Multi-sensor integrated intelligent detection robot control system

Abstract

The invention discloses a control system of a multi-sensor integrated intelligent detection robot, and provides a multi-sensor intelligent detection scheme integrating acoustics, infrared, vision, noise and laser radar aiming at the problems of low efficiency, high risk and more detection omission of traditional manual inspection. The control system processes multi-source information through a data fusion algorithm to realize accurate identification of equipment defects; an autonomous navigation and dynamic path planning algorithm is adopted, so that the robot can autonomously avoid obstacles and efficiently move in a complex environment; and a dynamic obstacle avoidance mechanism based on a virtual repulsive force field is introduced to realize real-time safe obstacle avoidance. The system supports remote monitoring and manual intervention, has modular expansion capability, is widely applicable to automatic detection and health management of industrial equipment such as pressure pipelines and pressure containers, and remarkably improves the safety, accuracy and intelligent level of detection.

Description

Technical Field

[0001] This invention belongs to the field of industrial robot control technology, and in particular relates to a control system for an intelligent inspection robot with multi-sensor integration. Background Technology

[0002] Currently, some intelligent inspection systems are gradually replacing manual inspections, but these systems still face some challenges. For example, existing inspection systems often rely on single sensor technology and lack multi-sensor fusion, which limits the accuracy and reliability of the inspection results. In addition, existing systems are mostly used for static equipment monitoring and lack adaptability to dynamic environments. Especially in complex industrial environments, they lack sufficient flexibility and mobility, making it difficult to meet the needs of different inspection scenarios.

[0003] Traditional inspection robots often rely on a single sensor, lacking effective data fusion and coordinated control. This leads to isolated information between different sensors, resulting in biased inspection results and making it difficult to form a comprehensive and accurate fault identification capability.

[0004] Existing control systems generally only support data processing and command issuance from a single sensor, lacking unified scheduling and fusion algorithms for multi-source data. This makes it impossible to achieve real-time perception and intelligent decision-making of environmental information in complex scenarios, resulting in difficulty in improving both robot detection efficiency and accuracy.

[0005] In complex industrial environments, there are dense pipelines, narrow spaces, many obstacles, and strong noise interference. Traditional control systems lack global coordination of sensor status and robot movement, which can easily lead to problems such as sensor signal distortion, unstable path planning, and untimely obstacle avoidance, affecting the continuity and safety of inspection operations.

[0006] Existing inspection robots mostly adopt a centralized control architecture, which lacks system flexibility and is difficult to adapt to the rapid adjustment of the number, type and functional modules of sensors under different working scenarios, thus limiting the scalability and adaptability of inspection robots.

[0007] Traditional control systems have limited capabilities in anomaly detection and safety protection. They lack real-time monitoring and emergency response mechanisms for situations such as sensor failures, abnormal robot postures, or power system failures, making it difficult to ensure the long-term stable operation of robots in complex scenarios.

[0008] The data processing and analysis capabilities of existing systems rely on manual or backend processing, resulting in poor real-time performance. They are unable to achieve edge computing and rapid fusion of data collected from multiple sensors, leading to delays in the feedback of detection results and reducing the practicality and value of on-site industrial testing.

[0009] Therefore, it is necessary to design a control system for an intelligent detection robot that integrates multiple sensors to solve the above problems. Summary of the Invention

[0010] The technical problem to be solved by this invention is to provide a control system for an intelligent inspection robot with multi-sensor integration. This system aims to provide an efficient, accurate, and safe inspection solution by integrating multiple high-precision sensors, such as acoustic imaging, infrared detection, visual imaging, and noise analysis, combined with robot autonomous navigation technology. The robot control system can automatically identify potential defects in equipment, such as weld leaks, loose bolts, and pipe deformation. Through highly sensitive sensors and data fusion algorithms, it generates inspection reports in real time, helping maintenance personnel to understand the equipment's operating status and ensuring its long-term stable operation.

[0011] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A control system for a multi-sensor integrated intelligent detection robot includes a sensor data acquisition module for real-time acquisition of environmental and equipment data from visual sensors, infrared sensors, acoustic imaging sensors, noise sensors, and lidar. The data processing and fusion module is used to fuse multi-source sensor data, extract features, and generate device status information. The path planning and navigation control module is used for path planning and dynamic obstacle avoidance based on environmental maps and sensor data. The task scheduling and execution module is used to schedule the robot to execute tasks according to the priority of the detection tasks; The safety protection and emergency response module is used to monitor the robot's status and trigger an emergency response in case of abnormalities. Remote monitoring and control interface, used to support remote manual control and real-time monitoring.

[0012] Preferably, the sensor data acquisition module further includes an inertial measurement unit (IMU) for acquiring the robot's posture and motion status in real time.

[0013] Preferably, the data processing and fusion module uses Kalman filtering or deep learning algorithms to fuse multi-source data, thereby enabling the identification and classification of equipment defects.

[0014] Preferably, the path planning and navigation control module uses SLAM technology and the A* algorithm for autonomous navigation, and supports the Dynamic Window Method (DWA) for path fine-tuning.

[0015] Furthermore, the system also includes a robotic arm control module, with a detection box integrated at the end of the robotic arm. The detection box integrates an acoustic imager, a camera, and a noise sensor, with a total weight not exceeding 4 kg.

[0016] Preferably, the system further includes a data storage and transmission module, which supports local database storage and encrypted transmission to the cloud platform, and supports industrial control computers to read and visualize the data through a network interface.

[0017] Preferably, a control method for a control system of a multi-sensor integrated intelligent detection robot includes the following steps: S1 collects environmental and equipment data in real time through multiple types of sensors; S2, fuses and processes the collected multi-source data to extract equipment status features; S3 performs path planning and navigation control based on fused data and environmental maps; S4, schedule the robot to perform the detection task according to the task priority; S5 monitors the robot's status in real time and triggers a safety response in case of anomalies; The S6 supports remote manual control and real-time data monitoring.

[0018] Preferably, the data fusion algorithm used in step S2 includes Kalman filtering or a convolutional neural network to identify at least one defect among air leaks, cracks, and loose bolts; the method of data fusion using a convolutional neural network is as follows: S201, Construct a multimodal feature fusion neural network, which includes: A sensor-specific feature extraction subnetwork is used to extract features from visual data, infrared thermal imaging data, acoustic and noise data, and lidar point cloud data, respectively. The cross-modal feature fusion layer is used to align and weight the features extracted by each sub-network; the global feature understanding and decision layer is used to classify equipment status and defects based on the fused features. S202, the data collected by various sensors are input into the corresponding feature extraction sub-networks to obtain the feature vectors of each sensor. Where k = 1, 2, ..., K, and K is the total number of sensor categories; S203, through a learnable weight matrix With bias vector Each feature vector Mapping to a uniform dimension D yields aligned feature vectors. : ; S204, Calculate the attention weights for each sensor feature. Then, weighted fusion is performed to obtain the global fusion feature. : ; in, and These are learnable fusion weight parameters; S205 inputs the globally fused features into the fully connected layer and the Softmax classifier, outputting the probability distribution y of device status and defect category: ; in, and These are the classifier parameters; S206, based on the output result y, identify whether the equipment is in normal condition or has at least one defect among air leakage, cracks, loose bolts, and corrosion.

[0019] Preferably, step S3 specifically includes the following steps: S301, Acquire input information, including: real-time environmental perception data obtained from multi-sensor data fusion, and robot fused pose estimation. and environmental maps containing semantic information. M ; S302, based on the environmental map M Given the sequence of task objective points G, an improved path search algorithm is used to calculate the global path. The improvement is reflected in the path cost function. Introducing semantic cost terms The semantic cost is dynamically adjusted based on the defect identification results of devices near the target point in the real-time environmental perception data; S303, along the global path During movement, based on the real-time environmental perception data, a dynamic window method is used in the robot's velocity space. Online trajectory planning is performed; the optimal speed pair is selected by optimizing the evaluation function. : ; Among them, semantic security evaluation item Calculations are made based on the category and state attributes of obstacles in the perception data; S304, the optimal speed pair The data is sent to the robot motion controller to execute navigation and the navigation status is continuously monitored. If the data deviates from the preset tolerance or encounters an impassable obstacle, the system returns to step S302 to perform global path replanning.

[0020] Preferably, step S3 further includes a dynamic obstacle avoidance step, which adjusts the path planning in real time when an obstacle is detected; specifically including: Real-time acquisition of environmental perception data obtained from multi-sensor data fusion, and extraction of state information of one or more obstacles from the data, wherein the state information includes at least: Obstacle categories Relative position Relative velocity and attribute tags ; For each obstacle i Dynamically calculate its corresponding safety radius : ; in, It is a basic safety distance based on obstacle category. and For speed compensation coefficient, It is a context-compensated distance based on attribute tags; When any obstacle is at a distance from the robot d Less than its dynamic safety radius When the obstacle avoidance response is triggered, it is based on the relative position of the obstacle. and distance d Calculate the virtual repulsive force exerted by the obstacle on the robot. : ; in, The maximum influence distance of the repulsive force. The repulsion gain is adjusted according to the obstacle threat level; Calculate the total repulsive force generated by all relevant obstacles. And convert it into the robot's current planned speed. Correction amount : ; in, To control the gain for obstacle avoidance; The corrected speed command The data is sent to the robot's motion controller to enable real-time obstacle avoidance and continuously monitor the obstacle status until the threat is eliminated.

[0021] Preferably, step S5 further includes a tilt detection and anti-fall mechanism, which automatically triggers emergency braking when the robot's posture is abnormal.

[0022] The beneficial effects of this invention are as follows: 1. This system is equipped with high-precision inspection equipment such as acoustic imagers, infrared cameras, high-definition cameras, and noise sensors, enabling comprehensive and accurate inspection of pressure pipelines and pressure vessels, identifying faults such as leaks, loose bolts, bracket deformation, corrosion, and cracks. Through highly integrated sensors and data fusion algorithms, the system achieves real-time monitoring and fault diagnosis of industrial equipment, greatly improving inspection efficiency and accuracy. The robot adopts a modular design, featuring a lifting arm and a robotic arm, allowing it to flexibly adapt to different operating scenarios. It also possesses autonomous navigation and path planning capabilities, enabling it to perform inspection tasks autonomously or remotely in complex environments. The control system, with its dual manual and automatic operating modes, provides high flexibility and safety, making it widely applicable to the periodic inspection and maintenance of industrial equipment.

[0023] 2. This invention utilizes a multi-sensor integrated intelligent inspection robot control system. The robot can autonomously or remotely control itself to complete comprehensive inspection tasks of pressure pipelines and pressure vessels, avoiding the safety risks associated with manual high-altitude operations and complex environments. The system combines multiple sensors, including high-definition vision, infrared imaging, acoustic imaging, and noise sensing, to achieve precise monitoring of equipment status and fault early warning, significantly improving inspection accuracy, especially in the early detection of minute defects such as leaks, cracks, and loose bolts, filling blind spots that traditional manual inspection cannot cover. The control system employs advanced autonomous navigation technology, utilizing lidar, IMU, and SLAM algorithms to achieve high-precision positioning and environmental perception. The robot can move flexibly in complex power plant environments, avoiding obstacles and accurately completing inspection tasks, especially in narrow, complex, and high-altitude environments, ensuring comprehensive and efficient inspection. The robot control system integrates intelligent task scheduling and data processing functions, automatically identifying defects and prioritizing tasks based on real-time multi-sensor data, ensuring timely completion of all inspection tasks, and using intelligent algorithms to analyze equipment health status and provide early warnings of potential faults. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the overall control system in an embodiment of the present invention; Figure 2 This is a schematic diagram of the robot display interface in an embodiment of the present invention; Figure 3 This is a schematic diagram of the user interface in an embodiment of the present invention; Figure 4 This is a schematic diagram of the remote operation of the bilateral robotic arm in an embodiment of the present invention; Figure 5 This is a schematic diagram of a remote control base station in an embodiment of the present invention. Detailed Implementation

[0025] Example 1: like Figures 1-5 As shown, a control system for a multi-sensor integrated intelligent detection robot includes a sensor data acquisition module for real-time acquisition of environmental and equipment data from a visual sensor, an infrared sensor, an acoustic imaging sensor, a noise sensor, and a lidar. The data processing and fusion module is used to fuse multi-source sensor data, extract features, and generate device status information. The path planning and navigation control module is used for path planning and dynamic obstacle avoidance based on environmental maps and sensor data. The task scheduling and execution module is used to schedule the robot to execute tasks according to the priority of the detection tasks; The safety protection and emergency response module is used to monitor the robot's status and trigger an emergency response in case of abnormalities. Remote monitoring and control interface, used to support remote manual control and real-time monitoring.

[0026] Preferably, the sensor data acquisition module further includes an inertial measurement unit (IMU) for acquiring the robot's posture and motion status in real time.

[0027] Preferably, the data processing and fusion module uses Kalman filtering or deep learning algorithms to fuse multi-source data, thereby enabling the identification and classification of equipment defects.

[0028] Preferably, the path planning and navigation control module uses SLAM technology and the A* algorithm for autonomous navigation, and supports the Dynamic Window Method (DWA) for path fine-tuning.

[0029] Furthermore, the system also includes a robotic arm control module, with a detection box integrated at the end of the robotic arm. The detection box integrates an acoustic imager, a camera, and a noise sensor, with a total weight not exceeding 4 kg.

[0030] Preferably, the system further includes a data storage and transmission module, which supports local database storage and encrypted transmission to the cloud platform, and supports industrial control computers to read and visualize the data through a network interface.

[0031] Preferably, a control method for a control system of a multi-sensor integrated intelligent detection robot includes the following steps: S1 collects environmental and equipment data in real time through multiple types of sensors; S2, fuses and processes the collected multi-source data to extract equipment status features; S3 performs path planning and navigation control based on fused data and environmental maps; S4, schedule the robot to perform the detection task according to the task priority; S5 monitors the robot's status in real time and triggers a safety response in case of anomalies; The S6 supports remote manual control and real-time data monitoring.

[0032] Preferably, the data fusion algorithm used in step S2 includes Kalman filtering or a convolutional neural network to identify at least one defect among air leaks, cracks, and loose bolts; the method of data fusion using a convolutional neural network is as follows: S201, Construct a multimodal feature fusion neural network, which includes: A sensor-specific feature extraction subnetwork is used to extract features from visual data, infrared thermal imaging data, acoustic and noise data, and lidar point cloud data, respectively. The cross-modal feature fusion layer is used to align and weight the features extracted by each sub-network; the global feature understanding and decision layer is used to classify equipment status and defects based on the fused features. S202, the data collected by various sensors are input into the corresponding feature extraction sub-networks to obtain the feature vectors of each sensor. Where k = 1, 2, ..., K, and K is the total number of sensor categories; S203, through a learnable weight matrix With bias vector Each feature vector Mapping to a uniform dimension D yields aligned feature vectors. : ; S204, Calculate the attention weights for each sensor feature. Then, weighted fusion is performed to obtain the global fusion feature. : ; in, and These are learnable fusion weight parameters; S205 inputs the globally fused features into the fully connected layer and the Softmax classifier, outputting the probability distribution y of device status and defect category: ; in, and These are the classifier parameters; S206, based on the output result y, identify whether the equipment is in normal condition or has at least one defect among air leakage, cracks, loose bolts, and corrosion.

[0033] Preferably, step S3 specifically includes the following steps: S301, Acquire input information, including: real-time environmental perception data obtained from multi-sensor data fusion, and robot fused pose estimation. and environmental maps containing semantic information. M ; S302, based on the environmental map M Given the sequence of task objective points G, an improved path search algorithm is used to calculate the global path. The improvement is reflected in the path cost function. Introducing semantic cost terms The semantic cost is dynamically adjusted based on the defect identification results of devices near the target point in the real-time environmental perception data; S303, along the global path During movement, based on the real-time environmental perception data, a dynamic window method is used in the robot's velocity space. Online trajectory planning is performed; the optimal speed pair is selected by optimizing the evaluation function. : ; Among them, semantic security evaluation item Calculations are made based on the category and state attributes of obstacles in the perception data; S304, the optimal speed pair The data is sent to the robot motion controller to execute navigation and the navigation status is continuously monitored. If the data deviates from the preset tolerance or encounters an impassable obstacle, the system returns to step S302 to perform global path replanning.

[0034] Preferably, step S3 further includes a dynamic obstacle avoidance step, which adjusts the path planning in real time when an obstacle is detected; specifically including: Real-time acquisition of environmental perception data obtained from multi-sensor data fusion, and extraction of state information of one or more obstacles from the data, wherein the state information includes at least: Obstacle categories Relative position Relative velocity and attribute tags ; For each obstacle i Dynamically calculate its corresponding safety radius : ; in, It is a basic safety distance based on obstacle category. and For speed compensation coefficient, It is a context-compensated distance based on attribute tags; When any obstacle is at a distance from the robot d Less than its dynamic safety radius When the obstacle avoidance response is triggered, it is based on the relative position of the obstacle. and distance d Calculate the virtual repulsive force exerted by the obstacle on the robot. : ; in, The maximum influence distance of the repulsive force. The repulsion gain is adjusted according to the obstacle threat level; Calculate the total repulsive force generated by all relevant obstacles. And convert it into the robot's current planned speed. Correction amount : ; in, To control the gain for obstacle avoidance; The corrected speed command The data is sent to the robot's motion controller to enable real-time obstacle avoidance and continuously monitor the obstacle status until the threat is eliminated.

[0035] Preferably, step S5 further includes a tilt detection and anti-fall mechanism, which automatically triggers emergency braking when the robot's posture is abnormal.

[0036] Example 2: This embodiment provides an application of the system in a large hydroelectric power plant in central China, as detailed below: 1. Control system architecture: 1.1 Sensor Data Acquisition Module: Responsible for real-time data acquisition from various sensors (such as vision, infrared, acoustic imaging, noise, etc.).

[0037] 1.2 Data Processing and Fusion Module: This module fuses, analyzes, and processes the collected data from various sensors to generate accurate equipment status information.

[0038] 1.3 Path planning and navigation control module: Based on sensor data and environmental maps, it plans paths to guide the robot to move safely and efficiently in complex environments.

[0039] 1.4 Task Scheduling and Execution Module: Schedules the robot to execute various detection tasks, ensuring that tasks are completed in order of priority.

[0040] 1.5 Safety Protection and Emergency Response Module: Monitors the robot's working status, prevents dangers, and automatically shuts down or alarms in emergency situations.

[0041] 2. Multi-sensor data acquisition and fusion: The control system integrates various sensors, including high-definition visual cameras, infrared sensors, acoustic sensors, and noise sensors. The data acquisition module collects information from these sensors in real time, processes it through data fusion algorithms (such as Kalman filtering and deep learning techniques), eliminates noise and extracts effective features, and finally sends the fused data to the data processing module for further analysis and decision-making.

[0042] Visual data acquisition: Environmental images are captured using high-definition cameras, and defect identification is performed using image processing technology.

[0043] Infrared and acoustic data acquisition: Infrared sensors are used to capture temperature changes, while acoustic sensors are used to detect possible minute leaks.

[0044] 3. Path planning and navigation control: Based on sensor data and environmental maps, the control system uses autonomous navigation algorithms (such as SLAM and A* algorithms) for path planning, ensuring that the robot can move freely within the work area, avoid obstacles, and perform complex tasks, such as working at heights.

[0045] 3.1 Dynamic obstacle avoidance and path fine-tuning: When the robot encounters dynamic obstacles while moving, the system can replan the path in real time and fine-tune the path through the dynamic window method (DWA) to ensure stable robot movement.

[0046] 3.2 Autonomous Navigation and Positioning: Utilizing lidar and IMU sensors to provide real-time location information, the control system achieves precise positioning and navigation in complex environments.

[0047] 4. Task scheduling and execution: The robot control system employs a task scheduling module, which uses intelligent algorithms to prioritize detection tasks and assign them to the robot for execution. Task scheduling takes into account environmental information, detection requirements, and equipment status to ensure high efficiency.

[0048] 4.1 Task allocation and priority scheduling: Based on different detection targets and areas, the task scheduling module automatically allocates corresponding tasks to the robot.

[0049] 4.2 Task Execution and Feedback: The robot provides feedback on task progress through the execution module, and the control system monitors the execution status in real time to ensure that the task is completed on time.

[0050] 5. Real-time monitoring and remote control: The system supports remote monitoring and manual control. Operators can view the robot's working status, environmental information, and data results in real time through the control platform. Simultaneously, operators can intervene manually to ensure the robot can adjust its operation as needed in complex or hazardous environments.

[0051] 5.1 Monitoring Platform: Provides image and data stream display, allows real-time viewing of the robot's current working status, and offers functions such as fault warning and progress monitoring.

[0052] 5.2 Remote control: Operators can precisely control the robot via remote control or computer terminal, especially when performing specific tasks in complex environments, where manual control provides additional flexibility.

[0053] 6. Safety Protection and Emergency Response: The control system incorporates a safety protection and emergency response module to respond promptly and prevent accidents when the robot encounters abnormal situations. This module integrates various safety sensors (such as tilt sensors and fall protection sensors) to monitor the robot's status in real time.

[0054] 6.1 Tilt Detection and Fall Prevention: The robot monitors its posture using tilt sensors to ensure it does not tip over when working at heights. If an unsafe tilt angle is detected, the system immediately triggers emergency braking.

[0055] 6.2 Emergency Braking and Stopping: In the event of sensor failure, path blockage, or emergency, the control system will automatically trigger the emergency stop function to prevent further damage or danger to the robot.

[0056] 7. Data storage and secure transmission: The robot control system supports data storage and secure transmission. All sensor data and detection results are stored in real time, ensuring data integrity. The control system uses encrypted transmission technology to ensure the security of data transmission, preventing data leakage or loss.

[0057] 7.1 Real-time storage and backup: All collected data will be uploaded to the cloud platform in real time for storage and backup, so as to facilitate subsequent analysis and backtracking.

[0058] 7.2 Data Encryption and Transmission: Secure data encryption protocols are used to ensure data security during transmission.

[0059] 8. System scalability and upgrades: This control system is highly scalable, supporting the addition of new sensors or detection modules in the future. Through its modular design, new functional modules or sensor devices can be easily added according to actual needs, expanding the robot's detection capabilities and control precision.

Claims

1. A control system for a multi-sensor integrated intelligent detection robot, characterized in that, It includes a sensor data acquisition module for real-time acquisition of environmental and equipment data from visual sensors, infrared sensors, acoustic imaging sensors, noise sensors, and lidar; The data processing and fusion module is used to fuse multi-source sensor data, extract features, and generate device status information. The path planning and navigation control module is used for path planning and dynamic obstacle avoidance based on environmental maps and sensor data. The task scheduling and execution module is used to schedule the robot to execute tasks according to the priority of the detection tasks; The safety protection and emergency response module is used to monitor the robot's status and trigger an emergency response in case of abnormalities. Remote monitoring and control interface, used to support remote manual control and real-time monitoring.

2. The control system for a multi-sensor integrated intelligent detection robot according to claim 1, characterized in that, The sensor data acquisition module also includes an inertial measurement unit (IMU) for real-time acquisition of the robot's posture and motion status.

3. The control system for a multi-sensor integrated intelligent detection robot according to claim 1, characterized in that, The data processing and fusion module uses Kalman filtering or deep learning algorithms to fuse multi-source data, thereby enabling the identification and classification of equipment defects.

4. The control system for a multi-sensor integrated intelligent detection robot according to claim 3, characterized in that, The path planning and navigation control module uses SLAM technology and the A* algorithm for autonomous navigation, and supports dynamic window method (DWA) for path fine-tuning.

5. The control system for a multi-sensor integrated intelligent detection robot according to claim 1, characterized in that, The system also includes a data storage and transmission module, which supports local database storage and encrypted transmission to the cloud platform, and supports industrial control computers to read and visualize the data through a network interface.

6. The control method for a control system of a multi-sensor integrated intelligent detection robot according to any one of claims 1-5, characterized in that, Includes the following steps: S1 collects environmental and equipment data in real time through multiple types of sensors; S2, fuses and processes the collected multi-source data to extract equipment status features; S3 performs path planning and navigation control based on fused data and environmental maps; S4, schedule the robot to perform the detection task according to the task priority; S5 monitors the robot's status in real time and triggers a safety response in case of anomalies; The S6 supports remote manual control and real-time data monitoring.

7. The control method for a control system of a multi-sensor integrated intelligent detection robot according to claim 6, characterized in that, The data fusion algorithm used in step S2 includes Kalman filtering or convolutional neural networks to identify at least one defect among air leaks, cracks, and loose bolts. The method for data fusion using convolutional neural networks is as follows: S201, Construct a multimodal feature fusion neural network, which includes: A sensor-specific feature extraction subnetwork is used to extract features from visual data, infrared thermal imaging data, acoustic and noise data, and lidar point cloud data, respectively. The cross-modal feature fusion layer is used to align and weight the features extracted by each sub-network; the global feature understanding and decision layer is used to classify equipment status and defects based on the fused features. S202, the data collected by various sensors are input into the corresponding feature extraction sub-networks to obtain the feature vectors of each sensor. Where k = 1, 2, ..., K, and K is the total number of sensor categories; S203, through a learnable weight matrix With bias vector Each feature vector Mapping to a uniform dimension D yields aligned feature vectors. : ； S204, Calculate the attention weights for each sensor feature. Then, weighted fusion is performed to obtain the global fusion feature. : ； in, and These are learnable fusion weight parameters; S205 inputs the globally fused features into the fully connected layer and the Softmax classifier, outputting the probability distribution y of device status and defect category: ； in, and These are the classifier parameters; S206, based on the output result y, identify whether the equipment is in normal condition or has at least one defect among air leakage, cracks, loose bolts, and corrosion.

8. The control method for a control system of a multi-sensor integrated intelligent detection robot according to claim 6, characterized in that, Step S3 specifically includes the following steps: S301, Acquire input information, including: real-time environmental perception data obtained from multi-sensor data fusion, and robot fused pose estimation. and environmental maps containing semantic information. M ; S302, based on the environmental map M Given the sequence of task objective points G, an improved path search algorithm is used to calculate the global path. The improvement is reflected in the path cost function. Introducing semantic cost terms The semantic cost is dynamically adjusted based on the defect identification results of devices near the target point in the real-time environmental perception data; S303, along the global path During movement, based on the real-time environmental perception data, a dynamic window method is used in the robot's velocity space. Online trajectory planning is performed; the optimal speed pair is selected by optimizing the evaluation function. : ； Among them, semantic security evaluation item Calculations are made based on the category and state attributes of obstacles in the perception data; S304, the optimal speed pair The data is sent to the robot motion controller to execute navigation and the navigation status is continuously monitored. If the data deviates from the preset tolerance or encounters an impassable obstacle, the system returns to step S302 to perform global path replanning.

9. The control method for a control system of a multi-sensor integrated intelligent detection robot according to claim 8, characterized in that, Step S3 further includes a dynamic obstacle avoidance step, which adjusts the path planning in real time when an obstacle is detected; specifically including: Real-time acquisition of environmental perception data obtained from multi-sensor data fusion, and extraction of state information of one or more obstacles from the data, wherein the state information includes at least: Obstacle categories Relative position Relative velocity and attribute tags ; For each obstacle i Dynamically calculate its corresponding safety radius : ； in, It is a basic safety distance based on obstacle category. and For speed compensation coefficient, It is a context-compensated distance based on attribute tags; When any obstacle is at a distance from the robot d Less than its dynamic safety radius When the obstacle avoidance response is triggered, it is based on the relative position of the obstacle. and distance d Calculate the virtual repulsive force exerted by the obstacle on the robot. : ； in, The maximum influence distance of the repulsive force. The repulsion gain is adjusted according to the obstacle threat level; Calculate the total repulsive force generated by all relevant obstacles. And convert it into the robot's current planned speed. Correction amount : ； in, To control the gain for obstacle avoidance; The corrected speed command The data is sent to the robot's motion controller to enable real-time obstacle avoidance and continuously monitor the obstacle status until the threat is eliminated.

10. The control method for a control system of a multi-sensor integrated intelligent detection robot according to claim 7, characterized in that, Step S5 also includes a tilt detection and anti-fall mechanism, which automatically triggers emergency braking when the robot's posture is abnormal.

Images