Uniformity online detection and control system and method

Abstract

The invention discloses a uniformity online detection and control system and method, and relates to the field of uniformity detection and control, the system comprises a characteristic parameter analysis module, a risk agglomeration determination module, an adhesion heat module and a control module, synchronous multi-source data is obtained, micro-areas are divided through droplet density analysis, and the uniformity of the micro-areas is determined; the method comprises the following steps of: establishing a plurality of micro-areas, respectively extracting data in each micro-area to calculate to obtain a physical characteristic parameter set, constructing an evolutionary hyperbola based on the physical characteristic parameter set of each micro-area, determining a high-risk aggregate source by searching a local large value of dynamic trigger strength in a high-potential interval, predicting an impact point based on the high-risk aggregate source to delimit a monitoring area, and determining a high-risk aggregate source according to the impact point. According to the method, a node adhesion heat distribution diagram is generated by judging and accumulating adhesion event intensity, response matrix control is executed according to the node adhesion heat distribution diagram to obtain a control strategy, and the risk of batch scrapping is avoided to the maximum extent.

Description

Technical Field

[0001] This invention relates to the field of uniformity detection and control, specifically to an online uniformity detection and control system and method. Background Technology

[0002] In powder mixing production in industries such as pharmaceuticals, food, and chemicals, it is often necessary to uniformly disperse extremely small but valuable or functionally critical liquid-phase active ingredients into large quantities of powder matrix materials to ensure the uniformity of content in the final product. However, in existing technologies, despite strictly adhering to set process parameters such as mixing time and speed, the final product frequently exhibits uneven distribution of active ingredients, resulting in hot spots with excessively high concentrations and cold spots with concentrations far below theoretical values, leading to the scrapping of entire batches. Traditional quality control methods, such as high-performance liquid chromatography (HPLC), typically involve offline sampling and analysis of the finished product after mixing. This "post-inspection" approach cannot achieve real-time monitoring and early warning during the process, nor can it trace the dynamic causes of problems for process optimization.

[0003] The root cause of this type of uniformity failure is not simply insufficient mixing, but a more complex failure mode involving dynamic adhesion and solidification. In the initial stages of mixing, the liquid additive forms viscous micro-agglomerates rich in active ingredients with some powder particles. Inside the mixer, intense friction and collisions between the powder and components such as the equipment walls and impellers generate a large amount of static electricity. These micro-agglomerates, with their high viscosity and unique charge characteristics, are easily attracted by static electricity, selectively adhering to "dead zones" such as the mixer's inner walls, the back of the impellers, and the discharge port. As mixing progresses, these adherents gradually solidify and harden under continuous shearing and drying. Finally, in the later stages of mixing or during discharge, they detach due to strong impact, randomly falling into the final product as hard lumps, thus disrupting the overall batch uniformity. Therefore, there is an urgent need for a system and method capable of online, real-time monitoring of this adhesion-solidification-detachment process, enabling early warning and intervention at the early stages of failure. However, currently, there is a lack of technical means to effectively identify this specific failure mode and implement closed-loop control.

[0004] To address the aforementioned shortcomings, a technical solution is provided. Summary of the Invention

[0005] To address the technical problems raised in the background section, this invention is proposed. Embodiments of this invention provide an online uniformity detection and control system and method.

[0006] The objective of this invention can be achieved through the following technical solution: an online uniformity detection and control method, comprising the following steps:

[0007] Acquire synchronous multi-source data, divide micro-regions through droplet density analysis, and extract data from each micro-region to calculate its set of physical characteristic parameters;

[0008] An evolutionary hyperbola is constructed based on the set of physical characteristic parameters of each micro-region, and the source of high-risk aggregates is determined by finding local large values ​​of dynamic triggering intensity in the high potential range.

[0009] The impact point is predicted based on high-risk aggregate sources to delineate the monitoring area, and the dual-modal signals collected within the monitoring area are analyzed. The node adhesion heat distribution map is generated by judging and accumulating the adhesion event intensity.

[0010] The control strategy is obtained by performing response matrix control based on the node adhesion heat distribution map.

[0011] Furthermore, the analysis steps for the physical feature parameter set are as follows:

[0012] Based on a subset of data with three-dimensional spatial information, texture structure features and information entropy analysis were performed to obtain the liquefaction entropy of each micro-region.

[0013] Droplet identification and binarization are performed on a subset of data based on three-dimensional spatial information to obtain a binarized three-dimensional image; volume counting is performed on the binarized three-dimensional image to obtain the initial droplet volume estimate for each micro-region;

[0014] The momentum contribution tensor is accumulated based on a subset of data containing three-dimensional velocity field data and three-dimensional spatial information to obtain the total momentum of all droplets in each micro-region;

[0015] Based on the data subset of triboelectric charge distribution image sequence and three-dimensional spatial information, time series data of internal charge peak decay over time are extracted. The time series data are then fitted to the exponential decay model using the nonlinear least squares method to solve for the relaxation time constant of each micro-region.

[0016] The liquefaction entropy, initial droplet volume estimate, total angular momentum of all droplets, and relaxation time constant of each microregion are collectively referred to as the set of physical characteristic parameters of each microregion.

[0017] Furthermore, the data subset analysis steps for the three-dimensional spatial information of each micro-region are as follows:

[0018] The continuous probability density field is transformed into a three-dimensional density surface. For the three-dimensional density surface, its Riemann tensor is calculated, and the arc length square of the differential displacement on the surface is calculated based on the Riemann tensor.

[0019] The geodesic distance on the three-dimensional density surface is defined based on the square of the arc length of the differential displacement on the surface. After initializing a set of generators on the three-dimensional density surface based on the geodesic distance, the Lloyd algorithm is used for iterative optimization to determine the final uniform irregular micro-region of droplet density.

[0020] Based on the final uniform irregular micro-regions of droplet density, the data and regions of the three-dimensional grayscale image data are matched and extracted to obtain a subset of the three-dimensional spatial information of each micro-region.

[0021] Furthermore, the continuous probability density field analysis steps are as follows:

[0022] Based on the three-dimensional grayscale image data obtained by laser-induced fluorescence scanning, the initial three-dimensional point cloud data of droplet dispersion is obtained by three-dimensional grayscale image data segmentation and centroid calculation; and it is then registered and synchronized in space and time with the three-dimensional velocity field data obtained by the particle image velocimetry system and the triboelectric charge distribution image sequence obtained by the imaging sensor array.

[0023] The initial droplet dispersion three-dimensional point cloud data is projected onto the target plane to obtain a two-dimensional discrete point set;

[0024] The kernel density estimation method is used to transform a two-dimensional discrete point set into a continuous probability density field.

[0025] Furthermore, the source analysis steps for the high-risk aggregates are as follows:

[0026] The structural incubation potential of each microregion is calculated based on the liquefaction entropy and initial droplet volume estimate of each microregion.

[0027] The dynamic triggering intensity value of each microregion is calculated based on the total angular momentum and relaxation time constant of each microregion.

[0028] Based on the structural incubation potential value and dynamic triggering intensity value of each microregion, an evolution hyperbola for each microregion is constructed, including the first structural incubation potential curve and the second dynamic triggering intensity curve.

[0029] The high potential range is locked in the first structure incubation potential curve. Based on the high potential range, the local large value of the second dynamic trigger intensity curve is found to determine the existence of resonance amplification effect. The micro-region corresponding to the local large value is obtained to obtain the high-risk aggregate source.

[0030] Furthermore, the analysis steps for the node adhesion heat distribution map are as follows:

[0031] The baseline difference and local minimum detection method is used to obtain the negative pulse amplitude for instantaneous temperature change signals;

[0032] High-pass filtering combined with signal energy integration is used to obtain the high-frequency energy bursting intensity of the stress wave signal;

[0033] Based on the negative pulse amplitude and high-frequency energy burst intensity, the temperature change condition and energy burst condition are judged respectively to obtain the adhesion event determination; for each monitoring node, when a valid adhesion event is detected on each monitoring node, the intensity of the valid adhesion event will be added to the existing cumulative intensity value of the corresponding node to classify the adhesion heat level and obtain the node adhesion heat distribution map.

[0034] Furthermore, the analysis steps for the instantaneous temperature change signal and stress wave signal are as follows:

[0035] The maximum deviation is calculated based on the centroid of the average impact position of each cluster, yielding the farthest distance of each cluster from its average impact position centroid. This farthest distance is then used as the base radius, with an additional safety factor added to obtain the uncertainty radius. Based on the radius of uncertainty Define each cluster as A circular area with radius [radius] is merged to obtain the inner wall monitoring area;

[0036] Based on the inner wall monitoring area, a uniform grid strategy is used to discretize the continuous monitoring surface to obtain a dual-modal sensor array; based on the dual-modal sensor array, instantaneous temperature change signals and stress wave signals are collected.

[0037] Furthermore, the centroid analysis steps for the average impact location of each cluster are as follows:

[0038] The geometric center of the high-risk agglomerate source is obtained as the initial position and its total moment of momentum is obtained to construct a momentum tensor. The momentum tensor is decomposed into eigenvalues ​​to determine the eigenvector corresponding to the largest eigenvalue as the main motion direction. A ray is emitted from the initial position along the main motion direction and its intersection with the inner wall of the equipment is calculated as the predicted inner wall impact point. Finally, the predicted inner wall impact points calculated from all high-risk agglomerate sources are collected to form a set of predicted inner wall impact points.

[0039] Based on the predicted set of impact points on the inner wall, the average impact position centroid of each cluster is obtained through data clustering and centroid calculation.

[0040] Furthermore, the control strategy analysis steps are as follows:

[0041] Perform a targeted, real-time physical state verification based on the node adhesion heat map to obtain the real-time state vector;

[0042] Two-dimensional response matrix control is executed based on instantaneous state vectors to obtain the control strategy.

[0043] A uniformity online detection and control system, the system comprising:

[0044] The feature parameter analysis module acquires synchronous multi-source data, divides micro-regions through droplet density analysis, and extracts data from each micro-region to calculate its physical feature parameter set;

[0045] The risk clustering determination module constructs an evolution hyperbola based on the physical characteristic parameter set of each micro-region, and determines the source of high-risk clusters by finding local large values ​​of dynamic triggering intensity in the high potential range.

[0046] The adhesion heat module predicts impact points based on high-risk aggregate sources to delineate monitoring areas, analyzes dual-modal signals collected within the monitoring areas, and generates a node adhesion heat distribution map by judging and accumulating the intensity of adhesion events.

[0047] The control module executes response matrix control based on the node adhesion heat distribution map to obtain the control strategy.

[0048] Compared with the prior art, the beneficial effects of the present invention are:

[0049] This invention integrates a high-sensitivity image sensor to collect laser-induced fluorescence and particle image velocimetry data in real time during the mixing process. Utilizing the high spatiotemporal resolution signals acquired by the image sensor, it performs texture structure feature and information entropy analysis, achieving non-contact online precision detection of material mixing uniformity and micro-agglomeration risk. This effectively overcomes the technical shortcomings of traditional contact sensors, such as susceptibility to interference and detection lag. By dividing the material into micro-regions through droplet density analysis and extracting data from each micro-region to calculate its physical characteristic parameter set, an evolutionary hyperbola is constructed based on the physical characteristic parameter set of each micro-region. High-risk agglomeration sources are identified by finding local maxima of dynamic triggering intensity within high-potential regions. Impact points are predicted based on these high-risk agglomeration sources to delineate the monitoring area, and the collected dual-modal signals within this area are analyzed. A node adhesion heat map is generated by judging and accumulating the intensity of adhesion events. Through the fusion of multi-source data such as laser-induced fluorescence, particle image velocimetry, and high-speed charge imaging, a comprehensive and precise quantification of the physical state of the initial stage of droplet spraying is achieved. More importantly, this invention proposes an analytical model for structural incubation potential and dynamic triggering intensity. By constructing an evolutionary hyperbola, it can accurately predict the formation of high-risk aggregate sources from complex initial data, thus identifying risks at the very beginning of a problem. This cause-based rather than effect-based diagnostic approach enables the system to anticipate failures rather than wait for them to occur, providing a solid data foundation for truly predictive maintenance and process optimization.

[0050] This invention employs response matrix control based on the nodal adhesion heat distribution map to derive a control strategy. This not only predicts the formation of high-risk agglomerates but also calculates their potential impact points on the equipment's inner wall, guiding dual-modal sensors (thermal and acoustic) for targeted and efficient deployment and monitoring. Upon confirming an adhesion event, the system does not simply issue an alarm but introduces secondary verification using ultrasonic impedance to accurately determine the physical state of the adhered material. Based on the instantaneous state vectors of "adhesion heat" and "hardness index," the system executes the most appropriate control strategy through a two-dimensional response matrix, achieving precise graded responses from "marking and monitoring" to "planned cleaning" and even "immediate intervention." This complete closed-loop logic of "prediction-detection-verification-action" ensures the timeliness, accuracy, and efficiency of intervention measures, minimizing batch spoilage risks and transforming the powder mixing process from an experience-dependent "black box" operation into a transparent, controllable, and adaptive intelligent production process. Attached Figure Description

[0051] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. The following drawings are not drawn to scale according to the actual size, but are intended to illustrate the main idea of ​​the present invention.

[0052] Figure 1 This is a flowchart of an online uniformity detection and control method according to the present invention;

[0053] Figure 2 This is a block diagram of an online uniformity detection and control system according to the present invention;

[0054] Figure 3 This is a flowchart for determining high-risk aggregate sources according to the present invention. Detailed Implementation

[0055] The technical solutions in 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 also within the scope of protection of the present invention.

[0056] Example 1:

[0057] like Figure 1 As shown, an online uniformity detection and control method includes:

[0058] Step S1: Acquire synchronous multi-source data, divide micro-regions through droplet density analysis, and extract data from each micro-region to calculate its physical characteristic parameter set.

[0059] It should be noted that the synchronized multi-source data includes three-dimensional grayscale image data, initial droplet dispersion three-dimensional point cloud data, three-dimensional velocity field data, and triboelectric charge distribution image sequence.

[0060] Specifically, step S1 also includes:

[0061] Step S11: Based on the three-dimensional grayscale image data obtained by laser-induced fluorescence scanning, the initial droplet dispersion three-dimensional point cloud data is obtained by three-dimensional grayscale image data segmentation and centroid calculation; and it is precisely registered and synchronized in space and time with the three-dimensional velocity field data obtained by the particle image velocimetry system and the triboelectric charge distribution image sequence obtained by the high charge imaging sensor array.

[0062] Directly below the liquid phase nozzle, two key non-contact optical diagnostic systems are deployed simultaneously: a laser-induced fluorescence (LIF) imaging system and a particle image velocimetry (PIV) system. The LIF imaging system specifically involves adding a specific concentration of fluorescent dye, such as rhodamine, to the liquid used for spraying. A laser beam shaped into a thin sheet is used to illuminate a spray field, exciting droplets within a two-dimensional plane. A high-speed camera, perpendicular to the laser sheet, captures the fluorescence signals emitted by the droplets on this two-dimensional slice. The acquired single-frame fluorescence images are processed, and an image segmentation algorithm is used to identify each individual fluorescence spot and calculate its centroid position in the image coordinate system. A high-speed scanning device is used to move the laser sheet along a path perpendicular to itself. The high-speed camera rapidly sweeps across the entire spray volume along the axial direction. During this process, the high-speed camera synchronously acquires a series of different data along with the scanning motion. Two-dimensional slice images of the location. Each acquired frame of the original two-dimensional slice image is preprocessed to obtain a two-dimensional grayscale slice image. This preprocessing step typically includes subtracting camera dark current and background noise, performing flat-field correction to eliminate the influence of optical path unevenness, and finally mapping the high-bit-depth original intensity values ​​to a normalized single-channel grayscale image, such as an 8-bit or 16-bit grayscale image. This series of two-dimensional grayscale slice images arranged in depth order are stacked along the z-axis to obtain three-dimensional grayscale image data. This data is essentially a three-dimensional array, where each element stores the fluorescence intensity value at the corresponding spatial location. Based on this three-dimensional grayscale image data, three-dimensional image segmentation algorithms, such as connected component analysis, are used to identify and separate the high-brightness regions representing individual droplets, i.e., the morphology of the fluorescent spots in three-dimensional space. For each segmented three-dimensional region, by calculating its centroid, the spatial position of the droplet in the three-dimensional Cartesian coordinate system can be accurately obtained. Where i is the index used to identify each droplet found. Through the above steps, the centroid coordinates of all identified droplets are summarized to obtain the initial 3D point cloud data of the droplet dispersion. This dataset consists of a large number of points... Composition, where each point This represents the i-th droplet, signifying an independent droplet. n represents the total number of droplets, including their unique three-dimensional spatial coordinates. Meanwhile, the Particle Image Velocimetry (PIV) system tracks tiny tracer particles suspended in the airflow, capturing two images in a very short time interval. By analyzing the displacement of the tracer particle swarm in these two images, it calculates a dense velocity vector covering the entire measurement area. After three-dimensional reconstruction, it obtains a three-dimensional velocity field that is completely synchronized with the three-dimensional grayscale image data in terms of spatial coordinates and time points. Each point in the velocity field... Both correspond to three-dimensional velocity vectors , representing the instantaneous velocity of the background airflow at that point, where It is a three-dimensional velocity vector In a Cartesian coordinate system, the components are along three mutually perpendicular coordinate axes. Simultaneously, a high-speed charge imaging sensor array deployed in the region below the liquid-phase nozzle of the mixer obtains a high-temporal-resolution sequence of triboelectric charge distribution images at the instant of spray initiation, for example, within the first 500 milliseconds. This triboelectric charge distribution image sequence is precisely spatially registered with both 3D grayscale image data and PIV data to ensure accuracy in the three 3D data volumes. The coordinates point to the same physical point in space, and the images are acquired at the same instant to ensure temporal synchronization. This image sequence records the charge distribution generated on a microsecond scale and its change over time when droplets impact a powder bed.

[0063] Step S12: Project the initial droplet dispersion 3D point cloud data onto the target plane to obtain a 2D discrete point set;

[0064] The kernel density estimation method is used to transform a two-dimensional discrete point set into a continuous probability density field;

[0065] The continuous probability density field is transformed into a three-dimensional density surface. For the three-dimensional density surface, its Riemann tensor is calculated, and the arc length square of the differential displacement on the surface is calculated based on the Riemann tensor.

[0066] The geodesic distance on the three-dimensional density surface is defined based on the square of the arc length of the differential displacement on the surface. After initializing a set of generators on the three-dimensional density surface based on the geodesic distance, the Lloyd algorithm is used for iterative optimization to determine the final uniform irregular micro-region of droplet density.

[0067] The initial droplet dispersion 3D point cloud data is projected onto a target plane, such as the impact surface, by precisely placing a calibration object with known geometric features at the intended location on the impact surface. The calibration object, such as a planar plate printed with a checkerboard or dot array, is scanned once or multiple times using a LIF scanning system. Since the feature points on the calibration plate, such as the checkerboard corners, are physically coplanar, their 3D coordinates can be extracted from the obtained 3D point cloud data. By performing a mathematical plane fitting on these feature points, for example using the least squares method, the precise mathematical equation of the plane in the measurement coordinate system is calculated. This equation defines the impact surface, resulting in a 2D discrete point set. The kernel density estimation method is applied to transform a two-dimensional discrete point set into a continuous probability density field. Its mathematical expression is:

[0068] ;

[0069] in, It is the total number of points; summation is performed by iterating through all data points. , It is a Gaussian kernel function. It is the bandwidth parameter that determines the smoothness. The value represents any point on the plane. The droplet distribution density in the vicinity transforms the continuous probability density field into a three-dimensional density surface, which is defined as follows: On this curved surface, and These are spatial coordinates, and The height value of the coordinate directly corresponds to the droplet density at that point. For example, for... The defined three-dimensional density surface is used to calculate its Riemann tensor g, which is derived from... It consists of symmetric positive definite matrices. The square of the arc length of the differential displacement on the surface is calculated based on the Riemann tensor. :

[0070] ;

[0071] The components of the metric tensor are derived from the continuous probability density field. The partial derivatives are calculated to obtain, ,in , and These are the components of the Riemann tensor g, specifically In mathematics, this precisely describes the local geometry of a surface. In regions with large density gradients, the component of g increases, thus increasing the arc length along that direction. It was stretched out.

[0072] The square of the arc length of the differential displacement on the three-dimensional density surface The square root yields the arc length infinitesimal element. , along the connection and Integrate along a specific path to obtain the length of that path. In all connections... and Among the possible paths, the minimum integral length corresponds to the geodesic distance between the two points. A set of generators is randomly selected within the spray influence area, denoted as set. Where a0 is the total number of generators, and also the total number of sub-regions to be divided. The three-dimensional density surface is divided into... 0 sub-regions , Where i0 represents the index, used to refer to a specific element in set G, and each region Both with generator Related, including all points on the surface that satisfy the following conditions :point arrive The geodesic distance is less than or equal to its distance to any other generator. The geodesic distance, where i1 represents the index. Its mathematical definition is: This yields initially uniform, irregular micro-regions of droplet density. The boundaries of these micro-regions are formed by two or more points with equidistant geodesic distances from the generators. Point arrive geodesic distance, Point arrive The geodesic distance. These micro-regions are on the physical plane ( - On a plane, the micro-regions exhibit irregular shapes and varying areas. However, in regions with high droplet density, the physical area of ​​the micro-regions is smaller, while in regions with sparse density, the physical area is larger, thus achieving an adaptive and high-fidelity discretization of the spray field.

[0073] Based on the initially uniform and irregular micro-regions of droplet density and a randomly selected set of generators, the final uniform and irregular micro-regions of droplet density are obtained using the Lloyd algorithm. Specifically, the Lloyd algorithm analysis steps are as follows: For each initially uniform and irregular micro-region of droplet density... The centroid is calculated on a three-dimensional density surface, and the generator elements of each region are calculated. Moving to the newly calculated centroid position in this region yields a new set of generators. Based on this new set of generators, the above operation is repeated to reanalyze the initially uniform and irregular micro-regions of droplet density. This process is continued until the maximum number of iterations for the generators reaches its upper limit, such as 100 or 500 iterations, resulting in the final uniform and irregular micro-regions of droplet density. These uniform and irregular micro-regions become highly homogeneous on the three-dimensional density surface, achieving the goal of adaptive and high-fidelity discretization.

[0074] Step S13: Based on the final uniform irregular micro-regions of droplet density, perform precise matching and extraction of data and regions on the three-dimensional grayscale image data to obtain a data subset of the three-dimensional spatial information of each micro-region;

[0075] Based on a subset of data with three-dimensional spatial information, texture structure features and information entropy analysis are performed to obtain the liquefaction entropy of each micro-region.

[0076] Specifically, each irregular micro-region defined on a two-dimensional plane is used as its projection base in the three-dimensional space of the three-dimensional grayscale image data. This two-dimensional projection base is stretched along a direction perpendicular to the plane, i.e., the z-axis, so that it penetrates the entire thickness of the three-dimensional grayscale image data, resulting in a three-dimensional columnar region with an irregular bottom shape. From the original, complete three-dimensional grayscale image data, all voxels that fall exactly within this columnar region and their corresponding grayscale values ​​are cut and extracted. Each two-dimensional micro-region obtains a unique data subset containing real three-dimensional spatial information. For the isolated three-dimensional data subset, its three-dimensional grayscale co-occurrence matrix is ​​calculated. This matrix is ​​essentially a detailed statistical table that records in detail the frequency of the occurrence of pairs of grayscale values ​​between any two voxels separated by a specific distance and direction within this data subset. For example, it counts how many times the situation occurs where a voxel with grayscale value b is paired with a voxel with grayscale value c to its right. By calculating multiple three-dimensional grayscale co-occurrence matrices with different directions and distances and averaging them, a comprehensive statistical matrix of the texture structure features within the region is obtained. The liquefaction entropy of each micro-region is calculated based on the gray-level co-occurrence matrix derived from this statistics. Specifically, the three-dimensional gray-level co-occurrence matrix is ​​normalized, meaning the sum of all elements in the matrix is ​​equal to 1. Each value in the matrix... This converts frequency into probability, representing grayscale value pairs. The probability of it appearing within this micro-region. Applying the classic formula of information entropy. The liquefaction entropy of each microregion This entropy value quantifies the uncertainty or complexity of the gray value combinations within the micro-region. A high liquefaction entropy value means that the combinations of gray value pairs within the region are very diverse and unpredictable. For example, droplets and gases may violently intertwine and break into a large number of fine structures, corresponding to a highly mixed and non-uniform physical state. Conversely, a low liquefaction entropy value indicates that the texture within the region is more ordered and homogeneous, and the combinations of gray value pairs are concentrated in a few cases, such as large areas of pure liquid or large areas of pure gas.

[0077] Step S14: Perform droplet identification and binarization on a subset of data based on three-dimensional spatial information to obtain a binarized three-dimensional image; perform voxel counting based on the binarized three-dimensional image to obtain the initial droplet volume estimate for each micro-region;

[0078] The momentum contribution tensor is accumulated based on a subset of data containing three-dimensional velocity field data and three-dimensional spatial information to obtain the total momentum of all droplets in each micro-region;

[0079] Based on the data subset of triboelectric charge distribution image sequence and three-dimensional spatial information, time series data of internal charge peak decay over time are extracted. The time series data is fitted to the exponential decay model by nonlinear least squares method, thereby solving for the relaxation time constant of each micro-region.

[0080] The liquefaction entropy, initial droplet volume estimate, total angular momentum of all droplets, and relaxation time constant of each microregion are collectively referred to as the set of physical characteristic parameters of each microregion.

[0081] By analyzing the grayscale histogram of the image or based on prior physical knowledge, a suitable grayscale threshold is selected as the boundary. Each voxel within the subset of 3D spatial information data of each micro-region is traversed. If the voxel's grayscale value is greater than or equal to this threshold, it is identified as a droplet voxel and assigned a new value, such as 1. If the voxel's grayscale value is less than this threshold, it is identified as a background voxel and assigned a new value, such as 0. Within this micro-region, a binary 3D image containing only 0s and 1s is obtained. The total number of voxels marked as 1 within the current micro-region is counted to obtain the initial droplet volume estimate. The continuous grayscale field is transformed into a discrete set composed of a large number of droplet voxels with precise 3D coordinates and voxels marked as 1. Based on the previously defined 2D irregular micro-region, it is extended vertically to form a 3D columnar analysis region. For a specific micro-region... Traverse all droplet voxels within its three-dimensional analysis region. For any droplet voxel e, record its three-dimensional spatial coordinates. Using this coordinate, interpolation is performed in the synchronized PIV three-dimensional velocity field data to obtain the instantaneous velocity vector at that location. ,in They are respectively droplet serotonin e in The velocity component in direction. Based on the key physical assumptions, the droplet is small enough to follow the background airflow well, so the airflow velocity can approximate the droplet velocity. Each identified droplet voxel is considered a unit contribution. Instantaneous velocity vector based on a single droplet voxel. Calculate its contribution to the total angular momentum, where angular momentum is a... The tensor describes the distribution and propagation of momentum in different directions. Treating the contribution of each voxel as 1, its expression is: ,in Representing the tensor cross product, specifically written in matrix form, it involves converting the instantaneous velocity vector... Multiply by its own transpose. Add the angular momentum contribution tensors of all droplets within the microregion d one by one to obtain the sum. The matrix represents the total angular momentum of all droplets within a micro-region d. It comprehensively reflects the magnitude of the total kinetic energy of the droplet group within that region, the degree of concentration or dispersion of their motion direction, and the intensity of momentum exchange.

[0082] For a specific micro-region To calculate its relaxation time constant, the triboelectric charge distribution image sequence of this micro-region was extracted from a subset of data based on three-dimensional spatial information in this image sequence. The extraction steps are the same as those analyzed above. Specifically, at each time point in the sequence... Analyze the micro-regions in the image Find all pixels and identify the maximum charge value among them; this value is denoted as the maximum charge value of the micro-region. Peak charge at time Arranging the charge peaks at all time points in chronological order yields the description of the micro-region. The set of data points showing how the peak internal charge decays over time. A mathematical model is applied to quantify this decay process, and its mathematical expression is as follows: In this formula, Represents any time The peak charge, It is the initial peak value at the start of decay. It is the baseline value of the final stable state reached by the charge, while The physical quantity we are looking for, the relaxation time constant, characterizes the rate of decay. To solve for the micro-region... relaxation time constant A nonlinear least squares numerical calculation method is employed. This method matches the previously extracted charge peak time series data points with an exponential decay model, and automatically adjusts the parameters in the model using an algorithm. , and The algorithm uses three parameters until it finds a theoretical curve that passes through all data points with minimal error. At this point, the algorithm provides the optimal result. The value is that of the micro-region. The relaxation time constant. By analyzing each micro-region By repeating this complete data extraction and curve fitting process, a set of relaxation time constants is obtained. Each constant value corresponds to a specific micro-region, thereby realizing the quantification of the charge relaxation characteristics of each micro-region.

[0083] A method for online detection and control of uniformity further includes:

[0084] like Figure 3 As shown, step S2: construct an evolutionary hyperbola based on the set of physical characteristic parameters of each micro-region, and determine the source of high-risk aggregates by finding local large values ​​of dynamic triggering intensity in the high potential range.

[0085] Specifically, step S2 also includes:

[0086] Step S21: Calculate the structural incubation potential value of each microregion based on the liquefaction entropy and initial droplet volume estimate of each microregion;

[0087] The dynamic triggering intensity value of each microregion is calculated based on the total angular momentum and relaxation time constant of each microregion.

[0088] Based on each micro-region Calculation of liquefaction entropy and initial droplet volume estimates for each micro-region The structural potential value is calculated using the formula: structural potential. , by micro-region liquefaction entropy Compared with the initial droplet volume estimate The data is constructed and then normalized by the maximum-minimum value to normalize the values. Removing the influence of dimensions, this represents the material basis for the formation of aggregates in micro-regions. High liquefaction entropy indicates that the internal material distribution of this region is chaotic and of varying sizes, suggesting the possibility of forming unstable large core structures. High initial droplet volume estimates mean that this region has sufficient material reserves to form a sizable aggregate. Only when a region is both chaotic and rich in material does its structural potential truly become high. Based on each micro-region... Calculation of total angular momentum and relaxation time constant for each micro-region Dynamic trigger strength value It is determined by the norm of total angular momentum. With relaxation time constant The data is constructed and then normalized by the maximum-minimum value to normalize the values. The dimensionless value is used to characterize the intensity and duration of the energy injection that causes aggregation. A high total angular momentum means that there is a violent velocity gradient and momentum exchange within the region, and the kinetic energy conditions for particle collisions and merging are sufficient. A high relaxation time constant means that this unstable dynamic (such as vortex, collision) lasts for a long time and is not fleeting. There is enough time to complete the physical process from collision to stable combination. Only when the energy exchange in a region is both intense and persistent is its dynamic triggering intensity truly threatening.

[0089] Step S22: Construct evolution hyperbolas for each microregion based on its structural gestation potential and dynamic triggering intensity values, including a first structural gestation potential curve and a second dynamic triggering intensity curve. Arrange multiple microregions in ascending order according to their structural gestation potential values ​​from low to high. Construct the first structural gestation potential curve with the microregion number as the abscissa and the structural gestation potential value of the microregion corresponding to each number as the ordinate. Construct the second dynamic triggering intensity curve with the same microregion number as the abscissa and the dynamic triggering intensity value of the microregion corresponding to each number as the ordinate.

[0090] All D microregions are arranged in ascending order according to their structural potential values, from low to high. This sorted sequence is the x-axis of the first structural potential curve. The x-axis of the first structural potential curve is the microregion sort number, from 1 to D, and the y-axis is the structural potential value of the corresponding sorted microregion. This curve is monotonically increasing, like an energy tank, showing the structural foundation from the most stable to the least stable. The x-axis of the second dynamic triggering intensity curve uses the exact same microregion sort number as the first structural potential curve, and the y-axis corresponds to the dynamic triggering intensity value of the sorted microregion. This curve usually fluctuates randomly, revealing the distribution of dynamic energy in different structural potential level regions.

[0091] Step S22 aims to establish the intrinsic relationship between the static foundation of the material's structural incubation potential and the dynamic triggering intensity energy injection. By arranging all micro-regions in ascending order of their structural incubation potential values, an ordered benchmark from stable to unstable is constructed. The dynamic triggering intensity is then superimposed on this benchmark, transforming two independent physical quantities into an evolutionary hyperbolic model that reveals the resonance effect. This makes it clear that the critical state where high energy injection acts precisely on a high-risk structural foundation becomes readily apparent. The irreplaceable nature of this method lies in the fact that simply and independently screening high potential and high intensity values ​​cannot capture this crucial coupling relationship. Only by first sorting and then superimposing can the local maximum value of the dynamic triggering intensity be accurately identified within the most unstable high-potential danger zone. This logic of discovering resonance points is far more profound and effective than isolatedly searching for the global maximum value of two indicators, and it is the core link in accurately determining the source of high-risk aggregates.

[0092] Step S23: Locate the high potential range in the first structural incubation potential curve, find the local maximum value of the second dynamic trigger intensity curve based on the high potential range to determine the existence of resonance amplification effect, and obtain the micro-region corresponding to the local maximum value to obtain the high-risk aggregate source.

[0093] Identify the high-potential region at the end of the first structural incubation potential curve, for example, the micro-region where the structural incubation potential ranks in the bottom 10%. This region represents the most unstable and high-risk area of ​​the structure. Within this high-potential region on the horizontal axis, find the local maximum value of the second dynamic triggering intensity curve, and identify the micro-region corresponding to this local maximum value. It was ultimately identified as a high-risk aggregate source because it precisely captured the critical state where a violent and sustained energy injection acted precisely on an unstable material structure. This resonance condition is highly likely to induce irreversible large-scale aggregation. A dynamic trigger intensity value greater than twice the average of all micro-region dynamic trigger intensity values ​​was considered a local large value. Choosing 2 as the multiplier is a simple and robust heuristic widely used in signal processing and statistics, similar to the 2σ principle in statistics. It assumes the data is approximately normally distributed, with about 95% of the data falling within the mean ± 2 standard deviations. The 2x average also acts as a powerful filter, aiming to identify extreme events where the energy injection intensity far exceeds the system's normal level. A value that is only 1.1 or 1.2 times the average is likely just random fluctuation; however, a value reaching 2x or more has a very high probability of marking a qualitative change in the physical mechanism—a truly violent and noteworthy dynamic event.

[0094] A method for online detection and control of uniformity further includes:

[0095] Step S3: Based on the predicted impact points of high-risk aggregate sources, the monitoring area is delineated, and the dual-modal signals collected within the monitoring area are analyzed. By judging and accumulating the intensity of adhesion events, a node adhesion heat map is generated.

[0096] Specifically, step S3 also includes:

[0097] Step S31: Obtain the geometric center of the high-risk agglomerate source as the initial position and obtain its total moment of momentum to construct a momentum tensor. Perform eigenvalue decomposition on the momentum tensor to determine the eigenvector corresponding to the largest eigenvalue as the main motion direction. Emit a ray from the initial position along the main motion direction and calculate its intersection with the inner wall of the equipment as the predicted inner wall impact point. Finally, gather the predicted inner wall impact points calculated from all high-risk agglomerate sources to form a set of predicted inner wall impact points.

[0098] The geometric center of the high-risk aggregate source was obtained as the initial position of the droplet cluster. The total angular momentum of all droplets in each microregion is retrieved to obtain the momentum tensor T. Since the momentum tensor T is a matrix, its direction cannot be directly indicated; therefore, its core motion information must be extracted through the crucial step of eigenvalue decomposition. By performing eigenvalue decomposition on T, three eigenvalues ​​are obtained. and its three corresponding mutually orthogonal eigenvectors Eigenvalue decomposition is an existing technique and will not be elaborated upon here. The purpose of eigenvalue decomposition is to transform and decompose the complex momentum tensor T, which cannot directly indicate direction, onto a set of mutually perpendicular principal axes, i.e., eigenvectors, thereby revealing its intrinsic core motion information. The resulting three eigenvalues ​​are... This precisely quantifies the strength or magnitude of the total momentum in each principal axis direction. Therefore, the larger the eigenvalue, the stronger the momentum component in that direction, and the more significant the motion trend. Among them, the largest eigenvalue... This represents the most significant component of momentum in its spatial distribution, and its corresponding eigenvector. This precisely indicates the most dominant and strongest initial ejection direction of the droplet cluster, yielding the initial motion direction vector. After determining this direction, the initial position... Starting from the main direction of motion, along the direction of motion A virtual ray is emitted, and the first intersection point between it and the inner wall of the equipment is calculated. This intersection point is the predicted impact point on the inner wall. The inner wall impact points were calculated from all high-risk micro-regions. By combining these sets, we obtain the predicted set of impact points on the inner wall.

[0099] Step S32: Based on the predicted set of inner wall impact points, the average impact position centroid of each cluster is obtained through data clustering and centroid calculation;

[0100] The maximum deviation is calculated based on the centroid of the average impact position of each cluster, yielding the farthest distance of each cluster from its average impact position centroid. This farthest distance is then used as the base radius, with an additional safety factor added to obtain the uncertainty radius. Based on the radius of uncertainty Define each cluster as A circular area with radius is merged to obtain a high-risk inner wall monitoring area.

[0101] The impact point set is clustered using the K-means clustering method. The points are divided into different clusters. For each point cluster, its geometric centroid is calculated and represented as... center of mass This represents the most likely average impact location within the cluster. After determining a point cluster and its centroid... Next, calculate the midpoint of the set of all predicted inner wall impact points in the cluster. To this center of mass Find the furthest distance among these distances. This furthest distance reflects the furthest deviation of the prediction from its mean position within that cluster. Use this furthest distance as the base radius, and add an additional safety factor. Therefore, radius The calculation formula is: ,in Point With the center of mass The Euclidean distance between them Belongs to the set of predicted inner wall impact points q represents the index. This means taking the maximum value of all these distances within the current cluster. This is a proactively set safety margin, the value of which depends on the tolerance for risk and the severity of the consequences of failing to detect a collision. For example, if the consequences of missing a collision are very serious, a larger margin should be selected. The value can be adjusted accordingly; conversely, it can be reduced appropriately. This margin can also be set based on the effective monitoring range of the installed sensors to ensure that the coverage area is larger than the sensor's blind zone. A value is defined for each point cluster with its centroid. With the center, and A circular high-risk area with a radius of 1 is defined. All such circular areas are merged. Any two or more overlapping or contacting circular areas are merged into a larger, irregularly shaped continuous area. Based on these finally formed, merged irregular areas, the high-risk inner wall monitoring area is obtained, transforming discrete and uncertain prediction points into continuous monitoring ranges with clear boundaries.

[0102] Step S33: Based on the high-risk inner wall monitoring area, a uniform grid strategy is used to discretize the continuous monitoring surface to obtain a dual-modal sensor array; based on the dual-modal sensor array, instantaneous temperature change signals and stress wave signals are collected.

[0103] In the homogenized grid strategy, the grid size is defined by the system's top-level monitoring accuracy requirement Δs. This accuracy Δs represents the minimum positioning resolution the system is expected to achieve or the smallest anomaly scale it can distinguish. Therefore, the grid side length L=Δs is set to ensure that the final deployment scheme meets the preset performance indicators. The specific generation steps are as follows: This homogenized grid is constructed on a two-dimensional plane containing the entire high-risk inner wall monitoring area. All grid nodes falling within the outline of the high-risk inner wall monitoring area are selected to form a preliminary candidate point set. Nodes that cannot be practically installed due to physical limitations, such as the presence of structural obstacles, reinforcing ribs, or unsuitable surface materials, are excluded. After these two layers of selection, all remaining valid nodes constitute a dual-modal sensor arrangement set, providing an accurate and feasible search space for subsequent sensor layout optimization algorithms. The dual-modal sensor arrangement set includes an infrared thermopile for non-contact measurement of the instantaneous temperature change signal of the inner wall. Piezoelectric acoustic emission probes are used to acquire stress wave signals generated by impact and adhesion. .

[0104] Step S34: Use baseline difference and local minimum detection method on the instantaneous temperature change signal to obtain the negative pulse amplitude;

[0105] High-pass filtering combined with signal energy integration is used to obtain the high-frequency energy burst intensity of the stress wave signal.

[0106] For signals from instantaneous temperature changes The amplitude of the negative temperature pulse is calculated using baseline difference and local minimum detection methods. A dynamic, real-time baseline is established for the instantaneous temperature change signal by performing a moving average on the signal; for example, the average of hundreds of sampling points prior to the current time point is calculated, and this smoothed value is used as the temperature baseline for the current moment. Within a very short sliding window, such as dozens of sampling points, the local minimum of the signal is continuously searched. Once a Calculate its relationship with the baseline at the corresponding time. The difference, negative pulse amplitude Defined as To ensure that what is detected is a real pulse and not random noise, only when the calculated negative pulse amplitude is... A temperature pulse event is only considered valid and its amplitude is determined when it exceeds a preset noise threshold.

[0107] For stress wave signals The intensity of high-frequency energy bursts is calculated using a high-pass filtering method combined with signal energy integration to obtain the original stress wave signal. The signal passes through a digital high-pass filter, such as a Butterworth or finite impulse response filter, with its cutoff frequency set at a high value (e.g., above 20kHz). The purpose is to filter out low-frequency mechanical vibrations and background noise generated during normal equipment operation, retaining only the high-frequency components produced by the impact event, thus obtaining the filtered signal. For this filtered signal Perform energy calculations. When the absolute value of the instantaneous amplitude of the signal... When a set trigger threshold is exceeded, integration begins within a fixed time window (e.g., a few milliseconds), determining the intensity of the high-frequency energy burst. It is calculated as the sum of the squares of the amplitudes of all sampling points within the window, i.e. This summation result directly quantifies the acoustic emission energy released at the moment of impact, and its magnitude is the high-frequency energy burst intensity.

[0108] Step S35: Based on the negative pulse amplitude and high-frequency energy burst intensity, the temperature change condition and energy burst condition are judged respectively to obtain the adhesion event determination; for each monitoring node, when a valid adhesion event is detected on each monitoring node, the intensity of the valid adhesion event will be added to the existing cumulative intensity value of the corresponding node to divide the adhesion heat level and obtain the node adhesion heat distribution map.

[0109] To determine an adhesion event, the system will assess the amplitude of the negative pulse based on spatiotemporal correlation. and high-frequency energy burst intensity A fusion criterion analysis is performed. Specifically, fusion is only considered when the sensors are on the same node and within a very short time window. Inside, for example A valid adhesion event can only be confirmed when two conditions are met simultaneously: one is a sudden temperature change, i.e., the calculated negative pulse amplitude. Exceeding a preset temperature threshold Secondly, the energy explosion conditions, namely the calculated high-frequency energy explosion intensity. Exceeding a preset energy threshold The setting of these two key thresholds is based on statistical analysis of background signals during normal device operation (when no adhesion events occur). This was achieved by collecting a large amount of baseline data and calculating their corresponding values. and By analyzing the values, we can obtain the statistical distribution of these noise metrics, such as their mean. and standard deviation The threshold is then set to a value higher than the vast majority of noise levels, for example, the threshold value. ,in It is a coefficient selected based on the required signal-to-noise ratio and the tolerance for false alarm rate, such as or For each monitoring node, whenever a new valid adhesion event that satisfies the spatiotemporal correlation conditions is detected at the node, the intensity of the event is added to the existing accumulated intensity value of that node. For example, the intensity of an energy burst. The value, where the monitoring node is the installation location of the dual-modal sensor determined after discretizing the monitoring area using a uniform grid strategy in step S33. This cumulative value can be considered as the total amount of adhesion damage suffered at that specific location, and it increases over time with the occurrence of new events. A three-color (green, orange, red) visualization scheme is used to classify adhesion heat levels through two key intensity thresholds, including a low-risk threshold. and high risk threshold The color displayed for each node is determined by comparing its real-time cumulative intensity value to two thresholds: a node is displayed green when its cumulative intensity value is zero, and green when its cumulative intensity value exceeds a low-risk threshold. However, it has not yet reached the high-risk threshold. At that time, that is The node is displayed in orange; when the node's cumulative intensity value exceeds the high-risk threshold... At that time, that is The node is displayed in red, representing a high-risk or critical state, indicating that the adhesion problem at that location is already quite serious. These two thresholds are used for visualization. and The settings are typically based on engineering experience, the fatigue limits of equipment and materials, or risk management strategies. For example, they can be based on historical data analysis. Set to the cumulative intensity value when a minor fault occurs for the first time in history, and then... This is set as the critical cumulative intensity value before a serious failure occurs. Through this dynamic accumulation and threshold judgment mechanism, the system obtains a node adhesion heat map from discrete, instantaneous adhesion event data. The color of each node in the map intuitively reflects the evolution of its adhesion risk, from safe green to warning orange, and then to dangerous red, providing a clear and direct decision-making basis for equipment condition monitoring and predictive maintenance.

[0110] Step S35 accurately captures the unique dual physical characteristics of adhesion events: the sudden drop in local temperature (negative pulse) caused by instantaneous heat transfer when droplets impact the wall surface, and the stress wave generated by the impact (high-frequency energy burst). Its function is to fuse these two heterogeneous signals for spatiotemporal correlation judgment, transforming instantaneous, discrete physical events into a quantitative accumulation of the risk level of each monitoring node. It not only determines the presence or absence of an event, but more importantly, it accumulates the intensity of each valid event, thereby generating a dynamically evolving node adhesion heat distribution map, intuitively revealing the entire process of risk from non-existence to presence, and from mild to severe. The irreplaceable nature of this method lies in the fact that the dual-modal fusion judgment mechanism greatly eliminates noise and false alarms that may be caused by a single signal source, ensuring high reliability of event judgment; while the intensity accumulation mechanism goes beyond simple event counting, able to distinguish the different levels of harm caused by a large number of minor adhesions and a few severe adhesions, quantifying and visualizing the invisible, time-accumulated damage history, providing a unique and indispensable data foundation for subsequent accurate prediction and graded response.

[0111] A method for online detection and control of uniformity further includes:

[0112] Step S4: Execute response matrix control based on the adhesion heat distribution map to obtain the control strategy.

[0113] Specifically, step S4 also includes:

[0114] Step S41: Perform a targeted real-time physical state verification based on the node adhesion heat distribution map to obtain the real-time state vector;

[0115] Step S42: Perform two-dimensional response matrix control based on the instantaneous state vector to obtain the control strategy.

[0116] All high-risk hotspots marked in orange and red on the node adhesion heat map are subjected to a targeted, real-time physical state verification. Specifically, the system guides ultrasonic sensors to precisely scan these identified hotspot locations, with the detection metric being acoustic impedance, denoted as [insert value here]. A lower A value indicates that the adhesive is still in a soft state, while a higher value indicates that the adhesive is still in a soft state. A value indicating solidification of the material signifies high hardness and a qualitative change in risk. This step generates a two-dimensional instantaneous state vector for each hotspot: one being the adhesion heat level inherited from the node adhesion heat distribution map. The first is the accumulation of quantity, and the second is the acoustic impedance obtained through ultrasonic detection. This represents the hardness index, a qualitative change, thus completing a comprehensive assessment of the current risk. Based on the acquired two-dimensional state vector... This will directly lead to the closed-loop decision-making and control execution phase to obtain the control strategy. The logic of this control strategy is embedded in a two-dimensional response matrix, ensuring that each risk combination corresponds to a specific instruction. The two-dimensional response matrix is ​​as follows:

[0117]

[0118] To effectively distinguish between low and high acoustic impedance, a specific acoustic impedance threshold is preset. In real-time detection, when the sensor measures the acoustic impedance value... satisfy When the adhesive is in a soft state, the system determines that the acoustic impedance is low; conversely, if the adhesive is in a soft state, the system determines that the acoustic impedance is low. If so, it is determined to be in a solidified state, with high acoustic impedance. Acoustic impedance threshold. This was achieved through an offline calibration process. During the experimental phase, samples representing two extreme physical states of the adhesive were collected or prepared: a typical soft sample and a fully cured sample. Using the same ultrasonic sensor system deployed in the field, acoustic impedance measurements were performed on both sets of samples, resulting in two sets of data distributions: one set corresponding to a cluster of lower acoustic impedance values ​​for the soft sample, and the other set corresponding to a cluster of higher acoustic impedance values ​​for the cured sample. A threshold was set through statistical analysis of these two sets of data, and the average values ​​of each set were calculated. Then, a midpoint, or intermediate value between these two averages, was taken as the threshold. This ensures that the two states can be distinguished as much as possible. The output of this process is specific and executable, either as automated control commands sent directly to the actuators or as intelligent alarms or maintenance work orders pushed to operators, containing precise location and risk assessment, resulting in a complete, efficient, and clear control loop from perception and assessment to response.

[0119] Example 2:

[0120] like Figure 2 As shown, an online uniformity detection and control system includes:

[0121] The feature parameter analysis module acquires synchronous multi-source data, divides micro-regions through droplet density analysis, and extracts data from each micro-region to calculate its physical feature parameter set;

[0122] The risk clustering determination module constructs an evolution hyperbola based on the physical characteristic parameter set of each micro-region, and determines the source of high-risk clusters by finding local large values ​​of dynamic triggering intensity in the high potential range.

[0123] The adhesion heat module predicts impact points based on high-risk aggregate sources to delineate monitoring areas, analyzes dual-modal signals collected within the monitoring areas, and generates a node adhesion heat distribution map by judging and accumulating the intensity of adhesion events.

[0124] The control module executes response matrix control based on the node adhesion heat distribution map to obtain the control strategy.

[0125] It should be understood that although the steps in the flowcharts of the various embodiments of the present invention are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the various embodiments may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least a portion of the sub-steps or stages of other steps.

[0126] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0127] The foregoing description is illustrative of the invention and should not be construed as limiting it. Although several exemplary embodiments of the invention have been described, those skilled in the art will readily understand that many modifications can be made to the exemplary embodiments without departing from the novel teachings and advantages of the invention. Therefore, all such modifications are intended to be included within the scope of the invention as defined in the claims. It should be understood that the foregoing description is illustrative of the invention and should not be construed as limiting it to the specific embodiments disclosed, and modifications to the disclosed embodiments and other embodiments are intended to be included within the scope of the appended claims. The invention is defined by the claims and their equivalents.

Claims

1. A method for online detection and control of uniformity, characterized in that, Includes the following steps: Acquire synchronous multi-source data, divide micro-regions through droplet density analysis, and extract data from each micro-region to calculate its set of physical characteristic parameters; An evolutionary hyperbola is constructed based on the set of physical characteristic parameters of each micro-region, and the source of high-risk aggregates is determined by finding local large values ​​of dynamic triggering intensity in the high potential range. The impact point is predicted based on high-risk aggregate sources to delineate the monitoring area, and the dual-modal signals collected within the monitoring area are analyzed. The node adhesion heat distribution map is generated by judging and accumulating the adhesion event intensity. The control strategy is obtained by performing response matrix control based on the node adhesion heat distribution map.

2. The method for online detection and control of uniformity according to claim 1, characterized in that, The analysis steps for the physical feature parameter set are as follows: Based on a subset of data with three-dimensional spatial information, texture structure features and information entropy analysis were performed to obtain the liquefaction entropy of each micro-region. Droplet identification and binarization are performed on a subset of data based on three-dimensional spatial information to obtain a binarized three-dimensional image; volume counting is performed on the binarized three-dimensional image to obtain the initial droplet volume estimate for each micro-region; The momentum contribution tensor is accumulated based on a subset of data containing three-dimensional velocity field data and three-dimensional spatial information to obtain the total momentum of all droplets in each micro-region; Based on the data subset of triboelectric charge distribution image sequence and three-dimensional spatial information, time series data of internal charge peak decay over time are extracted. The time series data are then fitted to the exponential decay model using the nonlinear least squares method to solve for the relaxation time constant of each micro-region. The liquefaction entropy, initial droplet volume estimate, total angular momentum of all droplets, and relaxation time constant of each microregion are collectively referred to as the set of physical characteristic parameters of each microregion.

3. The method for online detection and control of uniformity according to claim 2, characterized in that, The data subset analysis steps for the three-dimensional spatial information of each micro-region are as follows: The continuous probability density field is transformed into a three-dimensional density surface. For the three-dimensional density surface, its Riemann tensor is calculated, and the arc length square of the differential displacement on the surface is calculated based on the Riemann tensor. The geodesic distance on the three-dimensional density surface is defined based on the square of the arc length of the differential displacement on the surface. After initializing a set of generators on the three-dimensional density surface based on the geodesic distance, the Lloyd algorithm is used for iterative optimization to determine the final uniform irregular micro-region of droplet density. Based on the final uniform irregular micro-regions of droplet density, the data and regions of the three-dimensional grayscale image data are matched and extracted to obtain a subset of the three-dimensional spatial information of each micro-region.

4. The method for online detection and control of uniformity according to claim 3, characterized in that, The steps for the continuous probability density field analysis are as follows: Based on the three-dimensional grayscale image data obtained by laser-induced fluorescence scanning, the initial three-dimensional point cloud data of droplet dispersion is obtained by three-dimensional grayscale image data segmentation and centroid calculation; and it is then registered and synchronized in space and time with the three-dimensional velocity field data obtained by the particle image velocimetry system and the triboelectric charge distribution image sequence obtained by the imaging sensor array. The initial droplet dispersion three-dimensional point cloud data is projected onto the target plane to obtain a two-dimensional discrete point set; The kernel density estimation method is used to transform a two-dimensional discrete point set into a continuous probability density field.

5. The method for online detection and control of uniformity according to claim 4, characterized in that, The steps for analyzing the high-risk aggregate sources are as follows: The structural incubation potential of each microregion is calculated based on the liquefaction entropy and initial droplet volume estimate of each microregion. The dynamic triggering intensity value of each microregion is calculated based on the total angular momentum and relaxation time constant of each microregion. Based on the structural incubation potential value and dynamic triggering intensity value of each microregion, an evolution hyperbola for each microregion is constructed, including the first structural incubation potential curve and the second dynamic triggering intensity curve. The high potential range is locked in the first structure incubation potential curve. Based on the high potential range, the local large value of the second dynamic trigger intensity curve is found to determine the existence of resonance amplification effect. The micro-region corresponding to the local large value is obtained to obtain the high-risk aggregate source.

6. The method for online detection and control of uniformity according to claim 1, characterized in that, The steps for analyzing the node adhesion heat distribution map are as follows: The baseline difference and local minimum detection method is used to obtain the negative pulse amplitude for instantaneous temperature change signals; High-pass filtering combined with signal energy integration is used to obtain the high-frequency energy bursting intensity of the stress wave signal; Based on the negative pulse amplitude and high-frequency energy burst intensity, the temperature change condition and energy burst condition are judged respectively to obtain the adhesion event determination; for each monitoring node, when a valid adhesion event is detected on each monitoring node, the intensity of the valid adhesion event will be added to the existing cumulative intensity value of the corresponding node to classify the adhesion heat level and obtain the node adhesion heat distribution map.

7. The method for online detection and control of uniformity according to claim 6, characterized in that, The analysis steps for the instantaneous temperature change signal and stress wave signal are as follows: The maximum deviation is calculated based on the centroid of the average impact position of each cluster, yielding the farthest distance of each cluster from its average impact position centroid. This farthest distance is then used as the base radius, with an additional safety factor added to obtain the uncertainty radius. Based on the radius of uncertainty Define each cluster as A circular area with radius [radius] is merged to obtain the inner wall monitoring area; Based on the inner wall monitoring area, a uniform grid strategy is used to discretize the continuous monitoring surface to obtain a dual-modal sensor arrangement set; The system uses a dual-modal sensor array to collect instantaneous temperature change signals and stress wave signals.

8. The method for online detection and control of uniformity according to claim 7, characterized in that, The centroid analysis steps for the average impact location of each cluster are as follows: The geometric center of the high-risk agglomerate source is obtained as the initial position and its total moment of momentum is obtained to construct a momentum tensor. The momentum tensor is decomposed into eigenvalues ​​to determine the eigenvector corresponding to the largest eigenvalue as the main motion direction. A ray is emitted from the initial position along the main motion direction and its intersection with the inner wall of the equipment is calculated as the predicted inner wall impact point. Finally, the predicted inner wall impact points calculated from all high-risk agglomerate sources are collected to form a set of predicted inner wall impact points. Based on the predicted set of impact points on the inner wall, the average impact position centroid of each cluster is obtained through data clustering and centroid calculation.

9. The method for online detection and control of uniformity according to claim 1, characterized in that, The control strategy analysis steps are as follows: Perform a targeted, real-time physical state verification based on the node adhesion heat map to obtain the real-time state vector; Two-dimensional response matrix control is executed based on instantaneous state vectors to obtain the control strategy.

10. An online uniformity detection and control system, used to implement the online uniformity detection and control method according to any one of claims 1-9, characterized in that, The system includes: The feature parameter analysis module acquires synchronous multi-source data, divides micro-regions through droplet density analysis, and extracts data from each micro-region to calculate its physical feature parameter set; The risk clustering determination module constructs an evolution hyperbola based on the physical characteristic parameter set of each micro-region, and determines the source of high-risk clusters by finding local large values ​​of dynamic triggering intensity in the high potential range. The adhesion heat module predicts impact points based on high-risk aggregate sources to delineate monitoring areas, analyzes dual-modal signals collected within the monitoring areas, and generates a node adhesion heat distribution map by judging and accumulating the intensity of adhesion events. The control module executes response matrix control based on the node adhesion heat distribution map to obtain the control strategy.

Images