A non-contact full tailings paste stirring uniformity monitoring method and system

Abstract

The present invention provides a non-contact method and system for monitoring the mixing uniformity of tailings paste, belonging to the field of mine filling technology. The method comprises: continuously collecting paste surface images during paste preparation in a continuous mixer; processing the collected paste surface images using a deep learning method to segment non-paste areas and uneven paste areas; and determining the paste mixing uniformity based on the segmented non-paste and uneven paste areas. The present invention enables real-time quantification of paste mixing uniformity while ensuring the quantification accuracy of paste mixing uniformity.

Description

Technical Field

[0001] The present invention relates to the technical field of mine filling, and in particular to a non-contact full tailings paste stirring uniformity monitoring method and system. Background Art

[0002] Paste filling technology is a key technology in the field of mine filling, and includes four major process links: tailings thickening, paste mixing, pipeline transportation, and filling solidification. The mixing link requires the use of a continuous mixer to fully mix the whole tailings with cementitious materials such as binders, coarse aggregates and modifiers to disperse them evenly, thereby obtaining a high-quality paste with good fluidity and high filling strength. However, whole tailings are fine-grained materials, which are more difficult to disperse. The uneven paste surface will have blocks of material that are not completely broken up. Experienced engineers can estimate the mixing conditions by observing the surface characteristics of the prepared paste on site. So far, there is no method to achieve real-time quantification of the paste mixing uniformity, making it difficult to timely optimize the mixer configuration, power and other process parameters, and it is difficult to improve the paste mixing performance.

[0003] In summary, the current filling process for preparing pastes based on continuous mixers has the following problems:

[0004] (1) Fluctuations in upstream mining process parameters can affect the filling process, and the mixing process needs to be adjusted in a timely manner;

[0005] (2) The mixing process cannot achieve effective uniformity monitoring, the paste quality is difficult to quantify in real time, and the process control cannot be achieved quickly and effectively;

[0006] (3) The amount of stirring image data is insufficient, making it difficult to achieve high detection accuracy. Summary of the Invention

[0007] The embodiment of the present invention provides a non-contact full tailings paste stirring uniformity monitoring method and system, which can achieve real-time quantification of paste stirring uniformity while ensuring the quantification accuracy of paste stirring uniformity. The technical solution is as follows:

[0008] On the one hand, a non-contact method for monitoring stirring uniformity of a tailings paste is provided, comprising:

[0009] During the paste preparation process in the continuous mixer, the surface images of the paste are continuously collected;

[0010] Use deep learning methods to process the collected paste surface images and segment non-paste areas and paste uneven areas;

[0011] The paste mixing uniformity is determined based on the segmented non-paste area and the paste uneven area.

[0012] Furthermore, during the process of preparing the paste by the continuous mixer, continuously collecting the surface image of the paste includes:

[0013] During the process of preparing the paste by the continuous mixer, the surface image of the paste is continuously collected by an image acquisition module arranged above the continuous mixer.

[0014] Furthermore, the method of using a deep learning method to process the collected paste surface image and segmenting the non-paste area and the paste uneven area includes:

[0015] The collected paste surface image is preprocessed with histogram equalization and the image size is scaled to m*n. The scaled image is cropped into several k*k images using the overlap-tile strategy. After each cropped image is segmented using the pre-trained non-paste area segmentation model, the segmentation results are spliced back to the m*n size using the overlap-tile strategy to obtain the non-paste area in the image; where k≤m and k≤n;

[0016] The collected paste surface image is preprocessed into grayscale and scaled to p*q size. The scaled image is cropped into several r*r images using the overlap-tile strategy. Each cropped image is segmented using the pre-trained paste uneven region segmentation model, and the segmentation results are spliced back to p*q size using the overlap-tile strategy to obtain the paste uneven region in the image; where r≤p,r≤q.

[0017] Furthermore, the segmentation in the non-paste area segmentation model refers to: marking the non-paste area in the cropped image; wherein, the non-paste area refers to all areas in the image that do not belong to the paste part, including: the mixer blades, the edge of the mixing tank and other objects that block the camera's field of view.

[0018] Furthermore, the segmentation in the paste uneven area segmentation model refers to: marking the uneven areas in the cropped image through a deep learning method; wherein the uneven areas refer to the corrugated raised parts and layered parts on the paste surface.

[0019] Furthermore, determining the paste mixing uniformity according to the segmented non-paste area and the paste uneven area includes:

[0020] Calculate the ratio p of the paste uneven area to the total area of the paste in the paste surface image;

[0021] Determine the Gaussian distribution of the proportion p of the paste uneven area to the total area of the paste area at the current blade position according to the blade position;

[0022] According to the Gaussian distribution of the proportion of the paste uneven area at the current blade position, its probability distribution function F is obtained. The proportion p of the paste uneven area to the total area of the paste area is input into the probability distribution function F as an independent variable, and the paste mixing uniformity I=1-F(p) is calculated.

[0023] Furthermore, the calculating of the ratio p of the uneven paste area to the total area of the paste area in the paste surface image includes:

[0024] The result map of the paste uneven area segmentation is scaled to the same size as the non-paste area segmentation result map, and the overlapping portion of the paste uneven area and the non-paste area in the non-paste area segmentation result map is removed. The number of pixels of the remaining paste uneven area is calculated to obtain the area s0 of the paste uneven area;

[0025] Subtract the number of pixels in the non-paste area from the total number of pixels in the image to obtain the total area s1 of the paste area;

[0026] The ratio of the uneven paste area to the total area of the paste area is obtained by dividing the area s0 of the uneven paste area by the total area of the paste area = s0 / s1.

[0027] On the other hand, a non-contact full tailings paste mixing uniformity monitoring system is also provided, including:

[0028] An image acquisition module is used to continuously acquire images of the paste surface during the paste preparation process of the continuous mixer;

[0029] The image analysis module uses deep learning methods to process the collected paste surface images, segmenting the non-paste area and the paste uneven area, and determines the paste mixing uniformity based on the segmented non-paste area and the paste uneven area.

[0030] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0031] In an embodiment of the present invention, while a continuous mixer is preparing a paste, images of the paste surface are continuously collected. These images are processed using a deep learning method to segment non-paste areas and non-uniform paste areas. Based on these segmented non-paste and non-uniform paste areas, the paste mixing uniformity is determined. This approach, based on computer vision and deep learning, enables precise quantitative calculation of paste mixing uniformity and remote monitoring of the mixing process through a uniformity index, facilitating the refinement of paste filling preparation and resolving the difficulty of real-time quantitative monitoring of paste mixing uniformity. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0033] Figure 1 A schematic structural diagram of a non-contact full tailings paste stirring uniformity monitoring system provided by an embodiment of the present invention;

[0034] Figure 2 A schematic flow chart of a non-contact full tailings paste stirring uniformity monitoring method provided by an embodiment of the present invention;

[0035] Figure 3 A detailed flow chart of a non-contact full tailings paste stirring uniformity monitoring method provided by an embodiment of the present invention;

[0036] Figure 4 Schematic diagram of the training process of the non-paste area segmentation model, the paste uneven area segmentation model and the Gaussian process model provided in the embodiment of the present invention. DETAILED DESCRIPTION

[0037] To make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0038] The embodiment of the present invention provides a non-contact full tailings paste stirring uniformity monitoring method and a corresponding monitoring system. Figure 1 Shown, including:

[0039] An image acquisition module is used to continuously acquire images of the paste surface during the paste preparation process of the continuous mixer;

[0040] The image analysis module uses deep learning methods to process the collected paste surface images, segmenting the non-paste area and the paste uneven area, and determines the paste mixing uniformity based on the segmented non-paste area and the paste uneven area.

[0041] In this embodiment, Figure 1 As shown, the monitoring system also includes: a stirring system, wherein the stirring system includes: a continuous mixer; one end of the continuous mixer in the stirring system is connected to the material feeding end, and the other end is the paste discharge end, wherein the material feeding end is the feeding port for tailings slurry, cementitious materials and additives.

[0042] In this embodiment, the image acquisition module is connected to the image analysis module, which includes an industrial camera and an industrial internet communication component for data transmission. The camera is mounted above the continuous mixer and continuously captures images of the paste surface during preparation, aiming at the mixing tank. The industrial internet communication component is responsible for transmitting the captured images of the paste surface to the image analysis module. The image analysis module receives the images sent by the image acquisition module, analyzes them, and calculates a uniformity index. The industrial internet communication component also transmits control signals for changing the camera's sampling frequency and switching it on and off as needed.

[0043] like Figure 2 and Figure 3 As shown, an embodiment of the present invention provides a non-contact method for monitoring stirring uniformity of a full tailings paste, comprising:

[0044] S101, continuously collecting surface images of the paste during the paste preparation process in the continuous mixer;

[0045] In this embodiment, during the process of paste preparation by the continuous mixer, the camera in the image acquisition module arranged above the continuous mixer continuously captures the paste surface image, and the captured paste surface image is transmitted to the image analysis module through the industrial Internet communication component.

[0046] In this embodiment, the camera regularly samples at a certain interval (eg, one image every 10 seconds) to obtain an image of the paste surface.

[0047] S102, using a deep learning method to process the collected paste surface image to segment the non-paste area and the paste uneven area; specifically, the following steps may be included:

[0048] A1. Non-paste area segmentation: Perform histogram equalization preprocessing on the collected paste surface image and scale the image size to m*n (for example, 384*256). The scaled image is cropped into several k*k (k≤m, k≤n; for example, 256*256) images using the overlap-tile strategy. After segmenting each cropped image using the pre-trained non-paste area segmentation model, the segmentation results are spliced back to the m*n size using the overlap-tile strategy to obtain the non-paste area in the image.

[0049] In this embodiment, the segmentation in the non-paste area segmentation model refers to: marking the non-paste area in the cropped image, so as to identify the non-paste area; wherein, the non-paste area refers to all areas in the image that do not belong to the paste part, including: the mixer blades, the edge of the mixing tank and other objects that block the camera's field of view.

[0050] A2, paste uneven area segmentation: The collected paste surface image is grayscaled and preprocessed, and the image size is scaled to p*q (for example, 1536*1024). The scaled image is cropped into several r*r (r≤p, r≤q; for example, 512*512) images using the overlap-tile strategy. After each cropped image is segmented using the pre-trained paste uneven area segmentation model, the segmentation results are spliced back to p*q size using the overlap-tile strategy to obtain the paste uneven area in the image.

[0051] In this embodiment, the segmentation in the paste uneven area segmentation model refers to: marking the uneven areas in the cropped image through a deep learning method, thereby identifying the uneven areas of the paste; wherein the uneven areas refer to the corrugated raised parts and layered parts on the surface of the paste.

[0052] In this embodiment, the non-paste area segmentation model and the paste uneven area segmentation model that have been trained in advance are used to perform non-paste area segmentation and paste uneven area segmentation on the paste. Therefore, the non-paste area segmentation model and the paste uneven area segmentation model need to be trained before deployment. Figure 4 As shown, both models use the Deeplab v3 model, the unified focal loss loss function and the Adam optimizer, with a learning rate of 2*10 -5 The non-paste area segmentation model uses manually labeled non-paste area training data, performs histogram equalization preprocessing on the original image in the training data, and resizes the image to 384*256. Then, a 256*256 area is randomly cropped as input, and 75 rounds of training are performed. The paste uneven area segmentation model uses manually labeled paste uneven area training data, performs grayscale preprocessing on the original image in the training data, resizes the image to 1536*1024, and then randomly crops a 512*512 area as input, and 250 rounds of training are performed.

[0053] S103, determining the paste mixing uniformity according to the segmented non-paste area and the paste uneven area, which may specifically include the following steps:

[0054] B1, calculating the ratio p of the area of the uneven region in the paste region to the total area of the paste region (i.e., the ratio of the uneven paste region): calculating the ratio p of the uneven paste region in the paste surface image to the total area of the paste region; specifically, the following steps may be included:

[0055] B11, scaling the paste uneven region segmentation result image to the same size as the non-paste region segmentation result image, removing the overlapping portion of the paste uneven region and the non-paste region in the non-paste region segmentation result image, and calculating the number of pixels in the remaining paste uneven region to obtain the area s0 of the paste uneven region;

[0056] B12, subtract the number of pixels in the non-paste area from the non-paste area segmentation result from the total number of pixels in the image to obtain the total area s1 of the paste area;

[0057] B13, divide the area s0 of the paste uneven region by the total area of the paste region to obtain the ratio p=s0 / s1 of the paste uneven region to the total area of the paste region.

[0058] B2, determining the Gaussian distribution of the proportion p of the paste uneven area corresponding to the current paste surface image at the current blade position to the total area of the paste area according to the blade position; specifically, the following steps may be included:

[0059] B21, determine the blade position in the non-paste area: match the predetermined blade template with the non-paste area segmentation result to find the best match, and use the vertical coordinate value of the upper left corner of the matching position as the blade position value pos of the image;

[0060] B22, determine the distribution of the proportion of the paste uneven area at the current blade position: use the pre-trained Gaussian process model, take the blade position pos as input, and obtain a Gaussian distribution output; wherein, the Gaussian distribution is the Gaussian distribution of the proportion p of the paste uneven area at the current blade position.

[0061] In this embodiment, the Gaussian distribution corresponding to the blade position is found from the trained Gaussian process model.

[0062] In this embodiment, the Gaussian process model needs to be trained before deployment. Figure 4 As shown in the figure, the training data used was the ratio p of the uneven area of the paste region to the total area of the paste region, calculated in step B1 after image segmentation using the non-paste region segmentation model and the uneven paste region segmentation model. Using the GPyTorch framework, backpropagation was used to train hyperparameters such as length-scale and output-scale in the Gaussian process model. Training was performed using the Adam optimizer with a learning rate of 0.1 for 75 epochs.

[0063] B3. Calculate the unevenness index: According to the Gaussian distribution of the proportion of the paste uneven area at the current blade position, obtain its probability distribution function F, input the proportion p of the paste uneven area to the total area of the paste area as the independent variable into the probability distribution function F, and calculate the paste mixing uniformity I=1-F(p).

[0064] In this embodiment, the uniformity index I is obtained by calculating the probability distribution function of the current paste uneven area ratio p under the Gaussian distribution model. The uniformity of the paste is determined according to the degree to which the uniformity index I deviates from the mean of the Gaussian distribution; specifically:

[0065] If the probability distribution result is less than or equal to 0.682, it means that the current paste uneven area ratio p is less than the mean or the deviation from the mean is less than one standard deviation, and it is considered uniform;

[0066] If the probability distribution function result is greater than 0.682 and less than or equal to 0.954, it means that the current paste uneven area ratio p deviates from the mean by more than one standard deviation and less than two standard deviations, and is considered relatively uneven.

[0067] If the probability distribution function result is greater than 0.954, it means that the current paste uneven area ratio p deviates from the mean by more than two standard deviations and is considered uneven.

[0068] In this embodiment, S102 and S103 are executed in the image analysis module.

[0069] The non-contact full tailings paste stirring uniformity monitoring method described in the embodiment of the present invention has at least the following beneficial effects:

[0070] 1) The system and method provided in the embodiments of the present invention are a non-contact paste mixing uniformity monitoring system and method in a paste preparation scenario. The system and method are based on computer vision and deep learning, and realize accurate quantitative calculation of paste mixing uniformity and remote monitoring of the mixing process through uniformity indicators, which is conducive to improving the level of refined preparation of paste filling; thereby solving the problem of difficulty in real-time quantitative monitoring of paste mixing uniformity.

[0071] 2) The system and method provided in the embodiment of the present invention realize the visualization and digital mixing quality detection of the current mainstream continuous paste mixing equipment in mines. It has the characteristics of high degree of automation and fast detection speed. It can provide technical guidance for the quality control of mine paste filling slurry preparation. It has important practical value and theoretical significance, and is suitable for various mining enterprises that use cemented filling, such as non-ferrous, ferrous, precious metals, rare metals, etc.

[0072] 3) The system and method provided by the embodiment of the present invention have the advantages of high reliability and real-time monitoring, which improve the quality of the prepared paste and ensure the high-quality preparation of the filling slurry. It has the advantages of low energy consumption, high level of automation, and easy deployment, which can reduce the filling management cost.

[0073] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A non-contact method for monitoring stirring uniformity of full tailings paste, characterized in that: include: During the paste preparation process in the continuous mixer, the surface images of the paste are continuously collected; Use deep learning methods to process the collected paste surface images and segment non-paste areas and paste uneven areas; Determine the paste mixing uniformity based on the segmented non-paste area and the paste uneven area; The step of determining the paste mixing uniformity based on the segmented non-paste area and the paste uneven area includes: Calculate the ratio p of the paste uneven area to the total area of the paste in the paste surface image; Determine the Gaussian distribution of the proportion p of the paste uneven area to the total area of the paste area at the current blade position according to the blade position; According to the Gaussian distribution of the proportion of the paste uneven area at the current blade position, its probability distribution function F is obtained. The proportion p of the paste uneven area to the total area of the paste area is input into the probability distribution function F as an independent variable, and the paste mixing uniformity I=1-F(p) is calculated; The Gaussian distribution of the proportion p of the paste uneven area to the total area of the paste area at the current blade position is determined according to the blade position, including: Determine the blade position in the non-paste area: match the predetermined blade template with the non-paste area segmentation result, find the best match, and use the vertical coordinate value of the upper left corner of the matching position as the blade position value pos of the image; Determine the distribution of the paste uneven area ratio at the current blade position: Use a pre-trained Gaussian process model, take the blade position pos as input, and obtain a Gaussian distribution output; wherein, the Gaussian distribution is the Gaussian distribution of the paste uneven area ratio p at the current blade position.

2. The non-contact full tailings paste stirring uniformity monitoring method according to claim 1, characterized in that: During the process of preparing the paste by the continuous mixer, continuously collecting the paste surface image includes: During the process of preparing the paste by the continuous mixer, the surface image of the paste is continuously collected by an image acquisition module arranged above the continuous mixer.

3. The non-contact full tailings paste stirring uniformity monitoring method according to claim 1, characterized in that: The method of using a deep learning method to process the collected paste surface image and segment the non-paste area and the paste uneven area includes: The collected paste surface image is preprocessed with histogram equalization and the image size is scaled to m*n. The scaled image is cropped into several k*k images using the overlap-tile strategy. After each cropped image is segmented using the pre-trained non-paste area segmentation model, the segmentation results are spliced back to the m*n size using the overlap-tile strategy to obtain the non-paste area in the image; where k≤m and k≤n; The collected paste surface image is preprocessed into grayscale and scaled to p*q size. The scaled image is cropped into several r*r images using the overlap-tile strategy. Each cropped image is segmented using the pre-trained paste uneven region segmentation model, and the segmentation results are spliced back to p*q size using the overlap-tile strategy to obtain the paste uneven region in the image; where r≤p,r≤q.

4. The non-contact full tailings paste stirring uniformity monitoring method according to claim 3, characterized in that: Segmentation in the non-paste area segmentation model refers to marking the non-paste areas in the cropped image. The non-paste areas refer to all areas in the image that do not belong to the paste part, including the mixer blades, the edge of the mixing tank, and other objects that block the camera's field of view.

5. The non-contact full tailings paste stirring uniformity monitoring method according to claim 3, characterized in that: The segmentation in the paste uneven area segmentation model refers to marking the uneven areas in the cropped image through deep learning methods; among them, the uneven areas refer to the corrugated raised parts and layered parts on the paste surface.

6. The non-contact full tailings paste stirring uniformity monitoring method according to claim 1, characterized in that: The calculation of the ratio p of the paste uneven area to the total area of the paste area in the paste surface image includes: The result image of the paste uneven area segmentation is scaled to the same size as the non-paste area segmentation result image, and the overlapping part of the paste uneven area and the non-paste area in the non-paste area segmentation result image is removed. The number of pixels of the remaining paste uneven area is calculated to obtain the area s0 of the paste uneven area; Subtract the number of pixels in the non-paste area from the total number of pixels in the image to obtain the total area s1 of the paste area; The ratio of the uneven paste area to the total area of the paste area is obtained by dividing the area s0 of the uneven paste area by the total area of the paste area = s0 / s1.

7. A non-contact full tailings paste stirring uniformity monitoring system, characterized in that: include: An image acquisition module is used to continuously acquire images of the paste surface during the paste preparation process of the continuous mixer; The image analysis module uses deep learning methods to process the collected paste surface images, segmenting the non-paste area and the paste uneven area, and determines the paste mixing uniformity based on the segmented non-paste area and the paste uneven area; The step of determining the paste mixing uniformity based on the segmented non-paste area and the paste uneven area includes: Calculate the ratio p of the paste uneven area to the total area of the paste in the paste surface image; Determine the Gaussian distribution of the proportion p of the paste uneven area to the total area of the paste area at the current blade position according to the blade position; According to the Gaussian distribution of the proportion of the paste uneven area at the current blade position, its probability distribution function F is obtained. The proportion p of the paste uneven area to the total area of the paste area is input into the probability distribution function F as an independent variable, and the paste mixing uniformity I=1-F(p) is calculated; The Gaussian distribution of the proportion p of the paste uneven area to the total area of the paste area at the current blade position is determined according to the blade position, including: Determine the blade position in the non-paste area: match the predetermined blade template with the non-paste area segmentation result, find the best match, and use the vertical coordinate value of the upper left corner of the matching position as the blade position value pos of the image; Determine the distribution of the paste uneven area ratio at the current blade position: Use a pre-trained Gaussian process model, take the blade position pos as input, and obtain a Gaussian distribution output; wherein, the Gaussian distribution is the Gaussian distribution of the paste uneven area ratio p at the current blade position.

Images