A method for visual monitoring of a spray compounding process

Abstract

This invention discloses a visual monitoring method for a spray mixing process. This method relates to the field of image processing technology and includes the following steps: image acquisition, region segmentation to obtain droplet regions and background regions, mixing anomaly assessment, aggregation behavior assessment, dynamic uniformity monitoring, and graded control. The method divides the mixing stage into levels based on material mixing scores and droplet aggregation information, presets a target mixing score for each mixing stage level, and adjusts the spray pressure according to the deviation between the material mixing score and the target mixing score. It features accurate droplet contour identification through adaptive illumination segmentation, and the introduction of three-dimensional distance weights and color correction based on grid density gradients to locate mixing dead zones. It determines aggregation credibility based on contour overlap, movement rate, and density gradients to distinguish between stable homogeneous and over-mixing stages. The method dynamically adjusts the spray pressure according to the stage target score and its changing trend, improving monitoring accuracy and real-time control quality.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, specifically to a visual monitoring method for a spray mixing process. Background Technology

[0002] Spray mixing is a core process widely used in chemical, food, pharmaceutical, and agricultural fields. The uniformity of the mixture directly determines the product's performance, stability, and yield. For example, in pesticide formulation processing, uniform mixing of active ingredients and adjuvants is crucial to ensuring efficacy; in food additive preparation, the uniformity of spray mixing directly affects the product's taste and nutrient distribution. Therefore, achieving real-time and accurate monitoring of the spray mixing process is of great significance for optimizing process parameters and improving product quality.

[0003] Traditional visual monitoring relies mainly on sensor equipment, which can only acquire macroscopic process parameters such as spray pressure and flow rate. It lacks analysis of the microscopic behavior of droplets, such as droplet coalescence and local density distribution. It cannot analyze fine-grained issues such as mixing dead zones and excessive coalescence in local areas, which leads to the inability to accurately locate the specific location of uneven mixing, and the mixing state is prone to misjudgment, reducing the monitoring quality. At the same time, adjusting the spray pressure based only on static uniformity deviation does not take into account the changing trend of the mixing state, making it difficult to meet the real-time control requirements of dynamic mixing processes. Summary of the Invention

[0004] The purpose of this invention is to provide a visual monitoring method for spray mixing processes, which solves the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a visual monitoring method for a spray mixing process, comprising the following steps:

[0006] Step S1: Image acquisition. A sensor device is installed in the mixing chamber to acquire RGB images, depth maps, and light intensity of the mixing chamber. The sensor device is connected via a time synchronization trigger.

[0007] Step S2: Region segmentation. Based on the light intensity, and setting the light intensity update interval, the segmentation threshold is dynamically adjusted. The gray values ​​of the RGB image pixels are compared with the segmentation threshold. Based on the comparison result, the RGB image is divided into droplet region and background region.

[0008] Step S3: Mixing anomaly assessment. The monitoring area of ​​the mixing chamber is divided into several sub-regions according to an equidistant grid. The ratio of the droplet density difference to the distance difference between adjacent sub-regions is calculated to obtain the local density gradient of the droplets. The degree of material mixing anomaly risk is assessed based on the local density gradient of the droplets.

[0009] Step S4: Evaluation of droplet aggregation behavior. Based on the droplet region and depth map in the RGB image, the droplet contour is analyzed, and the ratio of the overlapping area of ​​adjacent droplet contours to their own area is calculated. Then, the movement rate of adjacent droplets is analyzed. Finally, the local density gradient of the droplet is introduced as an influencing factor to obtain the credibility of droplet aggregation behavior and to analyze whether the droplet aggregation behavior is effective.

[0010] Step S5: Dynamic monitoring of uniformity, calculating the static uniformity of droplet spatial distribution, and introducing the reliability of droplet coalescence behavior to dynamically adjust the static uniformity to obtain the material mixing score;

[0011] Step S6: Graded control, based on material mixing score and droplet coalescence information, divide the mixing stage into levels, preset the target mixing score for each mixing stage level, and adjust the spray pressure according to the deviation between the material mixing score and the target mixing score.

[0012] Optionally, in step S2, a standard segmentation threshold is first obtained through system preset. The grayscale values ​​of the RGB image are compared with the segmentation threshold to obtain an illumination ratio term, which is used to quantify the difference between real-time illumination and standard illumination. This ratio is then combined with a logarithmic function to obtain a logarithmic correction term, converting the linear illumination difference into a non-linear correction amount. Finally, an illumination sensitivity coefficient is introduced to control the magnitude of the standard segmentation threshold change with illumination intensity, ultimately obtaining the dynamic segmentation threshold T. adj ;

[0013] By analyzing the pixel grayscale values ​​in the RGB image, when the pixel grayscale value is greater than or equal to the dynamic segmentation threshold T... adj When the pixel grayscale value is less than the dynamic segmentation threshold T, it is identified as a droplet region. adj When the time is right, it is determined to be a background area.

[0014] Optionally, in step S3, the droplet distribution positions in the RGB image are first extracted, and the three-dimensional spatial coordinates of the droplets are obtained by combining the depth map. The three-dimensional spatial distance between adjacent droplets is calculated. Then, the droplet density difference in the sub-regions where adjacent droplets are located is calculated based on the droplet distribution positions in the RGB images in different sub-regions. The droplet density difference is compared with the three-dimensional spatial distance between adjacent droplets to obtain the spatial weighted density term.

[0015] Next, the color mean difference between adjacent droplets in the RGB image is calculated, and a color difference correction coefficient is introduced as a color difference correction term. The local density gradient is obtained by combining the spatial weighted density term and the color difference correction term, which reflects the local mixing uniformity of the material.

[0016] Optionally, in step S4, after each pixel of the RGB image is divided into a droplet region and a background region, contour extraction is performed: continuous edge pixels are converted into polygon contour coordinates through edge detection, and a single droplet contour is obtained through a data post-processing step.

[0017] Analyze the overlapping area of ​​the droplet profiles that exhibit coalescence behavior, compare the overlapping area with the area of ​​the droplet itself to obtain the profile overlap term, and determine the reliability of the coalescence behavior from the perspective of static morphology.

[0018] Further analysis of the droplet velocity during coalescence yields a motion trend term, validating the rationality of the coalescence behavior from the perspective of droplet dynamic motion.

[0019] Finally, the local density gradient is introduced as an influence term to obtain the confidence level of droplet aggregation behavior. When the confidence level of droplet aggregation behavior is ≥0.8, it is judged as effective aggregation behavior; when the confidence level of droplet aggregation behavior is <0.5, it is judged as invalid aggregation behavior; and when the confidence level of droplet aggregation behavior is 0.5≤0.8, it is judged as suspected aggregation behavior.

[0020] Optionally, when the droplet aggregation behavior is determined to be a suspected aggregation behavior, continuous frame tracking verification is performed, and the contour overlap change is calculated in the subsequent three frames. When the contour overlap term increases by more than 20% for three consecutive frames and the contour overlap term is ≥0.8, it is determined to be a valid aggregation behavior; otherwise, it is determined to be an invalid aggregation behavior.

[0021] Optionally, in step S5, firstly, the ratio of the number of droplets in each sub-region to the total number of droplets is analyzed to obtain the entropy value, which is used to reflect the static uniformity of the droplet distribution in space. Then, the ratio of effective aggregation behavior to the total number of droplets is calculated to obtain the aggregation rate. At the same time, the aggregation behavior influence coefficient is introduced to obtain the aggregation correction term. The entropy value is corrected by the aggregation correction term to obtain the material mixing score.

[0022] Optionally, in step S6, the mixing stage is divided into three levels based on the material mixing score and droplet coalescence information, resulting in the initial atomization period, rapid mixing period, and stable homogenization period.

[0023] A target mixing score is preset for each mixing stage level. The spray pressure is adjusted based on the deviation between the material mixing score and the target mixing score, as follows:

[0024]

[0025] In the above formula, ΔP t The amount of spray pressure adjustment over time t;

[0026] θ2 is the sensitivity adjustment coefficient;

[0027] E tar The target score for the material mixing stage is set, and the corresponding value is set according to different material mixing stages. The value is 0.6 when it is in the initial atomization stage, 0.8 when it is in the rapid mixing stage, and 0.95 when it is in the stable homogenization stage.

[0028] E t For the material mixing score at time t, (E) tar -E t () represents the scoring bias, used to quantify the gap between the current mixed state and the stage objective;

[0029] E t－1 For material mixing score at time t-1;

[0030] Δt is the time interval between two adjacent frames;

[0031] θ3 is the trend influence coefficient, ranging from 0 to 1, which controls the degree of influence of the trend on regulation. It is usually obtained by training multiple sets of samples, with an initial value of 0.3.

[0032] The mixed score change rate reflects the rate of change of the mixed state;

[0033] t time spray pressure adjustment amount ΔP t The maximum adjustment is 15%, when ΔP t When the value is greater than 0.15, the upper limit is 0.15. The system will recalculate and adjust it again in the next frame to ensure the stability of the spray mixing.

[0034] Optionally, data post-processing steps include noise removal, contour smoothing, and overlapping contour splitting;

[0035] The overlapping contour splitting step first examines the overlapping contours in the RGB image and, in conjunction with the Z-axis coordinate difference in the depth map, determines whether they are droplets in different spatial locations. When the depth difference is ≥2mm, they are determined to be independent droplets in different spatial locations, and at this point, they are split into two independent contours.

[0036] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0037] I. This invention dynamically adjusts the image segmentation threshold based on different lighting environments, thereby accurately identifying the outline of spray droplets. The monitoring area of ​​the mixing chamber is divided into several sub-regions according to an equidistant grid. The ratio of the distance difference to the density difference between droplets in adjacent sub-regions is calculated to obtain the local density gradient. The three-dimensional spatial distance between droplets is introduced to automatically amplify the weight of the density difference in nearby regions and reduce the influence of the density difference in distant regions. A color difference correction term is then introduced to correct the gradient value of droplet pairs with large color differences. Finally, the average gradient and the proportion of high gradients in the sub-regions are analyzed to determine whether the mixing is abnormal. This invention can effectively identify mixing dead zones in local areas, reduce misjudgments, and improve the accuracy of the results.

[0038] Second, this invention analyzes droplet contours based on droplet regions in RGB images, calculates the ratio of the overlapping area of ​​adjacent droplet contours to their own area, analyzes the movement rate of adjacent droplets, and finally introduces the local density gradient of droplets as an influencing factor to obtain the reliability of droplet aggregation behavior. It also analyzes whether droplet aggregation behavior is effective, which can determine whether the material is actually mixed. Furthermore, it analyzes the proportion of effective droplet aggregation behavior to determine whether excessive aggregation has occurred. This effectively distinguishes between the stable and uniform stages of material mixing and the stage of over-mixing, avoiding the problem of poor mixing effect caused by only looking at static results and ignoring the dynamic process, and improving the real-time performance and accuracy of uniformity assessment.

[0039] Third, this invention divides the mixing stage based on the material mixing score, and then presets the target mixing score for each mixing stage level. Based on the deviation between the material mixing score and the target mixing score, and by introducing the changing trend of the material mixing score, the spray pressure is dynamically adjusted, which can meet the real-time control requirements of the dynamic mixing process and improve the visual monitoring effect of the spray mixing process. Attached Figure Description

[0040] Figure 1 This is a flowchart of the method of the present invention;

[0041] Figure 2 This is a schematic diagram of the hierarchical control of the present invention. Detailed Implementation

[0042] 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.

[0043] For examples, please refer to Figure 1 and Figure 2 This embodiment provides a visual monitoring method for a spray mixing process, including the following steps:

[0044] Image acquisition: Sensor devices are installed inside the mixing chamber to acquire RGB images, depth maps, and light intensity of the mixing chamber. The sensor devices are connected through a time synchronization trigger to achieve time synchronization of multi-source data acquisition.

[0045] Region segmentation involves pixel-level alignment of the RGB image with the depth map; real-time adjustment of camera exposure compensation parameters based on ambient light data; and dynamic adjustment of the segmentation threshold based on illumination intensity data. Based on the segmentation threshold, the RGB image is divided into droplet regions and background regions. The region segmentation process is as follows:

[0046]

[0047] In the above formula, T adj This is a dynamic segmentation threshold used to distinguish between liquid and background regions in a region's RGB image.

[0048] T0 is the standard segmentation threshold, representing the base threshold under standard illumination, which is obtained through training with multiple sets of samples;

[0049] k is the light sensitivity coefficient, used to control the magnitude of the threshold change with light intensity. The value ranges from 0 to 1, with an initial value of 0.3, which was optimized through multiple sets of different liquid spraying experiments.

[0050] G real The real-time ambient light intensity is obtained through an ambient light sensor;

[0051] G0 is the standard ambient light intensity, which is obtained through system preset and represents the ideal light intensity for mixed lighting.

[0052] The illumination ratio is used to quantify the difference between real-time illumination and standard illumination, reflecting the extent to which the current ambient light intensity deviates from the ideal state. When the value of this item is <1, it is in a low-light environment, such as dark liquid spraying or cavity shading. When the value of this item is >1, it is in a high-light environment, such as light-colored liquid reflection or external strong light interference.

[0053] z is a correction value, which is 0.01, to avoid the case where the logarithmic function result is 0, which would cause calculation errors.

[0054] This is a logarithmic correction term that converts linear lighting differences into nonlinear correction values, which aligns with the camera's light-sensing characteristics. In low-light environments, this term is negative and changes more significantly, amplifying the threshold correction in low light conditions. In high-light environments, this term is positive and changes more gradually to avoid over-adjustment of the threshold in high light conditions.

[0055] Obtain the dynamic segmentation threshold T adj Next, the liquid and background in the RGB image are segmented. Because the grayscale value of the droplet region is generally higher than that of the background due to characteristics such as spray reflection and liquid color, a segmentation threshold T is used when the pixel grayscale value is greater than or equal to the dynamic segmentation threshold T. adj When the pixel grayscale value is less than the dynamic segmentation threshold T, it is identified as a droplet region. adj When the area is in bright light, it is considered a background area. In high-light environments, the logarithmic correction term is used. Since it is a positive number, the dynamic segmentation threshold T adj >Standard segmentation threshold T0, increasing the segmentation threshold filters high-brightness reflective noise. In low-light environments, the logarithmic correction term... Since it is a negative number, the dynamic segmentation threshold T adj<Standard segmentation threshold T0, reducing the segmentation threshold enhances the contour recognition capability of low-brightness droplets, thereby enabling dynamic adjustment of the threshold through real-time light intensity. Compared with the traditional fixed threshold, this improves the accuracy of region segmentation. Dynamic segmentation threshold T adj The system updates the data periodically according to the time intervals set by the system to ensure real-time data accuracy.

[0056] For the assessment of mixing anomalies, the monitoring area of ​​the mixing chamber is divided into several sub-regions according to an equidistant grid. The ratio of the density difference to the distance difference of droplets in adjacent sub-regions is calculated to obtain the local density gradient. The local mixing effect is then assessed based on the local density gradient. The process is as follows:

[0057]

[0058] In the above formula, G ij denoted as the local density gradient between droplet i and its adjacent droplet j, the larger the value, the higher the risk of uneven local mixing of materials;

[0059] P i The local droplet density of the sub-region where droplet i is located is obtained by calculating the ratio of the total number of droplets in the sub-region to the area of ​​the sub-region.

[0060] P j P represents the local droplet density in the subregion where droplet j is located. j -P i The density difference reflects the difference in the number of droplets in the regions where droplet i and droplet j are located. If the density difference P j -P i >0 indicates that the droplets in the sub-region where droplet j is located are more densely packed. Conversely, if the difference is negative, the droplets in the sub-region where droplet i is located are more densely packed, and droplets i and j do not belong to the same sub-region.

[0061] D ij Let be the three-dimensional spatial distance between droplet i and its adjacent droplet j, derived from the three-dimensional spatial coordinates of the two droplets. This is a spatially weighted density term, which correlates density difference with spatial distance. It indicates that the closer the droplets are, the greater the influence of density difference on the gradient. Since in spray mixing, large-scale density differences are typically seen in the spray diffusion process (e.g., low density at the spray edge), but small-scale density differences can easily lead to mixing dead zones. This term is expressed as the density difference value P. j -P i Divide by the three-dimensional spatial distance D between droplet i and its neighboring droplet j ij It automatically amplifies the weight of density differences in nearby regions and reduces the influence of density differences in distant regions, thereby reducing misjudgments and improving the accuracy of results;

[0062] β is the color difference correction coefficient, which is obtained through training with multiple sets of samples. Its value ranges from 0 to 1, with an initial value of 0.2. The greater the color difference of the liquid, the higher the value.

[0063] R i R is the average RGB color of droplet i. j Let be the average RGB color value of droplet j;

[0064] For color difference correction terms, gradient value correction is applied to droplet pairs with large color differences. The greater the difference between droplet i and droplet j, i.e., |R| > 0. i -R j The larger the value of |, the smaller the value of the correction term, thereby reducing the interference of color differences on density gradient calculation and avoiding segmentation errors caused by different liquid colors. For example, dark liquids may be misjudged as having low density, ensuring that the gradient value only reflects the actual droplet distribution differences, rather than visual imaging errors.

[0065] In spray mixing, the mixing dead zone usually appears in a small local space, such as the corner of the mixing chamber or near the nozzle. Traditional calculation of the global density mean is difficult to reflect the local unevenness. By dividing the mixing chamber into multiple sub-regions and analyzing the droplet density differences in the sub-regions, the accuracy of the results is improved. Furthermore, by introducing a color difference correction coefficient β, it can be adapted to complex scenarios such as multi-liquid mixing and highly reflective liquids, thus improving its applicability.

[0066] The local density gradient G between droplet i and its neighboring droplet j is obtained. ij Next, it is determined whether the sub-region is a mixing dead zone. Since the mixing process typically involves uneven distribution of droplets in a localized area, which cannot be homogenized through natural diffusion, it is necessary to calculate the average density gradient of all adjacent droplet pairs within the mixing chamber. If the number of adjacent droplet pairs in the sub-region is M, first calculate the density gradient of all adjacent droplet pairs in the sub-region, then divide by the number of adjacent droplet pairs M to obtain the average gradient G of the sub-region. avg ;

[0067] Next, we analyze the high gradient ratio in this sub-region. First, we obtain the number of droplet pairs with a local density gradient ≥ 0.7 in this sub-region, and compare it with the number of adjacent droplet pairs M to obtain the high gradient ratio GR in the sub-region.

[0068] Obtain the average gradient G of the sub-region avg When a region has a high gradient proportion (GR) within a sub-region, and the following conditions are met simultaneously, it is determined to be a mixed anomaly:

[0069] When the average gradient of the sub-region is G avg≥0.5 indicates that the overall density gradient of the region is high, exceeding the normal diffusion range, and the proportion of high gradient GR in the sub-region is ≥30%, indicating that the proportion of high gradient droplet pairs is too high. At this time, the sub-region is judged to be an abnormal mixing region.

[0070] For sub-regions identified as having mixed anomalies, continuous monitoring is performed. When the average gradient G of the sub-region is calculated over five consecutive calculations... avg If the fluctuation range is ≤10%, it indicates that the gradient value is stable and there is no natural diffusion trend. Alternatively, if the decrease in the proportion of high gradient GR in the sub-region is ≤5%, it indicates that the proportion of high gradient droplets has not decreased significantly. In this case, the sub-region is judged to be a mixing dead zone, and manual inspection should be notified immediately.

[0071] Clustering behavior evaluation is based on the analysis of droplet contours in the droplet region of the RGB image. Droplet contour extraction of the droplet region is performed by converting continuous edge pixels into polygon contour coordinates through edge detection and obtaining the individual droplet contour through data post-processing steps.

[0072] The data post-processing steps include noise removal, contour smoothing, and overlapping contour splitting. The overlapping contour splitting step first determines whether the overlapping contours in the RGB image are droplets in different spatial locations by combining the Z-axis coordinate difference of the depth map. When the depth difference is ≥2mm, they are determined to be independent droplets in different spatial locations, and at this time they are split into two independent contours.

[0073] After obtaining the droplet profile, the ratio of the overlapping area of ​​adjacent droplet profiles to their own area is calculated. Then, the movement velocities of adjacent droplets are analyzed. Finally, the local density gradient of the droplets is introduced as an influence term to obtain the confidence level of droplet coalescence behavior and analyze whether the droplet coalescence behavior is effective. The process is as follows:

[0074]

[0075] In the above formula, B ij The reliability of the coalescence behavior of droplets i and j is defined as follows: In the process of spray mixing, coalescence behavior is the process in which two or more independent droplets collide and come into contact during their movement, eventually merging into a larger droplet. It is the core of achieving material homogenization in spray mixing. The larger the value, the stronger the reliability, and it can basically eliminate misjudgment situations such as image overlap and noise interference.

[0076] S over Let be the area of ​​overlap between the outlines of droplets i and j;

[0077] S i S is the area of ​​the droplet i's own contour. j Let j be the area of ​​the droplet's own outline.

[0078] For the contour overlap term, the probability of aggregation behavior is determined from the perspective of static morphology. By using the ratio of the overlapping area of ​​the droplets to their own area, misjudgments caused by large droplets covering small droplets are avoided. When two droplets actually aggregate, the overlapping area will occupy a large proportion of their respective areas at the same time, and the result of this term is close to 1.

[0079] V i Let V be the velocity of droplet i. j Let j be the velocity of the droplet.

[0080] V max The maximum droplet velocity of the spray system is a preset parameter determined by the equipment performance.

[0081] The term represents the motion trend. It verifies the rationality of the aggregation behavior from the perspective of the dynamic motion of the droplets. If the difference in the motion speed of the two droplets is too large, such as one droplet moving backward and the other moving forward, the term approaches 0. Conversely, if the two droplets move in the same direction, the term approaches 1.

[0082] G ij Let be the local density gradient between droplet i and its neighboring droplet j;

[0083] γ is the local density influence coefficient of droplets, with a value ranging from 0 to 1 and an initial value of 0.5. It is obtained through system preset and the value of this coefficient is adjusted according to the local density gradient between droplets to reflect the influence of the risk of local mixing unevenness on the credibility of aggregation behavior. Droplets near the mixing dead zone are more likely to undergo abnormal aggregation. The aggregation judgment sensitivity in this area is increased by weighting terms to ensure that aggregation behavior in high-risk areas is not missed.

[0084] Agglomeration is a core process of material transfer and homogenization between droplets. Higher reliability of droplet aggregation behavior indicates more thorough droplet collisions and higher mixing efficiency; conversely, lower reliability may indicate mixing dead zones or unreasonable spray parameters. Accurately identifying true aggregation behavior improves the monitoring quality of the spray mixing process. When the reliability of the aggregation behavior between droplet i and droplet j is B... ij When the coefficient is ≥0.8, it is considered a valid aggregation behavior. When the aggregation behavior of droplet i and droplet j has a confidence level B ij When the value is less than 0.5, the contours overlap is small and the velocity difference is large, which is judged as invalid clustering behavior, such as droplet occlusion, splash collision or normal separation. Clustering behavior data judged as invalid clustering behavior will not be used for subsequent calculations to avoid interfering with the accuracy of subsequent results.

[0085] When 0.5 ≤ the confidence level of the coalescence behavior of droplet i and droplet j is B ijWhen the value is <0.8, it is judged as suspected clustering behavior, indicating that there is contour overlap between droplets, but the velocity difference is large or the local gradient has not reached the threshold. It may be delayed clustering or occlusion. At this time, continuous frame tracking verification is performed, and the contour overlap changes in the next three times are calculated. When the contour overlap term The growth exceeded 20% for three consecutive frames, and the outline overlap was significant. If the value is ≥0.8, it is considered a valid clustering action; otherwise, it is considered an invalid clustering action.

[0086] Uniformity is dynamically monitored by calculating the static uniformity of droplet spatial distribution and then dynamically adjusting the static uniformity by introducing droplet coalescence behavior confidence. The process is as follows:

[0087]

[0088] In the above formula, E t The value represents the material mixing score at time t. A higher value indicates a higher degree of material mixing uniformity and a better mixing effect, while a lower value indicates that the droplets are more concentrated and the mixing effect is worse.

[0089] n represents the total number of sub-regions, referring to the total number of all sub-regions within the mixing chamber;

[0090] PD y The droplet percentage in the y-th region is calculated by determining the ratio of the number of droplets in the sub-region to the total number of droplets.

[0091] This is the entropy term, used to measure the static uniformity of droplet spatial distribution. If the proportion of droplets in all sub-regions is equal, this term takes the maximum value of 1, indicating that the droplets are completely uniformly distributed. If the droplets are concentrated in a few sub-regions, this term approaches 0, indicating that the droplets are severely aggregated.

[0092] θ1 is the clustering behavior influence coefficient, ranging from 0.1 to 1, with an initial value of 0.2, obtained through sample training. It represents the degree of influence of clustering on the mixing effect. To determine the agglomeration rate, an agglomeration behavior influence coefficient θ1 can be introduced, which can be adjusted for different feed liquid characteristics: for high-viscosity feed liquids that are prone to agglomeration, the agglomeration behavior influence coefficient θ1 can be increased to improve the agglomeration rate. It has a greater impact on the entropy term; for low-viscosity liquids, it can reduce the influence coefficient θ1 of aggregation behavior and focus more on static distribution uniformity, thereby being able to cope with complex and ever-changing production environments and improve the adaptability to industrial scenarios.

[0093] N mer,t The total number of merging events at time t refers to the number of droplet pairs with all valid merging behaviors at time t;

[0094] N tot,t The total number of droplets at time t;

[0095] The clustering correction term is calculated using the clustering rate of the current frame. The entropy term is dynamically adjusted; a higher aggregation rate indicates more active droplet dynamic mixing, and a smaller entropy term value results in a lower entropy value. The greater the shrinkage, the lower the aggregation rate; conversely, the smaller the value of this term, the smaller its impact on the entropy term. In spray mixing, if the droplets have achieved static uniformity but still continue to aggregate in large quantities, it indicates that the material mixing has been overdone, and the spray intensity needs to be reduced. If the static distribution is uneven but aggregation is active, it indicates that mixing is in progress and no excessive intervention is needed; an aggregation correction term can be introduced. It effectively distinguishes between stable and uniform stages and over-mixed stages, avoiding the problem of poor mixing effect caused by only looking at static results and ignoring dynamic processes, and improving the real-time and accuracy of material mixing uniformity assessment.

[0096] Material mixing score E based on time t t And aggregation rate The mixing stage is divided as follows:

[0097] During the initial atomization period, the following must be met simultaneously: coalescence rate <0.2, material mixing score E for 3 consecutive frames at time t t <0.7;

[0098] During the rapid mixing period, the following condition must be met simultaneously: 0.2 ≤ aggregation rate. <0.6, material mixing score for 3 consecutive frames at time t: 0.7 ≤ E t <0.9;

[0099] For a stable homogeneous period to occur, the following conditions must be met simultaneously: aggregation rate ≥0.6, material mixing score E for 5 consecutive frames over time t t ≥0.9;

[0100] Stage switching is only allowed if multiple consecutive frames meet the entropy value range, filtering out abnormal fluctuations in a single frame to ensure the accuracy of the mixing stage division.

[0101] The process involves graded control, pre-setting target mixing scores for each mixing stage, and adjusting the spray pressure based on the deviation between the material mixing score and the target mixing score. The process is as follows:

[0102]

[0103] In the above formula, ΔP t This represents the spray pressure adjustment amount over time t. A positive value indicates that pressure needs to be increased, while a negative value indicates that pressure needs to be decreased.

[0104] θ2 is the adjustment sensitivity coefficient, which is dynamically adjusted based on the equipment's base pressure and spray characteristics. It is derived from multiple sample tests; the higher the equipment's base pressure, the higher the adjustment sensitivity coefficient should be. For example, if the equipment's base pressure is 11 bar, which falls within the high-pressure range, the adjustment sensitivity coefficient θ2 should be set to 2 to ensure the spray pressure adjustment ΔP at time t is within acceptable limits. t It can effectively change the mixing effect;

[0105] E tar The target score for the material mixing stage is set, and the corresponding value is set according to different material mixing stages. The value is 0.6 when it is in the initial atomization stage, 0.8 when it is in the rapid mixing stage, and 0.95 when it is in the stable homogenization stage.

[0106] E t For the material mixing score at time t, (E) tar -E t The score deviation is used to quantify the difference between the current mixing state and the stage target. A positive result indicates that the mixing has not met the target, and the droplet diffusion and coalescence effects need to be enhanced by increasing pressure. A negative result indicates that the mixing has been excessive, and the droplet coalescence needs to be reduced by decreasing pressure. To avoid minor adjustments, an absolute value of the score deviation, |E|, can be set. tar -E t Adjustment is only initiated when the pressure and flow rate exceed a preset threshold, thus avoiding frequent small changes in parameters such as nozzle pressure and flow rate, which could cause system oscillations and improve the stability of equipment operation.

[0107] E t－1 The material mixing score is given at time t-1, where Δt is the time interval between two adjacent frames.

[0108] θ3 is the trend influence coefficient, ranging from 0 to 1, which controls the degree of influence of the trend on regulation. It is usually obtained through training with multiple sets of samples, with an initial value of 0.3. The rate of change of the mixed score reflects the speed of change of the mixed state. The larger the value, the faster the change. A positive value indicates improvement of the mixture, and a negative value indicates deterioration of the mixture.

[0109] t time spray pressure adjustment amount ΔP t The maximum adjustment is 15%, that is, when ΔP t If the value is greater than 0.15, it will still be 0.15. Then the system will recalculate and adjust it again in the next frame to avoid problems such as liquid splashing and equipment impact caused by a single large adjustment, thereby improving the stability of industrial production.

[0110] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A visual monitoring method for a spray mixing process, characterized in that, Includes the following steps: Step S1: Image acquisition. A sensor device is installed in the mixing chamber to acquire RGB images, depth maps, and light intensity of the mixing chamber. The sensor device is connected via a time synchronization trigger. Step S2: Region segmentation. Based on the light intensity, and setting the light intensity update interval, the segmentation threshold is dynamically adjusted. The gray values ​​of the RGB image pixels are compared with the segmentation threshold. Based on the comparison result, the RGB image is divided into droplet region and background region. Step S3: Mixing anomaly assessment. The monitoring area of ​​the mixing chamber is divided into several sub-regions according to an equidistant grid. The ratio of the droplet density difference to the distance difference between adjacent sub-regions is calculated to obtain the local density gradient of the droplets. The degree of material mixing anomaly risk is assessed based on the local density gradient of the droplets. Step S4: Evaluation of droplet aggregation behavior. Based on the droplet region and depth map in the RGB image, the droplet contour is analyzed, and the ratio of the overlapping area of ​​adjacent droplet contours to their own area is calculated. Then, the movement rate of adjacent droplets is analyzed. Finally, the local density gradient of the droplet is introduced as an influencing factor to obtain the credibility of droplet aggregation behavior and to analyze whether the droplet aggregation behavior is effective. Step S5: Dynamic monitoring of uniformity, calculating the static uniformity of droplet spatial distribution, and introducing the reliability of droplet coalescence behavior to dynamically adjust the static uniformity to obtain the material mixing score; Step S6: Graded control, based on material mixing score and droplet coalescence information, divide the mixing stage into levels, preset the target mixing score for each mixing stage level, and adjust the spray pressure according to the deviation between the material mixing score and the target mixing score.

2. The visual monitoring method for the spray mixing process according to claim 1, characterized in that: In step S2, a standard segmentation threshold is first obtained through system preset. The grayscale values ​​of the RGB image are compared with the segmentation threshold to obtain an illumination ratio term, which is used to quantify the difference between real-time illumination and standard illumination. This ratio is then combined with a logarithmic function to obtain a logarithmic correction term, converting the linear illumination difference into a non-linear correction amount. Finally, an illumination sensitivity coefficient is introduced to control the magnitude of the standard segmentation threshold change with illumination intensity, ultimately yielding the dynamic segmentation threshold T. adj ; By analyzing the pixel grayscale values ​​in the RGB image, when the pixel grayscale value is greater than or equal to the dynamic segmentation threshold T... adj When the pixel grayscale value is less than the dynamic segmentation threshold T, it is identified as a droplet region. adj When the time is right, it is determined to be a background area.

3. The visual monitoring method for the spray mixing process according to claim 2, characterized in that: In step S3, the droplet distribution position in the RGB image is first extracted, and the three-dimensional spatial coordinates of the droplets are obtained by combining the depth map. The three-dimensional spatial distance between adjacent droplets is calculated. Then, the droplet density difference of the sub-region where the adjacent droplets are located is calculated based on the droplet distribution position in the RGB image in different sub-regions. The droplet density difference is compared with the three-dimensional spatial distance between adjacent droplets to obtain the spatial weighted density term. Next, the color mean difference between adjacent droplets in the RGB image is calculated, and a color difference correction coefficient is introduced as a color difference correction term. The local density gradient is obtained by combining the spatial weighted density term and the color difference correction term, which reflects the local mixing uniformity of the material.

4. The visual monitoring method for the spray mixing process according to claim 3, characterized in that: In step S4, after each pixel of the RGB image is divided into a droplet region and a background region, contour extraction is performed: continuous edge pixels are converted into polygon contour coordinates through edge detection, and a single droplet contour is obtained through a data post-processing step. Analyze the overlapping area of ​​the droplet profiles that exhibit coalescence behavior, compare the overlapping area with the area of ​​the droplet itself to obtain the profile overlap term, and determine the reliability of the coalescence behavior from the perspective of static morphology. Further analysis of the droplet velocity during coalescence yields a motion trend term, validating the rationality of the coalescence behavior from the perspective of droplet dynamic motion. Finally, the local density gradient is introduced as an influence term to obtain the confidence level of droplet aggregation behavior. When the confidence level of droplet aggregation behavior is ≥0.8, it is judged as effective aggregation behavior; when the confidence level of droplet aggregation behavior is <0.5, it is judged as invalid aggregation behavior; and when the confidence level of droplet aggregation behavior is 0.5≤0.8, it is judged as suspected aggregation behavior.

5. The visual monitoring method for the spray mixing process according to claim 4, characterized in that: When the droplet aggregation behavior is determined to be a suspected aggregation behavior, continuous frame tracking verification is performed, and the contour overlap change is calculated in the next three frames. When the contour overlap term increases by more than 20% for three consecutive frames and the contour overlap term is ≥0.8, it is determined to be a valid aggregation behavior; otherwise, it is determined to be an invalid aggregation behavior.

6. The visual monitoring method for the spray mixing process according to claim 5, characterized in that: In step S5, the ratio of the number of droplets in each sub-region to the total number of droplets is first analyzed to obtain an entropy value, which reflects the static uniformity of the droplet distribution in space. Then, the ratio of effective aggregation behavior to the total number of droplets is calculated to obtain the aggregation rate. At the same time, the aggregation behavior influence coefficient is introduced to obtain an aggregation correction term. The entropy value is corrected by the aggregation correction term to obtain the material mixing score.

7. The visual monitoring method for the spray mixing process according to claim 4, characterized in that: In step S6, the mixing stage is divided into three levels based on material mixing score and droplet coalescence information, resulting in the initial atomization period, rapid mixing period and stable homogenization period. A target mixing score is preset for each mixing stage level. The spray pressure is adjusted based on the deviation between the material mixing score and the target mixing score, as follows:

8. In the above formula, ΔP t The amount of spray pressure adjustment over time t; θ2 is the sensitivity adjustment coefficient; E tar The target score for the material mixing stage is set, and the corresponding value is set according to different material mixing stages. The value is 0.6 when it is in the initial atomization stage, 0.8 when it is in the rapid mixing stage, and 0.95 when it is in the stable homogenization stage. E t For the material mixing score at time t, (E) tar -E t () represents the scoring bias, used to quantify the gap between the current mixed state and the stage objective; E t－1 For material mixing score at time t-1; Δt is the time interval between two adjacent frames; θ3 is the trend influence coefficient, ranging from 0 to 1, which controls the degree of influence of the trend on regulation. It is usually obtained by training multiple sets of samples, with an initial value of 0.

3. The mixed score change rate reflects the rate of change of the mixed state; t time spray pressure adjustment amount ΔP t The maximum adjustment is 15%, when ΔP t If the value is greater than 0.15, it will still be 0.

15. The system will recalculate and adjust it again in the next frame to ensure the stability of the spray mixing.

9. The visual monitoring method for the spray mixing process according to claim 5, characterized in that: The data post-processing steps include noise removal, contour smoothing, and overlapping contour splitting. The overlapping contour splitting step first examines the overlapping contours in the RGB image and, in conjunction with the Z-axis coordinate difference in the depth map, determines whether they are droplets in different spatial locations. When the depth difference is ≥2mm, they are determined to be independent droplets in different spatial locations, and at this point, they are split into two independent contours.

Images