A spheronization and sorting device for one-shot formed glass microbeads

Abstract

The application relates to a one-time forming glass microsphere spheroidizing and sorting device, and belongs to the technical field of material sorting, which comprises the following: a feeding and melting module; the inlet end of a gradient spheroidizing module is connected with the outlet end of the feeding and melting module; a multistage precision sorting module is connected with the gradient spheroidizing module, is used for carrying out multilayer screening on the formed glass microspheres, is used for identifying the roundness parameters of the glass microspheres in real time and screening out unqualified products, and realizes the layered sorting of glass microspheres with different densities through a density gradient liquid; a sphericity detection and sorting module is connected with the multistage precision sorting module, transports the glass microspheres meeting the preset sphericity, and transports the glass microspheres not meeting the preset sphericity to a waste recovery channel; a finished product collecting unit is connected with the sphericity detection and sorting module and is used for collecting the glass microspheres meeting the preset sphericity; and the application has the beneficial effect that the sorting efficiency of the glass microspheres is improved through multilayer screening, roundness detection and density sorting.

Description

Technical Field

[0001] This invention belongs to the field of material sorting technology, and specifically relates to a spheroidizing and sorting device for one-time molding of glass microspheres. Background Technology

[0002] Glass microspheres are inorganic non-metallic materials with excellent properties, widely used in coatings, inks, plastics, rubber, building materials, and aerospace. In the production of glass microspheres, spheroidization and sorting are two key processes that directly affect the quality and application effects of the microspheres.

[0003] Currently, traditional glass microsphere production typically involves separate spheroidizing and sorting equipment. First, the glass raw material is melted and formed into glass microspheres using the spheroidizing equipment, then the microspheres are transported to the sorting equipment for grading and screening. This production method has the following drawbacks: First, the production process is cumbersome, requiring multiple transfers of the glass microspheres, which not only increases labor intensity but also easily leads to damage or contamination of the microspheres during transport. Second, production efficiency is low, as there is a time interval between the spheroidizing and sorting processes, making continuous production impossible. Third, the equipment requires a large footprint, necessitating separate spheroidizing and sorting workshops, increasing production costs. Fourth, the level of automation is low, with poor coordination between equipment, making it difficult to accurately control production parameters, resulting in unstable quality of the glass microspheres. Summary of the Invention

[0004] This invention provides a one-time molding glass microsphere spheroidizing and sorting device to solve the technical problem of low sorting efficiency of glass microspheres in the prior art. It improves the sorting efficiency of glass microspheres through multi-layer screening, roundness detection and density sorting.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solution:

[0006] A one-piece molded glass microsphere spheroidizing and sorting device, comprising:

[0007] The feeding and melting module is used for conveying glass raw materials, heating them to a molten state, and forming initial droplets;

[0008] The gradient spheroidizing module, whose inlet end is connected to the outlet end of the feeding and melting module, is used to receive the initial droplets and shape them into glass microspheres by surface tension during the falling process in a controlled high-temperature inert environment.

[0009] The multi-level precision sorting module, connected to the gradient spheroidization module, includes: a particle size screening unit with a multi-layer sieve structure for multi-layer screening of the formed glass microspheres; a roundness detection unit with built-in image acquisition components and intelligent image recognition algorithms for real-time identification of the roundness parameters of the glass microspheres and screening out unqualified products; and a density sorting unit that achieves layered sorting of glass microspheres of different densities through a density gradient liquid.

[0010] The sphericity detection and sorting module is connected to the multi-level precision sorting module to transport glass microspheres that meet the preset sphericity and glass microspheres that do not meet the preset sphericity to the waste recycling channel.

[0011] The finished product collection unit is connected to the sphericity detection and sorting module and is used to collect glass microspheres that meet the preset sphericity.

[0012] Optionally, for the feeding and melting module, the operation mode is as follows:

[0013] Step a: Precise delivery of glass raw materials to achieve quantity control, contamination prevention, and uniform feeding;

[0014] Step b: Heat the raw materials to a molten state to achieve high temperature, uniformity, and impurity removal;

[0015] Step c: Formation of the initial droplet; quantitative, shaped and stable droplet falling.

[0016] Optionally, inside the gradient spheroidizing module, after the initial droplet enters the module, due to the precise setting of the internal temperature gradient, the glass molecules on the surface of the droplet have different activities due to temperature differences during the droplet's descent. The glass molecules in the high-temperature region have high activity and strong fluidity. Under the action of surface tension, they gradually migrate to the region with lower energy, causing the droplet surface to become smooth and round. As the droplet continues to fall and enters the region with gradually decreasing temperature, the fluidity of the glass molecules gradually weakens, the spheroidization morphology is fixed, and finally, glass microspheres with smooth surfaces and high sphericity are formed.

[0017] Optionally, the screening efficiency of the particle size screening unit is calculated as the percentage of the mass of qualified particles through the target screen to the total mass of all qualified particles in the raw material. It is a core indicator for evaluating the performance of the screening unit, and the formula is as follows:

[0018] ;

[0019] in, For screening efficiency; The total mass of material passing through the sieve or particles that did not pass through the sieve. The total mass of the material that passes through the sieve or the particles that pass through the sieve. This represents the mass fraction of qualified particles in the raw material. This represents the mass fraction of qualified particles in the material remaining on the sieve; and how many qualified particles are mixed in with the unqualified particles remaining on the sieve. The higher the ratio, the more qualified particles remain on the sieve, resulting in a more failed sieve process. The screening efficiency is calculated by multiplying the proportion of qualified particles that actually pass through the sieve by 100% and converting it into a percentage.

[0020] Screening time is estimated as follows: too short a screening time will result in qualified particles not passing through the screen sufficiently, while too long a time will reduce production efficiency. Based on screening kinetics, the formula for estimating the effective screening time of a single-layer screen is:

[0021] ;

[0022] in, For filtering time; For equipment coefficients; The thickness of the particles deposited on the screen; The packing density of glass microspheres; The average velocity of the particles passing through the sieve openings; This refers to the utilization rate of the sieve aperture.

[0023] Optionally, in the image acquisition component and the image intelligent recognition algorithm, the image acquisition component acquires high-resolution, distortion-free, and high-contrast images of glass microbeads to provide a reliable data source for subsequent algorithms. The image acquisition component acquires the glass microbead images as follows:

[0024] ;

[0025] The target photocurrent is used to generate a target photocurrent distribution image; The background photocurrent is used to generate a background photocurrent distribution image, which ensures that the background of the image is not completely black. To increase the contrast ratio, the difference between the reflected light from the microbeads and the background light is forcibly increased, making the image outline sharper; The reflectivity of glass microspheres; Illuminance of the light source; The projected area of ​​the microspheres; The distance from the light source to the microbeads;

[0026] The contribution of the microbeads to the reflected light makes the outline of the microbeads clearly visible in the image; The total luminous flux reflected by the surface of the microspheres is calculated as reflectivity × incident light × projected area, which reflects how much light the microspheres can reflect. Following the inverse square law, light intensity decreases with the square of the distance. This simulates the light loss from the light source to the microspheres by controlling... This allows the microspheres to reflect light more strongly than they transmit light.

[0027] Optionally, the image intelligent recognition algorithm, corresponding to image-text comparison and text-image comparison, forces the model to align semantics in both directions:

[0028] ;

[0029] in, is the loss function, representing the bidirectional semantic alignment loss, used to describe the semantic binding tightness between images and text; For the first An index of image-text pairs, a batch of... For the sample; For the first Feature vectors of the image; For the first Feature vectors of the image; For the first Feature vectors of a paragraph text; For the first Feature vectors of a paragraph text; For temperature parameters;

[0030] For image-text comparison; where, It is an image With text The dot product similarity of matched features measures the degree of semantic matching. For image The sum of the exponential sum of the feature dot products of all texts constitutes global semantic competition.

[0031] For text-image contrast; The image-text contrast is symmetrical, which is the computation of text. Matching images The similarity probability;

[0032] The overall loss is the sum of negative logarithmic probabilities. ;

[0033] The negative sign maximizes the probability of a correct match in the model. It is a batch The average of each sample is used to stabilize the training gradient.

[0034] let and The similarity is higher than other mismatched pairs. By using global competition in the denominator of Softmax, the model is forced to learn to distinguish between semantically similar but mismatched samples.

[0035] Optionally, for the stratified sorting of glass microspheres with different densities, the sorting accuracy is directly determined by the linearity of the density gradient. A formula for accurately calculating the density gradient at any height is established through the coupling relationship between volume fraction, density, and height.

[0036] ;

[0037] in, For density gradient liquids at height Density at that location; The density of the low-density mother liquor; The density of the high-density mother liquor; The height at any position in the gradient liquid; This represents the total height of the density gradient liquid; This represents the total volume of the density gradient liquid; This is the gradient adjustment coefficient, which controls the nonlinear shape of the gradient; It is an exponential term that describes the rate at which the gradient changes with height. The larger the gradient, the more gradual the density change with height, enabling customization of the gradient morphology.

[0038] It describes the nonlinear law of density gradient change with height, the density difference between high-density mother liquor and low-density mother liquor; It is the density difference between the high-density mother liquor and the low-density mother liquor, which varies with height in a gradient liquid. The proportion of non-linear allocation.

[0039] Optionally, the sphericity detection and sorting module includes an optical imaging unit, which is used to acquire high-resolution and distortion-free real-time images of glass microspheres after preliminary screening by the multi-level precision sorting module.

[0040] Specifically, a high-brightness, low-flicker LED ring light source is used to eliminate reflective interference on the surface of the glass microspheres through diffuse reflection illumination, ensuring that the outline of the microspheres is clearly visible.

[0041] Optionally, the image analysis and processing unit interacts with the optical imaging unit to perform real-time processing of the microbead images, sphericity calculation, and conformity assessment. Its workflow is as follows:

[0042] Sphericity calculation is based on microsphere profile data and uses the sphericity calculation formula: Sphericity = Surface area of ​​a sphere with the same volume as the microsphere / Actual surface area of ​​the microsphere. The calculation is performed quantitatively, and the maximum circumcircle, minimum incircle, and equivalent diameter of the microsphere profile are automatically identified by image analysis software. The volume and actual surface area of ​​the microsphere are then calculated to obtain the sphericity value.

[0043] Optionally, the pneumatic sorting unit, based on the judgment results of the image analysis and processing unit, achieves rapid separation of qualified and unqualified microbeads. The specific operation is as follows:

[0044] The qualified product channel and the unqualified product channel are set up below the conveyor track. When an unqualified signal is received, the unqualified microbeads are sent to the unqualified product channel, while the qualified microbeads fall naturally into the qualified product channel under the action of gravity, thus achieving efficient separation between the two.

[0045] The beneficial effects of this invention are:

[0046] 1. The particle size screening unit of the present invention adopts a multi-layer sieve structure to perform multi-layer screening of the formed glass microspheres. The roundness detection unit has a built-in image acquisition component and an intelligent image recognition algorithm to identify the roundness parameters of the glass microspheres in real time and screen out unqualified products. The density sorting unit realizes the layered sorting of glass microspheres of different densities through a density gradient liquid. The sorting efficiency of glass microspheres is improved through multi-layer screening, roundness detection and density sorting.

[0047] 2. The image analysis and processing unit of the present invention completes real-time processing of microsphere images and sphericity calculation through data interaction with the optical imaging unit. Specifically, the sphericity calculation is performed based on the microsphere contour data using the sphericity calculation formula: Sphericity = Surface area of ​​a sphere with the same volume as the microsphere / Actual surface area of ​​the microsphere. The image analysis software automatically identifies the maximum circumcircle, minimum incircle, and equivalent diameter of the microsphere contour, calculates the volume and actual surface area of ​​the microsphere, and finally obtains the sphericity value. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0049] Figure 1 This is a schematic diagram of the overall structure of the present invention;

[0050] Figure 2 This is a schematic diagram of the internal structure of the multi-level precision sorting module of the present invention. Detailed Implementation

[0051] The embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0052] Example 1

[0053] like Figure 1As shown, this embodiment provides a one-time forming glass microsphere spheroidizing and sorting device, including: a feeding and melting module, a gradient spheroidizing module, a multi-stage precision sorting module, a sphericity detection and sorting module, and a finished product collection module connected in sequence. Each module forms an integrated operation process through a closed conveying channel, realizing the one-time forming of glass microspheres from raw material melting to spheroidizing and then to graded sorting.

[0054] The feeding and melting module is used for conveying glass raw materials, heating them to a molten state, and forming initial droplets;

[0055] The gradient spheroidizing module, whose inlet end is connected to the outlet end of the feeding and melting module, is used to receive the initial droplets and shape them into glass microspheres by surface tension during the falling process in a controlled high-temperature inert environment.

[0056] Multi-level precision sorting modules, such as Figure 2 As shown, connected to the gradient spheroidization module, it includes a particle size screening unit, a roundness detection unit, and a density sorting unit connected in sequence; the particle size screening unit adopts a multi-layer sieve structure for multi-layer screening of the formed glass microspheres; the roundness detection unit has a built-in image acquisition component and an intelligent image recognition algorithm for real-time identification of the roundness parameters of the glass microspheres and screening out unqualified products; the density sorting unit achieves layered sorting of glass microspheres of different densities through a density gradient liquid;

[0057] The sphericity detection and sorting module, connected to the multi-stage precision sorting module, includes an optical imaging unit, an image analysis and processing unit, and a pneumatic sorting unit. The optical imaging unit acquires real-time images of the glass microspheres after processing by the multi-stage precision sorting module. The image analysis and processing unit, connected to the optical imaging unit, calculates the sphericity of the glass microspheres in the image in real time and determines whether they meet the preset sphericity. The pneumatic sorting unit, connected to the image analysis and processing unit, sorts the microspheres based on the judgment results of the image analysis and processing module. Glass microspheres that meet the preset sphericity are transported to the finished product collection unit, while glass microspheres that do not meet the preset sphericity are transported to the waste recycling channel.

[0058] The finished product collection unit is connected to the sphericity detection and sorting module and is used to collect glass microspheres that meet the preset sphericity.

[0059] Example 2

[0060] Based on Example 1, the working principle of the feeding and melting module is as follows:

[0061] Step a: Precise delivery of glass raw materials to achieve quantity control, contamination prevention, and uniform feeding;

[0062] The mixed glass raw materials (usually dry powder or micron-sized particles) are fed to the heating zone at a stable and controllable rate to avoid raw material accumulation, jamming in the conveying channel, or mixing with impurities. At the same time, the feed rate is matched with the subsequent melting rate and droplet formation rate (if the feed is too fast, the melting will be insufficient, and if it is too slow, the production efficiency will be reduced).

[0063] The conveying method is based on the characteristics of the raw materials (such as particle size and flowability) and the production scale, as follows:

[0064] For small-batch / high-precision scenarios (e.g., optical glass production): screw conveyors are used. The raw materials are pushed by the uniform rotation of the screw, which can accurately control the feed rate (the error is usually ≤ ±1%). The conveying channel is closed, which can avoid external dust pollution.

[0065] For high-volume / continuous production scenarios (e.g., daily-use glass): a vibratory conveyor + silo buffer is used. The silo stores mixed raw materials, and the vibrator uses high-frequency, low-amplitude vibration to make the raw materials slide down evenly. The vibration frequency is adjusted in real time with the flow sensor to ensure stable feeding.

[0066] Step b: Heat the raw materials to a molten state to achieve high temperature, uniformity, and impurity removal;

[0067] The solid raw materials are completely melted and reacted in a high-temperature environment, and the bubbles and unmelted particles generated during the melting process are eliminated to form a glass melt with uniform composition. The melting temperature of the glass raw materials is between 1400 and 1600℃ (different types of glass vary greatly, such as quartz glass which requires a melting temperature of over 1700℃ and everyday soda-lime glass which is about 1450℃). Therefore, the heating system must meet the requirements of high-temperature stability and uniform temperature field.

[0068] Step c: Formation of the initial droplet; quantitative, shaped, and stable dropping;

[0069] The molten glass needs to be converted into initial droplets, the size, shape and drop frequency of which need to be matched with the subsequent forming process. The droplets are precisely controlled by forming nozzles and flow control.

[0070] Droplet formation principle: Molten glass has a certain viscosity (e.g., the viscosity decreases as the temperature changes). When the molten glass flows out of the nozzle, it will naturally shrink into a spherical droplet under the combined action of gravity and surface tension (surface tension makes the liquid tend to have the smallest surface area, i.e., spherical). By controlling the nozzle orifice diameter, the viscosity of the molten glass (i.e., adjusting the temperature), and the flow rate, the diameter of the droplet can be controlled (e.g., from a few millimeters to tens of millimeters).

[0071] Example 3

[0072] Based on Example 1, a precision temperature control system was constructed inside the gradient spheroidization module. Multiple sets of high-precision heating elements and temperature sensors were arranged vertically from the top to the bottom of the cavity. The heating elements worked together through an advanced intelligent control system to accurately create a high-temperature environment that changes in gradient from top to bottom.

[0073] During the preparation of glass microspheres, the temperature of the top region is maintained at around 1500°C to ensure that the initial droplets are in a good plastic state when they enter; the temperature of the middle region is reduced to around 1200°C, at which point the glass droplets begin to spheroidize under the action of surface tension; the temperature of the bottom region is reduced to around 900°C, which promotes the glass microspheres to basically complete spheroidization and gradually cool and solidify.

[0074] By precisely controlling the temperature in different areas, glass droplets can be effectively guided to gradually spherize according to a preset pattern during their descent, greatly improving the sphericity and quality stability of glass microspheres.

[0075] To create a stable high-temperature inert environment, the gradient spheroidizing module is equipped with an inert gas supply and circulation system. High-purity inert gases (such as argon and nitrogen) are injected from the top of the chamber at a uniform and stable flow rate through a designed inlet pipe. Inside the chamber, a multi-layer gas distribution device is installed to ensure that the inert gas evenly fills the entire space, effectively removing oxygen and preventing oxidation of the glass droplets during spheroidizing, which would affect product quality. Simultaneously, an outlet pipe is located at the bottom, connected to a gas purification and recovery device, to purify and enhance the used inert gas for easy recycling. This reduces production costs and aligns with environmental protection principles.

[0076] After the initial droplet enters the gradient spheroidization module, due to the precise setting of the internal temperature gradient, the glass molecules on the droplet surface exhibit different activities due to temperature differences during its descent. Glass molecules in the high-temperature region are highly active and fluid, gradually migrating to lower-energy regions under the influence of surface tension, causing the droplet surface to become smooth and rounded. As the droplet continues to fall and enters the region where the temperature gradually decreases, the fluidity of the glass molecules gradually weakens, the spheroidization morphology is fixed, and ultimately, smooth, highly spherical glass microspheres are formed.

[0077] Example 4

[0078] Based on Example 1, such as Figure 2 As shown, the essence of screening using the multi-layer screen structure of the particle size screening unit is that different layers of screens have different mesh counts, that is, the apertures of different layers of screens are different. The particles are stratified and screened by external forces such as vibration and airflow. The particle size screening unit works around the conversion between screen aperture and mesh count, the calculation of screening efficiency, and the estimation of screening time.

[0079] The particle size screening unit uses a conversion method between sieve aperture and mesh count. Mesh count is the core indicator of a sieve, referring to the number of sieve openings per inch (25.4 mm) of length. The mesh count of a sieve layer is inversely proportional to the aperture of that layer; the larger the mesh count, the smaller the aperture on the sieve. In practical applications, the mesh count of each sieve layer is determined by conversion based on the particle size range of the target glass microspheres, using the following formula:

[0080] ;

[0081] in, This represents the actual aperture of the sieve; 25.4 is the number of millimeters corresponding to inches. The diameter of the screen wire; The mesh size refers to the number of sieves. In the sieving process of glass microspheres (or other particulate materials), particles of different sizes are separated by multiple layers of sieves, so the mesh size must first be selected according to the target particle size.

[0082] The formula for calculating screening efficiency is as follows: Screening efficiency is the percentage of the mass of qualified particles through the target screen to the total mass of all qualified particles in the raw material. It is a core indicator for evaluating the performance of a screening unit. The formula is as follows:

[0083] ;

[0084] in, For screening efficiency; The total mass of material passing through the sieve or particles that did not pass through the sieve. The total mass of the material that passes through the sieve or the particles that pass through the sieve. This represents the mass fraction of qualified particles in the raw material. This represents the mass fraction of qualified particles in the sieve residue. and Used to measure the effectiveness of a screen in separating qualified particles or target particle size particles during the screening process of particulate materials. In industrial scenarios such as glass microsphere screening or powder classification, it is used to determine whether the performance of screening equipment (such as multi-layer screen units) meets the standards. First, observe how many qualified particles are mixed in with the unqualified particles that should remain on the screen. This part, The higher the ratio, the more qualified particles remain on the sieve, indicating a greater failure in the screening process. The screening efficiency is calculated by multiplying the percentage of qualified particles that actually pass through the sieve by 100%.

[0085] The screening time is estimated as follows: too short a screening time will result in qualified particles not passing through the screen sufficiently, while too long a time will reduce production efficiency. Based on screening kinetics, the formula for estimating the effective screening time of a single-layer screen is:

[0086] ;

[0087] in, For filtering time; For equipment coefficients (k=1.2-1.5 for vibrating screens, k=0.8-1.0 for air classifiers, determined by the equipment type); The thickness of particle accumulation on the screen should be controlled between 5 and 20 mm; if it is too thick, it will clog the screen holes. The bulk density of the glass microspheres is controlled between 0.4 and 0.6 g / cm³, which is determined by the uniformity of the microsphere particle size. The average velocity of particles passing through the sieve openings (5~10 mm / s for vibrating screens, 10~15 mm / s for airflow screens); The utilization rate of the screen openings is 60%~80%, which is related to the screen weaving method, with plain weave > twill weave.

[0088] The total screening time of multiple screens is the sum of the screening time of a single layer. However, in actual production, due to the top-down arrangement of screens (coarse screen → fine screen), the screening time of the upper coarse screen is usually shorter than that of the lower fine screen, and needs to be dynamically adjusted according to the particle quantity of each layer.

[0089] The particle size screening unit is integrated into the vibrating screen (mainstream) or air classifier, following the process of raw material pretreatment → layered screening → graded collection → equipment cleaning, ensuring no cross-contamination and no particle size mixing at each step.

[0090] In the image acquisition component and image intelligent recognition algorithm, the image acquisition component obtains high-resolution, distortion-free, and high-contrast images of glass microspheres, providing a reliable data source for subsequent algorithms. The image acquisition component acquires the glass microsphere images as follows:

[0091] ;

[0092] The target photocurrent (in A or mA) is used to generate a target photocurrent distribution image; The background photocurrent (in A or mA) is used to generate a background photocurrent distribution image, which ensures that the background of the image is not completely black. Conventional photocurrent imaging is a technique that captures the weak current signal generated by a material under illumination and converts the weak current signal into an intuitive image to reveal the photoelectric properties inside the material. The contrast coefficient is set to 1.5 to highlight the edges of the microbeads, thus forcibly increasing the difference between the microbead reflection and the background light, making the image outline sharper. The reflectivity of glass microspheres is approximately 0.08; high light transmittance requires a combination of side lighting and backlighting. Illuminance of the light source (e.g., 2000 lux); The projected area of ​​the microspheres; To determine the distance from the light source to the microbeads, a combination of 45° side lighting and backlighting is used to avoid blurring of the outline caused by light transmission.

[0093] The contribution of the microbeads to the reflected light makes the outline of the microbeads clearly visible in the image; The total luminous flux reflected by the surface of the microspheres is calculated as reflectivity × incident light × projected area, which reflects how much light the microspheres can reflect. Following the inverse square law, light intensity decreases with the square of the distance, simulating light loss from the light source to the microspheres. This is achieved by controlling... This allows the microspheres to reflect light more strongly than they transmit light, thus solving the problem of blurred outlines caused by light transmission in glass microspheres.

[0094] The image intelligent recognition algorithm consists of two parts, corresponding to image-text comparison and text-image comparison respectively, forcing the model to align semantics in both directions:

[0095] ;

[0096] in, Let be the loss function, representing the bidirectional semantic alignment loss, used to describe the semantic binding tightness between images and text. A smaller value indicates a tighter semantic binding between the image and the text. A larger value indicates a looser semantic binding between the image and the text; For the first An index of image-text pairs, a batch of... For the sample; For the first Feature vectors of the image (extracted by an image encoder, such as CLIP's Vision Transformer). For the first Feature vectors of the image; For the first Feature vectors of the segment text (extracted by a text encoder, such as CLIP's Text Transformer). For the first Feature vectors of a paragraph text; Temperature parameter (controls the sharpness of the probability distribution, reduces) This will make the probability of getting it right more prominent.

[0097] Part 1: Image-Text Contrast ;

[0098] in, It is an image With text The dot product of matched features (similarity) measures the degree of semantic matching.

[0099] For image The sum of the feature dot products of all texts (including itself and the other N−1 non-matching texts) is the global semantic competition.

[0100] It is a computational image Match the correct text The higher the probability (after Softmax normalization), the better the model can distinguish between matching and non-matching text.

[0101] Part Two: Text-Image Contrast ;

[0102] Symmetrical to the first part is the computational text. Matching images The similarity probability.

[0103] Forced bidirectional alignment of the model: It requires not only that the image can find the corresponding text, but also that the text can find the corresponding image, to avoid unidirectional skew.

[0104] Overall loss: sum of negative logarithmic probabilities ;

[0105] The negative sign maximizes the probability of a matching pair (equivalent to minimizing the probability of a negative logarithm). It is a batch The average of each sample is used to stabilize the training gradient.

[0106] let and The similarity is higher than other mismatched pairs. By using global competition in the denominator of Softmax, the model is forced to learn to distinguish between semantically similar but mismatched samples.

[0107] Images and text are data of different modalities (visual vs. language), but semantic relationships exist in reality. Models need to learn to recognize these relationships in order to achieve image-text retrieval and zero-shot recognition. This forces the model to bring matching image-text pairs closer together in the feature space and to keep unmatched pairs further apart, ultimately understanding cross-modal semantics.

[0108] For the stratified sorting of glass microspheres with different densities, the linearity of the density gradient directly determines the sorting accuracy. By leveraging the coupling relationship between volume fraction, density, and height, a formula for accurately calculating the density gradient at any height is established:

[0109] ;

[0110] in, For density gradient liquids at height Density at that location; The density of the low-density mother liquor; The density of the high-density mother liquor; The height at any position in the gradient liquid; This represents the total height of the density gradient liquid; This represents the total volume of the density gradient liquid; This is the gradient adjustment coefficient, which controls the nonlinear shape of the gradient. The smaller the gradient, the closer it is to linear, which is suitable for particles with uniform density distribution. The larger the value, the steeper the bottom gradient and the gentler the top gradient, making it suitable for scenarios requiring higher resolution in high-density regions (e.g., sorting samples containing a small amount of heavy particles). It is an exponential term that describes the rate at which the gradient changes with height. The larger the gradient, the more gradual the density change with height, enabling customization of the gradient morphology.

[0111] It describes the nonlinear law of density gradient change with height, the density difference between high-density mother liquor and low-density mother liquor; It is the density difference between the high-density mother liquor and the low-density mother liquor, which varies with height in a gradient liquid. The proportion of non-linear allocation;

[0112] when =0 (bottom of container), molecule →Correction term is 0→Gradient liquid density = Low-density mother liquor density ;

[0113] when = (Top of container), molecules Equal to the denominator → Correction term →Gradient liquid density = High-density mother liquor density ;

[0114] Mid-height The correction term controls the density from the exponential function. arrive The transition rate is used to achieve a nonlinear gradient (distinct from the traditional linear gradient assumption).

[0115] In a density gradient liquid, at any height Liquid density at the location Precise control is achieved through the ratio of high and low density mother liquor and nonlinear gradient adjustment.

[0116] by Based on (low-density mother liquor density), dynamic allocation is achieved through fractional terms (correction terms). The difference in density between high and low points causes density to vary with height. Nonlinear change;

[0117] Finally at the top of the liquid column = At that time, the density just reached That is, the high density of the mother liquor, which forms from arrive A continuous, controllable density field.

[0118] Example 5

[0119] Based on Example 1, the sphericity detection and sorting module includes an optical imaging unit. The optical imaging unit is used to acquire high-resolution and distortion-free real-time images of glass microspheres after preliminary screening by the multi-level precision sorting module, providing a reliable data basis for subsequent sphericity calculation. Specifically, a high-brightness, low-flicker LED ring light source is used to eliminate surface reflection interference of glass microspheres through diffuse reflection illumination, ensuring that the outline of the microspheres is clearly visible. The brightness of the light source can be automatically adjusted according to the particle size of the microspheres (usually adapted to the range of 1-100μm) to avoid image blurring or overexposure due to insufficient brightness.

[0120] The image analysis and processing unit interacts with the optical imaging unit to perform real-time processing of microbead images, sphericity calculation, and conformity assessment. Its workflow is as follows:

[0121] First, image preprocessing is performed on the acquired raw image, including noise reduction, grayscale conversion, and edge detection (using the Canny algorithm) to eliminate the influence of ambient light interference and image noise, and accurately extract the contour information of the microbeads.

[0122] Sphericity calculation is based on microsphere profile data and uses the internationally accepted sphericity calculation formula (sphericity = surface area of ​​a sphere with the same volume as the microsphere / actual surface area of ​​the microsphere) for quantitative calculation. Specifically, image analysis software automatically identifies the maximum circumcircle, minimum incircle, and equivalent diameter of the microsphere profile, calculates the volume and actual surface area of ​​the microsphere, and finally obtains the sphericity value; the calculation accuracy can reach ±0.01, meeting the requirements for high-precision sorting.

[0123] For the qualification judgment, the user can preset the sphericity qualification threshold (e.g., 90%, 95%, and 98%) through the module control system. The image analysis and processing unit will compare the real-time calculated sphericity value with the preset threshold to quickly determine whether the microbeads are qualified and generate a qualified or unqualified control signal. At the same time, it will automatically record the sphericity data of each microbead to form a sorting log, which will facilitate subsequent quality traceability and process optimization.

[0124] Based on the judgment results of the image analysis and processing unit, the pneumatic sorting unit achieves rapid separation of qualified and unqualified microbeads. The specific operation is as follows:

[0125] The pneumatic sorting unit communicates with the image analysis and processing unit via Ethernet in real time to receive qualified / unqualified control signals, ensuring that each microbead can be precisely controlled.

[0126] The system is divided into a qualified product channel and a non-qualified product channel, which are located at a certain angle (usually 30°-45°) below the conveyor track. When a non-qualified signal is received, the non-qualified microbeads are sent to the non-qualified product channel; the qualified microbeads fall naturally into the qualified product channel under the action of gravity, thus achieving efficient separation between the two.

[0127] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope described in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A spheroidizing and sorting device for one-piece molded glass microspheres, characterized in that, include: The feeding and melting module is used for conveying glass raw materials, heating them to a molten state, and forming initial droplets; The gradient spheroidizing module, whose inlet end is connected to the outlet end of the feeding and melting module, is used to receive the initial droplets and shape them into glass microspheres by surface tension during the falling process in a controlled high-temperature inert environment. The multi-level precision sorting module, connected to the gradient spheroidization module, includes: a particle size screening unit with a multi-layer sieve structure for multi-layer screening of the formed glass microspheres; a roundness detection unit with built-in image acquisition components and intelligent image recognition algorithms for real-time identification of the roundness parameters of the glass microspheres and screening out unqualified products; and a density sorting unit that achieves layered sorting of glass microspheres of different densities through a density gradient liquid. The sphericity detection and sorting module is connected to the multi-level precision sorting module to transport glass microspheres that meet the preset sphericity and glass microspheres that do not meet the preset sphericity to the waste recycling channel. The finished product collection unit is connected to the sphericity detection and sorting module and is used to collect glass microspheres that meet the preset sphericity.

2. The spheroidizing and sorting device for one-time molding of glass microspheres according to claim 1, characterized in that, The feeding and melting module operates as follows: Step a: Precise delivery of glass raw materials to achieve quantity control, contamination prevention, and uniform feeding; Step b: Heat the raw materials to a molten state to achieve high temperature, uniformity, and impurity removal; Step c: Formation of the initial droplet; quantitative, shaped and stable droplet falling.

3. The spheroidizing and sorting device for one-time molding of glass microspheres according to claim 1, characterized in that, Inside the gradient spheroidizing module, after the initial droplet enters the module, due to the precise setting of the internal temperature gradient, the glass molecules on the surface of the droplet have different activities due to temperature differences during the droplet's descent. The glass molecules in the high-temperature region have high activity and strong fluidity. Under the action of surface tension, they gradually migrate to the region with lower energy, causing the droplet surface to become smooth and round. As the droplet continues to fall and enters the region with gradually decreasing temperature, the fluidity of the glass molecules gradually weakens, the spheroidization morphology is fixed, and finally, glass microspheres with smooth surfaces and high sphericity are formed.

4. The spheroidizing and sorting device for one-time molding of glass microspheres according to claim 1, characterized in that, The sieving efficiency of the particle size screening unit is calculated as the percentage of the mass of qualified particles through the target screen to the total mass of all qualified particles in the raw material. It is a core indicator for evaluating the performance of the screening unit, and the formula is as follows: ； in, For screening efficiency; The total mass of material passing through the sieve or particles that did not pass through the sieve. The total mass of the material that passes through the sieve or the particles that pass through the sieve. This represents the mass fraction of qualified particles in the raw material. This represents the mass fraction of qualified particles in the material remaining on the sieve; and how many qualified particles are mixed in with the unqualified particles remaining on the sieve. The higher the ratio, the more qualified particles remain on the sieve, resulting in a more failed sieve process. The screening efficiency is calculated by multiplying the proportion of qualified particles that actually pass through the sieve by 100% and converting it into a percentage. Screening time is estimated as follows: too short a screening time will result in qualified particles not passing through the screen sufficiently, while too long a time will reduce production efficiency. Based on screening kinetics, the formula for estimating the effective screening time of a single-layer screen is: ； in, For filtering time; For equipment coefficients; The thickness of the particles deposited on the screen; The packing density of glass microspheres; The average velocity of the particles passing through the sieve openings; This refers to the utilization rate of the sieve aperture.

5. The spheroidizing and sorting device for one-time molding of glass microspheres according to claim 1, characterized in that, In the image acquisition component and image intelligent recognition algorithm, the image acquisition component acquires high-resolution, distortion-free, and high-contrast images of glass microspheres to provide a reliable data source for subsequent algorithms. The image acquisition component acquires the glass microsphere images as follows: ； The target photocurrent is used to generate a target photocurrent distribution image; The background photocurrent is used to generate a background photocurrent distribution image, which ensures that the background of the image is not completely black. To increase the contrast ratio, the difference between the reflected light from the microbeads and the background light is forcibly increased, making the image outline sharper; The reflectivity of glass microspheres; Illuminance of the light source; The projected area of ​​the microspheres; The distance from the light source to the microbeads; The contribution of the microbeads to the reflected light makes the outline of the microbeads clearly visible in the image; The total luminous flux reflected by the surface of the microspheres is calculated as reflectivity × incident light × projected area, which reflects how much light the microspheres can reflect. Following the inverse square law, light intensity decreases with the square of the distance. This simulates the light loss from the light source to the microspheres by controlling... This allows the microspheres to reflect light more strongly than they transmit light.

6. The spheroidizing and sorting device for one-time molding of glass microspheres according to claim 5, characterized in that, The image intelligent recognition algorithm, corresponding to image-text comparison and text-image comparison, forces the model to align semantics in both directions: ； in, is the loss function, representing the bidirectional semantic alignment loss, used to describe the semantic binding tightness between images and text; For the first An index of image-text pairs, a batch of... For the sample; For the first Feature vectors of the image; For the first Feature vectors of the image; For the first Feature vectors of a paragraph text; For the first Feature vectors of a paragraph text; For temperature parameters; For image-text comparison; where, It is an image With text The dot product similarity of matched features measures the degree of semantic matching. For image The sum of the exponential sum of the feature dot products of all texts constitutes global semantic competition. For text-image comparison; The image-text contrast is symmetrical, which is the computation of text. Matching images The similarity probability; The overall loss is the sum of negative logarithmic probabilities. ; The negative sign maximizes the probability of a correct match in the model. It is a batch The average of each sample is used to stabilize the training gradient. let and The similarity was higher than other mismatched pairs. By using global competition in the denominator of Softmax, the model is forced to learn to distinguish between semantically similar but mismatched samples.

7. The spheroidizing and sorting device for one-time molding of glass microspheres according to claim 5, characterized in that, For the stratified sorting of glass microspheres with different densities, the linearity of the density gradient directly determines the sorting accuracy. Based on the coupling relationship between volume fraction, density, and height, a formula for accurately calculating the density gradient at any height is established: ； in, For density gradient liquids at height Density at that location; The density of the low-density mother liquor; The density of the high-density mother liquor; The height at any position in the gradient liquid; This represents the total height of the density gradient liquid; This represents the total volume of the density gradient liquid; This is the gradient adjustment coefficient, which controls the nonlinear shape of the gradient; It is an exponential term that describes the rate at which the gradient changes with height. The larger the gradient, the more gradual the density change with height, enabling customization of the gradient pattern; It describes the nonlinear law of density gradient change with height, the density difference between high-density mother liquor and low-density mother liquor; It is the density difference between the high-density mother liquor and the low-density mother liquor, which varies with height in a gradient liquid. The proportion of non-linear allocation.

8. The spheroidizing and sorting device for one-time molding of glass microspheres according to claim 1, characterized in that, The sphericity detection and sorting module includes an optical imaging unit, which is used to acquire high-resolution and distortion-free real-time images of glass microspheres after preliminary screening by the multi-level precision sorting module. Specifically, a high-brightness, low-flicker LED ring light source is used to eliminate reflective interference on the surface of the glass microspheres through diffuse reflection illumination, ensuring that the outline of the microspheres is clearly visible.

9. The spheroidizing and sorting device for one-time molding of glass microspheres according to claim 8, characterized in that, The image analysis and processing unit, through data interaction with the optical imaging unit, completes real-time processing of the microbead images, sphericity calculation, and qualification judgment. Its workflow is as follows: Sphericity calculation is based on microsphere profile data and uses the sphericity calculation formula: Sphericity = Surface area of ​​a sphere with the same volume as the microsphere / Actual surface area of ​​the microsphere. The calculation is performed quantitatively, and the maximum circumcircle, minimum incircle, and equivalent diameter of the microsphere profile are automatically identified by image analysis software. The volume and actual surface area of ​​the microsphere are then calculated to obtain the sphericity value.

10. The spheroidizing and sorting device for one-time molding of glass microspheres according to claim 9, characterized in that, Based on the judgment results of the image analysis and processing unit, the pneumatic sorting unit achieves rapid separation of qualified and unqualified microbeads. The specific operation is as follows: The qualified product channel and the unqualified product channel are set up below the conveyor track. When an unqualified signal is received, the unqualified microbeads are sent to the unqualified product channel, while the qualified microbeads fall naturally into the qualified product channel under the action of gravity, thus achieving efficient separation between the two.

Images