Intelligent control system for flotation process based on froth image recognition

Abstract

The application discloses a flotation process intelligent regulation and control system based on foam image recognition, collects continuous images of a froth surface and an overflow weir area of a flotation tank, recognizes a local motion field after credibility determination, establishes an actual transport path and an along-path event chain containing migration, merging and rupture relations, calculates an effective speed to the weir, an apparent transport residence time, a bubble growth degree along the path, a merging degree, a rupture degree and a froth transport rate to the weir, compares the transport state with a current working condition stable baseline to determine the transport state, determines an adjustment direction and a magnitude according to historical responses or small amplitude control changes, adjusts aeration quantity, reagent addition quantity or pulp level, evaluates actual effects in a corresponding response window, and determines subsequent adjustment. The method is used for distinguishing different transport states with similar static appearances, and reducing repeated adjustment caused by environmental interference and response lag.

Description

Technical Field

[0001] This invention relates to the field of mineral flotation process detection and control technology, specifically to an intelligent control system for flotation processes based on foam image recognition. Background Technology

[0002] Foam flotation is one of the most widely used separation methods in mineral processing. During flotation, mineralized bubbles rise from the pulp into the froth layer, where they undergo drainage, migration, merging, and breakup before moving towards the overflow weir and forming the froth product. The size, number, texture, flow velocity, and stability of bubbles on the froth surface can reflect the current flotation conditions to a certain extent.

[0003] Existing visual inspection methods for flotation processes typically utilize camera devices to acquire foam images. Through image segmentation, target recognition, or feature extraction, parameters such as average foam diameter, foam density, color, texture, and movement speed are obtained. Based on these parameters, the flotation status is determined, and the aeration rate, reagent addition amount, or liquid level is adjusted.

[0004] The judgment criteria of the above methods are limited to the static appearance features in the current image, making it difficult to distinguish between the formation process and the subsequent trend. In a real production environment, mineralized bubbles need to travel a certain transport distance and residence time to move from the inside of the foam layer to the overflow weir. During this period, liquid film discharge, merging of adjacent bubbles, local bubble rupture, and changes in surface flow direction continuously occur. Two different operating conditions may have similar average bubble diameter, foam density, and surface texture at a certain moment, but the foam in one condition remains stable and is transported to the overflow weir rapidly, while the foam in the other condition has experienced a longer period of tank surface retention and continuous merging, and may even burst concentratedly before reaching the overflow weir. The foam on the surface of the flotation cell does not move uniformly in a single fixed direction. The tank structure, aeration state, foam layer thickness, slurry level, foam scraping process, and local foam congestion can all create flow deviation, lateral flow, stagnation zones, and local backflow. Using only the absolute value of the foam movement speed cannot determine whether the speed forms an effective transport towards the overflow weir.

[0005] The on-site video environment also affects the judgment results. Factors such as changes in lighting, water mist, slurry splashes, lens contamination, camera vibration, and periodic obstruction by the foam scraping mechanism, if directly identified as changes in foam, can easily lead to misjudgments. After the addition of reagents or adjustment of aeration, there is still a significant delivery, mixing, and foam response time. If the control system readjusts the control quantity before the control effect is transmitted to the foam observation area, it can easily lead to continuous over-adjustment.

[0006] Therefore, there is a need for an intelligent control scheme for the flotation process that can distinguish between the static appearance of foam and the actual transport state, eliminate non-flotation disturbances on site, and accurately correspond control actions with the actual response of foam. Summary of the Invention

[0007] The purpose of this invention is to provide an intelligent control system for the flotation process based on foam image recognition, so as to solve the problems in the prior art that it is difficult to distinguish between the static appearance of foam and the actual transport state, that non-flotation disturbances on site can easily cause misidentification, and that the control actions cannot accurately correspond to the actual response of foam.

[0008] To achieve the above objectives, the present invention adopts the following technical solution. A method for intelligent control of a flotation process based on foam image recognition includes: acquiring continuous foam images and establishing a spatial reference; determining the credibility of the images; identifying the local motion field and calculating the effective velocity to the weir; establishing the actual transport path and a chain of events along the path including migration, merging, and breakage relationships; calculating the apparent transport residence time, the degree of bubble growth along the path, the degree of merging along the path, the degree of breakage along the path, and the foam transport to the weir rate; comparing each indicator with the stable operating baseline of the current operating condition to determine the foam transport status and deviation; determining the adjustment direction and amplitude based on historical responses or small control changes, issuing control commands, and evaluating the actual response within the corresponding effective response window, thereby determining the adjustment direction and amplitude for the next control cycle.

[0009] Furthermore, the positional changes of each local foam unit on the foam surface between adjacent image frames are identified to obtain the motion vector of each local foam unit. The projection of the motion vector in the direction to the weir corresponding to the current position is determined as the effective velocity to the weir of the local foam unit. When the magnitude of the motion vector is higher than a preset velocity condition and the effective velocity to the weir is lower than a preset effective velocity condition, the corresponding local foam unit is identified as lateral movement or local backflow.

[0010] Further, the position of the bubble in the next image frame is predicted based on the local motion field, and candidate successor bubbles are searched in the neighborhood of the predicted position. When there is a candidate successor bubble in the neighborhood of the predicted position and its position change direction is consistent with the local motion field, a migration relationship is established. When there are multiple preceding bubbles in the neighborhood of the predicted position that correspond to a successor bubble, and the relative deviation between the apparent area of ​​the successor bubble and the sum of the apparent areas of the multiple preceding bubbles is within a preset deformation tolerance range, a merging relationship is established. When there are multiple smaller successor bubbles in the motion prediction area of ​​a preceding bubble, a rupture relationship is established. Based on the migration relationship, the merging relationship, and the rupture relationship, a bubble-along event chain with a time direction and allowing the bubble identifier in the event chain to change is generated.

[0011] Further, the process of calculating the flow transport index includes: for a flow event chain with an effective successor state reaching the vicinity of the overflow weir, determining the time of its first entry into the effective transport area and the time of its first arrival in the vicinity of the overflow weir, and determining the difference between the two as the apparent transport residence time of the flow event chain; determining the cumulative actual movement distance of the flow event chain, and determining the degree of bubble growth along the flow based on the ratio of the change in the equivalent diameter of the bubble along the flow event chain to the cumulative actual movement distance; determining the degree of merging and the degree of rupture along the flow based on the number of merging events and the number of rupture events in the flow event chain and the corresponding effective transport path length; and determining the foam transport to the weir rate based on the ratio of the number of flow event chains with an effective successor state reaching the vicinity of the overflow weir in the same entry batch to the number of entry flow event chains with a definite transport outcome.

[0012] Furthermore, when the effective velocity to the weir decreases, the apparent transport residence time increases, and the degree of merging along the flow increases, the current foam transport state is determined to be a stagnant merging state; when the degree of rupture along the flow increases and the foam transport to the weir rate decreases, the current foam transport state is determined to be a rupture along the flow state; when the area of ​​local foam units identified as lateral movement or local backflow within the effective transport area exceeds a preset proportion of the total area of ​​effective local foam units, the current foam transport state is determined to be a flow deviation and backflow state; when all transport indicators along the flow are within the fluctuation range of the stable operating baseline and the area proportion does not exceed a preset proportion, the current foam transport state is determined to be a normal transport state.

[0013] Furthermore, the stable operating baseline is the baseline value and its fluctuation range formed based on the multiple transport indicators along the flotation process when the process is in a stable operating stage; the deviation of the current transport state from the stable operating baseline is determined according to the fluctuation scale corresponding to each of the multiple transport indicators along the flotation process; and a new stable operating baseline corresponding to the changed operating conditions is formed in a new stable operating stage after the source of raw ore, the feed state or the operating conditions have undergone continuous changes.

[0014] Furthermore, the timing of the control action is recorded, the earliest effective response time is determined based on historical responses, and the effective response window is adjusted according to the apparent transport dwell time. After the window ends, the deviation before and after the control action is compared. If the deviation decreases, the original adjustment direction is retained as a candidate direction; if the deviation increases, the amplitude is reduced or the direction is changed. After entering the normal transport state, the control setpoint is maintained.

[0015] Furthermore, the adjusted value of the control variable should be between its minimum and maximum permissible control values, and the adjustment amplitude should meet the limits of the maximum change in a single cycle and the rate of change of continuous control. The actual time interval between two adjacent actions should not be less than the minimum evaluation time interval and should cover the actual duration from the previous action to the end of the complete response evaluation. When historical response data similar to the current operating condition is lacking, a small-amplitude single-variable control change that meets the above constraints should be used, and the suitability of the variable and its direction as a candidate for subsequent adjustment should be determined based on the complete and effective response evaluation.

[0016] Furthermore, during the credibility determination, within the original camera image coordinate system, areas within the fixed foam observation area that are stable in position, continuously blurred, or occluded at multiple consecutive sampling times, and where continuous foam movement is visible in the surrounding area, are designated as candidate areas for lens contamination. After excluding occlusion by the fixed tank structure and the bubble-scraping mechanism, a contamination candidate mask is formed, and lens contamination is not confirmed solely based on stillness or low speed. Candidate areas that cannot be distinguished from actual retained foam are marked as uncertain occlusions, and the usability of the image is determined based on the remaining usable area and visual credibility. Camera vibration is compensated based on the overall synchronous displacement of the fixed tank structure or the overflow weir edge; bubble-scraping occlusions are identified based on geometric shape, continuous movement direction, and periodic characteristics. Contamination candidates and bubble-scraping occluded areas are not included in the foam event statistics during the occlusion period; when the effective foam area ratio or visual credibility does not meet the usability conditions, the most recent credibility control setting is maintained, and the closed loop is restored after continuously meeting the effective image confirmation conditions.

[0017] Furthermore, the industrial vision acquisition device implementing this method is set above or to the side above the flotation cell, with a field of view covering a section of the foam transport area and the corresponding overflow weir boundary; the image processing and control device completes image processing, state determination and response evaluation, and is communicatively connected with the aeration volume, reagent addition volume and liquid level adjustment actuator.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. By determining the actual transport path and effective velocity of foam moving towards the overflow weir, the true directional transport of the product is distinguished from lateral movement and local backflow, giving the surface velocity of the foam a clear spatial physical meaning. This solves the problem of difficulty in distinguishing working conditions with similar static appearances but different transport processes. 2. By establishing a chain of events along the foam path that allows migration, merging, and bursting, the transport history can still be traced even if the bubble identifier changes. The apparent transport residence time, the degree of bubble growth, the degree of merging, the degree of bursting, and the rate of foam transport to the weir can be obtained along the actual transport path. This allows the state changes of the foam before it reaches the overflow weir to be quantified, rather than just observing the final foam morphology. 3. By identifying and eliminating lens contamination, camera vibration, bubble scraping mechanism obstruction, and lighting changes, the interference of the real industrial environment on the judgment of foam events is reduced; 4. By establishing a stable operating baseline corresponding to the current working conditions and comparing the degree of deviation, misjudgment caused by using a uniform absolute threshold under different mineral sources, reagent systems and operating conditions can be avoided; 5. By corresponding the control actions with the changes in the transport indicators along the path after passing through the actual response window, over-adjustment caused by continuous repeated adjustments before the control effect is manifested is avoided. The effective adjustment direction of the current working condition is determined based on the response of the flotation cell itself to the actual control actions, thereby reducing the adaptability problems caused by relying entirely on fixed manual experience control rules. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the flotation cell foam image acquisition and overflow direction reference in an embodiment of the present invention; Figure 2 This is an overall flowchart of the intelligent control system for the flotation process in an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the calculation of the effective velocity from foam to the weir, the transport path along the flow, and the event chain in an embodiment of the present invention. Figure 4 This is a schematic diagram illustrating bubble cross-frame migration, merging, and bursting events, as well as on-site interference identification, in an embodiment of the present invention. Figure 5 This is a schematic diagram illustrating the correspondence between control actions and foam response time, as well as the evaluation of deviation, in an embodiment of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] like Figure 1As shown, the system implementing the intelligent control method for flotation process based on foam image recognition described in this embodiment includes an image acquisition unit, an image reliability determination unit, a foam transport recognition unit, a state determination unit, a response evaluation unit, and a control execution unit. The image acquisition unit includes an industrial vision acquisition device positioned above or to the side of the flotation cell, its field of view covering at least a section of the foam transport area and the overflow weir boundary corresponding to that transport area. A foam layer is formed inside the flotation cell, and the foam on the cell surface crosses the overflow weir under the action of the foam scraping mechanism to form a foam product. The image reliability determination unit, foam transport recognition unit, state determination unit, and response evaluation unit are configured in the image processing and control device; the control execution unit is communicatively connected to the aeration volume adjustment actuator, the reagent addition actuator, and the liquid level adjustment actuator.

[0022] like Figure 2 As shown, the control process of this embodiment includes steps S110 to S200, as detailed below.

[0023] S110. Acquire continuous foam images and establish a spatial reference, such as Figure 1 As shown, an industrial vision acquisition device is installed above or to the side of the flotation cell, ensuring that the image acquisition range covers at least a section of the effective foam transport area and the corresponding overflow weir boundary. This embodiment does not require the camera device to simultaneously cover the entire flotation cell.

[0024] In the initial calibration stage, the position of the overflow weir in the image is determined based on the actual structure of the flotation cell, and a spatial mapping relationship is established between the image coordinates and the surface coordinates of the flotation cell. Dimensional calibration can be completed using the known overflow weir length, cell structure dimensions, or existing fixed structures on site, without requiring additional high-precision position sensors.

[0025] The region is calibrated in the reference image after location compensation and transformed to the same tank surface coordinate system through spatial mapping. The visible tank surface foam projection range is marked as the fixed foam observation area, excluding non-foam parts such as tank walls, fixed components, and product tanks; within this observation area, a fixed effective transport area is delineated from the upstream inlet boundary to the overflow weir. In this paper, "effective transport area" and "effective transport observation area" refer to the same fixed area. The overflow weir adjacent area is the inner strip-shaped area of ​​the tank within this area, not exceeding a preset bandwidth from the calibrated overflow weir boundary; the bandwidth is determined based on the calibration scale and tracking positioning error, and remains consistent in baseline acquisition and current evaluation. In this paper, "effective foam area" or "effective area" refers to the dynamically visible area obtained after excluding current contamination candidates, foam scraping obstructions, and parts that cannot be reliably imaged from the fixed foam observation area. The denominator of the area proportion is fixed as the area of ​​the foam observation area, not the area after elimination.

[0026] Specifically, for any position P in the image, the unit vector in the direction of the weir is determined based on the distance from that position to the nearest effective overflow weir, and is calculated using the following formula:

[0027] Where P represents the location point of the groove surface obtained through spatial mapping; P w Indicates the location of the nearest effective overflow weir corresponding to this location; n P This indicates a direction from point P to P. w The unit vector in the direction of the weir; ||·|| represents the vector magnitude. The position and displacement use the same trough surface coordinate system, and this unit vector serves as a reference for subsequent judgment on whether the foam movement belongs to the direction of effective product movement.

[0028] Both the location point and the motion vector use the calibrated trough surface coordinates. When the location point has already fallen on the target weir boundary, the direction vector at that point is no longer calculated, and it is directly registered to the weir event to avoid zero-distance division. Calibration uses at least four non-collinear fixed reference points with known trough surface coordinates to obtain the planar projection mapping; when the camera undergoes overall displacement, it is first registered to the calibration image using the fixed reference structure, and then the mapping is used.

[0029] Preferably, the installation height of the image acquisition device is such that the angle between its optical axis and the normal to the surface of the foam layer does not exceed 30°, to ensure that the projection deformation of the foam surface in the image is within an acceptable range; one feasible value is an installation height of 2.5m to 4.0m. The above values ​​are merely examples of one feasible method and do not constitute a limitation on the scope of protection of this invention.

[0030] S120. The original image is assessed for on-site credibility. Before identifying the foam state, not all images are assumed to be valid. For a real flotation site, this embodiment identifies the overall image clarity, sudden changes in global brightness, continuous obstruction caused by local water droplets, mist, or slurry contamination, the overall image displacement caused by camera vibration, periodic rigid obstruction caused by the passing of the foam scraping mechanism, and the proportion of the effective foam area to the fixed foam observation area. Based on these factors, the visual credibility of the current image or the current image window is determined.

[0031] Specifically, visual credibility is obtained by weighted fusion of the credibility of each of the above sub-items, and is given by the following formula:

[0032] in, Indicates visual credibility, with a value ranging from 0 to 1; This represents the weight of the k-th confidence item, and the sum of all weights is 1. The confidence level of the k-th confidence level component includes the clarity component, brightness stability component, pollution occlusion component, vibration component, bubble occlusion component, and effective area percentage component; K represents the total number of confidence level components.

[0033] One method for component normalization is as follows: Sharpness component: Calculate the Laplacian response variance of the grayscale image in the visible portion of the fixed bubble observation area, linearly normalize it using the calibrated unusable lower limit and sharpness reference value, and truncate it to 0-1. Brightness stability component: Calculate the absolute value of the average grayscale difference of the fixed reference area in adjacent frames, normalize it using the calibrated allowable brightness variation, subtract this ratio, and truncate it to 0-1. Contamination occlusion component: Subtract the ratio of the contamination candidate mask area to the fixed bubble observation area. Vibration component: Divide the residual displacement of the corresponding feature points after fixed structure registration by the allowable residual, subtract this ratio, and truncate it to 0-1. Bubble occlusion component: Subtract the ratio of the bubble occlusion mask area to the fixed bubble observation area. Effective area proportion component: Take the ratio of the dynamic effective bubble area area to the fixed bubble observation area. For each area, the intersection of the mask and the fixed foam observation area is calculated. Overlapping occlusion is deducted only once in the effective area. If the denominator is zero or registration cannot be completed, the current image is directly judged as unusable.

[0034] Preferably, the weights of each credibility item Based on the saliency of historical data from the site, one possible weighting is: 0.25 for sharpness, 0.15 for brightness stability, 0.20 for contamination / occlusion, 0.15 for vibration, 0.10 for bubble removal / occlusion, and 0.15 for effective area percentage. When visual credibility... If the value is below a preset confidence threshold, the current image window is determined not to meet the valid conditions. Preferably, the preset confidence threshold ranges from 0.6 to 0.8, and one possible value is 0.7. The above values ​​are merely examples of one possible implementation and do not constitute a limitation on the scope of protection of this invention.

[0035] Each normalized reference value is determined by manually labeled usable and unusable calibration images under the same installation conditions, and the calibration images and configuration versions are saved: the sharpness reference value is the median of the sharpness of the usable images; the lower limit of sharpness is selected between the upper limit of unusable sharpness and the lower limit of usable sharpness; the allowable variation in brightness and the allowable residual of vibration are the upper limits when manually determined to still be reliably segmented and correlated. If the two types of calibration samples cannot be distinguished, recalibration is performed, and thresholds are not set arbitrarily. If the sharpness is lower than the unusable lower limit or the registration residual exceeds the allowable upper limit, the current image is directly judged as invalid, and other higher sub-items are not allowed to offset the failure. An additional usable lower limit is set for the effective area ratio, which is selected from the calibration samples to complete the effective tracking of the area adjacent to the inlet to the weir; the number of consecutive effective frames before the loop closure can be uniformly set to 30 frames with the number of pollution candidate confirmation frames. The pollution and bubble scraping sub-items reflect the loss of usable information in the entire observation range, and the effective area ratio sub-item reflects the remaining information after removing all obstructions; the above-mentioned related sub-items participate in gating through weights, and the partial obstruction removal and the whole frame confidence gating are performed separately.

[0036] Lens contamination identification is performed within the original camera image coordinate system. First, calibrated tank walls, overflow weir edges, and fixed components are excluded. Then, areas that are persistently blurred or exhibit occlusion characteristics, whose positions remain approximately unchanged, and whose surrounding foam continues to move are marked as contamination candidate areas. A stationary position alone is insufficient to confirm lens contamination; genuine retained foam should not be eliminated solely due to low velocity. Candidate areas that cannot be distinguished from genuine retained foam are marked as uncertain occlusions, reducing their usable area, and, if necessary, are placed in an unreliable environmental state. Camera vibration is estimated and compensated for by the overall displacement of the fixed structure; the bubble-scraping mechanism is identified by its rigid shape, continuous movement direction, and periodicity. Contamination candidates or bubble-scraping occlusions are only removed from the foam event statistics of the corresponding area; tracking interruptions caused by occlusion are not considered natural breakage or natural disappearance.

[0037] Specifically, candidate contamination regions are determined according to the aforementioned joint conditions, and the regions corresponding to the candidate contamination masks are removed from the foam event statistics. For the overall synchronous displacement of the fixed overflow weir edge, it is treated as global motion caused by camera vibration, and position compensation is performed on the current frame before calculating the foam motion. For the motion of the long strip-shaped rigid region caused by the passing of the bubble-scraping mechanism, a bubble-scraping mask is formed by combining its geometry, continuous motion direction, and periodicity. During the period of coverage, the statistics of rupture and disappearance events in the corresponding region are suspended. Tracking interruptions caused by masking are not considered as natural ruptures.

[0038] Specifically, for the lens contamination candidate regions determined based on the aforementioned persistent blurring or occlusion features, their positional stability within the original camera image coordinate system is used as an auxiliary criterion and measured using the following formula:

[0039] S130. Identify the local motion field of the foam surface and calculate the effective velocity of the weir. Among the continuous images that have passed the confidence screening, identify the foam boundary, texture structure or continuously related local foam units, calculate the position change of each local foam unit in adjacent image frames, and form the foam surface motion vector.

[0040] For cases where the displacement per pixel is small, the cross-correlation peak can be interpolated using its neighboring sampling points at sub-pixel levels, or the speed can be calculated by accumulating the displacement of multiple frames and then converting it to the actual time interval; the corresponding time interval and search range must be updated synchronously. The numerical examples of 8.5 mm / s, 1.2 mm / pixel, and 10 frames / second correspond to approximately 0.708 pixels / frame, therefore, integer pixel displacement quantization alone cannot be used.

[0041] One method for local motion field recognition involves: grayscale conversion of adjacent valid images after position compensation; dividing overlapping local image blocks within the valid foam region; using each image block of the first frame as a template, calculating zero-mean normalized cross-correlation in the bounded neighborhood of the next frame; selecting the position with the highest similarity as a candidate displacement; and checking consistency through reverse matching. Matches with zero template variance, unclear correlation peaks, exceeding the search range, or reverse errors exceeding the calibration tolerance are discarded. The search range is determined based on the maximum allowable surface displacement and frame interval of the device, and the matching and reverse error thresholds are calibrated using manually labeled corresponding image blocks. The bit positions of each matched position after slot surface mapping are removed by the frame interval to form a set of local velocity vectors with position indices; pixel displacements that have not been time-normalized must not be directly used as mm / s velocity.

[0042] Specifically, for the motion vector of the i-th local foam element at time t, project it onto the direction to the weir corresponding to the current position to obtain the effective velocity to the weir, which is calculated as follows:

[0043] Among them, v L,i (t) represents the effective velocity of the i-th local foam element to the weir at time t; v i (t) represents the motion vector of this unit; n i (t) represents the unit direction vector from the current position toward the target overflow weir; · represents the vector dot product. The effective velocity toward the weir is the signed directional component, with negative values ​​indicating deviation from the target overflow weir.

[0044] Furthermore, to quantify the degree of lateral movement and backflow, the degree of lateral movement is defined by the following formula:

[0045] in, The value represents the degree of lateral movement of the i-th local foam unit at time t, and the value ranges from 0 to 2, where 0 indicates that the movement direction is completely along the direction to the weir, 1 indicates that the movement direction is perpendicular to the direction to the weir, and greater than 1 indicates that the movement direction has deviated from the overflow weir. This represents the magnitude of the motion vector. The larger the value, the more the foam movement direction deviates from the weir direction. Preferably, when When the value exceeds a preset lateral movement threshold, the local foam unit is determined to be in a lateral movement or recirculation state; the preset lateral movement threshold ranges from 0.5 to 0.75, and one possible value is 0.6. The above values ​​are merely examples of one possible implementation and do not constitute a limitation on the scope of protection of this invention.

[0046] When the magnitude of the motion vector is not greater than the upper limit of velocity noise determined by the registration residual of the stationary reference structure, the degree of lateral motion is not calculated and is marked as low speed or stagnation. Lateral motion or backflow is judged only in local foam elements where the magnitude exceeds the upper limit; a lateral motion degree exceeding 0.6 can be used as a criterion, and a degree exceeding 1 indicates that the motion direction is away from the overflow weir. When the denominator is zero, no normal direction value is assigned.

[0047] When using the above direction criterion, the preset velocity condition is taken as the upper limit of the velocity noise. Let the preset lateral motion threshold be η0, then the corresponding effective velocity condition at the weir is the motion vector magnitude multiplied by (1−η0). Provided the motion magnitude exceeds the noise upper limit, an effective velocity at the weir below this condition is equivalent to a lateral motion degree greater than η0. The regional flow deviation and return state is then determined according to the effective unit area ratio and continuous window condition that satisfy this local criterion, without confusing the direction ratio threshold with a fixed absolute velocity value.

[0048] S140. Establish the actual foam transport path by connecting the foam region from the direction inside the trough to the direction of the overflow weir according to the local motion vectors in multiple consecutive image frames, thus forming the actual foam transport path. The path is not preset to be a straight line, and is allowed to form curved paths, deflection paths, local backflow paths, and paths that re-enter the main transport direction after stagnation.

[0049] Specifically, for a transport path P k By continuously tracking the corresponding foam unit to obtain its position sequence along the route, the actual transport distance is accumulated according to the continuous position increments and calculated by the following formula:

[0050] in, This represents the cumulative actual distance traveled along the transport path; m represents the total number of location sampling points along the path. This represents the position coordinates of the k-th transport path at the j-th sampling time. The actual transport distance is accumulated using continuous position increments, rather than using the straight-line distance from the start to the end point.

[0051] Furthermore, to quantify the degree of path detour, a path detour degree is defined and determined as follows:

[0052] Where, γ P Indicates the path detour degree; L P D represents the cumulative actual distance traveled from the start to the end of the continuous path in the statistics. net For the net linear displacement from the same starting point to the same ending point, both use the same coordinate system and unit of length. Only in D... net Calculated when the resolution is greater than the positioning resolution, at which point γ P Not less than 1; when the endpoint has not yet reached the weir, the distance from the starting point to the overflow weir that has not yet been reached is not used to replace D. net For paths already reached from the weir, the endpoint is the first valid weir-reaching position. This index is not calculated if the net displacement is too small or the path is discontinuous. γ P The closer the value is to 1, the closer the observed path is to a straight line; an increase in the value indicates an increase in the detour of the path, and the backflow judgment still takes into account the local direction of motion.

[0053] S150. Establish cross-frame association of bubbles and the event chain along the bubble path, and identify the closed boundary of bubbles in each valid image frame. For a bubble at time t, record at least its bubble boundary, projected area, equivalent diameter, center position, and local motion direction. Predict the position of the bubble at the next time step using the current local motion vector, and search for candidate successor bubbles near the predicted position.

[0054] One bubble recognition method involves smoothing the grayscale image within the effective area, extracting the bubble boundary gradient, and then thresholding and morphologically closing it to form a boundary. Connected regions within the closed boundaries are marked, and adhering regions are segmented using local maxima of distance transformation as markers. Unclosed regions, regions located at occlusion edges, or regions smaller than the localization resolution are not considered complete bubbles. The area of ​​a region in the groove coordinate system is the apparent area, and the equivalent diameter is the diameter of a circle with the same area. The current local velocity is obtained by interpolating the center position. This velocity is multiplied by the time interval between adjacent frames and added to the current center position to obtain the predicted position for the next frame. This predicted position is then mapped back to image coordinates for candidate searching. The search radius at least covers the displacement caused by the localization error and calibration velocity error within the frame interval and is determined on manually labeled cross-frame association samples. When a match fails, the association is considered uncertain, and a successor node is not mandatory.

[0055] This embodiment does not only allow one-to-one target tracking, but also allows three types of relationships: migration, merging, and breaking.

[0056] Specifically, the criteria for determining the migration relationship are: a bubble in the previous image frame corresponds to a bubble in the next image frame, and the position change is consistent with the local motion direction.

[0057] Specifically, the criteria for determining the merging relationship are: multiple preceding bubbles in a subsequent image frame jointly correspond to a single subsequent bubble, and the subsequent bubble is located at the convergence point of the motion prediction regions of multiple preceding bubbles. Simultaneously, its apparent area change satisfies the allowable deformation tolerance, as determined by the following formula:

[0058] in, This represents the projected area of ​​the subsequent bubble in frame t+1; Let M represent the projected area of ​​the i-th preceding bubble in frame t; M represents the number of preceding bubbles involved in the merging. The allowable deformation tolerance is expressed as the ratio of the deviation to the sum of the apparent areas of the plurality of preceding bubbles. Preferably, The value ranges from 0.1 to 0.3, and one possible value is 0.2. The above values ​​are merely examples of one possible implementation and do not constitute a limitation on the scope of protection of this invention.

[0059] Specifically, the criteria for determining the breakage relationship are: a preceding bubble corresponds to two or more smaller successor bubbles in the next image frame, and the multiple successor bubbles are located within the boundary coverage area of ​​the preceding bubble or its motion prediction area.

[0060] Bubbles that leave the effective boundary of the image directly are not counted as bursts; bubbles that disappear within the area covered by the bubble scraping mechanism are not counted as natural bursts; bubbles that are obscured by lens contamination areas are not included in the burst statistics.

[0061] A time-oriented event chain is established for bubbles based on migration, merging, and bursting relationships. Even if a bubble merges or bursts during transport, its previous and subsequent states can still be linked through the event chain, rather than losing the entire transport history due to changes in the target number.

[0062] To avoid duplicate counting of many-to-one and one-to-many relationships, an independent root identifier is assigned upon first entry at the entry point. Merged nodes retain all preceding root identifiers, and branches inherit the original root identifiers. A unique event identifier is established for each real merge or breakage event, and it is counted only once within the same statistical window. The cumulative path length only counts the confirmed physical displacement between adjacent valid nodes, and shared edges are counted only once when summarizing event intensity. Pure number jumps caused by occlusion gaps and reconstructed centers after merging should not be directly used as movement distance.

[0063] S160. Calculate the transport indices along the route. For the first time a chain of events enters the fixed effective transport area from the upstream inlet, record the entry time t. in When at least one of its effective successor branches first enters the area adjacent to the overflow weir, the time t of the first effective arrival at the weir is recorded. out The apparent transport dwell time is calculated only for the event chain along the path that has already effectively reached the dam, and is determined by the following formula:

[0064] in, This indicates the apparent transport dwell time from the entrance to the effective dam; This is the moment of first entry; The first valid arrival time at the weir is recorded. The time when a node leaves the fixed effective transport area from other boundaries is recorded as the departure time. If no reliable successor node is detected consecutively within the visible area, and the situation remains unrecovered after at least a number of candidate confirmation frames, it is recorded as visibility termination. If the successor status cannot be determined due to occlusion or the end of observation, it is recorded as incomplete observation. The durations of departure, termination, and incomplete observation are not included in the weir arrival time. Statistics show that avoiding premature disappearance is interpreted as accelerated transportation. It is not equivalent to the residence time of air bubbles in the entire flotation cell.

[0065] Specifically, the apparent bubble size is determined at different locations or time points along the same transport event chain. For the change in the equivalent bubble diameter from the starting position to the current or ending position of the event chain, the degree of bubble growth along the path is calculated and determined according to the following relationship:

[0066] in, The apparent bubble growth rate indicates the actual transport distance per unit. and These are the equivalent diameters of the bubbles at the start and end points of the selected path, respectively. This represents the cumulative actual distance traveled along the same path. It is only applicable when the path is valid and... Calculations are performed when the positioning resolution is greater than the target resolution; length and diameter are in the same unit. When many-to-one merging occurs, if a single value is needed to represent the typical size of multiple bubbles before merging, this implementation takes the average of the equivalent diameters of each preceding bubble, weighted by their apparent area, and preserves the original root identifier and the states of each preceding bubble. This means it represents the area-weighted typical size of the group before merging, not a conserved gas volume quantity; compared with the equivalent diameter of a single bubble after merging, it can characterize the local size changes at the merging node. The entire path... The diameters of the starting and ending points of the path will still be used, instead of the sum of the starting diameters of multiple root chains. It also avoids mistaking locally weighted representative values ​​for new inlet diameters. When branches exist, the continuous path back to the weir branch from the first valid path is selected; when the branch is unclear, the chain is not calculated. .

[0067] Specifically, within a given transport path or time window, merging and breaking events are statistically analyzed to form the degree of merging and the degree of breaking per unit path, respectively, using the following formulas:

[0068]

[0069] Among them, R c With R b N represents the degree of merging and the degree of fragmentation of a unit effective transport path, respectively; c With N b The number of merged events and the number of broken events after deduplication within the same statistical window, the same region, and the same valid dataset; L P This is the sum of the lengths of non-repeating valid trajectory edges within the statistical set, in meters. Shared edges are counted only once, and events and paths use the same statistical metric. Calculations are performed when the total path length is greater than zero, with event intensity measured in times / m; otherwise, it is marked as insufficient data and not replaced with zero.

[0070] In another implementation of time normalization, only for the same event chain with a complete weir-to-weir record, the number of merging events and rupture events during the period from the chain's entry into the weir are divided by the chain's time factor (T). r The event intensity per unit time is obtained, with the unit being times / second. This time metric establishes a baseline with the aforementioned unit path metric. It does not divide the total number of events across multiple chains within the statistical window by the dwell time of a single chain, nor does it directly substitute the times / second metric into the state vector and baseline corresponding to the times / m metric.

[0071] For root event chains entering the same batch at the same inlet, if at least one valid successor branch reaches the vicinity of the overflow weir, it is recorded as having reached the weir; if all visible branches leave from other boundaries or terminate in the visible area, it is recorded as not having reached the weir. Root chains still moving within the area continue to be tracked; if the outcome cannot be determined due to occlusion, mismatch, or the end of observation, it is recorded as an incomplete observation and not directly recorded as not having reached the weir. The calculation of root chains with a definite outcome for this batch is as follows:

[0072] Among them, R s N represents the rate of foam transport to the weir within a single batch entering from the same inlet, representing the root event chain with a definable outcome; reach N represents the number of root chains that have been effectively connected to the weir in this batch. inThis is the sum of the number of root chains that have reached the weir and those that are not yet visible in the observation, excluding root chains still being tracked or whose fate cannot be determined due to obstruction, mismatch, or the end of observation. Each root chain, regardless of the number of branches, is counted at most once; after merging multiple roots, each retains its independent identity. The total number of root chains entering the batch and the proportion of incomplete observations are also recorded; if the sample size is insufficient or the proportion of incomplete observations does not meet the calibration requirements, this indicator will not be used for control evaluation. N in A value of zero indicates insufficient data. This indicator represents the apparent weir situation in the visually valid sample and is not equivalent to mineral recovery rate or actual air recovery rate.

[0073] S170. Construct a transport state vector and identify the true transport state of the foam. Based on the effective velocity to the weir, apparent transport residence time, bubble growth along the friction path, merging along the friction path, rupture along the friction path, and foam transport to the weir rate, form the current foam transport state vector, determined in the following manner:

[0074] Among them, X t This represents the bubble transport state vector at time t, where t is the time index; v L Indicates the effective velocity to the weir; T r Indicates apparent transport residence time; G L Indicates the degree of bubble growth along the path; R c Indicates the degree of merging along the route; R b Indicates the degree of fracture along the path; R s This indicates the rate at which foam is transported to the weir.

[0075] When constructing state vectors, a fixed-length statistical window and region are used uniformly: the effective velocity to the weir is taken as the area-weighted average of the effective local units; the apparent transport residence time is taken as the median residence time of the root chain that completes an effective journey to the weir within the window; the degree of growth along the path is taken as the median degree of growth of the chain with an effective continuous path; the event intensity is the sum of the total number of deduplicated events and the total length of the effective path within the window; and the arrival rate is calculated based on the identifiable outcomes of the same entry batch. The baseline and current value use the same caliber; if any necessary indicator is missing, it is not treated as a zero-entry state vector.

[0076] Based on the relationship between each indicator and the stable operating baseline, five states are identified: normal transport, stagnation and merging, friction rupture, flow deviation and backflow, and environmentally unreliable conditions. Environmentally unreliable conditions, belonging to the measurement quality state, take precedence over the foam process state determination. First, check the usability of images and statistical samples; if usable, then determine the process state. When multiple abnormal conditions coexist, retain the corresponding labels, without forcibly interpreting them as the sole physical cause.

[0077] One quantitative criterion is as follows: the normal range for each indicator is set at twice the normal fluctuation scale above and below the baseline mean. A continuous change is defined as three consecutive effective statistical windows exceeding the same boundary. "Stable" means within the normal range; "increasing" means continuously exceeding the upper limit; "decreasing" or "significantly decreasing" means continuously below the lower limit; "not continuously increasing" means not meeting the requirement of exceeding the upper limit for three consecutive windows. Normal transport requires all indicators to be within the normal range, and the area proportion of lateral movement or return flow units not exceeding a preset ratio.

[0078] Retention and Merging State: The effective velocity to the weir remains below its lower limit of the normal range, while the apparent transport residence time and the degree of merging along the friction remain above their respective upper limits. Increased bubble growth along the friction can serve as supplementary information. This state is used to trigger response selection aimed at improving effective transport to the weir and bringing residence time and merging back to the baseline; it cannot be assumed that this state will necessarily increase or decrease a certain control variable.

[0079] Friction-related rupture state: The degree of friction-related rupture consistently exceeds the upper limit of the normal range, while the foam transport rate to the weir consistently falls below its lower limit. This state is used to trigger response selection aimed at reducing abnormal ruptures and restoring the weir rate; termination due to image occlusion or associated uncertainty is not used for rupture counting in this state.

[0080] Deflection and backflow state: Within the effective transport area, if the velocity modulus exceeds the noise upper limit and the lateral movement exceeds the set threshold, the area of ​​the unit must exceed a preset proportion for three consecutive windows, representing the total area of ​​the effective local units; this proportion can be 50%. Here, "not low" velocity means its modulus exceeds the velocity noise upper limit, and "significantly decreased" velocity towards the weir is judged according to the lower limit of the normal range. Increased path tortuosity is used as auxiliary information. This state is used to select a control response that can increase the velocity component towards the overflow weir and reduce the proportion of deflection and backflow.

[0081] Unreliable environment state: Visual credibility or effective area percentage does not meet the gating conditions, or a complete state vector that meets the sample requirements cannot be formed. In this case, image closed-loop adjustment, state updates, and baseline updates are paused, and the most recent credible control setting is maintained; if no credible control value is available upon initial startup, the existing control setting on-site is used. Independent equipment limits and safety interlocks remain effective. After visual credibility is restored and the effective frame confirmation conditions are continuously met, a complete statistical window is re-established, and image closed-loop is then restored.

[0082] Under normal transport conditions, the current control setpoint is maintained, and no additional adjustments in the same direction are made. Control of abnormal conditions includes, in sequence, historical action screening, current action limitation, single-variable execution, and complete response evaluation. Each historical response record corresponds to one actually adjusted control variable, the setpoint before and after the action, its signed change, response delay, and transport indicators before and after the action under the same statistical caliber. The control variable is the aeration rate, reagent addition rate, or slurry level; a positive change indicates an increase in the variable's setpoint, and a negative change indicates a decrease. The control variable and the sign of the change together determine the candidate action; the name of the abnormal condition alone does not necessarily indicate that a certain control variable will increase or decrease.

[0083] For the stagnant and merged state, retrieve response records from the same ore source, similar feed, and control working range where only one variable is changed and there are no concurrent disturbances. Candidate records should simultaneously indicate that: the effective velocity to the weir below the normal range is recovering to the normal range; the apparent transport dwell time and degree of merging along the route above the normal range are declining towards the normal range; the overall deviation is decreasing; and no other indicators have newly exceeded limits. Read the control variables, change signs, and amplitudes from the records that meet the conditions, and after limiting the amplitude according to S200, form the current control command for that variable. After the complete evaluation, judge the actual improvement according to the same indicator conditions; if there is no improvement, do not continue to add the same action, and reduce the subsequent amplitude according to S200 or use other candidate directions.

[0084] For friction rupture conditions, retrieve univariate response records according to the above operating conditions. Candidate records should simultaneously indicate that: the friction rupture degree above the normal range is declining back to the normal range; the foam transport to the weir rate below the normal range is rising back to the normal range; the overall deviation is decreasing; and no other indicators have exceeded the limits. Read the controlled variable, the sign of the change, and the amplitude from the records, and execute after limiting the amplitude; after the complete evaluation, verify the friction rupture degree, weir rate, and overall deviation. The actual variables and their directions selected in the reagent addition amount, aeration amount, or slurry level are based on the screened records; friction rupture is not directly equated with the necessity of increasing reagents.

[0085] For the backflow and deviation state, retrieve single-variable response records under the same operating conditions. Candidate records should indicate that: the effective velocity to the weir, which was below the normal range, is recovering to the normal range; the proportion of lateral movement or backflow unit area statistically analyzed according to the same direction criteria is decreasing; the overall deviation is decreasing; and no other indicators have exceeded the limits. If the effective velocity to the weir was already in the normal range, it is required to remain in the normal range. Read the variables, change signs, and amplitudes from the records that meet the conditions, and execute the evaluation after amplitude limiting. After the complete evaluation, review the above-mentioned velocities, area proportions, and deviations. When the location of transport loss can be determined, use the location distribution described later to help filter the response records of relevant variables. The actual increase or decrease direction is still taken from the effective response under the current similar operating conditions.

[0086] For multiple matching records of the same control variable and in the same direction, the candidate with the smaller absolute change within the constraint is preferred. Since the changes in different control variables have different units, their numerical values ​​are not directly compared. When there are matching executable candidates for different variables, one variable is selected to implement the action for this cycle, while the remaining variables remain unchanged. In mixed abnormal states, candidate actions should improve the current target abnormality, prevent other triggered abnormalities from worsening, and reduce the overall deviation. If there are no matching historical actions, proceed to the small-amplitude single-variable response identification in S200; if the identification input conditions are not met, maintain the current settings. After resuming normal operation, maintain the set values ​​and do not repeatedly add actions in the same direction.

[0087] S180. Establish an adaptive baseline based on real operating conditions. Considering the differences in absolute foam size and motion state under different ore sources, pulp conditions, reagent regimes, and production shifts, this embodiment does not use a fixed foam diameter, velocity, or breakage rate as a unified judgment threshold for all operating conditions. Within a historical period confirming stable operation of the flotation process, establish a stable transport baseline for the current operating conditions, given by the following formula: The stable section is first selected based on independent production operation records: the ore source, feed rate and concentration, reagent system, aeration, and slurry level setpoints are within the normal fluctuation range specified on-site, no control jumps, alarms, or maintenance have occurred, the images are continuously usable, and the process covers several inlet to weir transport processes. Within this stable section, at least two valid window samples are obtained using the same caliber as online calculations; for example, continuous acquisition for 30 minutes is possible. The baseline mean for each item is the arithmetic mean of the index values ​​across all valid windows, and the sample standard deviation is obtained by dividing the sum of squares of each value minus the mean by the sample size minus one and then taking the square root; the stable interval is taken as twice the normal fluctuation scale above and below the mean. The on-site operating condition labels and allowable fluctuation ranges are derived from the production procedures and saved with the baselines. The baseline cannot be selected solely based on the claim of stability of the image to be judged; it is given by the following formula:

[0088] Among them, X base This represents the stable transport baseline vector for the current operating condition; v L,0 T r,0 G L,0 R c,0 R b,0 R s,0 The baseline values ​​represent, in order, the effective velocity to the weir during the stable operation phase, the apparent transport residence time, the degree of bubble growth along the path, the degree of merging along the path, the degree of rupture along the path, and the rate of foam transport to the weir.

[0089] Specifically, during actual operation, the deviation between the current transport status and the corresponding operating condition baseline is compared. For the i-th transport index, its standardized deviation is determined by the following formula:

[0090] in, This represents the standardized deviation of the i-th transport index; This represents the current value of the i-th transport index; This represents the baseline mean of the i-th transport index under the current stable operating conditions; This represents the normal fluctuation scale of the i-th transport index during the stable phase. Furthermore, the overall deviation of the current foam transport state from the current stable operating region is given by the following formula:

[0091] in represents the overall deviation at time t; I represents the number of transport indicators involved in the evaluation; This represents the standardized deviation of the i-th transport index. The smaller the value, the closer the current transportation status is to the stable operating range of the current working conditions.

[0092] When the source of raw ore, feeding conditions, or other operating conditions undergo continuous changes, a corresponding baseline is re-established in the new stable operating phase. This avoids absolute judgment methods such as "a certain fixed bubble diameter is definitely abnormal" or "below a certain fixed speed is definitely abnormal," which are not applicable to different types of ore and different production conditions.

[0093] The normal fluctuation scale is the maximum value among the sample standard deviation of the indicator, 3% of the absolute value of its baseline mean, and the measurement resolution of the indicator, to avoid division by zero due to a zero baseline or zero variance; the measurement resolution is determined by the image scale, sampling interval, and number of statistical samples. Automatic adjustment dependent on the baseline is not enabled when there are insufficient samples, the stable segment does not meet the conditions, or a new operating condition baseline has not yet been established. The baseline is frozen during anomaly or action evaluation; after a change in operating condition, the independent stable segment must be reconfirmed before updating to avoid absorbing anomalies into the baseline.

[0094] S190. The control action is time-aligned with the actual foam response. After the flotation control action occurs, its effect will not immediately appear on the foam surface observed by the camera. Taking reagent addition as an example, its process may include reagent delivery, slurry mixing, surface interaction of mineral particles, formation of mineralized bubbles, bubble rising, entry into the foam layer, and surface transport of the foam. If the control is deemed ineffective again based on the current foam image before the above process is completed, and the control quantity is repeatedly increased, continuous overshoot can easily occur.

[0095] like Figure 5 As shown, this embodiment records the timing of each control action. The earliest effective response time corresponding to the current control variable is established. This response time can be obtained based on the time relationship between historical control actions and the first sustained change in the foam transport index. The current foam transport residence time also participates in determining the evaluation window.

[0096] Specifically, the earliest effective response time of the j-th control variable is estimated based on historical control records and determined by the following formula:

[0097] in, The earliest effective response time of the j-th control variable is represented; L represents the number of historical control actions included in the statistics. This indicates the duration from the moment the l-th historical control action was issued until the first sustained change in the bubble transport index; This indicates median calculation. Using median statistics can reduce the interference of individual outlier responses on response time estimation.

[0098] When there are no historical control records, the earliest effective response time should not be set to zero or filled in with a fixed value based on experience. Under the condition that an effective stable baseline has been established, on-site conditions permit, and the variables and signed changes of this trial have been recorded, only a limited, small change is applied to that variable, while other variables remain unchanged. Using the values ​​of each indicator in the complete statistical window before the action as a reference, new changes relative to that reference are monitored using the complete effective window after the action. When the first instance of exceeding the normal fluctuation limit of an indicator occurs and three consecutive effective windows maintain the same direction of change, the start time of the first window is recorded as the starting point of this response, and this record is accepted after continuity is confirmed. It cannot be concluded that an action has produced a response simply because the indicator has deviated from the baseline before the action. The response delay is obtained by subtracting the action time from the response starting point. The upper limit of observation is determined by equipment data and on-site calibration based on the actual time required for reagent delivery, slurry mixing, bubble rising, and surface transport. If there is still no clear response after the timeout, or if the operating conditions and images during the period are invalid, it is marked as unidentifiable and no automatic adjustments are made. After obtaining a valid record, the response delay is updated using the median. When the upper limit of observation allowed by the equipment is unknown, calibration is completed before starting the trial.

[0099] The total waiting time for the j-th control variable from the issuance of the action to the end of the evaluation is determined by the response delay and the transport correction term:

[0100] in, This refers to the total waiting time from when the action is issued to when the evaluation ends. The earliest effective response delay; The apparent transport dwell time before the action is defined, and this window parameter remains constant during the evaluation period; α is the dwell time correction coefficient, ranging from 0.5 to 1.0, with one possible value of 0.8. The effective evaluation segment starts from the action time plus... Start, until the moment of action End, its length is α multiplied by It cannot be added again. The evaluation segment must accommodate the valid statistical window required for state determination and the event chain with a definable outcome; if insufficient, the evaluation should be extended and the actual end time recorded, without shortening it to zero duration.

[0101] Once the control action enters the effective evaluation phase, effect judgment data is collected. Only after the complete evaluation is finished, and the necessary effective statistical window and event chain for determining the outcome are sufficient, are the indicators before and after the action compared, and the next action determined. During the waiting and evaluation period, the current settings are maintained, and adjustments are not repeatedly increased based on the next frame of the image. The same area, indicator caliber, and frozen stable baseline are used before and after the action. If there are continuous changes in operating conditions such as ore feeding, or if the image or sample becomes invalid, this evaluation is not recorded as an effective improvement, and no action is automatically added due to the expiration of the nominal waiting time.

[0102] S200. The subsequent control direction is determined by utilizing the response of the flotation cell itself. This embodiment does not establish fixed empirical rules such as "large bubbles must reduce air" or "bubbling must increase frother" in advance. Instead, it prioritizes the use of historical response records formed by existing control actions on site for the current working conditions.

[0103] Specifically, for a certain control variable Record its changes And after an effective response time, the foam transport state changes from Change to Thus, we obtain the current operating conditions. to The true response relationship.

[0104] The effectiveness of the control action is evaluated according to the improvement criteria for the corresponding abnormal state in S170, combined with the overall deviation and whether other indicators have been added beyond the limits. Only when these conditions are met is the variable and direction considered as valid candidates under the current operating condition; if not, improvement is not confirmed based solely on a change in a single indicator, but subsequent actions are determined according to the complete evaluation and amplitude limit rules described later.

[0105] Specifically, the overall deviation is used as the basis for evaluating the control effect: when a control action satisfies the following relationship after passing through the complete response window, it indicates that the control action has brought the overall foam transport state closer to the current stable condition:

[0106] Among them, D after D represents the overall degree of deviation after this complete and valid evaluation. before This indicates the overall deviation of the complete window before the action; both use the same baseline. (D) after Less than D before Furthermore, if the improvement conditions for the corresponding abnormal state are met, the current variables and direction are saved as subsequent candidates; if normal transport has been restored, the current settings are maintained, and no further changes in the same direction are added. after If the increase or target anomaly is not improved, the original action is not repeated. Instead, the amplitude is reduced according to the constraints or another candidate direction is used within the subsequent complete control cycle. When data is insufficient, the missing deviation is not interpreted as improvement or deterioration. The deviation is used to evaluate the response of visual transport indicators and does not directly indicate an increase in mineral recovery.

[0107] When historical responses matching the current operating conditions are lacking, small control changes are used to identify local responses, without pre-assuming that the action will necessarily improve the current state. Before the trial, one permissible control variable and its signed small change are registered as the input for this identification; this variable must have an available actuator, the selected direction must have residual adjustment margin, and the amplitude must meet all equipment constraints in this step. If this input is not yet determined, or if equipment permissible conditions, effective baselines, and observation upper limits are lacking, the original control settings on site are maintained, and the trial is not automatically initiated. After the input is determined, other variables remain unchanged, and the response starting point and complete evaluation results are obtained according to S190; only when the response meets the improvement criteria for the current abnormal state is the variable and direction saved as subsequent candidates. If there is no clear response or the operating conditions are no longer comparable, it is not registered as a valid improvement record, and no additional action in the same direction is added due to timeout.

[0108] For selected control variables, signed changes in historical records meeting the conditions are used as candidates. The minimum of the following is taken as the current adjustment amplitude: the absolute value of the candidate, the remaining equipment adjustment margin in that direction, the maximum change in a single cycle, and the allowable rate of change multiplied by the current action interval; the direction is retained as the sign of the candidate change. The current setpoint equals the current setpoint of the variable plus the adjustment amplitude with the aforementioned sign. The upward adjustment margin is the maximum allowable value minus the current value; the downward adjustment margin is the current value minus the minimum allowable value; if the directional margin is zero, the candidate is not executed. Candidates and constraints all apply to the same variable and use the same units. When there is no historical response, the registered field-permitted trial increment that meets the minimum resolvable requirement of the equipment is used, and the same amplitude limit is applied. Amplitude limits ensure control constraints; actual improvement is still confirmed by a complete response evaluation. If a change in direction is required, it should be implemented after the current evaluation is completed, without simultaneously changing multiple variables and attributing the effect to one of them.

[0109] Specifically, each adjustment is constrained by the maximum permissible control value, the minimum permissible control value, the maximum change in a single period, the rate of change of continuous control, and the minimum evaluation time interval, as given by the following formula:

[0110] Among them, u j (t) represents the value of the j-th control variable in the t-th control cycle; u j,max with u j,min These are the maximum and minimum allowable control values ​​for this variable; Δu j,max The maximum change in a single period; r j,max Δt represents the upper limit of the rate of change of continuous control; Δt is the time interval between adjacent control cycles; Δt min The minimum evaluation time interval should be used. The actual interval between two adjacent actions should also cover the actual time from the issuance of the previous action to the completion of the full evaluation. If the evaluation is extended due to insufficient samples, the next action should not be issued in advance according to the original nominal waiting time.

[0111] Alternative implementation methods: In some embodiments, in addition to the location neighborhood search method based on motion prediction described above, the cross-frame association of bubbles can also adopt an association method based on appearance feature similarity. That is, based on the similarity of the equivalent diameter, area and boundary shape of the bubbles, candidate successor bubbles are searched in the next frame, and the one with the highest similarity is taken as the successor bubble. In other embodiments, location neighborhood search and feature similarity weighted combination can be used to improve the association stability when the bubble moves quickly or the image sampling interval is large.

[0112] In some embodiments, the image acquisition unit is further configured to dynamically adjust the image sampling interval based on the relationship between the bubble movement speed and the image sampling period, so that the bubble displacement in adjacent frames remains within a correlated range. Specifically, when the average bubble displacement in adjacent frames is greater than a preset upper displacement limit, the sampling interval is reduced; when the average bubble displacement in adjacent frames is less than a preset lower displacement limit, the sampling interval is increased, thereby reducing the computational load while ensuring the reliability of the correlation.

[0113] In some embodiments, the main transport loss areas are determined based on the distribution of the termination positions of the visible non-weir event chain, and the distribution of instability locations along the flow path is formed based on the merging and rupture statistics of different areas. This distribution serves as an auxiliary basis for screening control response records: when losses are concentrated in the area adjacent to the overflow weir, liquid level adjustment records are retrieved first; when losses are concentrated in the area within the tank, records of aeration or reagent addition adjustments are retrieved first. The specific positive or negative direction is still determined by the effective response of similar operating conditions, not solely by location.

[0114] In some embodiments, the stable operation baseline is not limited to the form of the mean and fluctuation scale of each indicator, but can also be in the form of a multivariate normal region. That is, in the feature space composed of multiple transport indicators along the route, a normal region boundary is determined based on historical data of the stable operation phase, and the transport status is determined to be normal based on whether the current state point is located within the boundary of the normal region.

[0115] End-to-end numerical examples: The following example uses a set of flotation scenario inputs to illustrate the state determination and control evaluation calculations of this method. All values ​​are exemplary calculation inputs.

[0116] An industrial vision acquisition device is installed 3.0m above the flotation cell, with a field of view covering the foam layer on the cell surface and the boundary of the overflow weir. The sampling frequency is 10 frames / second, and the resolution of a single frame is 1920×1080 pixels. After calibration, the overflow weir is on the right side of the image, and the scale of the area shown is 1.2mm per pixel.

[0117] Suppose that the following baseline input is obtained from the stable section using the aforementioned method: effective velocity v to the weir L The apparent transport residence time T is 8.5 mm / s. r The bubble growth rate along the path is 24.0 s. L The merging rate along the path is 0.012 mm / mm, R. c The rate is 0.35 times / m, and the degree of fracture along the friction is R. b The foam transport rate to the weir is 0.18 times / m. s The value is 0.86; the normal fluctuation scales of each indicator are 1.2 mm / s, 3.5 s, 0.004 mm / mm, 0.08 times / m, 0.05 times / m and 0.05, respectively. The stable segment can be collected using the aforementioned example condition of 30 consecutive minutes.

[0118] In the current scenario, the changes in average bubble diameter, foam coverage, and overall surface velocity are small. Relying solely on these static appearances or velocity moduli may not be sufficient to detect changes in weir transport.

[0119] In the transport variation scenario set in this example, the direction of foam movement changes from directly towards the weir to obliquely. The total velocity remains relatively unchanged, but the component reaching the weir decreases. The actual path length and apparent transport residence time increase, and the many-to-one correlation and unit distance merging within the path increase. The following calculations of this state are performed using given indices.

[0120] Given the current transport metric: v L =4.6mm / s, T r =38.5s, G L =0.026mm / mm, R c =0.62 times / m, R b=0.21 times / m, R s =0.68. The standardized deviations of each indicator and the overall degree of deviation are calculated using the aforementioned formula as follows: , , , , , Overall deviation

[0121] Under the condition that the above-mentioned abnormal index relationships continuously meet the valid window confirmation criteria, the effective velocity to the weir is lower than the lower limit of the normal range, and the apparent transport residence time and the degree of merging along the weir are higher than the upper limit of the normal range, thus triggering the stagnation and merging state. In this example, the standardized deviation of the degree of rupture along the weir is 0.60, which does not exceed twice the upper limit of the normal fluctuation scale, and therefore does not trigger the state of rupture along the weir.

[0122] Suppose that available historical response records show that the inflation rate was increased from 120 m³ / h to 132 m³ / h, a change of +12 m³ / h. Subsequently, the effective velocity at the weir rebounded, the apparent transport residence time and merging degree decreased, and the other candidate conditions mentioned above were met; the median response delay is 95 s. In this example, the inflation rate is selected as the control variable along the positive direction of this record. Assuming the single-cycle allowable change limit for this numerical calculation is 10 m³ / h, and the positive margin and allowable rate of change are both not less than 10 m³ / h, then the minimum allowable amplitude is taken according to S200, resulting in +10 m³ / h; this increment is added to the current inflation rate setpoint. This upper limit is a calculation condition; in actual use, it is provided by the equipment configuration.

[0123] This example gives τ j =95s, T before the action r =38.5s, α=0.8, the total waiting time T from the issuance of the action to the end of the scheduled evaluation. w =95 + 0.8 × 38.5 = 125.8s. The effective evaluation segment starts 95s after the action and ends at 125.8s, with a nominal length of 30.8s. If a sufficiently effective statistical window and an event chain that can determine the outcome have not yet been formed by this time, the evaluation is extended according to S190, and 125.8s is not considered a fixed duration that is sufficient under any working condition.

[0124] Assuming the following indicators are obtained after a complete and effective evaluation: v L =7.1mm / s, T r =28.6s, G L =0.016mm / mm, R c =0.44 times / m, R b =0.19 times / m, R s=0.80. The calculated standardized deviations for the six items are -1.17, 1.31, 1.00, 1.13, 0.20, and -1.20, respectively, with an overall deviation of D. after =2.61. The displayed value is rounded to two decimal places, and the overall deviation is calculated using the unrounded standardized deviation.

[0125] D is obtained from the above input. after =2.61, less than D before =8.04, indicating that this set of visual transport indicators is generally approaching the stable baseline. All six indicators have returned to within twice the normal fluctuation scale; if the lateral movement or backflow area ratio also meets the normal conditions, the current control settings will be maintained, and no further actions in the same direction will be added. The identified effective directions will only be saved as candidates for future anomalies.

[0126] The above calculations demonstrate that even in scenarios where the average bubble diameter or velocity modulus does not change significantly, stagnation can still be identified and the control response evaluated based on direction, path, and along-the-path event indicators. The change in the weir rate input from 0.68 to 0.80 in this example is solely for illustrating the evaluation process of visual indicators and does not constitute a field control test conclusion for conventional methods, nor is it equivalent to an improvement in mineral recovery rate or economic benefits.

[0127] It should be noted that the specific values, scenarios and parameters listed in the above embodiments are merely examples and do not constitute a limitation on the scope of protection of this invention. In practical applications, the relevant parameters can be adjusted according to the specifications of the flotation equipment, the properties of the ore, the production conditions and control requirements.

[0128] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0129] In one exemplary embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the program to perform the steps performed by the image processing and control device in the above-described intelligent control method for flotation process based on foam image recognition.

[0130] In one exemplary embodiment, a computer-readable storage medium is also provided, on which a computer program is stored, which, when executed by a processor, performs the steps performed by the image processing and control device in the above-described intelligent control method for flotation process based on foam image recognition.

[0131] In one exemplary embodiment, a computer program product is also provided, including a computer program that, when executed by a processor, performs the steps performed by the image processing and control device in the above-described intelligent control method for flotation process based on foam image recognition.

[0132] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. An intelligent control system for a flotation process based on foam image recognition, characterized in that, Includes the following steps: Acquire continuous images covering the foam surface of the flotation cell and the overflow weir area, establish the mapping relationship between image coordinates and flotation cell surface coordinates, and the direction to the weir with the overflow weir as the reference. The credibility of the continuous images is determined, and valid images are output. Based on the effective image, the local motion field of the foam surface is identified, the actual transport path of the foam to the overflow weir is determined, and cross-frame associations are established for bubbles in adjacent frames to form a path event chain that includes migration, merging, and bursting relationships. Based on the local motion field and the path event chain, path transport indices including the effective velocity to the weir, apparent transport residence time, degree of bubble growth along the path, degree of merging along the path, degree of bursting along the path, and foam transport to the weir rate are calculated. The foam transport status is determined by comparing the along-the-path transport index with the stable operating baseline of the current operating condition, and the degree of deviation of the current transport status from the stable operating baseline is determined according to the fluctuation scale corresponding to each of the along-the-path transport index. The adjustment direction and adjustment magnitude of the control variables are determined based on historical control response records similar to the current operating conditions; In the absence of the historical control response record, a small control change is implemented within the allowable adjustment range of the equipment to obtain the response direction under the current operating condition; after the control action is issued, the actual response is determined according to the change of the transport index along the path through the effective response window corresponding to the control action, and the adjustment direction and adjustment amplitude of the next control cycle are determined accordingly. Control commands are generated according to the adjustment direction and adjustment amplitude, and sent to at least one of the aeration volume adjustment actuator, reagent addition actuator and liquid level adjustment actuator of the flotation cell, to adjust at least one of the aeration volume, reagent addition amount and slurry level of the flotation cell.

2. The intelligent control system for flotation process based on foam image recognition according to claim 1, characterized in that, The process of identifying the local motion field on the foam surface and calculating the effective velocity of the weir includes: The positional changes of each local foam unit between adjacent image frames are identified. The positional change mapped to the coordinate system of the flotation cell surface is divided by the time interval between adjacent frames to obtain the motion vector of each local foam unit. The projection of the motion vector in the direction to the weir corresponding to the current position is determined as the effective velocity to the weir. When the magnitude of the motion vector is higher than the preset velocity condition and the effective velocity to the weir is lower than the preset effective velocity condition, the corresponding local foam unit will be identified as lateral motion or local backflow.

3. The intelligent control system for flotation process based on foam image recognition according to claim 1, characterized in that, The establishment of cross-frame associations to form a chain of events along the process includes: Predict the position of the bubble in the next image frame based on the local motion field, and search for candidate successor bubbles in the neighborhood of the predicted position. A migration relationship is established when there is a candidate successor bubble in the neighborhood of the predicted location and its position change direction is consistent with the local motion field. When multiple preceding bubbles correspond to a single subsequent bubble, and the relative deviation between the apparent area of ​​the subsequent bubble and the sum of the apparent areas of the multiple preceding bubbles is within a preset deformation tolerance range, a merging relationship is established. When multiple smaller successor bubbles exist within the motion prediction region of a preceding bubble, a rupture relationship is established. Based on the migration, merging, and breaking relationships, a process event chain with a time direction and allowing changes in bubble identifiers is generated.

4. The intelligent control system for flotation process based on foam image recognition according to claim 1, characterized in that, The calculation of the transport index along the route includes: For a chain of events along the spillway that has a valid successor state reaching the vicinity of the spillway, determine the time when it first enters the valid transport area and the time when it first arrives at the vicinity of the spillway, and determine the difference between the two as the apparent transport dwell time of the chain of events along the spillway. The cumulative actual movement distance of the event chain along the path is determined, and the degree of bubble growth along the path is determined based on the ratio of the change in the equivalent diameter of the bubble along the event chain to the cumulative actual movement distance. The degree of merging and the degree of breaking along the route are determined based on the number of merging events and the number of breaking events in the event chain along the route and the corresponding effective transport path length. The foam transport rate to the weir is determined by the ratio of the number of incident chains with valid successor states reaching the vicinity of the overflow weir in the same batch of entry to the number of incident chains with a definite transport outcome.

5. The intelligent control system for flotation process based on foam image recognition according to claim 1, characterized in that, Determining the foam transport state includes: When the effective velocity to the weir decreases, the apparent transport dwell time increases, and the degree of merging along the path increases, the current foam transport state is determined to be a stagnant and merged state. When the degree of frictional fracturing increases and the foam transport rate to the weir decreases, the current foam transport state is determined to be a frictional fracturing state. When the area of ​​local foam units identified as having lateral movement or local backflow within the effective transport area exceeds a preset proportion of the total area of ​​effective local foam units, the current foam transport state is determined to be a deflection and backflow state, where the preset proportion is 30% to 70%. When all the transport indicators along the route are within the fluctuation range of the stable operating baseline, and the area ratio of the local foam units with lateral movement or local backflow does not exceed the preset ratio, the current foam transport state is determined to be a normal transport state.

6. The intelligent control system for flotation process based on foam image recognition according to claim 1, characterized in that, The stable operating baseline is the baseline value and its fluctuation range formed based on the along-the-path transport index when the flotation process is in a stable operating phase. According to the fluctuation scale corresponding to each of the aforementioned transport indicators along the route, determine the degree of deviation of the current transport status from the stable operating baseline; In a new stable operating phase following continuous changes in the source of raw ore, ore feeding conditions, or operating conditions, a new stable operating baseline is formed that corresponds to the changed operating conditions.

7. The intelligent control system for flotation process based on foam image recognition according to claim 1, characterized in that, The determination of the actual response of the control action includes: Record the issuance time of the control action, and determine the earliest effective response time corresponding to the control action based on the time relationship between the historical control actions and the first continuous change of the transport index along the route; The duration of the effective response window is adjusted based on the apparent transport dwell time. After the effective response window ends, if the degree of deviation decreases compared to before the control action is issued, the adjustment direction of the control action is retained as a candidate direction for subsequent adjustment; if the degree of deviation increases compared to before the control action is issued, the adjustment amplitude of the subsequent control action is reduced or the subsequent adjustment direction is changed.

8. The intelligent control system for flotation process based on foam image recognition according to claim 1, characterized in that, The generation control instructions include: The adjusted value of the control variable is between the minimum and maximum allowable control values ​​of the control variable, and the adjustment amplitude meets the limits of the maximum change in a single cycle and the rate of change of continuous control. The time interval between two adjacent control actions is not less than the minimum evaluation time interval, and covers the actual time from the previous control action to the end of the complete response evaluation. When historical response data similar to the current operating condition is lacking, a control command is generated according to a control change amount less than a preset amplitude, and the response direction of the control variable under the current operating condition is determined based on the actual response result within the effective response window corresponding to the control command.

9. The intelligent control system for flotation process based on foam image recognition according to claim 1, characterized in that, The credibility determination includes: The area located within the pre-calibrated foam observation area, whose position remains stable at multiple consecutive sampling times in the camera's original image coordinate system and exhibits continuous blurring or occlusion, with visible continuous movement of surrounding foam, is designated as a candidate area for lens contamination. After excluding occlusion by the fixed tank structure and the bubble scraping mechanism, a contamination candidate mask is formed, and the area corresponding to the mask is removed from the foam event statistics. The area is not judged as lens contamination solely based on its stationary position or low movement speed. Based on the overall synchronous displacement of the fixed trough structure or the edge of the overflow weir in the image, the global motion caused by camera vibration is determined, and position compensation is performed on the current image; The area covered by the bubble-scraping mechanism is identified based on its geometric shape, continuous movement direction, and periodic characteristics, and the statistics of rupture and disappearance events are paused while the area is covered. When the effective bubble area ratio or visual credibility is lower than the available conditions, the most recent credible control command is maintained, and closed-loop control is resumed after visual credibility is restored and continuous effective images meet the preset confirmation conditions.

10. The intelligent control system for flotation process based on foam image recognition according to claim 1, characterized in that, The continuous images are acquired by an industrial vision acquisition device positioned above or to the side of the flotation cell, wherein the field of view of the industrial vision acquisition device covers at least a section of the foam transport area and the overflow weir boundary corresponding to that area. The credibility determination, local motion field identification, along-process event chain establishment, foam transport state determination, and actual response evaluation are implemented through an image processing and control device. Control commands are executed by the aeration volume adjustment actuator, reagent addition actuator, and liquid level adjustment actuator, which are communicatively connected to the image processing and control device. The actuators act on the aeration pipeline, reagent addition pipeline, and ore discharge or water replenishment pipeline of the flotation cell, respectively.

Images