An industrial intelligent visual detection and diagnosis system for vertical shaft guide rails

Abstract

The present application relates to the vertical shaft cage detection technical field, specifically, it relates to a kind of industrial intelligent vision detection and diagnosis system for vertical shaft cage, acquisition unit synchronously obtains cage flange surface image and cage car pitch angle, coal dust shadow quantization unit uses bimodal adaptive threshold segmentation combined with region growing algorithm to extract coal dust area, introduce space weight factor to calculate average gray value and coverage area proportion, dynamic coupling calibration unit relies on three-dimensional look-up table and nonlinear interpolation of multi-working condition calibration, match the optimal deviation compensation coefficient of the combination of jitter and coal dust, generate dynamic gray scale segmentation threshold, defect diagnosis decision unit is through zoned adaptive segmentation, multiscale morphological filtering denoising, combined with gradient amplitude and curvature double-constraint subpixel edge detection accurately calculate wear depth, according to real-time depth and historical trend output grading control instruction, eliminate coupling interference, improve cage wear detection accuracy, provide reliable basis for safe operation.

Description

Technical Field

[0001] This invention relates to the field of vertical shaft tank passage inspection technology, and more specifically, to an industrial intelligent vision inspection and diagnostic system for vertical shaft tank passages. Background Technology

[0002] Vertical shaft cage inspection is an important technology, specifically applied to the precise detection and diagnosis of cage flange wear. Its core principle is to achieve accurate measurement of wear depth by eliminating coupling interference and optimizing image segmentation and edge detection. This aligns with the core requirements of vertical shaft cage safety operation and maintenance for accuracy and reliability. During vertical shaft cage inspection, cage movement generates pitch and vibration, and coal dust easily adheres to the cage flange surface in the vertical shaft environment. These two factors create a coupling effect, altering the grayscale distribution of the cage surface image, leading to deviations in wear depth measurements. This, in turn, affects the accuracy of cage defect diagnosis and fails to provide a reliable basis for cage maintenance. To address this technical problem, we provide an industrial intelligent vision inspection and diagnosis system for vertical shaft cages. Summary of the Invention

[0003] The purpose of this invention is to provide an industrial intelligent vision inspection and diagnostic system for vertical shaft tank passages, so as to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, an industrial intelligent vision inspection and diagnostic system for vertical shaft tank passages is provided, comprising:

[0005] The cage image and attitude acquisition unit acquires images of the cage flange surface and measures the cage pitch and jitter angle in real time.

[0006] The coal dust shadow quantization unit extracts the coal dust attachment area from the surface image of the tank flange using a gray-level histogram segmentation algorithm, and calculates the average gray value and coverage area ratio of the coal dust attachment area.

[0007] The dynamic coupling calibration unit includes a jitter and shadow coupling relationship mapping module and a grayscale threshold dynamic correction module. The jitter and shadow coupling relationship mapping module records the deviation compensation coefficient of the wear depth measurement value relative to the true value under different combinations of cage pitch jitter angle and coal dust grayness through a three-dimensional lookup table established by the calibration experiment. The grayscale threshold dynamic correction module matches the optimal compensation coefficient in the three-dimensional lookup table according to the real-time input cage pitch jitter angle and coal dust grayness to generate a dynamic grayscale segmentation threshold.

[0008] The defect diagnosis decision unit uses a dynamic grayscale segmentation threshold to segment the surface image of the tank runner flange, combines morphological filtering to remove residual noise from coal dust shadows, calculates the wear groove depth through a sub-pixel edge detection algorithm, and outputs instructions based on the wear groove depth.

[0009] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0010] This invention simultaneously acquires images of the cage flange surface and the cage pitch and jitter angle through a cage image and attitude acquisition unit, providing complete data support for subsequent coupling interference processing. The coal dust shadow quantization unit uses a bimodal adaptive threshold segmentation combined with a region growing algorithm to accurately extract the coal dust attachment area. It introduces a spatial weight factor to optimize the calculation of average gray value and coverage area ratio, avoiding quantization errors caused by uneven coal dust distribution. The dynamic coupling calibration unit relies on a three-dimensional lookup table calibrated under multiple working conditions and a nonlinear interpolation mechanism to match the optimal deviation compensation coefficient for different jitter and coal dust combinations, generating a dynamic gray-scale segmentation threshold that adapts to the current coupling state, effectively eliminating the interference of the coupling between the two on the gray-scale distribution. The defect diagnosis decision unit removes residual noise through partitioned adaptive segmentation and multi-scale morphological filtering, and accurately calculates the wear depth by combining sub-pixel edge detection with gradient amplitude and curvature dual constraints. Finally, it outputs graded control commands based on real-time depth and historical wear trends, improving the accuracy of cage wear detection and providing a reliable basis for the safe operation and maintenance of vertical shaft cages. Attached Figure Description

[0011] Figure 1 This is an overall block diagram of the present invention.

[0012] The meanings of the labels in the diagram are as follows:

[0013] 1. Tank tunnel image and attitude acquisition unit; 2. Coal dust shadow quantization unit; 3. Dynamic coupling calibration unit; 4. Defect diagnosis decision unit. Detailed Implementation

[0014] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0015] This invention provides an industrial intelligent vision inspection and diagnostic system for vertical shaft tank passages. Please refer to [link / reference]. Figure 1 As shown, it includes:

[0016] The cage image and attitude acquisition unit 1 acquires images of the cage flange surface and measures the cage pitch and jitter angle in real time.

[0017] The coal dust shadow quantization unit 2 extracts the coal dust attachment area from the surface image of the tank flange using a gray-level histogram segmentation algorithm, and calculates the average gray value and coverage area ratio of the coal dust attachment area.

[0018] The dynamic coupling calibration unit 3 includes a jitter and shadow coupling relationship mapping module and a grayscale threshold dynamic correction module. The jitter and shadow coupling relationship mapping module records the deviation compensation coefficient of the wear depth measurement value relative to the true value under different combinations of cage pitch jitter angle and coal dust grayness through a three-dimensional lookup table established by the calibration experiment. The grayscale threshold dynamic correction module matches the optimal compensation coefficient in the three-dimensional lookup table according to the real-time input cage pitch jitter angle and coal dust grayness to generate a dynamic grayscale segmentation threshold.

[0019] The defect diagnosis decision unit 4 uses a dynamic grayscale segmentation threshold to segment the surface image of the tank runner flange, combines morphological filtering to remove residual noise from coal dust shadows, calculates the wear groove depth through a sub-pixel edge detection algorithm, and outputs instructions based on the wear groove depth.

[0020] When extracting the coal dust attachment region from the flange surface image of the tank runner using the gray-level histogram segmentation algorithm, the coal dust shadow quantization unit 2 employs a bimodal adaptive threshold segmentation technique, specifically including:

[0021] Gaussian filtering is applied to the surface image of the tank passage flange to eliminate high-frequency noise and generate a smooth gray-level histogram. Based on the peak distribution characteristics of the smooth gray-level histogram, the main gray-level peak representing the normal tank passage surface and the secondary gray-level peak representing the coal dust shadow are identified. The valley gray-level values ​​of the main gray-level peak and the secondary gray-level peak are used as the initial segmentation threshold. The initial segmentation result is then optimized by combining a region growing algorithm. Starting from the pixels below the initial segmentation threshold, the connected domain is dynamically expanded according to the gray-level gradient change of the eight neighboring pixels until the gradient change exceeds the preset tolerance. Finally, the coal dust attachment area is output.

[0022] After obtaining the coal dust attachment area, when calculating the average gray value and coverage area ratio of the coal dust attachment area, a spatial distribution weighting factor is introduced, which specifically includes:

[0023] The surface image of the cage flange is divided into several grid sub-regions. A weighting coefficient is assigned according to the proportion of coal dust mask pixels in each sub-region. The average gray value is calculated using a weighted average formula. The weighting coefficient is positively correlated with the coal dust coverage rate of the sub-region. The coverage area proportion is determined by the ratio of the total number of coal dust mask pixels to the total number of effective pixels in the detection area, and is associated with the real-time value of the cage pitch jitter angle. When the cage pitch jitter angle increases, the distortion correction coefficient is enabled for the calculation of the coverage area of ​​the edge region.

[0024] The jitter and shadow coupling relationship mapping module is implemented through a multi-condition calibration experimental platform when a 3D lookup table is established using calibration experiments. Specifically, it includes:

[0025] The calibration platform includes a simulated cage support with controllable pitch angle, a cage track test block with an adjustable grayscale coating, and a standard calibration module with a preset depth wear groove. Under each set of calibration conditions, the cage pitch shaking angle, simulated coal dust grayscale value, and wear depth imaging measurement value are collected simultaneously. By comparing the wear depth imaging measurement value with the actual depth of the standard calibration module, the deviation compensation coefficient under that condition is calculated, and finally a three-dimensional lookup table is formed with shaking angle and coal dust grayscale value as independent variables and deviation compensation coefficient as dependent variable.

[0026] A three-dimensional lookup table records the deviation compensation coefficient between the measured wear depth and the true value under different combinations of cage pitch and shaking angles and coal dust ash levels. A nonlinear interpolation mapping mechanism is used, specifically including:

[0027] For the combination points of shaking angle and coal dust ash level not covered by the calibration experiment, a three-dimensional Delaunay triangulation mesh is constructed based on the nearest neighbor calibration data points. The compensation coefficient is calculated by centroid coordinate interpolation within the triangular unit. At the same time, the correlation rule between the deviation compensation coefficient and the wear depth is established. The compensation coefficient changes piecewise with the increase of wear depth. When the wear depth is less than the critical value, the compensation coefficient has a linear relationship with the depth. After the critical value is exceeded, the quadratic polynomial compensation model is activated.

[0028] The grayscale threshold dynamic correction module performs time-series alignment and data filtering based on the real-time input cage pitch angle and coal dust grayscale, specifically including:

[0029] Hardware synchronization signals ensure that the jitter angle timestamps acquired by the gyroscope are strictly aligned with the image acquisition unit. Sliding window mean filtering is used to suppress transient noise for jitter angle data of multiple consecutive frames. For coal dust gray values, inter-frame weighted averaging is performed based on the obtained spatial weighting results. The filtered jitter angles and coal dust gray values, together with the initial estimate of the wear depth of the current detection area, are used as the input index for the three-dimensional lookup table.

[0030] When matching the optimal compensation coefficient in the three-dimensional lookup table, a two-parameter collaborative indexing strategy is executed, which includes:

[0031] Using the filtered jitter angle and coal dust ash value as reference coordinates, the nearest calibration point data is retrieved from the three-dimensional lookup table. When the input point is located within a triangular unit formed by multiple calibration data, the centroid coordinate interpolation is called to calculate the real-time compensation coefficient. The specific process of generating the dynamic grayscale segmentation threshold is as follows:

[0032] The real-time compensation coefficient is added to the basic static threshold, and then multiplied by the scaling factor modulated by the proportion of coal dust coverage area. Finally, a dynamic grayscale segmentation threshold that adapts to the current jitter and shadow coupling state is output.

[0033] When segmenting the surface image of the tank runner flange using a dynamic grayscale segmentation threshold, a partition-adaptive segmentation strategy is employed, specifically including:

[0034] Based on the divided grid sub-regions, a dynamic grayscale segmentation threshold is independently assigned to each sub-region. The dynamic grayscale segmentation threshold is adjusted by combining the generated global dynamic threshold with the local coal dust coverage rate and spatial location weight of the sub-region. For sub-regions located at the image edge, an additional threshold offset positively correlated with the jitter angle is introduced. After segmentation, the segmentation is verified by connected component labeling. For abnormal segmentation regions, a secondary segmentation based on grayscale texture features is initiated.

[0035] Combining morphological filtering to remove residual noise from coal dust shadows, a multi-scale adaptive selection mechanism for structural elements is employed, specifically including:

[0036] Based on the statistical characteristics of the noise connected domain in the obtained segmentation results, the shape and size of the structural element are dynamically selected. When the residual noise is mainly composed of small-area discrete points, a circular structural element is used for opening operation. When the residual noise is distributed in strips, a long strip structural element consistent with the direction of the tank passage texture is used to perform closing operation. The size of the structural element is jointly controlled by the proportion of coal dust coverage area and the shaking angle. When the shaking angle increases, the size of the structural element increases accordingly.

[0037] When calculating the depth of the wear groove using a sub-pixel edge detection algorithm, edge localization is implemented with dual constraints of gradient magnitude and curvature, specifically including:

[0038] In the denoised image, grayscale profile curves are extracted along the normal direction of the tank flange. Subpixel edge localization is achieved using cubic spline interpolation. Edge point screening must simultaneously meet the gradient magnitude threshold and local curvature threshold to exclude false edges generated by residual coal dust shadows. The calculated wear groove depth is calibrated by the deviation compensation coefficient and then input into the defect diagnosis decision unit 4. Based on the real-time depth value and historical wear trend, the defect diagnosis decision unit 4 outputs a graded control command and marks the wear position coordinates when the depth value exceeds the warning threshold multiple times consecutively.

[0039] Further explanation is needed: after the cage image and attitude acquisition unit completes the acquisition of images of the cage flange surface and the measurement of the cage pitch and shaking angle, the coal dust shadow quantification unit needs to accurately extract the coal dust attachment area. Since coal dust attachment in the vertical shaft cage environment will form a low grayscale shadow, which is significantly different from the high grayscale of the normal cage surface, a bimodal adaptive threshold segmentation technique is adopted. The specific implementation method is as follows:

[0040] When the coal dust shadow quantization unit extracts the coal dust attachment area using the gray-level histogram segmentation algorithm, it relies on the bimodal adaptive threshold segmentation technology. First, Gaussian filtering is applied to the surface image of the tank duct flange to eliminate high-frequency noise. Gaussian filtering is a linear smoothing filter based on a normal distribution, which can suppress high-frequency noise while preserving the main features of the image. Specifically, a 3×3 convolution kernel is selected, with a standard deviation of 0.8. The convolution kernel is then used to perform a weighted average operation with each pixel in the image and its neighboring pixels to generate a smoothed image. Based on this image, the number of pixels corresponding to each gray value is counted to form a smoothed gray-level histogram, laying the foundation for subsequent peak identification. After generating the smoothed gray-level histogram, the main gray-level peak representing the normal tank duct surface is identified based on its peak distribution characteristics. The smooth grayscale histogram, representing the shadow of coal dust, uses grayscale values ​​on the x-axis and the number of pixels corresponding to those values ​​on the y-axis. Normal tank surface materials reflect light uniformly, resulting in higher grayscale values ​​and forming the main grayscale peak with the most pixels. Areas with coal dust adhesion reflect less light, resulting in lower grayscale values ​​and forming the secondary grayscale peak with the second-highest number of pixels. During identification, a sliding window method (with a window size of 5 grayscale levels) is used to traverse the histogram. Peaks are identified when the number of pixels within the window first increases and then decreases. The peaks are sorted by pixel count, with the first peak being the mid-grayscale peak and the second peak the secondary grayscale peak. The grayscale value intervals corresponding to the two peaks are also recorded. The valley grayscale value between the main and secondary grayscale peaks is used as the initial segmentation threshold; the valley is the grayscale point with the fewest pixels between the two peaks. The initial segmentation threshold is determined by finding the minimum number of pixels within the grayscale range between the primary and secondary grayscale peaks. For example, the primary grayscale peak range is 150-200, the secondary grayscale peak range is 30-80, and the valley grayscale value is 100. This value serves as the initial segmentation threshold, used to initially distinguish between coal dust shadows (below the threshold) and normal tunnel surfaces (above the threshold). To correct potential boundary blurring issues in the initial segmentation, a region growing algorithm is used to optimize the initial segmentation results. The region growing algorithm starts with pixels below the initial segmentation threshold (i.e., coal dust candidate pixels), using each candidate pixel as a seed point. It then searches for its eight neighboring pixels (vertical, horizontal, left, right, and four diagonal directions) and calculates the grayscale gradient change between the seed point and its neighboring pixels, i.e., the grayscale values ​​of the two pixels. The absolute difference is used to dynamically expand the connected domain based on the gradient change magnitude. If the gradient change magnitude is less than or equal to the preset tolerance (set to 10), it is determined through experiments that this ensures the connectivity within the coal dust region while avoiding the erroneous inclusion of pixels from the normal tank surface. In this case, the neighboring pixel is included in the current connected domain, and it is used as a new seed point to continue expanding until the gray-scale gradient change of the neighboring pixel exceeds the preset tolerance. At this point, the expansion of the connected domain stops. After the connected domain expansion of all seed points is completed, all connected domains that meet the conditions are integrated, and the coal dust attachment region is finally output. This region is presented in the form of a binary mask, with the pixels in the coal dust attachment region marked as 1 and the pixels on the normal tank surface marked as 0. This provides an accurate regional basis for subsequent calculation of the average gray-scale value of coal dust and the coverage area ratio.

[0041] After extracting the binarized mask of the coal dust adhesion area using the bimodal adaptive threshold segmentation technique, to avoid quantization errors caused by uneven distribution of coal dust on the flange surface of the guide tube, a spatial distribution weighting factor needs to be introduced to accurately calculate the average gray value and the coverage area ratio, ensuring that the results closely match the actual coal dust adhesion state. The specific implementation method is as follows:

[0042] After obtaining the coal dust attachment area, when calculating the average gray value and coverage area ratio, the surface image of the tank runner flange is first divided into several grid sub-regions. The division is determined based on the image resolution. For example, for a 1920×1080 pixel image, a 32×18 grid division rule is used, with each sub-region measuring 60×60 pixels. This size avoids computational redundancy caused by excessively small sub-regions while accurately capturing local differences in coal dust distribution. During the division, grid lines are automatically generated using the image coordinate system, and each sub-region is assigned a unique coordinate identifier for easy subsequent weight matching and data traceability. A weight coefficient is assigned based on the proportion of coal dust mask pixels within each sub-region. The proportion of coal dust mask pixels is the percentage of coal dust pixels marked within the sub-region. The weighting factor is the ratio of the number of pixels (pixels with a value of 1 in the binary mask) to the total number of pixels in the sub-region. For example, if a sub-region has a total of 3600 pixels and 1800 coal dust pixels, accounting for 50%, the weighting factor is set according to the principle of positive correlation with the coal dust coverage of the sub-region. The weighting factor is determined through calibration experiments: 0.5 for 0-20% (sparse coal dust), 0.8 for 20%-50% (medium coal dust), and 1.2 for 50%-100% (dense coal dust). The higher the proportion, the greater the weight, ensuring that the dense coal dust area contributes more significantly to the calculation of the average gray value and avoiding the sparse area from lowering the overall gray value representativeness. The average gray value is calculated using a weighted average formula, and the specific process is as follows:

[0043] First, calculate the local average grayscale of the coal dust-attached pixels within each sub-region. This is the sum of the grayscale values ​​of all coal dust pixels within that sub-region divided by the number of coal dust pixels. Then, multiply the local average grayscale of each sub-region by its corresponding weight coefficient to obtain a weighted local grayscale value. Next, sum the weighted local grayscale values ​​of all sub-regions and divide by the sum of the weight coefficients of all sub-regions. The final result is the global average grayscale value of the coal dust-attached region. For example, if one sub-region has a local average grayscale of 50 and a weight of 0.8, and another sub-region has a local average grayscale of 40 and a weight of 1.2, the weighted sum is 50 × 0.8 + 40 × 1.2 = 40 + 48 = 88, the total weight is 2.0, and the global average grayscale value is 44. This value more accurately reflects the overall grayscale characteristics of the coal dust. The coverage area ratio is determined by the ratio of the total number of pixels in the coal dust mask to the total number of effective pixels in the detection area. The total number of pixels in the coal dust mask is the sum of the number of coal dust pixels in all sub-regions. The total number of effective pixels in the detection area needs to be reduced by the number of pixels at the image edges. Invalid pixels, for example, if the total number of pixels in an image is 2,073,600, and there are 118,800 invalid pixels, the total number of valid pixels is 1,954,800. If the total number of pixels in the coal dust mask is 97,740, the initial coverage area ratio is 5%. At the same time, this calculation needs to be associated with the real-time value of the cage pitch jitter angle, because the pitch jitter of the cage will cause geometric distortion in the edge area of ​​the image. When the real-time value collected by the jitter angle sensor exceeds the preset threshold, the distortion correction coefficient is enabled for the calculation of the coverage area of ​​the edge area. The distortion correction coefficient is obtained through multi-condition calibration experiments. For example, the correction coefficient for the edge sub-region is 1.1 when the jitter angle is 6 degrees, and 1.2 when it is 7 degrees. During correction, the number of coal dust pixels in the edge sub-region is multiplied by the corresponding correction coefficient and then included in the total number of pixels to ensure that the coverage area ratio is not affected by distortion. For example, if the original number of coal dust pixels in the edge sub-region is 1,000, the correction coefficient is 1.1, and after correction, it is 1,100. The final ratio of the total number of pixels to the total number of valid pixels is the corrected coverage area ratio.

[0044] The core of the dynamic coupling calibration unit is to establish the correlation between the cage pitch shaking angle, coal dust ash level, and wear depth measurement deviation. This relationship needs to be simulated through a multi-condition calibration experimental platform to generate a three-dimensional lookup table, providing data support for subsequent real-time calibration. The specific implementation method is as follows:

[0045] When the jitter and shadow coupling relationship mapping module establishes a 3D lookup table through calibration experiments, it relies on a multi-condition calibration experimental platform. This platform needs to fully simulate the key variables of vertical shaft cage inspection, specifically including three core components: First, a simulated cage support with controllable pitch angle. The support is driven by a stepper motor and can accurately adjust the pitch angle, with an adjustment range of 0-10 degrees, consistent with the actual cage jitter range. The support is equipped with the same industrial camera and gyroscope as the actual inspection, simulating image acquisition and attitude measurement during cage operation. Second, a cage test block with an adjustable grayscale coating. The test block material is consistent with the actual cage flange. The surface is coated with coal dust simulation paint of different concentrations to achieve adjustable grayscale values, with an adjustment range of 30-150, covering the grayscale range of actual coal dust adhesion. After each adjustment, the actual grayscale value is measured and recorded using a grayscale meter. Third, a standard calibration module with preset depth wear grooves. The module processes rectangular wear grooves of different depths on the cage test block. The actual depth of the grooves is calibrated by a laser thickness gauge as the benchmark value for wear depth measurement. Under each set of calibration conditions, three types of key data need to be collected simultaneously: First, a fixed pitch jitter angle is set using the stepper motor of the simulated cage support, and the gyroscope collects real-time angle values ​​at a sampling frequency of 10Hz. The average value of 10 samples is taken as the jitter angle for that condition. Second, an industrial camera is used to capture images of the surface of the cage guide test block, and the average gray value of the simulated coal dust area is calculated as the simulated gray value of the coal dust for that condition. Finally, the same industrial camera is used to capture images of the wear grooves, and the same edge detection algorithm as in actual testing is used to calculate the imaging measurement value of the grooves as the imaging measurement value of the wear depth for that condition. Each set of conditions needs to be collected repeatedly. The data is collected three times, and the average value is taken as the final data for this working condition to avoid single measurement errors. After the data acquisition is completed, the deviation compensation coefficient for this working condition is calculated by comparing the wear depth imaging measurement value with the actual depth of the standard calibration module. The deviation is the difference between the wear depth imaging measurement value and the actual depth. The calculation logic of the deviation compensation coefficient is compensation coefficient = 1 - (deviation / actual depth). For example, under a certain working condition, the actual depth is 1mm, the imaging measurement value is 0.8mm, the deviation is -0.2mm, and the compensation coefficient = 1 - (-0.2 / 1) = 1.2. This coefficient can correct the measurement value to 0.8 × 1.2 = 0.96mm, which is close to the actual value. Finally, the calibration data for all operating conditions were organized in a structured form to form a three-dimensional lookup table with the shaking angle and coal dust ash value as independent variables and the deviation compensation coefficient as the dependent variable. The row dimension of the table is the cage pitch shaking angle, and the column dimension is the coal dust ash value. Each cell of the table stores the deviation compensation coefficient for different wear depths under the corresponding shaking angle and ash value. This lookup table is pre-stored in the storage module of the dynamic coupling calibration unit and supports fast indexing by shaking angle and coal dust ash value, providing accurate compensation basis for subsequent real-time dynamic calibration.

[0046] After constructing a three-dimensional lookup table through multi-condition calibration experiments, the calibration conditions need to be set at fixed step sizes, which cannot cover all possible combinations of cage pitch shaking angle and coal dust ash level in actual testing. Directly discarding these uncovered points would lead to real-time calibration interruption or decreased accuracy. Therefore, a nonlinear interpolation mapping mechanism is required to calculate the deviation compensation coefficient of uncovered points based on existing calibration data. At the same time, the compensation rules are refined by combining the influence of wear depth to ensure the full coverage and calibration accuracy of the three-dimensional lookup table. The specific implementation method is as follows:

[0047] When recording the deviation compensation coefficients for different cage pitch shaking angles and coal dust ash levels using a 3D lookup table, the core mechanism fills in the gaps of uncovered combination points through a nonlinear interpolation mapping mechanism. Firstly, for the shaking angle and coal dust ash level combination points not covered in the calibration experiment, a 3D Delaunay triangulation mesh is constructed based on the nearest neighbor calibration data points. The 3D Delaunay triangulation mesh is a series of non-overlapping and non-intersecting tetrahedrons constructed using all calibration data points as vertices. Its core characteristic is that the circumsphere of each tetrahedron does not contain other calibration vertices, ensuring the uniqueness and smoothness of the interpolation results. The Qhull open-source algorithm library is used during construction, inputting the coordinates of all calibration data points into the algorithm to automatically generate the triangulation mesh. Simultaneously, a mesh validity verification module checks the volume of each tetrahedron, re-triangulating any unqualified tetrahedrons. To ensure mesh quality, after determining the triangular element containing the uncovered point, the deviation compensation coefficient of that point is calculated through barycentric coordinate interpolation. The barycentric coordinates represent the position of the uncovered point within the tetrahedron using the weights of the four vertices, with the sum of the four weights being 1. This intuitively reflects the distance relationship between the uncovered point and each vertex. In the specific calculation, the tetrahedron to which the uncovered point belongs is first determined by spatial coordinate comparison, and the jitter angle and coal dust ash coordinates of the four vertices of the tetrahedron, as well as the corresponding deviation compensation coefficients, are extracted. Then, the barycentric coordinate weights of the uncovered point relative to the four vertices are solved through linear algebra operations. Finally, the deviation compensation coefficient of each vertex is multiplied by the corresponding weight, and the product results are added together to obtain the deviation compensation coefficient of the uncovered point. At the same time, considering that the wear depth itself affects the deviation compensation coefficient, a correlation rule between the deviation compensation coefficient and the wear depth needs to be established.First, a supplementary calibration experiment was conducted to determine the critical value of the wear depth. The critical value is the depth at which the compensation coefficient changes with the wear depth at an inflection point. Specifically, the experiment was carried out using a standard calibration module. Under fixed shaking angles and coal dust ash levels, the imaging measurement values ​​and actual depths of grooves at different depths were measured, and the corresponding deviation compensation coefficients were calculated. The trend of the compensation coefficient with depth was observed. The experiment found that when the wear depth is less than 2mm, the compensation coefficient increases uniformly with increasing depth. When the depth exceeds 2mm, the increase in the compensation coefficient becomes larger. Therefore, 2mm was set as the critical value of the wear depth and pre-stored in the association rule base of the three-dimensional lookup table. Based on this critical value, the compensation coefficient changes piecewise with increasing wear depth: when the wear depth is less than the critical value, the compensation coefficient has a linear relationship with the depth. The slope and intercept of the linear relationship were determined by fitting the calibration data. For example, under the above fixed working conditions, a depth of 0.5mm corresponds to a compensation coefficient of 1.05, and 1mm corresponds to 1.1 and 1, respectively. 5mm corresponds to 1.15, and 2mm corresponds to 1.2. A linear formula is obtained by fitting (compensation coefficient = 0.95 + 0.2 × wear depth). Input any depth value less than 2mm to quickly calculate the corresponding compensation coefficient. When the wear depth exceeds the critical value (2mm), a quadratic polynomial compensation model is activated. The coefficients of this model (such as a, b, c) are also fitted by calibration data. For example, depths of 2mm (1.2), 3mm (1.28), and 5mm (1.4) are substituted into the quadratic polynomial (compensation coefficient = a × depth² + b × depth + c) to obtain a = 0.0125, b = 0.025, and c = 1.15. Substituting into the 3mm depth, we get 1.28, which is consistent with the calibration data. This ensures that the compensation coefficient can still accurately match the actual deviation after exceeding the critical value. Finally, through this segmented mechanism, the deviation compensation coefficient is fully correlated with the three parameters of shaking angle, coal dust ash, and wear depth, providing complete data support for subsequent dynamic calibration.

[0048] After the 3D lookup table is constructed, the grayscale threshold dynamic correction module needs to generate a dynamic grayscale segmentation threshold based on the real-time input cage pitch jitter angle and coal dust grayscale data. However, in actual detection, the jitter angle collected by the gyroscope and the coal dust grayscale data obtained by the image acquisition unit often have timestamp deviations due to differences in acquisition frequency, and real-time data is easily affected by transient noise. Directly inputting the data into the lookup table will lead to mismatch in the compensation coefficients. Therefore, it is necessary to first implement time alignment and data filtering to ensure that the input data is accurately synchronized, stable, and reliable. The specific implementation method is as follows:

[0049] The grayscale threshold dynamic correction module, based on the real-time input of the cage pitch jitter angle and coal dust grayscale, first ensures that the jitter angle timestamp collected by the gyroscope is strictly aligned with the image acquisition unit via a hardware synchronization signal. This hardware synchronization signal is generated by the system's built-in FPGA (Field-Programmable Gate Array) and uses a 1PPS synchronization pulse signal. This signal is simultaneously connected to the external trigger interfaces of both the gyroscope (used to measure the cage pitch jitter angle) and the industrial camera (image acquisition unit). When the synchronization pulse arrives, the gyroscope immediately records the current jitter angle and adds a timestamp accurate to milliseconds. The industrial camera simultaneously captures an image of the cage flange surface and adds the same millisecond-level timestamp to the image, ensuring that the jitter angle at the same time point is completely consistent with the timestamp of the coal dust grayscale data (extracted from the image). To avoid clock drift during long-term operation, the system calibrates the FPG every hour via NTP (Network Time Protocol) with a high-precision time server in the detection area. Clock A ensures that the timestamp error is always less than 1 millisecond, completely eliminating data misalignment caused by time asynchrony. After timing alignment, a sliding window mean filter is used to suppress transient noise for jitter angle data of multiple consecutive frames. The sliding window mean filter selects multiple consecutive frames of data to form a window, calculates the average value of the data within the window as the filtering result, and can effectively smooth transient spike noise. The window size is set to 5 frames. Combined with the dynamic characteristics of cage jitter, in specific operation, the system caches the most recent 5 frames of jitter angle data in the acquisition order. When a new frame of data arrives, the earliest frame of data is discarded, and the arithmetic mean of the remaining 5 frames of data is calculated. This average value is the filtered jitter angle data. At the same time, the filtering module has a built-in outlier detection function. If the difference between a frame of data and the adjacent frame of data exceeds 0.5 degrees (the preset transient noise judgment threshold), the outlier frame is directly discarded and replaced with the filtering result of the previous frame to avoid extreme noise affecting the filtering effect. The grayscale value of coal dust is then averaged across frames based on the previously calculated spatial weighting result. The spatial weighting result is the weighted local grayscale value of each grid sub-region. The purpose of the inter-frame weighted averaging is to suppress the error caused by the temporary local accumulation of coal dust in a single frame image. In practice, the system caches the spatial weighted grayscale results of the current frame and the previous frame, and calculates the inter-frame weighted average with a weight of 0.6 for the current frame and 0.4 for the previous frame. The current frame has a higher weight because it is closer to the real-time coal dust state than the previous frame. The weights are used to smooth inter-frame fluctuations. For example, if the global average grayscale value after spatial weighting of the current frame is 45 and that of the previous frame is 43, the inter-frame weighted average is 45 × 0.6 + 43 × 0.4 = 27 + 17.2 = 44.2. This value preserves the real-time nature of the current frame while avoiding grayscale jumps caused by noise in a single frame. If the difference between the grayscale values ​​of two consecutive frames exceeds 10 (the preset significant fluctuation threshold), the weight of the previous frame is automatically reduced to 0.2, and the weight of the current frame is increased to 0.8, prioritizing the dominance of real-time data.After time alignment and data filtering are completed, the filtered jitter angle and coal dust grayscale data, together with the initial estimate of the wear depth in the current detection area, are used as the input index of the three-dimensional lookup table. The initial estimate of the wear depth is based on the wear depth result calculated from the previous frame image. If it is the initial detection frame, the preset default value of 0.5mm is used. This value is the common initial wear depth in vertical shaft guideways. Its function is to supplement the depth dimension of the three-dimensional lookup table to ensure that the deviation compensation coefficient of the corresponding working condition can be accurately matched. When inputting, the system passes the three parameters to the lookup table retrieval module in the order of jitter angle → coal dust grayscale → wear depth. First, the subdivision grid corresponding to the jitter angle and coal dust grayscale is located, and then the specific compensation coefficient is determined in combination with the wear depth, providing accurate input for the subsequent generation of dynamic grayscale segmentation threshold.

[0050] After completing data time-series alignment and filtering, and determining the input index of the three-dimensional lookup table, the grayscale threshold dynamic correction module needs to accurately match the optimal deviation compensation coefficient through a two-parameter collaborative indexing strategy. This strategy uses the jitter angle and coal dust grayscale as the core collaborative dimensions to avoid compensation deviations caused by single-parameter retrieval. At the same time, it combines interpolation algorithms to cover uncalibrated working conditions, and finally generates a dynamic grayscale segmentation threshold that adapts to the current coupling state. The specific implementation method is as follows:

[0051] When matching the optimal compensation coefficient in the 3D lookup table, the core implements a two-parameter collaborative indexing strategy. First, the filtered jitter angle and coal dust ash value are used as reference coordinates. These reference coordinates map the two parameters to a two-dimensional retrieval dimension of the 3D lookup table (jitter angle as the X-axis, ranging from 0-10 degrees; coal dust ash value as the Y-axis, ranging from 30-150 degrees), forming spatial coordinate points. These coordinates are directly associated with the calibration point data stored in the lookup table. When retrieving the nearest calibration point data from the 3D lookup table, the Euclidean distance calculation method is used to calculate the spatial distance between the reference coordinates and all calibration point coordinates. The three calibration points with the smallest distances are selected to ensure they can form a two-dimensional triangular unit, providing sufficient reference for subsequent interpolation. For example, the three nearest calibration points corresponding to the reference coordinates (2.36, 44.2) are (2.2...). The system records the deviation compensation coefficients (e.g., 1.12, 1.16, 1.14) corresponding to the three calibration points (42, 42, 43, 45, 42, 43, 45, 42, 43, 43, 45 ...

[0052] The specific process of generating dynamic grayscale segmentation thresholds requires combining a base threshold and proportional modulation. First, a base static threshold is determined. This base static threshold is the baseline grayscale threshold used to distinguish normal cage surfaces from defective areas when there is no cage vibration or coal dust adhesion. It is determined through a calibration experiment under interference-free operating conditions. In a laboratory environment, images of the clean cage flange surface are taken, and a bimodal segmentation technique is used to find the grayscale valley value between the normal surface and the background. 1.1 times this value is taken as the base static threshold. This threshold is pre-stored in the system parameter library as a reference for the dynamic threshold. The real-time compensation coefficient is added to the base static threshold to obtain the initial adjustment threshold, for example, 99 + 14.2 = 113.2. The purpose of this step is to correct the grayscale shift caused by vibration and coal dust through the compensation coefficient, ensuring that the threshold can adapt to the grayscale distribution differences under the current operating conditions. Then, the initial... The threshold value is adjusted by multiplying it by a scaling factor modulated by the proportion of coal dust coverage area. The scaling factor is a correction coefficient dynamically adjusted according to the proportion of coal dust coverage area. It is calibrated through multi-condition experiments. When the proportion of coal dust coverage area is less than 5%, the scaling factor is set to 0.95; when the proportion is 5%-15%, it is set to 1.0; and when the proportion is greater than 15%, it is set to 1.05. For example, if the current proportion of coal dust coverage area is 8%, the scaling factor is 1.0. The final dynamic grayscale segmentation threshold is 113.2 × 1.0 = 113.2. The integer 113 is taken as the output value. This dynamic grayscale segmentation threshold can simultaneously adapt to the grayscale shift caused by cage pitch shaking, the low grayscale shadows formed by coal dust adhesion, and the segmentation requirements brought about by the difference in coal dust coverage area. Finally, it is output to the defect diagnosis decision unit, laying the foundation for the accurate segmentation of the subsequent cage flange surface image.

[0053] After generating a dynamic grayscale segmentation threshold that adapts to the coupling state of the current cage shaking and coal dust shadow, the defect diagnosis decision unit needs to use a partitioned adaptive segmentation strategy to segment the cage flange surface image. Since there are local differences in the distribution of coal dust on the cage flange surface, and the image edges are easily affected by distortion caused by shaking, if a single global threshold segmentation is used, edge missed detection or center false detection is likely to occur. Therefore, the threshold needs to be adjusted independently according to the grid sub-region. The specific implementation method is as follows:

[0054] When segmenting the surface image of the tank duct flange using dynamic grayscale segmentation thresholding, the dynamic grayscale segmentation threshold is first independently assigned to each sub-region based on the divided grid sub-regions. The allocation logic needs to be adjusted in conjunction with the generated global dynamic threshold, the local coal dust coverage rate of the sub-region, and the spatial position weight. The global dynamic threshold is the final threshold calculated by the compensation coefficient and the scaling factor, which serves as the benchmark for the thresholds of all sub-regions. The local coal dust coverage rate of the sub-region is the ratio of the number of coal dust mask pixels in the sub-region to the total number of pixels in the sub-region. The higher the coverage rate, the more significant the influence of coal dust in the region, and the threshold needs to be lowered to avoid misjudging dense coal dust as normal surface. The spatial position weight is set according to the position of the sub-region in the image, with the weight of the image center region set as follows. The threshold is set to 0.95, and the edge region weight is set to 0.85. The smaller the weight, the larger the threshold adjustment range, ensuring that the edge region threshold is more adapted to the distorted grayscale distribution. For example, with a global dynamic threshold of 113, a central sub-region with 8% coverage and a weight of 0.95 has an adjusted threshold of 113 - (8% ÷ 5% × 2) × 0.95 ≈ 113 - 3.04 = 109.96; an edge sub-region with 15% coverage and a weight of 0.85 has an adjusted threshold of 113 - (15% ÷ 5% × 2) × 0.85 ≈ 113 - 5.1 = 107.9. For sub-regions located at the image edge, an additional threshold offset positively correlated with the pitch jitter angle needs to be introduced, because the larger the jitter angle, the greater the geometric distortion and grayscale distribution at the image edge. The more severe the degree of shift, the greater the shift needs to be as the jitter angle increases. Experiments have shown that for every 1 degree increase in jitter angle, the shift increases by 2 grayscale units. Furthermore, the shift only affects the downward adjustment direction of the threshold; that is, the final threshold for the edge sub-region = adjusted threshold - shift. This avoids excessively high thresholds leading to missed detections. For example, in the above edge sub-region, the adjusted threshold is 107.9, and with an added shift of 4.72, the final threshold is 103.18, ensuring accurate segmentation of coal dust in distorted edge regions. After segmentation, connected component labeling verification ensures the validity of the segmentation results. Connected component labeling verification is performed on each connected region (a set of adjacent pixels with the same pixel value) in the segmented binary image (where pixels in the coal dust region are 1 and normal regions are 0). The system labels and analyzes features such as area and aspect ratio. If the area of ​​a connected region is less than 5 pixels (indicating it to noise, not real coal dust), or the aspect ratio is greater than 10 (indicating it to a pseudo-region caused by image edge distortion), it is labeled as an abnormal segmentation region. A secondary segmentation based on grayscale texture features is initiated for the abnormal segmentation region. The grayscale texture features of the abnormal region and its surrounding 10-pixel range are extracted and compared with the texture feature library of normal coal dust regions (calibrated through a large number of real coal dust images). If the texture feature matching degree is greater than or equal to 80%, the region is relabeled as a coal dust region. If the matching degree is less than 80%, it is judged as noise and removed. Finally, the accurate coal dust region segmentation result is output, laying the foundation for subsequent removal of residual noise.

[0055] After completing the adaptive segmentation of the tank guide flange surface image, residual noise from coal dust shadows may still exist in the segmentation results. This noise can interfere with subsequent wear depth detection, so morphological filtering is required to remove it. The core of this method is a multi-scale adaptive selection mechanism for structural elements, which dynamically matches structural elements according to the morphological characteristics of the noise, ensuring that the real coal dust area and tank guide defect features are not destroyed while removing noise. The specific implementation method is as follows:

[0056] When removing residual noise from coal dust shadows using morphological filtering, the shape and size of the structuring element are dynamically selected based on the statistical characteristics of the noise connected components in the obtained segmentation results. The statistical characteristics of the noise connected components are obtained through the connected component analysis module: All connected components marked as potential noise are traversed in the segmented binary image. For regions with an area less than 10 pixels or abnormal grayscale values, the area (to determine if it is a small discrete point) and aspect ratio (to determine if it is a strip) of each connected component are calculated. If the area of ​​the connected component is less than or equal to 5 pixels and the aspect ratio is 1 (close to a circle), it is determined to be small discrete point noise. If the area of ​​the connected component is 5-15 pixels and the aspect ratio is greater than or equal to 5 (appearing as a strip), it is determined to be strip noise (mostly caused by the superposition of surface texture and coal dust in the tank passage). (Forming), the corresponding structural element morphology is selected according to different types of noise: When the residual noise is mainly composed of small-area discrete points, a circular structural element is used for morphological opening operation. The morphological opening operation is a combination of erosion and expansion. The circular structural element can act evenly around the discrete points. The erosion process can remove discrete points smaller than the structural element, and the expansion process can restore the original contour of the real coal dust area, avoiding area shrinkage caused by excessive removal. The smooth edge characteristics of the circular structural element can also reduce the damage to the boundary of the coal dust area. When the residual noise is distributed in strips, a long strip structural element consistent with the direction of the shaft texture is used to perform morphological closing operation. The direction of the shaft texture is the length direction of the shaft flange (perpendicular to the vertical shaft height direction, as shown in the figure). (Similar to edge detection pre-calibration), the length direction of the elongated structural element is consistent with the texture direction. The closing operation (dilation followed by erosion) can fill the gaps in the strip noise while preserving the real coal dust area consistent with the texture direction, avoiding the destruction of the cage texture features by the lateral structural elements. The size of the structural element is jointly controlled by the proportion of coal dust coverage area and the pitch and shaking angle of the cage. The size adjustment rules are established through experimental calibration. When the proportion of coal dust coverage area is less than 5% (low total noise), the size of the circular structural element is set to 3×3 pixels and the size of the elongated structural element is set to 3×7 pixels. When the proportion is 5%-15% (medium total noise), the size is adjusted to 5×5 pixels and 5×11 pixels respectively. When the proportion is greater than 15% (high total noise), the size is adjusted accordingly. Do not adjust to 7×7 pixels or 7×15 pixels; at the same time, for every 1 degree increase in the pitch jitter angle of the cage, the size of the structural element increases by 1 pixel unit on the basis of the above. Because when the jitter angle increases, the area of ​​noise region caused by image edge distortion is larger, and a larger structural element is required to completely remove it. For example, when the coal dust coverage area accounts for 8% and the jitter angle is 2.36 degrees, the size of the circular structural element = 5×5 + 2×1 = 7×7 pixels (rounded to the nearest integer), and the size of the long strip structural element = 5×11 + 2×1 = 7×13 pixels. Ensure that the size of the structural element is adapted to the noise scale. After filtering, output the pure coal dust region segmentation result with residual noise removed, providing an interference-free image basis for subsequent sub-pixel edge detection and wear depth calculation.

[0057] After removing residual noise from coal dust shadows through multi-scale structural element morphological filtering, the depth of wear grooves on the flange surface of the tank passage needs to be accurately calculated. Since the depth of wear grooves is usually in the sub-millimeter range, conventional pixel-level edge detection is insufficient. Therefore, a sub-pixel edge detection algorithm is adopted, and false edges are eliminated through dual constraints of gradient magnitude and curvature to ensure accurate depth calculation. Finally, the defect diagnosis decision unit outputs control commands based on historical data. The specific implementation method is as follows:

[0058] When calculating the depth of the wear groove using the sub-pixel edge detection algorithm, the first step is to implement edge localization with dual constraints of gradient magnitude and curvature. The first step is to extract the grayscale profile curve along the normal direction of the tank passage flange in the denoised image. The normal direction of the tank passage flange is the direction perpendicular to the length of the tank passage flange. The tank passage flange extends along the height of the shaft, and the normal direction is horizontal. This direction can completely pass through the two sides and bottom structure of the wear groove, accurately reflecting the depth change. Before extraction, the flange length direction needs to be determined by the image preprocessing module. The Hough line detection algorithm is used to identify the straight line features of the flange edge, and then the normal direction is calculated. When extracting the grayscale profile curve, extraction lines are set at 5-pixel intervals along the flange length to ensure coverage of the entire detection area and avoid missing grooves. Each extraction line covers a 20-pixel range along the normal direction. The grayscale value of each pixel on the extraction line is recorded to form a grayscale profile curve, with the horizontal axis representing the pixel position and the vertical axis representing the grayscale value. The part of the curve where the grayscale value drops sharply and then rises again corresponds to the wear groove. After extracting the grayscale profile curve, cubic spline interpolation is used to achieve sub-pixel edge localization. Cubic spline interpolation is a method that uses known integer pixel grayscale values... This method, which uses mathematical methods to fit a smooth and continuous grayscale function curve, can calculate the grayscale value of non-integer pixels, i.e., sub-pixel positions, breaking through the accuracy limitations of pixel-level positioning. In practice, the integer pixel grayscale values ​​of each grayscale profile curve are input into the interpolation algorithm to generate a continuous grayscale function curve. Then, the inflection points of the curve are found; the steep changes in the grayscale curve at the two edges of the wear groove correspond to these inflection points. By calculating the first derivative of the curve (reflecting the rate of grayscale change), the position with the largest absolute value of the derivative is found; this position represents the sub-pixel level. Edge points, such as finding the left and right edge points at pixels 20.6 and 22.8, are used to achieve precise edge positioning. After obtaining sub-pixel edge points, a double constraint is used to filter false edges caused by residual coal dust shadows. The first constraint is the gradient magnitude threshold: the gradient magnitude is the absolute value of the first derivative of the grayscale profile curve at the edge point, reflecting the steepness of the grayscale change. The edge of a real wear groove has a large gradient magnitude due to the abrupt change in depth (e.g., when the grayscale decreases from 120 to 75, the gradient magnitude can reach 45), while false edges caused by residual coal dust are... Edges (e.g., slight fluctuations in local grayscale) have small gradient amplitudes (usually less than 20). Experimentally, a gradient amplitude threshold of 30 was set, retaining only edge points with gradient amplitudes exceeding 30. The second constraint is a local curvature threshold. Local curvature refers to the degree of bending of the grayscale curve near edge points. The curvature of true edge points has a fixed range (e.g., 0.1-0.5), while the curvature of pseudo-edge points is either too large (greater than 1.0) or too small (less than 0.05) due to irregular grayscale changes. Similarly, experimentally, a local curvature threshold range of 0.08-0.05 was set.6. Only edge points with curvature within this range are retained. After double-constraint screening, the remaining edge points are all edges of the actual wear grooves, effectively avoiding false edges interfering with depth calculation. The wear groove depth calculation needs to be based on the screened edge points and curve features. First, the bottom position of the groove is determined. In the grayscale profile curve between the left and right edge points, the point with the smallest grayscale value is found, which is the bottom of the groove, corresponding to the maximum depth. Then, through the mapping relationship between grayscale value and actual depth, the grayscale difference between the bottom of the groove and the normal surfaces on both sides (grayscale values ​​at the edge points) is calculated, thus obtaining the initial wear depth. Subsequently, the initial depth value is multiplied by the deviation compensation coefficient matched by the three-dimensional lookup table to complete the depth calibration. This compensation coefficient can eliminate the influence of cage pitch shaking and coal dust grayscale on imaging, ensuring that the calibrated depth value is close to the true value. The calibrated wear groove depth value is then input into the defect diagnosis decision unit 4. After receiving the calibrated depth value, the defect diagnosis decision unit 4 performs diagnostic analysis in combination with the real-time depth value and historical wear trend. The wear trend is the wear depth data of the detection point of the tank guide over the past 3-5 times. For example, if the depths of the past 3 times are 1.1mm, 1.2mm, and 1.37mm respectively, it can be determined that the wear is increasing. The unit pre-stores warning thresholds based on the safety standards of the tank guide. When the depth value exceeds a certain level warning threshold for 3 consecutive times, the corresponding graded control command is output. A mild warning outputs an adjustment of the detection frequency from once a week to once every 3 days. A moderate warning outputs a maintenance plan to be formulated within one week and the worn section of the tank guide is replaced. A severe warning outputs an immediate shutdown, prohibiting the operation of the cage, and completing emergency repairs within 24 hours. At the same time, the unit calculates the actual coordinates of the wear location through the mapping relationship between image coordinates and the actual location of the tank guide. For example, the coordinates of the groove edge point in the image are (1200 pixels, 800 pixels), which corresponds to the actual location of the tank guide at a length of 1200m and a height of 800mm in the vertical shaft tank guide. The coordinate information and graded commands are displayed synchronously on the monitoring terminal, which facilitates the staff to quickly locate and handle wear problems.

[0059] In this invention, the acquisition unit simultaneously acquires images of the cage flange surface and the cage pitch and jitter angle. The coal dust shadow quantification unit uses bimodal adaptive threshold segmentation combined with a region growing algorithm to extract the coal dust region, and introduces a spatial weighting factor to calculate the average gray value and coverage area ratio. The dynamic coupling calibration unit relies on a three-dimensional lookup table calibrated under multiple working conditions and nonlinear interpolation to match the optimal deviation compensation coefficient of the jitter and coal dust combination, and generates a dynamic gray-scale segmentation threshold. The defect diagnosis decision unit accurately calculates the wear depth through partitioned adaptive segmentation, multi-scale morphological filtering for noise reduction, and sub-pixel edge detection with gradient amplitude and curvature dual constraints. Based on the real-time depth and historical trends, it outputs hierarchical control commands to eliminate coupling interference, improve the accuracy of cage wear detection, and provide a reliable basis for safe operation and maintenance.

[0060] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. An industrial intelligent vision detection and diagnosis system for shaft guide, characterized in that, The application relates to a coal mine shaft lining wear defect diagnosis system based on dynamic coupling correction, which comprises the following parts: A shaft lining image and posture acquisition unit (1) acquires a shaft lining flange surface image and real-time measures a cage pitching angle; A coal dust shadow quantification unit (2) extracts a coal dust adhesion area from the shaft lining flange surface image through a gray scale histogram segmentation algorithm, calculates an average gray scale value of the coal dust adhesion area and a coverage area proportion; A dynamic coupling correction unit (3) comprises a pitching and shadow coupling relationship mapping module and a gray scale threshold dynamic correction module, wherein the pitching and shadow coupling relationship mapping module records a deviation compensation coefficient of a wear depth measurement value relative to a real value under different cage pitching angles and coal dust gray scale combinations through a three-dimensional lookup table established through a calibration experiment, and the gray scale threshold dynamic correction module generates a dynamic gray scale segmentation threshold by matching an optimal compensation coefficient in the three-dimensional lookup table according to real-time input of the cage pitching angle and the coal dust gray scale. When the three-dimensional lookup table of the pitching and shadow coupling relationship mapping module is established through a calibration experiment, the three-dimensional lookup table is realized through a multi-working condition calibration experiment platform, and specifically comprises the following steps: The calibration platform comprises a simulated cage support for controlling a pitching angle, a shaft lining test block with a surface covered with a gray scale coating and a standard calibration module with a pre-set depth wear groove, under each group of calibration working conditions, the cage pitching angle, the coal dust simulated gray scale value and the wear depth imaging measurement value are synchronously acquired, the deviation compensation coefficient under the working condition is calculated by comparing the wear depth imaging measurement value with the real depth of the standard calibration module, and finally a three-dimensional lookup table with the pitching angle and the coal dust gray scale value as independent variables and the deviation compensation coefficient as a dependent variable is formed; When the gray scale threshold dynamic correction module is used according to the real-time input of the cage pitching angle and the coal dust gray scale, time sequence alignment and data filtering are implemented, and specifically comprises the following steps: Through a hardware synchronous signal, the time stamp of the pitching angle collected by a gyroscope is strictly aligned with the image acquisition unit, sliding window mean filtering is adopted for the continuous multiple frames of the pitching angle data to suppress transient noise, and the coal dust gray scale value is based on the obtained spatial weighted result to perform interframe weighted average, the filtered pitching angle and coal dust gray scale data, together with the initial estimated value of the wear depth of the current detection area, are used as the input index of the three-dimensional lookup table; When the optimal compensation coefficient is matched in the three-dimensional lookup table, a double-parameter collaborative indexing strategy is executed, and specifically comprises the following steps: The filtered pitching angle and coal dust gray scale value are used as the reference coordinates to search the nearest calibration point data in the three-dimensional lookup table, when the input point is located in a triangular unit formed by multiple calibration data, the real-time compensation coefficient is calculated by calling the barycentric coordinate interpolation, and the specific process of generating the dynamic gray scale segmentation threshold is as follows: The real-time compensation coefficient is added to the basic static threshold, and then multiplied by a proportional factor modulated by the coal dust coverage area proportion, and finally the dynamic gray scale segmentation threshold adaptive to the current coupling state of the pitching and shadow is output. A defect diagnosis decision unit (4) is used to segment the shaft lining flange surface image by using the dynamic gray scale segmentation threshold, remove the coal dust shadow residual noise by combining morphological filtering, calculate the wear groove depth by using a sub-pixel edge detection algorithm, and output an instruction according to the wear groove depth.

2. The industrial intelligent vision detection and diagnosis system for shaft guide according to claim 1, characterized in that: When the coal dust shadow quantification unit (2) extracts the coal dust adhesion area from the flue surface image by the gray histogram segmentation algorithm, a double-peak adaptive threshold segmentation technology is adopted, specifically including: The flue surface image is subjected to Gaussian filtering to eliminate high-frequency noise, a smooth gray histogram is generated, the main gray peak representing the normal flue surface and the secondary gray peak representing the coal dust shadow are identified based on the peak distribution characteristics of the smooth gray histogram, and the valley gray value of the main gray peak and the secondary gray peak is taken as the initial segmentation threshold. Then, the initial segmentation result is optimized by combining the region growing algorithm, starting from the pixel points below the initial segmentation threshold, dynamically expanding the connected domain according to the gray gradient change amplitude of the eight-neighbor pixels, until the gradient change exceeds the preset tolerance, and finally outputting the coal dust adhesion area.

3. The industrial intelligent vision detection and diagnosis system for shaft guide according to claim 2, characterized in that: After obtaining the coal dust adhesion area, the average gray value and the coverage area ratio of the coal dust adhesion area are calculated, and a spatial distribution weight factor is introduced, specifically including: The flue surface image is divided into a plurality of grid sub-regions, a weight coefficient is assigned according to the proportion of coal dust mask pixels in each sub-region, the average gray value is calculated using a weighted average formula, the weight coefficient is positively correlated with the sub-region coal dust coverage, the coverage area ratio is determined by the ratio of the total number of coal dust mask pixels to the total number of effective pixels in the detection region, and is associated with the real-time value of the cage pitch oscillation angle. When the cage pitch oscillation angle increases, the coverage area calculation of the edge region is enabled with a distortion correction coefficient.

4. The industrial intelligent vision detection and diagnosis system for shaft guide according to claim 1, characterized in that: The three-dimensional lookup table records the deviation compensation coefficient of the wear depth measurement value relative to the true value under different cage pitch oscillation angles and coal dust gray combinations. A nonlinear interpolation mapping mechanism is adopted, specifically including: For the oscillation angle and coal dust gray combination points not covered by the calibration experiment, a three-dimensional Delaunay triangulation grid is constructed based on the nearest neighbor calibration data points, the compensation coefficient is calculated by barycentric coordinate interpolation in the triangular element, and an association rule between the deviation compensation coefficient and the wear depth is established. The compensation coefficient changes in sections with the increase of the wear depth. When the wear depth is less than the critical value, the compensation coefficient has a linear relationship with the depth. When the wear depth exceeds the critical value, a quadratic polynomial compensation model is enabled.

5. The industrial intelligent vision detection and diagnosis system for shaft guide according to claim 1, characterized in that: When the dynamic gray segmentation threshold is used to segment the flue surface image, a partition adaptive segmentation strategy is adopted, specifically including: According to the divided grid sub-regions, a dynamic gray segmentation threshold is independently assigned to each sub-region. The dynamic gray segmentation threshold is adjusted by the generated global dynamic threshold in combination with the local coal dust coverage and the spatial position weight of the sub-region. A threshold offset amount positively correlated with the oscillation angle is additionally introduced for the sub-regions located at the edge of the image. After segmentation, the connected domain marking verification is performed, and the secondary segmentation based on the gray texture features is started for the abnormal segmentation area.

6. The industrial intelligent vision detection and diagnosis system for shaft guide according to claim 5, characterized in that: The morphological filtering is combined to remove the residual noise of the coal dust shadow, and a multi-scale structure element adaptive selection mechanism is adopted, specifically including: According to the statistical characteristics of the noise connected domain in the obtained segmentation result, the shape and size of the structural element are dynamically selected, when the residual noise is mainly in the form of small area discrete points, the opening operation is performed by using the circular structural element, when the residual noise is distributed in the form of strips, the closing operation is performed by using the long strip structural element consistent with the texture direction of the flume, the size of the structural element is controlled by the coal dust coverage area ratio and the shaking angle, and the size of the structural element increases correspondingly when the shaking angle increases.

7. The industrial intelligent vision detection and diagnosis system for shaft guide according to claim 6, characterized in that: When the wear groove depth is calculated through the sub-pixel edge detection algorithm, the gradient amplitude and curvature double-constraint edge positioning are implemented, and specifically include: In the denoised image, the gray profile curve is extracted along the normal direction of the flume wing, the sub-pixel edge positioning is realized by using cubic spline interpolation, the edge point screening needs to satisfy the gradient amplitude threshold value and the local curvature threshold value at the same time, the pseudo-edge generated in the residual area of the coal dust shadow is excluded, the wear groove depth calculation value is calibrated through a deviation compensation coefficient, and then input into a defect diagnosis decision unit (4), the defect diagnosis decision unit (4) outputs a grading control instruction and marks the wear position coordinates according to the real-time depth value and the historical wear trend when the depth value exceeds the early warning threshold value for continuous multiple times.

Images