Belt carrier roller fault detection method and system based on inspection robot

Abstract

The invention relates to the technical field of tunnel monitoring and image processing, and discloses a belt carrier roller fault detection method and system based on an inspection robot, and the method comprises the steps: obtaining image data and video data of a TBM belt carrier roller, carrying out the preprocessing of the image data, and carrying out the positioning of the position of the carrier roller; inputting the preprocessed image data into a constructed belt carrier roller fault detection model, and training the model to obtain a fault carrier roller image; detecting the rotating speed type of the carrier roller according to the video data, and determining whether the rotating speed of the carrier roller is abnormal according to the rotating speed type of the carrier roller to obtain a rotating speed abnormal result of the carrier roller; and the state of the belt carrier roller is comprehensively judged according to the obtained fault carrier roller image and the rotating speed abnormal result of the carrier roller, the carrier roller with double fault indexes is obtained, an abnormal image frame is output, meanwhile, capturing time information of the image is associated, and the fault carrier roller with time information and position information is obtained. The fault carrier roller can be identified and positioned, and the inspection efficiency is high.

Description

Technical Field

[0001] The present invention relates to the technical field of tunnel monitoring and image processing, and in particular to a belt roller fault detection method and system based on a TBM long tunnel inspection robot. Background Art

[0002] During TBM tunneling, the broken rocks drilled by the tunnel boring machine need to be transported out through a belt conveyor. The belt conveyor rollers are key components. Roller failure and jamming will cause the roller temperature to rise sharply, causing fires. The tunnel inspection robot monitors the roller status to prevent accidents in advance. At present, the roller inspection work in TBM tunnels mainly adopts manual inspection. The tunnel mileage is long and there is water on the ground. Personnel can only follow the tool vehicle in and out, and the efficiency of manual inspection is low. It is necessary to study the fault detection method of belt rollers in long tunnels to analyze whether there are faults in belt rollers.

[0003] Most non-manual detection methods mainly use audio sensors, contact temperature sensors and infrared temperature sensors to detect faulty rollers. The audio method requires capturing noise signals from the belt conveyor and analyzing them to determine whether they are abnormal. However, in a long tunnel environment, due to the dense equipment and complex sound environment, the presence of a large number of interference signals reduces the accuracy of the above detection methods. The method of deploying temperature sensors inside the rollers is still relatively ideal, but the number of rollers is huge and wiring is required, which makes it difficult to implement. The roller temperature is monitored by thermal infrared imaging, and whether the roller is abnormally hot is determined by whether its highest temperature exceeds the set threshold. If the method of judging roller overheating only relies on the temperature threshold, it is easily affected by other heating equipment around it, thereby affecting the reliability of the detection. Summary of the invention

[0004] In view of the above problems, the purpose of the present invention is to provide a belt roller fault detection method and system based on a patrol robot, which can automatically patrol the rollers and identify and locate faulty rollers, replacing manual patrol and improving patrol efficiency.

[0005] To achieve the above-mentioned purpose, in the first aspect, the technical solution adopted by the present invention is: a belt roller fault detection method based on an inspection robot, which comprises: obtaining image data and video data of a TBM belt roller through the inspection robot, and locating the position of the roller after preprocessing the image data; inputting the preprocessed image data into a constructed belt conveyor roller fault detection model, training the model, obtaining a faulty roller image, and determining the position information of the faulty roller according to the faulty roller image; detecting the speed type of the roller according to the video data, determining whether the roller speed is abnormal according to the roller speed type, and obtaining an abnormal speed result of the roller; comprehensively judging the state of the belt roller according to the obtained faulty roller image and the abnormal speed result of the roller, obtaining a roller with dual fault indicators, and outputting an image frame confirming the abnormality, and at the same time associating the capture time information of the image to obtain a faulty roller with time information and position information.

[0006] In a preferred embodiment, the inspection robot obtains image data and video data of the TBM belt roller, and pre-processes the image data, including:

[0007] The inspection robot inspects along the track direction to obtain image data and video data of the TBM belt roller; wherein the image data includes RGB images and infrared temperature images;

[0008] The RGB image and the infrared temperature image are compressed to a preset size so that the sizes of the two images are consistent;

[0009] Perform target detection on the compressed temperature image data, filter out irrelevant background information, and locate the position of the roller.

[0010] In a preferred embodiment, the preprocessed image data is input into a constructed belt roller fault detection model, and the model is trained to obtain a fault roller image, including:

[0011] The belt roller fault detection model is constructed by using the salient target detection algorithm. The preprocessed image data is used as the input data of the belt roller fault detection model. The model is trained to obtain the optimal training weight. The training weight is used to test the roller fault detection model to obtain the faulty roller image, including:

[0012] Encode the input RGB image and temperature image, obtain high-level feature maps and low-level feature maps through the CAFM module and FFAM module, and fuse feature maps of different levels through the cross-modal interaction module;

[0013] The fused feature map is decoded, and the feature map output by the CAFM module and FFAM is combined with the features extracted after decoding to obtain the final abnormal temperature detection result of the roller.

[0014] In a preferred embodiment, the position information of the faulty roller is determined based on the image of the faulty roller, including: after acquiring the image of the faulty roller, determining which two RFID tags the grab time is within the time range of reading according to the grab time corresponding to the image of the faulty roller, and then determining the positions of the two RFID tags; using the positions of the two RFID tags and the time difference to determine the exact position of the faulty roller.

[0015] In a preferred embodiment, the speed type of the roller is detected according to the video data, and whether the speed of the roller is abnormal is determined according to the speed type of the roller, and the abnormal speed result of the roller is obtained, including:

[0016] The adjacent multiple frames of the video data are sequentially processed through convolution and pooling operations, and then fully connected to obtain the speed classification of the roller;

[0017] The rotational linear speed of the roller is obtained by the obtained classification category of the roller speed. At the same time, the linear speed of the belt is obtained by giving a rotational speed of the belt conveyor motor. The linear speed of the roller is compared with the linear speed of the belt. When the ratio of the linear speed of the roller to the linear speed of the belt is less than the set value, the roller speed is judged to be abnormal, and the obtained classification speed is further judged by speed identification to obtain the roller with abnormal speed.

[0018] In a preferred embodiment, comprehensively judging the state of the belt idler according to the obtained faulty idler image and the abnormal result of the rotation speed of the idler includes:

[0019] If the faulty roller image and the abnormal rotation speed result of the roller appear at the same time, the faulty roller image is output;

[0020] If only the image of the faulty roller or the abnormal result of the roller speed is found, the fault position will be inspected in the next inspection cycle, and a fault warning will be output at this time, but the fault image will not be output;

[0021] The image data are named according to the time when the image is captured, and the faulty roller with time information and position information is obtained through the faulty roller image name;

[0022] The fault information is organized into a fault report, and the monitoring report is output when the inspection robot moves to the end of the track.

[0023] In the second aspect, the technical solution adopted by the present invention is: a patrol robot, used to implement the above-mentioned belt roller fault detection method based on the TBM long tunnel patrol robot, which includes: a box body, which has a waterproof shell, and a drive motor and a control system are arranged in the waterproof shell, the drive motor is connected to the control system, and its operation is controlled by the control system; an upper connecting plate, which is arranged on the top of the box body, and a driving wheel and two sets of walking mechanisms are arranged on one side of the upper connecting plate, the driving wheel is arranged in the middle of the two sets of walking mechanisms, and the driving wheel is driven by the driving motor, thereby driving the two sets of walking mechanisms to move along the I-beam track suspended on the side of the TBM conveyor belt; wherein the driving wheel is arranged on the lower surface of the I-beam track and fits tightly with the lower surface to realize movement along the I-beam track; each set of walking mechanisms includes a driven wheel, a guide wheel, a bearing and a connecting hanging plate; the connecting plate The bottom of the hanging plate is installed on the upper connecting plate through a bearing to realize the steering of the robot; the connecting hanging plate adopts a groove structure, and a driven wheel is respectively arranged on the inner side of the vertical hanging plates on both sides of the connecting hanging plate of the groove structure, and the waist of the I-beam track is arranged between the two driven wheels, so that the box body can be hung on the I-beam track through the driven wheel; on the vertical hanging plates on both sides of the connecting hanging plate, two guide wheels are respectively arranged on both sides of the driven wheel, and the guide wheels are in contact with the side of the I-beam track to realize the movement of always in contact with the I-beam track; an RFID positioning sensor is arranged on the upper connecting plate; the RFID positioning sensor cooperates with each electronic tag arranged at intervals on the side of the I-beam track to transmit the read tag information to the control system; a dual-mode camera is arranged at the front of the box body to collect image data of the TBM belt roller and transmit it to the control system.

[0024] In a preferred embodiment, the clamping force between the driving wheel and the lower surface of the I-beam rail is adjusted by a spring, which is arranged between the driving wheel bracket and the upper connecting plate and sleeved on the bolt; an anti-loosening device is adopted on the connecting bolts used to connect the driving wheel bracket and the upper connecting plate, so that the driving wheel and the lower surface of the I-beam rail can always fit tightly to ensure power transmission;

[0025] An obstacle avoidance sensor is arranged at the front of the box, and an environmental monitoring sensor is arranged inside the box. The obstacle avoidance sensor and the environmental monitoring sensor transmit the collected data to the control system.

[0026] In the third aspect, the technical solution adopted by the present invention is: a belt roller fault detection system based on an inspection robot, comprising: a data processing module, which pre-processes the image data and video data of the TBM belt roller obtained by the inspection robot to locate the roller position; a target detection module, which inputs the pre-processed image data into the constructed belt roller fault detection model, trains the model, obtains the faulty roller image, and determines the position information of the faulty roller according to the faulty roller image; a speed detection module, which detects the speed type of the roller according to the video data, determines whether the roller speed is abnormal according to the roller speed type, and obtains the abnormal speed result of the roller; a comprehensive judgment module, which comprehensively judges the state of the belt roller according to the obtained faulty roller image and the abnormal speed result of the roller, obtains a roller with dual fault indicators, and outputs an image frame confirming the abnormality, and at the same time associates the capture time information of the image to obtain a faulty roller with time information and position information.

[0027] In a fourth aspect, the technical solution adopted by the present invention is: a computer-readable storage medium storing one or more programs, wherein the one or more programs include instructions, and when the instructions are executed by a computing device, the computing device executes any one of the above methods.

[0028] The present invention adopts the above technical solution, which has the following advantages:

[0029] 1. The present invention adopts an inspection robot running on an inspection track to automatically detect the belt conveyor rollers, avoiding manual inspection, reducing workload and labor costs, and adopts a tunnel conveyor belt roller image acquisition system to collect roller data and analyze whether the rollers are faulty. Finally, when the robot moves to the end, the inspection report is transmitted to the platform through wireless transmission, thereby improving the practicality of the system.

[0030] 2. The present invention establishes a tunnel belt roller fault detection model based on deep learning, which further improves the accuracy and efficiency of fault detection, and performs significant target detection and roller speed type detection on the roller, which can discover the latent fault of the roller and avoid further development and fire, thereby improving the practicality and timeliness of the method. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 This is a principle block diagram of a TBM conveyor belt roller fault detection system based on a patrol robot in an embodiment of the present invention;

[0032] Figure 2 Schematic diagram of the overall structure of the inspection robot in an embodiment of the present invention;

[0033] Figure 3 is a schematic diagram of the structure inside the inspection robot box in an embodiment of the present invention;

[0034] Figure 4 2 is a schematic diagram of the suspension walking structure of the inspection robot in an embodiment of the present invention;

[0035] Figure 5 1 is a schematic flow chart of a belt conveyor roller inspection method according to an embodiment of the present invention;

[0036] Figure 6 1 is a diagram of the infrared roller positioning detection process in an embodiment of the present invention;

[0037] Figure 7 is a structural diagram of a roller salient target detection model in an embodiment of the present invention;

[0038] Figure 8 is a structural diagram of a fast feature fusion module in an embodiment of the present invention;

[0039] Fig. 9 is a model diagram for judging roller speed type faults in an embodiment of the present invention;

[0040] Fig.10 : is a yolov5s network structure diagram used in target detection in an embodiment of the present invention;

[0041] Fig.11 Schematic diagram of the installation of the driving wheel and the I-beam track in an embodiment of the present invention;

[0042] Fig.12 : is a network structure diagram of a CAFM module and a FFAM module in an embodiment of the present invention;

[0043] Description of Figure Numbers:

[0044] 1- waterproof outer shell box, 2- upper connecting plate, 3- I-beam track, 4- driven wheel, 5- RFID positioning sensor, 6- driving wheel, 7- bearing, 8- obstacle avoidance sensor, 9- dual-mode camera, 10- driving motor, 11- sprocket, 12- battery, 13- wireless charging device, 14- environmental monitoring component, 15- PLC controller, 16- industrial computer, 17- guide wheel, 18- vertical hanging plate, 19- driven shaft, 20- connecting hanging plate, 21- spring. DETAILED DESCRIPTION

[0045] In order to solve the problems that the detection methods in the prior art are insufficient and easily affected by other surrounding heating equipment, further affecting the reliability and accuracy of the detection, the present invention provides a belt roller fault detection method and system based on a TBM long tunnel inspection robot. The present invention obtains the image data and video data of the TBM belt roller through the inspection robot, and locates the roller position after preprocessing the image data; the preprocessed image data is input into the constructed belt roller fault detection model, and the model is trained to obtain the faulty roller image; the roller speed type is detected according to the video data, and whether the roller speed is abnormal according to the roller speed type is determined to obtain the abnormal roller speed result; the state of the belt conveyor roller is comprehensively judged according to the obtained faulty roller image and the abnormal roller speed result, and the roller feature with dual fault indicators is obtained, and the image frame confirming the abnormality is output, and the capture time information of the image is associated at the same time to obtain the faulty roller with time information and position information.

[0046] The present invention is described in detail below with reference to the accompanying drawings and embodiments.

[0047] In order to make the purpose, technical solution and advantages of the embodiment of the present invention clearer, the technical solution of the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings of the embodiment of the present invention. Obviously, the described embodiment is a part of the embodiment of the present invention, not all of the embodiments. Based on the described embodiment of the present invention, all other embodiments obtained by ordinary technicians in this field belong to the scope of protection of the present invention.

[0048] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0049] Embodiment 1: In this embodiment, a TBM long tunnel inspection robot is provided, which is used to implement a belt roller fault detection method based on the TBM long tunnel inspection robot. Specifically, Figures 1 to 4 As shown, in this embodiment, the robot includes:

[0050] The box body has a waterproof shell 1, in which a driving motor 10 and a control system are arranged, and the driving motor 10 is connected to the control system, and its operation is controlled by the control system;

[0051] The upper connecting plate 2 is arranged on the top of the box body. A driving wheel 6 and two sets of traveling mechanisms are arranged on one side of the upper connecting plate 2. The driving wheel 6 is arranged in the middle of the two sets of traveling mechanisms. The driving wheel 6 is driven by a driving motor 10, and then drives the two sets of traveling mechanisms to move along the I-beam track 3 suspended on the side of the TBM conveyor belt; wherein the driving wheel 6 is arranged on the lower surface of the I-beam track 3 and is tightly fitted with the lower surface to realize movement along the I-beam track 3;

[0052] like Figure 4 As shown, each set of traveling mechanisms includes a driven wheel 4, a guide wheel 17, a bearing 7 and a connecting plate 20; the two sets of traveling mechanisms include four driven wheels and eight guide wheels, and the four driven wheels play a supporting role.

[0053] The bottom of the connecting hanging plate 20 is installed on the upper connecting plate 2 through the bearing 7 to realize the steering of the robot; the connecting hanging plate 20 adopts a groove-type structure, and a driven wheel 4 is respectively arranged on the inner side of the vertical hanging plates 18 on both sides of the connecting hanging plate 20 of the groove-type structure, and the two driven wheels 4 are respectively arranged on both sides of the waist of the I-beam track 3, so that the waist of the I-beam track 3 is between the two driven wheels 4, and the I-beam track 3 is located in the groove-type structure of the connecting hanging plate 20, so that the box body is hung on the I-beam track 3 through the driven wheel 4;

[0054] On the two side vertical hanging plates 18 of the connecting hanging plate 20, two guide wheels 17 are respectively arranged on both sides of the driven wheel, and the guide wheels 17 are in contact with the side edges of the rail waist of the I-beam rail 3 to achieve the movement along the rail in contact with the I-beam rail 3 all the time;

[0055] An RFID positioning sensor 5 is provided on the upper connecting plate 2; the RFID positioning sensor 5 cooperates with the electronic tags arranged at intervals on the side of the I-beam rail 3 to transmit the read tag information to the control system; preferably, the read tag information is stored in the control system in a table form;

[0056] A dual-mode camera 9 is arranged at the front of the box body for collecting image data of the TBM belt roller and transmitting it to the control system.

[0057] In this embodiment, preferably, the I-beam track 3 is made of No. 10 H steel, and a plurality of electronic tags are arranged at intervals on the side of the I-beam track 3; the distance between adjacent electronic tags is arranged correspondingly according to the number of rollers to be detected.

[0058] In this embodiment, preferably, the bearing 7 adopts a suspension bearing, which can help the robot to achieve flexible steering and easily pass through curves with a small curvature radius.

[0059] When in use, the I-beam track 3 is set on the side of the belt roller, so that the robot collects corresponding roller data along the I-beam track 3, pre-processes and detects and analyzes the collected data through the control system, and outputs a test report list.

[0060] In the above embodiment, Figure 1 , Figure 3 As shown, the control system includes an industrial computer 16 for data processing and a PLC controller 15 for controlling the speed and direction of the robot movement. The industrial computer 16 is used to receive the image data transmitted by the dual-mode camera 9, and analyze and process the image data and transmit the result to the PLC controller. The industrial computer 16 exchanges information with the monitoring platform through a wireless network. The PLC controller controls the motor 10 to work through the drive module, and collects data from the RFID positioning sensor 5, the obstacle avoidance sensor 8 and the environmental monitoring sensor assembly 14.

[0061] In the above embodiment, a battery pack 12 is arranged in the box, and the battery pack 12 can be charged by a wireless charging device 13 arranged on the side of the box; the wireless charging device 13 is monitored by a PLC controller, and the wireless charging device 13 can be charged by a charging pile arranged at the end of the I-beam rail 3.

[0062] The wireless charging device 13 integrates a wireless charger, a photoelectric sensor and a relay. When the robot reaches the preset charging pile position, the photoelectric sensor responds immediately and sends a signal to the PLC controller 5 to interrupt the motor power supply, so that the robot stops steadily and locks, and then through the set timing control, the relay connects the circuit for charging the battery. The PLC controller 15 monitors the charging and discharging status of the battery 12 throughout the process to ensure that it switches back to the motor power supply mode in time after the battery is fully charged, and after a proper delay, the robot automatically starts in reverse to continue its inspection task.

[0063] Optionally, the battery pack 12 uses lithium iron phosphate batteries.

[0064] In the above embodiment, Fig.11 As shown, the clamping force between the driving wheel 6 and the lower surface of the I-beam track 3 is adjusted by a spring. The spring 21 is arranged between the driving wheel bracket and the upper connecting plate 2 and is sleeved on the bolt. An anti-loosening setting is adopted on the connecting bolts used to connect the driving wheel bracket and the upper connecting plate 2, so that the driving wheel and the lower surface of the I-beam track 3 can always fit tightly to ensure power transmission.

[0065] like Figure 2 , Figure 3 As shown, an obstacle avoidance sensor 8 is arranged at the front of the box, and an environmental monitoring component 14 is arranged inside the box. The obstacle avoidance sensor 8 and the environmental monitoring sensor 14 transmit the data collected to the control system.

[0066] The environmental monitoring component 14 includes a variety of sensors such as temperature and humidity, carbon monoxide, carbon dioxide, and oxygen concentration. These sensors are connected to the PLC controller 15 to continuously monitor and feedback data in real time. Once it is detected that the temperature and humidity or carbon monoxide concentration deviates from the preset range of the safety standard, an alarm is immediately triggered to warn of possible safety hazards.

[0067] In the above embodiment, Figure 3 As shown, a sprocket 11 is provided on the driving shaft of the driving wheel 6 and the motor shaft of the driving motor 10 ; the driving motor 10 and the driving wheel 6 are connected by chain drive via the sprocket 11 , thereby transmitting power to the driving wheel 6 .

[0068] In the above embodiment, the dual-mode camera 9 is preferably a hemispherical dual-spectrum camera, which is arranged facing the rollers to collect the TBM belt conveyor roller data in real time and capture the key data information of the rollers.

[0069] In this embodiment, the dual-mode camera 9 is installed on the front of the box through a camera bracket. The dual-mode camera 9 has the dual ability to collect visible light and thermal infrared images. It can not only capture the real-time image of the roller, but also reveal the temperature extremes in the thermal infrared image, that is, the highest temperature and the lowest temperature. These roller image data are transmitted to the industrial computer 16 inside the robot via a network cable for in-depth data analysis and fault identification. The control system not only accurately diagnoses the image of the existing roller problem, but also synchronously records the timestamp of the image acquisition, thereby accurately locating the specific coordinates of the faulty roller, and all information is integrated into a detailed inspection report. When the inspection robot moves to the terminal of track 3, the report will be transmitted to the robot management platform wirelessly, and the operation and maintenance personnel will quickly organize precise maintenance operations based on the fault points clearly marked in the report.

[0070] In the above embodiment, an operation screen connected to the control system is also provided on the box. The interface of the operation screen is uniformly integrated in the console developed in C#, and has the functions of robot motion control, camera management and positioning. Through this interface, users can easily grasp the start and stop of the robot, speed adjustment, and real-time operating parameters of the motor, such as speed, current, voltage, and remaining power. In addition, the interface supports real-time preview of camera images and tag information read by RFID, allowing users to customize parameters and paths to capture specific images or record videos. All these data will be automatically archived to the specified folder, which will facilitate subsequent in-depth analysis and fault diagnosis.

[0071] When the robot of the present invention is in use, the robot is controlled and positioned through the PLC controller 15 and the RFID card reader, and the PLC controller 15 sends motion instructions to control the forward and reverse rotation of the servo motor 10 and the motor speed to achieve the reciprocating motion and speed control of the robot. At the same time, the RFID positioning sensor 5 is connected to the PLC controller 15. When the RFID positioning sensor 5 reads the electronic tag on the track 3, the read tag serial number and reading time are saved to obtain the current position of the robot.

[0072] Embodiment 2: In this embodiment, a belt roller fault detection method based on a TBM long tunnel inspection robot is provided, and the method is implemented based on the TBM long tunnel inspection robot in the above embodiments. The inspection robot collects relevant data of the belt conveyor roller in real time, and then preprocesses these raw data to build a fault detection model for the belt conveyor roller. After the model training is completed, the optimal training weights and their corresponding model files are loaded onto the robot's built-in development board to perform real-time preprocessing and detection on the acquired roller data. Once the image evidence of the faulty roller is detected, it is immediately output and recorded, and the specific position captured by the image of the faulty roller is analyzed to accurately locate the roller area where the fault occurs. When the robot drives to the end of the track, the complete detection report is sent wirelessly to the external monitoring platform. The platform arranges the corresponding maintenance team to repair the faulty roller in a timely manner accordingly.

[0073] Specifically, in this embodiment, Figure 5 As shown, the method comprises the following steps:

[0074] 1) Obtain image data and video data of TBM belt rollers through inspection robots, and locate the rollers after preprocessing the image data;

[0075] 2) Inputting the preprocessed image data into the constructed belt conveyor roller fault detection model, training the model, obtaining the faulty roller image, and determining the position information of the faulty roller according to the faulty roller image;

[0076] 3) Detect the speed type of the roller according to the video data, determine whether the speed of the roller is abnormal according to the speed type of the roller, and obtain the abnormal speed result of the roller;

[0077] 4) The state of the belt roller is comprehensively judged based on the obtained faulty roller image and the abnormal rotation speed result of the roller, a roller with dual fault indicators is obtained, and an image frame confirming the abnormality is output, and the capture time information of the image is associated at the same time to obtain a faulty roller with time information and position information.

[0078] In the above step 1), the image data and video data of the TBM belt roller are obtained by the inspection robot, and the image data is preprocessed, such as Figure 6 As shown, the following steps are included:

[0079] 1.1) The inspection robot inspects along the track direction to obtain image data and video data of the TBM belt roller; wherein the image data includes RGB images and infrared temperature images;

[0080] 1.2) compressing the RGB image and the infrared temperature image to a preset size so that the pixels and sizes of the two images are consistent;

[0081] In this embodiment, due to the difference in focal length and resolution between thermal infrared and visible light images, and the resulting field of view mismatch problem, it is necessary to perform fine calibration and alignment on these two types of images. Specifically, by cutting the RGB image to the field of view boundary of the temperature image, the pixels and sizes of the two are consistent, ensuring that the two are accurately aligned. After this step is completed, the optimized thermal infrared and visible light images are obtained.

[0082] 1.3) The compressed temperature image data is used for target detection, irrelevant background information is filtered out, and the position of the roller is located.

[0083] In this embodiment, the target detection uses the existing yolov5s to detect the roller and its position. Its network structure is as follows: Fig.10 Specifically, the target detection method includes the following steps:

[0084] 1.3.1) Input module: Input the infrared temperature image compressed to a specified size into the target detection model;

[0085] Backbone module: extracts key features from the input image;

[0086] 1.3.2) Neck module: Through a series of network layers, the features extracted by Backbone are further processed. The specific steps are as follows:

[0087] Use the Concat operation and the C3 network layer to fuse features, and use this operation multiple times;

[0088] Multi-scale features of the image are extracted through a fast version of the spatial pyramid pooling method.

[0089] 1.3.3) After multiple upsampling operations, the Concat operation and C3 network layer are used again to perform multiple feature fusion on the image.

[0090] 1.3.4) Head module: Target detection is performed based on the features output by the Neck part, including classification of faulty rollers and positioning of bounding boxes.

[0091] 1.3.5) Output the detected target roller image.

[0092] In the above step 2), for the abnormal temperature detection of the roller, the roller positioning is performed on the thermal infrared image of the roller (i.e., the temperature image) to enhance the accuracy and focus of the detection: the pre-processed image data is input into the constructed belt roller fault detection model, the model is trained, and the faulty roller image is obtained, specifically:

[0093] The belt roller fault detection model is constructed using the salient target detection algorithm. The preprocessed image data is used as the input data of the belt roller fault detection model. The model is trained to obtain the optimal training weight. The training weight is used to test the roller fault detection model to obtain the faulty roller image.

[0094] Among them, the network structure of the belt roller fault detection model includes an input end, an encoder, a decoder, and an output end, and the encoding stage uses a fast feature aggregation module and a cross-attention fusion module.

[0095] In this embodiment, Figure 7 As shown in the figure, a belt roller fault detection model is constructed using a salient target detection algorithm, which includes the following steps:

[0096] 2.1) Encode the input RGB image and temperature image, obtain high-level feature maps and low-level feature maps through the CAFM module and FFAM module, and fuse feature maps of different levels through the cross-modal interaction module;

[0097] In this embodiment, specifically, the input RGB image and temperature image are encoded by inputting two VGG backbones for the RGB image and the temperature image respectively for downsampling;

[0098] Through two VGG five convolutional blocks E i R and E i T (i=1,2,3,4,5), five feature maps with different resolutions are obtained; R represents RGB, T represents temperature;

[0099] For the different features contained in the high-level feature maps and the low-level feature maps, two cross-modal interaction modules are used to fuse the feature maps at different levels.

[0100] 2.2) Decode the fused feature map, and combine the feature map output by the CAFM module and the FFAM module with the features extracted after decoding to obtain the final abnormal temperature detection result of the roller, which can effectively reveal potential temperature abnormality problems;

[0101] In this embodiment, specifically, the fused feature map is decoded by setting five deconvolution blocks corresponding to the encoding layer, and each deconvolution block upsamples the feature map twice through a transposed convolution with learnable parameters.

[0102] Among them, the CAFM module is used to explore the complementarity between different modalities. The last three layers of high-level semantic features. In order to balance the overall efficiency, the FFAM module is used to complete the fusion and enhancement of the first two layers of low-level features. The features f from the two encoder branches are combined into a i R and f i T , i = 3, 4, 5; transmitted to the CAFM module, which is based on the Transformer mechanism and allows the feature maps of the two modalities to explore effective complementary information with each other and jointly perform the final salient target prediction through continuous training iterations. The features of the first two layers f i R and f i T , i = 1, 2; transmitted to the FFAM module to complete the perception and fusion of modal information. For the feature maps at the lower layers of the network structure, which contain information such as contours, edges, colors, textures, and shape features, particularly complex extraction operations will lead to a significant increase in parameters and computational consumption. In order to balance the efficiency of the model, a fast feature aggregation module FFAM is proposed to integrate and fuse the underlying feature maps in the network architecture, such as Figure 8 As shown in the figure. Network layers with smaller convolution kernels usually need to reduce the image resolution through pooling layers to expand the receptive field, which will inevitably lead to information loss. Dilated convolution increases the receptive field by introducing a dilation rate, but it will destroy the continuity between adjacent pixels. The fast aggregation module combines the advantages and disadvantages of both and completes the extraction of different receptive field features through two convolution branches with different dilation rates. Figure 8 The upper part of the figure shows a convolution branch with a dilation rate of 1, which aims to extract local information of the features, while the branch in the lower part has a dilation rate of 3 for the first convolution layer, which aims to capture relatively more macro feature information. BN with a ReLU layer is added after each convolution layer to improve the training speed and nonlinearity of the network. Subsequently, the output of the module is connected with the output of the corresponding decoder. Through this module, information from the underlying features can be aggregated into the final salient image. This module does not significantly increase the computational consumption or the number of network model parameters.

[0103] The FFAM module (cross-attention fusion module) is used to fuse the feature maps of the two modalities instead of directly adding and fusing them, such as Fig.12 As shown in Figure 2, the module consists of two inputs: an intermediate feature map f of size C×H×W from RGB and a hot branch. iR and f i T . First, the dimensions of the length and width of the image are flattened, and the number of channels of the feature map is reduced by half to reduce the amount of computation and network parameters. Then, a linear layer is used to generate queries r and kt, keywords kr and kt, and values ​​vr and vt of the two modalities, and they are input into a hybrid multi-head attention module developed by the self-attention mechanism. Hybrid multi-head attention is the integration of multiple self-attentions, producing multiple subspaces, splicing multiple attention matrices into a huge matrix, enabling the model to focus on different information elements. Then, residual connections and normalization are added. The goal of residual connections is to prevent degradation in deep neural network training. Normalization is used to accelerate training and increase training stability. Layer normalization (LN) is normalized across neurons in the same sample, while batch normalization (BN) is normalized across neurons in the same position in samples in the same batch.

[0104] In the above step 2), the position information of the faulty roller is determined according to the faulty roller image, specifically:

[0105] After acquiring the image of the faulty roller, determine which two RFID tags’ reading time range the grab time is within based on the grab time corresponding to the faulty roller image, and then determine the positions of the two RFID tags; determine the exact position of the faulty roller based on the positions of the two RFID tags and the time difference.

[0106] The time difference refers to the time it takes to read two RFID tags. Specifically:

[0107] First, the temperature anomaly image is detected based on the significant target. The image corresponds to a time. By comparing the time when the image is obtained between the times when two RFID tags are read, it can be determined that the faulty roller is within this interval. The specific roller is obtained by subtracting the time when the RFID tag is first read in this interval from the fault image time to obtain the time difference t. The distance between the faulty roller and the RFID tag can be obtained through the robot's movement speed v×t (for example, the RFID tag interval is set to L).

[0108] In the above step 3), the speed type of the roller is detected according to the video data, and whether the speed of the roller is abnormal is determined according to the speed type of the roller, and the abnormal speed result of the roller is obtained, such as Fig. 9 As shown, the following steps are included:

[0109] 3.1) Multiple adjacent frames of images in the video data are sequentially subjected to convolution and pooling operations, and then to full connection operations to obtain the speed classification category of the roller;

[0110] In this embodiment, K frames of images are preferably used to fully learn relevant features, and multiple 3D convolution (convolution kernel size is 3×3×3) and pooling (convolution kernel size is 2×2×2) operations are performed in sequence, and then multiple full connection operations are performed to obtain the rotation speed classification category of the roller.

[0111] For example, 16 adjacent frames of images in the video data are sequentially subjected to five three-dimensional convolution and pooling operations, and then three full-connection operations to obtain the rotation speed classification category of the roller.

[0112] In this embodiment, 16 adjacent frames of images in the video data are processed. The algorithm also supports other frame numbers, but a frame number higher than 16 will reduce the video processing efficiency, and a frame number lower than 16 cannot fully learn the relevant features of the roller, so 16 frames is preferred.

[0113] 3.2) The linear speed of the roller is obtained by the obtained classification category of the roller speed. At the same time, the linear speed of the belt can be obtained by giving the belt conveyor motor a rotation speed. The linear speed of the roller calculated by the roller speed is compared with the linear speed of the belt. When the ratio of the roller linear speed to the belt linear speed is less than the set value, it can be judged that the roller speed is abnormal. The obtained classification speed is further judged by speed identification to obtain the roller with abnormal speed.

[0114] The set value is set according to actual use and experience, and is not limited here. When the roller fails or the roller rotates very slowly, the roller line speed is much lower than the belt line speed.

[0115] In this embodiment, the speed identification is specifically as follows: predicting the speed type of the roller and determining the speed category of the roller, wherein the speeds of the rollers are set to be v1, v2, v3, v4, and v5. By comparing the classified speeds with the actual speed v0, the speed loss rate of the roller (v0-v i ) / v0, where (i=1,2,3,4,5), when the loss rate is greater than the preset value m, it can be considered that the speed is abnormal.

[0116] In the above step 4), the state of the belt idler is comprehensively judged based on the obtained faulty idler image and the abnormal result of the idler speed, including the following steps:

[0117] 4.1) If the faulty roller image and the abnormal rotation speed result of the roller appear at the same time, the faulty roller image is output;

[0118] 4.2) If only the image of the faulty roller or the abnormal speed result of the roller appears, the next inspection cycle will focus on the fault location; at this time, a fault warning is output, and no fault image is output. The final test report is determined by temperature detection and speed detection. A single method may misjudge the fault. The detection accuracy can be greatly improved by combining the two methods (using the RGB-T method as the main method and the abnormal speed detection as the auxiliary judgment method). The fault image is output, but the final fault is detected by both methods, and the faulty roller is found.

[0119] If only the image of the faulty roller or the abnormal result of the roller speed appears, an image warning will be output, and the fault image will not be output. There is a possibility of misjudgment through only a single fault. Only when the fault is detected by both temperature detection and speed detection methods can the faulty roller be found. Therefore, only a fault warning is output, and the fault location is detected in the next inspection cycle. If one of the following two situations occurs in the second detection, both methods detect the fault, and the fault in the previous detection still exists, the faulty roller is found and the faulty roller image is output.

[0120] Among them, the image data are named according to the time when the image is captured. The faulty roller with time information and position information is obtained through the image name of the faulty roller and the time information of the RFID tag reading;

[0121] 4.3) The fault information is organized into a fault report, and the monitoring report is output when the inspection robot moves to the end of the track. In summary, through the joint analysis of the abnormal temperature detection results and the abnormal speed detection results of the roller, the roller with dual fault indicators can be accurately identified, and the image frame confirming the abnormality can be output. At the same time, the capture time information of the image is associated to accurately determine when and where the robot detected the faulty roller. In this way, the specific location of the faulty roller can be accurately located, and a detailed report can be issued to the outside world to guide maintenance personnel to quickly go to the corresponding location to carry out maintenance work.

[0122] In one embodiment of the present invention, a belt roller fault detection system based on a TBM long tunnel inspection robot is provided, which comprises:

[0123] The data processing module pre-processes the image data and video data of the TBM belt roller obtained by the inspection robot to locate the roller position;

[0124] The target detection module inputs the preprocessed image data into the constructed belt roller fault detection model, trains the model, obtains the fault roller image, and determines the position information of the fault roller according to the fault roller image;

[0125] The speed detection module detects the speed type of the roller according to the video data, determines whether the speed of the roller is abnormal according to the speed type of the roller, and obtains the abnormal speed result of the roller;

[0126] The comprehensive judgment module comprehensively judges the state of the belt idler based on the obtained faulty roller image and the abnormal rotation speed result of the idler, obtains the idler with dual fault indicators, outputs the image frame confirming the abnormality, and associates the capture time information of the image to obtain the faulty roller with time information and position information.

[0127] In the above embodiment, the inspection robot obtains the image data and video data of the TBM belt roller, and pre-processes the image data, including:

[0128] The inspection robot inspects along the track direction to obtain image data and video data of the TBM belt roller; wherein the image data includes RGB images and infrared temperature images;

[0129] Compress the RGB image and the infrared temperature image to a preset size so that the sizes of the two images are consistent;

[0130] Perform target detection on the compressed temperature image data, filter out irrelevant background information, and locate the position of the roller.

[0131] In the above embodiment, the pre-processed image data is input into the constructed belt roller fault detection model, and the model is trained to obtain the fault roller image, including:

[0132] The belt roller fault detection model is constructed by using the salient target detection algorithm. The preprocessed image data is used as the input data of the belt roller fault detection model. The model is trained to obtain the optimal training weight. The training weight is used to test the roller fault detection model to obtain the faulty roller image, including:

[0133] Encode the input RGB image and temperature image, obtain high-level feature maps and low-level feature maps through the CAFM module and FFAM module, and fuse feature maps of different levels through the cross-modal interaction module;

[0134] The fused feature map is decoded, and the feature map output by the CAFM module and FFAM is combined with the features extracted after decoding to obtain the final roller detection result.

[0135] In the above embodiment, determining the position information of the faulty roller according to the faulty roller image includes:

[0136] After acquiring the image of the faulty roller, determine which two RFID tags’ reading time range the grab time is within based on the grab time corresponding to the faulty roller image, and then determine the positions of the two RFID tags; determine the exact position of the faulty roller based on the positions of the two RFID tags and the time difference.

[0137] In the above embodiment, the rotation speed type of the roller is detected according to the video data, and whether the rotation speed of the roller is abnormal is determined according to the rotation speed type of the roller, and the abnormal rotation speed result of the roller is obtained, including:

[0138] The adjacent multiple frames of images in the video data are sequentially subjected to convolution and pooling operations, and then to full connection operations to obtain the speed classification category of the roller;

[0139] The rotational linear speed of the roller is obtained by the obtained classification category of the roller speed. At the same time, the linear speed of the belt is obtained by giving a rotational speed of the belt conveyor motor. The linear speed of the roller is compared with the linear speed of the belt. When the ratio of the linear speed of the roller to the linear speed of the belt is less than the set value, the roller speed is judged to be abnormal, and the obtained classification speed is further judged by speed identification to obtain the roller with abnormal speed.

[0140] In the above embodiment, comprehensively judging the state of the belt idler according to the obtained faulty idler image and the abnormal result of the rotation speed of the idler includes:

[0141] If the faulty roller image and the abnormal rotation speed result of the roller appear at the same time, the faulty roller image is output;

[0142] If only the image of the faulty roller or the abnormal rotation speed of the roller appears, focus on the fault location in the next inspection cycle;

[0143] The image data are named according to the time when the image is captured, and the faulty roller with time information and position information is obtained through the faulty roller image name;

[0144] The fault information is organized into a fault report, and the monitoring report is output when the inspection robot moves to the end of the track.

[0145] The system provided in this embodiment is used to execute the above-mentioned method embodiments. Please refer to the above-mentioned embodiments for specific processes and detailed contents, which will not be repeated here.

[0146] In the computing device structure provided in one embodiment of the present invention, the computing device may be a terminal, which may include: a processor, a communications interface, a memory, a display screen and an input device. Among them, the processor, the communications interface and the memory communicate with each other through a communication bus. The processor is used to provide computing and control capabilities. The memory includes a non-volatile storage medium and an internal memory, the non-volatile storage medium stores an operating system and a computer program, and the computer program is executed by the processor to implement the above method; the internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be realized through WIFI, NFC (near field communication) or other technologies. The display screen may be a liquid crystal display screen or an electronic ink display screen, and the input device may be a touch layer covered on the display screen, or a key, trackball or touchpad provided on the housing of the computing device, or an external keyboard, touchpad or mouse, etc. The processor may call the logic instructions in the memory.

[0147] In addition, the logic instructions in the above-mentioned memory can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0148] In one embodiment of the present invention, a computer program product is provided, wherein the computer program product includes a computer program stored on a non-transitory computer-readable storage medium, wherein the computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the methods provided by the above-mentioned method embodiments.

[0149] In one embodiment of the present invention, a non-transitory computer-readable storage medium is provided, wherein the non-transitory computer-readable storage medium stores server instructions, and the computer instructions enable a computer to execute the methods provided in the above embodiments.

[0150] The above embodiment provides a computer-readable storage medium, whose implementation principle and technical effect are similar to those of the above method embodiment, and will not be repeated here.

[0151] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0152] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0153] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0154] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A belt roller fault detection method based on an inspection robot, characterized in that: include: The inspection robot obtains the image data and video data of the TBM belt roller, and pre-processes the image data to locate the roller position; The preprocessed image data is input into the constructed belt conveyor roller fault detection model, the model is trained to obtain the faulty roller image, and the position information of the faulty roller is determined according to the faulty roller image; The speed type of the roller is detected according to the video data, and whether the speed of the roller is abnormal is determined according to the speed type of the roller, and the abnormal speed result of the roller is obtained; The state of the belt idler is comprehensively judged based on the obtained faulty roller image and the abnormal rotation speed result of the roller, and a roller with dual fault indicators is obtained. An image frame confirming the abnormality is output, and the capture time information of the image is associated at the same time to obtain a faulty roller with time information and position information.

2. The belt roller fault detection method based on the inspection robot according to claim 1 is characterized in that: The inspection robot obtains the image data and video data of the TBM belt roller and pre-processes the image data, including: The inspection robot inspects along the track direction to obtain image data and video data of the TBM belt roller; wherein the image data includes RGB images and infrared temperature images; Compress the RGB image and the infrared temperature image to a preset size so that the sizes of the two images are consistent; Perform target detection on the compressed temperature image data, filter out irrelevant background information, and locate the position of the roller.

3. The belt roller fault detection method based on the inspection robot according to claim 1 is characterized in that: The preprocessed image data is input into the constructed belt roller fault detection model, and the model is trained to obtain the fault roller image, including: The belt roller fault detection model is constructed by using the salient target detection algorithm. The preprocessed image data is used as the input data of the belt roller fault detection model. The model is trained to obtain the optimal training weight. The training weight is used to test the roller fault detection model to obtain the faulty roller image, including: Encode the input RGB image and temperature image, obtain high-level feature maps and low-level feature maps through the CAFM module and FFAM module, and fuse feature maps of different levels through the cross-modal interaction module; The fused feature map is decoded, and the feature map output by the CAFM module and FFAM is combined with the features extracted after decoding to obtain the final abnormal temperature detection result of the roller.

4. The belt roller fault detection method based on the inspection robot as claimed in claim 1 is characterized in that: Determine the position information of the faulty roller according to the faulty roller image, including: After acquiring the image of the faulty roller, determine which two RFID tags’ reading time range the grab time is within based on the grab time corresponding to the faulty roller image, and then determine the positions of the two RFID tags; determine the exact position of the faulty roller based on the positions of the two RFID tags and the time difference.

5. The belt roller fault detection method based on the inspection robot as claimed in claim 1, characterized in that: The speed type of the roller is detected according to the video data, and whether the speed of the roller is abnormal is determined according to the speed type of the roller, and the abnormal speed result of the roller is obtained, including: The adjacent multiple frames of images in the video data are sequentially subjected to convolution and pooling operations, and then to full connection operations to obtain the speed classification category of the roller; The rotational linear speed of the roller is obtained by the obtained classification category of the roller speed. At the same time, the linear speed of the belt is obtained by giving a rotational speed of the belt conveyor motor. The linear speed of the roller is compared with the linear speed of the belt. When the ratio of the linear speed of the roller to the linear speed of the belt is less than the set value, the roller speed is judged to be abnormal, and the obtained classification speed is further judged by speed identification to obtain the roller with abnormal speed.

6. The belt roller fault detection method based on the inspection robot according to claim 1, characterized in that: The status of the belt idler is comprehensively judged based on the faulty idler image and the abnormal speed result of the idler, including: If the faulty roller image and the abnormal rotation speed result of the roller appear at the same time, the faulty roller image is output; If only the image of the faulty roller or the abnormal rotation speed of the roller appears, the fault position will be inspected in the next inspection cycle, and a fault warning will be output at this time, but the fault image will not be output; The image data are named according to the time when the image is captured, and the faulty roller with time information and position information is obtained through the faulty roller image name; The fault information is organized into a fault report, and the monitoring report is output when the inspection robot moves to the end of the track.

7. An inspection robot, used to implement the belt roller fault detection method based on a TBM long tunnel inspection robot as described in any one of claims 1 to 6, characterized in that: include: The box has a waterproof shell, in which a driving motor and a control system are arranged, and the driving motor is connected to the control system, and its operation is controlled by the control system; The upper connecting plate is arranged on the top of the box body, and a driving wheel and two sets of traveling mechanisms are arranged on one side of the upper connecting plate. The driving wheel is arranged in the middle of the two sets of traveling mechanisms, and the driving wheel is driven by a driving motor, thereby driving the two sets of traveling mechanisms to move along the I-beam track suspended on the side of the TBM conveyor belt; wherein the driving wheel is arranged on the lower surface of the I-beam track and fits closely to the lower surface to realize movement along the I-beam track; Each walking mechanism includes a driven wheel, a guide wheel, a bearing and a connecting plate; the bottom of the connecting plate is mounted on the upper connecting plate through a bearing to realize the steering of the robot; the connecting plate adopts a groove structure, and a driven wheel is respectively arranged on the inner side of the vertical plates on both sides of the connecting plate of the groove structure, and the waist of the I-beam track is arranged between the two driven wheels, so that the box can be hung on the I-beam track through the driven wheel; On the vertical hanging plates on both sides of the connecting hanging plate, two guide wheels are respectively arranged on both sides of the driven wheel, and the guide wheels are fitted with the side edges of the I-beam track to achieve the movement of always fitting with the I-beam track; An RFID positioning sensor is arranged on the upper connecting plate; the RFID positioning sensor cooperates with each electronic tag arranged at intervals on the side of the I-beam track to transmit the read tag information to the control system; A dual-mode camera is arranged at the front of the box to collect image data of the TBM belt rollers and transmit it to the control system.

8. The inspection robot according to claim 7, characterized in that: The clamping force between the driving wheel and the lower surface of the I-beam track is adjusted by a spring, which is arranged between the driving wheel bracket and the upper connecting plate and sleeved on the bolts; an anti-loosening device is adopted on the connecting bolts used to connect the driving wheel bracket and the upper connecting plate, so that the driving wheel and the lower surface of the I-beam track can always fit tightly to ensure power transmission; An obstacle avoidance sensor is arranged at the front of the box, and an environmental monitoring sensor is arranged inside the box. The obstacle avoidance sensor and the environmental monitoring sensor transmit the collected data to the control system.

9. A belt roller fault detection system based on a detection robot, characterized in that: include: The data processing module pre-processes the image data and video data of the TBM belt roller obtained by the inspection robot to locate the roller position; The target detection module inputs the preprocessed image data into the constructed belt roller fault detection model, trains the model, obtains the fault roller image, and determines the position information of the fault roller according to the fault roller image; The speed detection module detects the speed type of the roller according to the video data, determines whether the speed of the roller is abnormal according to the speed type of the roller, and obtains the abnormal speed result of the roller; The comprehensive judgment module comprehensively judges the state of the belt idler based on the obtained faulty roller image and the abnormal rotation speed result of the idler, obtains the idler with dual fault indicators, outputs the image frame confirming the abnormality, and associates the capture time information of the image to obtain the faulty roller with time information and position information.

10. A computer-readable storage medium storing one or more programs, characterized in that: The one or more programs include instructions, which, when executed by a computing device, cause the computing device to perform any one of the methods of claims 1 to 6.

Images