An image analysis-based screening machine operation monitoring and control system

Abstract

This invention relates to the fields of computer vision and industrial intelligent monitoring technology, and discloses a screening machine operation monitoring and control system based on image analysis. The system includes: an image acquisition module, which acquires time-series point cloud data of the feed flow, oversize material, and undersize material through a three-dimensional vision measurement device; a point cloud processing and particle size analysis module, which segments the time-series point cloud data, identifies material particle size information, and generates a particle size distribution vector; a process state quantification module, which quantifies the particle size distribution vector into a set of screening process state indicators, including a real-time screening probability matrix, an effective working area attenuation coefficient, and a material retention index; and a diagnostic decision module, which receives the screening process state indicator set, matches it with a preset diagnostic rule base, and outputs a fault diagnosis conclusion. This invention achieves full-process, non-contact, refined monitoring and closed-loop intelligent control of the screening machine's operating status, improving screening efficiency and the level of intelligent equipment maintenance.

Description

Technical Field

[0001] This invention relates to the fields of computer vision and industrial intelligent monitoring technology, and in particular to a screening machine operation monitoring and control system based on image analysis. Background Technology

[0002] In industrial production, traditional screening machine monitoring methods mainly rely on manual inspection, periodic shutdown inspection, or simple sensor monitoring, which cannot achieve real-time, non-contact assessment of the screen surface condition. Although some particle size analysis technologies based on two-dimensional vision or laser scanning have been developed, they usually only monitor the feed or a single outlet, and cannot comprehensively obtain multi-dimensional information on screening efficiency and screen surface condition.

[0003] In screening machine operation scenarios, materials are often in a state of high-speed movement and mutual occlusion and adhesion. When performing image segmentation, they are easily affected by factors such as motion blur, changes in lighting and particle overlap. Existing particle size analysis systems based on two-dimensional images often ignore the three-dimensional shape and motion trajectory of materials, resulting in low particle size recognition accuracy and high missegmentation rate, making it difficult to effectively adapt advanced three-dimensional vision technology to the screening process.

[0004] Wear, clogging, and damage to the screen surface are the main factors affecting screening efficiency and equipment life. Existing technologies mostly rely on indirect indicators such as temperature, vibration, or current fluctuations to judge the health status of the screen surface. They cannot distinguish the impact of screen surface degradation on screening efficiency from normal feeding fluctuations, nor can they pinpoint the specific area and type of fault. There is a lack of quantitative indicators and real-time evaluation systems for the health status of the screen surface that are available for engineering use.

[0005] Current screening machine monitoring systems are often single-function, data is isolated, and diagnostic rules are rigid, lacking adaptive optimization and intelligent decision-making capabilities. They are difficult to adapt to changes in different materials and working conditions. There is an urgent need for a full-process monitoring and control system that integrates three-dimensional vision, intelligent analysis, status quantification, and decision suggestions to improve the integration and automation level of screening machine operation monitoring and control. Summary of the Invention

[0006] This invention provides a screen machine operation monitoring and control system based on image analysis to solve existing technical problems.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention provides a screen machine operation monitoring and control system based on image analysis, comprising: The image acquisition module is used to acquire time-series point cloud data of the feed flow, the oversize material discharge end, and the undersize material discharge end of the screen machine through three-dimensional vision measurement equipment deployed at the feed end, the oversize material discharge end, and the undersize material discharge end of the screen machine. The point cloud processing and particle size analysis module is used to segment the time-series point cloud data, identify the particle size information of the material, and generate particle size distribution vectors of the feed flow, oversize material and undersize material based on the particle size information. The process state quantification module is used to quantify the particle size distribution vector into a set of screening process state indicators through a predefined calculation model, including: a real-time screening probability matrix characterizing the separation efficiency of different particle sizes, an effective working area attenuation coefficient characterizing the degradation of the screen surface function, and a material retention index characterizing the screen surface throughput performance. The diagnostic decision module is used to receive the set of status indicators of the screening process, match them with the preset diagnostic rule base, and output fault diagnosis conclusions for abnormal screening efficiency, screen blockage or damage.

[0008] The beneficial effects of the technical solution provided by this invention include at least the following: This invention constructs a three-dimensional visual monitoring system covering the entire screening process by simultaneously deploying three-dimensional vision measurement equipment at three points: the feed end of the screening machine, the discharge end of the oversize material, and the discharge end of the undersize material. This enables multi-angle synchronous acquisition of material flow, providing a complete data foundation for screening efficiency evaluation and fault diagnosis. It breaks through the limitations of single-point monitoring in traditional systems and significantly improves monitoring reliability.

[0009] This invention proposes a set of image analysis algorithms optimized for dynamic screening scenarios, including a material instance segmentation method based on spatiotemporal continuity, a weighted composite equivalent particle size calculation model, and a dynamic particle size distribution statistical method. These algorithms effectively overcome the interference of material particle adhesion under high-speed motion conditions, achieve high-precision particle segmentation and particle size extraction, provide reliable input for subsequent state quantification and diagnosis, and effectively improve the system's adaptability in complex industrial environments.

[0010] This invention innovatively proposes a method for quantifying the health status of screen surfaces based on a state-space model. By separating the inherent inertia of the screening process, feeding fluctuations, and health status disturbances, it accurately calculates multi-dimensional status indicators such as the effective working area attenuation coefficient and material retention index. This enables real-time identification of screen surface fault types, quantification of fault levels, and location of potential fault areas, effectively improving monitoring reliability and extending the service life of the screening equipment.

[0011] This invention integrates functions such as 3D vision acquisition, point cloud processing, state quantification, fault diagnosis and decision-making suggestions, and constructs a full-process control system from data acquisition to intelligent decision-making, forming an intelligent operation and maintenance closed loop, thereby improving the intelligence level and production efficiency of screening equipment. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a system structure diagram provided in an embodiment of the present invention. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0015] This embodiment provides a screening machine operation monitoring and control system based on image analysis. Please refer to... Figure 1 This is a system structure diagram provided in an embodiment of the present invention.

[0016] The image analysis-based screening machine operation monitoring and control system in this embodiment includes: The image acquisition module is used to acquire time-series point cloud data of the feed flow, the oversize material discharge end, and the undersize material discharge end of the screen machine through three-dimensional vision measurement equipment deployed at the feed end, the oversize material discharge end, and the undersize material discharge end of the screen machine. The point cloud processing and particle size analysis module is used to segment time-series point cloud data, identify the particle size information of materials, and generate particle size distribution vectors for feed flow, oversize material and undersize material based on the particle size information. The process state quantification module is used to quantify the particle size distribution vector into a set of screening process state indicators through a predefined calculation model, including: a real-time screening probability matrix characterizing the separation efficiency of different particle sizes, an effective working area attenuation coefficient characterizing the degradation of the screen surface function, and a material retention index characterizing the screen surface throughput performance. The diagnostic decision module receives a set of status indicators for the screening process, matches them with a preset diagnostic rule base, and outputs fault diagnosis conclusions for abnormal screening efficiency, screen blockage, or damage.

[0017] As one embodiment of the present invention, the image acquisition module includes a sensor array unit, a timing synchronization triggering unit, and a point cloud preprocessing unit; The sensor array unit consists of three-dimensional vision measurement devices deployed at the feed end, discharge end of the oversize material, and discharge end of the undersize material of the screening machine, used to acquire the original three-dimensional images of the material. The timing synchronization trigger unit is connected to the sensor array unit and is used to generate unified acquisition trigger commands according to a preset timing sequence to control each 3D vision measurement device to perform synchronous data acquisition. The point cloud preprocessing unit is used to perform noise reduction, filtering, and coordinate correction on the acquired raw 3D images to generate time-series point cloud data.

[0018] It should be noted that the sensor array unit is responsible for directly acquiring the original three-dimensional image of the material flow. In specific implementation, at least one three-dimensional vision measurement device is deployed at each of the three key workstations: the feed end of the screening machine, the discharge end of the oversize material, and the discharge end of the undersize material. Such devices can be structured light scanners, binocular stereo vision systems, or lidar, to form a vision sensor array covering the core material flow. These devices simultaneously capture dynamic three-dimensional images of the material flow from different perspectives, obtaining complete visual information of the feed flow, oversize material, and undersize material, providing a comprehensive data foundation for subsequent analysis.

[0019] The timing synchronization trigger unit ensures the time consistency and comparability of data collected from different measurement points. This unit is usually implemented by a central controller or a dedicated synchronization signal generator. It generates a unified pulse signal as a trigger command according to a preset timing sequence that matches the operating rhythm of the screening machine. This signal is sent to all three-dimensional vision measurement devices in the vision sensor array, forcing each device to start acquisition at the same time. This avoids data time misalignment caused by differences in the internal clock of the devices or software trigger delays, and ensures that the subsequently generated time-series point cloud data is aligned in the time dimension.

[0020] The point cloud preprocessing unit is responsible for receiving the original three-dimensional images from each sensor and performing a series of preprocessing steps, including: (1) Denoising: using algorithms such as statistical filtering or radius filtering to remove discrete noise points caused by ambient light interference, sensor noise itself, or floating dust; (2) Filtering: applying smoothing filtering methods such as Gaussian filtering or moving least squares to smooth the point cloud surface, reducing data fluctuations while retaining the main geometric features; (3) Coordinate correction: converting the data collected by all sensors to the same world coordinate system. Usually, it is necessary to perform joint calibration of the sensor array in advance to obtain the transformation matrix from each sensor coordinate system to the system reference coordinate system. Through coordinate correction, the point clouds collected at different locations are initially aligned in the spatial dimension.

[0021] As one embodiment of the present invention, the point cloud processing and particle size analysis module includes a point cloud registration and extraction unit, a material instance segmentation unit, a particle size feature extraction unit, and a dynamic distribution statistics unit. The point cloud registration and extraction unit is used to spatially align time-series point cloud data obtained from different measurement points, remove static background point clouds that characterize the structure of fixed equipment, and separate real-time material flow point clouds of feed flow, oversize material flow and undersize material flow. The material instance segmentation unit is used to segment the real-time material flow point cloud into instances, extract each independent material particle point cloud from the material flow point cloud and identify it. The particle size feature extraction unit is used to calculate the equivalent particle size of the point cloud for each individual material particle; The dynamic distribution statistics unit is used to perform statistical analysis on the equivalent particle size of all independent material particles in the feed stream, oversize stream, and undersize stream according to a preset time window and preset particle size interval, calculate the percentage of material quantity in each particle size interval, and generate particle size distribution vectors of the feed stream, oversize stream, and undersize stream as a function of time.

[0022] It should be noted that the point cloud registration and extraction unit is responsible for accurately separating the real-time material flow point cloud from the point cloud fused from multiple sensors, specifically including: Spatial alignment: The image acquisition module has performed preliminary coordinate correction. This unit further adopts a more refined point cloud registration algorithm, such as the Iterative Closest Point (ICP) algorithm or its variant, to fine-tune the alignment of point cloud frames from different measurement points but with the same timestamp, so as to eliminate registration errors caused by equipment vibration or slight deformation. Background removal and material flow separation: Typically, the system needs to pre-scan the unloaded screen to obtain a static background point cloud template of its fixed equipment structure. During real-time processing, by comparing the current frame point cloud with the background template, the differential method is used to quickly identify and remove point cloud clusters belonging to the static background. The remaining point cloud is then determined to be the real-time material flow point cloud. Finally, combined with the deployment location information of each sensor, the real-time material flow point cloud is further divided into feed flow, oversize material, and undersize material.

[0023] The dynamic distribution statistics unit is responsible for aggregating and statistically analyzing particle size distribution according to a preset time period, generating a standardized particle size distribution vector to achieve continuous quantitative monitoring of the process, including: (1) Presetting a time window and multiple particle size intervals according to process requirements or system initialization requirements. The time window determines the frequency of status updates and the smoothness of data. The particle size interval defines the precision and scope of particle size analysis, which is usually related to the screen hole size of the screen machine; (2) Quantity percentage statistics: Within each preset time window, the system traverses all independent material particles that have been identified and whose particle size features have been extracted. Based on the equivalent particle size of the particle, it is assigned to a preset particle size interval. The total number of material particles belonging to each particle size interval within the time window is counted. Finally, the percentage of the total number of material particles in each particle size interval to the sum of the total number of particles in all particle sizes within the time window is calculated.

[0024] The following is a specific example of a method to statistically analyze the equivalent particle size of all independent material particles in the feed stream, oversize stream, and undersize stream according to a preset time window and preset particle size interval, calculate the percentage of material quantity in each particle size interval, and generate particle size distribution vectors of the feed stream, oversize stream, and undersize stream as a function of time: Time window: Statistics are collected every 5 seconds; Grain size interval division: Interval 1: Equivalent particle size > 0 mm and ≤ 2 mm; Interval 2: Equivalent particle size > 2 mm and ≤ 5 mm; Interval 3: Equivalent particle size > 5 mm and ≤ 8 mm; Interval 4: Equivalent particle size > 8 mm and ≤ 12 mm; Statistical objects: feed flow, oversize material, undersize material; Statistical method: Percentage based on particle count; Table 1: Example Data Table for Granularity Distribution Vector Generation

[0025] As one embodiment of the present invention, the material instance segmentation unit performs the following: Voxel downsampling is performed on the real-time material flow point cloud, and the normal vector and curvature of each point in the downsampled real-time material flow point cloud are calculated. Based on the spatiotemporal continuity constraint of material particle motion, the real-time material flow point cloud of multiple consecutive frames is correlated and tracked. By comparing the consistency of the motion trajectory of the real-time material flow point cloud between adjacent frames, point cloud clusters with different motion trajectories are separated, and a temporary tracking label is assigned to each point cloud cluster. For point cloud clusters that remain stuck together in a single frame after separation, a region growing oversegmentation algorithm based on geodesic distance is executed with normal vector and curvature as constraints to further divide the stuck point cloud clusters into multiple possible granular regions, i.e. oversegmentation regions. The tracking tags are associated with oversegmented regions, and the missegmentation is corrected using graph optimization methods to obtain independent material particle point clouds. A global identifier is then assigned to each independent material particle point cloud.

[0026] It should be noted that the spatiotemporal continuity constraint of material particle motion refers to the assumption that the motion of the same particle should satisfy the continuity assumption under physical laws in a series of consecutive frames of images. This includes the smoothness of changes in position and velocity and the predictability of trajectory. This constraint assumes that the displacement, velocity or motion state of the particle will not change drastically within a very short time interval. Based on this constraint, the system establishes an association model of points or point clusters between adjacent frames. Even if particles are spatially close or visually attached within a single frame, their motion trajectories in consecutive frames will maintain their own independent and consistent physical motion patterns, thus providing a key temporal discrimination basis for instance segmentation.

[0027] The system models the position sequence of point cloud clusters in consecutive frames and extracts their motion features, such as displacement vectors and velocity vectors. It evaluates the consistency of trajectories by calculating the cosine similarity or velocity change rate of the displacement vectors of the same candidate particle between adjacent frames. If the angle between the displacement vectors or the velocity change rate of multiple points or point clusters in consecutive frames is less than the corresponding threshold, these points or point clusters are considered to exhibit a highly consistent motion pattern and belong to the same particle; otherwise, they are considered to belong to different particles.

[0028] Temporary tracking labels are temporary unique identifiers assigned to each point cloud cluster initially associated in consecutive frames. Their functions include: (1) serving as a carrier of spatiotemporal association, maintaining temporary identity tracking of the same particle in multiple consecutive frames; (2) providing structured input for subsequent oversegmentation and optimization, transforming the single-frame segmentation problem into a labeled cross-frame optimization problem; and (3) assisting in subsequent missegmentation correction, improving the physical rationality and stability of the segmentation results through the temporal continuity constraint of the labels.

[0029] The region growing oversegmentation algorithm based on geodesic distance specifically includes: (1) Seed point selection: Calculate the curvature of all points in the cluster of sticky points, identify the local maxima of curvature as candidate feature points, and select the points whose geodesic distance is greater than the preset geodesic distance threshold from the candidate feature points as initial seed points; (2) Constrained region growth: Starting with each seed point, based on the geodesic distance between the seed point and other points, and using the angle between the normal vectors less than the threshold as similarity constraint and the smooth change of curvature as region continuity constraint, gradually merge neighboring similar points; (3) Oversegmentation region generation: The above merging process continues until all points are assigned to a certain region, or no neighboring points that satisfy the constraints can be found; finally, the sticky cluster is subdivided into multiple sub-regions based on geometric feature continuity, and these sub-regions are oversegmentation regions.

[0030] The graph optimization method specifically includes: constructing a graph with oversegmented regions as nodes, calculating the similarity weights between nodes based on the normal vectors and curvature similarity of all points in each node, establishing temporal edges between nodes across frames based on tracking labels, defining an energy function that covers the probability (data item) of different nodes belonging to the same object and the consistency (smoothing item) of tracking labels of adjacent nodes, and finally minimizing the energy function using optimization algorithms such as graph cut or belief propagation, thereby merging misclassified oversegmented regions and separating adhered oversegmented regions.

[0031] As one embodiment of the present invention, when the particle size feature extraction unit calculates the equivalent particle size, it performs the following: For each independent material particle point cloud, first calculate its minimum bounding cuboid, and obtain the length, width, height and orientation of the minimum bounding cuboid as the length and orientation of the longest axis, intermediate axis and shortest axis of the corresponding independent material particle point cloud. Project the point cloud of the particle onto a plane perpendicular to the longest axis, and calculate the diameter of the equivalent circle of the projected area. The equivalent particle size is calculated using the following formula: ; In the formula, For equivalent particle size, and These represent the longest and shortest spatial axes of the corresponding individual material particles, respectively. The equivalent circle diameter, , and They are respectively , and The weighting coefficient is set according to the type of material.

[0032] As one embodiment of the present invention, the process status quantification module includes a screening efficiency calculation unit, an effective working area evaluation unit, a material retention analysis unit, and a process status fusion unit. The screening efficiency calculation unit is used to calculate the residual rate of each particle size in the oversize material and the extraction rate in the undersize material based on the particle size distribution vector of the feed, oversize material and undersize material, according to the preset particle size range, and generate a real-time screening probability matrix. The effective working area evaluation unit is used to analyze the temporal changes of the real-time screening probability matrix. When an unexpected decrease in the screening probability of the whole particle size or any particle size size is detected, the effective working area attenuation coefficient is calculated in combination with the screening machine operating parameters. The material retention analysis unit is used to calculate the proportion of retained fine particles relative to the total throughput by comparing the particle size distribution vectors of the feed and the oversize material, and this proportion is used as the material retention index. The process status fusion unit is used to integrate the real-time screening probability matrix, effective working area attenuation coefficient, and material retention index to generate a set of screening process status indicators.

[0033] It should be noted that the screening efficiency calculation unit obtains the particle size distribution vectors F_i, O_i, and U_i of the feed, oversize, and undersize materials within each particle size range from the point cloud particle size analysis module, where i represents the particle size range. The efficiency index is calculated based on the particle size distribution vectors, specifically including: (1) Residual rate: refers to the proportion of material that should have passed through the sieve but remains on the sieve. For a certain particle size i, its residual rate R_i can be approximately calculated as: R_i=(O_i) / (F_i), which reflects the proportion of the particle size that was not effectively separated. (2) Extraction rate: refers to the proportion of the target fine-grained material that is successfully screened into the undersize. For a certain particle size i, its extraction rate E_i can be approximately calculated as: E_i=(U_i) / (F_i), which reflects the proportion of the particle size that is effectively separated. The extraction rates of all particle size ranges are arranged in order of particle size to form a 1×N row vector or diagonal matrix, which is used to characterize the probability that each particle size material can pass through the sieve opening under the current operating state of the sieve.

[0034] Decreased probability of full-size sieving: This is manifested as a simultaneous decline in the extraction rate of all sizes, which usually means a reduction in the overall effective working area of ​​the sieve surface, which may result in large-area wear, uniform material layer coverage, or overall blockage. Decreased screening probability of any particle size: This refers to a significant decrease in the screening probability of one or more specific particle sizes, while other particle sizes may remain normal. This usually indicates a localized fault, such as abnormal passage of particles of a specific size due to localized damage to the screen, or obstruction of the passage of particles of a specific size due to localized blockage.

[0035] The determination of unexpected degradation specifically includes: First, under the initial healthy state or optimal operating condition of the screening machine, the system learns and stores a baseline screening probability matrix P_0. During real-time monitoring, the system compares the currently calculated screening probability matrix P_t with P_0. When the overall mean or the mean of a specific particle size of P_t decreases by more than the baseline mean in P_0 than the preset statistical control limit (usually using the 3σ principle) or the process allowable threshold, it is determined that "unexpected degradation" has occurred. This degradation is a continuous performance degradation signal that is different from normal feed fluctuations and is the starting point for triggering health assessment and fault diagnosis.

[0036] As one embodiment of the present invention, the effective working area evaluation unit performs the following when calculating the effective working area attenuation coefficient: The screening probability matrix for the k-th time period is modeled as a state-space model, as shown in the following formula: ; In the formula, For the sieving probability matrix, The state transition matrix is ​​used to characterize the inherent inertia of the sieving process. To characterize the process noise of feed fluctuations, To characterize the perturbation matrix of the screen surface health status caused by wear and clogging; Construct a matrix sequence from the screening probability matrices of K consecutive time periods. ,in ; Substituting the matrix sequence into the formula of the state-space model, the health state perturbation matrix is ​​calculated. The specific value; Extracting the health status perturbation matrix The Frobenius norm is calculated and normalized, and the normalized result is used as the effective working area attenuation coefficient.

[0037] It should be noted that the Frobenius norm is a commonly used method in matrix theory to measure the size or energy of a matrix. For an m×n matrix, its Frobenius norm is defined as the square root of the sum of the squares of the absolute values ​​of all its elements. In this patent, the Frobenius norm is derived from the healthy state perturbation matrix. The Frobenius norm is extracted from the matrix. Its physical meaning is: to accumulate the small perturbations (whether positive or negative) at all particle sizes and all potential locations in the matrix into a single positive scalar value. The larger the value, the more significant the overall deviation of the screen surface condition from the ideal healthy baseline, that is, the more serious the impact of wear or clogging problems.

[0038] As one embodiment of the present invention, the material retention analysis unit performs the following when calculating the material retention index: Within each analysis time window, the similarity measure between the particle size distribution vector of the feed and the particle size distribution vector of the material on the sieve is calculated, using the following formula: ; In the formula, This is the particle size distribution vector of the feed. Let be the particle size distribution vector of the material remaining on the sieve. For similarity measurement, It is particle size The weighting coefficient, for particle sizes smaller than the sieve aperture size, For particle sizes larger than the sieve aperture size, ; The material retention index is defined as follows: This is to amplify the effect of particle size retention smaller than the sieve aperture on the index.

[0039] As one embodiment of the present invention, the diagnostic decision module includes a rule matching unit, a fault assessment and tracing unit, a decision suggestion generation unit, and a diagnostic feedback and rule optimization unit. The rule matching unit receives the real-time screening probability matrix, the effective working area attenuation coefficient, and the material retention index, and performs item-by-item matching and reasoning according to the logical conditions and thresholds in the preset diagnostic rule base to generate the fault type. The fault assessment and tracing unit is used to quantify the fault level and locate the possible fault area based on the fault type, combined with the degradation rate and range of the real-time screening probability matrix and the attenuation coefficient of the effective working area. The decision suggestion generation unit is used to match maintenance suggestions from a preset maintenance strategy knowledge base based on the diagnosed fault type, fault level and possible fault area. Suggestions include adjusting the feed rate, stopping the machine for inspection, cleaning the designated area or replacing the screen plate. The diagnostic feedback and rule optimization unit is used to associate fault diagnosis conclusions, maintenance suggestions and actual maintenance effects to obtain closed-loop feedback data, and based on the closed-loop feedback data, automatically optimize and update the thresholds and logical relationships in the diagnostic rule base.

[0040] It should be noted that the diagnostic rule base is a structured "IF-THEN" logical set that maps quantitative indicators to specific fault types. The matching and reasoning process specifically includes: Condition matching: The system compares the three core indicators acquired in real time—real-time screening probability matrix, effective working area attenuation coefficient, and material retention index—with the logical conditions corresponding to each rule in the rule base. Specific example rules are provided below: Example rule 1: IF(effective working area decay coefficient > threshold A) AND(material retention index > threshold B) AND(decline rate of the smallest particle size element in the real-time screening probability matrix > threshold C) THEN Fault type = "screen blockage"; Example rule 2: IF (effective working area attenuation coefficient is greater than threshold D in local area) AND (the rate of increase of the value of the corresponding element of a certain intermediate particle size in the real-time screening probability matrix is ​​greater than threshold E) AND (material retention index is within the normal threshold range) THEN Fault type = "screen surface damage"; Each rule in the rule base contains a specific numerical threshold. The system compares real-time indicators with these thresholds to determine whether the conditions are met. It traverses all rules through forward chain reasoning to identify the rules that meet all conditions. If multiple rules are triggered simultaneously, they are sorted according to their priority, and one or more fault types with the highest priority are output. The numerical thresholds and priority of the rules are set based on the experience of professionals or industry paradigms.

[0041] The degradation rate of the real-time screening probability matrix can be obtained by calculating the derivative of the screening probability of the key granularity (usually the smallest granularity as the screening target) or the slope of the linear regression in the most recent time window (such as the past 10 minutes). A negative derivative or slope value indicates that the performance of the real-time screening probability matrix has deteriorated, and the larger the absolute value, the faster the performance deteriorates.

[0042] The scope refers to the breadth of granularity affected by performance degradation. By comparing the current matrix P_t with the healthy baseline matrix P_0, the relative decrease in probability of each granularity is calculated. A degradation judgment threshold is set (e.g., a decrease of more than 15% in the past 10 minutes). The number of granularities that exceed the degradation judgment threshold and their proportion in the total number of granularities are counted. Global degradation is characterized by most granularities (e.g., 80% or more of the total number of granularities) exceeding the degradation judgment threshold, while local degradation may only have 1-2 granularities exceeding the degradation judgment threshold.

[0043] Fault level determination can be achieved through a preset lookup table, while fault location relies on correlation analysis with the vision sensor array. For example, if the sieving probability matrix shows a severe decline in the extraction rate of a specific particle size, it can be inferred that the fault may occur in the main distribution area of ​​that particle size material.

[0044] As one embodiment of the present invention, the diagnostic feedback and rule optimization unit performs the following: The diagnostic process data of the current screening machine, including fault diagnosis conclusions, maintenance suggestions, and the actual effects after maintenance, are used as an incremental learning sample set. A pre-trained base diagnostic model is fine-tuned using few-shot transfer learning on this sample set. The decision boundary output by the fine-tuned basic diagnostic model is transformed into thresholds and logical relationships in the diagnostic rule base.

[0045] Furthermore, it should be noted that the present invention can be provided as a method, apparatus, or computer program product. Therefore, embodiments of the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, embodiments of the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code.

[0046] Embodiments of the present invention are described with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0047] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing terminal equipment to cause a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0048] It should also be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.

[0049] Finally, it should be noted that the above description represents a preferred embodiment of the present invention. It should be pointed out that although preferred embodiments have been described, those skilled in the art, once they understand the basic inventive concept of the present invention, can make various improvements and modifications without departing from the principles described herein. These improvements and modifications should also be considered within the scope of protection of the present invention. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the embodiments of the present invention.

Claims

1. A screening machine operation monitoring and control system based on image analysis, characterized in that, include: The image acquisition module is used to acquire time-series point cloud data of the feed flow, the oversize material discharge end, and the undersize material discharge end of the screen machine through three-dimensional vision measurement equipment deployed at the feed end, the oversize material discharge end, and the undersize material discharge end of the screen machine. The point cloud processing and particle size analysis module is used to segment the time-series point cloud data, identify the particle size information of the material, and generate particle size distribution vectors of the feed flow, oversize material and undersize material based on the particle size information. The process state quantification module is used to quantify the particle size distribution vector into a set of screening process state indicators through a predefined calculation model, including: a real-time screening probability matrix characterizing the separation efficiency of different particle sizes, an effective working area attenuation coefficient characterizing the degradation of the screen surface function, and a material retention index characterizing the screen surface throughput performance. The diagnostic decision module is used to receive the set of status indicators of the screening process, match them with the preset diagnostic rule base, and output fault diagnosis conclusions for abnormal screening efficiency, screen blockage or damage.

2. The image analysis-based screening machine operation monitoring and control system according to claim 1, characterized in that: The image acquisition module includes a sensor array unit, a timing synchronization triggering unit, and a point cloud preprocessing unit; The sensor array unit consists of three-dimensional vision measurement devices deployed at the feed end, the discharge end of the oversize material, and the discharge end of the undersize material of the screening machine, used to collect the original three-dimensional images of the material. The timing synchronization triggering unit is connected to the sensor array unit and is used to generate a unified acquisition triggering command according to a preset timing sequence to control each three-dimensional vision measurement device to perform synchronous data acquisition. The point cloud preprocessing unit is used to perform noise reduction, filtering, and coordinate correction on the acquired original 3D images to generate time-series point cloud data.

3. The image analysis-based screening machine operation monitoring and control system according to claim 1, characterized in that: The point cloud processing and particle size analysis module includes a point cloud registration and extraction unit, a material instance segmentation unit, a particle size feature extraction unit, and a dynamic distribution statistics unit. The point cloud registration and extraction unit is used to spatially align the time-series point cloud data obtained from different measurement points, remove the static background point cloud representing the structure of the fixed equipment, and separate the real-time material flow point cloud of the feed flow, the oversize material flow and the undersize material flow. The material instance segmentation unit is used to segment the real-time material flow point cloud into instances, extract each independent material particle point cloud from the material flow point cloud and identify it. The particle size feature extraction unit is used to calculate the equivalent particle size of each independent material particle point cloud; The dynamic distribution statistics unit is used to perform statistical analysis on the equivalent particle size of all independent material particles in the feed stream, oversize stream, and undersize stream according to a preset time window and a preset particle size interval, calculate the percentage of material quantity in each particle size interval, and generate particle size distribution vectors of the feed stream, oversize stream, and undersize stream as a function of time.

4. The image analysis-based screening machine operation monitoring and control system according to claim 3, characterized in that: The material instance segmentation unit performs the following: The real-time material flow point cloud is downsampled using voxelization, and the normal vector and curvature of each point are calculated after downsampling. Based on the spatiotemporal continuity constraint of material particle motion, the real-time material flow point cloud of multiple consecutive frames is correlated and tracked. By comparing the consistency of the motion trajectory of the real-time material flow point cloud between adjacent frames, point cloud clusters with different motion trajectories are separated, and a temporary tracking label is assigned to each point cloud cluster. For point cloud clusters that remain attached within a single frame after separation, the normal vector and curvature are constrained, and a region growing oversegmentation algorithm based on geodesic distance is executed to further divide the attached point cloud clusters into multiple possible granular regions, i.e. oversegmentation regions. The tracking tags are associated with oversegmented regions, and the missegmentation is corrected using a graph optimization method to obtain independent material particle point clouds. A global identifier is then assigned to each of the independent material particle point clouds.

5. The image analysis-based screening machine operation monitoring and control system according to claim 3, characterized in that: When the particle size feature extraction unit calculates the equivalent particle size, it performs the following: For each independent material particle point cloud, first calculate its minimum bounding cuboid, and obtain the length, width, height and orientation of the minimum bounding cuboid as the length and orientation of the longest, middle and shortest axes of the corresponding independent material particle point cloud. Project the point cloud of the particle onto a plane perpendicular to the longest axis, and calculate the equivalent circle diameter of the projected area; The equivalent particle size is calculated using the following formula: ； In the formula, The equivalent particle size is... and These are the longest and shortest spatial axes of the corresponding individual material particles, respectively. The equivalent circle diameter is... , and The respective , and The weighting coefficients are set according to the type of material.

6. The image analysis-based screening machine operation monitoring and control system according to claim 1, characterized in that: The process status quantification module includes a screening efficiency calculation unit, an effective working area evaluation unit, a material retention analysis unit, and a process status fusion unit. The screening efficiency calculation unit is used to calculate the residual rate of each particle size in the oversize material and the extraction rate in the undersize material based on the particle size distribution vector of the feed, oversize material and undersize material, according to the preset particle size range, and generate a real-time screening probability matrix. The effective working area evaluation unit is used to analyze the temporal changes of the real-time screening probability matrix. When an unexpected decrease in the screening probability of the whole particle size or any particle size is detected, the effective working area attenuation coefficient is calculated in combination with the screening machine operating parameters. The material retention analysis unit is used to calculate the proportion index of retained fine-grained materials relative to the total throughput by comparing the particle size distribution vectors of the feed and the oversize material, and use this as the material retention index. The process status fusion unit is used to integrate the real-time screening probability matrix, effective working area attenuation coefficient, and material retention index to generate a set of screening process status indicators.

7. The image analysis-based screening machine operation monitoring and control system according to claim 6, characterized in that: The effective working area evaluation unit performs the following when calculating the effective working area attenuation coefficient: The screening probability matrix for the k-th time period is modeled as a state-space model, as shown in the following formula: ； In the formula, For the sieving probability matrix, The state transition matrix is ​​used to characterize the inherent inertia of the sieving process. To characterize the process noise of feed fluctuations, To characterize the perturbation matrix of the screen surface health status caused by wear and clogging; Construct a matrix sequence from the screening probability matrices of K consecutive time periods. ,in ; Substituting the matrix sequence into the formula of the state-space model, the health state perturbation matrix is ​​calculated. The specific value; Extract the health state perturbation matrix The Frobenius norm is calculated and normalized, and the normalized result is used as the effective working area attenuation coefficient.

8. The image analysis-based screening machine operation monitoring and control system according to claim 6, characterized in that: The material retention analysis unit performs the following when calculating the material retention index: Within each analysis time window, the similarity measure between the particle size distribution vector of the feed and the particle size distribution vector of the material on the sieve is calculated, using the following formula: ； In the formula, This is the particle size distribution vector of the feed. Let be the particle size distribution vector of the material remaining on the sieve. For similarity measurement, It is particle size The weighting coefficient, for particle sizes smaller than the sieve aperture size, For particle sizes larger than the sieve aperture size, ; The material retention index is defined as follows: This is to amplify the effect of particle size retention smaller than the sieve aperture on the index.

9. The image analysis-based screening machine operation monitoring and control system according to claim 1, characterized in that: The diagnostic decision module includes a rule matching unit, a fault assessment and tracing unit, a decision suggestion generation unit, and a diagnostic feedback and rule optimization unit. The rule matching unit is used to receive the real-time screening probability matrix, the effective working area attenuation coefficient, and the material retention index, and perform item-by-item matching and reasoning according to the logical conditions and thresholds in the preset diagnostic rule base to generate the fault type. The fault assessment and tracing unit is used to quantify the fault level and locate the possible fault area based on the fault type, combined with the degradation rate and range of the real-time screening probability matrix and the effective working area attenuation coefficient. The decision suggestion generation unit is used to match maintenance suggestions from a preset maintenance strategy knowledge base based on the diagnosed fault type, fault level and possible fault area. The suggestions include adjusting the feed rate, stopping the machine for inspection, cleaning a designated area or replacing the screen plate. The diagnostic feedback and rule optimization unit is used to associate fault diagnosis conclusions, maintenance suggestions and actual maintenance effects to obtain closed-loop feedback data, and based on the closed-loop feedback data, automatically optimize and update the thresholds and logical relationships in the diagnostic rule base.

10. The image analysis-based screening machine operation monitoring and control system according to claim 9, characterized in that: The diagnostic feedback and rule optimization unit executes: The diagnostic process data of the current screening machine, including fault diagnosis conclusions, maintenance suggestions, and the actual effects after maintenance, are used as an incremental learning sample set. A pre-trained base diagnostic model is fine-tuned using few-shot transfer learning on this sample set. The decision boundary output by the fine-tuned basic diagnostic model is transformed into thresholds and logical relationships in the diagnostic rule base.

Images