Coal slime flotation tailing ash content instrument based on three-channel image recognition and detection method

Abstract

The application discloses a coal slime flotation tailing ash content instrument based on three-channel image recognition and a detection method, and relates to the technical field of coal slime flotation tailing ash content instruments.The three-channel cyclone is used for separating tailings and outputting from three channels; the data acquisition card is used for collecting data and sending the collected data to a control chip; the control chip is configured with an ash content analysis model, and is used for inputting the collected data into the ash content analysis model to obtain an ash content analysis result corresponding to the tailings.

Description

Technical Field

[0001] This invention relates to the field of coal washing and beneficiation, specifically to a coal slime flotation tailings ash analyzer and detection method based on three-channel image recognition. Background Technology

[0002] Coal slime is a byproduct of coal washing, a semi-solid substance formed by mixing coal powder and water. Coal slime enters the mixing tank as a slurry, where reagents (foaming agents and collectors) are added and the mixture is thoroughly stirred. The stirred slurry then enters the flotation machine, where the rotating impeller generates strong agitation. Combined with aeration, this produces numerous bubbles of varying sizes. Hydrophobic coal particles, due to the adsorption of reagents (collectors), adhere to the bubbles and are carried to the surface of the slurry, aggregating into a mineralized froth layer. This layer is scraped off by a froth scraper as clean coal. Hydrophilic gangue particles do not react with the reagents and do not adhere to the bubbles, remaining in the slurry as flotation tailings. A coal slime flotation tailings ash analyzer can measure the ash content of the coal slime flotation tailings.

[0003] Existing tailings ash analysis devices often suffer from the following technical problems: First, existing tailings ash analyzers primarily utilize visible light to photograph water or dry samples, then use the image information to measure ash content. The accuracy of this method depends on the correlation between image information and ash content. However, in actual production processes, especially in multi-coal washing and beneficiation, it is often necessary to continuously adjust the coal blending ratio, leading to drastic changes in coal quality (such as the presence of black or other colored gangue). In such cases, the correlation between image information and ash content changes, ultimately resulting in a sharp decrease in the accuracy of ash content measurement results. Secondly, because the filming was done in the visible light band, large particles were not easily identified, making it impossible to determine whether the upstream equipment of the flotation was damaged (screen breakage), resulting in the leakage of coarse clean coal (coarse coal leakage). Third, in the process of recognition using three-channel images, the weighting of the three channels directly affects the accuracy of the ash content measurement results. Specifically, substances that undergo changes (such as black gangue) tend to accumulate in large quantities in one of the channels. If the weight of that channel is too high, the accuracy of the ash content measurement results will decrease sharply. Currently, there is a lack of effective solutions for setting reasonable weights. Summary of the Invention

[0004] The summary section of this invention provides a brief overview of the concepts, which will be described in detail in the detailed description section below. This summary section is not intended to identify key or essential features of the claimed technical solutions, nor is it intended to limit the scope of the claimed technical solutions.

[0005] This invention proposes a coal slime flotation tailings ash analyzer and detection method based on three-channel image recognition to solve one or more of the technical problems mentioned in the background section above.

[0006] This invention provides a coal slime flotation tailings ash analyzer based on three-channel image recognition, comprising: a three-channel hydrocyclone, used to separate tailings entering from the feed inlet into a first type of precipitate, a second type of precipitate, and a third type of precipitate, outputting the first type of precipitate from the first channel, the second type of precipitate from the second channel, and the third type of precipitate from the third channel; wherein, the first channel is equipped with a first flow sensor, a first image acquisition device, and a first concentration sensor; the second channel is equipped with a second flow sensor, a second image acquisition device, and a second concentration sensor; and the third channel is equipped with a third flow sensor, a third image acquisition device, and a third concentration sensor; and a data acquisition card, used to acquire the first flow rate measured by the first flow sensor, the first image acquired by the first image acquisition device, the first concentration measured by the first concentration sensor, and the second flow rate measured by the second flow sensor. The second image acquired by the second image acquisition device, the second concentration measured by the second concentration sensor, the third flow rate measured by the third flow sensor, the third image acquired by the third image acquisition device, the third concentration measured by the third concentration sensor, and the first flow rate, first image, first concentration, second flow rate, second image, second concentration, third flow rate, third image, and third concentration are sent to the control chip. The control chip is equipped with an ash analysis model and is used to extract features from the first image, second image, and third image respectively to obtain the first image features corresponding to the first image, the second image features corresponding to the second image, and the third image features corresponding to the third image. The first flow rate, first concentration, first image features, second flow rate, second concentration, second image features, third flow rate, third concentration, and third image features are input into the ash analysis model to obtain the ash analysis results corresponding to the tailings.

[0007] Optionally, the coal slime flotation tailings ash analyzer further includes: a first screen disposed below the first channel and a second screen disposed below the second channel, wherein the solid particles filtered by the first and second screens are transported to the collection container by a weighing electronic belt conveyor. The weighing electronic belt conveyor is used to send the weighing data of the solid particles to the control chip. The control chip is also used to determine the real-time flow rate of the solid particles based on the weighing data. When the real-time flow rate is greater than a preset flow rate threshold, a prompt message indicating damage to the upstream equipment is generated and sent to the maintenance terminal so that the maintenance personnel at the maintenance terminal can perform maintenance on the upstream equipment.

[0008] Optionally, since the product corresponding to the first channel is a high coal content substance, if the difference between the first concentration and the first flow rate of the first channel and the preset conventional value is greater than the preset threshold and both the first concentration and the first flow rate are greater than the preset threshold, a prompt message indicating damage to the upstream equipment can be generated and sent to the maintenance terminal so that the maintenance personnel corresponding to the maintenance terminal can carry out maintenance on the upstream equipment.

[0009] This invention provides a method for detecting ash content in coal slime flotation tailings based on three-channel image recognition, comprising: acquiring sample data from three channels of a three-channel hydrocyclone included in a coal slime flotation tailings ash analyzer at multiple time points within a historical time interval, obtaining a sample dataset, the sample data including a first sample flow rate, a first sample image, a first sample concentration, a second sample flow rate, a second sample image, a second sample concentration, a third sample flow rate, a third sample image, a third sample concentration, and sample ash content; extracting features from the first, second, and third sample images in each sample dataset, obtaining corresponding first sample image features, second sample image features, and third sample image features, and updating the sample data using the first, second, and third sample image features to generate an updated sample dataset. This data is used to obtain an updated sample dataset. An ash analysis model is constructed based on the Gaussian process regression algorithm and the updated sample dataset. Current measurement data is acquired, including first real-time flow rate, first real-time image, first real-time concentration, second real-time flow rate, second real-time image, second real-time concentration, third real-time flow rate, third real-time image, and third real-time concentration. Feature extraction is performed on the first, second, and third real-time images to obtain first, second, and third real-time image features. These first, second, and third real-time image features are then used to replace the first, second, and third real-time images in the current measurement data, resulting in updated measurement data. The updated measurement data is then input into the ash analysis model to obtain the ash analysis results.

[0010] Optionally, the ash analysis model includes a first solid component calculation network, a second solid component calculation network, a third solid component calculation network, a first feature fusion network, a second feature fusion network, a third feature fusion network, a first prediction network, a second prediction network, a third prediction network, and a fourth prediction network. The first solid component calculation network takes a first real-time flow rate and a first real-time concentration as input and outputs a first real-time solid component; the second solid component calculation network takes a second real-time flow rate and a second real-time concentration as input and outputs a second real-time solid component; and the third solid component calculation network takes a third real-time flow rate and a third real-time concentration as input and outputs a third real-time solid component. The feature fusion network takes the first real-time solid component as input and the first real-time image features as input and outputs the first fused feature. The second feature fusion network takes the second real-time solid component and the second real-time image features as input and outputs the second fused feature. The first prediction network takes the first fused feature as input and outputs the first gray analysis result. The second prediction network takes the second fused feature and outputs the second gray analysis result. The third prediction network takes the third fused feature as input and outputs the third gray analysis result. The fourth prediction network takes the first gray analysis result, the second gray analysis result, and the third gray analysis result as input and outputs the gray analysis result.

[0011] Optionally, before inputting the updated measurement data into the ash analysis model to obtain the ash analysis results, the method further includes: post-processing the updated measurement data to obtain processed measurement data; and inputting the updated measurement data into the ash analysis model to obtain the ash analysis results, including: inputting the processed measurement data into the ash analysis model to obtain the ash analysis results.

[0012] Optionally, the updated measurement data is post-processed to obtain processed measurement data, including: obtaining historical updated measurement data corresponding to a time point before the current time point corresponding to the updated measurement data, and using the historical updated measurement data to predict the updated measurement data at the current time point to obtain predicted updated measurement data; determining the new error based on the error covariance and process noise corresponding to a time point before the current time point; determining the gain based on the new error and the parameter matrix of the coal slime flotation tailings ash analyzer; and generating processed measurement data using the gain, parameter matrix, updated measurement data, and predicted updated measurement data.

[0013] Optionally, the fourth prediction network is used to weight the first gray analysis result, the second gray analysis result, and the third gray analysis result to obtain a weighted result, and the weighted result is determined as the gray analysis result. The weighting weights are determined by: determining the confidence levels of the first gray analysis result, the second gray analysis result, and the third gray analysis result respectively; determining the first image and the second image respectively; and determining the weighting weights based on the confidence levels and image scores.

[0014] The present invention has the following beneficial effects: 1. First, a three-channel hydrocyclone is used to separate the tailings. This separation utilizes the density difference between coal and gangue in the tailings to classify the tailings slurry, which is then output through three channels. The three channels yield materials with high coal content, high ash content, and suspended matter content, respectively. The high coal content material contains coal with a coal content greater than or equal to 80%, and the high ash content material contains ash with a ash content greater than or equal to 80%. Based on this, compared to existing technologies that use images of unsorted tailings for ash content measurement, the separation process causes altered substances (such as black gangue) to accumulate in large quantities in the product of one channel, while the characteristics (such as color) of the products in the other two channels remain largely unchanged. Therefore, by collecting and integrating data from the products of the three channels, the influence of changing substances on the ash analysis results can be reduced, and the influence of fluctuations caused by coal quality changes on the measurement can be eliminated through comprehensive judgment, thereby improving the prediction accuracy of ash analysis results. In addition, by introducing the flow rate and (solid mass) concentration of each channel on the basis of sorting, the types and amount of data information are increased, further improving the prediction accuracy of ash analysis results. 2. It can accurately identify coarse coal leakage in the flotation system and generate prompts to remind maintenance personnel to troubleshoot the problem in a timely manner, thus preventing leakage of coarse clean coal particles. 3. By comparing the distance between image features and average image features, the images of channels with large changes can be identified, indicating that the image of the substance in that channel has changed drastically. Therefore, the weight of the image of that channel is reduced to avoid conflict with the mapping relationship established in the stable state, which would lead to a decrease in prediction accuracy. Correspondingly, the weights of the images of the other two channels are increased, thereby improving the prediction accuracy by setting the weights reasonably. Attached Figure Description

[0015] The above and other features, advantages, and aspects of the various embodiments of the present invention will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and elements are not necessarily drawn to scale.

[0016] Figure 1 This is a linear relationship diagram between ash content and slurry image color in existing technologies; Figure 2 This is a schematic diagram of the structure of the three-channel hydrocyclone included in the coal slime flotation tailings ash analyzer based on three-channel image recognition of the present invention. Figure 3 This is a schematic diagram of the structure of the coal slime flotation tailings ash analyzer based on three-channel image recognition of the present invention; Figure 4This is a flowchart of the coal slime flotation tailings ash content detection method based on three-channel image recognition of the present invention. Detailed Implementation

[0017] The invention will now be described in more detail with reference to the accompanying drawings. While some embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the invention. It should be understood that the drawings and embodiments of the invention are for illustrative purposes only and are not intended to limit the scope of protection of the invention.

[0018] It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described herein can be combined with each other.

[0019] It should be noted that the concepts of "first" and "second" mentioned in this invention are only used to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0020] It should be noted that the terms "a" and "a plurality of" used in this invention are illustrative rather than restrictive. Those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0021] The names of messages or information exchanged between the various devices of this invention are for illustrative purposes only and are not intended to limit the scope of these messages or information.

[0022] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0023] To better illustrate this invention, the working principle of existing tailings ash analyzers is first introduced: Under normal conditions, the size of solids in flotation tailings slurry should be less than 0.25 mm, including coal and gangue. Generally, after the flotation system is started, without changing the coal quality, reagent concentration, feed rate, feed concentration, and other production conditions, the color (grayscale) of the (tailings) slurry image has a certain linear correlation with the ash content; that is, the lower the ash content, the darker the slurry color, and the higher the ash content, the lighter the slurry color. Based on this, existing tailings ash analyzers can be fitted using a linear regression model to obtain the linear relationship between ash content (Ash) and slurry image color (grayscale), Ash = k*th + b, as shown below. Figure 1 As shown, th represents the grayscale of the slurry image. The slurry composition under stable production conditions can be analyzed and mathematically processed to obtain the values ​​of k and b, thus enabling the determination of tailings ash content based on the color of the slurry image.

[0024] However, when the coal quality changes, such as the presence of black or other colored gangue, the parameters k and b, determined under the condition of constant coal quality, cannot accurately reflect the changes in ash content during the production process. Therefore, in such cases, the accuracy of existing tailings ash analyzers for ash content measurement decreases sharply.

[0025] Based on this, the present invention provides a coal slime flotation tailings ash analyzer based on three-channel image recognition, specifically including a three-channel hydrocyclone, a data acquisition card, and a control chip. The three-channel hydrocyclone (also called a bipolar hydrocyclone) is a separation device that separates particles based on density distribution. Figure 2 The diagram shows a schematic of a three-channel hydrocyclone. The three-channel hydrocyclone includes an inlet 1, a first-stage cylinder 2, a first-stage compound cone 3, a sleeve 4, a side overflow pipe 5, a first-stage overflow pipe 6, a side overflow connecting section 7, a second-stage overflow pipe 8, a second-stage cylinder 9, and a second-stage long cone 10.

[0026] like Figure 3 As shown, the coal slime flotation tailings ash analyzer based on three-channel image recognition includes: a three-channel hydrocyclone 301, a data acquisition card 302, and a control chip 303. The three-channel hydrocyclone 301 is used to separate the tailings entering from the feed inlet into three types of sediment: a first type (high coal content material), a second type (high ash content material), and a third type (floating matter material). The first type of sediment is output from the first channel, the second type from the second channel, and the third type from the third channel. The first channel is equipped with a first flow sensor, a first image acquisition device, and a first concentration sensor; the second channel is equipped with a second flow sensor, a second image acquisition device, and a second concentration sensor; and the third channel is equipped with a third flow sensor, a third image acquisition device, and a third concentration sensor.

[0027] The data acquisition card 302 is used to acquire data from various channels, including: the first flow rate measured by the first flow sensor, the first image acquired by the first image acquisition device, the first concentration measured by the first concentration sensor, the second flow rate measured by the second flow sensor, the second image acquired by the second image acquisition device, the second concentration measured by the second concentration sensor, the third flow rate measured by the third flow sensor, the third image acquired by the third image acquisition device, the third concentration measured by the third concentration sensor, and to send the first flow rate, the first image, the first concentration, the second flow rate, the second image, the second concentration, the third flow rate, the third image, and the third concentration to the control chip 303.

[0028] The control chip 303 is equipped with an ash analysis model. The control chip is used to extract features from the first image, the second image, and the third image respectively to obtain the first image features corresponding to the first image, the second image features corresponding to the second image, and the third image features corresponding to the third image. The first flow rate, the first concentration, the first image features, the second flow rate, the second concentration, the second image features, the third flow rate, the third concentration, and the third image features are input into the ash analysis model to obtain the ash analysis results corresponding to the tailings.

[0029] As an example, features can be extracted from the first, second, and third images using a convolutional neural network to obtain the features of the first, second, and third images, respectively. As another example, the grayscale analysis results can be constructed based on a Gaussian process regression algorithm.

[0030] Optionally, the ash analysis model includes a first solid component calculation network, a second solid component calculation network, a third solid component calculation network, a first feature fusion network, a second feature fusion network, a third feature fusion network, a first prediction network, a second prediction network, a third prediction network, and a fourth prediction network. The first solid component calculation network takes a first real-time flow rate and a first real-time concentration as input and outputs a first real-time solid component; the second solid component calculation network takes a second real-time flow rate and a second real-time concentration as input and outputs a second real-time solid component; and the third solid component calculation network takes a third real-time flow rate and a third real-time concentration as input and outputs a third real-time solid component. The feature fusion network takes the first real-time solid component as input and the first real-time image features as input and outputs the first fused feature. The second feature fusion network takes the second real-time solid component and the second real-time image features as input and outputs the second fused feature. The first prediction network takes the first fused feature as input and outputs the first gray analysis result. The second prediction network takes the second fused feature and outputs the second gray analysis result. The third prediction network takes the third fused feature as input and outputs the third gray analysis result. The fourth prediction network takes the first gray analysis result, the second gray analysis result, and the third gray analysis result as input and outputs the gray analysis result.

[0031] In these optional implementations, the network structure and parameters of each sub-network included in the gray analysis model can be adjusted as needed. For example, fully connected networks, recurrent neural networks, residual networks, etc., can be used, and nonlinear activation functions can also be included as needed. Based on this, the model is trained using a sample dataset and deep learning algorithms (including backpropagation and stochastic gradient descent) until the training cutoff condition is met (e.g., the number of iterations reaches a preset number), thus obtaining the gray analysis model. Specifically, the sample data in the sample dataset includes first sample flow rate, first sample image, first sample concentration, second sample flow rate, second sample image, second sample concentration, third sample flow rate, third sample image, third sample concentration, and sample gray value. Based on this, the first sample flow rate, first sample image, first sample concentration, second sample flow rate, second sample image, second sample concentration, third sample flow rate, third sample image, and third sample concentration are used as outputs, and the sample gray value is used as the expected output. The difference between the expected output and the actual output is calculated, and the difference is fed back into the gray analysis model to adjust the parameters of each layer, thus completing one iteration. Similarly, multiple iterations can be performed to obtain the gray analysis model.

[0032] In some embodiments, a three-channel hydrocyclone is first used to separate the tailings, thereby utilizing the density difference between coal and gangue in the tailings to separate the tailings slurry and output it from three channels. The three channels respectively yield high coal content, high ash content, and floating matter content, wherein the high coal content material has a coal content greater than or equal to 80%, and the high ash content material has an ash content greater than or equal to 80%. Based on this, compared with the prior art of using unsorted tailings images for ash content measurement, because of the sorting, substances that have changed in the tailings (such as black gangue) will accumulate in large quantities in the product of one channel, while the characteristics (such as color) of the products in the other two channels remain unchanged. Therefore, by collecting and integrating data from the products of the three channels, the influence of changed substances on the ash analysis results can be reduced, and the influence of fluctuations caused by changes in coal quality on the measurement can be eliminated through comprehensive judgment, thereby improving the prediction accuracy of ash analysis results. In addition, by introducing the flow rate and (solid mass) concentration of each channel on the basis of sorting, the variety and amount of data information are increased, further improving the prediction accuracy of ash analysis results.

[0033] In some embodiments, to further address the technical problem described in the background section, namely, "because the image is taken in the visible light band, large particles are not easily identified, making it impossible to determine whether the upstream equipment of the flotation is damaged (screen breakage), resulting in leakage of coarse clean coal (coarse coal runoff)," in some embodiments of the present invention, the coal slime flotation tailings ash analyzer further includes: A first screen is installed below the first channel, and a second screen is installed below the second channel. Solid particles filtered by the first and second screens are transported to a collection container by a weighing electronic belt conveyor. The weighing electronic belt conveyor is used to send the weighing data of the solid particles to a control chip. The control chip is also used to determine the real-time flow rate of the solid particles based on the weighing data. When the real-time flow rate is greater than a preset flow rate threshold, a prompt message indicating damage to the upstream equipment is generated and sent to the maintenance terminal so that the maintenance personnel at the maintenance terminal can perform maintenance on the upstream equipment.

[0034] The first and second screens are 200-mesh screens, capable of filtering out solid particles larger than 200 mesh. The filtered solid particles are then transported to a collection container via a weighing electronic belt conveyor, thus recovering the coarse coal particles. In addition, the weighing electronic belt conveyor measures the weighing data and sends it to a control chip. The control chip then determines the real-time flow rate of the solid particles based on the weighing data and speed signal. When the real-time flow rate exceeds a preset flow threshold, it is considered that upstream equipment has malfunctioned, causing leakage of coarse coal particles. This generates a warning message indicating upstream equipment malfunction and sends it to the maintenance terminal, allowing the corresponding maintenance personnel to inspect and repair the upstream equipment.

[0035] In these embodiments, the output products of the first and second channels are filtered by a first screen and a second screen, thereby promptly detecting the leakage of coarse coal particles and eliminating system malfunctions in a timely manner.

[0036] Furthermore, such as Figure 4 The diagram shows a flowchart of the coal slime flotation tailings ash content detection method based on three-channel image recognition according to the present invention. The coal slime flotation tailings ash content detection method based on three-channel image recognition includes the following steps: Step 401: Obtain sample data from the three channels of the three-channel hydrocyclone included in the coal slime flotation tailings ash analyzer at multiple time points within the historical time interval, and obtain a sample dataset. The sample data includes the first sample flow rate, the first sample image, the first sample concentration, the second sample flow rate, the second sample image, the second sample concentration, the third sample flow rate, the third sample image, the third sample concentration, and the sample ash content.

[0037] Step 402: For each sample data, feature extraction is performed on the first sample image, the second sample image, and the third sample image to obtain the corresponding first sample image features, second sample image features, and third sample image features. The sample data is then updated using the first sample image features, second sample image features, and third sample image features to generate updated sample data, thus obtaining the updated sample dataset. Specifically, the original first sample image, second sample image, and third sample image can be replaced with the first sample image features, second sample image features, and third sample image features to obtain the updated sample data.

[0038] Step 403: Construct a grey analysis model based on the Gaussian process regression algorithm and the updated sample dataset. Gaussian process regression (GPR) is a nonparametric model that uses Gaussian process (GP) priors to perform regression analysis on the data.

[0039] Optionally, the ash analysis model can also be a neural network model, specifically including a first solid component calculation network, a second solid component calculation network, a third solid component calculation network, a first feature fusion network, a second feature fusion network, a third feature fusion network, a first prediction network, a second prediction network, a third prediction network, and a fourth prediction network. The first solid component calculation network takes a first real-time flow rate and a first real-time concentration as input and outputs a first real-time solid component; the second solid component calculation network takes a second real-time flow rate and a second real-time concentration as input and outputs a second real-time solid component; and the third solid component calculation network takes a third real-time flow rate and a third real-time concentration as input and outputs a third real-time solid component. The system uses real-time solid components as input and outputs a first fused feature network. The second feature fusion network uses the first real-time solid components and the first real-time image features as input and outputs a second fused feature network. The first prediction network uses the first fused feature as input and outputs a first gray analysis result. The second prediction network uses the second fused feature and outputs a second gray analysis result. The third prediction network uses the third fused feature as input and outputs a third gray analysis result. The fourth prediction network uses the first gray analysis result, the second gray analysis result, and the third gray analysis result as input and outputs a gray analysis result.

[0040] Optionally, the fourth prediction network is used to weight the first, second, and third gray analysis results to obtain a weighted result, and this weighted result is determined as the gray analysis result. The weighting weights can be determined manually or through the following methods: determining the confidence levels of the first, second, and third gray analysis results respectively; determining the image scores of the first, second, and third images respectively; and determining the weighting weights based on the confidence levels and image scores. Specifically, the weighting weights can be determined based on the confidence levels and image scores using a preset calculation formula. The higher the confidence level and image score, the greater the corresponding weight, and vice versa.

[0041] Step 404: Obtain the current measurement data, which includes the first real-time flow rate, the first real-time image, the first real-time concentration, the second real-time flow rate, the second real-time image, the second real-time concentration, the third real-time flow rate, the third real-time image, and the third real-time concentration. Extract features from the first real-time image, the second real-time image, and the third real-time image to obtain the first real-time image features, the second real-time image features, and the third real-time image features. Replace the first real-time image, the second real-time image, and the third real-time image in the current measurement data with the first real-time image features, the second real-time image features, and the third real-time image features to obtain updated measurement data.

[0042] Step 405: Input the updated measurement data into the ash analysis model to obtain the ash analysis results.

[0043] Optionally, before inputting the updated measurement data into the ash analysis model to obtain the ash analysis results, the method further includes: post-processing the updated measurement data to obtain processed measurement data; and inputting the updated measurement data into the ash analysis model to obtain the ash analysis results, including: inputting the processed measurement data into the ash analysis model to obtain the ash analysis results.

[0044] Optionally, the updated measurement data is post-processed to obtain processed measurement data. This includes: acquiring historical updated measurement data corresponding to a time point prior to the current time point corresponding to the updated measurement data, and using the historical updated measurement data to predict the updated measurement data at the current time point, thus obtaining predicted updated measurement data; determining the new error based on the error covariance and process noise corresponding to a time point prior to the current time point; determining the gain based on the new error and the parameter matrix of the coal slime flotation tailings ash analyzer; and generating processed measurement data using the gain, parameter matrix, updated measurement data, and predicted updated measurement data. This prevents drastic fluctuations in the measurement data.

[0045] In these embodiments, compared with the prior art of using unsorted tailings images for ash content measurement, by collecting and integrating data from the products of the three channels, the influence of changed substances on the ash content analysis results can be reduced, and the influence of fluctuations caused by coal quality changes on the measurement can be eliminated through comprehensive judgment, thereby improving the prediction accuracy of ash content analysis results. In addition, by introducing the flow rate and (solid mass) concentration of each channel on the basis of sorting, the types and amount of data information are increased, further improving the prediction accuracy of ash content analysis results.

[0046] In some embodiments, to further address the third technical problem described in the background section, namely, "In the process of recognition using three-channel images, how the weights of the three channels are set directly affects the accuracy of the ash content measurement results. Specifically, substances that have changed (e.g., black gangue) tend to accumulate in large quantities in one of the channels. If the weight of that channel is too large, the accuracy of the ash content measurement results will decrease sharply. Currently, there is a lack of effective solutions to set reasonable weights," in some embodiments of the present invention, the weighting weights involved in the fourth prediction network can be determined in the following way: Step 1: Determine the confidence levels of the first ash analysis result, the second ash analysis result, and the third ash analysis result, respectively; Step two involves determining the image scores for the first, second, and third images, respectively. These images can be any images acquired at any point during the actual measurement. Generally, the weights can be periodically redefined; when a re-determination of weighting is needed, the weights can be determined using the most recently acquired images from the three channels.

[0047] Step 3: Obtain the first historical image set corresponding to the first channel, the second historical image set corresponding to the second channel, and the third historical image set corresponding to the third channel. For the first historical image set, determine the corresponding first average image feature; for the second historical image set, determine the corresponding second average image feature; and for the third historical image set, determine the corresponding third average image feature. Specifically, the first image feature can be determined for each image in the first historical image set, and then the average of the first image features corresponding to each image can be calculated to obtain the first average image feature. Similarly, the second and third average image features can be obtained.

[0048] Step 4: Determine the first feature, second feature, and third feature corresponding to the first image, second image, and third image, respectively; determine the first distance between the first feature and the first average image feature; the second distance between the second feature and the second average image feature; and the third distance between the third feature and the third average image feature. Sort the first distance, second distance, and third distance and determine the largest of the three as the target distance. Determine the image corresponding to the target distance as the target image. Step 5: Set the weight corresponding to the target image to the first preset weight value, which is the lower limit of the weight.

[0049] Step 6: Determine the weighted weights of the two images other than the target image based on the confidence level and image score.

[0050] Specifically, weighted weights can be determined based on confidence level and image score using a preset calculation formula. Higher confidence level and image score correspond to higher weights, and vice versa. For example, the weight of an image can be obtained by multiplying the confidence level and image score by a preset coefficient. The weight of the final image can be obtained by subtracting the known weights of the two channels. In practice, the channel weights can be considered equivalent to the image weights.

[0051] In these implementations, by comparing the distance between image features and average image features, the images of channels with large changes can be identified, indicating that the image of the substance in that channel has changed drastically. As a result, the weight of the image of that channel is reduced to avoid conflict with the mapping relationship established in the stable state, which would lead to a decrease in prediction accuracy. Correspondingly, the weights of the images of the other two channels are increased, thereby improving the prediction accuracy by setting the weights reasonably.

[0052] The above description is merely a selection of preferred embodiments of the present invention and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention is not limited to specific combinations of the above-described technical features, but also includes other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in this invention.

Claims

1. A coal slime flotation tailings ash analyzer based on three-channel image recognition, characterized in that, include: The three-channel hydrocyclone is used to separate tailings entering from the feed inlet into three types of sediments: a first type, a second type, and a third type. The first type of sediment is output from the first channel, the second type from the second channel, and the third type from the third channel. The three channels respectively produce high-coal-content substances, high-ash-content substances, and floating matter-content substances. The high-coal-content substances have a coal content greater than or equal to 80%, and the high-ash-content substances have an ash content greater than or equal to 80%. Substances that have undergone changes in the tailings tend to accumulate in large quantities in the product of one channel, while the characteristics of the products in the other two channels remain largely unchanged. The first channel is equipped with a first flow sensor, a first image acquisition device, and a first concentration sensor; the second channel is equipped with a second flow sensor, a second image acquisition device, and a second concentration sensor; and the third channel is equipped with a third flow sensor, a third image acquisition device, and a third concentration sensor. A data acquisition card is used to acquire a first flow rate measured by a first flow sensor, a first image acquired by a first image acquisition device, a first concentration measured by a first concentration sensor, a second flow rate measured by a second flow sensor, a second image acquired by a second image acquisition device, a second concentration measured by a second concentration sensor, a third flow rate measured by a third flow sensor, a third image acquired by a third image acquisition device, and a third concentration measured by a third concentration sensor, and to send the first flow rate, the first image, the first concentration, the second flow rate, the second image, the second concentration, the third flow rate, the third image, and the third concentration to a control chip; The control chip is equipped with an ash analysis model. The control chip is used to extract features from the first image, the second image, and the third image respectively to obtain the first image features corresponding to the first image, the second image features corresponding to the second image, and the third image features corresponding to the third image. The first flow rate, the first concentration, the first image features, the second flow rate, the second concentration, the second image features, the third flow rate, the third concentration, and the third image features are input into the ash analysis model to obtain the first ash analysis result, the second ash analysis result, and the third ash analysis result corresponding to the tailings. The first ash analysis result, the second ash analysis result, and the third ash analysis result are weighted to obtain a weighted result, and the weighted result is determined as the ash analysis result. The weighting is determined in the following way: The confidence levels of the first ash analysis result, the second ash analysis result, and the third ash analysis result are determined respectively. Determine the image score values ​​for the first image, the second image, and the third image respectively; The weighting weights are determined based on the confidence level and the image score.

2. The coal slime flotation tailings ash analyzer based on three-channel image recognition according to claim 1, characterized in that, The coal slime flotation tailings ash analyzer also includes: A first screen and a second screen are respectively set below the first channel and below the second channel. Solid particles filtered by the first screen and the second screen are transported to a collection container by a weighing electronic belt conveyor. The weighing electronic belt conveyor is used to send the weighing data of the solid particles to the control chip. The control chip is also used to determine the real-time flow rate of the solid particles based on the weighing data. When the real-time flow rate is greater than a preset flow rate threshold, a prompt message indicating damage to the upstream equipment is generated and sent to the maintenance terminal so that the maintenance personnel corresponding to the maintenance terminal can perform maintenance on the upstream equipment.

3. A method for detecting ash content in coal slime flotation tailings based on three-channel image recognition, applied to the coal slime flotation tailings ash analyzer as described in claim 1, comprising: The sample data of the three channels of the three-channel hydrocyclone included in the coal slime flotation tailings ash analyzer are obtained at multiple time points within a historical time interval to obtain a sample dataset. The sample data includes the first sample flow rate, the first sample image, the first sample concentration, the second sample flow rate, the second sample image, the second sample concentration, the third sample flow rate, the third sample image, the third sample concentration, and the sample ash content. For each sample data, feature extraction is performed on the first sample image, the second sample image, and the third sample image to obtain the corresponding first sample image features, second sample image features, and third sample image features. The sample data is then updated using the first sample image features, second sample image features, and third sample image features to generate updated sample data, thus obtaining the updated sample dataset. A grey analysis model is constructed based on the Gaussian process regression algorithm and the updated sample dataset. Acquire current measurement data, which includes a first real-time flow rate, a first real-time image, a first real-time concentration, a second real-time flow rate, a second real-time image, a second real-time concentration, a third real-time flow rate, a third real-time image, and a third real-time concentration. Extract features from the first real-time image, the second real-time image, and the third real-time image to obtain first real-time image features, second real-time image features, and third real-time image features. Replace the first real-time image, the second real-time image, and the third real-time image in the current measurement data with the first real-time image features, the second real-time image features, and the third real-time image features to obtain updated measurement data. The updated measurement data is input into the ash analysis model to obtain the ash analysis results.

4. The method for detecting ash content in coal slime flotation tailings based on three-channel image recognition according to claim 3, characterized in that: The ash analysis model includes a first solid component calculation network, a second solid component calculation network, a third solid component calculation network, a first feature fusion network, a second feature fusion network, a third feature fusion network, a first prediction network, a second prediction network, a third prediction network, and a fourth prediction network. The first solid component calculation network takes the first real-time flow rate and the first real-time concentration as input and outputs the first real-time solid component. The second solid component calculation network takes the second real-time flow rate and the second real-time concentration as input and outputs the second real-time solid component. The third solid component calculation network takes the third real-time flow rate and the third real-time concentration as input and outputs the third real-time solid component. The first feature fusion network takes the first real-time solid component and the first real-time image features as input and outputs a first fused feature. The second feature fusion network takes the second real-time solid component and the second real-time image features as input and outputs a second fused feature. The first prediction network takes the first fused feature as input and outputs a first ash analysis result. The second prediction network takes the second fused feature as input and outputs a second ash analysis result. The third prediction network takes a third fused feature as input and outputs a third ash analysis result. The fourth prediction network takes the first ash analysis result, the second ash analysis result, and the third ash analysis result as input and outputs the ash analysis result.

5. The method for detecting ash content in coal slime flotation tailings based on three-channel image recognition according to claim 4, characterized in that, Before inputting the updated measurement data into the ash analysis model to obtain the ash analysis results, the method further includes: The updated measurement data is post-processed to obtain processed measurement data; and The step of inputting the updated measurement data into the ash analysis model to obtain the ash analysis results includes: The processed measurement data is input into the ash analysis model to obtain the ash analysis results.

6. The method for detecting ash content in coal slime flotation tailings based on three-channel image recognition according to claim 5, characterized in that, The post-processing of the updated measurement data to obtain processed measurement data includes: Obtain the historical updated measurement data corresponding to a time point before the current time point corresponding to the updated measurement data, and use the historical updated measurement data to predict the updated measurement data at the current time point to obtain the predicted updated measurement data; The new error is determined based on the error covariance and process noise corresponding to a time point prior to the current time point; The gain is determined based on the new error and the parameter matrix of the coal slime flotation tailings ash analyzer; The processing measurement data is generated using the gain, the parameter matrix, the updated measurement data, and the predicted updated measurement data.

7. The method for detecting ash content in coal slime flotation tailings based on three-channel image recognition according to claim 6, characterized in that, The fourth prediction network is used to weight the first ash analysis result, the second ash analysis result, and the third ash analysis result to obtain a weighted result, and the weighted result is determined as the ash analysis result.

Images