and s is the standard deviation of a certain original data.

[0111] The real-time measured flow field data set is divided into training set and test set, 80% for training set and 20% for test set. For the training set and test set, StandardScaler object is created and fit_transform and transform methods are used for processing respectively. The prediction training of flow field data at different positions is carried out using early stopping method and three parallel convolution layers with different expansion rates, the loss function is monitored, the prediction results are compared with the test set results, and the parameter optimization of CNN model is carried out. For the time series historical data of the measured flow field characteristics, a cycle time network LSTM model is established, and the historical data is analyzed to realize the time series prediction of the flow field characteristic monitoring value, and the model parameters are optimized according to the model effect, and then the cycle time network model corresponding to the flow field characteristics at different positions of the grading chamber is established to realize the advanced intelligent prediction of all measured values, and the advanced prediction values are recorded on the parameter module corresponding to the Fluent model, and the established three-dimensional model is optimized.

[0112] Again, for the flow field model simulation calculation data at different positions, the model is generalized to form a spatial prediction model to realize the prediction of the flow rate, flow rate and airflow pressure distribution of the gas at all positions.

[0113] S5, based on the trained flow field characteristic prediction model, the flow field characteristics are predicted, when the predicted flow field characteristics are abnormal, the flow field characteristics are adjusted by adjusting the key process parameter values to make the strong vortex airflow classifier work normally.

[0114] By predicting the flow field characteristics, the working condition of the classifier can be perceived in advance, when the flow rate, flow rate and airflow pressure of the gas change and reach the working failure threshold, the key process parameter values of the classifier are adjusted in advance, such as the flow rate, flow rate and airflow pressure of the gas are reduced to adjust the feeding speed, the inclination angle of the classifier and the rotation speed of the disturbance structure 10, so that the flow rate, flow rate and airflow pressure of the gas fed by the reverse classifier remain normal state.

[0115] Compared with the prior art, the embodiment can provide theoretical support for analysis and verification and optimization of the flow field of the strong cyclone airflow screening classifier by three-dimensional modeling and meshing of the strong cyclone airflow screening classifier, and facilitate popularization and use of the air classifier. The laser Doppler velocimetry technology adopted in the embodiment is a non-contact measurement, which reduces the interference to the fluid; the emitted signal can be connected with a photodiode receiver and a signal processing device for use, which is beneficial to real-time monitoring; has good directional sensitivity, can measure the components of the speed, and realizes multidimensional measurement; and the measurement accuracy is as high as 0.025%. The CNN-LSTM-Attention model established in the embodiment can realize intelligent prediction of the flow field characteristics under different positions and different process parameters. The grading system and the digital twin visualization model can realize multi-parameter collaborative regulation and optimization, and the design of the center disturbance structure 10 and the screen cage 11 in the classification chamber can ensure that the airflow fully disperses and classifies the powder material, thereby improving the working efficiency and classification effect of the classifier. The prediction method provided in the embodiment can make an overall and systematic prediction of the flow field variation law of the classifier under different states, and through a physical model, a numerical model and data driving, an ultimate prediction model is established, thereby meeting the real-time monitoring and regulation of the flow field characteristics in the classification chamber.

[0116] Those skilled in the art can understand that all or part of the processes of the above-mentioned embodiment methods can be completed by a computer program instructing related hardware, and the program can be stored in a computer readable storage medium. The computer readable storage medium is a disk, an optical disk, a read-only memory or a random access memory, etc.

[0117] The above is only a preferred specific embodiment of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical range disclosed in the present application, which should be covered within the protection scope of the present application.

Claims

1. A method for intelligent monitoring and control of a strong swirling airflow sieving and classification process, characterized in that, include: A digital twin visualization model of a strong swirling airflow classifier was established. Two-phase flow simulation was performed through simulation calculations to obtain flow field data at different locations in the classification chamber under different process parameters, thereby determining the key process parameters that affect the flow field characteristics of the airflow classifier. The flow field data of the high-intensity cyclone airflow classifier was measured using measuring equipment to obtain flow field data at different locations and under different key process parameters within the classifier chamber, which were then used as measured data. The measured data and simulation results were compared, and the flow field model was verified and corrected based on the comparison results. The measuring equipment included a laser Doppler velocimeter and a wind pressure measuring device. The laser Doppler velocimeter used non-contact measurement, with its optical receiving window placed at the measuring port outside the classifier chamber to measure the gas velocity and flow rate at different locations within the classifier chamber. The receiving end of the wind pressure measuring device extended into the high-intensity cyclone airflow classifier to measure the airflow pressure within the classifier chamber. Simulation data were obtained from different locations and key process parameters in the graded chamber based on the modified flow field model. Based on machine learning, a flow field characteristic prediction model is established. The flow field characteristics are trained using measured data and simulated data to obtain a well-trained flow field characteristic prediction model. The flow field characteristics are predicted based on the trained flow field characteristic prediction model. When the predicted flow field characteristics are abnormal, the flow field characteristics are regulated by adjusting the key process parameter values ​​so that the strong vortex airflow screening and classifying machine can operate normally.

2. The intelligent monitoring and control method according to claim 1, characterized in that, The flow field characteristic prediction model is a CNN-LSTM-Attention model, which sequentially includes an input layer, a CNN convolutional layer, a pooling layer, an LSTM layer, an attention layer, a dropout layer, and a dense fully connected layer; the CNN convolutional layer includes three parallel convolutional layers with different dilation rates.

3. The intelligent monitoring and control method according to claim 2, characterized in that, The gas velocity, flow rate, and air pressure at different locations in the grading chamber are normalized and then processed by sliding window, before being input into the CNN-LSTM-Attention model. The sliding window size is set to the first threshold.

4. The intelligent monitoring and control method according to claim 2, characterized in that, Three parallel convolutional layers with different dilation rates each have a filter of size 3. The filter stride threshold and the dilation rates of the three convolutional layers are set, with sigmoid as the activation function and same padding.

5. The intelligent monitoring and control method according to claim 1, characterized in that, The high-intensity cyclone airflow sieving and classifying machine includes a screw conveyor, a disturbance structure, a screen cage, a drive motor, a coarse discharge port, and a fine discharge port. The left and right ends of the screen cage are the discharge end and the feed end, respectively. The feed end receives the powder conveyed by the screw conveyor. The screw conveyor and the disturbance structure are mounted on the same rotating shaft. The screen cage surrounds the disturbance structure. The drive motor drives the rotating shaft to rotate, thereby driving the screw conveyor and the disturbance structure to rotate. The powder is conveyed into the screen cage through the screw conveyor. Under the disturbance of the disturbance structure, coarse particles undergo centrifugal motion along the screen wall and enter the coarse discharge port at the discharge end. Fine particles enter the fine discharge port through the side wall of the screen cage.

6. The intelligent monitoring and control method according to claim 5, characterized in that, The high-intensity cyclone airflow screening and grading machine also includes a belt device, which connects the drive motor lead screw and the rotating shaft. The drive motor drives the belt device to rotate, which in turn drives the rotating shaft to rotate.

7. The intelligent monitoring and control method according to claim 6, characterized in that, The disturbance structure includes three blades evenly arranged on the rotating shaft, each blade having teeth on the side facing the screen cage; the disturbance structure also includes an air classifier wheel, which is located on the rotating shaft and on the side of the blades away from the spiral conveyor.

8. The intelligent monitoring and control method according to claim 1, characterized in that, The wind pressure measurement device includes a differential pressure sensor, a tubing, and a data acquisition card. The differential pressure sensor has a pagoda-shaped double nozzle. The inlet of the tubing extends into the grading chamber, and the outlet of the tubing is connected to the long nozzle of the pagoda-shaped double nozzle of the differential pressure sensor. The outlet of the differential pressure sensor is connected to the data acquisition card, and the other end of the acquisition card is connected to a computer to achieve signal transmission.

9. The intelligent monitoring and control method according to claim 1, characterized in that, The laser Doppler velocimetry device includes a laser, an optical system, and a signal processing device; the laser is a helium-neon laser; the optical system includes a collimating lens, a grating, a first aperture, a total reflection mirror, a converging lens, a collecting lens, a second aperture, a collecting lens, and a photodiode detector arranged sequentially along the laser incident direction; the signal processing device includes an amplification circuit, a filtering circuit, and a signal acquisition circuit.

Images