A safety inspection robot applied to cloud construction factory

Abstract

The application relates to a safety inspection robot applied to a cloud construction factory, relates to the technical field of intelligent sensing and control equipment, and comprises a track assembly and an inspection robot body arranged on the track assembly, wherein the track assembly comprises a plurality of transverse tracks, the plurality of transverse tracks are arranged at different heights respectively, and the inspection robot body walks on the transverse tracks; the inspection robot body comprises a fuselage, a multi-sensor fusion sensing module and a controller; the multi-sensor fusion sensing module is used for collecting multi-dimensional data; the controller is used for track detection, automatic obstacle avoidance, construction safety supervision and construction progress and quality identification according to the multi-dimensional data; at least part of the multi-sensor fusion sensing module is arranged on the fuselage; the controller is further used for interfacing with an attendance system to perform personnel management; and the controller is further used for data interaction with a cloud construction factory management platform, so that the inspection automation level of the cloud construction factory is improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent sensing and control equipment technology, and in particular to a safety inspection robot for building factories in the cloud. Background Technology

[0002] With the continuous innovation of the construction industry, cloud-based construction factories are gradually emerging as a new construction model. This model achieves efficient and intensive construction by organizing and advancing the construction process in a vertical space. In a cloud-based construction factory, the building structure continuously rises as construction progresses, creating a dynamically changing construction environment. Due to the unique and complex nature of cloud-based construction factories, comprehensive, timely, and accurate safety inspections are crucial. Safety inspections can not only promptly identify potential safety hazards during construction, such as structural defects, equipment malfunctions, and violations of operating procedures, but also effectively prevent accidents, ensuring the safety of construction workers and the smooth progress of the project. However, existing inspection methods have revealed many problems and limitations when facing the unique environment of cloud-based construction factories, failing to meet the actual needs of safety inspections.

[0003] Currently, most inspection robots on the market are ground-based. These robots perform well in traditional flat construction sites, completing inspection tasks through pre-set paths or autonomous navigation systems. However, in cloud-based construction plants, as construction progresses and building floors rise, ground-based robots cannot keep up with the ascending structure to reach different floors for inspection. This results in blind spots in high-altitude work areas, making it impossible to perform real-time and effective safety monitoring of these critical parts, increasing the safety risks of working at heights. For example, during the steel structure installation phase of high-rise buildings, ground-based robots cannot inspect the welding quality and bolt connections at heights; any quality issues could lead to serious structural safety accidents.

[0004] Therefore, there is a need to provide a safety inspection robot for cloud-based factory construction to improve the level of inspection automation in cloud-based factory construction. Summary of the Invention

[0005] This invention provides a safety inspection robot for cloud-based construction factories, comprising a track assembly and a robot body mounted on the track assembly. The track assembly includes multiple transverse tracks at different heights, on which the robot body moves. The robot body includes a fuselage, a multi-sensor fusion sensing module, and a controller. The multi-sensor fusion sensing module collects multi-dimensional data, and the controller performs track detection, automatic obstacle avoidance, construction safety monitoring, and construction progress and quality identification based on the multi-dimensional data. At least a portion of the multi-sensor fusion sensing module is mounted on the fuselage. The controller also interfaces with an attendance system for personnel management and interacts with a cloud-based construction factory management platform.

[0006] Furthermore, the multi-sensor fusion perception module includes at least a lidar component, an ultrasonic sensor, and an image acquisition component. The multi-dimensional data includes obstacle distance information acquired by the lidar component and the ultrasonic sensor. The controller performs automatic obstacle avoidance based on the multi-dimensional data, including: determining the current obstacle avoidance distance threshold based on the obstacle avoidance scenario issued by the cloud-based factory management platform; determining whether to perform obstacle avoidance based on the obstacle distance information acquired by the lidar component and the ultrasonic sensor and the current obstacle avoidance distance threshold; if obstacle avoidance is determined, acquiring obstacle images through the image acquisition component to determine the obstacle type; and executing obstacle avoidance actions based on the obstacle type.

[0007] Furthermore, the multi-sensor fusion sensing module includes at least an infrared thermal imaging camera, and the multi-dimensional data includes infrared thermal images of the hot work area collected by the infrared thermal imaging camera; the controller performs construction safety supervision based on the multi-dimensional data, including: flame identification based on the infrared thermal images of the hot work area; and identification of violations based on images of the work area collected by the image acquisition component.

[0008] Furthermore, the controller identifies construction progress and quality based on the multi-dimensional data, including: acquiring multiple processes issued by the cloud-based construction factory management platform; acquiring images of each process through an image acquisition component to identify construction progress and quality, wherein the multi-dimensional data includes at least images of each process.

[0009] Furthermore, the controller interfaces with the attendance system to manage personnel, including: obtaining attendance information from the attendance system; acquiring personnel images through an image acquisition component; and managing personnel based on the personnel images and attendance information.

[0010] Furthermore, the multi-sensor fusion sensing module includes vibration monitoring devices installed at multiple different locations on the machine body. The installation positions of the multiple vibration monitoring devices are determined based on the following process: determining multiple test transverse tracks, wherein the wear amount of the track grooves of any two transverse tracks is different; determining multiple initial positions on the machine body, and installing vibration monitoring devices at each initial position; for each test transverse track, collecting vibration monitoring data corresponding to each initial position through the vibration monitoring devices installed at each initial position; filtering valid test transverse tracks and valid initial positions based on the vibration monitoring data corresponding to each initial position for each test transverse track; filtering the installation positions of the multiple vibration monitoring devices based on the vibration monitoring data corresponding to each valid initial position for each valid test transverse track.

[0011] Furthermore, the multi-dimensional data includes vibration monitoring data collected by each vibration monitoring device; the controller performs track detection based on the multi-dimensional data, including: extracting vibration monitoring data corresponding to the installation position of each vibration monitoring device for each valid test transverse track from the vibration monitoring data corresponding to each valid initial position; determining the vibration characteristics corresponding to different track groove wear amounts based on the vibration monitoring data corresponding to the installation position of each vibration monitoring device for each valid test transverse track; acquiring historical wear data of the transverse track; determining multiple key wear positions of the transverse track and wear correlation characteristics between multiple key wear positions based on the historical wear data of the transverse track; and performing track detection based on the vibration monitoring data collected by each vibration monitoring device corresponding to each key wear position and the wear correlation characteristics between multiple key wear positions.

[0012] Furthermore, the controller performs track detection based on vibration monitoring data collected by each vibration monitoring device corresponding to each critical wear location and wear correlation characteristics between multiple critical wear locations. This includes: predicting the wear risk of each critical wear location based on vibration monitoring data collected by each vibration monitoring device corresponding to each critical wear location and wear correlation characteristics between multiple critical wear locations; filtering risk locations from multiple critical wear locations based on the wear risk of each critical wear location; determining the optimal detection path based on the wear risk of multiple risk locations; and for each risk location on the optimal detection path, acquiring an image corresponding to the risk location through an image acquisition component, and determining the wear amount of the risk location based on the image corresponding to the risk location.

[0013] Furthermore, the multi-sensor fusion sensing module includes an RFID positioning component, wherein the RFID positioning component includes multiple RFID readers and RFID positioning tags installed on the device body, and the multi-dimensional data includes the device body position collected by the RFID positioning component; the controller performs track detection based on the multi-dimensional data, including: determining the positioning error based on multiple positioning test positions and the RFID positioning component.

[0014] Furthermore, the controller determines the positioning error based on multiple positioning test locations and the RFID positioning component, including: acquiring historical positioning error data; determining multiple target error locations based on the historical positioning error data; performing error correlation analysis and average deviation accuracy analysis on the multiple target error locations based on the historical positioning error data to obtain error correlation analysis results and average deviation accuracy analysis results; determining multiple positioning test locations based on the error correlation analysis results and average deviation accuracy analysis results; for each positioning test location, determining the positioning deviation corresponding to the positioning test location based on the body position collected by the RFID positioning component at the positioning test location; and determining the positioning error based on the positioning deviation corresponding to each positioning test location.

[0015] Compared to existing technologies, the safety inspection robot for building factories in the cloud, as described in this specification, has at least the following advantages: 1. The safety inspection robot's track assembly comprises multiple transverse tracks at varying heights, upon which the robot itself can move. This unique design breaks the limitations of traditional single-height inspections, constructing a comprehensive, three-dimensional inspection system. In cloud-based factory construction, different construction areas at different heights present varying safety hazards and monitoring needs. Low-height tracks facilitate detailed inspections of basic construction areas, such as ground material stacks and equipment foundation installations, promptly identifying issues like improper material placement and non-standard foundation installations. High-height tracks allow the robot to have a panoramic view of the entire construction site, providing macroscopic monitoring of high-altitude operations, such as steel structure erection and exterior wall construction, effectively preventing accidents such as falls from heights and being struck by objects. The combination of multiple-height tracks ensures comprehensive inspection without blind spots, providing a solid guarantee for safe production in cloud-based factory construction. 2. Multi-sensor fusion sensing module collects multi-dimensional data. A single sensor often only acquires limited information and is easily affected by environmental interference or its own limitations. Multi-sensor fusion, however, integrates data from multiple sensors such as vision, hearing, and touch, achieving information complementarity and enhancement. For example, vision sensors can capture image information of the construction scene, identifying whether personnel operations are standardized and whether equipment is operating normally; hearing sensors can detect abnormal sounds, promptly identifying equipment malfunctions or potential safety hazards; touch sensors can perceive the physical characteristics of objects, assisting in judging material quality. Based on this rich multi-dimensional data, the controller can more accurately perform track detection, automatic obstacle avoidance, construction safety monitoring, and construction progress and quality identification, greatly improving the accuracy and comprehensiveness of inspections. 3. The controller integrates with the attendance system and the cloud-based construction factory management platform. After integrating with the attendance system, the controller can acquire personnel attendance information in real time and, combined with personnel operation data collected during inspections, achieve precise personnel management. For example, it can determine whether personnel are working in designated areas and whether their operations comply with safety regulations, issuing timely alerts for violations. Data interaction with the cloud-based construction factory management platform makes the inspection robot a crucial node in the factory's information management. It can upload various data collected during inspections, such as safety status, construction progress, and quality information, to the management platform in real time, providing management with comprehensive and accurate decision-making support. Simultaneously, the management platform can also issue instructions to the controller to adjust inspection strategies and priorities, achieving intelligent and efficient factory management. Attached Figure Description

[0016] This specification will be further described by way of exemplary embodiments, which will be described in detail with reference to the accompanying drawings. These embodiments are not limiting; in these embodiments, the same reference numerals denote the same structures, wherein: Figure 1 This is a block diagram of the inspection robot body shown in one embodiment of this application; Figure 2 This is a flowchart illustrating the determination of the installation positions of multiple vibration monitoring devices in one embodiment of this application; Figure 3 This is a flowchart illustrating orbit detection based on multi-dimensional data in one embodiment of this application; Figure 4 This is a flowchart illustrating the determination of positioning error in one embodiment of this application; Figure 5 This is a structural diagram of the inspection robot body shown in one embodiment of this application. Detailed Implementation

[0017] To more clearly illustrate the technical solutions of the embodiments in this specification, the accompanying drawings used in the description of the embodiments will be briefly introduced below.

[0018] A safety inspection robot for building factories in the cloud is disclosed, relating to the field of intelligent sensing and control equipment technology. It includes a track assembly and an inspection robot body set on the track assembly. The track assembly includes multiple transverse tracks located at different heights, and the inspection robot body moves on the transverse tracks.

[0019] The horizontal rails are made of lightweight aluminum alloy (single section weight ≤20kg, suitable for the load-bearing requirements of cloud construction factories). Each section of the horizontal rail is 2m long and is connected by "bolts + quick clips". This meets the progress requirement of cloud construction factories climbing 1 floor (about 3m) every 3-5 days, and the disassembly and reassembly time is ≤2 hours / floor.

[0020] The inspection robot is equipped with a clamp, eliminating the need to drill holes in the main body of the cloud-based factory, thus avoiding structural damage. The clamp is compatible with H-beams (200×200mm-300×300mm) used in cloud-based factory construction, with a clamping force of ≥500N to prevent the track from falling off.

[0021] Figure 1 This is a flowchart illustrating the determination of the installation positions of multiple vibration monitoring devices in one embodiment of this application, as shown below. Figure 1 As shown, the inspection robot body includes a body, a multi-sensor fusion perception module, and a controller. The multi-sensor fusion perception module is used to collect multi-dimensional data, and the controller is used to perform track detection, automatic obstacle avoidance, construction safety supervision, and construction progress and quality identification based on the multi-dimensional data. At least part of the multi-sensor fusion perception module is set on the body.

[0022] like Figure 1 As shown, in some embodiments, the multi-sensor fusion perception module includes at least a lidar component, an ultrasonic sensor, and an image acquisition component, and the multi-dimensional data includes obstacle distance information acquired by the lidar component and the ultrasonic sensor.

[0023] In some embodiments, the controller performs automatic obstacle avoidance based on multi-dimensional data, including: Based on the obstacle avoidance scenario issued by the cloud-based factory management platform, determine the current obstacle avoidance distance threshold; Based on the obstacle distance information collected by the lidar components and ultrasonic sensors and the current obstacle avoidance distance threshold, it is determined whether to perform obstacle avoidance. If obstacle avoidance is required, images of the obstacles are acquired using the image acquisition component to determine the type of obstacle. Based on the type of obstacle, perform obstacle avoidance actions.

[0024] Specifically, the lidar component generates point cloud data of the surrounding environment by emitting laser beams and measuring reflection time, accurately calculating the distance between obstacles and the inspection robot. By emitting ultrasonic waves and receiving echoes, it detects nearby obstacles, supplementing the lidar's perception blind spots at short distances or on non-reflective surfaces.

[0025] The lidar needs to be calibrated monthly to ensure its ranging accuracy meets the performance requirements of the robot's environmental perception system. The specific operation procedure is as follows: First, a standard calibration board is selected. This board is 1m × 1m in size, with a black and white checkerboard pattern design with 50% reflectivity. It is precisely placed on a horizontal surface 5m directly in front of the lidar, ensuring that the center of the calibration board is aligned with the lidar's optical axis to reduce the impact of angular deviations on the ranging results. Then, the "Sensor Calibration" function module is accessed through the robot's local operating interface. The "LiDAR Calibration" option is selected in the submenu to start the automated calibration program. The system will control the lidar to scan the calibration board from multiple angles, collect reflected point cloud data, and calculate the average ranging value. Simultaneously, the actual distance parameter (5m) on the calibration board is read and compared for analysis. If the absolute value of the deviation between the detected distance and the actual distance is ≤2cm, the calibration is considered qualified, the system records the current parameters, and exits the calibration mode. If the deviation exceeds the allowable range, the system will automatically trigger a parameter correction mechanism, adjusting key parameters such as the lidar's internal delay and beam divergence angle through an iterative algorithm until the ranging error converges to within the acceptable threshold. The entire calibration process is monitored by the robot control system. Calibration results are automatically logged, containing information such as calibration time, ambient temperature, initial error value, and corrected parameters, for subsequent maintenance and analysis. Operators must ensure the calibration board is placed stably to avoid calibration failure due to ambient light interference or uneven ground.

[0026] The cloud-based factory management platform sends obstacle avoidance distance thresholds for the current scene to the inspection robot based on real-time factory conditions (such as temporary building material stacking and construction elevator operation), serving as a benchmark for determining whether obstacle avoidance is necessary. For example: Temporary building material stacking scenario: The cloud-based construction factory management platform issues a static obstacle avoidance threshold of 1.5 meters (i.e., obstacle avoidance is triggered when the distance to the obstacle is ≤1.5 meters).

[0027] Construction elevator operation scenario: The cloud-based construction factory management platform issues a dynamic obstacle avoidance threshold of 3 meters (due to the high speed of the elevator, it is necessary to avoid obstacles in advance).

[0028] Both lidar and ultrasonic sensors simultaneously collect obstacle distance data. The controller combines the results from both and compares them with a threshold issued by the cloud-based factory management platform to determine whether to trigger obstacle avoidance.

[0029] For example: LiDAR detection revealed dense reflective points (suspected to be bundles of steel bars) 1.2 meters to the left of the track.

[0030] The ultrasonic sensor was used to detect and supplement short-range data, confirming that the distance between the steel bar bundles was 1.3 meters (to avoid the error of lidar at close range).

[0031] The current threshold is 1.5m ≤ 1.5m and 1.3m ≤ 1.5m, which triggers obstacle avoidance.

[0032] LiDAR provides long-range, high-precision detection, while ultrasonic sensors fill in short-range blind spots; data fusion improves reliability. If data from either sensor exceeds a threshold, obstacle avoidance is triggered.

[0033] If obstacle avoidance is triggered, the image acquisition component takes a picture of the obstacle and identifies its type (such as a bundle of steel bars, a construction elevator, or a person) through an obstacle recognition model, providing a basis for subsequent obstacle avoidance actions. The obstacle recognition model can be a convolutional neural network model.

[0034] The controller invokes preset obstacle avoidance strategies based on the type of obstacle to ensure a balance between safety and efficiency. For example: 1. Static obstacles (e.g., bundles of steel bars): Action: The inspection robot temporarily pauses, and a voice prompt says, "There are building materials ahead, which need to be cleaned manually."

[0035] Cloud push: Sends obstacle locations and images to operators.

[0036] Recovery condition: After manual cleaning, press the "Continue Inspection" button on the inspection robot.

[0037] 2. Dynamic obstacles (e.g., construction elevators): Predicted Time to Collision (TTC): Elevator distance 3 meters, speed 1 m / s → TTC = 3 seconds.

[0038] Action: If TTC ≤ 2 seconds, the inspection robot will stop abruptly and send a "dodge request" to the elevator control system via the industrial bus. Upon receiving the request, the elevator will automatically decelerate to 0.5 m / s.

[0039] Recovery condition: The inspection robot resumes operation after the elevator moves away (distance > 3 meters).

[0040] like Figure 1 As shown, in some embodiments, the multi-sensor fusion sensing module includes at least an infrared thermal imaging camera, and the multi-dimensional data includes infrared thermal images of the hot work area collected by the infrared thermal imaging camera.

[0041] In some embodiments, the controller performs construction safety monitoring based on multi-dimensional data, including: Flame identification is performed based on infrared thermal images of the hot work area. Violations are identified by capturing images of the work area using an image acquisition component.

[0042] Specifically, the core area of ​​the flame is accurately located through dynamic temperature threshold analysis (e.g., automatically dropping from 300℃ to 200℃ in hot work mode). Combined with flame morphological characteristics (e.g., flashing frequency, area expansion rate, etc.) and temperature gradient distribution, it effectively distinguishes between normal welding sparks and abnormal fire spread. When a sustained high temperature area (>200℃ and area growth rate >5% / second) is detected, a first-level warning is immediately triggered. Simultaneously, a panoramic view of the work area is acquired through a visible light image acquisition component, and deep learning algorithms are used to analyze personnel behavior in real time. On the one hand, unsafe behaviors such as not wearing protective equipment (such as flame-retardant clothing and protective masks) or standing in violation of regulations (downwind of the flame) are identified. On the other hand, environmental hazards such as the accumulation of flammable materials or obstruction of fire-fighting facilities are monitored within the work area. When violations or environmental risks are detected, visible light and infrared images are superimposed to generate composite alarm information. Multimodal data fusion is used to improve the accuracy of risk assessment and realize the four-level linkage of "thermal imaging temperature measurement - flame recognition - behavior analysis - environmental monitoring". The sensitivity of flame recognition can be optimized by adjusting the temperature threshold, and visible light images can be used to supplement behavioral and environmental details, forming a three-dimensional safety management mechanism with all elements and multiple levels for hot work, effectively reducing the risk of fire accidents at the construction site.

[0043] The system acquires real-time visible light images of key areas such as hot work and high-altitude operations using an image acquisition component. An AI recognition engine then performs deep analysis of the images to intelligently identify violations. The specific process is as follows: the image acquisition component captures the work scene at a rate of 15 frames per second. The data is synchronously transmitted to the AI ​​model on the cloud-based factory management platform for real-time processing. Based on a quarterly updated hazard source sample library (e.g., newly added images of loose scaffolding, illegal stacking, etc.) and a false alarm correction mechanism (e.g., labeling squatting work as normal behavior), the AI ​​model uses target detection algorithms to locate personnel and equipment. Combined with a behavior analysis model, it determines whether more than 20 types of pre-defined violations exist, such as not wearing safety helmets, illegally crossing guardrails, and not having fire extinguishers in hot work areas. When a suspected violation is detected, a secondary verification is immediately triggered: the persistence of the behavior is confirmed by comparing multiple frames of images, and infrared thermal imaging data is correlated to eliminate false positives (e.g., distinguishing between normal work sparks and fire hazards). If the violation is verified as valid, the controller automatically generates a structured record, including a violation timestamp, GPS location information, and violation type code. Managers can use a cloud-based factory management platform to filter records by time range, violation type, and other dimensions, and export Excel reports containing detailed processing fields for safety training, accountability, and performance evaluation. The entire identification process is fully automated, from image acquisition to model reasoning, behavior verification, and record generation, ensuring that violations are detected within ≤5 seconds and the accuracy rate is ≥95%, effectively supporting dynamic safety management at construction sites.

[0044] The cloud-based factory safety monitoring system constructs a comprehensive and intelligent safety management system, with a focus on enhancing risk prevention and alarm response efficiency for high-altitude operations. Through AI models, it implements real-time monitoring of high-altitude work areas and features a newly added "high-altitude guardrail detection" function, which can accurately identify potential hazards such as missing or loose guardrails with an accuracy rate exceeding 93%, triggering serious alarms and sending notifications to responsible personnel.

[0045] Alarm permissions are managed hierarchically by role: ordinary operators can handle general alarms such as smoking alarms, and can view records and issue on-site voice warnings; safety administrators are responsible for handling more serious / critical alarms, and have the authority to send SMS reminders and activate emergency plans; project managers approve the activation of emergency plans through the "User Management" module of the cloud-based factory management platform, and can suspend all operations in the factory if necessary. For critical alarms, a standardized handling procedure is initiated: after triggering the strong on-site audible and visual alarm and the pop-up reminder on the cloud-based factory management platform, the safety administrator must confirm the handling within 5 minutes; if personnel are injured, the system automatically dials the preset first aid station number through the "Emergency Linkage" function, and simultaneously pushes the location of the injured person and on-site images to the medical team.

[0046] Furthermore, the cloud-based construction factory management platform supports in-depth analysis of regulatory data, generating risk reports by floor. It provides detailed statistics on the distribution of hazards such as unauthorized hot work and loose scaffolding. These reports can be exported in Excel and PDF formats and seamlessly integrated with project management systems, providing data support for safety management decisions. Through a closed-loop management system of "monitoring-alarm-response-analysis," the platform increases the detection rate of safety hazards by 40% and reduces emergency response time to within 5 minutes, effectively ensuring construction safety in the cloud-based construction factory.

[0047] In some embodiments, the controller identifies construction progress and quality based on multi-dimensional data, including: Obtain multiple work processes issued by the cloud-based factory management platform; Images of each process are acquired using an image acquisition component to identify construction progress and quality. The multi-dimensional data includes images of at least each process.

[0048] Specifically, the controller first obtains a complete list of process steps from the cloud-based factory management platform, covering key steps such as "rebar tying → formwork installation → concrete pouring", and presets the completion standards for each process in the cloud (for example, rebar tying must cover the entire work surface and the spacing between tying points must be ≤20cm), forming a quantifiable basis for judging progress.

[0049] In terms of construction progress recognition, the controller relies on high-definition image acquisition components deployed on the robot to capture images of the work surface of each process at a rate of 15 frames per second. For the rebar tying process, the visual recognition algorithm analyzes the coverage of the rebar and the distribution of tying points. When it is detected that the rebar completely covers the work surface and the spacing between more than 90% of the tying points is ≤20cm, the cloud-based construction factory management platform marks the process as "complete" on the progress Gantt chart and updates it to the project management system simultaneously. If the concrete pouring process is delayed by 24 hours, a three-level response mechanism is immediately triggered: First, an early warning message is pushed to the project manager, and a checklist containing six common reasons such as "delay in rebar acceptance" and "concrete supply interruption" is generated. After the management personnel select the actual reasons, rectification suggestions are generated based on historical data (e.g., coordinating with suppliers to prioritize delivery) and the rectification task is pushed to the responsible team's terminal.

[0050] In the quality identification link, a detection strategy of "three-zone coverage imaging + algorithm comparison" is adopted. After the inspection robot body enters the steel bar binding area, the operator selects the "steel bar detection mode", and the controller automatically controls the pan-tilt to capture images of the left, middle and right three areas, and collects raw data with 4K resolution in each area. The center coordinates of the steel bars are extracted through a dimension measurement algorithm, the spacing between adjacent steel bars is calculated and compared with the design value of 20cm. When the deviation exceeds ±2cm, the system marks the problem position with a red box in the image and generates a defect report including "problem type (e.g., out-of-tolerance spacing)", "deviation value (e.g., 25cm)", "position coordinates (e.g., XYZ three-dimensional coordinates)". After the quality management personnel receive the report, they click "rectification assignment" on the cloud construction factory management platform, select the responsible team and set a 24-hour rectification deadline. After the responsible person uploads the post-rectification image, the inspection robot body re-captures images of the same area for re-inspection. When the re-inspection pass rate is ≥90%, it is marked as "rectified"; otherwise, rectification reminders are continuously pushed.

[0051] When quality defects such as cracks occur in the concrete pouring process and rework is required, the controller automatically executes three-step linked operations: First, interact with the cloud construction factory management platform, and the cloud construction factory management platform postpones the subsequent "formwork removal process" in the progress Gantt chart, and the postponement time is accurately calculated as the expected rework duration (e.g., 48 hours); Second, push adjustment notifications to the construction team through the cloud construction factory management platform, including information such as "original planned time", "adjusted time", and "rework area map"; Third, synchronously update the resource allocation plan in the project management system to ensure that the service cycle of turnover materials such as formwork and scaffolding matches the adjusted progress. This mechanism realizes automatic compensation of progress affected by quality defects, and avoids the problem of construction period out of control caused by the separation of progress and quality in traditional management.

[0052] Through the technical path of "process standard preset - real-time image acquisition - algorithm intelligent identification - data closed-loop linkage", the accuracy of progress identification is increased to 98%, the missed detection rate of quality defects is reduced to below 3%, and the construction period delay caused by rework is reduced by an average of 65%, which provides an efficient and accurate construction management solution for cloud construction factories.

[0053] As Figure 1 shows, in some embodiments, the multi-sensor fusion perception module comprises vibration monitoring devices arranged at a plurality of different positions on the body.

[0054] Figure 2 is a flow chart showing determination of the mounting positions of a plurality of vibration monitoring devices in an embodiment of the present application. In some embodiments, the mounting positions of the plurality of vibration monitoring devices are determined based on the following flow: determining a plurality of transverse rails for testing, wherein the wear amounts of the rail grooves of any two transverse rails are different; Determine multiple initial positions on the machine body, and install vibration monitoring positions at each initial position. For example, multiple initial positions on the machine body can be determined according to a preset interval (e.g., 5cm, etc.). For each test transverse track, vibration monitoring data for each initial position of the test transverse track is collected through vibration monitoring positions installed at each initial position. Based on the vibration monitoring data of each test transverse track corresponding to each initial position, valid test transverse tracks and valid initial positions are selected. Based on the vibration monitoring data corresponding to each valid test transverse track at each valid initial position, the installation locations of multiple vibration monitoring devices are selected.

[0055] Specifically, the following process can be used to filter valid test transverse tracks and valid initial positions based on the vibration monitoring data corresponding to each initial position for each test transverse track: S11. For each initial position, extract the vibration characteristics of the test transverse track corresponding to the initial position from the vibration monitoring data of the test transverse track corresponding to the initial position. For example, time domain characteristics (e.g., peak value, root mean square value, etc.) and frequency domain characteristics (e.g., vibration amplitude at different frequencies, etc.) are used to form the vibration characteristic vector of the test transverse track corresponding to the initial position. Calculate the cosine distance between the vibration characteristic vectors of any two test transverse tracks corresponding to the initial position. Calculate the standard deviation of the cosine distance between the vibration characteristic vectors of any two test transverse tracks corresponding to the initial position. Initial positions with a standard deviation greater than a standard deviation threshold (e.g., 0.5) are taken as valid initial positions. S12. For each test transverse track and each valid initial position, extract the vibration characteristics of the test transverse track corresponding to the valid initial position from the vibration monitoring data of the test transverse track corresponding to the valid initial position, and form the vibration characteristic vector of the test transverse track corresponding to the valid initial position. S13. For each test transverse track, based on the vibration feature vector of the test transverse track corresponding to each valid initial position, form the first vibration feature matrix corresponding to the test transverse track, wherein the row vector of the first vibration feature matrix is ​​the vibration feature vector of the test transverse track corresponding to a valid initial position. S14. For any two transverse test tracks, calculate the cosine distance between the first vibration characteristic matrices corresponding to the two transverse test tracks. If there exists a cosine distance between any two transverse test tracks that is less than the first cosine distance threshold (e.g., 1), then retain one of the transverse test tracks. S15. Determine whether the cosine distance between the vibration characteristic matrices corresponding to any two remaining transverse test tracks is greater than or equal to the first cosine distance threshold. If yes, use the remaining transverse test tracks as valid transverse test tracks. If no, execute S14. S16. For each valid initial position, extract the vibration features of the valid test transverse track corresponding to the valid initial position from the vibration monitoring data of the valid test transverse track corresponding to the valid initial position, form the vibration feature vector of the valid test transverse track corresponding to the valid initial position, and form the second vibration feature matrix corresponding to the valid initial position based on the vibration feature vector of each valid test transverse track corresponding to the valid initial position. The row vector of the second vibration feature matrix is ​​the vibration feature vector of a valid test transverse track corresponding to the valid initial position. S17. For any two valid initial positions, calculate the cosine distance between the second vibration feature matrices corresponding to the two valid initial positions. If there exists a cosine distance between any two valid initial positions that is less than the second cosine distance threshold (e.g., 1), then retain one of the valid initial positions. S18. Determine whether the cosine distance between the second vibration feature matrix corresponding to any two remaining valid initial positions is greater than or equal to the second cosine distance threshold. If yes, use the remaining valid initial positions as the installation positions of the vibration monitoring device. If not, execute S17.

[0056] By selecting transverse test tracks with varying wear levels in the grooves, the impact of different wear conditions on vibration can be comprehensively considered. By determining multiple initial positions on the machine body at preset intervals and installing vibration monitoring devices to collect data, data information from multiple positions can be obtained. By screening effective transverse test tracks and initial positions, invalid data interference can be eliminated. Finally, the installation positions of vibration monitoring devices can be selected based on this, making the installation positions more representative and effective, and accurately capturing key vibration information.

[0057] In some embodiments, the multidimensional data includes vibration monitoring data collected by each vibration monitoring device; Figure 3 This is a flowchart illustrating orbit detection based on multi-dimensional data in one embodiment of this application, such as... Figure 3 As shown, in some embodiments, the controller performs track detection based on multi-dimensional data, including: From the vibration monitoring data of each valid test transverse track corresponding to each valid initial position, extract the vibration monitoring data of each vibration monitoring device corresponding to each valid test transverse track at each valid initial position; Based on the vibration monitoring data of each valid test transverse track corresponding to the installation position of each vibration monitoring device, the vibration characteristics corresponding to different track groove wear amounts are determined. Obtain historical wear data for the transverse track; Based on historical wear data of the transverse track, multiple key wear locations of the transverse track and the wear correlation characteristics between these key wear locations were determined. Track inspection is performed based on vibration monitoring data collected by each vibration monitoring device at each critical wear location and wear correlation characteristics between multiple critical wear locations.

[0058] Specifically, the historical wear data of the transverse track can include multiple historical time points and the wear amount at different locations on the transverse track. For each location, the time point when the wear amount is greater than the wear amount threshold (e.g., 2 mm) is taken as the wear time point. The total number of wear time points and the total number of historical time points are calculated as the wear frequency of the location. Locations with a wear frequency greater than the wear frequency threshold (e.g., 50%) are taken as critical wear locations.

[0059] For any two critical wear locations, the wear amounts at multiple historical time points of the two critical wear locations are substituted as two variables into the correlation coefficient (e.g., Pearson correlation coefficient) calculation formula to obtain the wear correlation coefficient between the two critical wear locations. The wear association characteristics between multiple critical wear locations can include the wear correlation coefficient between any two critical wear locations.

[0060] In some embodiments, the controller performs track detection based on vibration monitoring data collected by each vibration monitoring device corresponding to each critical wear location and wear correlation characteristics between multiple critical wear locations, including: Based on the vibration monitoring data collected by each vibration monitoring device at each critical wear location and the wear correlation characteristics between multiple critical wear locations, the wear risk at each critical wear location is predicted. Based on the wear risk of each critical wear location, risky locations are selected from multiple critical wear locations; Determine the optimal detection path based on the wear risk at multiple risk locations; For each risk location in the optimal detection path, an image corresponding to the risk location is acquired by the image acquisition component, and the wear amount at the risk location is determined based on the image corresponding to the risk location.

[0061] Specifically, the wear risk at each critical wear location can be predicted through the following process: S21. For each critical wear location, calculate the cosine similarity between the vibration monitoring data collected by each vibration monitoring device corresponding to the critical wear location and the vibration characteristics corresponding to different track groove wear amounts, and take the track groove wear amount with the largest cosine similarity as the initial wear amount of the critical wear location. S22. For each critical wear location, the wear correlation coefficient between the critical wear location and other critical wear locations is used as a weight to calculate the risk correction value of the critical wear location by weighted summing of the initial wear amounts of the other critical wear locations. S23. For each critical wear location, sum the initial wear amount and risk correction value of the critical wear location to obtain the wear risk of the critical wear location.

[0062] Key wear locations where the wear risk is greater than the wear risk threshold (e.g., 2) are designated as risk locations.

[0063] The optimal detection path can be determined through the following process: S31. Generate multiple detection paths; S32. Define a fitness function where the higher the wear risk, the earlier the detection sequence of high-risk locations and the shorter the detection path, resulting in a larger fitness function value. This function considers two key factors: first, high-risk locations should be detected earlier because these locations need timely detection to identify potential problems early and take countermeasures to ensure equipment safety; second, the detection path should be short, as shorter paths reduce inspection time and resource consumption, improving inspection efficiency. The fitness function quantifies and integrates these two factors, ensuring that paths meeting the condition of "higher wear risk locations being detected earlier and shorter detection paths" have a larger fitness function value, meaning that the path performs better overall. S33. The Particle Swarm Optimization (PSO) algorithm determines the optimal detection path based on multiple detection paths and a fitness function. These paths are considered different locations within the search space, with each path corresponding to a "particle." The fitness function serves as the standard for evaluating particle performance. The fitness value of each detection path is calculated based on the fitness function, reflecting its comprehensive performance in terms of detection order and path length. During algorithm execution, particles continuously move within the search space, adjusting their flight speed and direction based on their own historical best position and the group's historical best position. Through this iterative update, the particle swarm gradually converges towards regions with higher fitness values. After multiple iterations, the algorithm converges to a globally optimal solution or a near-global optimal solution, thus determining the optimal detection path that satisfies the conditions of prioritizing high-risk locations and minimizing path length. This effectively improves the scientific rigor and rationality of inspection path planning, laying the foundation for efficient inspection work.

[0064] By extracting key information from vibration monitoring data of the effective initial position and test track, the vibration characteristics corresponding to different wear amounts in the track grooves are determined, providing a data foundation for accurately understanding the track condition. Combining historical wear data of the transverse track, key wear locations and wear correlation characteristics are identified, enabling a more comprehensive and in-depth understanding of track wear. In the specific implementation of track inspection, the risk of key wear locations is predicted based on vibration monitoring data and wear correlation characteristics, and risky locations are screened out, making inspection more targeted and avoiding resource waste. The optimal inspection path is determined based on the risk locations, improving inspection efficiency and ensuring that high-risk locations are inspected in a timely manner. For risky locations on the optimal inspection path, images are acquired using an image acquisition component to determine the wear amount, realizing a closed loop from data prediction to actual inspection, accurately obtaining the actual situation of track wear.

[0065] like Figure 1 As shown, in some embodiments, the multi-sensor fusion sensing module includes an RFID positioning component, wherein the RFID positioning component includes multiple RFID readers and RFID positioning tags disposed on the device body, and the multi-dimensional data includes the device body position collected by the RFID positioning component.

[0066] Figure 4 This is a flowchart illustrating the determination of positioning error in one embodiment of this application, as shown below. Figure 4 As shown, in some embodiments, the controller determines the positioning error based on multiple positioning test locations and RFID positioning components, including: Acquire historical positioning error data, which includes positioning error values ​​of RFID positioning components at multiple locations corresponding to multiple historical time points; Based on historical positioning error data, multiple target error locations can be determined. For example, the positioning error frequency of a location can be calculated based on historical positioning error data, and the location with a positioning error frequency greater than the positioning error frequency threshold (e.g., 50%) can be used as the target error location. Based on historical positioning error data, error correlation analysis and average deviation accuracy analysis are performed on the error locations of multiple targets to obtain the results of error correlation analysis and average deviation accuracy analysis. Based on the results of error correlation analysis and average deviation accuracy analysis, multiple positioning test locations were determined. For each positioning test location, the positioning deviation corresponding to the positioning test location is determined based on the body position collected by the RFID positioning component at the positioning test location; The positioning error is determined based on the positioning deviation corresponding to each positioning test position. For example, the average of the positioning deviations corresponding to the positioning test positions is taken as the positioning error. If the positioning error is greater than the positioning error threshold (e.g., 5cm), the positioning data of the RFID positioning component is corrected based on the positioning error.

[0067] Specifically, performing error correlation analysis and average deviation accuracy analysis on multiple target error locations can include the following process: S411. For any two target error locations, the positioning error values ​​of the RFID positioning components at the two target error locations corresponding to multiple historical time points are taken as two variables and substituted into the correlation coefficient calculation formula to obtain the positioning error correlation coefficient of the two target error locations. The error correlation analysis results include the positioning error correlation coefficient of any two target error locations. S42. Using a clustering algorithm, multiple target error locations are grouped according to the correlation coefficient of the positioning errors between any two target error locations to determine multiple location groups. Among them, the larger the correlation coefficient of the positioning errors, the greater the probability that two target error locations are clustered into the same location group. S43. Based on multiple position groups, generate multiple position units, wherein the multiple position units include multiple target error positions from different position groups respectively; S44. For each location unit and each historical time point, the average value of the positioning error value of each target error position included in the location unit is calculated to obtain the error compensation value. The positioning error value of each target error position included in the location unit is corrected by the error compensation value to obtain the corrected positioning error value of the target error position. The corrected positioning error values ​​of each target error position included in the location unit are summed to obtain the total corrected positioning error value. S45. For each location unit, the average of the sum of the corrected positioning error values ​​for each historical time point corresponding to the location unit is calculated to obtain the average of the sum of positioning error values ​​corresponding to the location unit. The average deviation accuracy analysis result includes the average of the sum of positioning error values ​​corresponding to each location unit.

[0068] The target error locations included in the location unit with the smallest sum of positioning error values ​​are used as multiple positioning test locations.

[0069] Utilizing historical positioning error data allows for the full extraction of value from past positioning information. Determining target error locations by calculating positioning error frequency enables precise focusing on areas prone to errors, making subsequent analysis more targeted. Error correlation analysis and average deviation accuracy analysis of target error locations allow for in-depth exploration of the inherent patterns and characteristics of error generation, providing a scientific basis for rationally determining positioning test locations and ensuring that the selected test locations effectively reflect the overall positioning error situation. Determining positioning deviation based on the RFID positioning component's acquisition of the device's position at the positioning test location, and further calculating the positioning error, is a direct and effective method. Using the average positioning deviation as the positioning error provides a simple and intuitive quantification of the overall positioning accuracy. Correcting the positioning data when the positioning error exceeds a set threshold allows for timely calibration of the RFID positioning component's positioning results, ensuring the accuracy and reliability of the positioning data.

[0070] The controller is also used to interface with the attendance system for personnel management.

[0071] Specifically, it includes: Retrieve attendance information from the attendance system; Images of people are captured using an image acquisition component; Personnel management is carried out based on personnel images and attendance information.

[0072] Specifically, the controller first establishes a secure connection with the real-name attendance system to obtain basic personnel information (e.g., name, job type, attendance status, etc.) and attendance records in real time, forming a dynamic personnel database. Simultaneously, the image acquisition component of the inspection robot captures personnel images at a rate of 30 frames per second, extracts facial features using a facial recognition algorithm, and compares them with the registration information in the attendance database for verification. Firstly, during the daily attendance verification phase, the system automatically compares the personnel's "attendance check-in time" with their "entry time into the work area". When it detects that the actual entry time is later than the planned working hours or that the number of people does not match the schedule, it is immediately marked as "pending verification" and pushed to the administrator's terminal. Secondly, during the unauthorized personnel control phase, if an unregistered visitor is detected, the inspection robot will immediately trigger an audible and visual alarm and simultaneously open a "temporary authorization" channel for the operator. The operator can input visitor information (name, reason for visit) and authorization time limit (maximum 2 hours) through the cloud interface. During the authorization period, the robot will pause its alarm and record the visitor's movement trajectory. Third, in the data closed-loop stage, all verification records (e.g., normal attendance, abnormal events, temporary authorizations, etc.) generate structured logs, which support the generation of visual reports by job type, time, floor and other dimensions, and can be exported to the project management system for efficiency analysis.

[0073] The system automates the entire process of personnel management, from identification and verification to handling and analysis, reducing the response time to unauthorized entry incidents to within 3 seconds and lowering the rate of missed attendance reports to less than 2%, thus effectively improving the efficiency of personnel management at construction sites.

[0074] The controller is also used to interact with cloud-based factory management platforms.

[0075] Specifically, the controller achieves deep data integration with the cloud-based construction plant management platform through standardized data interfaces and multi-dimensional interaction mechanisms. At the data integration level, it supports bidirectional interaction with the BIM model: managers can select "BIM Integration" in the cloud's "System Integration" module, enter the BIM system's IP address and port, and upload a floor model in .ifc format. Defects such as concrete cracks and steel corrosion discovered during inspections are automatically and accurately mapped to the corresponding locations in the BIM model using 3D markers. Clicking on the markers allows users to view detailed information such as defect type, discovery time, and processing status, forming an intuitive comparison between "digital twin" and "on-site reality."

[0076] In terms of remote control, the controller has established a tiered emergency response system: when an emergency such as a tower crane malfunction is detected, the manager can trigger a "full-area emergency stop" command through the cloud-based "emergency control" module. The robot immediately stops its inspection and triggers on-site audible and visual alarms. Simultaneously, the system requires the administrator's password for secondary confirmation, effectively preventing accidental operation. This function achieves a delay of ≤3 seconds from command issuance to equipment response, ensuring rapid handling in emergency scenarios.

[0077] In terms of historical data management, the system provides multi-condition combined query capabilities: users can filter inspection records in the "Historical Data" module by floor (e.g., 11th floor) and time range (e.g., the last 7 days). The data covers 12 categories of information, including safety alarms, quality defects, and personnel attendance. All data is retained for at least one year. Query results can be exported in Excel and PDF formats. The exported content includes structured information such as defect location coordinates, responsible persons for handling, and before-and-after comparison images, which can be directly used for archiving project completion documents, achieving full lifecycle management of construction data. Through these functions, the controller processes an average of 23,000 data interactions per day, with a data synchronization accuracy rate maintained above 99.7%.

[0078] In summary, the inspection robot itself is an intelligent device integrating multiple functional modules, designed to achieve efficient and accurate factory environment inspection, such as... Figure 5As shown, its core is an ARM control motherboard, responsible for coordinating the operation and data interaction of various modules. The robot is equipped with encoders, lifting and walking drivers, and corresponding motors, enabling precise control of its movement trajectory and position, ensuring the accuracy and flexibility of the inspection route. In terms of perception, it is equipped with a sound pickup module, distance sensor, photoelectric switch, and vision camera, which can collect multi-dimensional information such as audio, distance, obstacles, and images in real time, comprehensively perceiving the surrounding environment. Simultaneously, through an Ethernet module, the robot can conduct stable data transmission with the industrial control computer in the central control room. The industrial control computer is connected to the management platform via a wireless router, enabling remote monitoring and command issuance. For power management, the battery manager effectively manages battery charging and discharging, ensuring a stable power supply, while the power relay ensures circuit safety. Furthermore, the robot has rich interfaces, such as AI, DO, IP, PO, and I / O, facilitating connection to various external devices and expanding its functionality. In case of emergencies, managers can send commands through the management platform, which are transmitted wirelessly to cause the robot to perform operations such as a full-area emergency stop.

[0079] Finally, it should be understood that the embodiments described in this specification are merely illustrative of the principles of the embodiments described herein. Other variations may also fall within the scope of this specification. Therefore, alternative configurations of the embodiments described herein are intended to be illustrative rather than limiting, and should be considered consistent with the teachings of this specification. Accordingly, the embodiments described herein are not limited to those explicitly introduced and described herein.

Claims

1. A safety inspection robot for building factories in the cloud, characterized in that, The system includes a track assembly and an inspection robot body mounted on the track assembly. The track assembly includes multiple transverse tracks located at different heights, and the inspection robot body moves on the transverse tracks. The inspection robot body includes a body, a multi-sensor fusion perception module and a controller. The multi-sensor fusion perception module is used to collect multi-dimensional data, and the controller is used to perform track detection, automatic obstacle avoidance, construction safety supervision and construction progress and quality identification based on the multi-dimensional data. At least part of the multi-sensor fusion perception module is set on the body. The controller is also used to interface with the attendance system for personnel management; The controller is also used to interact with a cloud-based factory management platform; The multi-sensor fusion sensing module includes vibration monitoring devices installed at multiple different locations on the machine body; The installation locations of multiple vibration monitoring devices are determined based on the following process: Multiple transverse tracks were identified for testing, wherein the wear amount of the track grooves of any two transverse tracks was different; Determine multiple initial positions on the fuselage, and install vibration monitoring positions at each initial position; For each test transverse track, vibration monitoring data for each initial position of the test transverse track is collected through vibration monitoring positions installed at each initial position. Based on the vibration monitoring data of each test transverse track corresponding to each initial position, valid test transverse tracks and valid initial positions are selected. Based on the vibration monitoring data corresponding to each valid test transverse track at each valid initial position, the installation locations of multiple vibration monitoring devices are selected.

2. The safety inspection robot for building factories in the cloud, as described in claim 1, is characterized in that... The multi-sensor fusion perception module includes at least a lidar component, an ultrasonic sensor, and an image acquisition component, and the multi-dimensional data includes obstacle distance information acquired by the lidar component and the ultrasonic sensor. The controller performs automatic obstacle avoidance based on the multi-dimensional data, including: Based on the obstacle avoidance scenario issued by the cloud-based factory management platform, determine the current obstacle avoidance distance threshold; Based on the obstacle distance information collected by the lidar components and ultrasonic sensors and the current obstacle avoidance distance threshold, it is determined whether to perform obstacle avoidance. If obstacle avoidance is required, images of the obstacles are acquired using the image acquisition component to determine the type of obstacle. Based on the type of obstacle, perform obstacle avoidance actions.

3. A safety inspection robot for building factories in the cloud, as described in claim 2, is characterized in that... The multi-sensor fusion sensing module includes at least an infrared thermal imaging camera, and the multi-dimensional data includes infrared thermal images of the hot work area collected by the infrared thermal imaging camera. The controller performs construction safety supervision based on the multi-dimensional data, including: Flame identification is performed based on infrared thermal images of the hot work area. Violations are identified by capturing images of the work area using an image acquisition component.

4. A safety inspection robot for building factories in the cloud, as described in claim 1, is characterized in that... The controller identifies construction progress and quality based on the multi-dimensional data, including: Obtain multiple work processes issued by the cloud-based factory management platform; Images of each process are acquired using an image acquisition component to identify construction progress and quality. The multi-dimensional data includes at least images of each process.

5. A safety inspection robot for building factories in the cloud, as described in any one of claims 1-4, characterized in that, The controller interfaces with the attendance system for personnel management, including: Retrieve attendance information from the attendance system; Images of people are captured using an image acquisition component; Personnel management is carried out based on personnel images and attendance information.

6. A safety inspection robot for building factories in the cloud, as described in claim 1, is characterized in that... The multi-dimensional data includes vibration monitoring data collected by each vibration monitoring device; The controller performs track detection based on the multi-dimensional data, including: From the vibration monitoring data of each valid test transverse track corresponding to each valid initial position, extract the vibration monitoring data of each vibration monitoring device corresponding to each valid test transverse track at each valid initial position; Based on the vibration monitoring data of each valid test transverse track corresponding to the installation position of each vibration monitoring device, the vibration characteristics corresponding to different track groove wear amounts are determined. Obtain historical wear data for the transverse track; Based on historical wear data of the transverse track, multiple key wear locations of the transverse track and the wear correlation characteristics between these key wear locations were determined. Track inspection is performed based on vibration monitoring data collected by each vibration monitoring device at each critical wear location and wear correlation characteristics between multiple critical wear locations.

7. A safety inspection robot for building factories in the cloud, as described in claim 6, is characterized in that... The controller performs track detection based on vibration monitoring data collected by each vibration monitoring device corresponding to each critical wear location and wear correlation characteristics between multiple critical wear locations, including: Based on the vibration monitoring data collected by each vibration monitoring device at each critical wear location and the wear correlation characteristics between multiple critical wear locations, the wear risk at each critical wear location is predicted. Based on the wear risk of each critical wear location, risky locations are selected from multiple critical wear locations; Determine the optimal detection path based on the wear risk at multiple risk locations; For each risk location in the optimal detection path, an image corresponding to the risk location is acquired by the image acquisition component, and the wear amount at the risk location is determined based on the image corresponding to the risk location.

8. A safety inspection robot for building factories in the cloud, as described in any one of claims 2-4, characterized in that, The multi-sensor fusion sensing module includes an RFID positioning component, wherein the RFID positioning component includes multiple RFID readers and RFID positioning tags installed on the device body, and the multi-dimensional data includes the device body position collected by the RFID positioning component. The controller performs track detection based on the multi-dimensional data, including: The positioning error was determined based on multiple positioning test locations and RFID positioning components.

9. A safety inspection robot for building factories in the cloud, as described in claim 8, is characterized in that... The controller determines the positioning error based on multiple positioning test locations and RFID positioning components, including: Obtain historical positioning error data; Based on historical positioning error data, the error locations of multiple targets were determined; Based on historical positioning error data, error correlation analysis and average deviation accuracy analysis are performed on the error locations of multiple targets to obtain the results of error correlation analysis and average deviation accuracy analysis. Based on the results of error correlation analysis and average deviation accuracy analysis, multiple positioning test locations were determined. For each positioning test location, the positioning deviation corresponding to the positioning test location is determined based on the body position collected by the RFID positioning component at the positioning test location; The positioning error is determined based on the positioning deviation corresponding to each positioning test position.

Images