A coal mine scraper inner wall defect visual detection method and system

Abstract

The present application provides a kind of coal mine scraper inner wall defect visual detection method and system, the present application relates to scraper defect detection technical field, the method includes obtaining the visual imaging of coal mine scraper inner wall;Determine a plurality of first defect suspected areas and the inner wall defect suspicious degree distribution map of each first defect suspected area based on the visual imaging of the coal mine scraper inner wall;Determine a plurality of basic detection points of each first defect suspected area based on the visual imaging of a plurality of first defect suspected areas and the inner wall defect suspicious degree distribution map of each first defect suspected area;Determine the core defect area of scraper inner wall based on the inner wall state perception data of a plurality of basic detection points of second defect suspected area, the inner wall state perception data of a plurality of advanced detection points of second defect suspected area, this method can quickly and accurately determine the core defect area of coal mine scraper inner wall.

Description

Technical Field

[0001] This invention relates to the field of scraper conveyor defect detection, specifically to a visual inspection method and system for defects on the inner wall of a coal mine scraper conveyor. Background Technology

[0002] In coal mining operations, scraper conveyors are critical transportation equipment, and the integrity of their internal structure directly affects the safety and efficiency of the entire production system. During long-term operation, the inner wall of the scraper conveyor is susceptible to impact from coal gangue, material abrasion, and corrosion, easily leading to cracks and wear depressions. Traditional methods for detecting defects in the inner wall of coal mine scraper conveyors mainly rely on manual inspection and offline disassembly. Manual inspection requires personnel to enter the equipment with tools or check the inner wall through observation ports while the scraper conveyor is stopped. This method is limited by underground lighting conditions, inspection angle, and personnel experience, making it difficult to detect hidden or minor early defects. Furthermore, it is inefficient, labor-intensive, and poses personnel safety risks. Offline disassembly requires disassembling key components of the scraper conveyor and conducting comprehensive inspections using specialized equipment. While this improves inspection accuracy, it leads to prolonged equipment downtime, severely impacting continuous coal mine production. Moreover, the disassembly and reassembly process can cause secondary damage to the equipment, increasing maintenance costs. While existing automated inspection technologies incorporate image acquisition and sensor monitoring, they are susceptible to noise contamination and blurred features due to complex environments such as dust obstruction, uneven lighting, and electromagnetic interference underground. Sensor data also suffers from fluctuations in accuracy due to vibration and temperature interference, making it difficult to accurately identify and locate defects. The lack of specific adaptability in existing technologies leads to low detection coverage and a high false positive rate.

[0003] Therefore, how to quickly and accurately determine the core defect area of ​​the inner wall of a coal mine scraper conveyor is an urgent problem to be solved. Summary of the Invention

[0004] The main technical problem solved by this invention is how to quickly and accurately determine the core defect area of ​​the inner wall of a coal mine scraper conveyor.

[0005] According to a first aspect, the present invention provides a visual inspection method for defects on the inner wall of a coal mine scraper conveyor, comprising: acquiring a visual image of the inner wall of the coal mine scraper conveyor; determining multiple first suspected defect areas and a distribution map of the degree of doubt for each first suspected defect area based on the visual image of the inner wall of the coal mine scraper conveyor; determining multiple basic detection points for each first suspected defect area based on the visual image of the multiple first suspected defect areas and the distribution map of the degree of doubt for each first suspected defect area, and acquiring the inner wall state perception data of the multiple basic detection points for each first suspected defect area; constructing an inner wall detection map, wherein the inner wall detection map includes multiple basic detection point nodes and edges between multiple nodes, the node features of each basic detection point node are the position of the basic detection point, the inner wall state perception data of the basic detection point, and the edges between the detection nodes are the direction and distance between the basic detection point nodes, and the phase of the inner wall state perception data. Similarity; Based on graph neural network, the inner wall detection map is processed to determine multiple advanced detection points for each first suspected defect area, and the inner wall state perception data of multiple advanced detection points for each first suspected defect area is obtained; Based on the inner wall state perception data of multiple basic detection points for each first suspected defect area and the inner wall state perception data of multiple advanced detection points for each first suspected defect area, K clusters are obtained; Based on the K clusters, the inner wall defect signal feedback doubt distribution map of each first suspected defect area is determined; Based on the inner wall defect doubt distribution map of each first suspected defect area and the inner wall defect signal feedback doubt distribution map of each first suspected defect area, the second suspected defect area is determined; Based on the inner wall state perception data of multiple basic detection points for the second suspected defect area and the inner wall state perception data of multiple advanced detection points for the second suspected defect area, the core defect area of ​​the scraper conveyor inner wall is determined.

[0006] In one possible implementation, determining the core defect area of ​​the scraper conveyor's inner wall based on the inner wall state perception data of multiple basic detection points in the second defect suspected area and the inner wall state perception data of multiple advanced detection points in the second defect suspected area includes: using a defect verification determination model to determine multiple defect verification points in each second defect suspected area based on the inner wall state perception data of multiple basic detection points in the second defect suspected area and the inner wall state perception data of multiple advanced detection points in the second defect suspected area; and determining the core defect area of ​​the scraper conveyor's inner wall based on the precise detection data of inner wall defects from the multiple defect verification points in each second defect suspected area.

[0007] In one possible implementation, the inner wall state sensing data includes ultrasonic vibration signal data of the inner wall surface and inner wall material density feedback data.

[0008] In one possible implementation, the input to the graph neural network is the inner wall detection map, and the output of the graph neural network is multiple advanced detection points for each first suspected defect area.

[0009] According to a second aspect, the present invention provides a visual inspection system for defects on the inner wall of a coal mine scraper conveyor, comprising: an imaging acquisition module for acquiring a visual image of the inner wall of the coal mine scraper conveyor; a region positioning module for determining multiple first suspected defect areas and a distribution map of the degree of doubt for each first suspected defect area based on the visual image of the inner wall of the coal mine scraper conveyor; a basic detection module for determining multiple basic detection points for each first suspected defect area based on the visual image of the multiple first suspected defect areas and the distribution map of the degree of doubt for each first suspected defect area, and acquiring the inner wall state perception data of the multiple basic detection points for each first suspected defect area; a construction module for constructing an inner wall detection map, wherein the inner wall detection map includes multiple basic detection point nodes and edges between multiple nodes, the node features of each basic detection point node are the position of the basic detection point, the inner wall state perception data of the basic detection point, and the edges between detection nodes are the direction and distance between the basic detection point nodes, and the similarity of the inner wall state perception data; and an advanced detection module. The system comprises the following modules: a module for processing the inner wall detection map based on a graph neural network to determine multiple advanced detection points for each suspected first defect area, and acquiring inner wall state perception data for each advanced detection point in each suspected first defect area; a clustering analysis module for clustering the inner wall state perception data of the multiple basic detection points and the multiple advanced detection points in each suspected first defect area to obtain K clusters; a signal analysis module for determining the inner wall defect signal feedback doubt distribution map for each suspected first defect area based on the K clusters; a suspected area screening module for determining a second suspected defect area based on the inner wall defect doubt distribution map and the inner wall defect signal feedback doubt distribution map for each suspected first defect area; and a core defect determination module for determining the core defect area of ​​the scraper conveyor's inner wall based on the inner wall state perception data of the multiple basic detection points and the multiple advanced detection points in the second suspected second defect area.

[0010] In one possible implementation, the core defect determination module is further configured to: use a defect verification determination model to determine multiple defect verification points for each second defect suspected area based on the inner wall state perception data of multiple basic detection points in the second defect suspected area and the inner wall state perception data of multiple advanced detection points in the second defect suspected area; and determine the core defect area of ​​the scraper conveyor inner wall based on the precise detection data of the inner wall defects of the multiple defect verification points in each second defect suspected area.

[0011] In one possible implementation, the inner wall state sensing data includes ultrasonic vibration signal data of the inner wall surface and inner wall material density feedback data.

[0012] In one possible implementation, the input to the graph neural network is the inner wall detection map, and the output of the graph neural network is multiple advanced detection points for each first suspected defect area.

[0013] According to a third aspect, embodiments of the present invention provide an electronic device, including: a processor; a memory; and a computer program; wherein the computer program is stored in the memory and configured to be executed by the processor to implement the method as described above, the method including: acquiring a visual image of the inner wall of a coal mine scraper conveyor; determining a plurality of first suspected defect areas and an inner wall defect suspicion distribution map for each first suspected defect area based on the visual image of the inner wall of the coal mine scraper conveyor; determining a plurality of basic detection points for each first suspected defect area based on the visual image of the plurality of first suspected defect areas and the inner wall defect suspicion distribution map for each first suspected defect area, and acquiring inner wall state perception data of the plurality of basic detection points for each first suspected defect area; constructing an inner wall detection map, the inner wall detection map including a plurality of basic detection point nodes and edges between the nodes, wherein the node features of each basic detection point node are the position of the basic detection point, the inner wall state perception data of the basic detection point, and the edges between the detection nodes. The method involves determining the direction and distance between basic detection point nodes and the similarity of inner wall state perception data. A graph neural network is used to process the inner wall detection map to determine multiple advanced detection points for each first suspected defect area, and inner wall state perception data for these advanced detection points is obtained. K clusters are then formed based on the inner wall state perception data of the basic detection points and the advanced detection points for each first suspected defect area. A distribution map of the doubtfulness of inner wall defect signal feedback for each first suspected defect area is determined based on these K clusters. A second suspected defect area is determined based on the distribution map of doubtfulness of inner wall defect signal feedback for each first suspected defect area. Finally, the core defect area of ​​the scraper conveyor's inner wall is determined based on the inner wall state perception data of the basic detection points and the advanced detection points for the second suspected defect area.

[0014] According to the fourth aspect, this embodiment provides a computer-readable storage medium storing a computer program thereon. When executed by a processor, the program implements the aforementioned visual detection method for inner wall defects of a coal mine scraper conveyor. The method includes: acquiring a visual image of the inner wall of the coal mine scraper conveyor; determining multiple first suspected defect areas and an inner wall defect suspicion distribution map for each first suspected defect area based on the visual image of the inner wall of the coal mine scraper conveyor; determining multiple basic detection points for each first suspected defect area based on the visual image of the multiple first suspected defect areas and the inner wall defect suspicion distribution map for each first suspected defect area, and acquiring inner wall state perception data for the multiple basic detection points in each first suspected defect area; constructing an inner wall detection map, the inner wall detection map including multiple basic detection point nodes and edges between multiple nodes, wherein the node features of each basic detection point node are the location of the basic detection point, the inner wall state perception data of the basic detection point, and the edges between the detection nodes are the basic detection points. The direction and distance between nodes, and the similarity of the inner wall state perception data; the inner wall detection map is processed based on a graph neural network to determine multiple advanced detection points for each first suspected defect area, and the inner wall state perception data of the multiple advanced detection points for each first suspected defect area is obtained; K clusters are obtained based on the inner wall state perception data of the multiple basic detection points and the multiple advanced detection points for each first suspected defect area; the inner wall defect signal feedback doubt distribution map of each first suspected defect area is determined based on the K clusters; the inner wall defect doubt distribution map and the inner wall defect signal feedback doubt distribution map of each first suspected defect area are determined to identify the second suspected defect area; the inner wall state perception data of the multiple basic detection points and the multiple advanced detection points for the second suspected defect area are determined to identify the core defect area of ​​the scraper conveyor inner wall.

[0015] This invention provides a visual inspection method and system for the inner wall defects of a coal mine scraper conveyor. The method includes: acquiring a visual image of the inner wall of the coal mine scraper conveyor; determining multiple first suspected defect areas and a defect suspicion distribution map for each first suspected defect area based on the visual image of the inner wall; determining multiple basic detection points for each first suspected defect area based on the visual image of the multiple first suspected defect areas and the defect suspicion distribution map for each first suspected defect area, and acquiring inner wall state perception data for the multiple basic detection points in each first suspected defect area; constructing an inner wall detection map, wherein the inner wall detection map includes multiple basic detection point nodes and edges between multiple nodes, the node features of each basic detection point node are the position of the basic detection point, the inner wall state perception data of the basic detection point, and the edges between detection nodes are the direction and distance between the basic detection point nodes, and the similarity of the inner wall state perception data; and performing visual inspection of the inner wall defects using a graph neural network. The wall detection map is processed to determine multiple advanced detection points for each first suspected defect area, and the inner wall state perception data of the multiple advanced detection points for each first suspected defect area is obtained. Based on the inner wall state perception data of the multiple basic detection points of each first suspected defect area and the inner wall state perception data of the multiple advanced detection points of each first suspected defect area, K clusters are obtained. Based on the K clusters, the inner wall defect signal feedback doubt distribution map of each first suspected defect area is determined. Based on the inner wall defect doubt distribution map of each first suspected defect area and the inner wall defect signal feedback doubt distribution map of each first suspected defect area, a second suspected defect area is determined. Based on the inner wall state perception data of the multiple basic detection points of the second suspected defect area and the inner wall state perception data of the multiple advanced detection points of the second suspected defect area, the core defect area of ​​the scraper conveyor inner wall is determined. This method can quickly and accurately determine the core defect area of ​​the inner wall of the coal mine scraper conveyor. Attached Figure Description

[0016] Figure 1 A flowchart illustrating a visual inspection method for defects on the inner wall of a coal mine scraper conveyor, provided in an embodiment of the present invention.

[0017] Figure 2 A schematic diagram of the inner wall of a coal mine scraper conveyor provided in an embodiment of the present invention;

[0018] Figure 3 A schematic diagram of an industrial high-definition camera provided for an embodiment of the present invention;

[0019] Figure 4 A flowchart illustrating the process of determining the core defect region of the inner wall of a scraper conveyor, as provided in this embodiment of the invention.

[0020] Figure 5 A schematic diagram of a visual inspection system for defects on the inner wall of a coal mine scraper conveyor, provided in an embodiment of the present invention; Detailed Implementation

[0021] The present invention will now be described in further detail with reference to specific embodiments and accompanying drawings. Similar elements in different embodiments are referred to by associated similar element reference numerals. In the following embodiments, many details are described to facilitate a better understanding of the invention. However, those skilled in the art will readily recognize that some features may be omitted in different situations, or may be replaced by other elements, materials, or methods. In some cases, certain operations related to the present invention are not shown or described in the specification. This is to avoid obscuring the core parts of the invention with excessive description. For those skilled in the art, detailed description of these related operations is not necessary; they can fully understand the related operations based on the description in the specification and general technical knowledge in the art.

[0022] In this embodiment of the invention, the following are provided: Figure 1 The method shown is a visual inspection method for defects on the inner wall of a coal mine scraper conveyor, comprising steps S1 to S9:

[0023] Step S1: Obtain a visual image of the inner wall of the coal mine scraper conveyor.

[0024] The visual image of the inner wall of a coal mine scraper conveyor is an image of the inner wall of the scraper conveyor captured by a high-definition industrial camera installed around the scraper conveyor. Figure 2 This is a schematic diagram of the inner wall of a coal mine scraper conveyor provided in an embodiment of the present invention. Figure 3 This is a schematic diagram of an industrial high-definition camera provided in an embodiment of the present invention.

[0025] The visual imaging of the inner wall of a coal mine scraper conveyor can present the visual characteristics of the inner wall surface, including color, texture, and whether there are obvious depressions or protrusions.

[0026] The visual image of the inner wall of a coal mine scraper conveyor can be a static image taken when the scraper conveyor is stopped, or a dynamic image sequence taken when the scraper conveyor is running at low load. For example, images obtained by a high-definition industrial camera installed at the head of the scraper conveyor taking pictures of the inner wall from multiple angles during the scraper conveyor's shutdown for maintenance.

[0027] Step S2: Based on the visual imaging of the inner wall of the coal mine scraper conveyor, determine multiple first defect suspected areas and the inner wall defect suspicion distribution map of each first defect suspected area.

[0028] In some embodiments, a defect feature analysis model can be used to determine multiple first suspected defect areas and a distribution map of the degree of doubt for defects on the inner wall of each first suspected defect area. The defect feature analysis model is a convolutional neural network model. The input to the defect feature analysis model is a visual image of the inner wall of the coal mine scraper conveyor, and the output of the defect feature analysis model is multiple first suspected defect areas and a distribution map of the degree of doubt for defects on the inner wall of each first suspected defect area.

[0029] Convolutional neural network models include Convolutional Neural Networks (CNNs). A CNN contains structures such as convolutional layers, pooling layers, and fully connected layers. CNNs can learn deep features of images step-by-step through a multi-layered network structure to achieve tasks such as object recognition, classification, or region segmentation in images.

[0030] Multiple suspected first-defect areas are a set of regions that may have internal wall defects, determined by analyzing the visual imaging of the inner wall of the coal mine scraper conveyor using a defect feature analysis model.

[0031] Each suspected first defect area corresponds to a specific region in the visual imaging of the inner wall of a coal mine scraper conveyor. The suspected first defect area differs from the normal inner wall area in visual characteristics, such as abnormal color or irregular outline.

[0032] The inner wall defect suspicion distribution map for each first suspected defect zone is a data chart output by the defect feature analysis model, representing the probability distribution of the presence of inner wall defects at various locations within that first suspected defect zone. In the inner wall defect suspicion distribution map for each first suspected defect zone, different locations correspond to a probability value between 0 and 1. The closer the probability value is to 1, the higher the probability that a defect exists at that location.

[0033] The visual imaging of the inner wall of a coal mine scraper conveyor can completely record the surface visual information of the inner wall, including the visual differences between normal areas and potentially defective areas. For example, normal inner wall areas have uniform color and consistent texture, while potentially defective areas may exhibit visual features such as darker color, uneven surface, scratches, or cracks. These visual differences are crucial for defect feature analysis models to identify defects. By learning and analyzing these features, the model can filter out areas with defective characteristics and determine the probability of defects at different locations within each area.

[0034] Convolutional neural networks (CNNs) possess multi-layer feature extraction and spatial correlation preservation capabilities. CNNs can use different convolutional kernels to capture basic features such as inner wall edges and textures, and can distinguish the visual boundaries between defects and normal areas. Subsequent convolutional layers of CNNs can combine abstract features and extract advanced defect features such as crack morphology and corrosion patches, while filtering out interference such as dust and minor scratches. Furthermore, the local receptive field characteristic of CNNs can preserve the spatial location of features. Combined with fully connected layers and activation functions, defect features can be mapped to spatial probabilities. The model can calculate the defect probability at each location based on the matching degree between the region and the defect sample features. Continuous regions with probabilities reaching a threshold are designated as the first suspected defect area. The set of probabilities at all locations can then constitute a distribution map of the suspected inner wall defects.

[0035] In some embodiments, determining multiple first suspected defect areas and a distribution map of the degree of doubt for each first suspected defect area based on the visual imaging image of the inner wall of the coal mine scraper conveyor includes steps S21 to S23:

[0036] Step S21: Based on the visual imaging image of the inner wall of the coal mine scraper conveyor, determine the location of multiple visual feature anomalies, the grayscale deviation value between each visual feature anomaly and the surrounding normal inner wall area, and the pixel distribution uniformity of the surrounding normal inner wall area of ​​each visual feature anomaly.

[0037] In some embodiments, a convolutional neural network can be used to determine the locations of multiple visual feature anomalies, the grayscale deviation between each visual feature anomaly and the surrounding normal inner wall region, and the pixel distribution uniformity of the surrounding normal inner wall region for each visual feature anomaly.

[0038] The locations of multiple visual feature anomalies were determined by a convolutional neural network. These points are located in the visual imaging image of the inner wall of a coal mine scraper conveyor, and their coordinates differ significantly from the surrounding normal areas due to differences in visual features such as grayscale and texture.

[0039] The grayscale deviation value between each visual feature anomaly point and the surrounding normal inner wall area is determined by a convolutional neural network. It is the difference between the average grayscale value of each visual feature anomaly point and the average grayscale value of the surrounding inner wall area that has been determined to be defect-free and visually stable. The sign and magnitude of the difference reflect the direction and degree of the grayscale deviation of the anomaly point from the normal range.

[0040] The pixel distribution uniformity of the normal inner wall region surrounding each visual feature anomalous point is an index value determined by a convolutional neural network, which measures the stability of the spatial distribution of gray values ​​of all pixels within the normal inner wall region surrounding a single visual feature anomalous point.

[0041] Convolutional neural networks (CNNs) can perform layer-by-layer feature mining on the visual image of the inner wall of a coal mine scraper conveyor through sliding convolution operations. They can progressively capture visual differences related to defects in the inner wall, such as abrupt changes in grayscale and texture breaks often observed in defective areas. These features are key to identifying visual anomalies, allowing the CNN to pinpoint their coordinates. Leveraging the local receptive field characteristic, the CNN can simultaneously correlate a certain range of pixels around a single pixel when analyzing it. By calculating the difference in grayscale mean between the central pixel and the surrounding area, it can determine the grayscale deviation between each visual anomaly and the surrounding normal inner wall area. Furthermore, the CNN can analyze the dispersion of grayscale values ​​in surrounding areas. For example, by extracting grayscale fluctuation features within the region using convolutional kernels and combining this with quantization calculations in subsequent fully connected layers, it can accurately determine the uniformity of pixel distribution in the normal inner wall area surrounding the anomaly.

[0042] Step S22: Based on the locations of the multiple visual feature anomalies, the grayscale deviation between each visual feature anomaly and the surrounding normal inner wall area, and the pixel distribution uniformity of the surrounding normal inner wall area, determine the boundary range of multiple initially selected first defect suspected areas, the distribution information of visual feature anomalies in each initially selected first defect suspected area, the defect risk coefficient of each visual feature anomaly, and the proportion of visual feature anomalies with high defect risk coefficients in each initially selected first defect suspected area.

[0043] In some embodiments, a deep neural network can be used to determine the boundary range of multiple preliminary first suspected defect areas, the distribution information of visual feature anomalies in each preliminary first suspected defect area, the defect risk coefficient of each visual feature anomaly, and the proportion of visual feature anomalies with high defect risk coefficients in each preliminary first suspected defect area.

[0044] A deep neural network (DNN) is a neural network model composed of multiple hidden layers. The structure of a deep neural network includes an input layer, multiple hidden layers, and an output layer. The input layer receives raw data, and each hidden layer consists of multiple neurons. These neurons use activation functions such as ReLU and sigmoid to perform nonlinear transformations on the input signal, thereby gradually extracting and abstracting features. The output layer outputs results according to the task requirements. By stacking multiple hidden layers, deep neural networks can learn complex, deep features of data.

[0045] The boundary range of multiple preliminary suspected defect areas is the spatial range covered by the closed contour lines of multiple consecutive image regions that are divided by a deep neural network for preliminary determination of possible defects.

[0046] The distribution information of visual feature anomalies within each preliminary first suspected defect area is a set of information output by a deep neural network, which includes the specific coordinates of all visual feature anomalies within a single preliminary first suspected defect area, the spatial arrangement of each visual feature anomaly within the area, and the relative distance between each anomaly.

[0047] The spatial arrangement of each visual feature anomaly point within the region can be clustered, linear, or discrete.

[0048] The defect risk coefficient for each visual feature anomaly is a numerical index determined by a deep neural network, representing the probability of association between each visual feature anomaly and the actual defects on the inner wall of the coal mine scraper conveyor.

[0049] The percentage of visual feature anomalies with high defect risk coefficients within each preliminary first defect suspected area is the ratio of the number of visual feature anomalies with defect risk coefficients reaching a preset high-risk threshold within a single preliminary first defect suspected area, as output by a deep neural network, to the total number of all visual feature anomalies within that preliminary first defect suspected area.

[0050] Deep neural networks can perform spatial correlation analysis on the coordinate data of multiple visual feature anomalies through hidden layers to identify the clustering patterns of these anomalies and delineate continuous region boundaries, thereby obtaining the boundary range of multiple preliminary suspected defect areas. Deep neural networks can simultaneously calculate the coordinate arrangement of anomalies within each region and generate visual feature anomaly distribution information. Furthermore, deep neural networks can use the grayscale deviation value of each visual feature anomaly and the uniformity of surrounding pixel distribution as input features, and then establish a mapping relationship with defect correlation through multi-layer nonlinear transformations, thereby quantifying the defect risk coefficient of each anomaly. Simultaneously, the model can statistically analyze the defect risk coefficients within each preliminary region to calculate the ratio of the number of high-risk anomalies to the total number of anomalies in the region, thus determining the proportion of high-risk anomalies.

[0051] Step S23: Based on the boundary range of the multiple preliminary first suspected defect areas, the distribution information of visual feature anomalies in each preliminary first suspected defect area, the defect risk coefficient of each visual feature anomaly, and the proportion of visual feature anomalies with high defect risk coefficients in each preliminary first suspected defect area, determine the multiple first suspected defect areas and the distribution map of the degree of doubt for defects on the inner wall of each first suspected defect area.

[0052] In some embodiments, a deep neural network can be used to determine multiple first defect suspected areas and a distribution map of the degree of doubt for defects on the inner wall of each first defect suspected area.

[0053] Deep neural networks can analyze the spatial integrity and independence of the boundary range of each initially selected first-stage suspected defect area through hidden layer operations. They can also identify areas conforming to typical spatial morphology of defects by combining the distribution information of visual feature anomalies within each initially selected area. Deep neural networks can transform the defect risk coefficient of each visual feature anomaly into a key weight for area determination. After deep integration with anomaly distribution information, the model can use nonlinear transformations to highlight the impact of high-risk anomalies on the overall defect probability of the area. Then, combined with a quantitative assessment of the proportion of high-risk anomalies within each initially selected area, multiple first-stage suspected defect areas with acceptable probabilities can be selected. Furthermore, deep neural networks can also construct a continuous doubt data matrix covering the entire first-stage suspected defect area by interpolating the spatial distribution patterns of anomaly risk coefficients within the area to complete the doubt value for locations without anomalies. This allows the model to transform the continuous doubt data matrix into an intuitive distribution map of inner wall defect doubts.

[0054] Step S3: Based on the visual imaging images of multiple suspected first defect areas and the distribution map of the doubt about the inner wall defects of each suspected first defect area, determine multiple basic detection points for each suspected first defect area, and obtain the inner wall state perception data of the multiple basic detection points for each suspected first defect area.

[0055] The visual imaging image of the first suspected defect area is a set of local imaging images corresponding to the first suspected defect area, which are extracted from the visual imaging image of the inner wall of the coal mine scraper conveyor.

[0056] In some embodiments, a basic detection determination model can be used to determine multiple basic detection points for each first suspected defect area. The basic detection determination model is a Transformer model. The inputs to the basic detection determination model are visual images of the multiple first suspected defect areas and a distribution map of the degree of suspicion of internal wall defects in each first suspected defect area. The output of the basic detection determination model is multiple basic detection points for each first suspected defect area.

[0057] The Transformer model is a deep learning model based on a self-attention mechanism. The Transformer model architecture mainly consists of two parts: an encoder and a decoder. The encoder consists of multiple identical encoding layers, each containing a multi-head self-attention mechanism and a feedforward neural network. The multi-head self-attention mechanism allows the model to simultaneously focus on information from different positions in the input data to capture the internal dependencies within the data. The feedforward neural network can perform independent non-linear transformations on the features at each position. The decoder adds a multi-head cross-attention mechanism to the encoder, allowing the decoder to focus on the correlation between the features output by the encoder and the current input to the decoder.

[0058] The multiple basic detection points in each first defect suspected area are the set of specific location points within each first defect suspected area that need to be detected by the basic detection determination model.

[0059] Each basic detection point corresponds to a precise spatial location within the first suspected defect area. These locations can be those with a high degree of suspicion of defect, obvious visual abnormalities, or representative locations.

[0060] The inner wall condition sensing data for each suspected first defect area is obtained by using sensor detection equipment to detect multiple basic detection points in each suspected first defect area. This data reflects the specific inner wall condition at the basic detection points. The inner wall condition sensing data includes ultrasonic vibration signal data of the inner wall surface and inner wall material density feedback data.

[0061] The ultrasonic vibration signal data of the inner wall surface is the signal data generated by the vibration of the inner wall surface at the basic detection point by the ultrasonic sensor. The ultrasonic vibration signal data can reflect whether there are defects such as internal cracks or loosening in the inner wall.

[0062] The inner wall material density feedback data is obtained by detecting the density of the inner wall material at the basic detection point using an ultrasonic density detector. It can reflect whether there are defects such as material corrosion or voids in the inner wall that cause abnormal density.

[0063] The visual image of the first suspected defect area clearly presents the visual details within it, including the shape, size, location, and surface texture changes of the abnormal area. This information helps the model understand the spatial structural characteristics of the suspected area. The defect suspicion distribution map of the inner wall of each first suspected defect area clarifies the probability of defects at different locations within each area. Locations with higher defect suspicion are more likely to contain actual defects, enabling the model to accurately locate the basic detection points that need to be detected.

[0064] The Transformer model can transform visual images of multiple suspected first-order defect areas and internal wall defect suspicion distribution maps into sequences with location encoding. Then, through multi-head self-attention, it can capture global correlations between the data, such as the correspondence between defect features in the visual images and high-probability locations in the suspicion distribution map. This allows for precise localization of key locations with visual anomalies and high suspicion levels. The Transformer model can focus on local feature details in the visual images of each suspected first-order defect area, such as the possible defect edge morphology and abnormal surface texture areas within the suspected area. Simultaneously, it can combine this with the global distribution patterns presented by the internal wall defect suspicion distribution map of each suspected first-order defect area, such as the spatial distribution range of high-suspicious areas within the suspected area and the existence of continuous high-suspicious areas. Through this comprehensive analysis of local and global information, the Transformer model can rationally plan the distribution of detection points, ultimately selecting multiple basic detection points that are both representative of defects and cover key parts of the suspected area.

[0065] Step S4: Construct an inner wall detection map. The inner wall detection map includes multiple basic detection point nodes and edges between multiple nodes. The node features of each basic detection point node are the location of the basic detection point and the inner wall state perception data of the basic detection point. The edges between detection nodes are the direction and distance between the basic detection point nodes and the similarity of the inner wall state perception data.

[0066] The inner wall inspection map is a map data constructed based on multiple basic inspection points in each suspected first defect area and their inner wall state perception data. It can be used to represent the spatial and state correlations between the basic inspection points. The inner wall inspection map can organize the scattered basic inspection point data in a structured form, thereby intuitively presenting the connections between the basic inspection points.

[0067] Multiple basic detection point nodes are the fundamental building blocks of the inner wall detection map, with each node corresponding to a basic detection point. The node features of each basic detection point node include its location and the perceived inner wall condition data. The location of the basic detection point is its precise coordinates within the coordinate system of the coal mine scraper conveyor's inner wall. The perceived inner wall condition data includes ultrasonic vibration signal data of the inner wall surface and feedback data of the inner wall material density. The node features of each basic detection point node comprehensively reflect its spatial location and inner wall condition.

[0068] Edges between multiple nodes can be used to represent the association attributes between basic detection point nodes. The edges include the direction and distance between the basic detection point nodes, as well as the similarity of the inner wall state perception data between them.

[0069] The similarity of the inner wall state sensing data is obtained by calculating the similarity value of the inner wall state sensing data of two basic detection point nodes. The higher the similarity, the closer the inner wall state of the two detection points is.

[0070] In some embodiments, a deep neural network can be used to determine the similarity of the inner wall state perception data between basic detection point nodes.

[0071] Step S5: Process the inner wall detection map based on the graph neural network to determine multiple advanced detection points for each first suspected defect area, and obtain the inner wall state perception data of the multiple advanced detection points for each first suspected defect area.

[0072] Graph Neural Networks (GNNs) are a class of deep learning models specifically designed for processing graph data. GNNs update node representations by aggregating the features of a node itself and its neighbors, thereby capturing the relationships between nodes and the overall structural information of the graph. The input to the GNN is the inner wall detection graph, and the output is multiple advanced detection points for each suspected first defect area.

[0073] Multiple advanced detection points in each first suspected defect area are a set of additional locations within each first suspected defect area that require further detection, obtained by processing the inner wall detection map using a graph neural network.

[0074] Advanced detection points can be located in areas where there is a close correlation between basic detection points, abnormal correlation in the inner wall condition, or insufficient coverage by basic detection points. These areas may contain defect information that has not been captured by basic detection points. By detecting advanced detection points, the inner wall condition data of the first suspected defect area can be further supplemented and improved, thereby enhancing the comprehensiveness and accuracy of defect detection.

[0075] The inner wall condition sensing data for each suspected first defect area is obtained by sensor detection equipment after detecting multiple advanced detection points in each suspected first defect area. This data reflects the specific inner wall condition at each advanced detection point. The inner wall condition sensing data for the advanced detection points includes ultrasonic vibration signal data of the inner wall surface and inner wall material density feedback data.

[0076] By constructing an inner wall detection map, the spatial and state association networks among multiple basic detection points can be clearly reflected. This association information is crucial for determining advanced detection points, as the rationality of the spatial distribution of basic detection points and the existence of state correlations directly affect the complete judgment of the defect distribution in the first suspected defect area. By using the location of basic detection points and inner wall state perception data as node features, and the directional distance and state similarity between basic detection points as edge features, the information in the state perception data can be utilized more fully. This helps the graph neural network better understand the spatial positional relationships and state association patterns between basic detection points, thereby avoiding the omission of potential defect areas due to isolated analysis of individual detection points. Processing inner wall detection map data based on graph neural networks can effectively learn the complex associations and information transmission between basic detection points, thereby more accurately identifying blank areas not fully covered by basic detection points. Compared with traditional single-point analysis methods, graph neural networks have better global association representation capabilities and region analysis capabilities when processing map data.

[0077] Graph neural networks excel at uncovering the relationships between nodes and edges in a graph. In the inner wall detection graph, basic detection points are nodes, and edges include positional distances and state similarities. Graph neural networks can capture the relationships between basic detection points. For example, if a node has a similar state and is spatially continuous with its surrounding nodes, it indicates the possible existence of a continuous defect region. Alternatively, if two nodes are close but have significantly different states, it suggests the possible existence of a defect boundary. Through this relationship analysis, graph neural networks can identify blank areas with insufficient coverage by basic detection points and potential defect associations, requiring supplementary detection. Furthermore, graph neural networks can combine the global graph structure to determine which blank areas are crucial to the completeness of defect detection, thereby identifying multiple advanced detection points.

[0078] Step S6: Cluster the inner wall state perception data of multiple basic detection points in each first defect suspected area and the inner wall state perception data of multiple advanced detection points in each first defect suspected area to obtain K clusters.

[0079] The clustering algorithm described is K-means clustering. K-means clustering is an unsupervised learning algorithm that divides an input dataset into K distinct clusters based on the similarity between data points. This results in high similarity among data points within the same cluster and low similarity among data points in different clusters. K is a preset number of clusters; in some embodiments, K can be pre-set manually.

[0080] The K clusters are datasets obtained by clustering the internal wall condition sensing data of multiple basic detection points and multiple advanced detection points in each suspected first defect area using a clustering algorithm. Each cluster contains internal wall condition sensing data from several basic or advanced detection points. Internal wall condition sensing data within the same cluster exhibit high similarity; for example, if the ultrasonic vibration signal waveforms and material density values ​​of all detection points within a cluster are similar, it indicates that the internal wall conditions corresponding to these detection points are similar. Internal wall condition sensing data within different clusters show low similarity; for example, if the material density of detection points in one cluster is significantly lower than that in another cluster, it indicates a significant difference in the internal wall conditions between the two clusters.

[0081] The process of clustering the inner wall condition sensing data of multiple basic detection points and multiple advanced detection points in each first suspected defect area using the K-means clustering algorithm is as follows: First, K data samples are randomly selected as initial cluster centers from the two types of inner wall condition sensing data sets of each first suspected defect area. Next, for the inner wall condition sensing data of each detection point in the first suspected defect area, the distance between it and these K initial cluster centers is measured and calculated using Euclidean distance, and the inner wall condition sensing data of the detection point is assigned to the corresponding cluster according to the principle of closest proximity. After the inner wall condition sensing data of all detection points has been divided, the average values ​​of various features in the inner wall condition sensing data of all detection points in each cluster, such as the amplitude and frequency of ultrasonic vibration signals and the material density, are recalculated, and the cluster centers of each cluster are updated accordingly. The above steps of data partitioning and cluster center updating are repeated until the change in cluster centers between two iterations is minimal, such as less than a preset threshold. At this point, the clustering process is considered to have converged, thus completing the K-means clustering of the inner wall state perception data of the two types of detection points in the first defect suspected area.

[0082] By dividing the data into K clusters, the scattered detection data can be categorized according to state similarity, thus providing a classification basis for analyzing the defect signal distribution in the first suspected defect area. Clustering can effectively integrate the complex detection data of each first suspected defect area. Since there are many basic and advanced detection points within each first suspected defect area, and the characteristics of the inner wall state perception data of the two types of detection points are diverse, directly analyzing the data of each detection point individually makes it difficult to quickly grasp the overall defect distribution trend. However, clustering can group detection points with similar inner wall state perception data characteristics into one category and form multiple clusters, which can greatly simplify the data structure and make the originally scattered detection data exhibit obvious category correlations.

[0083] Step S7: Determine the distribution map of doubt for the inner wall defect signal feedback for each first defect suspected area based on the K clusters.

[0084] In some embodiments, a defect signal analysis model can be used to determine the distribution map of doubtful feedback of inner wall defect signals for each first suspected defect area. The defect signal analysis model is a Transformer model. The input to the defect signal analysis model is the K clusters, and the output of the defect signal analysis model is the distribution map of doubtful feedback of inner wall defect signals for each first suspected defect area.

[0085] The distribution map of the doubtfulness of the inner wall defect signal feedback in each first defect suspected area is a data chart generated by analyzing K clusters through a defect signal analysis model. It is used to represent the probability distribution of the presence of inner wall defects at each location in each first defect suspected area based on the detection signal data.

[0086] K clusters can visually represent the different internal wall condition types and spatial distribution of detection points within the suspected area. By analyzing these clusters, the model can quickly determine the internal wall condition characteristics corresponding to each cluster. For example, if all detection points in one cluster show low material density, while another cluster shows abnormal ultrasonic vibration signals, it can then determine whether each cluster is related to a defect. By combining the distribution patterns of the clusters, the model can also preliminarily locate concentrated areas of suspected defects, thus providing a clear classification basis for determining the distribution map of doubt in the internal wall defect signal feedback, and avoiding defect judgment bias caused by data clutter.

[0087] The Transformer model extracts the location information of all detection points within each cluster, the corresponding inner wall state-sensing data features, and the number of detection points within the cluster. This information is then organized into sequence data according to cluster category, with each cluster corresponding to a data sequence unit, thus forming the cluster sequence data for the entire first suspected defect area. Next, the Transformer model inputs the cluster sequence data into the encoder for processing. The encoder's multi-head self-attention mechanism calculates the attention weights between each cluster data sequence unit and other cluster data sequence units, and captures the correlations between different clusters, such as the spatial positional relationship between a potential defect cluster and surrounding normal clusters, and the degree of difference in their inner wall state features. Simultaneously, the self-attention mechanism also focuses on the location distribution and state feature consistency of detection points within the same cluster, such as whether detection points within a defect cluster are spatially continuous and whether the state features of the detection points within the cluster are highly consistent. Then, the encoder's feedforward neural network performs a nonlinear transformation on the cluster features processed by the self-attention mechanism to enhance the defect feature information of the clusters. Subsequently, the model can use interpolation algorithms to derive the doubt level of defect signal feedback at locations without detection points within the first suspected defect zone, based on the processed cluster features and the location distribution of detection points within each cluster. For the location of a detection point, the model can directly determine the doubt level based on the defect probability of its cluster. For example, the doubt level for a detection point belonging to a severe defect cluster is set to 0.8-1.0, while the doubt level for a detection point belonging to a normal cluster is set to 0-0.2. For locations without detection points, the model can calculate the doubt level based on the defect probability of the clusters to which surrounding detection points belong and the distance weight. For example, if a location is surrounded by many high-doubt detection points and is relatively close, then the doubt level of that location is high. Finally, the model can integrate the doubt levels of defect signal feedback at all locations within the first suspected defect zone to generate a distribution map of the doubt level of defect signal feedback on the inner wall of each first suspected defect zone.

[0088] Step S8: Determine the second suspected defect area based on the inner wall defect doubt distribution map of each first suspected defect area and the inner wall defect signal feedback doubt distribution map of each first suspected defect area.

[0089] In some embodiments, a defect region screening model can be used to determine a second suspected defect region. The defect region screening model is a convolutional neural network model. The inputs to the defect region screening model are the inner wall defect suspicion distribution map and the inner wall defect signal feedback suspicion distribution map of each first suspected defect region, and the output of the defect region screening model is the second suspected defect region.

[0090] The second suspected defect area is a region that is more likely to have a real defect after analyzing the distribution map of doubt about the inner wall defects and the distribution map of doubt about the feedback of the inner wall defects in each of the first suspected defect areas using the defect area screening model.

[0091] The second suspected defect zone is smaller than the first suspected defect zone. The second suspected defect zone is determined based on the first suspected defect zone by judging visual features and measured signal features. The visual features refer to the distribution of doubt about the inner wall defect, and the measured signal features refer to the distribution of doubt about the inner wall defect signal feedback. The second suspected defect zone simultaneously meets the conditions of high visual doubt about the defect and high doubt about the defect in the measured signal.

[0092] Convolutional neural networks (CNNs) can perform feature mining on both the distribution map of doubtful internal wall defects and the distribution map of doubtful internal wall defect signal feedback. Furthermore, convolutional layers can capture key features such as the edge contours and spatial distribution patterns of highly doubtful areas in the distribution map. The model combines the extracted features from the distribution map through a feature fusion layer, ensuring that the fused features simultaneously cover highly doubtful information from both the visual and signal dimensions, avoiding misjudgments caused by single-dimensional analysis. Fully connected layers and activation functions can classify the fused feature map and identify pixels with double high doubt as secondary suspected defect areas, ultimately accurately determining secondary suspected defect areas that meet both conditions.

[0093] Step S9: Determine the core defect area of ​​the scraper conveyor's inner wall based on the inner wall state perception data of multiple basic detection points in the second defect suspected area and the inner wall state perception data of multiple advanced detection points in the second defect suspected area.

[0094] In some embodiments, Figure 4 This is a flowchart illustrating a process for determining the core defect region of the inner wall of a scraper conveyor, as provided in an embodiment of the present invention. The determination of the core defect region includes steps S31-S32:

[0095] Step S31: Based on the inner wall state perception data of multiple basic detection points in the second defect suspected area and the inner wall state perception data of multiple advanced detection points in the second defect suspected area, the defect verification determination model is used to determine multiple defect verification points in each second defect suspected area.

[0096] The defect verification and determination model is a deep neural network model. The input of the defect verification and determination model is the inner wall state perception data of multiple basic detection points in the second defect suspected area and the inner wall state perception data of multiple advanced detection points in the second defect suspected area. The output of the defect verification and determination model is multiple defect verification points in each second defect suspected area.

[0097] The multiple defect verification points in each second defect suspected area are points determined by the defect verification determination model to accurately verify the existence of defects.

[0098] The inner wall state perception data of multiple basic detection points in the second suspected defect area and the inner wall state perception data of multiple advanced detection points in the second suspected defect area correspond to the preliminary detection and supplementary detection results of the second suspected defect area, respectively, and can achieve complete coverage of all detection locations in the suspected area. The model can analyze the distribution patterns of abnormal information in the input data, such as whether abnormalities at different detection locations show a continuous distribution and whether abnormalities recorded at different detection stages have common characteristics, thereby determining which locations' abnormalities are more valuable for defect verification.

[0099] Deep neural networks possess multi-layer feature extraction and complex correlation analysis capabilities, accurately identifying points valuable for defect verification. They can process the inner wall state perception data of multiple basic detection points and multiple advanced detection points in the suspected second defect area in multiple layers. The model extracts basic abnormal features from single-class data through shallow networks, and then mines the correlation patterns between the two classes of data through deep networks. For example, it can determine whether anomalies at different detection locations exhibit spatial continuity, and whether there are common trends in anomalies between preliminary and supplementary detection records. This allows for the elimination of isolated and unrelated abnormal data, and the focus on anomalies that point to potential defects. The model assigns differentiated weights to anomaly features at different locations through nonlinear transformations and weight allocation mechanisms. Higher weights are given to locations with significant anomaly features that spatially represent the core of potential defects, while lower weights are given to locations with weak or unrelated features. Finally, by combining preset defect verification point judgment criteria, the model can filter out high-weight locations, which are the defect verification points that accurately point to potential defects and require further verification.

[0100] Step S32: Determine the core defect area of ​​the scraper conveyor's inner wall based on the precise detection data of inner wall defects from multiple defect verification points in each second suspected defect area.

[0101] Precise detection data of internal wall defects at multiple defect verification points in each suspected second defect area were obtained by using an electromagnetic pulse detector to conduct targeted detection at multiple defect verification points in each suspected second defect area. This data accurately reflects the actual condition of internal wall defects within a local area centered on the defect verification point. The precise detection data of internal wall defects includes the specific spatial coordinates of the defect verification point, the quantitative parameters of the internal wall defects within the local area centered on the defect verification point, and the physical property parameters of the internal wall material within that local area.

[0102] The quantitative parameters of internal wall defects include the depth, length and width of cracks, the diameter, depth and volume of cavities, and the area, corrosion depth and corrosion degree of the corrosion zone.

[0103] Physical performance parameters include the material's density, hardness, and elastic modulus.

[0104] In some embodiments, a core defect determination model can be used to determine the core defect region on the inner wall of the scraper conveyor. The core defect determination model is a deep neural network model. The input to the core defect determination model is precise detection data of inner wall defects from multiple defect verification points in each second suspected defect area, and the output of the core defect determination model is the core defect region on the inner wall of the scraper conveyor.

[0105] The core defect area of ​​the scraper conveyor inner wall is determined by analyzing the precise detection data of multiple defect verification points in each second suspected defect area using the core defect determination model. This area is the scraper conveyor inner wall area where defects actually exist and whose defect severity meets the standards for attention.

[0106] The precise detection data of internal wall defects at multiple defect verification points in each second suspected defect zone, centered on the defect verification points, accurately records the quantitative parameters of internal wall defects and the physical property parameters of surrounding materials within the corresponding local area. This data offers higher precision and detail, directly revealing the true morphology, severity, and spatial distribution characteristics of the defects. The model can accurately identify the boundary range of defects and determine their type and attributes based on this data. By analyzing the correlation of defect data at different verification points, the model can also determine whether defects exhibit continuous or concentrated distribution characteristics, thereby accurately delineating the core areas where defects truly exist and require focused attention.

[0107] Deep neural networks can input precise detection data of inner wall defects from multiple defect verification points in each suspected second defect area into the input layer for preprocessing, ensuring that the numerical ranges of different parameter types are within the range suitable for model analysis. After entering the hidden layers, the deep neural network can progressively analyze the defect features in the precise detection data of inner wall defects through multi-layer nonlinear transformations. The model can identify whether features conforming to the defect definition exist within a local region centered on each verification point. Simultaneously, by combining the spatial coordinate information of the verification points, it can analyze the spatial correlation of defect features at different verification points, thereby determining whether defects at adjacent verification points belong to the same continuous defect system. For example, it can determine whether defect features at different verification points exhibit morphological continuity or distributional correlation. The deep neural network can calculate weights for the defect features corresponding to each verification point. The weight is determined based on the significance and spatial correlation of the defect; verification points with more obvious defect features and closer correlation to surrounding defects receive higher weights. The model then categorizes high-weight verification points, grouping those that are spatially close and have the same defect type into the same defect group. Next, it fits the defect region boundary of each group based on the coordinates of the verification points and the defect features, ensuring that the boundary completely covers the defect range reflected by all verification points within the group. Finally, the deep neural network determines whether the defect features within a region are continuous and whether there are obvious data gaps. It ultimately retains regions with continuous defect features and clearly defined coverage areas, identifying these regions as the core defect areas of the scraper conveyor's inner wall.

[0108] Based on the same inventive concept Figure 5 This is a schematic diagram of a visual inspection system for defects on the inner wall of a coal mine scraper conveyor, provided in an embodiment of the present invention. The visual inspection system for defects on the inner wall of a coal mine scraper conveyor includes:

[0109] The imaging acquisition module 41 is used to acquire a visual image of the inner wall of the coal mine scraper conveyor.

[0110] The area positioning module 42 is used to determine multiple first defect suspected areas and the inner wall defect suspicion distribution map of each first defect suspected area based on the visual imaging map of the inner wall of the coal mine scraper conveyor.

[0111] The basic detection module 43 is used to determine multiple basic detection points for each first defect suspected area based on the visual imaging map of multiple first defect suspected areas and the inner wall defect doubt distribution map of each first defect suspected area, and to obtain the inner wall state perception data of the multiple basic detection points for each first defect suspected area.

[0112] Module 44 is used to construct an inner wall detection map. The inner wall detection map includes multiple basic detection point nodes and edges between multiple nodes. The node features of each basic detection point node are the position of the basic detection point and the inner wall state perception data of the basic detection point. The edges between detection nodes are the direction and distance between the basic detection point nodes and the similarity of the inner wall state perception data.

[0113] The advanced detection module 45 is used to process the inner wall detection map based on the graph neural network to determine multiple advanced detection points for each first defect suspected area, and to acquire the inner wall state perception data of the multiple advanced detection points for each first defect suspected area.

[0114] Clustering analysis module 46 is used to cluster K clusters based on the inner wall state perception data of multiple basic detection points in each first defect suspected area and the inner wall state perception data of multiple advanced detection points in each first defect suspected area.

[0115] Signal analysis module 47 is used to determine the distribution map of doubt of the inner wall defect signal feedback for each first defect suspected area based on the K clusters;

[0116] The suspected area screening module 48 is used to determine the second suspected defect area based on the inner wall defect doubt distribution map of each first suspected defect area and the inner wall defect signal feedback doubt distribution map of each first suspected defect area.

[0117] The core defect determination module 49 is used to determine the core defect area of ​​the scraper machine's inner wall based on the inner wall state perception data of multiple basic detection points in the second defect suspected area and the inner wall state perception data of multiple advanced detection points in the second defect suspected area.

[0118] It should be noted that, in order to simplify the descriptions disclosed herein and thus aid in the understanding of one or more embodiments of the invention, the foregoing description of embodiments of this specification may sometimes combine multiple features into a single embodiment, drawing, or description thereof. However, this method of disclosure does not imply that the subject matter of this specification requires more features than those mentioned in the claims. In fact, the embodiments contain fewer features than all the features of a single embodiment disclosed above.

[0119] Finally, it should be understood that the embodiments described in this specification are merely illustrative of the principles of the embodiments described herein. Other variations may also fall within the scope of this specification. Therefore, alternative configurations of the embodiments described herein are intended to be illustrative rather than limiting, and should be considered consistent with the teachings of this specification. Accordingly, the embodiments described herein are not limited to those explicitly introduced and described herein.

Claims

1. A visual inspection method for defects on the inner wall of a coal mine scraper conveyor, characterized in that, include: Obtain a visual image of the inner wall of a coal mine scraper conveyor; Based on the visual imaging of the inner wall of the coal mine scraper conveyor, multiple first defect suspected areas and a distribution map of the degree of doubt of inner wall defects in each first defect suspected area are determined. Based on the visual imaging images of multiple suspected first defect areas and the distribution map of the doubt of inner wall defects in each suspected first defect area, multiple basic detection points are determined for each suspected first defect area, and the inner wall state perception data of the multiple basic detection points of each suspected first defect area is obtained. The inner wall state perception data includes the ultrasonic vibration signal data of the inner wall surface and the inner wall material density feedback data. An inner wall detection map is constructed, which includes multiple basic detection point nodes and edges between multiple nodes. The node features of each basic detection point node are the position of the basic detection point and the inner wall state perception data of the basic detection point. The edges between detection nodes are the direction and distance between the basic detection point nodes and the similarity of the inner wall state perception data. The inner wall detection map is processed based on a graph neural network to determine multiple advanced detection points for each first suspected defect area, and the inner wall state perception data of the multiple advanced detection points for each first suspected defect area is obtained. The advanced detection points are newly added location points in each first suspected defect area that need further detection. K clusters are obtained by clustering the inner wall state perception data of multiple basic detection points in each first defect suspected area and the inner wall state perception data of multiple advanced detection points in each first defect suspected area. Based on the K clusters, determine the distribution map of doubt for the feedback of inner wall defect signals in each first suspected defect area; The second suspected defect area is determined based on the inner wall defect doubt distribution map of each first suspected defect area and the inner wall defect signal feedback doubt distribution map of each first suspected defect area. Based on the inner wall condition perception data of multiple basic detection points in the second defect suspected area and the inner wall condition perception data of multiple advanced detection points in the second defect suspected area, the core defect area of ​​the scraper conveyor inner wall is determined.

2. The visual inspection method for defects on the inner wall of a coal mine scraper conveyor as described in claim 1, characterized in that, The determination of the core defect area of ​​the scraper conveyor's inner wall based on the inner wall state perception data of multiple basic detection points in the second defect suspected area and the inner wall state perception data of multiple advanced detection points in the second defect suspected area includes: Based on the inner wall state perception data of multiple basic detection points in the second defect suspected area and the inner wall state perception data of multiple advanced detection points in the second defect suspected area, the defect verification determination model is used to determine multiple defect verification points in each second defect suspected area. Based on the precise detection data of inner wall defects from multiple defect verification points in each second suspected defect area, the core defect area of ​​the scraper conveyor's inner wall is determined.

3. The visual inspection method for defects on the inner wall of a coal mine scraper conveyor as described in claim 1, characterized in that, The input to the graph neural network is the inner wall detection map, and the output of the graph neural network is multiple advanced detection points for each first suspected defect area.

4. A visual inspection system for defects on the inner wall of a coal mine scraper conveyor, characterized in that, include: The imaging acquisition module is used to acquire visual images of the inner wall of the coal mine scraper conveyor. The area positioning module is used to determine multiple first defect suspected areas and the inner wall defect suspicion distribution map of each first defect suspected area based on the visual imaging map of the inner wall of the coal mine scraper conveyor. The basic detection module is used to determine multiple basic detection points for each first defect suspected area based on visual imaging images of multiple first defect suspected areas and the distribution map of doubt of inner wall defects in each first defect suspected area, and to acquire inner wall state perception data of multiple basic detection points in each first defect suspected area. The inner wall state perception data includes ultrasonic vibration signal data of inner wall surface and inner wall material density feedback data. The construction module is used to construct an inner wall detection map. The inner wall detection map includes multiple basic detection point nodes and edges between multiple nodes. The node features of each basic detection point node are the position of the basic detection point and the inner wall state perception data of the basic detection point. The edges between detection nodes are the direction and distance between the basic detection point nodes and the similarity of the inner wall state perception data. The advanced detection module is used to process the inner wall detection map based on the graph neural network to determine multiple advanced detection points for each first suspected defect area, and to acquire the inner wall state perception data of multiple advanced detection points for each first suspected defect area. The advanced detection points are newly added location points in each first suspected defect area that need to be further detected. The clustering analysis module is used to cluster K clusters based on the inner wall state perception data of multiple basic detection points in each first defect suspected area and the inner wall state perception data of multiple advanced detection points in each first defect suspected area. The signal analysis module is used to determine the distribution map of doubt for the feedback of inner wall defect signals in each first defect suspected area based on the K clusters. The suspected area screening module is used to determine the second suspected defect area based on the inner wall defect doubt distribution map of each first suspected defect area and the inner wall defect signal feedback doubt distribution map of each first suspected defect area. The core defect determination module is used to determine the core defect area of ​​the scraper conveyor's inner wall based on the inner wall state perception data of multiple basic detection points in the second defect suspected area and the inner wall state perception data of multiple advanced detection points in the second defect suspected area.

5. The visual inspection system for defects on the inner wall of a coal mine scraper conveyor as described in claim 4, characterized in that, The core defect determination module is also used for: Based on the inner wall state perception data of multiple basic detection points in the second defect suspected area and the inner wall state perception data of multiple advanced detection points in the second defect suspected area, the defect verification determination model is used to determine multiple defect verification points in each second defect suspected area. Based on the precise detection data of inner wall defects from multiple defect verification points in each second suspected defect area, the core defect area of ​​the scraper conveyor's inner wall is determined.

6. The visual inspection system for defects on the inner wall of a coal mine scraper conveyor as described in claim 4, characterized in that, The input to the graph neural network is the inner wall detection map, and the output of the graph neural network is multiple advanced detection points for each first suspected defect area.

7. An electronic device, characterized in that, include: processor; Memory; And a computer program; wherein the computer program is stored in the memory and configured to be executed by the processor to implement the visual inspection method for defects on the inner wall of a coal mine scraper conveyor as described in any one of claims 1 to 3.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the visual inspection method for defects on the inner wall of a coal mine scraper conveyor as described in any one of claims 1 to 3.

Images