An intelligent detection method for spherical alumina quality

Abstract

The present invention relates to the technical field of spherical alumina quality detection technology, and in particular to an intelligent detection method for spherical alumina quality, comprising: collecting a microscopic image and chemical purity of a spherical alumina sample; outputting the sphericity and defect ratio of the spherical alumina sample; outputting a quality grade assessment result; performing a fluidity test on the spherical alumina sample, sequentially collecting, comparing and auditing image data of the spherical alumina sample, and calculating the flow velocity of the spherical alumina sample; determining a detection accuracy adjustment method based on the velocity variance of the flow velocity of several spherical alumina samples, including adjusting the image data comparison type or the number of audit intervals of the image data; wherein the shooting frequency of collecting the image data is determined based on the instantaneous velocity change of the spherical alumina sample collision. The present invention improves the accuracy of alumina detection.

Description

Technical Field

[0001] The present invention relates to the technical field of spherical alumina quality detection, and in particular to an intelligent detection method for spherical alumina quality. Background Art

[0002] In the existing technology, high-resolution industrial cameras are used to capture microscopic images, and image processing algorithms are used to calculate the area-to-perimeter ratio to determine the qualified spherical standard; by evaluating the roughness and measuring the specific surface area, it is suitable for high-performance application scenarios such as catalyst carriers.

[0003] Chinese Patent Publication No.: CN119580366A discloses an intelligent data analysis system and method based on alumina production, including: a parameter acquisition module, which acquires ball mill parameters, raw material parameters and spherical alumina impurity thresholds; identifies the ball mill parameters and raw material parameters according to a threshold prediction model to determine a standard threshold; a sample acquisition module, which processes the alumina raw material through a ball mill to obtain spherical alumina; detects the purity of the spherical alumina, and samples the spherical alumina by random sampling to obtain multiple samples; observes the samples through an electron microscope to obtain a morphology image set; an image segmentation module, which constructs a morphology image segmentation model to segment the morphology image set to obtain a first image set and a second image set; the first image set is the microscopic morphology of spherical alumina, and the second image set is the microscopic morphology of impurities; a first recognition module, which constructs a first image recognition model to perform segmentation on the first image set The system and method for intelligent data analysis of alumina production have the following problems: the spherical alumina parameter set includes the particle size data, sphericity data and smoothness coefficient of the spherical alumina particles; the first coefficient is calculated based on the spherical alumina parameter set; the second recognition module constructs a second image recognition model to identify the second image set, and obtains the impurity parameter set; the impurity parameter set includes the impurity particle size, impurity sphericity and impurity smoothness coefficient; the second coefficient is calculated based on the impurity parameter set; the comprehensive evaluation module calculates the quality coefficient through the first coefficient, the second coefficient, the spherical alumina impurity threshold and the spherical alumina purity. It can be seen that the system and method for intelligent data analysis of alumina production have the following problems: the image acquisition environment is unstable, resulting in inconsistent interference of the surrounding structures of the acquisition environment on the image acquisition, which leads to insufficient comprehensiveness of data acquisition; the data involved in modeling or the data involved in model updating are redundant, resulting in chaotic modeling and low efficiency; or the data has abnormal data that is missed during the modeling stage, resulting in reduced modeling accuracy. Summary of the Invention

[0004] To this end, the present invention provides an intelligent detection method for the quality of spherical alumina, which is used to overcome the problems in the prior art such as inconsistent interference of the surrounding structures of the acquisition environment on the image acquisition due to the instability of the image acquisition environment, resulting in insufficient comprehensiveness of data acquisition, and the modeling efficiency is low due to the redundancy of the data involved in modeling or the data involved in model updating, or the decrease in modeling accuracy due to the omission of abnormal data in the modeling stage.

[0005] To achieve the above object, the present invention provides an intelligent detection method for the quality of spherical alumina, comprising:

[0006] Collect microscopic images and chemical purity of spherical alumina samples;

[0007] Segmenting and comparing the microscopic images to obtain characteristic parameters of the spherical alumina sample to output the sphericity and defect ratio of the spherical alumina sample, wherein the characteristic parameters include the particle size and defect area of ​​the spherical alumina sample;

[0008] Inputting the sphericity, the defect ratio, and the chemical composition purity of the spherical alumina sample into a multimodal fusion model respectively, and outputting a quality grade evaluation result;

[0009] Performing a fluidity test on the spherical alumina sample, sequentially collecting, comparing and auditing image data of the spherical alumina sample, and calculating the flow velocity of the spherical alumina sample;

[0010] Determining a detection accuracy adjustment method based on a velocity variance of the flow velocities of the plurality of spherical alumina samples, including adjusting a comparison type of the image data, or adjusting a number of audit intervals of the image data based on a difference between an output grade of a model and an actual quality grade of the spherical alumina;

[0011] The shooting frequency of collecting the image data is determined according to the instantaneous velocity change of the spherical aluminum oxide sample collision.

[0012] Furthermore, the process of adjusting the contrast type of the image data includes:

[0013] Under the same conditions, the flow velocity of the same batch of spherical alumina samples on the inclined slide was collected several times;

[0014] calculating a velocity variance based on a number of said flow velocities;

[0015] If the speed variance is greater than or equal to the preset first variance, it is determined that the accuracy of the model establishment does not meet the requirements;

[0016] If the speed variance is greater than or equal to a preset second variance, it is determined that the comprehensiveness of the data collection does not meet the requirements, and the number of comparison types of the collected image data is increased;

[0017] The preset first variance is smaller than the preset second variance; and the number of contrast types of the collected image data is positively correlated with the speed variance.

[0018] Furthermore, the same conditions include that the volumes of the spherical alumina samples of the same batch undergoing the fluidity test are the same, the inclination angles of the inclined slideways are the same, and the ambient temperature of the fluidity test is the same.

[0019] Furthermore, the process of adjusting the number of audit intervals for image data includes:

[0020] If the speed variance is greater than or equal to the preset first variance and less than or equal to the preset second variance, it is preliminarily determined that the accuracy of the data involved in the modeling does not meet the requirements;

[0021] Obtain the output grade of the model and the actual quality grade of spherical alumina;

[0022] According to the difference between the output grade of the model and the actual quality grade of spherical alumina;

[0023] If the level difference is greater than the preset level difference, it is determined that the accuracy of the data involved in the modeling does not meet the requirements, and the number of audit intervals of the data is increased;

[0024] The number of audit intervals of the data is positively correlated with the level difference.

[0025] Furthermore, the grade difference is the absolute value of the difference between the output grade of the model and the actual quality grade of the spherical alumina.

[0026] Furthermore, the process of determining the shooting frequency of collecting the image data according to the instantaneous velocity change of the spherical aluminum oxide sample collision includes:

[0027] respectively obtaining a first displacement of the aluminum oxide that has collided within a unit fluidity test time before the collision and a second displacement of the aluminum oxide within a unit fluidity test time after the collision;

[0028] Calculating a change in the instantaneous velocity according to the first displacement and the second displacement;

[0029] If the variation of the instantaneous speed is greater than the preset variation, it is determined that the real-time performance of image acquisition does not meet the requirement, and the shooting frequency of acquiring the image data is increased.

[0030] Furthermore, the change in instantaneous speed is the difference between the instantaneous speed before the collision and the instantaneous speed after the collision.

[0031] Furthermore, the instantaneous rate before the collision is the ratio of the first displacement to the unit fluidity test duration; and the instantaneous rate after the collision is the ratio of the second displacement to the unit fluidity test duration.

[0032] Furthermore, the process of determining the unit liquidity test duration includes:

[0033] Obtaining the passing moments of the marked particle flows of the spherical aluminum oxide samples at the end of several inclined slides;

[0034] Calculating a standard deviation of time intervals between a plurality of said passing moments;

[0035] If the standard deviation of the time interval is greater than or equal to the preset first standard deviation and less than the preset second standard deviation, the unit liquidity test duration is set to the first-level duration;

[0036] If the standard deviation of the time interval is greater than or equal to the preset second standard deviation and less than the preset third standard deviation, the unit liquidity test duration is set to the secondary duration;

[0037] If the standard deviation of the time interval is greater than or equal to the preset third standard deviation, the unit liquidity test duration is set to the third level duration.

[0038] Furthermore, the first-level duration is greater than the second-level duration, and the second-level duration is greater than the third-level duration.

[0039] Compared with the prior art, the beneficial effect of the present invention lies in that the method of the present invention constructs an intelligent detection system for multimodal fusion; by cross-modal data fusion of high-resolution industrial cameras, laser scattering instruments and X-ray fluorescence spectroscopy, synchronous analysis of the morphological characteristics, particle size distribution and chemical composition of spherical alumina is achieved; the dynamic parameters of flow velocity are associated with the image acquisition frequency to achieve modeling, and the elastic deformation after particle collision causes the loss of key deformation frames, thereby resulting in a decrease in the surface defect detection rate, or a large number of repeated images are introduced when the fluidity is stable, which increases the computational burden and introduces noise, and the displacement measurement window is too long, resulting in the inability to capture the transient switching of particle stagnation-flow. The modeling accuracy is reduced due to the omission of abnormal data in the data modeling stage. The shooting frequency is adjusted in real time by the instantaneous velocity change before and after the collision, thereby improving the effectiveness of image acquisition, and by adjusting the number of audit intervals of the data, the misjudgment of a single indicator is reduced, thereby improving the accuracy of modeling, and by adjusting the comparison type of the image data, the comprehensiveness of data acquisition is improved.

[0040] Furthermore, the method of the present invention sets a first preset variance and a second preset variance. Since only bright field imaging is used, surface microcracks under dark field and grain boundary anomalies under polarized light are prone to missed detection. For example, different acquisition angles cause inconsistent interference of surrounding structures on image acquisition, and the missing features caused by missed detection increase the model misjudgment rate, which in turn leads to the inability to match the morphology and composition data of the same particle, and a decrease in the accuracy of multimodal fusion. By increasing the contrast types, the accuracy and comprehensiveness of image acquisition are improved.

[0041] Furthermore, the method of the present invention sets a preset level difference, and can reduce the problem of decreased accuracy of model output results due to misjudgment of a single indicator by increasing the number of audit intervals. This improves the accuracy of the model.

[0042] Furthermore, the method of the present invention sets a preset change amount, and the elastic deformation after particle collision causes the loss of key deformation frames, which leads to a decrease in the surface defect detection rate, or the introduction of a large number of repeated images when the fluidity is stable, which increases the computational burden and introduces noise, resulting in delays or errors in image acquisition, and thus leads to an iterative increase in the error amount of the participating data in the early stage of modeling. By increasing the shooting frequency, the modeling accuracy is increased.

[0043] Furthermore, the present invention determines the unit fluidity test duration, calculates the standard deviation of the particle flow time interval, and establishes a three-level duration determination method. Through adaptive timing control, the optimal sampling rate can be maintained when powder fluidity fluctuations are detected. By constructing a reverse duration adjustment mechanism, the sampling conditions are extended when the fluidity is poor to ensure data integrity and increase detection accuracy. When the fluidity is good, the duration is shortened to improve timeliness and reduce system power consumption. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 This is an overall flow chart of an intelligent detection method for spherical alumina quality according to an embodiment of the present invention;

[0045] Figure 2 This is a comparison type adjustment flow chart of an intelligent detection method for spherical alumina quality according to an embodiment of the present invention;

[0046] Figure 3 A flow chart of the number of audit intervals for adjusting image data in an intelligent detection method for spherical alumina quality according to an embodiment of the present invention;

[0047] Figure 4 This is a flow chart of determining the shooting frequency of an industrial camera in an intelligent detection method for spherical alumina quality according to an embodiment of the present invention. DETAILED DESCRIPTION

[0048] In order to make the objects and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are merely used to explain the present invention and are not intended to limit the present invention.

[0049] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0050] It should be noted that, in the description of the present invention, terms such as "up", "down", "left", "right", "inside", and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and does not indicate or imply that the device or element must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation on the present invention.

[0051] Furthermore, it should be noted that, in the description of the present invention, unless otherwise expressly specified or limited, the terms "mounted," "connected," and "connected" should be understood in a broad sense. For example, they may refer to fixed connections, detachable connections, or integral connections; mechanical connections or electrical connections; direct connections or indirect connections through an intermediate medium; and internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0052] See also Figure 1 、 Figure 2 、 Figure 3 as well as Figure 4 As shown in the figure, they are respectively an overall flow chart of the intelligent detection method for spherical alumina quality according to an embodiment of the present invention, a flow chart for adjusting the comparison type, a flow chart for adjusting the audit interval number of image data, and a flow chart for determining the shooting frequency of an industrial camera. An intelligent detection method for spherical alumina quality according to an embodiment of the present invention includes:

[0053] Collect microscopic images and chemical purity of spherical alumina samples;

[0054] Segmenting and comparing the microscopic images to obtain characteristic parameters of the spherical alumina sample to output the sphericity and defect ratio of the spherical alumina sample, wherein the characteristic parameters include the particle size and defect area of ​​the spherical alumina sample;

[0055] Inputting the sphericity, the defect ratio, and the chemical composition purity of the spherical alumina sample into a multimodal fusion model respectively, and outputting a quality grade evaluation result;

[0056] Performing a fluidity test on the spherical alumina sample, sequentially collecting, comparing and auditing image data of the spherical alumina sample, and calculating the flow velocity of the spherical alumina sample;

[0057] Determining a detection accuracy adjustment method based on a velocity variance of the flow velocities of the plurality of spherical alumina samples, including adjusting a comparison type of the image data, or adjusting a number of audit intervals of the image data based on a difference between an output grade of a model and an actual quality grade of the spherical alumina;

[0058] The shooting frequency of collecting the image data is determined according to the instantaneous velocity change of the spherical aluminum oxide sample collision.

[0059] Specifically, the spherical alumina samples are in a granular form.

[0060] Specifically, a high-resolution industrial camera is used to capture microscopic images of the spherical aluminum oxide sample. Examples of high-resolution industrial camera models include Photron FASTCAM Nova S12 and UC1400 industrial microscope cameras.

[0061] Specifically, the chemical composition purity of the spherical alumina sample is the molar mass content of alumina in the spherical alumina sample. The spherical alumina sample also contains silicon dioxide and iron oxide. The chemical composition purity of the spherical alumina sample is collected by an elemental analyzer.

[0062] Specifically, the microscopic images are segmented and compared to extract characteristic parameters of the spherical alumina sample, and the sphericity, surface roughness, and defect ratio of the spherical alumina sample are calculated based on a convolutional neural network model, including:

[0063] Acquire a microscopic image of a spherical alumina sample where a single particle occupies an area of ​​at least 50 × 50 pixels;

[0064] The U-Net segmentation network segments the microscopic image and compares it with the spherical aluminum oxide sample images in the historical database according to the comparison type to obtain the defect area;

[0065] The sphericity of the segmented microscopic image is calculated according to the geometric method, and a sphere is determined to be qualified when the sphericity is greater than or equal to 0.92;

[0066] Use Mask R-CNN to determine the defect area of ​​the segmented microscopic image,

[0067] Among them, the defect area includes the area with crack aspect ratio greater than or equal to 3, the area with hole circularity greater than or equal to 0.85, and the area with edge gradient greater than or equal to 15%; the defect ratio is the ratio of the total area of ​​the defect area to the area of ​​the spherical alumina sample in the microscopic image.

[0068] Specifically, the defective areas include spherical aluminum oxide missing and concave areas.

[0069] Specifically, the sphericity, defect ratio, and chemical purity of the spherical alumina sample are input into a multimodal fusion model, and a quality grade evaluation result is output, including:

[0070] Standardize the data of spherical alumina samples in terms of sphericity, defect ratio and chemical purity;

[0071] The weights of the sphericity, defect ratio, and chemical purity after data standardization are calculated using a two-layer neural network and the weighted average method is used:

[0072] The quality level is determined based on the calculated results.

[0073] In practice, the method of the present invention divides the quality levels into one to five levels.

[0074] The weighted fusion calculation result of the fifth quality level is greater than or equal to 0.95, corresponding to a sphericity greater than or equal to 0.93, a defect ratio less than or equal to 0.1%, and a purity greater than or equal to 99.5%;

[0075] The weighted fusion calculation result of the fourth-level quality grade is greater than or equal to 0.85, corresponding to a sphericity greater than or equal to 0.9, a defect ratio less than or equal to 0.3%, and a purity greater than or equal to 99%;

[0076] The weighted fusion calculation result of the third-level quality grade is greater than or equal to 0.7, corresponding to a sphericity greater than or equal to 0.85, a defect ratio less than or equal to 0.5%, and a purity greater than or equal to 98%;

[0077] The weighted fusion calculation result of quality grades below level three is less than 0.7, which is unqualified spherical alumina.

[0078] Those skilled in the art can adjust the quality grade classification standards based on actual test results.

[0079] In implementation, the method of the present invention constructs an intelligent detection system for multimodal fusion; by cross-modal data fusion of high-resolution industrial cameras, laser scattering instruments and X-ray fluorescence spectroscopy, synchronous analysis of the morphological characteristics, particle size distribution and chemical composition of spherical alumina is achieved; the dynamic parameters of flow velocity are associated with the image acquisition frequency to achieve modeling, and the elastic deformation after particle collision causes the loss of key deformation frames, thereby reducing the surface defect detection rate, or because a large number of repeated images are introduced when the fluidity is stable, which increases the computational burden and introduces noise, and the displacement measurement window is too long, resulting in the inability to capture the transient switching of particle stagnation-flow. The modeling accuracy is reduced due to the omission of abnormal data in the data modeling stage. The shooting frequency is adjusted in real time by the instantaneous velocity change before and after the collision, thereby improving the effectiveness of image acquisition, and by adjusting the number of audit intervals of the data, thereby reducing the misjudgment of a single indicator and improving the accuracy of modeling, and by adjusting the comparison type of the image data, the comprehensiveness of data acquisition is improved.

[0080] Specifically, the process of adjusting the contrast type of the image data includes:

[0081] Under the same conditions, the flow velocity of the same batch of spherical alumina samples on the inclined slide was collected several times;

[0082] calculating a velocity variance based on a number of said flow velocities;

[0083] If the speed variance is greater than or equal to the preset first variance, it is determined that the accuracy of the model establishment does not meet the requirements;

[0084] If the speed variance is greater than or equal to a preset second variance, it is determined that the comprehensiveness of the data collection does not meet the requirements, and the number of comparison types of the collected image data is increased;

[0085] The preset first variance is smaller than the preset second variance; and the number of contrast types of the collected image data is positively correlated with the speed variance.

[0086] Specifically, a number of industrial cameras are arranged at equal intervals on the edge of the inclined surface of the inclined slide.

[0087] Specifically, the spherical alumina samples from the same batch are a number of spherical alumina samples of equal volume sampled from a production batch of spherical alumina.

[0088] Specifically, the velocity variance is the variance of the instantaneous velocity of several spherical alumina samples of equal volume flowing on the same inclined slide at the middle position of the inclined surface of the inclined slide at one time. The calculation process of the variance is a technical means well known to those skilled in the art and will not be described in detail here.

[0089] Specifically, the instantaneous speed is acquired by an industrial camera set at the middle position of the inclined surface of the inclined slide.

[0090] Specifically, the same conditions include that the volumes of the spherical aluminum oxide samples of the same batch undergoing the fluidity test are the same, the inclination angles of the inclined slides are the same, and the ambient temperature of the fluidity test is the same.

[0091] Specifically, the same conditions also include the same vertical height of falling on the inclined slide and the same surface humidity on the inclined slide.

[0092] Specifically, the comparison types of image data include minimum particle size value, maximum particle size value, median particle size, cumulative particle size distribution, specific surface area, apparent density, and particle shape.

[0093] Specifically, under the condition that the inclination angle of the inclined slide is 10°, the general value range of the preset first variance is [0.4m² / s², 0.6m² / s²], and the general value range of the preset second variance is [1m² / s², 1.5m² / s²].

[0094] Preferably, the preferred embodiment of the preset first variance is 0.5 m² / s², and the preferred embodiment of the preset second variance is 1.2 m² / s².

[0095] In implementation, when the difference between the speed variance and the preset second variance is within 0.01 m² / s², the number of comparison types of the image data is increased by 4. If the difference between the speed variance and the preset second variance exceeds 0.01 m² / s², the increase in the number of comparison types and the difference between the speed variance and the preset second variance satisfy a linear relationship N=3×D(v), where N is the number of comparison types, and D(v) is the ratio of the difference between the speed variance and the preset second variance to 0.01 m² / s². For example, if the difference between the speed variance and the preset second variance is 0.03 m² / s², the current number of image comparison types is 20, and the number of image comparison types increases to 29.

[0096] In implementation, the method of the present invention sets a first preset variance and a second preset variance. Since only bright field imaging is used, surface microcracks under dark field and grain boundary anomalies under polarized light are prone to missed detection. For example, different acquisition angles cause inconsistent interference of surrounding structures on image acquisition, and the missing features caused by missed detection increase the model misjudgment rate, which in turn leads to the inability to match the morphology and composition data of the same particle, and a decrease in the accuracy of multimodal fusion. By increasing the contrast type, the accuracy and comprehensiveness of image acquisition are improved.

[0097] Specifically, the process of adjusting the number of audit intervals of image data includes:

[0098] If the speed variance is greater than or equal to the preset first variance and less than or equal to the preset second variance, it is preliminarily determined that the accuracy of the data involved in the modeling does not meet the requirements;

[0099] Obtain the output grade of the model and the actual quality grade of spherical alumina;

[0100] According to the difference between the output grade of the model and the actual quality grade of spherical alumina;

[0101] If the level difference is greater than the preset level difference, it is determined that the accuracy of the data involved in the modeling does not meet the requirements, and the number of audit intervals of the data is increased;

[0102] The number of audit intervals of the data is positively correlated with the level difference.

[0103] Specifically, the audit is to review whether the image data is correct, such as eliminating or marking redundant and erroneous image data through data cleaning and verification rules; the audit interval number is the number of images between the triggering of two audits, for example, an audit is performed after every 100 new images are added.

[0104] Specifically, the grade difference is the absolute value of the difference between the output grade of the model and the actual quality grade of the spherical alumina.

[0105] Specifically, under the condition that the pixel size of the microscopic image of the spherical aluminum oxide sample captured by a 12-bit industrial camera is 3.45 μm, the preset level difference is 1 level.

[0106] During implementation, if the difference between the level difference amount and the preset level difference amount is greater than or equal to 1 level, the number of audit intervals of the data will be increased according to the ratio of the difference between the level difference amount and the preset level difference amount to 0.5. For example, if the difference between the level difference amount and the preset level difference amount is 2 levels, the number of audit intervals of the current data is 6, and the number of audit intervals of the data is increased to 6 + 2 / 0.5 = 10.

[0107] In implementation, the method of the present invention sets a preset level difference, and the micro-displacement at the moment of collision, such as elastic deformation, cannot be accurately captured, and the component misdetection caused by XRF drift is reduced by increasing the number of audit intervals. This reduces the problem of decreased accuracy of the model output results caused by misjudgment of a single indicator, thereby achieving improved accuracy of the model.

[0108] Specifically, the process of determining the shooting frequency of collecting the image data according to the instantaneous velocity change of the spherical aluminum oxide sample collision includes:

[0109] respectively obtaining a first displacement of the aluminum oxide that has collided within a unit fluidity test time before the collision and a second displacement of the aluminum oxide within a unit fluidity test time after the collision;

[0110] Calculating a change in the instantaneous velocity according to the first displacement and the second displacement;

[0111] If the variation of the instantaneous speed is greater than the preset variation, it is determined that the real-time performance of image acquisition does not meet the requirement, and the shooting frequency of acquiring the image data is increased.

[0112] Specifically, the change in instantaneous speed is the difference between the instantaneous speed before the collision and the instantaneous speed after the collision.

[0113] Specifically, under the condition that the resolution of the industrial camera is 1 μm / px, the general value range of the preset change amount is [0.1 m / s, 0.5 m / s], and the preferred embodiment of the preset change amount is 0.3 m / s.

[0114] In implementation, when the difference between the change in instantaneous speed and the preset change is within 0.1m / s, the shooting frequency of the industrial camera will increase by 100fps. When the difference between the change in instantaneous speed and the preset change exceeds 0.1m / s, the shooting frequency of the industrial camera will increase by 120fps for every 0.1m / s increase.

[0115] Specifically, the instantaneous rate before the collision is the ratio of the first displacement to the unit fluidity test duration; the instantaneous rate after the collision is the ratio of the second displacement to the unit fluidity test duration.

[0116] In implementation, the method of the present invention sets a preset change amount. Due to the elastic deformation after particle collision, the key deformation frame is lost, which leads to a decrease in the surface defect detection rate, or a large number of repeated images are introduced when the fluidity is stable, which increases the calculation burden and introduces noise, resulting in delays or errors in image acquisition, and then leads to an iterative increase in the error amount of the participating data in the early stage of modeling. By increasing the shooting frequency, the modeling accuracy is increased.

[0117] Specifically, the process of determining the unit liquidity test duration includes:

[0118] Obtaining the passing moments of the marked particle flows of the spherical aluminum oxide samples at the end of several inclined slides;

[0119] Calculating a standard deviation of time intervals between a plurality of said passing moments;

[0120] If the standard deviation of the time interval is greater than or equal to the preset first standard deviation and less than the preset second standard deviation, the unit liquidity test duration is set to the first-level duration;

[0121] If the standard deviation of the time interval is greater than or equal to the preset second standard deviation and less than the preset third standard deviation, the unit liquidity test duration is set to the secondary duration;

[0122] If the standard deviation of the time interval is greater than or equal to the preset third standard deviation, the unit liquidity test duration is set to the third level duration.

[0123] Specifically, the labeled particle flow of the spherical alumina sample is a group of spherical alumina particles with a surface sprayed with a fluorescent dye or a magnetic coating, such as Fe3O4-coated spherical alumina particles.

[0124] Specifically, the first-level duration is greater than the second-level duration, and the second-level duration is greater than the third-level duration.

[0125] Specifically, the time interval of the passing moments is the interval between the adjacent marked particle flows of the spherical aluminum oxide samples that pass through the end of the inclined slide in sequence.

[0126] Specifically, under the condition that the inclination angle of the inclined slide is 10°, the general value range of the preset first standard deviation is [0.08s, 0.12s], the general value range of the preset second standard deviation is [0.25s, 0.4s], and the general value range of the preset third standard deviation is [0.48s, 0.55s]; the preferred embodiment of the preset first standard deviation is 0.1s, the preferred embodiment of the preset second standard deviation is 0.3s, and the preferred embodiment of the preset third standard deviation is 0.5s.

[0127] Specifically, under the condition that the inclination angle of the inclined slide is 10°, the general value range of the first-level time length is [100ms, 120ms], the general value range of the first-level time length is [40ms, 60ms], and the general value range of the first-level time length is [8ms, 30ms]; the preferred embodiment of the first-level time length is 100ms, the preferred embodiment of the second-level time length is 50ms, and the preferred embodiment of the third-level time length is 20ms.

[0128] In implementation, the present invention determines the unit fluidity test duration, calculates the standard deviation of the particle flow time interval, and establishes a three-level duration determination method. Through adaptive timing control, the optimal sampling rate can be maintained when powder fluidity fluctuations are detected. By constructing a reverse duration adjustment mechanism, the sampling conditions are extended when the fluidity is poor to ensure data integrity and increase detection accuracy. When the fluidity is good, the duration is shortened to improve timeliness and reduce system power consumption.

[0129] Thus far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art may make equivalent changes or substitutions to the relevant technical features, and the technical solutions after such changes or substitutions will fall within the scope of protection of the present invention.

Claims

1. An intelligent detection method for the quality of spherical alumina, characterized in that: include: Collect microscopic images and chemical purity of spherical alumina samples; Segmenting and comparing the microscopic images to obtain characteristic parameters of the spherical alumina sample to output the sphericity and defect ratio of the spherical alumina sample, wherein the characteristic parameters include the particle size and defect area of ​​the spherical alumina sample; Inputting the sphericity, the defect ratio, and the chemical composition purity of the spherical alumina sample into a multimodal fusion model respectively, and outputting a quality grade evaluation result; Performing a fluidity test on the spherical alumina sample, sequentially collecting, comparing and auditing image data of the spherical alumina sample, and calculating the flow velocity of the spherical alumina sample; Determining a detection accuracy adjustment method based on a velocity variance of the flow velocities of the plurality of spherical alumina samples, including adjusting a comparison type of the image data, or adjusting a number of audit intervals of the image data based on a difference between an output grade of a model and an actual quality grade of the spherical alumina; The shooting frequency of collecting the image data is determined according to the instantaneous velocity change of the spherical aluminum oxide sample collision; The process of adjusting the contrast type of the image data includes: Under the same conditions, the flow velocity of the same batch of spherical alumina samples on the inclined slide was collected several times; calculating a velocity variance based on a number of said flow velocities; If the speed variance is greater than or equal to the preset first variance, it is determined that the accuracy of the model establishment does not meet the requirements; If the speed variance is greater than or equal to a preset second variance, it is determined that the comprehensiveness of the data collection does not meet the requirements, and the number of comparison types of the collected image data is increased; Wherein, the preset first variance is smaller than the preset second variance; the number of contrast types of the collected image data is positively correlated with the speed variance; The process of adjusting the number of audit intervals for image data involves: If the speed variance is greater than or equal to the preset first variance and less than or equal to the preset second variance, it is preliminarily determined that the accuracy of the data involved in the modeling does not meet the requirements; Obtain the output grade of the model and the actual quality grade of spherical alumina; According to the difference between the output grade of the model and the actual quality grade of spherical alumina; If the level difference is greater than the preset level difference, it is determined that the accuracy of the data involved in the modeling does not meet the requirements, and the number of audit intervals of the data is increased; The number of audit intervals of the data is positively correlated with the level difference.

2. The intelligent detection method for spherical alumina quality according to claim 1, characterized in that: The same conditions include that the volumes of the spherical aluminum oxide samples of the same batch for the fluidity test are the same, the inclination angles of the inclined slides are the same, and the ambient temperature of the fluidity test is the same.

3. The intelligent detection method for spherical alumina quality according to claim 1, characterized in that: The grade difference is the absolute value of the difference between the output grade of the model and the actual quality grade of the spherical alumina.

4. The intelligent detection method for spherical alumina quality according to claim 1, characterized in that: The process of determining the shooting frequency of collecting the image data according to the instantaneous velocity change of the spherical aluminum oxide sample collision includes: respectively obtaining a first displacement of the aluminum oxide that has collided within a unit fluidity test time before the collision and a second displacement of the aluminum oxide within a unit fluidity test time after the collision; Calculating a change in the instantaneous velocity according to the first displacement and the second displacement; If the variation of the instantaneous speed is greater than the preset variation, it is determined that the real-time performance of image acquisition does not meet the requirement, and the shooting frequency of acquiring the image data is increased.

5. The intelligent detection method for spherical alumina quality according to claim 4, characterized in that: The change in instantaneous speed is the difference between the instantaneous speed before the collision and the instantaneous speed after the collision.

6. The intelligent detection method for spherical alumina quality according to claim 5, characterized in that: The instantaneous rate before the collision is the ratio of the first displacement to the unit fluidity test duration; the instantaneous rate after the collision is the ratio of the second displacement to the unit fluidity test duration.

7. The intelligent detection method for spherical alumina quality according to claim 6, characterized in that: The process of determining the unit liquidity test duration includes: Obtaining the passing moments of the marked particle flows of the spherical aluminum oxide samples at the end of several inclined slides; Calculating a standard deviation of time intervals between a plurality of said passing moments; If the standard deviation of the time interval is greater than or equal to the preset first standard deviation and less than the preset second standard deviation, the unit liquidity test duration is set to the first-level duration; If the standard deviation of the time interval is greater than or equal to the preset second standard deviation and less than the preset third standard deviation, the unit liquidity test duration is set to the secondary duration; If the standard deviation of the time interval is greater than or equal to the preset third standard deviation, the unit liquidity test duration is set to the third level duration.

8. The intelligent detection method for spherical alumina quality according to claim 7, characterized in that: The first-level duration is greater than the second-level duration, and the second-level duration is greater than the third-level duration.

Images