Mining area production safety incident intelligent early warning method fused with large model reasoning

Abstract

The invention relates to the technical field of mine safety production, in particular to a mining area production safety incident intelligent early warning method fused with large model reasoning, which comprises the following steps: respectively acquiring a surface image and an infrared feature image of a supporting device steel structure and identifying point features of all suspected damages; determining a false noisy point and a real corrosion point, and determining a region where the real corrosion point is located as a target damage region; obtaining a false noisy point set, respectively extracting the false noisy points in the false noisy point set and the infrared characteristic spectrum of the target damage area, and adjusting the exposure time of the infrared imager; calculating a structural stiffness reduction coefficient and an updated inherent frequency; when the vibration sensor monitors a rock mass vibration event, an alarm threshold value of rock mass vibration is adjusted; and continuing to monitor the rock mass so as to complete intelligent early warning of rock mass stability and production safety events in the mining area. According to the invention, the early warning accuracy of the mining area production safety event is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of mine safety production, and in particular to a mine production safety event intelligent early warning method fusing large model reasoning. BACKGROUND

[0002] Mine production safety is the lifeline in the field of mineral resources development, and is related to personnel life safety and national property safety. However, the underground environment of the mine area is complex and changeable, the geological conditions are poor, and the dangerous factors such as water, gas, ground pressure and dust are intertwined, so that production safety events such as roof falling, rib spalling, water inrush and gas outburst frequently occur. When the monitoring data exceeds the preset experience threshold, an alarm is triggered, wherein the threshold setting depends on manual experience, which is difficult to adapt to the dynamic geological conditions of different mine areas, and is prone to false alarms or missed alarms. Secondly, this method can only provide a binary judgment of whether it is over standard, and cannot reveal the risk evolution process and internal causes. The early warning information lacks interpretability, and is limited to the visible surface level through computer vision algorithms. The microcrack development and stress redistribution inside the rock mass will cause errors in triggering the alarm. Therefore, an intelligent mine production safety event early warning method that can fuse multi-source information, quantify risk levels, and realize self-evolution and closed-loop control is urgently needed to overcome the shortcomings of the above-mentioned prior art.

[0003] Chinese patent publication No. CN114331151A discloses a mine underground mining and transportation operation environment risk early warning method, comprising the following steps: obtaining risk perception data and associated parameter data of a group of continuous sampling periods in a region to be warned; there are multiple regions to be warned in the entire mine area; data filtering is performed on the risk perception data to generate effective risk perception data; the regional risk value of each region to be warned is calculated according to the effective risk perception data, and a multi-parameter risk matrix of the entire mine area is constructed; the multi-parameter risk matrix of the entire mine area is input into a risk level judgment model, and the risk warning level of the operation environment of the entire mine area is output. By monitoring the risk perception data such as explosive gas, dust, toxic and harmful gas and other risk perception data in the operation environment, a risk matrix is established, and the total danger level of the entire mine area is determined.

[0004] It can be seen that the mine underground mining and transportation operation environment risk early warning method has the problems that due to the rust of the supporting device of the rock mass, the structure dynamic characteristics of the supporting device change and additional noise are caused due to rust during the vibration monitoring process when rock mass instability occurs, and when the rust or other attached pollutants similar in material characteristics to the rust fall off onto the sensor monitoring interface or the structure surface of the supporting device, the falling off of the rust is misjudged as the falling off of the rust area due to the instability of the falling off of the rust, thereby causing the failure of the adaptive adjustment of the alarm threshold for monitoring vibration. SUMMARY

[0005] To address this, the present invention provides an intelligent early warning method for mining production safety incidents that integrates large-scale model reasoning. This method overcomes the problems in existing technologies, such as the alteration of the structural dynamic characteristics and additional noise of rock mass support devices due to corrosion during vibration monitoring of rock mass instability, and the failure of adaptive adjustment of the alarm threshold for vibration monitoring due to the instability of rust or other similar material adhering to the sensor monitoring interface or the structural surface of the support device when rust or other similar materials fall off.

[0006] To achieve the above objectives, this invention provides an intelligent early warning method for mining production safety incidents that integrates large-scale model reasoning, comprising: Surface images and infrared feature images of the steel structure of the support device under the rock in the mining area were collected respectively; All suspected point features of damage were identified based on the surface image; Based on the percentage of overlapping areas of the maximum change in the dot infrared feature regions of the suspected damage as the camera distance changes, false noise caused by attached contaminants and potential real corrosion points are determined, and the area where the potential real corrosion points are located is determined as the target damage area. Obtain a set of false noise points whose straight-line distances to the target damage area satisfy a preset straight-line distance condition, and extract the infrared feature maps of the false noise points and the target damage area from the set of false noise points respectively; The exposure time of the infrared imager is adjusted based on the overlap between the first infrared feature of the false noise in the infrared feature map and the second infrared feature of the target damage area, and the false noise that meets the preset overlap condition is removed from the surface image to form a damage state dataset. Based on the damage state dataset and the initial digital model of the support device, the structural stiffness reduction factor and the updated natural frequency of the support device are calculated. When the vibration sensor detects a rock mass vibration event, the alarm threshold for rock mass vibration is adjusted based on the updated natural frequency. Based on the adjusted alarm threshold, the rock mass will continue to be monitored to achieve intelligent early warning of rock mass stability and production safety events in the mining area.

[0007] Furthermore, the step of identifying all suspected point features of damage based on the surface image includes: Obtain the contrast of each pixel region in the surface image; The contrast ratio is compared with the preset contrast ratio; If the contrast ratio is greater than the preset contrast ratio, then the pixel region is identified as a suspected point-like feature of damage. The contrast ratio is the ratio of the local brightness standard deviation of a pixel region to the local brightness mean of the pixel region.

[0008] Further, the step of determining false noise caused by adhering contaminants and potential real corrosion points based on the percentage of overlapping area of ​​the maximum change region of the point-like infrared feature region of the suspected damage with varying camera distance, and determining the area where the potential real corrosion points are located as the target damage area, includes: Obtain an infrared image of the pixel region of the suspected damage's dot-like features; The largest region enclosed by all connected pixels in the infrared image whose surface temperature value differs from the average temperature of the background area by a preset temperature difference is defined as the suspected damaged infrared feature region of the point feature. If the proportion of the suspected damaged infrared feature area with the largest change as the camera changes is less than the preset proportion, it is determined to be a false noise caused by attached contaminants. If the proportion of the suspected damaged infrared feature area that changes the most with the camera is greater than a preset proportion, it is determined to be a potential real corrosion point.

[0009] Furthermore, the percentage of the area of ​​maximum change in the suspected damaged infrared feature region as the camera changes is the ratio of the overlapping area of ​​the suspected damaged infrared feature region and the area of ​​maximum change to the area of ​​the suspected damaged infrared feature region, wherein, The region of maximum change is the union of the suspected damaged infrared feature regions calculated at all sampling distances when the camera is moved to the first distance and the second distance, respectively, and the infrared images corresponding to the point features at the first distance and the second distance are acquired.

[0010] Further, the exposure time of the infrared imager is adjusted based on the overlap between the first infrared feature of false noise points in the infrared feature map and the second infrared feature of the target damage area, and false noise points that meet the preset overlap condition are removed from the surface image to form a damage state dataset, including: Extract the infrared feature maps of the false noise points and the target damage area from the set of false noise points whose straight-line distance is less than a preset straight-line distance; The overlap between the first infrared feature and the second infrared feature is compared with a preset overlap. If the overlap is greater than a preset overlap, the exposure time of the infrared imager is increased, and false noise points with an overlap greater than the preset overlap are removed from the surface image to form a damage state dataset.

[0011] Furthermore, the overlap is the ratio of the number of shared characteristic peaks of the first infrared feature and the second infrared feature within a unit wavenumber range to the total number of characteristic peaks of the first infrared feature within the same unit wavenumber range.

[0012] Furthermore, the exposure time is positively correlated with the degree of overlap.

[0013] Furthermore, the step of calculating the structural stiffness reduction factor and updated natural frequency of the support device based on the damage state dataset and the initial digital model of the support device includes: Based on the proportion of rust area in the damage state dataset and the initial structural stiffness value of the support device, the structural stiffness reduction factor of the support device is calculated. Based on the structural stiffness reduction factor and the initial natural frequency value of the support device, the updated natural frequency of the support device is calculated, wherein... The structural stiffness reduction factor is the product of the ratio of the corrosion area to the total surface area of ​​the support device and the ratio of the corrosion depth to the thickness of the support device. The updated natural frequency is the difference between the initial natural frequency value of the support device and the frequency offset, wherein, The frequency offset is the difference between the product of the initial natural frequency value and the square root of the structural stiffness reduction factor.

[0014] Furthermore, when the vibration sensor detects a rock mass vibration event, adjusting the alarm threshold for rock mass vibration based on the updated natural frequency includes: The updated intrinsic frequency is compared with the preset intrinsic frequency; If the updated natural frequency is less than the preset natural frequency, the alarm threshold for rock mass vibration is reduced.

[0015] Furthermore, the alarm threshold is positively correlated with the updated inherent frequency.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention acquires surface images and infrared feature images of the steel structure of the support device under the rock in the mining area. When the local contrast of a pixel area in the surface image is greater than a preset contrast, it indicates that the area has a significant difference from the surrounding background in terms of texture, color, or brightness. Therefore, isolated points or small areas with such high contrast characteristics are identified as pixel areas with suspected damage point-like features. By acquiring infrared images of pixel areas with suspected damage point-like features, when the difference between the surface temperature value and the average temperature of the background area in the infrared image is greater than a preset temperature difference, it indicates that there is a thermal property anomaly between the point-like feature area and the surrounding intact steel structure, such as anomalies in heat capacity or thermal conductivity. This may be caused by damage or attachments. When the camera is moved to a first distance and a second distance, the apparent size and geometric contour of the point-like features in the image will change with the change in shooting distance. Infrared images corresponding to the point-like features at the first distance and the second distance are acquired respectively. The area represented by this feature in the infrared image will also change accordingly. Therefore, by calculating the union of these infrared feature areas at different distances, the union area of ​​the suspected damaged infrared feature areas at all sampling distances can be calculated, which is the maximum range of change that it can exhibit under the current observation conditions. When the proportion of this union area of ​​the suspected damaged infrared feature areas changing with the camera is less than the preset proportion, it indicates that the thermal anomaly area at this point has a weak correlation with the boundary of its physical shape, and its thermal characteristics are unstable, which is consistent with the characteristics of two-dimensional surface contaminants, such as dust and water stains. When this proportion is greater than the preset proportion, it indicates that the thermal anomaly area at this point highly coincides with the boundary of its physical three-dimensional shape, the heat source is stable and coupled with the physical entity, which is consistent with the characteristics of real rust pits. By collecting surface images and infrared feature images of the steel structure of the support device under the rock in the mining area, false noise caused by attached contaminants and potential real rust points are identified, further eliminating the misjudgment error caused by the visual similarity between attached contaminants and rust.

[0017] Furthermore, this invention extracts the infrared feature spectra of false noise points with a linear distance less than a preset linear distance from the target damaged area. When the linear distance is less than the preset linear distance, it indicates that the false noise points and the target damaged area are spatially close, making it difficult to distinguish whether the false noise point is an independent attachment or debris detached from the target damaged area during initial screening. By comparing the overlap between the first infrared feature of the false noise point and the second infrared feature of the target damaged area with a preset overlap, when the overlap is greater than the preset overlap, it indicates that the chemical composition of the false noise point is highly similar to the corrosion products of the target damaged area, and it is highly likely to be debris detached from the corrosion area. Using contaminants with the same composition further enhances the signal-to-noise ratio and feature resolution of infrared images, increasing the difference in absorption rates in specific bands between the first and second infrared features. This allows for better screening of false noise and differences in corrosion products within the target damage area. False noise with an overlap greater than a preset overlap is removed from the surface image to form a damage state dataset. At this point, the surface image has removed attached contaminants with the same composition as the corrosion area but belonging to the non-structural body, and only contains the true damage features under the influence of structural corrosion. Therefore, the accuracy of subsequent structural stiffness reduction calculations and natural frequency updates based on the damage state dataset is improved, ultimately making the filtering and reconstruction of rock mass vibration signals and safety early warning more reliable.

[0018] Furthermore, this invention calculates the structural stiffness reduction factor of the support device based on the proportion of rust area in the damage state dataset and the initial structural stiffness value of the support device. The proportion of rust area represents the degree of material loss on the steel structure surface, thereby allowing the calculation of the structural stiffness reduction factor of the support device determined by the weakening of the material cross-section. Based on the structural stiffness reduction factor and the initial natural frequency value of the support device, the updated natural frequency of the support device is calculated. The relationship between the structural stiffness reduction factor and the initial natural frequency value of the support device is based on the basic principle of structural dynamics, namely, the natural frequency of a structure is proportional to the square root of its stiffness, further improving the accuracy of the dynamic characteristics of the support device in the digital twin model.

[0019] Furthermore, when the vibration sensor detects a rock mass vibration event, the present invention compares the updated natural frequency with the preset natural frequency, i.e. the natural frequency under healthy conditions. When the updated natural frequency is less than the preset natural frequency, it indicates that the support structure has decreased stiffness due to corrosion, its dynamic characteristics have changed, and it is more sensitive to vibration. This is judged as a deterioration of the structural health status. At this time, by lowering the alarm threshold for rock mass vibration, the system's ability to capture real rock mass instability precursor signals under structural corrosion conditions is further improved, thereby significantly improving the timeliness and reliability of the early warning. Attached Figure Description

[0020] Figure 1 This is an overall flowchart of the intelligent early warning method for mining production safety incidents that integrates large model reasoning, as described in this embodiment of the invention. Figure 2 This is a flowchart illustrating the identification of point features of suspected damage in an intelligent early warning method for mining production safety incidents that integrates large-model reasoning, as described in an embodiment of the present invention. Figure 3 This is a flowchart illustrating the determination of the target damage area in the intelligent early warning method for mining production safety incidents that integrates large model reasoning, as described in an embodiment of the present invention. Figure 4 This is a flowchart illustrating the formation of a damage state dataset in the intelligent early warning method for mining production safety incidents that integrates large model reasoning, as described in an embodiment of the present invention. Detailed Implementation

[0021] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0022] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0023] It will be understood by those skilled in the art that, unless explicitly stated otherwise, the singular forms “a,” “an,” and “the” used herein may also include the plural forms. It should be further understood that the word “comprising” as used herein means the presence of features, integers, steps, operations, elements / components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements / components. It should be understood that when we say a module is “connected” or “coupled” to another module, it can be directly connected or coupled to the other module, or there may be intermediate units. Furthermore, the terms “connected” or “coupled” as used herein can include wireless connections or wireless coupling.

[0024] Please see Figure 1 The diagram shown is an overall flowchart of the intelligent early warning method for mining production safety incidents integrating large-scale model reasoning according to an embodiment of the present invention. The intelligent early warning method for mining production safety incidents integrating large-scale model reasoning according to an embodiment of the present invention includes: Step S1: Collect surface images and infrared feature images of the steel structure of the support device under the rock in the mining area; Step S2: Identify all suspected point features of damage based on the surface image; Step S3: Based on the percentage of overlapping areas of the maximum change area of ​​the dot infrared feature region of the suspected damage with the camera distance, determine the false noise caused by the attached contaminants and the potential real corrosion points, and determine the area where the potential real corrosion points are located as the target damage area. Step S4: Obtain a set of false noise points whose straight-line distances to the target damage area satisfy a preset straight-line distance condition, and extract the infrared feature maps of the false noise points and the target damage area from the set of false noise points respectively. Step S5: Adjust the exposure time of the infrared imager based on the overlap between the first infrared feature of the false noise in the infrared feature map and the second infrared feature of the target damage area, and remove the false noise that meets the preset overlap condition from the surface image to form a damage state dataset. Step S6: Based on the damage state dataset and the initial digital model of the support device, calculate the structural stiffness reduction factor and the updated natural frequency of the support device. Step S7: When the vibration sensor detects a rock mass vibration event, adjust the alarm threshold for rock mass vibration based on the updated natural frequency. Step S8: Based on the adjusted alarm threshold, continue to monitor the rock mass to complete the intelligent early warning of rock mass stability and production safety events in the mining area.

[0025] Specifically, explosion-proof industrial cameras installed on the roof of mine roadways in the mining area are used to collect surface images of the steel structure of the support device under the rock in the mining area.

[0026] Specifically, infrared thermal imagers installed on the roof of mine roadways in the mining area are used to collect infrared feature images of the steel structure supporting the rock under the mine.

[0027] In practice, this invention acquires surface images and infrared feature images of the steel structure of the support device under the rock in the mining area. When the local contrast of a pixel area in the surface image is greater than a preset contrast, it indicates that the area has a significant difference from the surrounding background in terms of texture, color, or brightness. Therefore, isolated points or small areas with such high contrast characteristics are identified as pixel areas with suspected damage point features. By acquiring infrared images of pixel areas with suspected damage point features, when the difference between the surface temperature value and the average temperature of the background area in the infrared image is greater than a preset temperature difference, it indicates that there is a thermal property anomaly between the point feature area and the surrounding intact steel structure, such as anomalies in heat capacity or thermal conductivity. This may be caused by damage or attachments. When the camera is moved to a first distance and a second distance, the apparent size and geometric contour of the point feature in the image will change with the shooting distance. Infrared images corresponding to the point feature at the first distance and the second distance are acquired respectively. At this time, the feature in the infrared... The areas represented in the image will also change accordingly. Therefore, by calculating the union of infrared feature areas at different distances, the union area of ​​suspected damaged infrared feature areas at all sampling distances can be calculated, which is the maximum range of change that it can exhibit under the current observation conditions. When the proportion of this union area of ​​suspected damaged infrared feature areas changing with the camera is less than the preset proportion, it indicates that the thermal anomaly area at this point has a weak correlation with the boundary of its physical shape, and its thermal characteristics are unstable, which is consistent with the characteristics of two-dimensional surface contaminants, such as dust and water stains. When this proportion is greater than the preset proportion, it indicates that the thermal anomaly area at this point highly coincides with the boundary of its physical three-dimensional shape, the heat source is stable and coupled with the physical entity, which is consistent with the characteristics of real rust pits. By collecting surface images and infrared feature images of the steel structure of the support device under the rock in the mining area, false noise caused by attached contaminants and potential real rust points are identified, further eliminating the misjudgment error caused by the visual similarity between attached contaminants and rust.

[0028] Please see Figure 2 The flowchart shown is a process for identifying point features of suspected damage in an intelligent early warning method for mining production safety incidents that integrates large-model reasoning, according to an embodiment of the present invention. The step of identifying point features of all suspected damage based on the surface image includes: Obtain the contrast of each pixel region in the surface image; The contrast ratio is compared with the preset contrast ratio; If the contrast ratio is greater than the preset contrast ratio, then the pixel region is identified as a suspected point-like feature of damage. The contrast ratio is the ratio of the local brightness standard deviation of a pixel region to the local brightness mean of the pixel region.

[0029] In practice, based on factors affecting contrast in the lighting environment of the mining area, including but not limited to mine lamp spots, shadows, and dust coverage, the spatial resolution of the explosion-proof industrial camera on the roof of the mine roadway is selected. The center of the pixel area is used as the reference, and a preset straight-line distance is extended in all directions to cover a local area of ​​the pixel area.

[0030] Optionally, the preset straight-line distance is selected based on the area of ​​the pixel region of the suspected damage point feature, and the selectable range of the preset straight-line distance is [2 pixels, 5 pixels].

[0031] Preferably, the preset straight-line distance in a specific embodiment of the present invention is 3 pixels.

[0032] Optionally, under local region calculation conditions in the pixel area, the preset contrast can be selected in the range of [15%, 40%].

[0033] Preferably, the preset contrast ratio is 25%.

[0034] Those skilled in the art will understand that the selectable range and preferred embodiment of the preset contrast are based on the area of ​​the pixel region of the suspected damage point features and the local window size. Those skilled in the art can adaptively adjust the preset contrast according to the specific application scenario.

[0035] Please see Figure 3 The flowchart shown is a process for determining the target damage area in the intelligent early warning method for mining production safety incidents that integrates large model inference according to an embodiment of the present invention. The process involves determining false noise points caused by adhering contaminants and potential real corrosion points based on the percentage of overlapping areas of the maximum change in the point-like infrared feature regions of the suspected damage as the camera distance changes, and then determining the area where the potential real corrosion points are located as the target damage area. This includes: Obtain an infrared image of the pixel region of the suspected damage's dot-like features; The largest region enclosed by all connected pixels in the infrared image whose surface temperature value differs from the average temperature of the background area by a preset temperature difference is defined as the suspected damaged infrared feature region of the point feature. If the proportion of the suspected damaged infrared feature area with the largest change as the camera changes is less than the preset proportion, it is determined to be a false noise caused by attached contaminants. If the proportion of the suspected damaged infrared feature area that changes the most with the camera is greater than a preset proportion, it is determined to be a potential real corrosion point.

[0036] Optionally, provided that the spatial resolution of the infrared imager is not less than 640×480 pixels, the preset temperature difference value can be selected within the range of [2℃, 5℃]; provided that the distance difference of the camera movement is not less than 20% of the detection distance, the preset percentage can be selected within the range of [60%, 85%].

[0037] Preferably, the preset temperature difference is 2.5℃ in the preferred embodiment; the preset percentage is 70% in the preferred embodiment.

[0038] Specifically, the percentage of the area of ​​maximum change in the suspected damaged infrared feature region as the camera changes is the ratio of the overlapping area between the suspected damaged infrared feature region and the area of ​​maximum change to the area of ​​the suspected damaged infrared feature region, where, The region of maximum change is the union of the suspected damaged infrared feature regions calculated at all sampling distances when the camera is moved to the first distance and the second distance, respectively, and the infrared images corresponding to the point features at the first distance and the second distance are acquired.

[0039] Optionally, the selectable range of the first distance and the second distance is selected based on the straight-line distance between the surface of the camera and the infrared feature area of ​​the suspected damage. The selectable range of the first distance is [1.0m, 1.7m], and the selectable range of the second distance is [1.8m, 2.5m].

[0040] Preferably, the first distance is the median of the straight-line distance between the surface of the camera and the infrared feature area of ​​the suspected damage. That is, when the detection distance is in the range of 1.0m to 2.5m, the preferred embodiment of the first distance is 1.65m, and the preferred embodiment of the second distance is 2.0m.

[0041] Those skilled in the art will understand that the selectable range and preferred embodiments of the first and second distances are based on the straight-line distance between the surface of the camera and the infrared feature area of ​​the suspected damage. Those skilled in the art can adaptively adjust the first and second distances according to specific application scenarios.

[0042] Please see Figure 4 The flowchart shown is a process for forming a damage state dataset in the intelligent early warning method for mining production safety incidents that integrates large model inference according to an embodiment of the present invention. The process involves adjusting the exposure time of the infrared imager based on the overlap between the first infrared feature of false noise points in the infrared feature spectrum and the second infrared feature of the target damage area, and removing false noise points that meet preset overlap conditions from the surface image to form a damage state dataset. This includes: Extract the infrared feature maps of the false noise points and the target damage area from the set of false noise points whose straight-line distance is less than a preset straight-line distance; The overlap between the first infrared feature and the second infrared feature is compared with a preset overlap. If the overlap is greater than a preset overlap, the exposure time of the infrared imager is increased, and false noise points with an overlap greater than the preset overlap are removed from the surface image to form a damage state dataset.

[0043] Optionally, the absolute value of the temperature fluctuation in the mine environment is less than 5℃, and the dust concentration in the mine is less than 50mg / m³. 3 Under stable environmental conditions, the preset overlap range is [85%, 95%].

[0044] Preferably, the preferred embodiment with a preset overlap ratio is 90%.

[0045] In implementation, when the overlap is greater than the preset overlap value by less than 1%, the exposure time of the infrared imager is adjusted to 1.1 times the current exposure time of the infrared imager. When the overlap is greater than the preset overlap value by more than 1%, the exposure time of the infrared imager is increased by 0.05% for every 0.05% exceeding 1%. In a specific embodiment, the current overlap is 93%, and the current exposure time of the infrared imager is 100ms. The increased exposure time of the infrared imager is 100ms × 1.1 × (1 + 2.0%) = 112.2ms. When the calculated exposure time has more than two decimal places, it is rounded to one decimal place, i.e., 112.2ms.

[0046] In practice, this invention extracts the infrared feature spectra of false noise points with a linear distance less than a preset linear distance from the target damaged area. When the linear distance is less than the preset linear distance, it indicates that the false noise points and the target damaged area are spatially close, making it difficult to distinguish whether the false noise point is an independent attachment or debris detached from the target damaged area during initial screening. By comparing the overlap between the first infrared feature of the false noise point and the second infrared feature of the target damaged area with a preset overlap, when the overlap is greater than the preset overlap, it indicates that the chemical composition of the false noise point is highly similar to the corrosion products of the target damaged area, and it is highly likely to be debris detached from the corrosion area. Using contaminants with the same composition further enhances the signal-to-noise ratio and feature resolution of infrared images, increasing the difference in absorption rates in specific bands between the first and second infrared features. This allows for better screening of false noise and differences in corrosion products within the target damage area. False noise with an overlap greater than a preset overlap is removed from the surface image to form a damage state dataset. At this point, the surface image has removed attached contaminants with the same composition as the corrosion area but belonging to the non-structural body, and only contains the true damage features under the influence of structural corrosion. Therefore, the accuracy of subsequent structural stiffness reduction calculations and natural frequency updates based on the damage state dataset is improved, ultimately making the filtering and reconstruction of rock mass vibration signals and safety early warning more reliable.

[0047] Specifically, the overlap is the ratio of the number of shared characteristic peaks of the first infrared feature and the second infrared feature within a unit wavenumber range to the total number of characteristic peaks of the first infrared feature within the same unit wavenumber range.

[0048] Specifically, the exposure time is positively correlated with the degree of overlap.

[0049] Specifically, the calculation of the structural stiffness reduction factor and the updated natural frequency of the support device based on the damage state dataset and the initial digital model of the support device includes: Based on the proportion of rust area in the damage state dataset and the initial structural stiffness value of the support device, the structural stiffness reduction factor of the support device is calculated. Based on the structural stiffness reduction factor and the initial natural frequency value of the support device, the updated natural frequency of the support device is calculated, wherein... The structural stiffness reduction factor is the product of the ratio of the corrosion area to the total surface area of ​​the support device and the ratio of the corrosion depth to the thickness of the support device. The updated natural frequency is the difference between the initial natural frequency value of the support device and the frequency offset, wherein, The frequency offset is the difference between the product of the initial natural frequency value and the square root of the structural stiffness reduction factor.

[0050] Specifically, the damage status dataset includes, but is not limited to, corrosion area, corrosion depth, corrosion type, corrosion location coordinates, corrosion distribution density, and data confidence level.

[0051] Specifically, the initial digital model of the support device includes, but is not limited to, the three-dimensional geometric model of the support device, material property parameters, mechanical boundary conditions, initial structural stiffness value, initial natural frequency value, and material constitutive relation.

[0052] In implementation, this invention calculates the structural stiffness reduction factor of the support device based on the proportion of rust area in the damage state dataset and the initial structural stiffness value of the support device. The proportion of rust area represents the degree of material loss on the steel structure surface, thereby allowing the calculation of the structural stiffness reduction factor of the support device determined by the weakening of the material cross-section. Based on the structural stiffness reduction factor and the initial natural frequency value of the support device, the updated natural frequency of the support device is calculated. The relationship between the structural stiffness reduction factor and the initial natural frequency value of the support device is based on the fundamental principle of structural dynamics, namely, the natural frequency of a structure is proportional to the square root of its stiffness, further improving the accuracy of the dynamic characteristics of the support device in the digital twin model.

[0053] Specifically, when the vibration sensor detects a rock mass vibration event, adjusting the alarm threshold for rock mass vibration based on the updated natural frequency includes: The updated intrinsic frequency is compared with the preset intrinsic frequency; If the updated natural frequency is less than the preset natural frequency, the alarm threshold for rock mass vibration is reduced.

[0054] Optionally, based on the typical fundamental frequency distribution characteristics of commonly used U-shaped steel arch frames and I-beam supports in mines under healthy conditions, the preset range of natural frequencies is [20Hz, 45Hz].

[0055] In a preferred embodiment of the U-shaped steel arch frame and I-beam support made of Q235 steel, a preset natural frequency of 35Hz is used.

[0056] Those skilled in the art will understand that the range of preset inherent frequencies and preferred embodiments are based on the material of the support device, and those skilled in the art can make adaptive adjustments to the preset inherent frequencies according to specific application scenarios.

[0057] In implementation, when the updated natural frequency is less than the preset natural frequency by less than 5 Hz, the alarm threshold for rock vibration is adjusted to 95% of the current alarm threshold. When the updated natural frequency is less than the preset natural frequency by more than 5 Hz, the alarm threshold is reduced by 0.05% for every 1 Hz exceeding 2 Hz. In a specific embodiment, the current updated natural frequency is 28 Hz and the current alarm threshold is 100 mm / s. The reduced alarm threshold is 100 mm / s × 95% × (1 - 0.1%) = 94.905 mm / s. When the calculated exposure time has more than three decimal places, it is rounded to two decimal places, i.e., 94.91 mm / s.

[0058] Specifically, the alarm threshold is positively correlated with the updated inherent frequency.

[0059] In practice, this invention compares the updated natural frequency with the preset natural frequency (i.e., the natural frequency under healthy conditions) when the vibration sensor detects a rock mass vibration event. When the updated natural frequency is less than the preset natural frequency, it indicates that the support structure has suffered a decrease in stiffness due to corrosion, its dynamic characteristics have changed, and it is more sensitive to vibration. This is judged as a deterioration in the structural health status. At this time, by lowering the alarm threshold for rock mass vibration, the system's ability to capture real rock mass instability precursor signals under structural corrosion conditions is further improved, thereby significantly improving the timeliness and reliability of the early warning.

[0060] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A method for intelligent early warning of mining production safety incidents integrating large-scale model reasoning, characterized in that, include: Surface images and infrared feature images of the steel structure of the support device under the rock in the mining area were collected respectively; All suspected point features of damage were identified based on the surface image; Based on the percentage of overlapping areas of the maximum change in the dot infrared feature regions of the suspected damage as the camera distance changes, false noise caused by attached contaminants and potential real corrosion points are determined, and the area where the potential real corrosion points are located is determined as the target damage area. Obtain a set of false noise points whose straight-line distances to the target damage area satisfy a preset straight-line distance condition, and extract the infrared feature maps of the false noise points and the target damage area from the set of false noise points respectively; The exposure time of the infrared imager is adjusted based on the overlap between the first infrared feature of the false noise in the infrared feature map and the second infrared feature of the target damage area, and the false noise that meets the preset overlap condition is removed from the surface image to form a damage state dataset. Based on the damage state dataset and the initial digital model of the support device, the structural stiffness reduction factor and the updated natural frequency of the support device are calculated. When the vibration sensor detects a rock mass vibration event, the alarm threshold for rock mass vibration is adjusted based on the updated natural frequency. Based on the adjusted alarm threshold, the rock mass will continue to be monitored to achieve intelligent early warning of rock mass stability and production safety events in the mining area.

2. The intelligent early warning method for mining production safety incidents based on large-scale model reasoning as described in claim 1, characterized in that, The step of identifying all suspected damage point features based on the surface image includes: Obtain the contrast of each pixel region in the surface image; The contrast ratio is compared with the preset contrast ratio; If the contrast ratio is greater than the preset contrast ratio, then the pixel region is identified as a suspected point-like feature of damage. The contrast ratio is the ratio of the local brightness standard deviation of a pixel region to the local brightness mean of the pixel region.

3. The intelligent early warning method for mining production safety incidents based on large-scale model reasoning as described in claim 2, characterized in that, The method of determining false noise caused by adhering contaminants and potential real corrosion points based on the percentage of overlapping areas of the maximum change in the point-like infrared feature regions of the suspected damage with varying camera distance, and defining the area where the potential real corrosion points are located as the target damage area, includes: Obtain an infrared image of the pixel region of the suspected damage's dot-like features; The largest region enclosed by all connected pixels in the infrared image whose surface temperature value differs from the average temperature of the background area by a preset temperature difference is defined as the suspected damaged infrared feature region of the point feature. If the proportion of the suspected damaged infrared feature area with the largest change as the camera changes is less than the preset proportion, it is determined to be a false noise caused by attached contaminants. If the proportion of the suspected damaged infrared feature area that changes the most with the camera is greater than a preset proportion, it is determined to be a potential real corrosion point.

4. The intelligent early warning method for mining production safety events based on large-scale model reasoning as described in claim 3, characterized in that, The percentage of the area of ​​maximum change in the suspected damaged infrared feature region as the camera changes is the ratio of the overlapping area between the suspected damaged infrared feature region and the area of ​​maximum change to the area of ​​the suspected damaged infrared feature region, where, The region of maximum change is the union of the suspected damaged infrared feature regions calculated at all sampling distances when the camera is moved to the first distance and the second distance, respectively, and the infrared images corresponding to the point features at the first distance and the second distance are acquired.

5. The intelligent early warning method for mining production safety events based on large-scale model reasoning as described in claim 4, characterized in that, The exposure time of the infrared imager is adjusted based on the overlap between the first infrared feature of false noise points in the infrared feature spectrum and the second infrared feature of the target damage area, and false noise points that meet the preset overlap conditions are removed from the surface image to form a damage state dataset, including: Extract the infrared feature maps of the false noise points and the target damage area from the set of false noise points whose straight-line distance is less than a preset straight-line distance; The overlap between the first infrared feature and the second infrared feature is compared with a preset overlap. If the overlap is greater than a preset overlap, the exposure time of the infrared imager is increased, and false noise points with an overlap greater than the preset overlap are removed from the surface image to form a damage state dataset.

6. The intelligent early warning method for mining production safety incidents based on large-scale model reasoning as described in claim 5, characterized in that, The overlap is the ratio of the number of shared characteristic peaks of the first infrared feature and the second infrared feature within a unit wavenumber range to the total number of characteristic peaks of the first infrared feature within the same unit wavenumber range.

7. The intelligent early warning method for mining production safety events based on large-scale model reasoning as described in claim 6, characterized in that, The exposure time is positively correlated with the degree of overlap.

8. The intelligent early warning method for mining production safety events based on large-scale model reasoning as described in claim 7, characterized in that, The calculation of the structural stiffness reduction factor and updated natural frequency of the support device based on the damage state dataset and the initial digital model of the support device includes: Based on the proportion of rusted area in the damage state dataset and the initial structural stiffness value of the support device, the structural stiffness reduction factor of the support device is calculated. Based on the structural stiffness reduction factor and the initial natural frequency value of the support device, the updated natural frequency of the support device is calculated, wherein... The structural stiffness reduction factor is the product of the ratio of the corrosion area to the total surface area of ​​the support device and the ratio of the corrosion depth to the thickness of the support device. The updated natural frequency is the difference between the initial natural frequency value of the support device and the frequency offset, wherein, The frequency offset is the difference between the product of the initial natural frequency value and the square root of the structural stiffness reduction factor.

9. The intelligent early warning method for mining production safety incidents based on large-scale model reasoning as described in claim 8, characterized in that, When the vibration sensor detects a rock mass vibration event, the alarm threshold for rock mass vibration is adjusted based on the updated natural frequency, including: The updated intrinsic frequency is compared with the preset intrinsic frequency; If the updated natural frequency is less than the preset natural frequency, the alarm threshold for rock mass vibration is reduced.

10. The intelligent early warning method for mining production safety incidents based on large-scale model reasoning according to claim 9, characterized in that, The alarm threshold is positively correlated with the updated inherent frequency.

Images