Underground shield tunneling machine tunnel face cutter wear detection system and method

Abstract

The embodiment of the invention provides an underground shield tunneling machine tunnel face cutter wear detection system and method, and relates to the technical field of engineering machinery detection. The system comprises an unmanned aerial vehicle, a detection assembly, a transmission assembly and a data processing device. Wherein the unmanned aerial vehicle can carry the detection assembly to perform image acquisition on the tunnel face cutter of the shield tunneling machine in an underground construction area so as to obtain detection image data. And the transmission assembly transmits the detection image data and the control instruction through the optical fiber transmission line, so that the data processing equipment can identify a cutter abrasion area and defect information in the detection image data by using a cutter abrasion identification algorithm, and a structured detection result is generated. The system can perform real-time image transmission by combining the unmanned aerial vehicle with the optical fiber, adopts MAML meta-learning and an improved Swindow-Transform double-stage architecture, adapts to a small sample scene and 1080p cutter wear real-time detection, can adapt to a high-dust construction area environment with a short distance and a limited space, and improves the safety, adaptability and reliability of a detection process.

Description

Technical Field

[0001] This application relates to the field of engineering machinery testing technology, and in particular to a system and method for detecting wear of cutting tools at the working face of an underground tunnel boring machine. Background Technology

[0002] As a core component of the tunnel boring machine (TBM), the wear condition of the cutterhead directly affects construction efficiency and project costs. To ensure construction efficiency, regular inspection, maintenance, and replacement of the cutterhead are necessary. In some engineering practices, the inspection, maintenance, and replacement of the cutterheads account for approximately 1 / 4 to 1 / 3 of the total tunnel excavation time. TBM cutterhead inspection can promptly detect defects such as cutter wear and mud cake formation, preventing cascading damage, reducing maintenance costs, and minimizing project delays.

[0003] Shield tunneling cutter inspection can rely on manual methods, where inspectors enter the construction area to conduct close-range visual inspections or simple tool measurements, and judge the wear condition of the cutters based on the results. However, the shield tunneling area is a high-risk and space-constrained area, posing safety hazards, and manual inspection is inefficient and highly subjective.

[0004] Therefore, robots can be used to replace manual labor for cutterhead inspection. Wheeled or tracked robots move within the construction area and acquire visual images of the cutters using cameras and other image-gathering devices. Image analysis is then used to determine the cutterhead's wear condition. However, the muddy, complex, and space-constrained environment of the tunnel boring machine's construction area limits the robot's inspection capabilities. It struggles to fully cover the large-diameter cutterhead, particularly the top and edge areas, leading to inspection difficulties and reducing the safety, adaptability, and reliability of cutterhead inspection. Summary of the Invention

[0005] In view of this, embodiments of this application provide a system and method for detecting cutter wear at the working face of an underground tunnel boring machine, in order to solve the problems of low safety, adaptability and reliability of tunnel boring machine cutter detection.

[0006] According to a first aspect of this application, a wear detection system for the cutting tools at the working face of an underground tunnel boring machine is provided, the system comprising: The drone includes a fusion positioning unit configured to calculate the drone's positioning data using an extended Kalman filter fusion algorithm. A detection component is mounted on the UAV; the detection component includes a detection camera, a camera gimbal, and a narrow-spectrum infrared supplementary lighting module; the detection camera is mounted on the camera gimbal; the detection camera is equipped with an anti-fog and anti-fog coated lens and a dust cover; the detection camera is configured to perform image acquisition on the tunnel boring machine face cutter in the underground construction area to obtain detection image data. A transmission component is configured to transmit the detected image data and control commands; the transmission component includes an optical fiber transmission line and an optical transceiver; the optical fiber transmission line includes a detection end and a control end; the detection end of the optical fiber transmission line is connected to the UAV and the detection component via the optical transceiver. The data processing device is connected to the control terminal of the optical fiber transmission line via the optical transceiver; the data processing device is configured as follows: Acquire detection image data, the detection image data including tool images; A tool wear identification algorithm is invoked, which includes a defect detection model. This defect detection model is a neural network model trained using sample defect data. The defect detection model employs a two-stage training architecture combining MAML meta-learning and an improved Swin-Transformer. The defect detection model includes a dark channel prior dehazing module and an improved median filtering module to remove image blur and noise caused by dust. The sample defect data includes sample images containing typical tool defects and defect labels for the sample images. The defect labels include defect type labels and defect level labels. The tool wear recognition algorithm is used to identify the tool head wear area and defect information in the detection image data. The defect information includes the defect type and defect level output by the defect detection model based on the tool image. A structured detection result is generated based on the wear area of ​​the cutter head and the defect information. The structured detection result includes a defect location image, defect coordinates, defect size, and defect level.

[0007] In some embodiments, the fusion positioning unit includes an inertial navigation module, an optical flow sensor, and a laser ranging module; the inertial navigation module is configured to detect the attitude of the UAV; the optical flow sensor is configured to detect the position of the UAV by analyzing the motion patterns of pixels between consecutive image frames; the laser ranging module is configured to detect the position of the UAV by laser ranging; the data processing device is further configured to: The positioning data collected by the fusion positioning unit is obtained, and the positioning data includes the original detection data of the fusion positioning unit, the position of the UAV, and the attitude of the UAV. A construction area model is constructed based on the positioning data and the detected image data; The undetected areas are determined according to the construction area model and the detection image data; A detection control command is generated based on the undetected area, and the detection control command includes a planned path for controlling the flight of the UAV; The detection and control commands are sent to the drone so that the drone flies along the planned path.

[0008] In some embodiments, the system further includes a display terminal connected to the data processing device; the data processing device is further configured to: The interface rendering background is generated based on the construction area model and the detected image data; Based on the construction area model and the structured detection results, an algorithm marking result is generated, which includes a color-coded visual identifier. The algorithm's labeling results are overlaid onto the interface rendering background in real time to generate a result display interface; The results are displayed on the display terminal.

[0009] In some embodiments, the display terminal is a virtual reality device; the data processing device is further configured to: The system acquires interactive commands input by the user through the virtual reality device, the interactive commands including at least one of a review command, a correction command, and a confirmation command; According to the interaction instructions, the detection result information corresponding to the structured detection result is recorded. The detection result information includes the confirmed defect location image, defect coordinates, and defect level. A structured test report is generated based on the test results.

[0010] In some embodiments, the virtual reality device includes a pose sensor configured to detect real-time pose data of a user; the data processing device is further configured to: The real-time pose data and historical pose data are acquired, wherein the historical pose data is obtained by recording the real-time pose data during the user's wearing process; The motion amplitude is calculated based on the real-time pose data and the historical pose data. The motion amplitude is calculated based on the weighted sum of the differences in each dimension of the real-time pose data and the historical pose data, combined with the construction area model. If the amplitude of the movement is greater than a preset amplitude threshold, a field-of-view following instruction is generated based on the real-time pose data. The field-of-view following instruction is used to control the UAV to fly according to the real-time pose data. The vision-following command is sent to the drone.

[0011] In some embodiments, the transmission component further includes an image transmission encoder connected to the detection camera; the data processing device is further configured to: Based on the structured detection results, a region of interest is defined. The region of interest is a defect region in the tool image where the number of samples is less than a threshold, or a defect region detected for the first time, or a defect region in the tool image where the defect level is higher than a preset level threshold. A region of non-interest is determined based on the region of interest, and the region of non-interest is the area outside the region of interest in the tool image; Generate encoding optimization instructions, the encoding optimization instructions including a first encoding method for the region of interest and a second encoding method for the region of non-interest; The encoding optimization instruction is sent to the image transmission encoder so that the image transmission encoder performs image encoding on the raw image data acquired by the detection camera according to the encoding optimization instruction to generate detection image data.

[0012] In some embodiments, the data processing device is further configured to: Obtain the transmission data stream corresponding to the detected image data; A first decoding method is determined based on the first encoding method, and a second decoding method is determined based on the second encoding method; According to the region of interest, the key region image is decoded in the transmitted data stream using the first decoding method; According to the non-interest region, the second decoding method is used to decode the non-critical region image in the transmitted data stream; The key region image and the non-key region image are stitched together to generate the tool image.

[0013] In some embodiments, the data processing device is further configured to: In response to a remote control command input by the user, flight control parameters are read from the remote control command; A detection strategy is obtained, which includes tool detection position and flight restriction conditions; The flight control parameters are corrected based on the detection strategy, and flight control commands are generated based on the corrected flight control parameters. The flight control command is sent to the drone to control the drone to fly to the tool detection position according to the flight restrictions.

[0014] In some embodiments, the data processing device is further configured to: Multiple tool images are extracted from the detection image data according to the detection time; Acquire multiple positioning data points that were collected at the same detection time as the tool image; Multiple tool images are stitched together based on the positioning data to generate an overview image; Using the overview image as the rendering background, the defect location image is generated by overlaying the wear area of ​​the cutter head and the defect information in real time.

[0015] According to a second aspect of this application, a method for detecting cutter wear at the face of an underground tunnel boring machine is provided, applied to the system described in the first aspect; the method includes: Acquire detection image data, the detection image data including tool images; A tool wear identification algorithm is invoked, which includes a defect detection model. This defect detection model is a neural network model trained using sample defect data. The defect detection model employs a two-stage training architecture combining MAML meta-learning and an improved Swin-Transformer. The defect detection model includes a dark channel prior dehazing module and an improved median filtering module to remove image blur and noise caused by dust. The sample defect data includes sample images containing typical tool defects and defect labels for the sample images. The defect labels include defect type labels and defect level labels. The tool wear recognition algorithm is used to identify the tool head wear area and defect information in the detection image data. The defect information includes the defect type and defect level output by the defect detection model based on the tool image. A structured detection result is generated based on the wear area of ​​the cutter head and the defect information. The structured detection result includes a defect location image, defect coordinates, defect size, and defect level.

[0016] According to a third aspect of this application, a computer device is provided, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein the processor executes the program to implement the above-described method for detecting cutter wear at the face of an underground tunnel boring machine.

[0017] According to a fourth aspect of this application, a storage medium is provided that stores a computer program thereon, which, when executed by a processor, implements the above-described method for detecting cutter wear at the face of an underground tunnel boring machine.

[0018] Based on the above technical solutions, this application provides a system and method for detecting cutter wear on the face of an underground tunnel boring machine (TBM). The system includes a drone, a detection component, a transmission component, and a data processing device. The drone, equipped with the detection component, can acquire images of the TBM face cutter in the underground construction area to obtain detection image data. The transmission component transmits the detection image data and control commands via fiber optic transmission lines, enabling the data processing device to use a cutter wear recognition algorithm to identify wear areas and defects in the detection image data, generating structured detection results. The system can acquire images via a drone and transmit them in real-time via fiber optics, identifying wear, cracks, and other defects in the cutterhead surface detection image data according to the cutter wear recognition algorithm. The system adopts a two-stage architecture of MAML meta-learning and an improved Swin-Transformer, adapting to small sample scenarios and 1080p real-time detection requirements. It can adapt to high-dust construction environments in close-range, confined spaces, improving the safety, adaptability, and reliability of the detection process.

[0019] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0020] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 A schematic diagram of the structure of the underground shield tunneling machine face cutter wear detection system provided in this application embodiment; Figure 2 A schematic diagram of the process for detecting cutter wear at the face of an underground tunnel boring machine provided in this application embodiment; Figure 3 This is a schematic diagram illustrating the process of identifying wear areas and defect information of the cutter head provided in an embodiment of this application; Figure 4 This is a schematic diagram of the process for generating defect location images provided in an embodiment of this application; Figure 5 This is a schematic diagram of the ROI encoding optimization process provided in an embodiment of this application. Detailed Implementation

[0021] The present application will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the embodiments of the present application can be combined with each other.

[0022] In this embodiment, a tunnel boring machine (TBM) is a full-face tunneling machine integrating excavation, support, muck removal, and guidance. The TBM may include a tunneling system, a support system, a muck removal system, a propulsion system, a guidance system, and an after-sales support system. The core component of the tunneling system is the cutterhead. The cutterhead includes a cutterhead and multiple cutters mounted on it, such as roller cutters and scrapers. During construction, the cutterhead can be driven to rotate by a high-power hydraulic device or an electric motor to cut the ground and excavate the tunnel.

[0023] The wear condition of tunnel boring machine (TBM) cutters directly affects the construction efficiency and project cost. In engineering practice, the inspection, maintenance, and replacement of cutters account for approximately one-third of the total tunnel excavation time. Therefore, to ensure construction efficiency, it is necessary to regularly inspect, maintain, and replace the TBM cutters. Inspection of the cutters can promptly detect defects such as cutter wear and mud cake formation, thereby preventing cascading damage and reducing maintenance costs.

[0024] In order to inspect the tunnel boring machine cutters, some embodiments may rely on manual inspection methods, in which inspectors enter the construction area to conduct close-range visual inspection or simple tool measurement, and judge the wear condition of the cutters based on the visual inspection or tool measurement results.

[0025] During tunnel excavation, the exposed surface of the soil or rock strata to be excavated is called the working face. Located directly in front of the tunnel boring machine's cutterhead, the working face is a dynamic interface transitioning from a stable state (original strata) to a disturbed and disrupted state, bearing the original ground stress and groundwater pressure. The construction area corresponding to the working face has a relatively small spatial volume, representing a close-range, confined construction environment.

[0026] Because the tunnel boring machine (TBM) construction area is a high-risk and space-constrained area with safety hazards, and manual inspection is inefficient and highly subjective, robots can be used to replace manual labor for cutter inspection. Specifically, in some embodiments, wheeled or tracked robots can move within the construction area and use image acquisition devices such as cameras to capture visual images of the cutters, thereby determining the cutter wear condition through image analysis.

[0027] However, the construction environment of tunnel boring machines (TBMs) is often muddy, complex, and spatially limited, which restricts the robot's detection capabilities and makes it difficult to fully cover the large-diameter cutterhead, especially the top and edge areas, thus reducing the safety, adaptability, and reliability of cutterhead inspection. Therefore, in some embodiments, drones can also be used for inspection. Drones can carry image acquisition equipment such as cameras and fly into the corresponding construction area at the tunnel face to collect images.

[0028] During the inspection of tunnel boring machine (TBM) cutterheads by UAVs, the underground environment of the TBM construction environment, lacking GPS and other positioning signals as well as wireless network signals, causes severe interference to wireless communication. Furthermore, the confined space of the construction area at the tunnel face prevents the UAV from hovering stably and transmitting images in real time, reducing the reliability of cutterhead inspection.

[0029] In some embodiments, fixed probes can also be used for detection. This involves installing monitoring probes in the construction area where the tunnel boring machine (TBM) face is located to capture images of the face. However, the fixed probes have fixed installation positions, making them unsuitable for the continuously advancing tunneling process. Furthermore, fixed probes have a single viewing angle, easily resulting in blind spots and failing to achieve full coverage of the TBM cutterhead. Therefore, the TBM cutterhead detection process described in the above embodiments suffers from low safety, adaptability, and reliability.

[0030] To address the issues of low safety, adaptability, and reliability in shield tunneling machine (TBM) cutterhead inspection, this application provides a cutterhead wear detection system for the tunnel face in some embodiments. This system, equipped with fiber optic real-time image transmission and a dedicated cutterhead wear recognition algorithm, acquires real-time images of the cutterhead surface and identifies defects such as wear and cracks. Combined with human-machine interaction for verification and judgment of the cutterhead status, it can solve the problems of low safety, delayed detection, insufficient accuracy, and unstable image transmission in the TBM cutterhead inspection process. This provides reliable assurance for TBM cutterhead operation and maintenance decisions and construction safety in engineering projects.

[0031] like Figure 1 As shown, the underground shield machine face cutter wear detection system includes a drone, detection components, transmission components, and data processing equipment.

[0032] The drone may include the drone body and a fusion positioning unit mounted on the drone body. To adapt to the flight requirements in confined spaces, the width of the drone body should be less than the width of the flight deck, such as ≤30cm. The fusion positioning unit can detect the drone's position and attitude in real time based on position and attitude data collected by multiple sensors. Therefore, the fusion positioning unit is configured to calculate the drone's positioning data using an Extended Kalman Filter (EKF) fusion algorithm.

[0033] Extended Kalman Filter (EKF) approximates a nonlinear model as a linear model by performing a first-order Taylor expansion at the optimal point of the state estimate (i.e., the current estimate), and then applies the prediction update framework of the standard Kalman Filter (KF). For example, when a fusion positioning unit calculates the UAV's positioning data using the EKF fusion algorithm, it can first set an initial state estimate and an initial error covariance, then read the Inertial Measurement Unit (IMU) data collected by the sensors, and calculate the Jacobian matrix based on the IMU data to update the prediction covariance. When the observed position is reached, the observation residual is calculated, and the Jacobian matrix and Kalman gain are calculated based on the observation residual. Then, the covariance is updated by correcting the state estimate. After multiple iterations, the accurate positioning data of the UAV is obtained.

[0034] In some embodiments, the fusion positioning unit includes an inertial navigation module, an optical flow sensor, and a laser ranging module. The inertial navigation module is configured to detect the drone's attitude; the optical flow sensor is configured to detect the drone's position by analyzing the motion patterns of pixels between consecutive image frames; and the laser ranging module is configured to detect the drone's position using laser ranging.

[0035] For example, a GPS-free fusion positioning unit can be constructed by integrating an IMU inertial navigation module, an optical flow sensor, and a laser ranging module onto the UAV itself. This GPS-free fusion positioning unit can achieve centimeter-level positioning accuracy through an extended Kalman filter fusion algorithm. Combined with the Z-axis absolute height reference provided by the laser ranging module, it is suitable for close-range tool head detection at 30-50cm. Furthermore, the optical flow module optimizes XY plane displacement measurement in low-texture, high-dust environments, and the IMU inertial navigation module can suppress high-frequency drift.

[0036] The detection component is mounted on a drone and is used to acquire images within the construction area where the tunnel boring machine (TBM) face is located. Therefore, the detection component includes a detection camera, a camera gimbal, and a narrow-spectrum infrared illumination module. The detection camera is mounted on the camera gimbal and is equipped with an anti-fog and anti-fouling coated lens and a dust cover. For example, the detection camera uses a 1080p@25fps resolution and is equipped with an anti-fog and anti-fouling coated lens and a miniature dust cover with a 0.1mm aperture to prevent dust adhesion and block large dust particles.

[0037] The camera pan-tilt head is used to adjust the shooting angle of the inspection camera according to the design degrees of freedom. The inspection camera is configured to perform image acquisition on the tunnel boring machine face cutter in the underground construction area to obtain inspection image data.

[0038] For example, the detection components include a detection camera with low-light sensitivity. The camera gimbal can be a miniature, high-precision, dual-axis stabilized gimbal. The camera gimbal can use a lightweight frame and be driven by a small-angle, high-torque miniature digital servo motor, whose yaw and pitch rotation range is adapted to the limited field of view of the tunnel boring machine's observation hole.

[0039] Narrow-spectrum infrared illumination modules are used for infrared illumination to improve the image quality of acquired images. For example, a detection component can integrate an 850nm narrow-spectrum infrared illumination module. This wavelength of light has strong penetrating power through dust, which can reduce the image graying problem caused by dust scattering.

[0040] In some embodiments, to adapt to the dim and dusty environment of the construction area, the detection component may also include an infrared illumination module. This module can actively emit infrared light perceptible to the detection camera via an infrared sensor, illuminating the target scene in low-light or no-light conditions. This achieves adaptation to dim and dusty environments, ensuring clear detection images.

[0041] The transmission component is used for signal transmission between the UAV (including the detection component) and the data processing equipment. The signals transmitted by the transmission component can include detected image data and control commands. To adapt to the underground tunneling environment, the transmission component includes fiber optic transmission lines and optical transceivers. The optical transceiver is the terminal device in the fiber optic communication system responsible for photoelectric signal conversion. The optical transceiver can convert the electrical signals from the detection component and the UAV control system into optical signals that can be transmitted in the fiber optic cable. The optical transceiver can also convert the optical signals transmitted from the fiber optic cable back into electrical signals that can be recognized by other components.

[0042] Fiber optic transmission lines can extend from data processing equipment to drones for data transmission between the drone and the data processing equipment. The fiber optic transmission line includes a detection end and a control end. The detection end of the fiber optic transmission line connects the drone and the detection components via an optical transceiver. The control end of the fiber optic transmission line connects to the optical transceiver data processing equipment.

[0043] For example, the transmission component can be a functional module that integrates 4K image transmission and control. The transmission component can be an optical transceiver with bidirectional transmission through a single optical fiber, and can achieve image transmission capability with an end-to-end delay of less than 50ms through an optical fiber transmission line.

[0044] In some embodiments, the transmission component may also support Region of Interest (ROI) encoding optimization for image data. ROI encoding optimization refers to the differential processing of ROIs and non-ROIs during video or image encoding to maximize the visual quality of key areas within limited transmission capacity.

[0045] For this purpose, the transmission component may also include an image encoder. The image encoder can be a standalone image encoding component or integrated into the optical transceiver. The image encoder can connect to a detection camera to encode the raw images captured by the camera into image data.

[0046] Image transmission encoders can perform ROI encoding optimization by running ROI encoding optimization algorithms. For example, the encoder can be configured to acquire the raw video stream captured by the detection camera, where the raw video stream includes multiple raw images. Raw images are then extracted from the raw video stream based on a preset sampling interval. Regions of interest (ROIs) are then identified from the raw images, and non-ROIs are determined based on the ROIs. The ROIs are then encoded using a first encoding method, and the non-ROIs are encoded using a second encoding method to generate detection image data.

[0047] The first and second encoding methods are two different encoding methods, differing in bit rate, encoding resolution, and encoding tools. To meet the combined requirements of image recognition and data transmission, the first encoding method requires higher encoding quality, larger data volume, and more resources than the second encoding method.

[0048] For example, by setting smaller quantization parameters (QP) for macroblocks or coding units (CUs) within a Region of Interest (ROI), the encoding becomes finer and the bit rate higher. Conversely, setting larger quantization parameters for non-ROI regions results in coarser encoding and a lower bit rate.

[0049] To adapt to the construction environment at the tunnel face, some embodiments may incorporate integrated structures and vibration damping designs for the UAV and transmission components. Specifically, the transmission components and the UAV are fixed using embedded mounting brackets; the brackets have pre-drilled ventilation holes, and thick silicone pads are attached to the inside to enclose the image transmission module, absorbing flight vibrations, protecting delicate electronic components, and improving reliability in vibrating environments.

[0050] The data processing equipment is used to identify defects in the inspection image data and to control the drone during the inspection process. Specifically, it generates structured inspection results based on the inspection image data and sends control commands to the drone or inspection unit in response to user interaction.

[0051] A data processing device can be any electronic device with data processing capabilities. This electronic device includes, but is not limited to, computers, servers, mobile terminals, smart wearable devices, and industrial control machines. For ease of description, this application embodiment uses a data processing device as the execution subject of the method. It should be understood that the method can also be applied to other types of execution subjects, which will not be shown in detail in this application embodiment.

[0052] For example, the data processing equipment has a built-in image recognition and processing unit that can be set up on the ground. By receiving video streams transmitted back via fiber optic cables in real time and integrating a dedicated shield cutter wear recognition algorithm, it can automatically identify the wear area of ​​the cutterhead in real time and mark the defect type and severity level. This allows the data processing equipment to output structured inspection results including defect location images, coordinates, dimensions, and severity levels, and then transmit the recognition results back via a data link for overlay display.

[0053] In some embodiments, the system further includes a display terminal for visually displaying the recognition results generated by the data processing device. To this end, the display terminal can be connected to the data processing device to receive the structured detection results generated by the data processing device and render a result display interface based on the structured detection results.

[0054] The display terminal may include a display component, which is used to present a visual interactive interface. Within this interface, the structured detection results are displayed using visual elements such as graphics, text, and colors. For example, if the display terminal is a smartphone, tablet, personal computer, virtual reality (VR) device, remote control display, etc., the detection parameters, positioning coordinates, and structured detection results can be overlaid and displayed through the interactive interface.

[0055] It should be noted that the display terminal can be a terminal device independent of the data processing device, such as a smartphone as the display terminal and a server as the data processing device. Alternatively, the display terminal can be integrated with the data processing device, in which case the data processing device is an electronic device with display capabilities, such as a computer monitor as the display terminal and the computer's main unit as the data processing device.

[0056] In some embodiments, the display terminal is a virtual reality device. The virtual reality device may include a processor, a display module, a spatial tracking module, and a communication module. The processor is a main processor such as a system-on-chip (SoC). The processor may include a CPU, GPU, and NPU, responsible for running the operating system, processing sensor data, and performing AI tasks. The display module includes a display screen and optical components for presenting virtual reality images. The spatial tracking module includes an inertial measurement unit, spatial positioning sensors, cameras, and other pose sensors for detecting the user's real-time pose data. The communication module includes a DisplayPort, a USB interface, and wireless communication circuitry for communicating with external devices.

[0057] During the process of detecting cutter wear at the face of an underground tunnel boring machine (TBM), the data processing equipment can be configured to execute a method for detecting cutter wear at the face of an underground TBM, such as... Figure 2 As shown, the method includes: S101. Acquire detection image data.

[0058] When performing cutter wear detection, detection image data can be acquired first. This detection image data includes cutter images. A cutter image refers to a detection image containing the entire or part of the cutterhead of the tunnel boring machine, obtained by the detection component through image acquisition.

[0059] The data processing device can acquire detection image data by sending a data acquisition command to the detection component. Specifically, in some embodiments, the data processing device can respond to a user-inputted start detection command by sending a data acquisition command to the drone and the detection component. After receiving the data acquisition command, the drone can fly to a suitable image acquisition area and hover stably at a specific location within that area. The detection component, after the drone has hovered stably, responds to the image acquisition command by initiating image capture to obtain at least one detection image, and then sends the captured detection image to the data processing device, enabling the data processing device to detect and acquire the detection image data.

[0060] To enable the drone to hover stably at a specific location within the image acquisition area, the data processing device can respond to user-input remote control commands, read flight control parameters from the commands, and obtain a detection strategy. This detection strategy includes the tool detection position and flight constraints. The flight control parameters are then corrected based on the detection strategy, and flight control commands are generated based on these corrected parameters. These commands are then sent to the drone to control it to fly to the tool detection position according to the flight constraints.

[0061] The data processing device can also obtain detection image data by frame extraction from the video stream. That is, in some embodiments, the data processing device can receive a video stream captured by the detection component in response to a user-input start detection command, and perform frame extraction on the video stream to extract at least one detection image from the video stream, thereby obtaining detection image data.

[0062] For example, when performing tool wear detection, the entire drone and fiber optic transmission line can be deployed to the working face detection area first, and the equipment can be initialized and calibrated using VR glasses or other display terminals. After the initial calibration is complete, the positioning and flight phase begins. The drone activates its onboard GPS-free fusion positioning unit and flexibly flies to the tool head detection position within a confined space, entering the multi-angle detection phase. During flight and detection, a laser ranging module can maintain a detection distance of 30-50cm from the tool head.

[0063] During the multi-angle inspection phase, the UAV can hover stably in front of the cutter head and scan in multiple directions (up, down, left, and right) by controlling the rotation of the camera gimbal. This allows the inspection camera to capture images of the cutter from various angles. The images are then transmitted in real time to ground data processing equipment and display terminals via fiber optic transmission lines to obtain the inspection image data.

[0064] In some embodiments, the UAV can also perform autonomous detection according to a planned path. That is, the data processing device is also configured to acquire positioning data collected by the fusion positioning unit. The positioning data includes the raw detection data from the fusion positioning unit, the UAV's position, and the UAV's attitude.

[0065] A construction area model is then constructed based on the positioning data and detection image data. Undetected areas are determined according to the construction area model and detection image data. Detection control instructions are then generated based on the undetected areas. These detection control instructions include a planned path for controlling the flight of the UAV. The detection control instructions are then sent to the UAV so that the UAV flies according to the planned path.

[0066] For example, data processing equipment can acquire positioning data and build a 3D model based on the original detection data, UAV position and attitude in the positioning data. The original detection data is superimposed and modeled according to the UAV position and attitude to obtain a three-dimensional point cloud model. Then, the detection image data is added to the three-dimensional point cloud model according to the UAV position and attitude to form a construction area model.

[0067] Then, based on the construction area model, undetected areas that the drone has not yet flown over are identified. Based on the shape and location of these undetected areas, a flight path is planned using the principle of minimum path or optimal shooting angle, and detection control commands containing the planned path are generated. By sending these detection control commands to the drone, it is controlled to fly along the planned path, completing the image detection of the undetected areas.

[0068] S102, invoke the tool wear identification algorithm.

[0069] like Figure 3 As shown, after acquiring the detection image data, the data processing device can perform tool wear identification based on the detection image data, thus enabling the invocation of a tool wear identification algorithm. This tool wear identification algorithm includes a defect detection model, which is a neural network model trained using sample defect data. The sample defect data includes sample images containing typical tool defects and defect labels for the sample images. The defect labels include defect type labels and defect level labels.

[0070] The defect detection model can employ a two-stage training architecture combining MAML meta-learning and an improved Swin-Transformer. To this end, before performing wear detection, the defect detection model can include an offline meta-training stage (referred to as the meta-training stage) and an engineering small-sample fine-tuning stage (referred to as the meta-fine-tuning stage). In the offline meta-training stage, the MAML algorithm can learn general defect features on a common tool defect dataset containing defects such as wear, cracks, and chipping. These features serve as meta-knowledge for rapid adaptation to new scenarios, outputting pre-trained weights.

[0071] During the small-sample fine-tuning phase, 30-50 actual construction scene defect samples from dusty environments can be used to fine-tune the pre-trained weights and generate customized weights. Alternatively, 5-10 small samples from the construction site with defect type or level labels can be used to quickly adjust model parameters to adapt to the tool characteristics of the project. During online detection, the model loads the fine-tuned weights and performs real-time inference using the improved Swin-Transformer. Furthermore, the defect detection model includes a dark channel prior dehazing module and an improved median filtering module. The dark channel prior dehazing module removes image blurring caused by dust, while the improved median filtering module filters dust noise to extract defect features.

[0072] For example, a tool wear identification algorithm can be used to call a defect detection model and then use that model to identify tool wear. The defect detection model, as a pre-trained neural network model, can be trained based on typical defect samples such as tool cracks, wear, and chipping. After training, the defect detection model automatically identifies the wear area of ​​the tool head in real time based on the input image data and labels the defect type and severity level. For example, defect types are labeled as cracks, wear, and chipping; severity levels are labeled as slight, moderate, and severe.

[0073] S103. Use a tool wear recognition algorithm to identify the wear area and defect information of the tool head in the detection image data.

[0074] After invoking the tool wear recognition algorithm, the data processing equipment can run the corresponding control program to perform tool wear recognition in the inspection image data, identifying the tool head wear area and defect information from the inspection image data. The defect information includes the defect type and defect level output by the defect detection model based on the tool image.

[0075] For example, a data processing device can extract tool images from detected image data and input the tool images into a defect detection model. The defect detection model can use neural network units to encode features of the input tool images to generate a feature matrix. Based on the feature matrix, deep learning is then used to transform the object detection problem into a single regression problem to predict bounding boxes and categories from the image.

[0076] Next, classifiers such as Support Vector Machine and Softmax are used to calculate the classification probability of the input tool image relative to each defect type label and defect level label. Then, the defect type label and defect level label with the highest classification probability are used to determine the defect type and defect level corresponding to the tool image, thus obtaining the defect information.

[0077] In some embodiments, when the defect detection model includes a dark channel prior dehazing module and an improved median filtering module, the tool wear recognition algorithm can be used to identify the tool head wear area and defect information in the detection image data. This can be achieved by first removing dust and fog through the dark channel prior dehazing module, then filtering dust noise through the improved median filtering module, and finally extracting defect features to obtain defect information.

[0078] S104. Generate structured inspection results based on the wear area and defect information of the cutter head.

[0079] After identifying the wear area and defect information of the cutter head, the data processing equipment generates structured inspection results based on the wear area and defect information. These structured inspection results include defect location images, defect coordinates, defect dimensions, and defect level.

[0080] For example, a data processing device can extract defects from a tool image based on the wear area of ​​the identified tool head, crop the defect location image from the tool image, and then record the defect coordinates and defect size based on the position of the defect in the tool image.

[0081] Simultaneously, the data processing equipment can generate type labels based on the defect type in the defect information, such as labeling the defect type as one of the types like crack, wear, or chipping. It can also label the defect level based on the severity level in the defect information, such as minor crack, moderate crack, or severe crack. Therefore, by fusing the generated defect location image, defect coordinates, defect size, and defect level data, structured inspection results can be generated for transmission back via the data link and overlay display.

[0082] In some embodiments, after generating structured detection results, the data processing device can display the generated structured detection results through a display terminal. Specifically, the data processing device is further configured to generate an interface rendering background based on the construction area model and the detection image data, and to generate algorithm labeling results based on the construction area model and the structured detection results. The algorithm labeling results include color-coded visual identifiers. Then, the algorithm labeling results are overlaid onto the interface rendering background in real time to generate a result display interface, which is then displayed on the display terminal.

[0083] For example, in the identification and evaluation stage of tool wear detection, the data processing equipment can automatically analyze the detection image data using a dedicated wear identification algorithm and mark the wear area and degree using a defect detection model. The data processing equipment can overlay the algorithm identification results (including defect type and degree level) onto the display interface of the display terminal in real time, and use a color-coded visualization scheme to generate the interface rendering background and algorithm marking results. The interface rendering background can be the original image content of the detection image, and the algorithm marking results can be represented by red bounding boxes to indicate severe wear or chipping defects, yellow bounding boxes to indicate moderate wear, and green bounding boxes to indicate minor cracks or wear, with a defect type label and severity level overlaid next to each bounding box.

[0084] In some embodiments, while displaying the results on the display terminal, real-time verification by inspection personnel is also supported, and defect location information is recorded based on the verification results. When the display terminal is a virtual reality device, verification and recording can be performed from the user's first-person perspective using VR devices. That is, by equipping a display terminal such as VR glasses, inspection personnel can observe the drone's image transmission in real-time from a first-person perspective and perform real-time verification. For this purpose, the data processing device is also configured to acquire interactive commands input by the user through the virtual reality device. These interactive commands include at least one of verification commands, correction commands, and confirmation commands.

[0085] Following the interactive instructions, the system records the corresponding detection result information for the structured detection results and generates a structured detection report based on this information. The detection result information includes the confirmed defect location image, defect coordinates, and defect level.

[0086] For example, inspectors can perform interactive operations based on virtual reality devices through the interface of VR glasses, such as gaze confirmation, gesture operation, or voice commands. These interactive operations can form interactive instructions used to review, correct, or confirm the algorithmic labeling results. The data processing equipment then responds to the user's interactive operations, automatically recording the final confirmed defect location coordinates, images, and evaluation levels to generate a structured inspection report.

[0087] In some embodiments, when the display terminal is a virtual reality device, the virtual reality device and the drone can also be linked for control. Specifically, the data processing device is configured to acquire real-time pose data and historical pose data. The real-time pose data is the data detected in real-time by the pose sensor in the virtual reality device. Historical pose data can be obtained by recording the real-time pose data during the user's wearing of the device.

[0088] After acquiring real-time pose data and historical pose data, the motion amplitude is calculated based on these data. The motion amplitude is calculated using a weighted sum of the differences in each dimension between the real-time and historical pose data, combined with a construction area model.

[0089] For example, VR devices can use built-in pose sensors such as gyroscopes and accelerometers to collect real-time user pose data, including position coordinates (x, y, z) and attitude data. The attitude data is represented using Euler angles, i.e., (α, β, γ), where α, β, and γ represent rotation angles centered on the x, y, and z axes, respectively. Historical pose data from the last acquisition is then retrieved from the storage unit. This storage unit can be used to store previously acquired pose data in chronological order for comparison and calculation.

[0090] After data preprocessing including calibration, filtering, and synchronization, the differences are calculated. These differences can include positional and attitude differences. For positional differences, the differences between the real-time pose data (x, y, z) and the corresponding position coordinates (x0, y0, z0) in the historical pose data are calculated in each dimension: Δx = x - x0; Δy = y - y0; Δz = z - z0. For attitude differences, if the attitude data uses Euler angles, the differences between the rotation angles (α, β, γ) in the real-time pose data and the corresponding rotation angles (α0, β0, γ0) in the historical pose data are calculated in each dimension: Δα = α - α0; Δβ = β - β0; Δγ = γ - γ0.

[0091] Then, a weighted summation is performed based on the difference calculation results. During the weighted summation, weights are first determined, that is, weights are assigned to each dimension of the position difference and posture difference based on the needs of motion analysis and the contribution of each dimension's data to the motion amplitude. For example, in a construction area scenario with a narrow detection space, positional changes have a greater impact on the motion amplitude, so the position difference can be assigned a relatively higher weight. In a construction area scenario with an open detection space, posture changes are also important, so appropriate weights are assigned to the posture difference. Weight values ​​can be determined based on experience, experimental data, or through machine learning methods. Weight values ​​are between 0 and 1, and the sum of all weight values ​​is 1. Then, a weighted summation is performed based on the determined weights, multiplying each dimension's difference by its corresponding weight and then summing them to obtain the weighted total difference, i.e.: Δs = w x Δx+w y Δy+w z Δz+w α Δα+w β Δβ+w γ Δγ.

[0092] Based on the weighted total difference, the data processing device can combine it with the construction area model to map the total difference in the VR device's coordinate space to the coordinate space corresponding to the construction area model, in order to calculate the motion amplitude. During coordinate mapping, a mapping function can first be established between the VR device's coordinate space and the coordinate space corresponding to the construction area model, and then the total difference can be converted into motion amplitude based on the mapping function.

[0093] Then, based on the application scenario, one or more motion amplitude thresholds are set to determine the degree of the motion. When the motion amplitude is less than or equal to a certain preset amplitude threshold, it is considered a small motion, and the drone can maintain its current state and only respond to the VR screen. When the motion amplitude is greater than a certain preset amplitude threshold, it is considered that the user has made a large motion. At this time, a field-of-view following command can be generated based on real-time pose data and sent to the drone to control the drone to fly according to the real-time pose data.

[0094] After generating a structured inspection report, the data processing equipment can output the report to obtain the tool wear inspection results. Furthermore, after the inspection is completed, the drone can autonomously return to its home location along the return path, allowing inspection personnel to disassemble the transmission components, collect and organize the fiber optic cables, and perform maintenance and cleaning.

[0095] By applying the technical solutions of the above embodiments, the underground shield machine face cutter wear detection system and method described in the above embodiments eliminate the need for inspection personnel to enter the dangerous area of ​​the face during defect detection, avoiding safety risks such as collapse and dust, and improving the safety of the inspection process. Furthermore, fiber optic image transmission overcomes signal attenuation, and combined with GPS-free fusion positioning achieves stable hovering. With a body width ≤30cm and supplemented by soft shell protection, the detection system can reliably operate in the harsh underground environment of confined spaces, high dust levels, and strong interference, improving the adaptability of the inspection process. Fiber optic transmission achieves zero packet loss, and multi-sensor fusion positioning solves the GPS-free hovering problem, ensuring uninterrupted inspection and improving the reliability of the inspection process. In addition, the system can achieve multi-angle shooting, avoiding blind spots, and uses a dedicated recognition algorithm for automatic wear marking, enabling comprehensive and accurate detection. Combined with VR first-person perspective real-time evaluation and automatic marking, it significantly improves inspection and verification efficiency, enhancing the efficiency of the inspection operation.

[0096] In some embodiments, as a refinement and extension of the specific implementation of the above embodiments, and to fully illustrate the specific implementation process of this embodiment, some embodiments of this application also provide a method for detecting the wear of cutters at the face of an underground tunnel boring machine. The difference between this method and the above embodiments is that, during the image detection process, multiple images can be acquired and superimposed to obtain an overview image containing the entire cutter face of the tunnel boring machine, such as... Figure 4As shown, the method includes: S201. Extract multiple tool images from the detection image data according to the detection time; S202. Acquire multiple positioning data points collected at the same detection time as the tool image; S203. Based on the positioning data, stitch together multiple tool images to generate an overview image; S204. Using the overview image as the rendering background, generate a defect location image by overlaying the tool disc wear area and defect information in real time.

[0097] In order to perform image overlay synthesis, the data processing device can extract multiple tool images from the detection image data according to the detection time. The multiple tool images include images taken by the drone at different times and locations.

[0098] Then, multiple positioning data collected at the same detection time as the tool image are acquired, and multiple tool images are stitched together based on the positioning data to generate an overview image. Then, using the overview image as the rendering background, a defect location image is generated by superimposing the tool disc wear area and defect information in real time.

[0099] For example, at detection time T1, a high-resolution industrial camera is used to photograph the tunnel boring machine cutters. Since the cutters are quite long, a single shot cannot cover the entire cutter; therefore, segmented shooting is necessary. Thus, at time T1, the camera starts shooting from the left end of the cutter, obtaining image A1. Then, a drone moves the camera to the middle position to capture image A2, and finally moves the camera to the right end of the cutter to capture image A3. This results in three cutter images (A1, A2, A3).

[0100] While capturing images of each tool, the inertial navigation module, optical flow sensor, and laser ranging module in the fusion positioning unit record the UAV's positioning data for subsequent image stitching. Specifically, based on the positioning data (P1, P2, P3), the relative positional relationships between images (A1, A2, A3) are determined. Then, an image stitching algorithm based on feature point matching is used to stitch images A1, A2, and A3 together according to the positioning data. Finally, the stitched images undergo fusion processing to eliminate stitching seams, generating a complete overview image B.

[0101] Then, image processing algorithms such as edge detection and threshold segmentation are used to analyze the overview image B to detect the wear areas and defect locations of the tool. The detected wear areas and defect information are then overlaid onto the overview image B with different colors or markers, and these markers are updated in real time to generate the defect location image C.

[0102] In some embodiments, as a refinement and extension of the specific implementation of the above embodiments, and to fully illustrate the specific implementation process of this embodiment, some embodiments of this application also provide a method for detecting cutter wear at the face of an underground tunnel boring machine. The difference between this method and the above embodiments is that, when the transmission component supports ROI encoding optimization, the recognition algorithm can be linked with the image transmission encoding. For severely defective areas marked by the algorithm, ROI encoding optimization is automatically enabled to improve the image quality of that area. For example... Figure 5 As shown, the method includes: S301. Delineate the region of interest based on the structured detection results; S302. Determine the non-interested regions based on the region of interest; S303, Generate encoding optimization instructions; S304. Send the encoding optimization instruction to the image transmission encoder so that the image transmission encoder performs image encoding on the raw image data acquired by the detection camera according to the encoding optimization instruction to generate detection image data.

[0103] When the transmission component supports ROI encoding optimization, the data processing device can first delineate the region of interest (ROI) based on the structured detection results, and then determine the non-ROI based on the ROI. The ROI is a defect region in the tool image where the number of samples is less than a threshold, or a defect region detected for the first time, or a defect region in the tool image where the defect level is higher than a preset level threshold. The non-ROI is the region in the tool image outside the ROI.

[0104] Different processing priorities can be set for different types of regions of interest. For example, small-sample defect regions can be set as the first priority. A small-sample defect region refers to a defect region corresponding to a defect type with fewer than 10 samples in the sample library, or a newly detected defect region. In this case, small-sample defect regions are designated as regions of interest regardless of their defect level. Correspondingly, non-small-sample defect regions with defect levels higher than a preset threshold are set as the second priority.

[0105] Then, encoding optimization instructions are generated, which include a first encoding method for the region of interest and a second encoding method for the region of non-interest. These encoding optimization instructions are then sent to the image transmission encoder, causing the encoder to perform image encoding on the raw image data acquired by the detection camera according to the instructions, thereby generating detection image data.

[0106] For example, after a drone equipped with a detection camera captures a detection image (image number Image_001) containing a tool target, image processing algorithms such as edge detection and template matching can be used to analyze Image_001 and identify the area where the tool is located. If the upper left corner of the area where the tool is located is (100, 150) and the lower right corner is (200, 250), then this area can be marked as the region of interest. That is, in image Image_001, the defined region of interest is a rectangular area with coordinates ranging from (100, 150) to (200, 250).

[0107] After identifying the region of interest (ROI) containing the tool target, the remaining portion of the detected image can be designated as the non-ROI. Specifically, the non-ROI comprises all areas of the detected image except for those between (100, 150) and (200, 250). Then, based on the division of the ROI and non-ROI, encoding optimization instructions are generated. High-resolution and high-quality encoding parameters are used for the ROI (tool region) to ensure clear details. Low-resolution and low-quality encoding parameters are used for the non-ROI (background region) to reduce data volume. For example, the encoding optimization instructions might include ROI encoding parameters of 1024×768 resolution and a quality factor of 95, and non-ROI encoding parameters of 640×480 resolution and a quality factor of 50.

[0108] The encoding optimization command is then sent to the image encoder via the communication interface. Upon receiving the command, the image encoder encodes the raw image data acquired by the detection camera according to the instructions. Specifically, for the detection image Image_002, the tool area is encoded using high-resolution and high-quality encoding parameters according to the region of interest (ROI). For non-ROI regions, low-resolution and low-quality encoding parameters are used. After encoding, the optimized detection image data Encoded_Image_002 is generated. The detection image optimized by ROI encoding makes the tool area clearly visible, while significantly reducing the amount of data in the background area, thus improving transmission efficiency.

[0109] In accordance with ROI encoding optimization, the data processing device can perform corresponding decoding based on ROI encoding optimization rules. That is, the data processing device is also configured to acquire the transmission data stream corresponding to the detected image data, and determine a first decoding method based on a first encoding method, and a second decoding method based on a second encoding method.

[0110] Next, according to the region of interest, the key region image is decoded in the transmission data stream using the first decoding method, and according to the non-key region of interest, the non-key region image is decoded in the transmission data stream using the second decoding method. Then, the tool image is generated by stitching the key region image and the non-key region image together.

[0111] By applying the technical solutions of the above embodiments, this application provides a system and method for detecting cutter wear on the face of an underground tunnel boring machine (TBM). The system includes a drone, a detection component, a transmission component, and a data processing device. The drone can carry the detection component to perform image acquisition on the cutter face of the TBM in the underground construction area to obtain detection image data. The transmission component transmits the detection image data and control commands via fiber optic transmission lines, enabling the data processing device to use a cutter wear recognition algorithm to identify wear areas and defect information on the cutterhead in the detection image data, generating structured detection results. The system can acquire images via a drone and transmit them in real time via fiber optics, identifying wear, cracks, and other defects in the cutterhead surface detection image data according to the cutter wear recognition algorithm. The system adopts a two-stage architecture of MAML meta-learning and an improved Swin-Transformer, adapting to small sample scenarios and 1080p real-time detection requirements. It can adapt to high-dust construction environments in close-range, confined spaces, improving the safety, adaptability, and reliability of the detection process.

[0112] This application also provides a computer device, specifically a personal computer, server, network device, etc. The computer device includes a bus, processor, memory, and communication interface, and may also include input / output interfaces and a display device. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the computer device stores location information. The network interface of the computer device is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements the steps in the various method embodiments.

[0113] Those skilled in the art will understand that the structure of the computer device described above is only a partial structure related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied. A specific computer device may include more or fewer components, or combine certain components, or have different component arrangements.

[0114] In one embodiment, a computer-readable storage medium is also provided, which may be non-volatile or volatile, and a computer program is stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0115] In one embodiment, a computer program product is also provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0116] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0117] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods.

[0118] Any references to memory, database, or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc.

[0119] Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can take many forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0120] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchain. The processors involved in the embodiments provided in this application may be, but are not limited to, general-purpose processors, graphics processors, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc.

[0121] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0122] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A wear detection system for cutting tools at the face of an underground tunnel boring machine, characterized in that, The system includes: The drone includes a fusion positioning unit configured to calculate the drone's positioning data using an extended Kalman filter fusion algorithm. A detection component is mounted on the UAV; the detection component includes a detection camera, a camera gimbal, and a narrow-spectrum infrared supplementary lighting module; the detection camera is mounted on the camera gimbal; the detection camera is equipped with an anti-fog and anti-fog coated lens and a dust cover; the detection camera is configured to perform image acquisition on the tunnel boring machine face cutter in the underground construction area to obtain detection image data. A transmission component is configured to transmit the detected image data and control commands; the transmission component includes an optical fiber transmission line and an optical transceiver; the optical fiber transmission line includes a detection end and a control end; the detection end of the optical fiber transmission line is connected to the UAV and the detection component via the optical transceiver. The data processing device is connected to the control terminal of the optical fiber transmission line via the optical transceiver; the data processing device is configured as follows: Acquire detection image data, the detection image data including tool images; A tool wear identification algorithm is invoked, which includes a defect detection model. This defect detection model is a neural network model trained using sample defect data. The defect detection model employs a two-stage training architecture combining MAML meta-learning and an improved Swin-Transformer. The defect detection model includes a dark channel prior dehazing module and an improved median filtering module to remove image blur and noise caused by dust. The sample defect data includes sample images containing typical tool defects and defect labels for the sample images. The defect labels include defect type labels and defect level labels. The tool wear recognition algorithm is used to identify the tool head wear area and defect information in the detection image data. The defect information includes the defect type and defect level output by the defect detection model based on the tool image. A structured detection result is generated based on the wear area of ​​the cutter head and the defect information. The structured detection result includes a defect location image, defect coordinates, defect size, and defect level.

2. The system according to claim 1, characterized in that, The fusion positioning unit includes an inertial navigation module, an optical flow sensor, and a laser ranging module; the inertial navigation module is configured to detect the attitude of the UAV; the optical flow sensor is configured to detect the position of the UAV by analyzing the motion patterns of pixels between consecutive image frames; the laser ranging module is configured to detect the position of the UAV by laser ranging; the data processing device is further configured to: The positioning data collected by the fusion positioning unit is obtained, and the positioning data includes the original detection data of the fusion positioning unit, the position of the UAV, and the attitude of the UAV. A construction area model is constructed based on the positioning data and the detected image data; The undetected areas are determined according to the construction area model and the detection image data; A detection control command is generated based on the undetected area, and the detection control command includes a planned path for controlling the flight of the UAV; The detection and control commands are sent to the drone so that the drone flies along the planned path.

3. The system according to claim 2, characterized in that, The system further includes a display terminal connected to the data processing device; the data processing device is further configured to: The interface rendering background is generated based on the construction area model and the detected image data; Based on the construction area model and the structured detection results, an algorithm marking result is generated, which includes a color-coded visual identifier. The algorithm's labeling results are overlaid onto the interface rendering background in real time to generate a result display interface; The results are displayed on the display terminal.

4. The system according to claim 3, characterized in that, The display terminal is a virtual reality device; the data processing device is further configured to: The system acquires interactive commands input by the user through the virtual reality device, the interactive commands including at least one of a review command, a correction command, and a confirmation command; According to the interaction instructions, the detection result information corresponding to the structured detection result is recorded. The detection result information includes the confirmed defect location image, defect coordinates, and defect level. A structured test report is generated based on the test results.

5. The system according to claim 4, characterized in that, The virtual reality device includes a pose sensor configured to detect real-time pose data of the user; the data processing device is further configured to: The real-time pose data and historical pose data are acquired, wherein the historical pose data is obtained by recording the real-time pose data during the user's wearing process; The motion amplitude is calculated based on the real-time pose data and the historical pose data. The motion amplitude is calculated based on the weighted sum of the differences in each dimension of the real-time pose data and the historical pose data, combined with the construction area model. If the amplitude of the movement is greater than a preset amplitude threshold, a field-of-view following instruction is generated based on the real-time pose data. The field-of-view following instruction is used to control the UAV to fly according to the real-time pose data. The vision-following command is sent to the drone.

6. The system according to claim 1, characterized in that, The transmission component further includes an image transmission encoder connected to the detection camera; the data processing device is further configured to: Based on the structured detection results, a region of interest is defined. The region of interest is a defect region in the tool image where the number of samples is less than a threshold, or a defect region detected for the first time, or a defect region in the tool image where the defect level is higher than a preset level threshold. A region of non-interest is determined based on the region of interest, and the region of non-interest is the area outside the region of interest in the tool image; Generate encoding optimization instructions, the encoding optimization instructions including a first encoding method for the region of interest and a second encoding method for the region of non-interest; The encoding optimization instruction is sent to the image transmission encoder so that the image transmission encoder performs image encoding on the raw image data acquired by the detection camera according to the encoding optimization instruction to generate detection image data.

7. The system according to claim 6, characterized in that, The data processing device is also configured to: Obtain the transmission data stream corresponding to the detected image data; A first decoding method is determined based on the first encoding method, and a second decoding method is determined based on the second encoding method; According to the region of interest, the key region image is decoded in the transmitted data stream using the first decoding method; According to the non-interest region, the second decoding method is used to decode the non-critical region image in the transmitted data stream; The key region image and the non-key region image are stitched together to generate the tool image.

8. The system according to claim 1, characterized in that, The data processing device is also configured to: In response to a remote control command input by the user, flight control parameters are read from the remote control command; A detection strategy is obtained, which includes tool detection position and flight restriction conditions; The flight control parameters are corrected based on the detection strategy, and flight control commands are generated based on the corrected flight control parameters. The flight control command is sent to the drone to control the drone to fly to the tool detection position according to the flight restrictions.

9. The system according to claim 1, characterized in that, The data processing device is also configured to: Multiple tool images are extracted from the detection image data according to the detection time; Acquire multiple positioning data points that were collected at the same detection time as the tool image; Multiple tool images are stitched together based on the positioning data to generate an overview image; Using the overview image as the rendering background, the defect location image is generated by overlaying the wear area of ​​the cutter head and the defect information in real time.

10. A method for detecting wear of cutting tools at the face of an underground tunnel boring machine, characterized in that, Applied to the system according to any one of claims 1-9; the method comprises: Acquire detection image data, the detection image data including tool images; A tool wear identification algorithm is invoked, which includes a defect detection model. This defect detection model is a neural network model trained using sample defect data. The defect detection model employs a two-stage training architecture combining MAML meta-learning and an improved Swin-Transformer. The defect detection model includes a dark channel prior dehazing module and an improved median filtering module to remove image blur and noise caused by dust. The sample defect data includes sample images containing typical tool defects and defect labels for the sample images. The defect labels include defect type labels and defect level labels. The tool wear recognition algorithm is used to identify the tool head wear area and defect information in the detection image data. The defect information includes the defect type and defect level output by the defect detection model based on the tool image. A structured detection result is generated based on the wear area of ​​the cutter head and the defect information. The structured detection result includes a defect location image, defect coordinates, defect size, and defect level.

Images