Image recognition-based rotary drill bit wear state detection method and system

Abstract

This invention belongs to the field of image processing technology, specifically relating to a method and system for detecting the wear state of rotary drilling bits based on image recognition. The method includes: acquiring scalar brightness field and global potential flow gradient magnitude field data of the drill bit surface image; calculating the local potential well smoothing index and potential flow divergence coefficient based on the scalar brightness and potential flow gradient magnitude within a local window of each pixel; performing a top-hat transformation on the scalar brightness field to obtain the entity protrusion weight of each pixel, and weighting it with the local potential well smoothing index and potential flow divergence coefficient to obtain the degradation retention weight; mapping all pixels to graph network nodes, calculating the walk transition probability of a node moving to any of its neighboring nodes to construct a global graph transition probability matrix; performing PageRank iteration based on the transition probability matrix until convergence to obtain the retention probability of each pixel, thereby determining the wear state of the drill bit. This invention effectively suppresses mud interference and improves detection accuracy.

Description

Technical Field

[0001] This invention relates to the field of image processing technology. More specifically, this invention relates to a method and system for detecting the wear condition of rotary drilling bits based on image recognition. Background Technology

[0002] With the large-scale development of underground space, ultra-long hollow piles are widely used in deep and large foundation pit projects. During the drilling stage, the dynamic cutting reaction force generated by uneven strata can easily cause the alloy drill teeth of rotary drill bits to suffer severe physical wear or breakage. Once the drill teeth fail, it will not only reduce drilling efficiency, but also cause the drill bit to be subjected to uneven load and deflection, which in turn leads to extremely serious pile position deviation quality accidents.

[0003] Currently, visual detection of drill bit wear largely relies on the PageRank algorithm. Its core logic is to construct network nodes from pixels in an image and statically assign the probability of virtual particles moving between pixels based solely on the two-dimensional geometric distance between nodes and the similarity of surface brightness. This attempts to make particles aggregate and find areas of significant wear with strong contrast.

[0004] However, in the scenario of rotary drilling for ultra-long hollow piles, when the drill bit is pulled out of the hole, the surface will inevitably be covered with thick and unevenly flowing mud and sediment. The random fluid characteristics of the mud and the reflective surface water film will produce a large number of false bright spots, which will cover up the true mechanical wear pattern of the drill teeth. This will cause the underlying walking transfer network of the traditional PageRank algorithm to be severely disturbed: the virtual walking particles are easily intercepted by large areas of highly reflective continuous mud, while the wear areas that are actually flattened are smoothed by the mud, resulting in a very low probability of transfer, and thus being completely missed by the algorithm. This leads to extremely low accuracy in detecting the wear state and cannot provide reliable early warning for engineering construction. Summary of the Invention

[0005] To address the technical problem that the presence of thick mud causes the traditional PageRank algorithm's walk-transfer network to fail, resulting in inaccurate drill bit wear detection, this invention provides solutions in the following aspects.

[0006] In a first aspect, the present invention provides a method for detecting the wear state of rotary drilling bits based on image recognition, comprising: acquiring scalar brightness field data of an image of the drill bit surface, and calculating the potential current gradient magnitude field of the entire surface to obtain the potential current gradient magnitude of each pixel; calculating a local potential well smoothing index based on the difference between the maximum and mean values ​​of the scalar brightness within a local window of each pixel; calculating the variance of the potential current gradient magnitude of all pixels within a local window of each pixel to obtain the potential current divergence coefficient; performing a top-hat transformation on the scalar brightness field to obtain the solid protrusion weight of each pixel; and calculating the local potential well smoothing index and the potential current gradient magnitude of each pixel. The flow divergence coefficient is used to weight the entity protrusion weights and calculate the degradation retention weights. All pixels are mapped to nodes in the graph network. Based on the distance between each node and any of its neighboring nodes, and the degradation retention weights of the corresponding pixels of the neighboring nodes, the pixel walk transition probability of each node to any of its neighboring nodes is calculated, and a full graph transition probability matrix is ​​constructed. Based on the transition probability matrix, PageRank iteration is performed until convergence to obtain the retention probability of each pixel. Based on the statistical result that the normalized value of the retention probability of all pixels is greater than a set threshold, the wear state of the drill bit is determined.

[0007] This invention evaluates the physical wear and breakage characteristics of drill teeth by calculating the local potential well smoothing index and potential flow divergence coefficient; and extracts the solid protrusion weight by combining morphological top cap transformation to calculate the degradation retention weight. The degradation retention weight is then used to reconstruct the underlying Markov transition probability matrix, breaking the static transition rule and forcibly guiding wandering particles to converge at the real defect, effectively suppressing the background noise of complex mud movement, and significantly improving the detection accuracy under extreme working conditions.

[0008] Preferably, the local potential well smoothing exponent satisfies the expression: In the formula, For the first Local potential well smoothness index of each pixel; , For the first The maximum and average values ​​of all scalar brightness values ​​within a local window of each pixel; It is a natural exponential function; This is the preset amplification factor.

[0009] This invention utilizes the extreme difference in local microscopic brightness. By combining negative exponential decay with an amplification factor, when the drill bit becomes dull and the brightness distribution tends to be uniform, the exponent produces a cliff-like nonlinear surge, effectively extracting the flat features covered by mud and significantly improving the model's sensitivity to early physical wear.

[0010] Preferably, the calculation of the global current gradient magnitude field includes: using the Sobel operator to convolve and differentiate the scalar brightness field to obtain the horizontal and vertical gradients of each pixel; taking the square root of the sum of the squares of the horizontal and vertical gradients to obtain the gradient magnitude of each pixel, and constructing the global current gradient magnitude field.

[0011] Preferably, the potential flow divergence coefficient is equal to the normalized value of the variance of the potential flow gradient magnitude of all pixels within the local window of each pixel.

[0012] Preferably, obtaining the entity bump weight of each pixel includes: performing a top-hat transformation on the scalar brightness field using a preset fixed-size filter kernel to obtain a bump distribution map; and performing Gaussian filtering and minimax normalization on the bump distribution map to obtain an entity bump weight matrix, wherein the value of the corresponding pixel in the matrix is ​​the bump weight of that pixel.

[0013] This invention constructs a rigorous physical topology mask through macroscopic spatial filtering and performs morphological top-hat transformation using a filter kernel of a specific size to peel off suspected drill bit entities from the mud. This effectively eliminates false interference from large areas of flat mud and high-intensity water film reflection, ensuring that subsequent feature fusion is only performed within the actual drill bit area and reducing the background false recognition rate.

[0014] Preferably, the degradation retention weight satisfies the expression: In the formula, For the first Degradation retention weight of each pixel; For the corresponding convexity weight matrix of the entity The convexity weight value of each pixel; For the first Local potential well smoothness index of each pixel; For the first Potential current divergence coefficient of each pixel; This is the standard normalization function.

[0015] This invention integrates macroscopic masking and microscopic wear characteristics, constructs a physical topological mask through a product term, and effectively filters out background interference in non-drill tooth solid areas. At the same time, it introduces a third-order nonlinear polarization suppression mechanism, which utilizes the expansion characteristics of the power function to rapidly attenuate the low-amplitude noise generated by water film reflection, while significantly enhancing the feature weight of the real defects.

[0016] Preferably, the method for obtaining adjacent nodes includes: obtaining all nodes within the local window of each node as adjacent nodes.

[0017] Preferably, the pixel walk transition probability of each node moving to any of its neighboring nodes satisfies the expression: In the formula, For nodes Transfer to its first The pixel walkover transition probability of each adjacent node; For nodes Its first The Euclidean distance between adjacent nodes; , These are the index values ​​and total number of adjacent nodes; For nodes The The degradation retention weight of each neighboring node corresponding to a pixel; The preset distance attenuation coefficient; The preset intervention coefficient; It is a natural exponential function.

[0018] This invention uses the degenerate retention weight as an exogenous perturbation multiplier, giving real defects a strong attraction. By combining the dual factors of Euclidean distance loss and defect-targeted traction, virtual wandering particles can break through weak mud noise interference and spontaneously converge towards the real drill tooth defect area, thus solving the particle trapping problem.

[0019] Preferably, determining the wear state of the drill bit includes: calculating the percentage of pixels whose normalized retention probability is greater than a set threshold relative to the total number of pixels in the entire image; if the percentage exceeds the process safety ratio, the wear state is determined to be severe.

[0020] Secondly, the present invention provides an image recognition-based rotary drilling bit wear condition detection system, including a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned image recognition-based rotary drilling bit wear condition detection method is implemented.

[0021] By adopting the above technical solution, the image recognition-based rotary drill bit wear condition detection method is generated into a computer program and stored in a memory for loading and execution by a processor. This allows for the creation of a terminal device based on the memory and processor, facilitating its use.

[0022] The beneficial effects of this invention are as follows:

[0023] This invention extracts the physical characteristics of drill teeth that have been mechanically worn down and flattened through morphological cap transformation and local extremum drop calculation, and introduces the potential flow gradient amplitude variance feature to extract the characteristics of the drill tooth fracture region.

[0024] Further fusion of extracted wear features and cubic nonlinear polarization are performed to reconstruct the pixel transition probability matrix of the underlying random walk, giving the real defect region a strong gravitational effect, guiding wandering particles to converge towards the real defect, thereby improving the accuracy of drill bit wear condition detection. Attached Figure Description

[0025] Figure 1 This is a flowchart illustrating the image recognition-based rotary drilling bit wear condition detection method of the present invention;

[0026] Figure 2 This is a schematic diagram showing the binarized image after processing by the traditional PageRank algorithm;

[0027] Figure 3 This is the binarized result after processing according to the present invention. Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0030] This invention discloses a method for detecting the wear condition of rotary drilling bits based on image recognition, referring to... Figure 1 This includes steps S1-S5:

[0031] S1. Obtain scalar brightness field data of the drill bit surface image.

[0032] It should be noted that under harsh working conditions where the drill bit is encased in mud, RGB color information suffers from severe redundancy and interference. Tiny brightness fluctuations on the drill bit surface can directly reflect its true three-dimensional physical topology. Therefore, extracting a single-channel grayscale image can meet the feature requirements for defect detection. However, due to the significant differences in ambient light intensity at different construction periods, directly using 0... Calculating with absolute integer pixel values ​​of 255 will cause the built-in model parameters to become invalid. Therefore, it is necessary to normalize the grayscale image to map discrete absolute pixel values ​​to 0. A scalar brightness field is constructed from consecutive floating-point numbers between 1 and 2.

[0033] Specifically, camera equipment is installed at the borehole opening. The gap between the rotary drilling rig and the borehole opening when it is suspended and unloading soil is triggered to capture images of the drill bit surface. The images are then converted to grayscale to obtain a grayscale image. The grayscale value of each pixel in the grayscale image is extracted, converted to a 32-bit floating-point number, and normalized by dividing by 255 to construct a global scalar brightness field.

[0034] At this point, scalar brightness field data of the drill bit surface image has been obtained.

[0035] S2. Obtain the global potential current gradient magnitude field. Based on the scalar brightness and potential current gradient magnitude within the local window of each pixel, calculate the local potential well smoothing index and potential current divergence coefficient respectively.

[0036] It should be noted that a healthy and sharp alloy drill bit is similar to a steep mountain peak, with a huge brightness difference between the maximum brightness value of the local micro-region at its tip and the average brightness value of the surrounding area. When the drill bit is subjected to extreme cutting force and becomes dull, its tip is flattened into a flat section, resulting in a more uniform brightness distribution in the local micro-region. Therefore, by analyzing the difference in local brightness of a pixel, the state characteristics of the region where the pixel is located can be reflected.

[0037] Specifically, for any pixel, a local window with a fixed side length centered on that pixel is constructed, and the maximum and average values ​​of all scalar brightness values ​​within that local window are obtained.

[0038] It should be added that the fixed side length reflects the spatial receptive field range for capturing microstructural features. If the value is too large, it will inevitably be affected by low-brightness data from a broad background, diluting the difference between the maximum and the mean, and failing to truly reflect the smoothness of the tip. If the value is too small, it is easily affected by single high-frequency light noise points or local reflective water film patches, losing the statistically significant comparative value of the mean. Therefore, the fixed side length must be an odd number, and the value range should be 15. 35 pixels, or 25 in this embodiment of the invention, ensures that the analysis window covers the core light and shadow change zone of the micro-region at the end of the drill bit. The implementer can make a proportional dynamic adjustment based on the focal length resolution of the actual camera equipment and the imaging ratio of the drill bit in the image. For pixels that cannot obtain a complete local window, the nearest neighbor interpolation method is used to fill them in, so as to ensure the integrity and continuity of the data.

[0039] The local potential well smoothness index of any pixel is calculated based on the difference between the maximum and mean values ​​of all scalar brightness values ​​within a local window. This local potential well smoothness index satisfies the following expression:

[0040]

[0041] In the formula, For the first Local potential well smoothness index of each pixel; , For the first The maximum and average values ​​of all scalar brightness values ​​within a local window of each pixel; It is a natural exponential function; This is the preset amplification factor.

[0042] in, Reflects the first The difference in scalar brightness values ​​within a local window of a pixel, when the... When a sharp object exists within a local window of a pixel, the scalar brightness value changes significantly, resulting in a large difference in brightness, after passing through a negative exponential... After attenuation Approaching 0; while when the first When a local window containing a pixel undergoes severe physical smoothing to form a flat cross-section, the scalar brightness value distribution tends to be uniform, resulting in minimal drop. At this point, the negative exponential term... Approaching the maximum value 1, in summary, if The larger it is, the more likely it is to be the first The smoother the potential energy of the local scalar brightness value of a pixel, the more likely it is to lose its peak physical structure. The higher the probability that a pixel is located in the dull area of ​​the drill bit, the more likely it is to be in the dull area.

[0043] It should be added that the amplification factor reflects the nonlinear sensitivity of the mathematical model to the shrinking trend of local extreme value difference. In the scalar brightness field of [0,1], the local extreme value difference of a healthy sharp drill tooth is usually between 0.2 and 0.4, while the difference in the blunt region is close to 0. If the amplification factor is not introduced, the difference between the natural exponential decay results of the two is extremely small and is easily submerged by the background noise of the mud. Therefore, the amplification factor is taken in the range of (5,20), and is taken as 12 in this embodiment of the invention to ensure that when the extreme value difference is close to 0, a cliff-like exponential surge response that meets the system's expectations can be generated. The implementer can adjust it according to the response effect.

[0044] It should be further noted that in real hard rock drilling environments, in addition to the dulling caused by continuous friction, drill bits are also prone to instantaneous fracture failure when cutting through hard pebble layers. The fracture surface is often uneven and full of jagged fracture crystal phase textures. In this state, the originally smooth and continuous light and shadow potential flow on the healthy metal surface is completely destroyed, resulting in extremely violent and disordered random fluctuations in the magnitude and direction of the micro potential flow gradient. Therefore, the potential well smoothness index alone cannot identify such defects. By analyzing the discrete variance of the local potential flow gradient amplitude, the unique physical structure of the fracture surface can be captured, thereby ensuring the accuracy of wear state identification.

[0045] Specifically, to obtain the global potential current gradient magnitude field, the calculation process includes: using the Sobel operator to perform convolution and differentiation on each pixel in the scalar brightness field in the horizontal and vertical directions to obtain the horizontal potential current gradient and vertical potential current gradient of each pixel, and taking the square root of the sum of their squares to obtain the potential current gradient magnitude of each pixel, thus obtaining the global potential current gradient magnitude field.

[0046] Extract the potential flow gradient magnitude of all pixels within the local window of any pixel, and perform maximum and minimum normalization on the variance of all potential flow gradient magnitudes to obtain the potential flow divergence coefficient of any pixel.

[0047] Among them, the variance value reflects the dispersion of the potential flow gradient amplitude in the local area of ​​the pixel. If the pixel is in the healthy drill tooth area, its surface morphology is continuous and the gradient transition is smooth, resulting in a very low variance of the potential flow gradient amplitude in the local area. However, if the pixel is in the fractured area, due to the uneven physical cross-section and the chaotic optical reflection, the gradient abrupt change between adjacent pixels is extremely frequent, resulting in a very high dispersion of variance. It is believed that the drill bit surface in the area where the pixel is located is more likely to fracture.

[0048] At this point, the local potential well smoothness index and potential flow divergence coefficient of each pixel are obtained.

[0049] S3. Perform a top-hat transformation on the scalar brightness field to obtain the entity bump weight for each pixel, and then combine the local potential well smoothing index and potential flow divergence coefficient to weight the entity bump weight to obtain the degradation retention weight.

[0050] It should be noted that, considering the false characteristics of the high-intensity reflectivity of the mud film, the calculated parameters may be interfered with, leading to misidentification. From the perspective of the two-dimensional projection of the actual three-dimensional physical topology, regardless of the health of the drill bit, it appears as a macroscopic solid protrusion relative to the flat mud background. Therefore, a solid protrusion weight matrix is ​​constructed by macroscopic spatial filtering to separate the suspected drill bit solid region from the mud background, which serves as the basic mask for subsequent feature fusion. Furthermore, the target region must satisfy at least one of the two degradation characteristics of flattening or breakage. At the same time, in order to prevent the weak features of the residual water film in the healthy region from accumulating and trapping random wandering particles, a high-order nonlinear polarization suppression mechanism is introduced to significantly suppress the background interference region.

[0051] Specifically, the entity convexity weight matrix is ​​obtained through the following process: obtaining a fixed-size filter kernel; performing a morphological top-hat transformation on the scalar brightness field using the filter kernel to obtain a convexity distribution map; and applying a Gaussian filter and min-max normalization to the convexity distribution map to obtain the entity convexity weight matrix. For the corresponding convexity weight matrix of the entity The convex weight of each pixel.

[0052] It should be added that the size of the morphological filter kernel reflects the largest macroscopic spatial scale boundary in the image that can be completely contained and identified as an independent entity protrusion. In drill bit images that have undergone size normalization, the physical projection width of a single standard alloy drill tooth is typically around 60. Fluctuations within a 75-pixel range: If the kernel size is smaller than the lower limit of the actual physical pixel width, it will destroy the topological integrity of the drill tooth entity, mistakenly fragmenting the internal highlight patches into fragmented independent backgrounds; if the kernel size is too large, it will excessively introduce low-frequency fluctuation interference caused by the continuous undulation of a large area of ​​mud. Therefore, the size of the morphological filter kernel must be strictly slightly larger than the maximum projected pixel width of a single real alloy drill tooth. In this embodiment of the invention, 80×80 pixels is used to ensure that the entire drill tooth entity can be regarded as a complete topology in the top cap transformation. Implementers can adjust it according to the physical parameters of different drill bits.

[0053] Based on the local potential well smoothness index and potential flow divergence coefficient of any pixel, the convexity weight value is weighted to obtain the degradation retention weight of any pixel, which satisfies the expression:

[0054]

[0055] In the formula, For the first Degradation retention weight of each pixel; For the corresponding convexity weight matrix of the entity The convexity weight value of each pixel; For the first Local potential well smoothness index of each pixel; For the first Potential current divergence coefficient of each pixel; This is the standard normalization function.

[0056] in, Reflects the first The degree of overall physical structural degradation of a pixel, including dulling, balding, or breakage. A larger value indicates a more severe degradation. The local regions of each pixel exhibit severe loss of mechanical structure, both in terms of the smoothness of the potential energy drop and the roughness of the potential flow divergence; the product term The introduction of this method establishes a rigorous physical topological masking mechanism, ensuring that only degenerate features located within the actual drill bit body are preserved, eliminating interference from non-drill bit mud backgrounds. Furthermore, the cubic approach reflects a nonlinear polarization suppression mechanism, the logic of which is that the dispersed low-noise score generated by normal water film reflection collapses to zero after cubic polarization, while the highly concentrated score of the actual defect area remains high. In summary, The larger the value, the more significant the effect. The probability of a pixel experiencing real physical wear and tear is extremely high.

[0057] At this point, the degradation retention weight of each pixel has been obtained.

[0058] S4. Map all pixels to nodes of the graph network, calculate the pixel walk transition probability of each node to any of its neighboring nodes, and construct the full graph transition probability matrix.

[0059] It should be noted that the static rules of the traditional PageRank algorithm are prone to failure in rotary drill bit images filled with mud and water film. This causes wandering particles to be deflected or stuck by large areas of mud reflective patches, resulting in aimless full-image wandering. In order to ensure that wandering particles in the network can accurately converge to the actual drill tooth defects, the degenerate retention weights extracted above are used to reconstruct the underlying Markov transition probability matrix.

[0060] Specifically, the drill bit surface image is mapped into a two-dimensional topological graph network. Each pixel in the image is regarded as an independent node, and the spatial coordinates of each pixel in the image grid are extracted. For any node in the graph network, all adjacent nodes within its local window are obtained.

[0061] Calculate the Euclidean distance between any node and any of its neighboring nodes within its local window. Based on this distance and the degradation retention weight of the corresponding pixel of any neighboring node, calculate the pixel walk transition probability of any node in the graph network moving to any of its neighboring nodes. The pixel walk transition probability satisfies the expression:

[0062]

[0063] In the formula, For the node Transfer to its first The pixel walkover transition probability of each adjacent node; For nodes Its first The Euclidean distance between adjacent nodes; , These are the index values ​​and total number of adjacent nodes; For nodes The The degradation retention weight of each neighboring node corresponding to a pixel; The preset distance attenuation coefficient; The preset intervention coefficient; It is a natural exponential function.

[0064] in, Reflecting nodes The The ability of a neighboring node to force network flow intervention on wandering particles acts as an exogenous perturbation multiplier. The larger the value, the more severe the physical wear at the neighboring node, and the stronger the forced siphon and traction effect it exerts on the surrounding virtual particles as a defective gravitational black hole. Reflected in the combined effect of spatial distance loss and the gravitational pull of adjacent node defects, the particle's movement from the node... Inclined toward its first The potential energy driving force for the movement of adjacent nodes; Reflects the nodes The sum of the potential driving forces for transfer to all adjacent nodes within its local window is used to normalize the transfer probability within the locally connected domain; in summary, if a node The The stronger the forced traction force of the virtual particles from adjacent nodes under the targeted siphon of the central defect, the more likely it is that the particles in the graph network are drawn from the nodes. Transfer to its first The higher the pixel walk transfer probability of each adjacent node, the better, so that the virtual walking particles in the network can break through the interference of complex mud and water film with high reflectivity and spontaneously converge towards the real drill bit wear or breakage area.

[0065] It should be added that the intervention coefficient reflects the strength of the forced suppression of the traditional geometric distance walk rule by the a priori features of the defective black hole. The distance attenuation coefficient will cause an exponential decay of nearly one order of magnitude to the far nodes at the edge of the local window. To ensure that the comprehensive traction force generated by the real defects at the far end can overcome the distance advantage of the adjacent weak mud noise points, the theoretical amplification lower limit of the intervention coefficient needs to exceed 50. Considering the engineering safety margin of the extreme high reflectivity conditions on site, the value is taken as 80 in this embodiment of the invention. The implementer can adjust it based on the on-site conditions. The distance attenuation coefficient reflects the energy loss rate and basic wandering intention of wandering particles when crossing physical distances in the two-dimensional image space. If the value is too large, the distance penalty is too severe, which will cause the virtual particles to be locked in a very small spatial neighborhood and unable to perform large-scale macroscopic feature optimization. If the value is too small, the probability of particles blindly jumping over long distances is too high, which will cause the local topological details of the image to be completely ignored. Therefore, the value range is set to (0.1,1). In this embodiment of the invention, it is taken as 0.5 to ensure that while the particles have a certain neighborhood exploration activity, they are still limited by local physical topological constraints. The implementer can make corresponding balance adjustments through the neighborhood connectivity scale when actually constructing the graph network.

[0066] Thus, the pixel walk transition probabilities of any node in the graph network moving to all its neighboring nodes are obtained, and the whole graph transition probability matrix is ​​constructed.

[0067] S5. Perform PageRank iterations based on the transition probability matrix until convergence, obtain the retention probability of each pixel, and determine the wear state of the drill bit based on the statistical results of the retention probabilities of all pixels.

[0068] It should be noted that the traditional PageRank algorithm uses Markov iteration, and wandering particles will converge towards the wear area. Therefore, once the global probability distribution reaches a steady state, high-risk wear patches can be extracted and warnings can be output.

[0069] Specifically, the pixel walk transition probability matrix of the graph network is used as the underlying iterative operator. The existing PageRank global iterative algorithm is used to continuously deduce until the particle residence probability distribution in each node of the entire graph converges, and the residence probability of each pixel is extracted at this time. The residence probability of the entire graph is normalized by the maximum and minimum normalization to obtain the normalized value of the residence probability of each pixel. The percentage of pixels with a normalized residence probability greater than a set threshold is obtained. If the percentage exceeds the process safety ratio, the wear condition is determined to be severe. The system then draws a bright red box to mark the wear area corresponding to the front-end image and reminds the user to replace the tooth for maintenance in time to avoid serious hole deviation accidents.

[0070] It should be added that the threshold setting reflects the rigor of the system's defect feature truncation. If the value is too low, it is easy to mix in the wandering noise left over from the normal environment, causing false alarms; if the value is too high, the judgment condition is extremely strict, and it is easy to miss some real defects that are in the critical wear state. Since the pre-transfer matrix reconstruction has given real defects a very strong siphoning ability, the threshold in this embodiment of the invention is set to 0.6. The implementer can dynamically fine-tune it according to the sensitivity requirements of the actual identification on site. The process safety ratio is to convert the area of ​​the upper limit of wear allowed by the process into the corresponding pixel number ratio in the image space in proportion to the camera resolution calibration coefficient.

[0071] For example, Figure 2 The binary image processed by the traditional PageRank algorithm is affected by the reflection of mud and water film. In addition to some real wear signals, it is also filled with a large number of disordered scattered white noise and continuous interference, resulting in serious false alarms in the detection results.

[0072] Figure 3 The image shows the binarized result after processing according to the present invention. The background is pure black, indicating that the noise of the complex mud flow has been perfectly suppressed by the algorithm. At the same time, the image clearly shows bright white connected regions. These white regions accurately correspond to the actual dulling and breakage failure locations of the drill teeth. Compared with the random walk method of traditional algorithms, the present invention has stronger ability to suppress complex backgrounds and target mechanical defects, significantly improving the accuracy of detection.

[0073] This invention also discloses an image recognition-based rotary drilling bit wear condition detection system, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement the image recognition-based rotary drilling bit wear condition detection method according to this invention.

[0074] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

Claims

1. A method for detecting the wear condition of rotary drilling bits based on image recognition, characterized in that, include: Acquire scalar brightness field data of the drill bit surface image and calculate the global potential current gradient magnitude field, including: using the Sobel operator to convolve and differentiate the scalar brightness field to obtain the horizontal and vertical gradients of each pixel; taking the square root of the sum of the squares of the horizontal and vertical gradients to obtain the gradient magnitude of each pixel, and constructing the global potential current gradient magnitude field to obtain the potential current gradient magnitude of each pixel. The local potential well smoothness index is calculated based on the difference between the maximum and mean scalar brightness within the local window of each pixel. , ; , For the first The maximum and average values ​​of all scalar brightness values ​​within a local window of each pixel; It is a natural exponential function; The preset amplification factor is used; the variance of the potential current gradient magnitude of all pixels within the local window of each pixel is calculated to obtain the potential current divergence coefficient, which is equal to the normalized value of the variance of the potential current gradient magnitude of all pixels within the local window of each pixel. To obtain the entity bump weights for each pixel, a top-hat transform is performed on the scalar brightness field. This includes: performing a top-hat transform on the scalar brightness field using a preset fixed-size filter kernel to obtain a bump distribution map; applying a Gaussian filter to the bump distribution map and performing min-max normalization to obtain an entity bump weight matrix, where the value of each pixel in the matrix is ​​its bump weight; and weighting the entity bump weights according to the local potential well smoothing exponent and the potential flow divergence coefficient to calculate the degradation retention weight. , ; For the corresponding convexity weight matrix of the entity The convexity weight value of each pixel; For the first Potential current divergence coefficient of each pixel; The standard normalization function is used; all pixels are mapped to nodes in the graph network, and the pixel walk transition probability of each node to any of its neighboring nodes is calculated based on the distance of each node to any of its neighboring nodes and the degradation retention weight of the corresponding pixels of the neighboring nodes. , ; For nodes Its first The Euclidean distance between adjacent nodes; , These are the index values ​​and total number of adjacent nodes; For nodes The The degradation retention weight of each neighboring node corresponding to a pixel; The preset distance attenuation coefficient; Preset intervention coefficients and construct a full-map transition probability matrix; PageRank iteration is performed based on the transition probability matrix until convergence, the retention probability of each pixel is obtained, and the wear state of the drill bit is determined based on the statistical result that the normalized value of the retention probability of all pixels is greater than a set threshold.

2. The method for detecting the wear condition of rotary drilling bits based on image recognition according to claim 1, characterized in that, The methods for obtaining adjacent nodes include: Get all nodes within the local window of each node as adjacent nodes.

3. The method for detecting the wear condition of rotary drilling bits based on image recognition according to claim 1, characterized in that, The determination of the wear condition of the drill bit includes: The percentage of pixels whose normalized retention probability is greater than a set threshold is calculated out of the total number of pixels in the entire image. If the percentage exceeds the process safety ratio, the wear condition is determined to be severe.

4. A rotary drilling bit wear condition detection system based on image recognition, characterized in that, include: The processor and memory, wherein the memory stores computer program instructions that, when executed by the processor, implement the image recognition-based rotary drilling bit wear condition detection method according to any one of claims 1-3.

Images