A vibration screen mesh hole state monitoring method based on image processing

Abstract

The present application relates to the technical field of image processing, and discloses a vibration screen mesh hole state monitoring method based on image processing, comprising: acquiring a continuous image sequence of the screen mesh through a fixed camera, dividing an analysis area and generating a brightness oscillation time sequence of each area, adaptively extracting a reference oscillation sequence representing the overall vibration rhythm, and determining the abnormal state of the mesh hole through phase synchronism analysis, wherein the dynamic image sequence is converted into a time sequence signal analysis, early abnormality detection is realized through capturing the phase asynchronism of brightness oscillation, the limitations of traditional static form analysis are avoided, the fault type is distinguished based on harmonic characteristics, the vibration signal is reused to warn the health state of the whole machine, and the detection reliability and maintenance efficiency are improved without physical marker points.

Description

Technical Field

[0001] The invention relates to a method for monitoring mesh status of a vibrating screen based on image processing, and belongs to the technical field of image processing. Background Art

[0002] The current mainstream screen monitoring technology mainly attempts to capture changes in mesh morphology to determine blockage or damage by performing grayscale threshold segmentation or edge detection on static image frames. This type of method implicitly regards the high-frequency vibrating screen as an instantaneous static object, ignoring the complexity of the dynamic system composed of material flow, elastic deformation and environmental interference.

[0003] In typical working conditions such as mineral processing wet screening, when dust adhesion causes image blur or a thin layer of material adheres to the mesh but is not completely blocked, existing technologies face three fundamental bottlenecks: 1. The static analysis paradigm has difficulty distinguishing between real damage and morphological artifacts caused by instantaneous occlusion. Its essence is to simplify multi-dimensional dynamic problems into two-dimensional static recognition, resulting in a high false alarm rate; 2. In order to improve accuracy, high-speed cameras or deep learning models are introduced. Although they can partially improve the recognition effect, they significantly increase hardware costs and computing load, and cannot solve the systematic interference problem of dynamic interference; 3. Existing solutions can only determine the existence of anomalies, and cannot distinguish between structural fatigue cracks and viscosity-induced blinding blockages, two types of faults with completely different mechanisms and completely different treatment strategies.

[0004] While recent research has attempted to add vibration sensors to screen frames to aid analysis, this approach requires modifications to the equipment itself and is difficult to accurately synchronize with visual data, introducing new signal coupling errors. Therefore, the goal of this invention is to develop a method based on dynamic image time series analysis, without requiring physical markers, to achieve early detection of screen anomalies and accurate diagnosis of fault types under complex operating conditions. Summary of the Invention

[0005] The present invention provides a method for monitoring the mesh status of a vibrating screen based on image processing, the main purpose of which is to solve the problems that the existing static image analysis method is difficult to capture dynamic anomalies, the high-cost solution increases the complexity of the system, and is unable to accurately distinguish the fault type.

[0006] To achieve the above object, the present invention provides a method for monitoring the mesh status of a vibrating screen based on image processing, comprising the following steps:

[0007] Step a, obtaining a continuous image sequence, using a fixed image acquisition device to obtain a continuous image sequence of the vibrating screen when it is working;

[0008] Step b, dividing the analysis area, dividing the image area of ​​the screen into multiple analysis areas based on the continuous image sequence, and the division is completed when the method is initialized;

[0009] Step c: generating a brightness oscillation time series, and calculating the average brightness value of the pixels within each of the multiple analysis areas in real time, thereby obtaining a brightness oscillation time series corresponding to each analysis area;

[0010] Step d: Determine a reference oscillation sequence. Based on the brightness oscillation time series of multiple analysis areas, determine a reference oscillation sequence that characterizes the overall vibration beat of the vibrating screen. The determination process is to perform a fast Fourier transform on multiple brightness oscillation time series and extract the frequency with the strongest energy in the spectrum as the main operating frequency of the vibrating screen, thereby constructing a reference oscillation sequence.

[0011] Step e, determine the mesh state, compare the phase of the brightness oscillation time series of each analysis area with the reference oscillation sequence, and determine the state of the screen mesh corresponding to the analysis area in the following way: if the quantized value of the mutual relationship between the brightness oscillation time series of each analysis area and the reference oscillation sequence is lower than the preset synchronization threshold, then the screen mesh corresponding to the analysis area is determined to be in an abnormal state.

[0012] Preferably, in step d, the method for determining the reference oscillation sequence includes: performing spectral analysis on the brightness oscillation time series of multiple analysis areas that have been confirmed to be in a healthy state, and extracting the frequency peak with the strongest energy as the main working frequency of the vibrating screen; or, performing spectral analysis after overall averaging processing on the brightness oscillation time series of all analysis areas to extract the frequency peak with the strongest energy as the main working frequency of the vibrating screen.

[0013] Preferably, the phase comparison is performed by calculating the brightness oscillation time series of each analyzed region With the reference oscillation sequence The mutual correlation coefficient Quantify the correlation coefficient The calculation method is: in, is the length of the time series, is the time sampling point index, is the brightness oscillation time series The arithmetic mean of The reference oscillation sequence When the correlation coefficient is lower than the preset synchronization threshold, the mesh size of the sieve corresponding to the analysis area is determined to be in an abnormal state.

[0014] Preferably, the abnormal state includes an imminent breakage state or a blocked state; when the oscillation amplitude of the brightness oscillation time series of the analysis area decays below a preset amplitude attenuation threshold or loses periodic oscillation, the mesh size of the sieve corresponding to the analysis area is determined to be in a blocked state.

[0015] Preferably, the method does not require setting any physical marking points on the vibrating screen body before execution, and does not require obtaining a synchronization signal from an external device.

[0016] Preferably, after determining that the mesh size of the sieve corresponding to the analysis area is in an abnormal state, the method further includes the following steps: obtaining harmonic energy characteristics, performing spectral analysis on the brightness oscillation time series of the analysis area in the abnormal state to obtain the energy distribution characteristics of its harmonic components, the harmonic components including the fundamental frequency energy and the energy at the integer multiple frequency positions of the fundamental frequency; qualitatively diagnosing the abnormal type, and performing a qualitative diagnosis on the type of abnormal state based on the energy distribution characteristics of the harmonic components. The diagnosis is based on the following rules: if the energy proportion of the high-order harmonic components exceeds the preset first harmonic energy proportion threshold, it is determined to be a structural looseness or fatigue crack risk; if the energy proportion of the high-order harmonic components is lower than the preset second harmonic energy proportion threshold, and the ratio of the energy of the even-order harmonic components to the fundamental wave energy exceeds the preset ratio threshold, it is determined to be a viscous blinding blockage.

[0017] Preferably, the method also includes the following steps: periodically acquiring a reference oscillation sequence, repeating step d to periodically obtain a series of reference oscillation sequences arranged in chronological order; analyzing the time series evolution characteristics, performing time series statistical analysis on the frequency values ​​of a series of reference oscillation sequences to determine the long-term drift rate or short-term instability of the frequency value, the long-term drift rate is quantified by the slope of the frequency value changing with time, and the short-term instability is quantified by the standard deviation of the frequency value; warning the health status of the whole machine, based on the long-term drift rate or short-term instability of the frequency value, warning the health status of the whole system of the vibrating screen, and the warning is based on the following rules: if the absolute value of the long-term drift rate exceeds the preset drift rate threshold, the drive system performance attenuation warning is triggered; if the short-term instability exceeds the preset instability threshold, the transmission or support component abnormality warning is triggered.

[0018] Preferably, in step a, the image acquisition device is an industrial camera; the industrial camera is fixedly installed and facing the vibrating screen mesh.

[0019] Preferably, in step b, the analysis area is divided so that each analysis area covers at least one mesh; the division is performed by performing threshold segmentation or edge detection on an initial static image frame in a continuous image sequence to determine the distribution of the meshes of the screen.

[0020] Preferably, in step c, the generation process of the brightness oscillation time series is a large-scale parallel processing; the average brightness value of the pixel is the average grayscale value of the pixel.

[0021] Compared with the prior art, the present invention has the following beneficial effects:

[0022] 1. By converting the pixel brightness oscillations in the screen area into time series signals, the system directly captures the phase synchronization changes in the mesh vibration behavior. When the vibration beat is out of step due to structural fatigue or material adhesion in a local area, the cross-correlation between the brightness oscillation sequence and the reference beat naturally decays. This detection mechanism based on the asynchrony of motion behavior breaks away from the limitations of traditional static morphological analysis, enabling the system to identify early anomalies before the mesh undergoes physical deformation, providing a technical window for preventive maintenance.

[0023] 2. The system autonomously extracts the main vibration frequency from the brightness oscillations in the normal area and generates a global reference sequence as a dynamic reference system. This process does not rely on external synchronization signals and is naturally robust to gradual changes in illumination. Because the core of the algorithm focuses on phase correlation rather than the absolute value of brightness, this self-reference mechanism enables the system to maintain the stability of abnormality judgment under conditions of dust obstruction or light fluctuations, significantly reducing the false alarm rate caused by environmental interference. When a phase anomaly is detected, the system automatically triggers a deep spectral analysis of the brightness oscillation waveform. The surge in high-order harmonic energy caused by structural looseness and the enhancement of even-order harmonic characteristics caused by sticky and wet blockage form a fault fingerprint with a clear physical mechanism. This two-layer analysis of phase anomaly triggering + waveform feature diagnosis converts a single detection signal into a diagnostic basis that can distinguish fault types and guide targeted maintenance strategies.

[0024] 3. The periodically updated reference frequency sequence synchronously becomes a beat recorder of the operating status of the entire system. The long-term drift of the main frequency reflects the performance degradation of the drive system, while short-term fluctuations reveal loose transmission components. Without adding new sensors, this mechanism converts the by-products of screen monitoring into key indicators for the health assessment of the entire machine, achieving a capability leap from component-level to system-level monitoring. There is no need to set physical marking points on the screen throughout the process. Dynamic brightness changes are only captured by ordinary industrial cameras. The analysis area is divided using a simple segmentation of the initial frame. The core calculation focuses on one-dimensional signal processing, so that the system can significantly reduce hardware costs and computing power requirements while maintaining high robustness, which meets the universal deployment requirements of industrial scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 This is a flow chart of the health monitoring data flow and diagnostic analysis of the present invention;

[0026] Figure 2 This is a correlation coefficient change diagram of the vibrating screen mesh state monitoring system of the present invention;

[0027] Figure 3 This is a flowchart of abnormal diagnosis of the vibrating screen mesh status monitoring system of the present invention.

[0028] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0029] In order to make the purpose, features and advantages of the present invention more clearly understood, the present invention will be further elaborated below.

[0030] The present invention discloses a method for monitoring the mesh status of a vibrating screen based on image processing. The overall architecture of its technical solution begins with capturing a continuous dynamic image sequence of the screen working area through a fixedly deployed image acquisition device. Then, in the system initialization stage, the image is accurately deconstructed into a plurality of independent analysis areas, and for each analysis area, its real-time two-dimensional image information is converted into a one-dimensional brightness oscillation time series. Finally, through rigorous timing signal analysis, that is, comparing the phase synchronization between each local time series and an adaptively generated reference oscillation sequence, online monitoring of the microscopic state of the screen mesh, fault diagnosis and trend warning of the health status of the whole machine are realized.

[0031] Specifically, in order to ensure the objectivity of the monitoring results and the non-invasiveness of the deployment, this method does not require the deployment of any physical marking points on the vibrating screen body or reliance on external synchronization signals before execution. The process begins with an industrial camera facing the vibrating screen in a fixed posture. This unchanging geometric relationship ensures a constant mapping of image pixels and physical points on the screen. The installation parameters of the image acquisition equipment, including the viewing range, shooting angle, and imaging coverage area, are configured based on the size of the screen unit. The industrial camera should be fixed within a range of 0.8 to 1.0 meters from the screen plane, and its optical axis must maintain an angle of deviation of less than 3 degrees from the screen plane. When the camera resolution is set to 1280×720 pixels, its field of view should cover a mesh area of ​​no less than 10 rows and 10 columns, so that the minimum side length of each mesh in the image is no less than 30 pixels to meet the contour recognition accuracy required for subsequent analysis area division. This configuration is initialized during system In this stage, the coverage and imaging contrast are verified to meet the following conditions based on the grayscale histogram of the collected static image: the grayscale mean value of the mesh opening area is in the range of 30%-60% of the image dynamic range, and the grayscale mean value of the metal screen area is more than 50% higher than the upper limit of the grayscale range, so as to ensure that the vibration brightness oscillation sequence has sufficient dynamic response space and avoid subsequent data saturation. In order to accurately capture local anomalies and take into account the effective use of computing resources, the system performs a one-time analysis area division during initialization. This procedure is not a rough segmentation, but collects a frame of high-contrast image of the screen in a static state, and then calls the threshold segmentation or edge detection algorithm to perform pixel-level recognition on the physical contour of the screen mesh, and generates a standardized analysis area matrix based on the principle of ensuring that each analysis area completely covers at least one mesh. Once the matrix is ​​established, it becomes an immutable spatial reference benchmark for all subsequent dynamic analyses.

[0032] Since direct analysis of image morphology under high-frequency vibration is highly susceptible to artifacts introduced by transient material flow or ambient light changes, the present invention shifts the processing focus from traditional morphological analysis to analysis of the physical characteristics of the brightness oscillation time series. Specifically, for each analysis area divided above, the system calculates the average grayscale value of all pixels in the area in real time at each image sampling moment, and constructs the brightness mean value of this series evolving over time as a brightness oscillation time series that characterizes the vibration behavior of the area. ,in is the region index, is the time point; in order to cope with the real-time computing needs of massive pixel data, the brightness mean generation process adopts a large-scale parallel processing architecture, thereby ensuring the real-time response capability of the entire process. The essence of this conversion is to successfully reduce the high-dimensional visual pattern recognition problem to the analysis of the intrinsic rhythm of the one-dimensional time series signal, thereby fundamentally avoiding the limitations of static image analysis. In order to accurately distinguish between the real faults of local meshes and global disturbances such as frequency changes of the drive system, the system must establish a dynamic reference that can characterize the ideal synchronous vibration state of the whole machine; to this end, the system determines the reference oscillation sequence through a set of adaptive procedures The specific procedure is: first, the multiple brightness oscillation time series Perform fast Fourier transform to analyze its spectral composition. To ensure the global representativeness and robustness of the benchmark, the preferred solution is to select multiple analysis areas that have been confirmed to be in a healthy state in the initial stage or based on historical data, and then perform ensemble average of their brightness oscillation time series. Then, perform spectral analysis on the average series and extract the frequency peak with the most concentrated energy, which is calibrated as the main working frequency of the vibrating screen in the current working condition. Alternatively, in scenarios with higher automation requirements, the main frequency can be extracted after global averaging of the brightness oscillation time series of all analysis areas. Finally, the system constructs a pure benchmark oscillation sequence based on the accurately obtained main working frequency. The generation of this sequence provides a stable endogenous reference frame for subsequent synchronization comparison that is not affected by local illumination or uniform attachments and only concerns the vibration phase.

[0033] In the final judgment link of the mesh state, the core criterion is whether the phase synchronization of the local vibration beat and the global ideal beat is consistent; in order to quantitatively evaluate this synchronization, the system converts the time series of each analysis area into With the reference oscillation sequence Perform phase comparison by calculating the correlation coefficient between the two The calculation strictly follows the following formula: ,in, is the calculated length of the time series, and are the local and reference instantaneous brightness values, respectively, and and is the arithmetic mean of their respective time series; the synchronization threshold adopted by the system is set according to a rigorous calibration procedure: under the condition of healthy operation of the equipment, a period of baseline data is collected, the correlation coefficient of all healthy areas is calculated and its statistical distribution model is constructed, and the value at a specific standard deviation (such as three times the standard deviation) below the mean of the distribution is taken as the threshold; in continuous monitoring, if any area If the calculated value is lower than this threshold, the system will determine that the screen mesh corresponding to the area has entered an abnormal state. This detection mechanism based on the desynchronization of motion behavior ensures that the vibration anomaly caused by early structural fatigue or material adhesion can be captured before the macroscopic physical damage occurs to the mesh. When a certain analysis area is identified as an abnormal state, in order to provide more accurate guidance on the fault type for maintenance decisions, the system will automatically start a deep diagnosis program for the brightness oscillation time series of the abnormal area; the program first performs a spectrum analysis on the abnormal sequence again to obtain the energy distribution characteristics of its fundamental frequency and each harmonic component; then, the system performs qualitative diagnosis based on a rule base based on physical mechanisms. The establishment of this rule base is based on the analysis of a large amount of test data: if high-order If the total energy proportion of harmonics (such as the third, fourth, and fifth) shows a significant jump compared to the normal state and exceeds the preset threshold value of the first harmonic energy proportion, the anomaly is characterized by structural looseness or fatigue cracks, because structural nonlinearity is the direct cause of inducing high-order harmonics; conversely, if the high-order harmonic energy proportion is stable, but the ratio of the energy of the even harmonics to the fundamental wave energy exceeds another preset ratio threshold, the fault type is determined to be a blinding blockage caused by viscous materials. The physical basis for this is that viscous damping will introduce an asymmetric vibration recovery process, thereby enhancing the even harmonics in the spectrum; at the same time, if the oscillation amplitude of the sequence itself continues to decay to below the preset amplitude attenuation threshold or its periodicity is completely lost, it is directly confirmed as a serious blockage.

[0034] The initialization calibration procedure of the method starts with a set of reference image sequences acquired during the stable operation of the device in a healthy state, and based on this sequence, the following steps are completed: First, the statistical characteristics of the brightness oscillation time series of all analysis areas are calculated, and the overall mutual correlation values ​​are converted to The synchronization threshold is determined by subtracting three times the standard deviation from the mean of Secondly, the screen tension in a specific area is reduced by 10% in steps using a torque wrench, and the proportion of high-order harmonic energy is recorded at each level. , and independently spray standard viscosity fluid quantitatively to simulate the blockage accumulation process, and record the corresponding even harmonic and fundamental energy ratio , and then all the collected data are Projected into the two-dimensional feature space; finally, the linear discriminant analysis method is used to calculate the decision boundary that can effectively separate the two types of fault sample data point sets, structural looseness and viscous and wet blockage. The intercepts of this boundary on the two coordinate axes are defined as the first harmonic energy ratio threshold and the even harmonic and fundamental wave energy ratio threshold, respectively. During the viscous and wet blockage simulation process, the state corresponding to the oscillation amplitude decaying to less than 30% of the healthy state mean defines the specific value of the amplitude attenuation threshold; during the operation of the system, each core operation parameter is deterministically generated by the output of the previous link, among which the time series length used for spectrum analysis and cross-correlation calculation is , which is greater than the main vibration frequency The number of sampling points contained in 10 cycles is the smallest integer power of 2, the main frequency It is determined by the frequency peak with the strongest energy after fast Fourier transform of the average brightness oscillation sequence of all healthy areas or the entire area; the synchronization threshold The system starts periodic adaptive update and automatically filters out the The value is continuously higher than the current The analysis area of ​​​​the dynamic health reference group is formed, and the mean and standard deviation are recalculated based on the latest data of the group, and then updated. value, thus forming a complete, self-correcting closed-loop monitoring logic.

[0035] The spatial partitioning matrix used to construct the analysis area is based on the boundary extraction results of the physical structure of the screen in the initial static image. The partitioning procedure first uses the Otsu threshold segmentation method to preliminarily separate the screen and the background in the image. Then, the Sobel edge operator is used to extract the grid structure composed of metal wires in the screen area, identify all closed areas and calculate their area and boundary shape. All areas between The enclosed area between will be identified as a single mesh unit, where and Represent the mean and standard deviation of the area of ​​all closed areas respectively. The system then expands 10% edge redundancy with each mesh unit as the center to construct a matrix containing non-overlapping analysis areas. Each analysis area in the matrix contains only one physical mesh and completely coincides with its position. The matrix maintains spatial consistency during system operation. If the system detects that the global matching offset of the image exceeds 3 pixels, it will automatically relocate the coordinate reference of the analysis area to eliminate long-term error accumulation; the update period of the reference oscillation sequence and the dynamic adjustment parameters of the synchronization threshold are determined by the frequency stability and correlation distribution statistics within the preset time window. The update period is set to 3600 seconds by default. The system selects the cross-correlation in each period. A healthy data set is constructed from all analysis areas whose coefficients are greater than the threshold of the previous cycle. A fast Fourier transform is performed on the main vibration frequency of each analysis area in the data set within the cycle, and its most energetic frequency component is extracted. When the standard deviation of all main frequency value sets does not exceed 2% of the original mean, the frequency is considered stable, and a new benchmark oscillation sequence is allowed to be constructed with the mean frequency of the cycle. The cross-correlation synchronization threshold is composed of the distribution of the cross-correlation coefficients of all areas in the healthy data set under the benchmark sequence. The mean of this distribution minus three times the standard deviation is used as the judgment threshold for the next cycle. If the number of samples in the healthy data set is less than 30% of the total number of analysis areas, the benchmark update is not performed in the current cycle, and the benchmark sequence and threshold of the previous cycle are used.

[0036] The diagnostic model parameters and related thresholds of this method are deterministically generated in a set of offline systematic calibration protocols. This protocol begins with the preparation of a standard contamination liquid mixed with water, glycerin and 300 mesh quartz powder in a specific proportion. Its kinematic viscosity is calibrated and recorded using a rotational viscometer; then, for the simulation of structural loosening faults, the torque of a specific bolt is gradually reduced by 10% of the standard tightening torque of the equipment through a calibrated torque wrench, and the brightness oscillation time series is collected at each torque level. For the simulation of viscous blockage faults, after restoring the standard tension, the aforementioned standard contamination liquid is sprayed on the same area in a fixed dose of 5 ml each time, and the time series is also collected at each dose level; after obtaining sample data covering healthy and two types of faults with different severity levels, the system performs spectral analysis on each sample sequence to extract its high-order harmonic energy proportion. Energy ratio of even harmonics to fundamental wave , thereby projecting all samples into a two-dimensional feature space, and based on this data point set, the linear discriminant analysis method is used to calculate the decision boundary that can separate the two types of fault sample data point sets. The linear equation parameters of this boundary directly define the first harmonic energy ratio threshold and the even harmonic to fundamental energy ratio threshold. At the same time, during the viscous blockage simulation process, the state corresponding to the first stable decrease in the oscillation amplitude to below 30% of the average amplitude of the healthy state is defined as the amplitude attenuation threshold. The benchmark oscillation sequence that characterizes the overall vibration beat Synchronicity threshold is periodically calculated and adjusted in a closed-loop adaptive update procedure, where It is not a preset ideal function, but a time series of brightness oscillations of all analysis areas in a dynamic healthy reference group. It is generated in real time by performing ensemble averaging, and its calculation formula is: , is the number of members in the reference group; the membership and synchronization threshold of the dynamic health reference group At the end of each preset update period (for example, 3600 seconds), an iterative update is performed. The procedure is to first make the average value of the correlation coefficient in the previous period continuously higher than the current value. The regions were selected as candidate health groups, and then the validity of the candidate groups was verified. The verification conditions included that the number of its members was not less than 30% of the total number of analysis regions and the standard deviation of the mutual correlation coefficient within the group was not greater than 1.5 times the standard deviation of the initial health baseline. Only when both conditions were met at the same time, the candidate group was confirmed as the new dynamic health reference group, and the new mean of the mutual correlation coefficient was calculated based on its latest data. and standard deviation , and then update the synchronization threshold of the next cycle to If any verification condition is not met, the current cycle will not perform the update and the reference group and threshold of the previous cycle will be used.

[0037] Example 1: In a non-ferrous metal beneficiation wet screening process that operates continuously at high load, a key high-frequency vibrating screen faces a common operational and maintenance dilemma. Its screen is continuously subjected to the erosion of high-viscosity and highly abrasive slurry, which induces two failure modes with very different mechanisms: one is progressive viscous blinding blockage, in which fine mineral mud accumulates at the mesh. However, before it is completely blocked, traditional static visual systems are difficult to identify due to the insignificant morphological changes; the other is structural fatigue cracks under high stress. These microcracks are almost invisible under the slurry cover until they expand into tearing damage, causing unplanned downtime. To avoid such failures, the operation and maintenance party has to adopt experience-based preventive maintenance that far exceeds actual needs, resulting in a double loss of efficiency and resources. In this scenario, the accurate identification of early and distinguishable types of anomalies constitutes a challenge that existing technical paths cannot effectively address.

[0038] When the monitoring method of the present invention is applied to this working condition, it does not attempt to directly identify the geometry of the mesh covered by the slurry. Instead, it redefines the problem as a quantitative assessment of the health of the vibration behavior of each area of ​​the screen, thereby fundamentally changing the analytical paradigm. After the system is initialized, the continuous image sequence is converted into a brightness oscillation time series matrix covering the entire screen, and a baseline oscillation sequence representing the current overall vibration beat of the screening machine is adaptively constructed. ; When the device is running continuously for 72 hours, the corresponding brightness oscillation time series of an analysis area located in the center of the screen is and The mutual correlation coefficient , its calculated value is lower than the preset synchronization threshold for the first time, and the system determines that the area has entered an abnormal state. However, in the original image at this time, the mesh morphology of the area does not have any clearly identifiable blockage or damage.

[0039] The determination of this phase asynchronous abnormal state is not only an independent alarm signal, but also a high-quality precondition for triggering the in-depth diagnosis procedure. The system then The data stream is imported into the harmonic energy characteristic analysis module, which further verifies the inherent synergistic effect among the multiple technical features of the present invention; the spectrum analysis results show that the The energy distribution of the sequence showed a significant enhancement of even-order harmonics, while the energy of high-order harmonics remained within the normal range. Based on this clear fault fingerprint of the physical mechanism and according to the internal qualitative diagnostic rules, the system finally output a specific diagnostic conclusion: there is an early risk of sticky and wet blinding blockage in the center of the screen, rather than structural fatigue; based on this early warning, the operation and maintenance team carried out targeted flushing of the designated area during the intervals of planned team rotation, and quickly restored the normal vibration state of the area, thereby avoiding a potential blockage event that would cause a decrease in production line efficiency at a very low cost. Furthermore, on another vibrating screen of the same model running in parallel, the system also captured an abnormality in an edge analysis area through phase synchronization analysis after about 500 hours of continuous operation, but the subsequent harmonic energy characteristic analysis presented a completely different spectrum: the area In the spectrum, the energy proportion of the high-order harmonic components showed a large jump and exceeded the preset first harmonic energy proportion threshold. The system therefore determined that there was a risk of structural loosening or fatigue cracking. This diagnostic conclusion enabled the maintenance team to replace the entire network in a timely manner before the crack expanded to an irreversible level, avoiding a potential production safety accident. This operation example confirmed how the solution of the present invention utilizes a two-layer analysis architecture of phase anomaly triggering and waveform feature diagnosis to provide a reliable technical path for effectively distinguishing between two faults with different physical causes: sticky blockage and fatigue cracking within a single technical system.

[0040] Example 2: The core purpose of the experiment is to verify whether the monitoring method can accurately distinguish between structural looseness and sticky blinding blockage, two typical early faults that are easily confused in industrial sites, based solely on non-contact visual information in the initial stage of the abnormality; the test platform is built on a standard industrial vibrating screen with independently adjustable tension, and its main operating frequency is measured to be 15Hz. An industrial camera is fixed directly above the screen, and its field of view covers multiple analysis areas including the target test area; the image acquisition frame rate is set to ensure an optimized balance between distortion-free sampling of higher-order harmonics of vibration and the data processing load of the control system. Given that the diagnosis requires analysis up to the fifth harmonic of the fundamental frequency, that is, 75Hz, in order to meet the sampling theorem and retain sufficient waveform details, the sampling frequency should be higher than its Nyquist frequency of 150Hz. Therefore, in this experiment, the image acquisition frame rate is set to 200fps. This setting provides high-fidelity data input for subsequent harmonic energy characteristic analysis.

[0041] The test process begins with the vibrating screen running in a healthy state for 30 minutes to collect baseline data, from which the system calculates the The healthy mean and standard deviation were calculated, and the synchronization threshold was set at 0.92. Subsequently, the test was carried out in two stages. In the first stage, the torque of the adjacent tensioning bolts in the analysis area numbered 112 was gradually reduced by a calibrated torque wrench to simulate the accumulation process of structural loosening. In the second stage, after the equipment was restored to a healthy state, a controllable viscosity fluid mixed with water, glycerin and mineral fine powder was used to spray the same analysis area in a quantitative manner to simulate the gradual process of viscous blinding blockage. During both stages, the system continuously recorded the changes in key monitoring indicators. The evolution trend and quantitative data at key nodes can be seen in Table 1 below.

[0042] Table 1: Evolution of key monitoring indicators under different failure modes.

[0043]

[0044] The experimental data show that both failure modes cause the phase asynchrony between the brightness oscillation time series of the corresponding analysis area and the reference vibration series in their early stages of development, which is reflected in their mutual correlation coefficients. The phase synchronization analysis is a universal and effective means of early anomaly detection. After the anomaly is detected, the harmonic energy feature analysis shows a clear and differentiated fingerprint. As shown in Table 1, under the structural loosening condition, the energy proportion of high-order harmonics increases sharply with the increase in the degree of loosening, while the energy ratio of even-order harmonics does not change significantly. This phenomenon reflects the nonlinear impact effect introduced by the structural loosening. In contrast, under the viscous and wet blockage condition, the energy proportion of high-order harmonics remains at a low level, while the energy ratio of even-order harmonics to the fundamental wave increases significantly. This is consistent with the physical mechanism of the asymmetry of the vibration recovery process caused by viscous damping. This harmonic feature difference based on the physical mechanism provides a reliable quantitative basis for the subsequent qualitative diagnosis of faults.

[0045] Example 3: This example combines Figures 1 to 3 , a method for monitoring the mesh status of a vibrating screen based on image processing is described, such as Figure 1 As shown, first, in step a, image acquisition, an industrial camera is installed between the industrial camera and the polarizer to acquire images and obtain a continuous image sequence of the metal powder area; then, in step b, the region division phase is analyzed, the acquired image is preliminarily divided into regions, and a region coverage grid for each region is generated; in step c, a polarization information entropy map is generated, and the corresponding brightness amplitude sequence is generated by calculating the average brightness value of each region. After entering the time series information flow, step d determines the reference amplitude sequence, and extracts the main frequency of the reference amplitude through FFT analysis. Sequence; Subsequently, step e phase synchronization determination stage calculates the correlation coefficient ,like If the value is lower than the threshold, it is judged as an abnormal state and the deep diagnosis phase is triggered. If it is a normal state, continuous monitoring is performed. In the neighborhood energy feature analysis of step g, the polarization information entropy map and entropy value comparison analysis and structural feature analysis are used to identify the relevant features of structural looseness and stress changes, and finally obtain the diagnosis results, including abnormalities such as structural looseness, structural fatigue, or equipment failure caused by viscosity and wetness.

[0046] like Figure 2 As shown in the figure, the horizontal axis is the operating time in hours, which shows the performance of the vibration screen mesh condition monitoring system under different operating times; the vertical axis is the correlation coefficient , indicating the degree of correlation of each monitoring indicator. Several curves in the figure represent the changes in correlation coefficients under different states. Specifically, the thick solid line of the health state curve indicates the stability of the correlation coefficient when the screen is in a normal and healthy state; the dotted line of the structural looseness curve indicates the trend of the correlation coefficient gradually decreasing when the screen undergoes structural loosening; the dotted line of the viscosity and moisture blockage curve reflects the change of the correlation coefficient when the screen undergoes viscosity and moisture blockage.

[0047] like Figure 3 As shown, the abnormal area is detected by the monitoring system and sent to the spectrum analysis module. The spectrum analysis module receives data from the abnormal area and performs signal analysis. During this process, the spectrum analysis determines whether there is structural looseness or sticky and wet blockage by identifying and analyzing the changes in high-order harmonics and even-order harmonics. According to the results of the spectrum analysis, the system will generate a corresponding alarm report and send it to the operation and maintenance center for processing. The operation and maintenance center will further analyze and take corresponding measures based on the alarm content, including early warning of structural looseness of the mesh or timely handling of sticky and wet blockage problems.

[0048] Example 4: In industrial deployment, the problem faced is how to make the judgment base of the monitoring system adapt to the slow, non-fault drift of the monitored object due to normal wear or fine-tuning of the working conditions. If a judgment threshold that is calibrated and fixed in the initial state of the equipment is used, it may cause false alarms over time due to benchmark mismatch. For this purpose, this method configures a set of adaptive update procedures for synchronization thresholds. The procedure runs continuously in the background of the system. Its core is to maintain a dynamic health reference group composed of multiple analysis areas that continuously show high synchronization. The system recalculates the health reference group of all members in the health reference group at a preset update period, whose typical value is one hour. Mean of the values and standard deviation , and then the real-time synchronization threshold Dynamic adjustment is performed based on the following relationship: in, is a confidence coefficient whose value is related to the desired statistical confidence level. When it is set to 3, it corresponds to a confidence interval of approximately 99.7% under the normal distribution. This dynamic threshold mechanism enables the judgment benchmark to follow the overall distribution of the healthy data group, thereby effectively filtering out the slow, global normal state evolution of the entire system and only judging the local asynchronous behavior that significantly deviates from the current healthy group.

[0049] Furthermore, in order to establish an objective basis for the qualitative diagnosis of the two core fault types of structural looseness or fatigue crack risk and viscosity-induced blinding blockage, the present invention includes a systematic offline calibration protocol, which is executed when a new device is initially deployed or after an overhaul. The first stage of the protocol is the construction of a fault feature sample library. Under controlled conditions, the operator implants two types of faults into the standard screen. For structural looseness, the torque of a specific tensioning bolt is reduced in precise steps to simulate it. For viscosity-induced blockage, it is achieved by spraying a standard viscosity fluid in increments. At each severity level of each fault, the system collects and stores the brightness oscillation time series and its complete spectrum data for multiple vibration cycles. After obtaining the data containing the health, After obtaining a complete sample library of different fault types and degrees, the protocol enters the second stage, which is the statistical determination of the diagnostic boundary. The system projects all sample data into a two-dimensional feature space composed of the energy proportion of high-order harmonics and the energy ratio of even harmonics to fundamental waves. Since different fault types will form different data point clusters in this space, the system then uses linear discriminant analysis or other equivalent statistical classification algorithms to calculate the decision boundary that can effectively separate these fault clusters. The parameters of this boundary directly constitute the core parameters of diagnostic rules such as the first harmonic energy proportion threshold and the second harmonic energy proportion threshold. This calibration procedure based on data-driven and statistical principles provides a reproducible quantitative basis for qualitative fault diagnosis.

[0050] Similarly, the long-term drift rate threshold and short-term instability threshold on which the overall machine health status warning depends also follow a baseline calibration process based on the equipment's own operating characteristics. After the vibrating screen is installed and debugged and enters stable production, the system will start a baseline performance evaluation cycle of no less than one hundred effective working hours. During this period, the system continuously records the working main frequency sequence obtained by analyzing the benchmark oscillation sequence. After the evaluation cycle, the system performs statistical analysis on the frequency data for this long period of time and calculates its overall average fluctuation range and standard deviation. Whether the subsequent operating status monitoring value deviates significantly from the statistical characteristics of this baseline distribution becomes the basis for triggering the corresponding warning. In this way, the warning system of each device is configured according to its own initial operating characteristics, thereby improving the accuracy of the warning information.

[0051] Example 5: During the initial deployment and debugging phase of the system, an image quality calibration procedure is required to ensure that the image acquisition device obtains a brightness oscillation signal with an effective dynamic range and signal-to-noise ratio. The procedure first sets the exposure mode of the industrial camera to manual to prevent the automatic exposure adjustment caused by changes in ambient light from interfering with the intrinsic rhythm of the brightness oscillation time series; then, under standard lighting conditions, the shutter and aperture of the lens are adjusted to image a healthy screen in a stationary state until the grayscale histogram of its image presents a wide and centered distribution, that is, it is necessary to ensure that the pixel mean value of the light-transmitting area of ​​the mesh is significantly lower than the saturation value, while the pixel mean value of the screen wire area is significantly higher than the minimum value. This reserves sufficient quantization space for the subsequent capture of brightness oscillations caused by the passage of materials, thereby avoiding nonlinear distortion caused by saturation or cutoff of the signal.

[0052] In order to cope with the slight spatial drift of the entire screen relative to the camera field of view caused by extreme vibration or accidental impact, and thereby ensure the continuous and accurate correspondence between the analysis area and the physical mesh, the system can also be configured with a periodic spatial calibration and verification mechanism; this mechanism runs at a low frequency in the background and can be triggered at the beginning of each work shift or after the system detects abnormal vibration impact. After triggering, the system will use the known working main frequency information to predict and select an instantaneous frame at the peak of the vibration displacement and the clearest image, and perform template matching or phase correlation operation on the structural features of the frame with the static image template of the screen stored during the initial calibration. If the spatial displacement offset obtained by the operation exceeds the preset pixel tolerance, the system will automatically update the coordinate reference of all analysis areas or issue a prompt to the operator for manual fine calibration. This mechanism fundamentally regulates the monitoring errors that may be caused by the failure of the spatial mapping relationship, and is an extended implementation method known to ordinary technicians in this field.

[0053] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0054] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for monitoring the mesh status of a vibrating screen based on image processing, characterized in that: The following steps are involved: Step a, obtaining a continuous image sequence of the vibrating screen mesh when it is working by a fixedly set image acquisition device; Step b, dividing the image area of ​​the screen into a plurality of analysis areas based on the continuous image sequence, the division being completed when the method is initialized; Step c: for each of the multiple analysis areas, calculating the average brightness value of the pixels within the analysis area in real time, thereby obtaining a brightness oscillation time series corresponding to each analysis area; Step d: Based on the brightness oscillation time series of multiple analysis areas, a reference oscillation sequence that characterizes the overall vibration beat of the vibrating screen is determined. The determination process is carried out by performing a fast Fourier transform on the multiple brightness oscillation time series and extracting the frequency with the strongest energy in the spectrum as the main operating frequency of the vibrating screen, thereby constructing the reference oscillation sequence; Step e: performing a phase comparison between the brightness oscillation time series of each analysis area and the reference oscillation sequence, and determining the state of the mesh of the screen corresponding to the analysis area in the following manner: if the quantized value of the mutual relationship between the brightness oscillation time series of each analysis area and the reference oscillation sequence is lower than a preset synchronization threshold, then determining that the mesh of the screen corresponding to the analysis area is in an abnormal state; Phase comparison is performed by calculating the time series of brightness oscillations in each analyzed area. With the reference oscillation sequence The mutual correlation coefficient Quantify the correlation coefficient The calculation method is: ,in, is the length of the time series, is the time sampling point index, is the brightness oscillation time series The arithmetic mean of The reference oscillation sequence When the correlation coefficient is lower than the preset synchronization threshold, the mesh size of the sieve corresponding to the analysis area is determined to be in an abnormal state.

2. The method for monitoring the mesh status of a vibrating screen based on image processing according to claim 1, characterized in that: In step d, the method for determining the reference oscillation sequence includes: performing spectral analysis on the brightness oscillation time series of multiple analysis areas that have been confirmed to be in a healthy state, and extracting the frequency peak with the strongest energy as the main working frequency of the vibrating screen; or, performing spectral analysis after overall averaging processing on the brightness oscillation time series of all analysis areas to extract the frequency peak with the strongest energy as the main working frequency of the vibrating screen.

3. The method for monitoring the mesh status of a vibrating screen based on image processing according to claim 1, characterized in that: Abnormal states include imminent breakage or blockage. When the oscillation amplitude of the brightness oscillation time series in the analysis area decays below a preset amplitude decay threshold or loses periodic oscillation, the mesh size of the sieve corresponding to the analysis area is determined to be in a blockage state.

4. The method for monitoring the mesh status of a vibrating screen based on image processing according to claim 1, characterized in that: After determining that the mesh size of the sieve corresponding to the analysis area is in an abnormal state, the method further includes the following steps: performing spectral analysis on the brightness oscillation time series of the analysis area in the abnormal state to obtain the energy distribution characteristics of its harmonic components, the harmonic components including the fundamental frequency energy and the energy at the frequency position of integer multiples of the fundamental frequency; based on the energy distribution characteristics of the harmonic components, performing a qualitative diagnosis of the type of abnormal state, and the diagnosis is based on the following rules: if the energy proportion of the high-order harmonic components exceeds the preset first harmonic energy proportion threshold, it is determined to be a structural looseness or fatigue crack risk; if the energy proportion of the high-order harmonic components is lower than the preset second harmonic energy proportion threshold, and the ratio of the energy of the even-order harmonic components to the fundamental wave energy exceeds the preset ratio threshold, it is determined to be a viscous blinding blockage.

5. The method for monitoring the mesh status of a vibrating screen based on image processing according to claim 1, characterized in that: The method also includes the following steps: repeatedly executing step d to periodically obtain a series of reference oscillation sequences arranged in chronological order; performing time series statistical analysis on the frequency values ​​of the series of reference oscillation sequences to determine the long-term drift rate or short-term instability of the frequency values, the long-term drift rate is quantified by the slope of the frequency value changing with time, and the short-term instability is quantified by the standard deviation of the frequency value; based on the long-term drift rate or short-term instability of the frequency value, an early warning is issued for the health status of the entire system of the vibrating screen, and the early warning is based on the following rules: if the absolute value of the long-term drift rate exceeds a preset drift rate threshold, a drive system performance degradation early warning is triggered; if the short-term instability exceeds a preset instability threshold, a transmission or support component abnormality early warning is triggered.

6. The method for monitoring the mesh status of a vibrating screen based on image processing according to claim 1, characterized in that: In step a, the image acquisition device is an industrial camera; the industrial camera is fixedly installed and faces the vibrating screen mesh.

7. The method for monitoring the mesh status of a vibrating screen based on image processing according to claim 1, characterized in that: In step b, the analysis area is divided so that each analysis area covers at least one mesh; the distribution of the meshes of the screen is determined by performing threshold segmentation or edge detection on an initial static image frame in a continuous image sequence.

8. The method for monitoring the mesh status of a vibrating screen based on image processing according to claim 1, characterized in that: In step c, the generation process of the brightness oscillation time series is a large-scale parallel processing; the average brightness value of the pixel is the average grayscale value of the pixel.

Images