A Vision-Based Decision-Making System and Method for Predicting and Replacing Tunnel Boring Machine Cutting Teeth Based on Life Prediction

Abstract

This invention provides a vision-based recognition-based system and method for predicting and replacing cutting teeth on a tunneling machine, relating to the field of intelligent mining equipment condition monitoring technology. The system includes: a cutting tooth recognition module that identifies pixel-level masks of each cutting tooth in a pre-acquired image of the tunneling machine's cutting head based on an instance segmentation neural network; a health assessment module that extracts the geometric features of each cutting tooth based on the pixel-level masks and generates a health score for each tooth; a lifespan prediction module that predicts the remaining lifespan of each cutting tooth based on a time-series prediction model, using the health score and corresponding cumulative cutting load data; a lifespan comparison module that compares the remaining lifespan with a preset lifespan threshold to obtain a comparison result; and a replacement decision module that generates a cutting tooth replacement command based on the comparison result. This system enables independent tracking and condition assessment of individual cutting teeth, avoiding waste caused by batch replacement of entire cycles and reducing cutting tooth consumption costs.

Description

Technical Field

[0001] This invention relates to the field of intelligent equipment condition monitoring technology in mining, and in particular to a decision-making system and method for predicting and replacing the life of cutting teeth of a tunneling machine based on visual recognition. Background Technology

[0002] Cantilever roadheaders are widely used in coal mines and hard rock tunneling operations. The cutting teeth on their cutting heads, as key wear parts that directly contact the rock mass, bear the core function of breaking the rock. Due to the complex underground working conditions and varying rock types, the loads borne by the cutting teeth at different positions differ significantly, resulting in highly uneven wear rates.

[0003] Currently, there are several methods for monitoring and maintaining cutting teeth, including whole-round replacement, manual visual inspection, multi-sensor monitoring, and physical signal monitoring. The traditional whole-round replacement strategy ignores the wear differences between individual cutting teeth, resulting in the premature discarding of many teeth still capable of effective cutting, leading to unnecessary material waste. Manual visual inspection is highly subjective and lacks precision, making it difficult to detect minute chipping or internal cracks. This can easily lead to severely worn cutting teeth continuing to operate, reducing cutting efficiency, causing severe vibrations in the cutting arm, accelerating fatigue damage to core components such as bearings and reducers, and even causing safety accidents.

[0004] In recent years, although some studies have attempted to monitor tool condition through indirect signals such as vibration, current, and temperature, these methods rely on macroscopic behavioral signals of the entire machine or subsystem, making it impossible to accurately locate individual cutting teeth. Furthermore, they are susceptible to interference from factors such as changes in geological conditions and operating techniques, lacking intuitiveness and interpretability. Monitoring based on special physical signals such as acoustic emission is adept at detecting internal hidden damage, but it is insufficient for quantitative assessment of surface geometric wear, and the sensor installation requirements are stringent, making it unsuitable for large-scale daily inspections. Existing machine vision technologies mostly focus on lithological identification of the working face or large-scale target detection, failing to achieve individualized, continuous, and automated condition tracking and life prediction for dozens to hundreds of cutting teeth on the cutting head. Summary of the Invention

[0005] This invention provides a vision recognition-based tunneling machine cutting tooth life prediction and replacement decision system and method to solve the technical problem that the existing technology cannot accurately predict the life of a single cutting tooth.

[0006] On one hand, the present invention provides a vision recognition-based tunneling machine cutting tooth life prediction and replacement decision system, comprising: The cutting tooth recognition module is used to identify the pixel-level mask of each cutting tooth in a pre-acquired image of the cutting head of a tunneling machine based on an instance segmentation neural network. The health assessment module is used to extract the geometric morphological features of each cut tooth based on the pixel-level mask and generate a health score for each cut tooth. The life prediction module is used to predict the remaining life of each cutting tooth based on the time-series prediction model, according to the health score of each cutting tooth and the corresponding cumulative cutting load data. The lifespan comparison module is used to compare the remaining lifespan with a preset lifespan threshold to obtain a comparison result. The replacement decision module is used to generate a tooth replacement instruction based on the comparison results; The decision display module is used to visually display the cutting tooth position information and operation instructions corresponding to the replacement command.

[0007] Optionally, the step of extracting the geometric morphological features of each cut tooth based on the pixel-level mask and generating a health score for each cut tooth includes: Based on the pixel-level mask, the wear area ratio, chipping index, and symmetry deviation of each cutting tooth are determined; A health score for the cutting tooth is generated based on the wear area ratio, the chipping index, and the symmetry deviation.

[0008] Optionally, generating a health score for the cutting tooth based on the wear area ratio, the chipping index, and the symmetry deviation includes: H = 100 - (α × W + β × C + γ × D); Where H is the health score; W is the wear area ratio; C is the chipping index; D is the symmetry deviation; α, β, and γ are their respective weighting coefficients, and α+β+γ=1.

[0009] Optionally, the prediction of the remaining service life of each cutting tooth based on the time-series prediction model, according to the health score of each cutting tooth and the corresponding cumulative cutting load data, includes: The health score sequence of each cutting tooth is concatenated with the cumulative cutting load sequence to form a two-dimensional matrix; A sliding window is used to extract historical data with a preset time step from the two-dimensional matrix; Based on the extracted historical data, a Long Short-Term Memory (LSTM) network is used to predict the remaining service life of the cutting teeth.

[0010] Optionally, the preset lifespan threshold includes a safety threshold and an economic threshold, and the safety threshold is less than the economic threshold; generating a cutter replacement command based on the comparison result includes: A replacement instruction is generated when the remaining service life is less than or equal to the safety threshold. When the remaining useful life is greater than the safety threshold and less than or equal to the economic threshold, a replacement recommendation instruction is generated. When the remaining useful life is greater than the economic threshold, it is marked as an observation state.

[0011] Optionally, the instance segmentation neural network is an improved YOLOv11n instance segmentation model, and the improvement of the improved YOLOv11n instance segmentation model includes: Replace the C3k2 module in the backbone network with the C3k2_SAConv module; A lightweight multi-scale spatial attention module, MSAM-Lite, is introduced into the feature fusion path of the neck network. Replace the CIoU loss function with the PioU loss function.

[0012] Optionally, the cumulative cut load data is obtained in the following way: Monitor the dynamic load signal of the cutting motor; Based on the motor dynamic load signal, the instantaneous cutting power consumption is determined; Based on the three-dimensional model of the cutting head and the installation radius and instantaneous position angle of each cutting tooth, the load distribution coefficient of each cutting tooth is determined. Based on the load distribution coefficient, the instantaneous cutting power consumption is decomposed into each cutting tooth to obtain the equivalent individual power consumption of each cutting tooth. The equivalent individual power consumption is accumulated over time to generate cumulative cutting load data corresponding to each cutting tooth.

[0013] Optionally, the cutting tooth is pre-set with a unique identification mark, which is an ArUco mark embedded in the non-direct working surface area of ​​the cutting tooth; The identification mark has a side length of 2.5 mm, an embedding depth of 0.3 to 0.5 mm, and is coated with a tungsten carbide wear-resistant coating.

[0014] Optionally, the cut-off head image is acquired in the following ways: When the tunneling machine is stopped or the cutting head is rotating at a speed not exceeding a preset speed, a camera mounted on the cutting arm of the tunneling machine captures images of the cutting head at a frame rate not lower than a preset frequency; wherein, the images of the cutting head include the identification mark; The acquired cropping head images are denoised, contrast enhanced, and lens distortion corrected.

[0015] On the other hand, the present invention also provides a vision-based method for predicting and replacing the lifespan of tunneling machine cutting teeth, the method being applied to the vision-based tunneling machine cutting tooth lifespan prediction and replacement decision system described in any of the above claims, the method comprising: A neural network based on instance segmentation is used to identify the pixel-level mask of each cutting tooth in a pre-acquired image of the cutting head of a tunneling machine. Based on the pixel-level mask, the geometric morphological features of each cutting tooth are extracted to generate a health score for each cutting tooth. Based on the time-series prediction model, the remaining service life of each cutting tooth is predicted according to the health score of each cutting tooth and the corresponding cumulative cutting load data. The remaining service life is compared with a preset service life threshold to obtain the comparison result; Based on the comparison results, a cutting tooth replacement command is generated; The tooth position information and operation instructions corresponding to the replacement command are displayed visually.

[0016] This invention provides a vision-based tunneling machine cutting tooth life prediction and replacement decision system and method. The health assessment module extracts the geometric features of each cutting tooth based on pixel-level masks to generate a health score for each tooth. The life prediction module predicts the remaining service life of each cutting tooth based on a time-series prediction model, using the health score and corresponding cumulative cutting load data. The life comparison module compares the remaining service life with a preset life threshold to obtain a comparison result. The replacement decision module generates a cutting tooth replacement command based on the comparison result. This system enables independent tracking and status assessment of individual cutting teeth, avoiding waste caused by batch replacement of entire laps and reducing cutting tooth consumption costs. The decision display module visualizes the cutting tooth location information and operation instructions corresponding to the replacement command, reducing the subjectivity of manual inspection and the technical threshold for maintenance personnel, and improving the intelligence level of tunneling operations and the accuracy of replacement operations. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the structure of the tunneling machine cutting tooth life prediction and replacement decision system based on visual recognition provided in an embodiment of the present invention; Figure 2 This is a flowchart illustrating the vision recognition-based method for predicting and replacing the life of tunneling machine cutting teeth, as provided in an embodiment of the present invention. Figure 3 This is a schematic diagram of the position of the industrial camera provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of the improved SMP-YOLO instance segmentation network provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the process for scoring the health of cutting teeth according to an embodiment of the present invention; Figure 6This is a schematic diagram of the input and output of the LSTM lifetime prediction model provided in an embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this 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 this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0020] Figure 1 This is a schematic diagram of the structure of the tunneling machine cutting tooth life prediction and replacement decision system based on visual recognition provided in an embodiment of the present invention.

[0021] See Figure 1 The vision recognition-based tunneling machine cutting tooth life prediction and replacement decision system may include a cutting tooth recognition module 101, a health assessment module 102, a life prediction module 103, a life comparison module 104, a replacement decision module 105, and a decision display module 106.

[0022] The cutting tooth recognition module 101 is used to identify the pixel-level mask of each cutting tooth in a pre-acquired image of the cutting head of a tunneling machine based on an instance segmentation neural network.

[0023] Each cutting tooth is pre-set with a unique identification mark, which is an ArUco mark embedded in the non-direct working surface area of ​​the cutting tooth. The identification mark has a side length of 2.5mm, an embedding depth of 0.3 to 0.5mm, and is coated with a tungsten carbide wear-resistant coating.

[0024] The non-direct working surface area refers to the part of the cutting tooth that is far from the cutting edge and the main impact area, to avoid the marking falling off or deforming due to scratching or impact during the cutting operation. ArUco markings are generated based on the OpenCV standard dictionary, with an internal 4×4 binary encoded grid, providing more than 100 unique IDs for permanent binding of individual cutting teeth. The tungsten carbide wear-resistant coating is deposited through a supersonic flame spraying process, with a thickness of 20–30 μm and a hardness of HRA90 or higher. It is metallurgically bonded to the cutting tooth substrate, maintaining the black and white contrast of the marking pattern while resisting dust erosion, metal scratches, and water vapor corrosion, ensuring the long-term legibility of the markings in the high-wear environment downhole.

[0025] The cutter head images were acquired in the following ways: When the tunneling machine 310 is stopped or the cutting head is rotating at a speed not exceeding a preset speed, a camera mounted on the cutting arm 330 of the tunneling machine acquires images of the cutting head at a frame rate not lower than a preset frequency; wherein, the images of the cutting head may include identification tags. The acquired cropping head images are denoised, contrast enhanced, and lens distortion corrected.

[0026] Specifically, the preset rotation speed can be 10 rpm. The camera is an explosion-proof industrial camera 340, such as... Figure 3 As shown. The preset frequency is generally 10Hz or 15Hz. For example, the explosion-proof industrial camera 340 is a 5-megapixel intrinsically safe explosion-proof industrial camera. Its lens optical axis is directly aligned with the center of the cutting disk, and the distance between the front end of the camera and the end face of the cutting head 320 is 600mm (selectable range is 500–700mm), ensuring that the entire cutting disk area falls completely within the field of view. To overcome the problems of insufficient underground lighting and coal dust scattering, a ring of white LED light source (color temperature 5500K, power 15W) is integrated around the camera. It has IP68 protection rating and Ex d I Mb explosion-proof certification. The light source triggers a flash 10ms before image acquisition, providing instantaneous high-brightness illumination, suppressing motion blur and enhancing the contrast of identification marks and cutting tooth contours. The acquired multi-frame images are denoised, contrast-enhanced, and lens distortion-corrected. The multi-frame local images are then stitched together into a corrected image covering the entire cutting disk using a cylindrical projection model. Noise reduction employs a non-local means algorithm to suppress LED flicker and sensor thermal noise; contrast enhancement utilizes the CLAHE (Constrained Contrast Adaptive Histogram Equalization) algorithm to improve the visibility of serrated edges and identification markers; lens distortion correction uses pre-calibrated camera intrinsic parameters and distortion coefficients to perform radial and tangential distortion correction on each frame. All raw images, corrected images, and processing logs are stored locally on the solid-state drive of the explosion-proof industrial control computer downhole, eliminating the need to upload them to the ground or cloud. Motion blur is avoided by limiting image acquisition to a stopped or low-speed rotation state.

[0027] The instance segmentation neural network is an improved YOLOv11n instance segmentation model. Improvements to the YOLOv11n instance segmentation model can include: Replace the C3k2 module in the backbone network with the C3k2_SAConv module; A lightweight multi-scale spatial attention module, MSAM-Lite, is introduced into the feature fusion path of the neck network. Replace the CIoU loss function with the PioU loss function.

[0028] Based on the improved YOLOv11n SMP-YOLO instance segmentation model, see [link / reference]. Figure 4The original C3k2 module in the backbone network was replaced with the C3k2_SAConv module. This module utilizes a mechanism that incorporates switchable dilated convolutions to enhance the model's ability to perceive features at different wear scales of the cutting teeth. A lightweight multi-scale spatial attention module (MSAM-Lite) was introduced into the feature fusion path of the neck network. This module recalibrates the spatial dimensions of the feature map, making the network more focused on the cutting edge of the cutting teeth and the ArUco identification marker region. The original CIoU loss function was replaced with the PioU loss function to improve the overall detection performance of the model. After training and testing, the improved model can stably output high-precision pixel-level localization and reliable identification information of the cutting teeth under typical coal mine dust interference.

[0029] The health assessment module 102 is used to extract the geometric features of each cutting tooth based on the pixel-level mask and generate a health score for each cutting tooth.

[0030] See Figure 5 Based on the pixel-level mask, the geometric morphological features of each cut tooth are extracted to generate a health score for each cut tooth, which may include: Based on pixel-level masks, the wear area ratio, chipping index, and symmetry deviation of each cutting tooth are determined. A health score for the cutting teeth is generated based on the wear area ratio, chipping index, and symmetry deviation.

[0031] Based on the wear area ratio, chipping index, and symmetry deviation, a health score for the cutting tooth is generated, which may include the following formula (1): H=100-(α×W+β×C+γ×D)(1); Where H is the health score; W is the wear area ratio; C is the chipping index; D is the symmetry deviation; α, β, and γ are their respective weighting coefficients, and α+β+γ=1.

[0032] The wear area ratio is the proportion of the mask pixel area of ​​the current effective cutting edge region of the cutting tooth relative to the standard area of ​​the new state, as shown in the following formula (2): (2); in, The standard mask area of ​​the cutting edge of this type of cutting tooth in its factory state is pre-stored in the system database through calibration images or 3D CAD model projections. This is the pixel area of ​​the corresponding region extracted from the segmentation mask in the current detection period, obtained after ArUco marker pose correction.

[0033] The chipping index is used to characterize the severity of local breakage or missing parts of the cutting edge of the cutting tooth. It is defined as the proportion of the missing pixel area in the sensitive zone at the front end of the cutting edge to the theoretical total area of ​​the sensitive zone, as shown in the following formula (3): (3); in, The standard mask representing a brand new cut tooth. This indicates the currently detected truncated mask. FSB This represents a binary mask for the sensitive zone at the tip of the cutting edge. This indicates the calculation of the pixel area of ​​the binary mask.

[0034] Symmetry deviation is used to measure the symmetry distortion of the overall geometry of the cutting tooth. It is obtained by calculating the angle between the current mask principal axis direction and the cutting tooth mounting reference direction. Specifically, firstly, principal component analysis (PCA) is performed on the mask to fit its principal direction vector, and then the cutting tooth spatial pose calculated by ArUco markers is combined to determine the theoretical symmetry axis. The absolute value of the angle between the two is the symmetry deviation, as shown in the following formula (4): (4); in, Indicates the current mask. This represents the mirror image of the current mask obtained along the standard axis of symmetry. This represents the self-mirror intersection ratio of the new cutting teeth, typically taken as 0.98. For intersection, union, and comparison.

[0035] After normalizing the wear area ratio, chipping index, and symmetry deviation to the interval [0,1], they were substituted into the health scoring model. In this embodiment, based on 300 sets of full-life-cycle monitoring data of cutting teeth, multiple linear regression was used to fit the relationship between remaining life and various damage indicators, and finally α=0.65, β=0.25, and γ=0.10 were selected. It can accurately capture the minute wear, local chipping, and overall deformation of the cutting edge area of ​​the cutting teeth, avoiding the problems of strong subjectivity and low accuracy of manual judgment; the wear area ratio, chipping index, and symmetry deviation characterize the degree of damage of the cutting teeth from three dimensions: area loss, edge integrity, and shape symmetry, respectively, improving the accuracy of cutting tooth condition assessment.

[0036] The life prediction module 103 is used to predict the remaining life of each cutting tooth based on the time-series prediction model, according to the health score of each cutting tooth and the corresponding cumulative cutting load data.

[0037] Based on a time-series prediction model, the remaining service life of each cutting tooth is predicted according to its health score and corresponding cumulative cutting load data. This can include: The health score sequence of each cutting tooth is concatenated with the cumulative cutting load sequence to form a two-dimensional matrix; A sliding window is used to extract historical data at a preset time step from a two-dimensional matrix; Based on the extracted historical data, a Long Short-Term Memory (LSTM) network is used to predict the remaining service life of the cutting teeth.

[0038] Specifically, the health score sequence and the cumulative cut load sequence can be used as different columns in a two-dimensional matrix. The health score sequence reflects the degradation trend of the cutter's geometry, while the cumulative cut load sequence reflects the cumulative energy input borne by the cutter. A sliding window is used to extract historical data from the two-dimensional matrix for the most recent preset time step (e.g., the last 5 time steps), where each time step corresponds to one shutdown detection cycle. Based on the extracted historical data, a Long Short-Term Memory (LSTM) network is used as a time-series prediction model to predict the remaining useful life (RUL) of the cutter. Specifically, this LSTM model adopts a two-layer stacked structure, with 64 hidden units in each layer, using the tanh activation function, followed by a fully connected layer (128, 64, 1) to output a scalar RUL value. The loss function is the SmoothL1Loss, which is robust to outliers, and the optimizer is Adam with a learning rate of 1×10⁻⁶. 3 .

[0039] like Figure 6 As shown, the health score sequence H, taking the most recent 5 time steps, can include H(t), H(t-1), H(t-2), H(t-3), and H(t-4). The load sequence L can include L(t), L(t-1), L(t-2), L(t-3), and L(t-4). The LSTM processes a sliding window sequence of length 5. At time step t-4, H(t-4) and L(t-4) are used; at time step t-3, H(t-3) and L(t-3) are used; at time step t-2, H(t-2) and L(t-2) are used; at time step t-1, H(t-1) and L(t-1) are used; and at time step t, H(t) and L(t) are used. The LSTM network, through its internal memory chain, sequentially analyzes the health and load data at the 5 consecutive time points, learns patterns from historical degradation trajectories, and predicts the final remaining useful life (RUL).

[0040] The aforementioned model takes the historical health status and load sequence of the cutting tooth as input, and can learn the dynamic law of wear evolution over time and the accelerating effect of load on wear, thereby outputting an independent prediction of the remaining service life of the cutting tooth. Using the full life cycle data of 200 failed cutting teeth in the historical database, the initial model is trained. When a cutting tooth enters the recommended replacement state, if the subsequent actual running time deviates from the prediction by more than 20%, incremental learning is triggered for that individual tooth to improve the prediction accuracy. The model is quantized using the TensorRT engine and runs on an explosion-proof industrial control computer in the well, with a single inference time of less than 15ms.

[0041] The lifespan comparison module 104 is used to compare the remaining lifespan with a preset lifespan threshold to obtain a comparison result.

[0042] The preset lifespan threshold can be set according to actual conditions.

[0043] The replacement decision module 105 is used to generate a tooth replacement instruction based on the comparison results.

[0044] Specifically, the preset lifespan threshold may include a safety threshold and an economic threshold, with the safety threshold being less than the economic threshold; based on the comparison result, a cutter replacement instruction is generated, which may include: A replacement instruction is generated when the remaining service life is less than or equal to the safety threshold. When the remaining useful life is greater than the safety threshold and less than or equal to the economic threshold, a replacement recommendation instruction is generated. When the remaining useful life is greater than the economic threshold, it is marked as an observation state.

[0045] The safety threshold ranges from 1 to 3 hours, preferably 2 hours; the economic threshold ranges from 6 to 10 hours, preferably 8 hours. When the remaining useful life (RUL) is less than or equal to the safety threshold, it indicates a risk of sudden breakage of the cutting tooth, which may damage the cutting head or cause a safety accident, thus generating an immediate replacement instruction. When the remaining useful life is greater than the safety threshold but less than or equal to the economic threshold, it indicates that the cutting tooth is nearing the end of its useful life but has not yet reached a dangerous level. In this case, if a matching planned shutdown window exists in the future (e.g., the window end time is greater than or equal to the current time + RUL), a recommended replacement instruction is generated; otherwise, the cutting tooth is marked as being under observation. When the remaining useful life is greater than the economic threshold, it indicates that continued use is still economically beneficial, and premature replacement would result in resource waste, thus being marked as being under observation. This approach maximizes the utilization value of the cutting tooth while ensuring equipment safety, avoiding unnecessary premature replacement.

[0046] The decision display module 106 is used to visually display the cutting tooth position information and operation instructions corresponding to the replacement command.

[0047] Specifically, visualization can be achieved through a human-computer interaction terminal. The terminal, such as an explosion-proof industrial tablet or AR glasses, is used to register the cutting head's 3D model with a real-time image or 3D model via SLAM positioning, achieving millimeter-level spatial overlay of replacement guidance. In practice, when the operator approaches the cutting head to prepare for replacement, the terminal activates the visual-inertial SLAM system. It continuously acquires RGB-D images of the cutting head area using a front-facing camera, extracts stable geometric features on the cutting disc, matches them with corresponding features in the 3D model, and uses the PnP (Perspective-n-Point) algorithm to calculate the 6-DOF pose of the current camera relative to the cutting head coordinate system. After registration, the 3D position of the cutting tooth to be replaced is projected onto the image plane, obtaining its pixel coordinates on the screen. This information is then overlaid and displayed in the form of highlights, color coding, or labels. Specifically, the terminal surrounds the cutting tooth to be replaced with a flashing red border on the screen, simultaneously displaying the cutting tooth identifier, remaining service life value, and replacement type (e.g., immediate replacement or recommended replacement) operation guidance. All AR rendering and SLAM calculations are completed locally on the terminal, without relying on the network. The 3D model and replacement instructions are synchronized from the explosion-proof industrial control computer via encrypted USB or underground LAN, without being uploaded to the cloud. The equipment has a strong light readability mode, adapting to environments with insufficient roadway lighting or reflective surfaces. The visual display method lowers the technical threshold for maintenance personnel and improves the efficiency and accuracy of replacement operations.

[0048] In this embodiment, the health assessment module 102 extracts the geometric features of each cutting tooth based on pixel-level masks and generates a health score for each cutting tooth; the life prediction module 103 predicts the remaining service life of each cutting tooth based on a time-series prediction model, according to the health score of each cutting tooth and the corresponding cumulative cutting load data; the life comparison module 104 compares the remaining service life with a preset life threshold to obtain a comparison result; the replacement decision module 105 generates a cutting tooth replacement command based on the comparison result; this achieves independent tracking and status assessment of individual cutting teeth, avoiding waste caused by batch replacement of the entire circle and reducing the cost of cutting tooth consumption; the decision display module 106 visualizes the cutting tooth position information and operation instructions corresponding to the replacement command, reducing the subjectivity of manual inspection and the technical threshold for maintenance personnel, and improving the intelligence level of tunneling operations and the accuracy of replacement operations.

[0049] In one embodiment of this specification, the cumulative cut load data is obtained in the following manner: Monitor the dynamic load signal of the cutting motor; Determine the instantaneous cutting power consumption based on the dynamic load signal of the motor; Based on the three-dimensional model of the cutting head and the installation radius and instantaneous position angle of each cutting tooth, the load distribution coefficient of each cutting tooth is determined. Based on the load distribution coefficient, the instantaneous cutting power consumption is decomposed into each cutting tooth to obtain the equivalent individual power consumption of each cutting tooth. The equivalent individual power consumption is accumulated over time to generate cumulative cutting load data corresponding to each cutting tooth.

[0050] In this embodiment, the total output torque and speed of the cutting motor are monitored in real time, and the instantaneous cutting power consumption of the motor is calculated as shown in the following formula (5): (5); in, T(t) Represents the total output torque. ω(t) Represents rotational speed. Representative moment Instantaneous cutoff power consumption.

[0051] Based on the three-dimensional model of the cutting head, a load distribution coefficient related to the installation radius and instantaneous position angle is defined for the k-th cutting tooth. The equivalent individual power consumption borne by the cutting tooth in the time interval Δt is calculated as shown in the following formula (6): (6); in, η_k(t) Indicates the load distribution factor; This represents the equivalent individual power consumption. The time sampling interval is denoted as .

[0052] Integrating over all operation times yields the cumulative cutting load sequence uniquely bound to that cutting tooth ID.

[0053] The cumulative equivalent cutting power consumption corresponding to the cutting tooth ID can be obtained from the cutting motor monitoring unit and used as an external load factor. If the current health score deviates significantly from the historical trend (e.g., the health score is abnormally high under high load), a manual review alarm is triggered. The system obtains the cumulative cutting hardness integral value bound to the cutting tooth ID from the external lithology identification system via the RS485 interface. For any cutting tooth, the system continuously maintains two synchronized time-series data: a health score sequence, with a sampling interval of each shutdown detection cycle; and a cumulative cutting load sequence, which represents the equivalent cutting power consumption borne by the cutting tooth during each detection.

[0054] Based on the same general inventive concept, this invention also protects a vision-based method for predicting and replacing the lifespan of tunneling machine cutting teeth. Figure 2 This is a flowchart illustrating the vision recognition-based tunneling machine cutter life prediction and replacement decision method provided in this embodiment of the invention. The vision recognition-based tunneling machine cutter life prediction and replacement decision method provided by this invention will be described below. The vision recognition-based tunneling machine cutter life prediction and replacement decision method described below can be referred to in conjunction with the vision recognition-based tunneling machine cutter life prediction and replacement decision system described above.

[0055] See Figure 2 A vision-based method for predicting and replacing the life of tunnel boring machine cutting teeth may include the following steps.

[0056] Step 201: Based on the instance segmentation neural network, identify the pixel-level mask of each cutting tooth in the pre-acquired cutting head image of the tunneling machine.

[0057] Step 202: Based on the pixel-level mask, extract the geometric morphological features of each cutting tooth and generate a health score for each cutting tooth.

[0058] Step 203: Based on the time-series prediction model, predict the remaining service life of each cutting tooth according to the health score of each cutting tooth and the corresponding cumulative cutting load data.

[0059] Step 204: Compare the remaining service life with the preset service life threshold to obtain the comparison result.

[0060] Step 205: Generate a cutting tooth replacement command based on the comparison results.

[0061] Step 206: Visualize the tooth position information and operation instructions corresponding to the replacement command.

[0062] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units.

[0063] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as OM / AM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0064] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A visual recognition-based roadheader pick life prediction and replacement decision system, characterized in that, include: The cutting tooth recognition module is used to identify the pixel-level mask of each cutting tooth in a pre-acquired image of the cutting head of a tunneling machine based on an instance segmentation neural network. The health assessment module is used to extract the geometric morphological features of each cut tooth based on the pixel-level mask and generate a health score for each cut tooth. The life prediction module is used to predict the remaining life of each cutting tooth based on the time-series prediction model, according to the health score of each cutting tooth and the corresponding cumulative cutting load data. The lifespan comparison module is used to compare the remaining lifespan with a preset lifespan threshold to obtain a comparison result. The replacement decision module is used to generate a tooth replacement instruction based on the comparison results; The decision display module is used to visually display the cutting tooth position information and operation instructions corresponding to the replacement command.

2. The visual recognition based roadheader pick life prediction and replacement decision system according to claim 1, characterized in that, The step of extracting the geometric morphological features of each cut tooth based on the pixel-level mask and generating a health score for each cut tooth includes: Based on the pixel-level mask, the wear area ratio, chipping index, and symmetry deviation of each cutting tooth are determined; A health score for the cutting tooth is generated based on the wear area ratio, the chipping index, and the symmetry deviation.

3. The tunneling machine cutting tooth life prediction and replacement decision system based on visual recognition according to claim 2, characterized in that, The process of generating a health score for the cutting tooth based on the wear area ratio, the chipping index, and the symmetry deviation includes: H = 100 - (α × W + β × C + γ × D); Where H is the health score; W is the wear area ratio; C is the chipping index; D is the symmetry deviation; α, β, and γ are their respective weighting coefficients, and α+β+γ=1.

4. The tunneling machine cutting tooth life prediction and replacement decision system based on vision recognition according to claim 1, characterized in that, The time-series prediction model predicts the remaining service life of each cutting tooth based on its health score and corresponding cumulative cutting load data, including: The health score sequence of each cutting tooth is concatenated with the cumulative cutting load sequence to form a two-dimensional matrix; A sliding window is used to extract historical data with a preset time step from the two-dimensional matrix; Based on the extracted historical data, a Long Short-Term Memory (LSTM) network is used to predict the remaining service life of the cutting teeth.

5. The tunneling machine cutting tooth life prediction and replacement decision system based on vision recognition according to claim 1, characterized in that, The preset lifespan threshold includes a safety threshold and an economic threshold, and the safety threshold is less than the economic threshold; the step of generating a cutter replacement instruction based on the comparison result includes: A replacement instruction is generated when the remaining service life is less than or equal to the safety threshold. When the remaining useful life is greater than the safety threshold and less than or equal to the economic threshold, a replacement recommendation instruction is generated. When the remaining useful life is greater than the economic threshold, it is marked as an observation state.

6. The tunneling machine cutting tooth life prediction and replacement decision system based on visual recognition according to claim 1, characterized in that, The instance segmentation neural network is an improved YOLOv11n instance segmentation model, and the improvements to the improved YOLOv11n instance segmentation model include: Replace the C3k2 module in the backbone network with the C3k2_SAConv module; A lightweight multi-scale spatial attention module, MSAM-Lite, is introduced into the feature fusion path of the neck network. Replace the CIoU loss function with the PioU loss function.

7. The tunneling machine cutting tooth life prediction and replacement decision system based on vision recognition according to claim 1, characterized in that, The cumulative cut load data is obtained through the following methods: Monitor the dynamic load signal of the cutting motor; Based on the motor dynamic load signal, the instantaneous cutting power consumption is determined; Based on the three-dimensional model of the cutting head and the installation radius and instantaneous position angle of each cutting tooth, the load distribution coefficient of each cutting tooth is determined. Based on the load distribution coefficient, the instantaneous cutting power consumption is decomposed into each cutting tooth to obtain the equivalent individual power consumption of each cutting tooth. The equivalent individual power consumption is accumulated over time to generate cumulative cutting load data corresponding to each cutting tooth.

8. The tunneling machine cutting tooth life prediction and replacement decision system based on visual recognition according to claim 1, characterized in that, The cutting tooth is pre-set with a unique identification mark, which is an ArUco mark embedded in the non-direct working surface area of ​​the cutting tooth; The identification mark has a side length of 2.5 mm, an embedding depth of 0.3 to 0.5 mm, and is coated with a tungsten carbide wear-resistant coating.

9. The tunneling machine cutting tooth life prediction and replacement decision system based on visual recognition according to claim 8, characterized in that, The cutter head images were acquired in the following ways: When the tunneling machine is stopped or the cutting head is rotating at a speed not exceeding a preset speed, a camera mounted on the cutting arm of the tunneling machine captures images of the cutting head at a frame rate not lower than a preset frequency; wherein, the images of the cutting head include the identification mark; The acquired cropping head images are denoised, contrast enhanced, and lens distortion corrected.

10. A method for predicting and replacing the life of tunnel boring machine cutting teeth based on visual recognition, characterized in that, The method is applied to the vision recognition-based tunneling machine cutter life prediction and replacement decision system according to any one of claims 1 to 9, and the method includes: A pixel-level mask for each cutting tooth in a pre-acquired image of a tunneling machine's cutting head is identified using an instance segmentation neural network. Based on the pixel-level mask, the geometric morphological features of each cutting tooth are extracted to generate a health score for each cutting tooth. Based on the time-series prediction model, the remaining service life of each cutting tooth is predicted according to the health score of each cutting tooth and the corresponding cumulative cutting load data. The remaining service life is compared with a preset service life threshold to obtain the comparison result; Based on the comparison results, a cutting tooth replacement command is generated; The tooth position information and operation instructions corresponding to the replacement command are displayed visually.

Images