Intelligent fluidized bed jet mill

Abstract

The invention discloses an intelligent fluidized bed jet mill, and relates to the technical field of material crushing and grading equipment. The equipment comprises a hardware part and a control module, wherein the hardware part consists of a crushing area, a feeding area and a grading area. Low-temperature and high-pressure nozzles uniformly distributed on an annular gas distribution disc are adopted in the crushing area; the grading area is provided with a flow dividing cylinder and a conical flow guide plate for separating ascending airflow and falling coarse particles; the control module collects particle size data of powder in real time, predicts future particle size distribution in combination with an LSTM network, dynamically optimizes the rotating speed of a grading impeller through a particle swarm algorithm, and achieves cooperative control over smashing efficiency, particle size precision and energy consumption with the effective output per unit energy consumption as the optimization target. Through low-temperature embrittlement, an anti-pollution structure and an intelligent algorithm, the problems that traditional equipment is high in energy consumption, extensive in particle size control and prone to contaminating materials are solved, and the device is suitable for fine smashing of high-purity or high-hardness materials.

Description

Technical Field

[0001] The present invention relates to the technical field of material crushing and classification equipment, and specifically to an intelligent fluidized bed air classifier mill. Background Art

[0002] An air classifier mill is a device that uses high-speed airflows to impact, collide with, and shear materials to achieve material crushing. Its working principle is to accelerate the materials to supersonic speeds through compressed air, causing the materials to repeatedly impact and collide within the crushing chamber to achieve the crushing effect. Subsequently, the crushed materials enter the classification chamber with the upward airflow. Through the action of the classification impeller, fine particles and coarse particles are separated. The fine particles are discharged with the airflow, while the coarse particles return to the crushing area for further crushing.

[0003] Traditional air classifier mill equipment has certain limitations in terms of crushing efficiency and classification accuracy. For example, the airflow distribution within the crushing area is uneven, resulting in an unsatisfactory material crushing effect. In addition, there are problems such as relying on manual adjustment of the classification impeller speed, high energy consumption, low particle size control accuracy, and insufficient efficiency. For example, the rotation speed of the classification impeller is fixed or only adjusted through simple feedback, unable to adapt to changes in material characteristics and complex working conditions, making it difficult to balance energy consumption and crushing effect. Existing equipment has limited capabilities for real-time monitoring and dynamic optimization of particle size distribution, and it is difficult to meet the requirements for high-precision and narrow-distribution powders in the high-end manufacturing field. Therefore, we provide an intelligent fluidized bed air classifier mill. Summary of the Invention

[0004] The purpose of the present invention is to provide an intelligent fluidized bed air classifier mill to solve the above problems.

[0005] The present invention can be achieved through the following technical solutions: An intelligent fluidized bed air classifier mill includes a hardware part and a control module. The hardware part is provided with a crushing area and a classification area. The crushing area includes a crushing cylinder body, on one side of which a screw conveyor is fixedly arranged. An annular air distribution plate is installed outside the crushing cylinder body, and a plurality of equally angularly distributed crushing nozzles are communicated with the inside thereof. The classification area is located above the crushing area, and a plurality of classification impellers driven by a variable-frequency motor are equally angularly installed in the classification cylinder body of the classification area. A classification chamber is coaxially arranged inside the classification cylinder body, and the classification impellers are rotatably installed between the classification chamber and the classification cylinder body;

[0006] The control module includes a data acquisition unit and an optimization control unit. The data acquisition unit is used to obtain the particle size distribution of the particle group in the classification area and downstream of the classification chamber in real time, and record the operating data of this device, including gas pressure, air temperature, feeding speed, rotation speed of the classification impeller, and working current of the frequency modulation motor;

[0007] Based on the above operating data and the particle size distribution of the particle swarm, an LSTM prediction network is built to predict the powder particle size distribution in a future period, and an optimization function with maximizing the time efficiency ratio as the optimization objective is constructed according to the prediction data. The particle swarm algorithm is used to calculate the optimal rotation speed sequence, and the rolling horizon control mode is used for optimization and update.

[0008] A further technical improvement of the present invention lies in that: the annular air distribution disk is connected to the high-pressure gas path, and the high-pressure gas path uses low-temperature helium or low-temperature nitrogen, and the temperature range is -20°C to -150°C.

[0009] A further technical improvement of the present invention lies in that: a flow dividing cylinder is provided in the connecting transition section between the crushing area and the classification area, and a conical guide plate is arranged at the top of the classification chamber for separating the rising air flow and the falling-back coarse particles.

[0010] A further technical improvement of the present invention lies in that: ceramic tiles are attached to the inner walls of both the crushing area and the classification area, and the crushing nozzles are also made of ceramic materials.

[0011] A further technical improvement of the present invention lies in that: before constructing the LSTM prediction network, the historical operation is preprocessed to construct a sequence of feature vectors with time series, and the feature vector is X(t) = [D in (t), D out (t), P(t), N(t), V(t)];

[0012] Among them, D in (t) = [d 10 , d 50 , d 90 T , representing the particle size distribution matrix of the particle swarm in the classification area at the corresponding moment; D out (t) represents the average particle size of the powder particle swarm collected at the rear end of the classification chamber at the corresponding moment; P(t), N(t), V(t) respectively represent the inlet pressure, the rotation speed of the classification impeller, and the feeding speed.

[0013] A further technical improvement of the present invention lies in that: the process of building the LSTM prediction network includes:

[0014] Constructing a sliding window to move on the sequence of feature vectors, the width of the window is set to W time steps, and the step size is set to time steps;

[0015] Taking the data within the sliding window as the input, and taking the data corresponding to the first M time steps in the future as the output [D in (t + 1), D in (t + 2),..., D in (t + M)] and [D out (t + 1), D out ​(t + 2),..., D out Use (t + M)] as the output; then divide the training set and the test set to perform batch training on the above LSTM prediction network to obtain a network model that meets the performance requirements.

[0016] 8. A further technical improvement of the present invention lies in: the optimization function is

[0017] where Q(t) represents the effective output that meets the target particle size neighborhood range; E(t) represents the comprehensive energy consumption; d 50 (t) represents the median particle size at the current moment, d s represents the target particle size; a, b, and c are weighting factors used to balance the priorities of output, energy consumption, and particle size deviation;

[0018] The constraint conditions include:

[0019] ① Particle size upper limit constraint: d 90 (t + i) ≤ d max + Δd safe d max represents the maximum particle size allowed in this comminution process, and Δd safe represents the safety margin;

[0020] ② Equipment safety constraint: I(t) ≤ I 额定 , indicating that the current of the variable-frequency motor does not exceed the rated current;

[0021] ③ Rotation speed range constraint: N min ≤ N(t) ≤ N max , that is, the rotation speed of the classification impeller is within the allowable range of design;

[0022] The decision variable is set as the rotation speed N(t) of the classification impeller, and the particle swarm optimization algorithm is used to solve the function to obtain the optimal rotation speed sequence N * (t).

[0023] A further technical improvement of the present invention lies in: the rolling horizon control mode is to select the first five rotation speed commands from the optimal rotation speed sequence to control the frequency modulation motor, and then move the sliding window to the next position based on the new operation data for re-optimization.

[0024] Compared with the prior art, the present invention has the following beneficial effects:

[0025] 1. In the present invention, ceramic tiles are attached to the inner walls of the comminution area and the classification area, and the nozzle is made of ceramic material to avoid metal contamination, especially suitable for high-purity fields such as medicine and food; at the same time, the design of the shunt cylinder and the conical deflector effectively isolates the rising air flow and the falling particles, reduces air flow disorder, effectively improves the comminution efficiency and classification efficiency of the powder, and also reduces the energy consumption.

[0026] 2. The present invention uses low-temperature helium / nitrogen at -20°C to -150°C to embrittle the material, improve crushing, and avoid denaturation of heat-sensitive materials.

[0027] 3. The present invention predicts the rotational speed of the classification impeller through an LSTM network and dynamically optimizes it using a particle swarm algorithm. Taking "effective output per unit energy consumption" as an index, it reduces energy consumption while ensuring the particle size accuracy of the product; and the entire control process is fully autonomously optimized and controlled without human intervention, achieving dynamic real-time optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] For the convenience of those skilled in the art to understand, the present invention will be further described below in conjunction with the accompanying drawings.

[0029] Figure 1 is a schematic diagram of the external structure of the whole machine of the present invention;

[0030] Figure 2 is a schematic diagram of a partial cross-section of the whole machine of the present invention;

[0031] Figure 3 is a schematic diagram of the intelligent control execution process of the control module of the present invention.

[0032] In the figure: 1. Crushing area; 2. Feeding area; 3. Classification area; 101. Crushing cylinder; 102. Annular gas distribution plate; 103. Crushing nozzle; 301. Classification cylinder; 302. Classification chamber; 303. Classification impeller; 304. Variable-frequency motor; 305. Shunt cylinder; 306. Conical deflector. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0033] To further elaborate on the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following will describe in detail the specific embodiments, structures, features, and effects of the present invention in conjunction with the accompanying drawings and preferred embodiments.

[0034] Please refer to Figures 1-3 As shown, an intelligent fluidized bed air classifier includes a hardware part and a control module; wherein, the hardware part includes a whole machine box, and the whole machine box is divided into a crushing area 1, a feeding area 2, and a classification area 3; the control module includes a data acquisition unit, a data storage unit, and an optimization control unit.

[0035] The crushing zone 1 includes a crushing cylinder body 101. An annular air distribution disc 102 is installed outside the crushing cylinder body 101. A plurality of crushing nozzles 103 are evenly distributed at equal angles on the inner side of the annular air distribution disc 102. Each crushing nozzle 103 penetrates and is fixed on the side wall of the crushing cylinder body 101, and the extension lines of the air outlet ends of all the crushing nozzles 103 intersect at a point on the central axis of the crushing cylinder body 101. One end far from the air outlet end is hermetically connected to the annular air distribution disc 102 by a threaded connection with the crushing cylinder body 101; the annular air distribution disc 102 is connected to a high-pressure gas path. The high-pressure gas path uses compressed low-temperature helium or low-temperature nitrogen, generally set at (-20°C to -150°C), so as to embrittle the material and achieve good crushing effect;

[0036] The feeding zone includes a screw conveyor fixedly connected to the flange on the side wall of the crushing cylinder body 101. The screw conveyor evenly and continuously feeds the material into the crushing zone;

[0037] The classification zone is arranged above the crushing zone and is connected to the classification zone. The classification zone includes a classification cylinder body 301. A classification chamber 302 is coaxially arranged at the central position inside the classification cylinder body 301. A plurality of classification impellers 303 are arranged at equal angles between the classification chamber 302 and the classification cylinder body 301. Both ends of the classification impeller 303 are rotatably connected to the classification chamber 302 and the classification cylinder body 301 respectively. A variable-frequency motor 304 is arranged on one side of each classification impeller 303 and drives the classification impeller 303 to rotate through the variable-frequency motor 304. The variable-frequency motor 304 is fixed on the outer side wall of the classification cylinder body 301.

[0038] During operation, the material to be crushed is evenly fed into the crushing zone through the screw conveyor, and is accelerated by the high-speed low-temperature air flow in the crushing nozzle 103, and impacts and collides repeatedly at the center of the crushing zone, so as to achieve the purpose of crushing; the crushed material enters the classification zone along with the upward air flow. The classification impeller 303 rotates under the power of the frequency modulation motor 304. The fine particles in the material enter the classification chamber 302 under the action of the centripetal force of the air flow and are collected after being discharged with the air flow. The coarse particles in the material are thrown away under the action of the centrifugal force of the classification impeller 303 and fall back to the crushing zone to continue crushing.

[0039] To avoid material contamination during the crushing process, ceramic tiles are attached to the inner walls of the crushing zone 1 and the classification zone 3, and the crushing nozzle 103 is also made of ceramic material, and zirconia or silicon carbide ceramic can be specifically selected;

[0040] It should be noted that the number, gas pressure, and gas temperature of the crushing nozzles 103 are adjusted according to the material hardness and crushing requirements.

[0041] Furthermore, to prevent the coarse particles from interfering with the mixed particles in the upward airflow during the falling process and causing airflow disorder, a shunt cylinder 305 is provided in the connecting transition section between the crushing area and the classification area. The shunt cylinder 305 is coaxially arranged with the classification chamber 302, and a conical deflector 306 is provided at the top of the classification chamber 302, so that the upward airflow can smoothly discharge from the shunt cylinder 305 and reach the classification impeller 303 smoothly. The coarse particles thrown out enter the annular chamber between the shunt cylinder and the crushing cylinder body 101 and fall back to the crushing area, avoiding the mutual interference between the upward airflow and the falling particles.

[0042] During the production process, the control module uses the data acquisition unit to measure the particle size distribution (d10, d50, d90, etc.) of the powder in the classification area 3 and the particle size distribution of the powder at the collection end at the rear of the classification chamber 302, and simultaneously records the gas pressure, airflow temperature, feed rate, rotation speed of the classification impeller 303, and operating current of the frequency modulation motor 304 in the whole machine chamber in real time;

[0043] Among them, a high-frame-rate laser scattering instrument is used in the classification area to obtain the particle size distribution matrix D of the particle group in the classification area in real time in (t)=[d 10 ,d 50 ,d 90 T ; at the rear of the classification chamber 302, an ultrasonic particle size analyzer is used to obtain the average particle size D of the collected powder particle group in real time out (t);

[0044] The optimization control unit extracts the historical operation data from the data storage unit and normalizes the data. Specifically, Min-Max normalization processing is used; according to the Pearson correlation coefficient method, the features with a correlation exceeding the set value (set to 0.4 here) with D out (t) are selected to construct the feature vector X(t)=[D in (t),D out (t),P(t),N(t),V(t)]; where P(t), N(t), and V(t) represent the intake pressure, classification impeller rotation speed, and feed rate respectively.

[0045] For a specific material, a large number of feature vectors are arranged in time sequence to obtain a feature vector sequence. An LSTM prediction network is built to predict the particle size distribution of the powder in a future period based on real-time data. Specifically;

[0046] A sliding window is constructed to move on the feature vector sequence based on the historical operation data. The width of the window is set to W (usually taking 50-100) time steps, and the step size is set to time step; the data within the sliding window is used as the input, and the corresponding data for the first M (usually taking 5-10) time steps in the future is used as the output [D​in (t + 1), D in (t + 2),..., D in (t + M)] and [D out (t + 1), D out (t + 2),..., D out (t + M)] as the output; then divide the training set and the test set to conduct batch training on the above LSTM prediction network to obtain a network model that meets the performance requirements, such as the root mean square error (RMSE) and the mean absolute percentage error (MAPE) meeting the set indicators.

[0047] Define the effective output under unit energy consumption as the aging ratio, and construct an optimization function with maximizing the aging ratio as the optimization goal:

[0048] Among them, Q(t) represents the effective output that meets the neighborhood range of the target particle size; E(t) represents the comprehensive energy consumption; d 50 (t) represents the median particle size at the current moment, d s represents the target particle size; a, b, c are weight factors used to balance the priorities of output, energy consumption, and particle size deviation;

[0049] The constraint conditions include:

[0050] ① Particle size upper limit constraint: d 90 (t + i) ≤ d max +Δd safe d max represents the maximum particle size allowed in this comminution process, and Δd safe represents the safety margin to prevent constraint failure caused by prediction errors, generally taking 0.5 μm;

[0051] ② Equipment safety constraint: I(t) ≤ I 额定 , indicating that the current of the variable-frequency motor 304 does not exceed the rated current;

[0052] ③ Rotation speed range constraint: N min ≤ N(t) ≤ N max , that is, the rotation speed of the classification impeller 303 is within the design allowable range;

[0053] In this embodiment, the rotation speed N(t) of the classification impeller 303 is selected as the decision variable, and the particle swarm algorithm is used to solve the above optimization problem to obtain the optimal rotation speed sequence N * (t);

[0054] After obtaining the optimal rotation speed sequence, control the frequency conversion motor 304 according to the first five rotation speed commands in the sequence, and then move the sliding window to the next position based on the new operation data for re-optimization; repeat the operation in turn to complete the real-time control of the rotation speed of the classification impeller 30.

[0055] The above are only the preferred embodiments of the present invention, and do not impose any form of limitation on the present invention. Although the present invention has been disclosed above with the preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some changes or modifications to equivalent embodiments by using the technical content disclosed above without departing from the technical solution of the present invention. However, as long as it does not depart from the technical solution content of the present invention, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention still fall within the scope of the technical solution of the present invention.

Claims

1. An intelligent fluidized bed airflow pulverizer, characterized in that: The invention comprises a hardware part and a control module, wherein the hardware part is provided with a crushing area (1) and a classification area (3), wherein the crushing area (1) comprises a crushing cylinder (101), one side of which is fixedly provided with an auger conveyor, an annular gas distribution plate (102) is installed outside the crushing cylinder (101), and a plurality of crushing nozzles (103) evenly distributed at equal angles are connected to the inner side of the crushing cylinder (101), the classification area (3) is located above the crushing area (1), and a plurality of classification impellers (303) driven by a variable frequency motor (304) are installed at equal angles inside the classification cylinder (301) of the classification area (3), a classification chamber (302) is coaxially provided inside the classification cylinder (301), and the classification impeller (303) is rotatably installed between the classification chamber (302) and the classification cylinder (301); The control module comprises a data acquisition unit and an optimization control unit, wherein the data acquisition unit is used to obtain in real time the particle size distribution of the particle group in the classification area (3) and downstream of the classification chamber (302), and simultaneously record the operation data of the device, including gas pressure, gas flow temperature, feed speed, rotation speed of the classification impeller (303) and working current of the frequency modulation motor (304); Based on the above operating data and the particle size distribution of the particle group, an LSTM prediction network is built to predict the powder particle size distribution in a certain period of time in the future. An optimization function with maximizing the time efficiency ratio as the optimization goal is constructed based on the predicted data. The particle swarm algorithm is used to calculate the optimal speed sequence, and the rolling time domain control mode is used for optimization and update.

2. The intelligent fluidized bed airflow pulverizer according to claim 1, characterized in that: The annular gas distribution plate (102) is connected to a high-pressure gas circuit, and the high-pressure gas circuit uses low-temperature helium or low-temperature nitrogen with a temperature range of -20°C to -150°C.

3. The intelligent fluidized bed airflow pulverizer according to claim 1, characterized in that: A flow divider (305) is provided in the connecting transition section between the pulverizing zone (1) and the classification zone (3), and a conical flow guide plate (306) is provided on the top of the classification chamber (302) for separating the rising airflow from the falling coarse particles.

4. The intelligent fluidized bed airflow mill according to claim 1, characterized in that: The inner walls of the crushing area (1) and the classification area (3) are both affixed with ceramic tiles, and the crushing nozzle (103) is also made of ceramic material.

5. The intelligent fluidized bed airflow mill according to claim 1, characterized in that: Before building the LSTM prediction network, the historical operation is preprocessed to construct a feature vector sequence with time series. The feature vector is X(t) = [D in (t),D out (t), P(t), N(t), V(t)]; Among them, D in (t) = [d 10 ,d 50 ,d 90 ] T , represents the particle size distribution matrix of the particle group in the classification area at the corresponding time; D out (t) represents the average particle size of the powder particle group collected at the rear end of the classification chamber (302) at the corresponding moment; P(t), N(t), and V(t) represent the intake pressure, the rotation speed of the classification impeller (303), and the feed speed, respectively.

6. The intelligent fluidized bed airflow pulverizer according to claim 5, characterized in that: The process of building the LSTM prediction network includes: Construct a sliding window to move on the feature vector sequence, the window width is set to W time steps, and the step size is set to W time steps; Take the data in the sliding window as input and the corresponding data of the first M time steps in the future as output and As output; then divide the training set and test set to perform batch training on the LSTM prediction network to obtain a network model that meets the performance requirements.

7. The intelligent fluidized bed airflow pulverizer according to claim 6, characterized in that: The optimization function is Where Q(t) represents the effective yield within the target particle size range; E(t) represents the comprehensive energy consumption; d 50 (t) represents the median particle size at the current moment, d s Indicates the target particle size; a, b, c are weight factors used to balance the priority of yield, energy consumption and particle size deviation; The constraints include: ①Particle size upper limit constraint: d 90 (t+i)≤d max +Δd safe , d max Indicates the maximum particle size allowed in this pulverization process, Δd safe Indicates safety margin; ② Equipment safety constraints: I(t)≤I 额定 , indicating that the current of the variable frequency motor (304) does not exceed the rated current; ③Speed ​​range constraint: N min ≤N(t)≤N max , that is, the rotation speed of the classifying impeller (303) is within the design allowable range; The decision variable is set as the speed N(t) of the classifying impeller (303), and the particle swarm algorithm is used to solve the function to obtain the optimal speed sequence N * (t).

8. The intelligent fluidized bed airflow pulverizer according to claim 1, characterized in that: The rolling time domain control mode is to select the first five speed instructions in the optimal speed sequence to control the frequency modulation motor (304), and then move the sliding window to the next position based on the new operation data for re-optimization.

Images