Industrial intelligent detection robot

Abstract

The invention relates to the technical field of intelligent industrial robots, in particular to an industrial intelligent detection robot which comprises a robot body, a walking module is installed at the bottom of the robot body, a rotating platform is arranged in the center of the robot body, a sensing detection module and a detection scanning module are integrated on the rotating platform, and a main control module is arranged in the robot body. The main control module is in signal connection with the walking module, the sensing detection module and the detection scanning module and used for receiving data collected by all the modules and issuing a control instruction, and the functions of welding seam positioning, path planning, walking control and defect detection are achieved. Through the integrated design of multi-source sensing fusion positioning, dual-mode cooperative detection, bionic self-adaptive walking, multi-machine intelligent scheduling and dynamic parameter adaptation, precise positioning and detection of the industrial detection robot under the complex tank wall working condition are achieved, the whole machine operation stability and the multi-machine cooperative efficiency are comprehensively improved, and the working efficiency of the industrial detection robot is improved. And the harsh requirements of high-end industrial detection are met.

Description

Technical Field

[0001] This invention relates to the field of intelligent industrial robot technology, and in particular to an intelligent industrial inspection robot. Background Technology

[0002] Industrial inspection robots are automated equipment integrating mechanical transmission, sensing and detection, intelligent control, and communication technologies. Their core function is to replace manual labor in various inspection tasks in industrial settings. By being equipped with different types of inspection modules and walking mechanisms, they can perform defect detection, condition monitoring, and data collection on industrial equipment and structures. Compared to traditional manual inspection, industrial inspection robots offer advantages such as high operating efficiency, stable inspection accuracy, and low safety risks. They can adapt to harsh or dangerous operating environments such as high temperature, high pressure, high humidity, and confined spaces, making them key equipment for industrial intelligent upgrading and safe production assurance.

[0003] Currently, industrial inspection robots are widely used in many fields such as petrochemicals, power, metallurgy, and construction. Chinese invention patent application number CN202211392623.9 discloses a wall-climbing robot for weld inspection, which includes a top plate, a flaw detection mechanism, a magnetic wheel mechanism, and a sensor assembly. It uses four sets of magnetic wheel mechanisms to achieve wall walking and adsorption, uses an industrial camera to detect welds and complete tracking, and is equipped with a magnetic particle flaw detector to perform weld inspection. It is also equipped with a gravity feedback mechanism and a liquid supply mechanism to ensure the reliability of the inspection.

[0004] However, the magnetic particle flaw detectors carried by the aforementioned existing industrial inspection robots are only suitable for detecting surface and near-surface defects in welds of Grade III and IV, and cannot meet the quality assessment requirements of higher-level welds of Grade I and II. Furthermore, their technical solutions use single vision positioning and single-type defect detection, lacking a dynamic adaptation mechanism for scanning step distance and walking speed, resulting in low positioning and detection accuracy. At the same time, the existing solutions generally use a single magnetic wheel adsorption and drive structure, which is incompatible with both metal and non-metal tank walls, resulting in poor fit to curved and steeply sloped tank walls. During walking, there is a lack of multi-module collaborative anti-instability design, making it prone to tilting or getting stuck on rough or obstacle-prone tank wall surfaces. Moreover, the existing technology only supports independent operation of a single device, lacking an effective identity authentication and path scheduling mechanism. When multiple devices operate simultaneously, problems such as collisions, overlapping work areas, or omissions are prone to occur, seriously affecting the overall inspection efficiency.

[0005] The shortcomings of the existing technologies mentioned above mean that industrial inspection robots cannot meet the stringent requirements of high-end industrial inspection in terms of inspection accuracy, operational stability, safety and reliability, and overall efficiency under complex working conditions. In particular, in the inspection of large steel structures such as ships and pressure vessels, there are core pain points such as the contradiction between high-altitude movement efficiency and safety, insufficient dynamic inspection accuracy, low efficiency of multi-machine collaboration, and poor functional adaptability. Therefore, developing an industrial inspection robot technology solution that can solve the above problems has become a technical bottleneck that urgently needs to be overcome by those skilled in the art. Summary of the Invention

[0006] The main technical problem to be solved by this invention is to provide an industrial intelligent inspection robot. Through the integrated design of multi-source sensor fusion positioning, dual-mode collaborative detection, biomimetic adaptive walking, multi-machine intelligent scheduling and dynamic parameter adaptation, the industrial inspection robot can achieve accurate positioning and detection in complex tank wall conditions, and the overall operation stability and multi-machine collaborative efficiency are comprehensively improved.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: An industrial intelligent inspection robot includes a body, a walking module installed at the bottom of the body, a rotating platform at the center of the body, a sensing and detection module and a detection and scanning module integrated on the rotating platform, and a main control module inside the body. The main control module is connected to the walking module, the sensing and detection module and the detection and scanning module respectively, and is used to receive data collected by each module and issue control commands to realize weld positioning, path planning, walking control and defect detection functions.

[0008] The following are further optimizations of the above technical solution by the present invention: The main control module adopts a dual-main-control architecture, including a data processing unit and a motion control unit. The data processing unit is used for multi-source data fusion, algorithm calculation and global decision scheduling, while the motion control unit is used to parse control commands and generate motion drive signals.

[0009] Further optimization: The walking module includes multiple walking legs, which are designed based on the biomimetic structure of a human leg, including an upper support leg, a lower support leg, and a support foot. The hip, knee, and ankle joints are articulated to achieve adaptive posture adjustment. The support foot integrates an auxiliary walking device, a main adsorption device, and an emergency adsorption device at its bottom. The main adsorption device and the emergency adsorption device have identical structures and employ an electromagnetic-vacuum dual-mode switching system. The three walking legs use a staggered control strategy; when one walking leg loses power, the other two remain in an adsorption state. Torque sensors are installed at the joints of the walking legs, and electromagnetic brake locking devices are installed in the support feet. These, along with the emergency adsorption device and the dynamic fall protection redundancy control unit of the main control module, achieve fall protection.

[0010] Further optimization: The rotating platform includes a platform base connected to the fuselage. The platform base has a built-in torque sensor to collect the eccentric torque and rotational resistance torque of the platform in real time. A differential rotation mechanism is set below the platform base. The output axis of the differential rotation mechanism extends upward and is connected to the sensing and detection module and the scanning module. An angle sensor is set at the connection point. The main control module has a built-in high-precision dynamic balance control algorithm.

[0011] Further optimization: The sensing and detection module is equipped with a ranging device, a visual positioning device, a weld feature point recognition device, an attitude perception compensation device, and a tank wall three-dimensional coordinate calibration device; the multi-anchor point linkage positioning scheme realizes fuzzy positioning and precise positioning of the weld, and the deep learning model is combined to predict the weld inflection point to ensure the continuity of detection.

[0012] Further optimization: The detection scanning module includes a detection arm and an array-type defect detection device. The detection arm can realize Z-axis lifting and circumferential rotation to form a spiral three-dimensional scanning path. The defect detection device adopts a dual-mode synchronous detection method to simultaneously collect internal and external defects and cross-sectional parameters of the weld, realizing overlapping and cross-verification of the detection area.

[0013] Further optimization: The main control module incorporates a high-precision closed-loop control algorithm and an adaptive gait adjustment algorithm. The high-precision closed-loop control algorithm integrates torque, displacement, and speed sensor data to achieve precise adjustment of the walking module, while the adaptive gait adjustment algorithm automatically switches the walking gait based on the tank wall working condition data.

[0014] Further optimizations: The machine body is equipped with a wireless communication module and a QR code identification mark, which supports self-organizing network communication and identity authentication of multiple devices. The main control module has a built-in multi-machine collaborative walking scheduling algorithm to realize the sharing of location information of multiple devices, dynamic path adjustment and return timing coordination, so as to avoid collisions and omissions or overlaps in the work area.

[0015] Further optimization: The main control module can dynamically adjust the scanning step distance based on the weld width data collected by the sensor detection module, and optimize the scanning frequency based on the walking speed data of the walking module, balancing detection accuracy and work efficiency.

[0016] Further optimizations: The robot supports 5G / Wi-Fi dual-mode remote communication and encrypted control, and has functions such as remote emergency stop, inspection data visualization, and defect marking.

[0017] The present invention, by adopting the above technical solution, has the following beneficial effects: This invention employs a multi-anchor point linkage scheme in the positioning stage, combining infrared ranging, AprilTag visual positioning, and ultrasonic phased array feature point recognition. This, coupled with a deep learning model to predict weld inflection points, enables full-process positioning of the weld, from fuzzy to precise, ensuring continuous detection of complex trajectories. Simultaneously, a real-time acoustic path compensation algorithm with a residual path length is used to dynamically correct the ultrasonic detection signal, eliminating acoustic path errors caused by movement and achieving millimeter-level positioning accuracy during dynamic movement. In the detection stage, a dual-mode synchronous detection method combining phased array ultrasound and high-frequency eddy current is used. Data collected simultaneously from laser scanning sensors is used to construct a three-dimensional visualization model, enabling comprehensive acquisition of internal and external defects and cross-sectional parameters. Overlapping and cross-verifying detection areas ensure no blind spots. For efficiency, the main control module dynamically adjusts the scanning step distance based on the weld width and optimizes the scanning frequency based on the walking speed, balancing detection accuracy and operational efficiency.

[0018] The walking leg of this invention is based on the biomimetic structure of a human leg. The hinged design of the hip, knee and ankle joints, combined with the universal ball joint ankle joint, can adaptively adjust the fitting angle according to the curved surface of the tank wall. The electromagnetic-vacuum dual-mode switching system of the main adsorption device is compatible with both metal and non-metal tank walls. The three walking legs adopt a staggered control strategy. When a single module is powered off and walking, the other two remain adsorbed. Combined with a high-precision closed-loop control algorithm and an adaptive gait adjustment algorithm, walking stability is guaranteed in all aspects.

[0019] The main control module of this invention has a built-in dynamic fall protection redundant control unit. When the body tilts beyond the threshold, the emergency adsorption device will be activated and the electromagnetic brake locking device will be triggered simultaneously, forming a dual protection of "adsorption + braking". The torque sensor at the joint monitors the force in real time and automatically reduces torque and alarms when abnormalities occur. It supports remote emergency stop and multi-terminal encrypted control, which comprehensively improves the safety of the equipment.

[0020] This invention enables real-time data interaction among multiple devices through QR code authentication on the side of the device and self-organizing network communication. The multi-machine collaborative walking scheduling algorithm built into the main control module allows the devices to share location information when bypassing obstacles. Other devices dynamically adjust their speed and path to avoid collisions. When a single device returns to its original path, the main control module coordinates the timing and speed of the return to prevent omissions or overlaps in the work area, effectively improving the overall work efficiency.

[0021] The present invention will be further described below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0022] Figure 1 This is a three-dimensional view of the overall structure of the intelligent detection robot of this invention; Figure 2 This is a control principle diagram of the intelligent detection robot of the present invention; Figure 3 This is a flowchart of the time-sharing control of the intelligent detection robot's adsorption and walking in this invention; Figure 4 This is a flowchart of the intelligent inspection robot's weld seam fuzzy positioning dual-anchor point linkage positioning process of the present invention; Figure 5 This is a flowchart of the high-precision closed-loop control algorithm for the intelligent detection robot of this invention; Figure 6 This is a flowchart of the adaptive gait adjustment algorithm for the intelligent detection robot of the present invention; Figure 7 This is a flowchart of the pre-planning and real-time correction of the intelligent detection robot's walking path according to the present invention.

[0023] In the diagram: 1. Body; 2. Walking module; 201. Upper support leg; 202. Lower support leg; 203. Support foot; 204. Walking drive motor; 3. Sensor detection module; 4. Detection and scanning module; 5. Rotating platform; 6. Wireless communication module. Detailed Implementation

[0024] like Figure 1-2 The industrial intelligent inspection robot shown includes a body 1, a walking module 2 installed at the bottom of the body 1, and a rotating platform 5 located at the center of the body 1. The rotating platform 5 integrates a sensing module 3 and a detection scanning module 4. The sensing module 3 is used to collect real-time images and working parameters of the object to be inspected, providing data support for weld identification, path planning, and real-time correction. The detection scanning module 4 is dedicated to the precise scanning and inspection of welds, realizing the identification, location, and quantitative analysis of weld defects. The rotating platform 5 can rotate 360°, enabling the sensing module 3 and the detection scanning module 4 to achieve circumferential coverage inspection without moving the entire machine. This inspection robot is suitable for weld inspection scenarios of large welded components such as large pressure tanks and ship gangways. The following detailed description uses the weld inspection of large pressure tanks as a typical application case.

[0025] The main control module is located inside the fuselage 1. The main control module adopts a dual main control architecture, including a data processing unit and a motion control unit. The data processing unit is used to receive monitoring data collected by various sensors, instruction information sent by remote terminals, and input signals from the human-machine interface. It undertakes high-performance algorithm calculation and global decision scheduling functions. The motion control unit is used to parse the control instructions output by the data processing unit, generate precise motion drive signals and send them to the execution components, and at the same time, provide feedback on the execution status data.

[0026] This design separates computationally intensive algorithm tasks and real-time control tasks into different main control units, avoiding response delays caused by resource contention. It effectively ensures the efficiency of data processing and the real-time accuracy of motion control during weld inspection, adapting to the high real-time and high reliability requirements of weld inspection in industrial scenarios.

[0027] In this embodiment, the data processing unit uses Jetson Xavier NX to perform multi-source data fusion processing, target recognition and behavior decision-making algorithm calculations, and generate motion control commands; the motion control unit uses an STM32H743 microcontroller to receive motion control commands sent by Jetson Xavier NX, parse and generate motion control logic adapted to each actuator, and drive the robot's various mechanisms to perform precise movements.

[0028] To further decouple the algorithmic and control tasks within the data processing unit and motion control unit, and to achieve layered collaboration between high-performance algorithm computation, global decision scheduling, and high-real-time motion execution, thus adapting to complex detection scenarios with high precision and multiple operating conditions, the data processing unit can also adopt a dual-chip collaborative architecture of Raspberry Pi 4B and Jetson Xavier NX. The two communicate via a PCIe interface. The Raspberry Pi 4B handles non-real-time, highly complex global task planning and logical decisions, is responsible for robot task flow scheduling and multi-module data integration, and also functions as a human-machine interface and remote communication command transmission and reception function; the Jetson Xavier... NX, as the dedicated AI computing core, focuses on image recognition, target detection, and machine learning algorithm calculations, outputting accurate environmental perception and target localization results. The motion control unit is simultaneously upgraded to a layered architecture of Siemens S7-1200 series programmable controller (PLC) and STM32H743 microcontroller. The PLC, as a motion coordination component, receives global motion commands from the Raspberry Pi 4B, combines the robot's real-time pose data for path planning and motion logic optimization, and generates a standardized set of motion control commands. The microcontroller, as a motion execution component, outputs high-precision pulse and switching signals based on the motion control logic generated by the PLC, controlling the robot's drive motors, actuators, and other components to complete precise movements.

[0029] The data processing unit incorporates a Kalman filter fusion module and a residual path compensation module to achieve IMU-CCD-ultrasound multi-source data fusion and dynamic compensation of ultrasound signals. The residual path compensation module detects residual path deviations during transmission or conversion, generates a reverse compensation amount, and actively cancels the deviations to restore the integrity of the signal or physical quantity. Its core algorithm flow is as follows: detect the residual path error at the output of the acquisition system and transmit the signal to the control unit; the control unit calculates the compensation parameters in conjunction with a preset compensation model; the execution unit controls the system to make corrections based on the compensation parameters; continuously monitor the effect after compensation, dynamically adjust the compensation parameters, and form a closed-loop control.

[0030] The rotating platform 5 includes a platform base connected to the body 1. The platform base has a built-in torque sensor to collect the eccentric torque and rotational resistance torque of the platform in real time. A differential rotation mechanism is set under the platform base to eliminate jamming and offset problems during the platform rotation process. A platform surface is set on the top of the platform base. Two symmetrically arranged counterweights are set on the platform surface. The counterweights are slidably mounted on guide rails. The counterweight drive motor drives the counterweights to move along the corresponding guide rails to adjust the center of gravity.

[0031] The output axis of the differential rotation mechanism extends upward and connects to the sensing module 3 and the detection scanning module 4. An angle sensor is installed at the connection point. The detection scanning module 4 includes a multi-joint folding detection arm. The angle sensor is used to collect the rotation angle and extension length of the detection arm. The main control module has a built-in high-precision dynamic balance control algorithm to solve the problems of center of gravity shift and rotational instability of the rotating platform 5 during the full-attitude movement of the detection scanning module 4. The specific algorithm flow is as follows: M1, Data Acquisition: Angle sensor acquires the rotation angle and extension length of the detection arm, torque sensor acquires the eccentric torque and rotational resistance torque of the rotating platform 5, and transmits the data to the main control module; M2, Center of gravity offset calculation: Based on the above posture parameters of the detection arm and combined with the overall center of gravity model, the main control module calculates the center of gravity offset and the required compensation balance torque. M3, Counterweight Adjustment: The high-precision dynamic balance control algorithm built into the main control module outputs the counterweight movement distance based on the center of gravity offset. The counterweight drive motor drives the counterweight to move along the guide rail, changing the position of the platform's center of gravity. M4, Real-time Torque Calibration: The main control module compares the actual eccentric torque collected by the torque sensor with the preset threshold. If there is a deviation, the position of the counterweight is adjusted, and the adjustment is repeated until the eccentric torque is less than the set threshold. In step M4, the main control module is equipped with a torque dead zone to prevent frequent operation of mechanical parts from causing instability of the center of gravity and affecting the overall operation of the robot. M5, Rotation Coordination Control: The main control module combines the preset resistance torque-speed mapping table, compares the actual rotational resistance torque with the reference value, and dynamically adjusts the platform speed according to the degree of deviation. When there is severe jamming, the anti-spin difference mechanism is activated to assist, and the rotation angle is corrected in a closed loop through the angle sensor.

[0032] like Figure 1As shown, the walking module 2 includes three walking legs evenly spaced apart. Two of the walking legs are installed at the bottom of the body 1 as the main walking legs, and the other is installed on the side of the body 1 as the auxiliary walking leg. The walking legs are biomimetic structures based on human legs, including an upper support leg 201, a lower support leg 202, and a support foot 203. The top of the upper support leg 201 is hinged to the body 1 to form a hip joint, the bottom of the upper support leg 201 is hinged to the top of the lower support leg 202 to form a knee joint, and the bottom of the lower support leg 202 is hinged to the support foot 203 to form an ankle joint. Torque sensors are integrated and installed on the inner side of the bearings at each joint.

[0033] A walking drive motor 204 is installed at the hip joint to drive the upper support leg 201 to swing around the hip joint, thereby driving the lower support leg 202 and support foot 203 to complete basic walking movements such as lifting the leg and stepping. At the same time, it can achieve precise adjustment of the leg posture. The output shaft of the walking drive motor 204 is connected to the upper support leg 201 through a flexible coupling to ensure smooth power transmission and to offset vibration and impact.

[0034] The walking drive motor 204 is connected to the main control module via a motor drive controller. The motor drive controller receives the control commands from the main control module and converts them into electrical signals for the walking drive motor 204 to operate. At the same time, it receives feedback signals from the walking drive motor 204 to achieve low-level closed-loop regulation of the motor speed.

[0035] In this embodiment, the walking drive motor 204 is a servo motor, and the motor drive controller is a servo driver. The two, together with the main control module, form a closed-loop system. The motor drive controller can directly adjust the output of the servo motor by modifying the given pulse frequency or analog voltage.

[0036] In addition to this embodiment, the walking drive motor 204 can also be a stepper motor or a brushless DC motor, which, together with the motor drive controller, can be used to adjust the walking speed.

[0037] A damping torsion spring joint is installed at the knee joint. The one-way damper is used to limit the speed of joint bending and extension, while the torsion spring is used to provide preload for joint reset. The ankle joint adopts a universal ball joint structure to realize multi-directional rotation of the support foot 203. The ball joint has an internal elastic damping pad to buffer the impact of the ground and limit excessive swing. This design allows the support foot 203 to adaptively adjust the fitting angle according to the curvature of the can wall. Furthermore, the adsorption surface at the bottom of the support foot 203 can also adopt an arc shape that adapts to the shape of the can wall so as to better fit the surface of the can wall and ensure adsorption stability.

[0038] The bottom of the support foot 203 integrates an auxiliary walking device, a main adsorption device, and an emergency adsorption device. The auxiliary walking device is equipped with a speed sensor. The auxiliary walking device can be a magnetic track or a Mecanum wheel, which transforms the sliding friction between the bottom surface of the support foot 203 and the tank wall into rolling friction, reducing movement resistance while improving stability and smoothness during walking. The magnetic track adheres to the surface of the tank wall through magnetic attraction and is suitable for walking on curved surfaces and tank walls with large slopes. The Mecanum wheel relies on its omnidirectional movement characteristics to achieve flexible movement without a turning radius. Both structures are commonly used auxiliary walking devices in existing industrial robots and will not be described in detail in this invention.

[0039] The main adsorption device employs an electromagnetic-vacuum dual-mode switching adsorption system, including an electromagnet, a vacuum pump, and an electromagnetic reversing valve. The electromagnetic reversing valve is a two-position three-way valve. The electromagnet and vacuum pump are controlled by this valve to open and close the circuit and gas path. When the tank wall material is metal, the main control module energizes the coil corresponding to the magnetic adsorption station, causing the electromagnetic reversing valve core to switch and move to the magnetic adsorption station, connecting the power supply circuit to the electromagnet and simultaneously cutting off the gas path to the vacuum pump. At this time, the electromagnet is energized to perform the adsorption action against the tank wall, while the vacuum pump is in standby mode. Machine status: When the tank wall material is non-metallic, the main control module disconnects the power supply to the coil corresponding to the magnetic adsorption station and simultaneously energizes the coil corresponding to the vacuum adsorption station. The solenoid reversing valve core reverses to the vacuum adsorption station, cutting off the power supply circuit of the electromagnet and simultaneously connecting the air passage of the vacuum pump. At this time, the vacuum pump works to perform the adsorption action with the tank wall, and the electromagnet is de-energized and in standby mode. When the detection ends and the equipment detaches from the tank wall, the main control module sends a power-off signal to the coil, the solenoid reversing valve core resets to the initial position, and both the electromagnet and the vacuum pump stop working, and the adsorption effect is released.

[0040] The main adsorption unit has a built-in pressure sensor that can provide real-time feedback on the adsorption status.

[0041] like Figure 3 As shown, for a single walking leg, the auxiliary walking device and the main adsorption device are controlled in a time-sharing manner, operating according to the following timing sequence: L1. Adsorption: When the main adsorption device is powered on, the electromagnetic reversing valve is switched according to the different materials of the tank wall, so that the electromagnet or vacuum pump works to complete the adsorption. At the same time, the pressure sensor built into the main adsorption device confirms the adsorption stability. L2, Power off: The main adsorption device is powered off, adsorption is canceled, and the walking drive motor 204 is ready to start. L3, Walking: The walking drive motor 204 starts, driving the upper support leg 201 to swing around the hinge point, thereby driving the lower support leg 202 and support foot 203 to complete the walking action; L4. Adsorption: The main adsorption unit is powered on again to complete the adsorption process, and the pressure sensor built into the main adsorption unit confirms the adsorption stability.

[0042] By repeating the above steps L1-L4 in a loop, the movement of a single walking module 2 can be achieved.

[0043] The three walking legs are controlled in a staggered manner, alternating between power-off walking and adsorption actions. For example, when one main walking leg is powered off and walking, the other main walking leg and the auxiliary walking leg remain in an adsorption and locking state to ensure the robot's adsorption stability. Within a certain time period, the three walking legs perform staggered actions according to Table 1 to complete the overall movement of the robot.

[0044] Table 1. Decomposition of walking leg staggered control movements To achieve accurate weld seam detection, the robot's working path needs to be precisely matched with the distribution direction of the weld seam. To this end, the sensing and detection module 3 is equipped with a variety of dedicated detection devices according to the detection requirements. In this embodiment, the sensing and detection module 3 is integrated on the rotating platform 5 and connected to the main control module for signal transmission. It is used to collect real-time images and various parameters of the wall tank and transmit them to the data processing unit of the main control module. This provides basic data support for the realization of functions such as weld seam positioning and recognition, path pre-planning, real-time path correction, and multi-machine collaborative scheduling during the robot's operation.

[0045] The weld seam location and identification includes core components such as fuzzy positioning, precise positioning and tracking, data optimization, defect marking and remote calibration. Each component works in concert to achieve full-process data processing of the weld seam from initial identification to precise positioning and defect marking.

[0046] The robot is equipped with a wireless communication module 6, which is an external high-gain wireless communication antenna. It adopts directional transmission technology that resists metal interference, and uses 5G / Wi-Fi dual-mode communication. It is equipped with encryption algorithms to ensure data transmission security. It supports remote control command transmission, remote emergency stop, and remote operation on multiple devices (computers, tablets, mobile phones). It can also display the device status and detection data in real time on a visual remote control interface, and realize functions such as three-dimensional reconstruction of scanning trajectory and defect location marking. At the same time, it supports remote import of standard data and real-time positioning data for comparison and calibration, and automatically records calibration logs. Moreover, this communication module is integrated into the three-dimensional model, which comprehensively ensures the controllability of the detection process and the consistency of positioning accuracy.

[0047] To further ensure the spatial accuracy and data consistency of weld seam positioning and identification, the sensing and detection module 3 is equipped with a tank wall three-dimensional coordinate calibration device and an attitude perception compensation device. The tank wall three-dimensional coordinate calibration device provides a unified and standard tank wall three-dimensional coordinate reference for the entire weld seam positioning process. All positioning data and ranging data are mapped to this calibration coordinate system to ensure the spatial consistency of the data.

[0048] The attitude perception and compensation device uses an IMU (Inertial Measurement Unit) as its core, and is equipped with auxiliary sensors such as tilt sensors and attitude gyroscopes. It collects data such as the robot's tilt angle, offset, and motion posture in real time to provide real-time attitude compensation for the detection data of each device, ensuring the spatial matching and detection accuracy of various data during the weld positioning process.

[0049] The main control module has a built-in dynamic fall protection redundancy control unit. In conjunction with the attitude perception and compensation device, it collects real-time tilt angle and attitude change data of the robot body 1 to realize dynamic attitude monitoring of body 1 and fall protection emergency linkage function: when the attitude sensor of body 1 detects that the tilt angle of body 1 exceeds the built-in tilt angle threshold, the emergency adsorption device is immediately activated. The structure and working principle of the emergency adsorption device are the same as those of the main adsorption device. It can select electromagnetic adsorption or vacuum adsorption mode according to the tank wall material, effectively improving the robot's operation stability and fall protection redundancy under complex tank wall conditions, and avoiding the risk of equipment falling due to the tilt instability of body 1.

[0050] Furthermore, the bottom surface of the support foot 203 is also equipped with an electromagnetic brake locking device. This device consists of an electromagnetic coil and a brake actuator. When energized, the electromagnetic coil generates a magnetic field, driving the brake actuator (usually a brake pawl or friction pad) to extend outward, forming a rigid connection with the tank wall surface or locking by friction. When de-energized, the magnetic field of the electromagnetic coil disappears, driving the actuator to return to its original position, without affecting the robot's normal walking and operation. The electromagnetic brake locking device is linked with the fuselage 1 attitude detection device. When the fuselage 1 tilt angle exceeds the angle threshold, the emergency adsorption device and the electromagnetic brake locking device are activated simultaneously, forming a dual protection of adsorption and braking, further reducing the risk of falling.

[0051] To achieve the fuzzy positioning function of weld seams, the sensing and detection module 3 is equipped with a ranging device, an AprilTag visual positioning device, and a weld seam feature point recognition device.

[0052] The ranging device is an infrared ranging sensor used in the process of fuzzy positioning of weld seams to detect in real time the relative distance between the robot and the tank wall, the flatness of the wall, and the contour height difference of the weld seam area. By sensing the difference in physical shape between the weld seam and the surrounding base material, the approximate direction and distribution of the weld seam can be preliminarily determined, and suspected weld seam areas can be identified in advance and a pre-positioning area for the weld seam can be generated. This provides initial spatial coordinates for subsequent precise positioning of the weld seam, greatly shortening the calculation and detection time for subsequent fine positioning and improving the overall positioning efficiency.

[0053] In this embodiment, six infrared ranging sensors are arranged in an arc array in the detection area at the front end of the robot. The included angle between two adjacent infrared ranging sensors is 20°, forming a wide-angle detection range of 120°. This layout can achieve no blind spots in the detection area and effectively capture the spatial distance information around the weld in the direction of robot movement, providing comprehensive and continuous ranging data support for fuzzy positioning of the weld.

[0054] For the multi-channel distance data acquired by the infrared ranging sensor array, the system employs a data stitching algorithm for integrated processing: The raw ranging data undergoes noise reduction, calibration, and synchronization preprocessing to eliminate the effects of individual sensor errors, environmental interference, and data acquisition timing differences. Combining the reference parameters of the tank wall 3D coordinate calibration device and the real-time attitude data of the attitude perception compensation device, and based on the layout parameters of the infrared ranging sensor arc array and the spatial coordinate relationship, the discrete single-point ranging data is spatially stitched and contour-fitted to reconstruct the overall 3D contour morphology of the wall surface and suspected welds within the detection range. Standardized weld pre-positioning area data is output, providing reliable basic data for subsequent precise inspection operations.

[0055] In the weld seam fuzzy positioning stage, the AprilTag visual positioning device and the weld seam feature point recognition device are used for multi-marker point linkage positioning of the weld seam. The AprilTag visual positioning device is installed on the top of the robot and consists of a visual acquisition unit and a marker point calculation unit. The weld seam feature point recognition device is an ultrasonic phased array sensor installed at the front of the robot. Both establish signal connections with the main control module to ensure real-time data transmission. The two form a "dual positioning anchor point". When a single AprilTag marker point is occluded, the system automatically switches to weld seam feature point positioning. Figure 4 As shown, the specific steps are as follows: N1. Positioning preparation and marking layout: Paste AprilTag visual markers at the preset positions on the tank wall. Multiple AprilTag visual markers are arranged at intervals to ensure coverage of the entire detection area, providing a stable reference anchor point for visual positioning. In step N1 above, the tank wall three-dimensional coordinate calibration device simultaneously completes the calibration of the overall three-dimensional coordinate system of the tank wall, maps the physical position of all AprilTag visual markers to the calibration coordinate system, generates standardized three-dimensional coordinate data of marker points, and provides a unified coordinate reference for visual positioning. N2. Visual positioning-led linkage positioning: After the robot starts the fuzzy positioning process, the AprilTag visual positioning device collects AprilTag mark images on the tank wall in real time, and analyzes and processes the image data to obtain the robot's 6-DOF pose information relative to the mark point. At this time, the weld feature point recognition sensor is in real-time standby detection state, only collecting weld feature point information as redundant data, and does not participate in the main positioning solution; In step N2 above, the attitude perception and compensation device collects the robot's own attitude data in real time and dynamically compensates and corrects the 6-DOF pose information to eliminate the positioning error caused by the robot's own attitude deviation. The obtained 6-DOF pose information can clearly indicate the three-dimensional positional relationship of the robot relative to the tank wall reference mark point and the weld pre-positioning area. It can not only complement the weld pre-positioning area initially generated by the infrared ranging sensor array, providing a precise initial reference for subsequent positioning work, but also allow the main control module to grasp the robot's own attitude status in real time, ensuring the stability of the linkage work in the vision positioning-dominated mode. N3. Positioning Status Monitoring and Mode Switching Judgment: The main control module determines whether the AprilTag marker is obscured. If the number of validly identified markers is lower than the preset fault tolerance threshold, the AprilTag visual positioning mode will be maintained, and the AprilTag visual positioning device will continuously output pose information. The weld feature point recognition device can simultaneously collect weld feature point data as auxiliary verification. If the number of validly identified AprilTag markers is lower than the preset fault tolerance threshold, the system will automatically trigger a positioning mode switching command, pause the visual positioning data relying on AprilTag, and seamlessly switch to the weld feature point positioning mode dominated by the ultrasonic phased array sensor to ensure that the positioning work is not interrupted. In step N3 above, whether the AprilTag marker is occluded is determined by the main control module based on the robot's 6-DOF pose information relative to the marker output in step N2, and by comprehensively monitoring the marker recognition success rate, signal strength, and pose data stability. N4. Weld Feature Point Positioning Calibration and Replacement: The weld feature point recognition device starts working, captures the spatial position coordinates of the weld toe and weld root based on the unified coordinate system of the tank wall three-dimensional coordinate calibration device, completes the positioning and replacement of weld feature points when AprilTag visual positioning fails, and calibrates the overall positioning result based on the unified coordinate system. In step N4 above, the spatial coordinates of the weld toe and weld root are captured by the ultrasonic phased array sensor relying on the penetrability and reflection characteristics of the ultrasonic signal. After the ultrasonic signal emitted by the sensor acts on the weld detection area, it can accurately identify the interface morphology difference and material height difference at the junction of the weld and the base material. At the same time, by real-time acquisition and professional analysis of the reflection duration and intensity changes of the echo signal, the spatial coordinates of the feature points of the weld toe and weld root can be accurately extracted and determined. After the switch, the industrial camera of the AprilTag vision positioning module continues to scan the tank wall in real time. Once a valid AprilTag marker image is acquired and reliable pose data is calculated, the system will automatically switch back to the AprilTag vision positioning-based mode without any awareness, restoring global positioning calibration. N5. Positioning Data Fusion and Optimization Output: The main control module integrates and optimizes the positioning data from the AprilTag visual positioning device and the weld feature point recognition device, ultimately forming a complete and accurate weld fuzzy positioning dataset, providing reliable support for subsequent accurate weld positioning, path pre-planning, and real-time correction.

[0056] Based on the aforementioned fuzzy positioning data of the weld, the sensing and detection module 3 is also equipped with a positioning device, a weld feature detection device, a visual detection device, and a curvature detection device 13, which are used to collect key feature parameters of the weld. The data collected by each device is based on a unified three-dimensional coordinate reference of the tank wall and attitude perception compensation data, and work together to achieve accurate positioning of the weld and adaptation to complex shapes.

[0057] The positioning device includes a GPS / BeiDou positioning unit, a UWB positioning unit, and a mobile tag unit. The GPS / BeiDou positioning unit simultaneously receives GPS and BeiDou satellite signals to acquire global satellite positioning data for the robot and the weld area. The UWB positioning unit and the mobile tag unit acquire near-range positioning data. The data acquired by both units are synchronously transmitted to the main control module for fusion processing to establish a precise mapping relationship between the weld position and the three-dimensional coordinate system of the tank wall, thereby achieving real-time tracking and global positioning of the weld position.

[0058] The weld feature detection device includes a laser scanning sensor and an ultrasonic array probe. The laser scanning sensor scans the weld area in real time, collecting data on the weld's appearance contour, surface flatness, and morphological differences at the weld-base material junction, and simultaneously transmits the data to the main control module. The ultrasonic array probe, relying on the penetrating and reflecting characteristics of ultrasonic signals, emits ultrasonic signals into the weld. By receiving and analyzing the reflection duration and intensity changes of the echo signals, it obtains internal structural data such as the distribution of defects and the material height difference of the weld cross-section. The main control module uses a real-time path compensation algorithm to dynamically correct the raw path data collected by the ultrasonic array probe in real time, eliminating path errors caused by movement. The robot is equipped with an inkjet marking device at its tail. When a weld defect is detected, it immediately sprays a biodegradable marking liquid to facilitate subsequent repairs.

[0059] The main control module integrates the appearance and morphological features of the weld with the data on internal structural defects to build a complete three-dimensional visualization model of the weld. Based on this model data, the main control module can accurately identify the actual distribution and direction of the weld, providing real-time and accurate decision-making basis for equipment travel path planning and real-time correction.

[0060] The main control module can also adjust the focusing depth and scanning angle of the ultrasonic array probe to achieve precise coverage of key areas such as the root and sidewall of the weld bevel by the ultrasonic beam, so as to adapt to different types of welds.

[0061] While inspecting the internal structure of the weld, the ultrasonic array probe collects cross-sectional dimension data at different locations of the weld, including key parameters such as weld width and height. This data is then transmitted synchronously to the main control module, which analyzes the weld cross-sectional dimension data to identify the range of variation in the weld cross-sectional dimensions. When a change in the weld cross-sectional dimensions is detected, causing the deviation between the robot's positioning center and the weld center to exceed a set threshold, the robot's posture is automatically adjusted to ensure precise alignment between the identification positioning center and the weld center, thus ensuring the positioning accuracy and inspection reliability of the variable cross-section weld area.

[0062] The visual detection device is used to acquire image data of the weld area in real time and transmit the image data to the main control module for preprocessing. In this embodiment, the visual detection device is an industrial CCD camera, equipped with a ring light source set coaxially with it to eliminate astigmatism on the surface of the object being measured. The industrial CCD camera is configured in wide-angle detection mode, which can meet the large field-of-view detection requirements of the front end of the body 1 for the surrounding environment. In addition to this embodiment, the visual detection device can also be an existing device such as an industrial CMOS camera or an industrial area array camera, used to acquire and transmit weld image data to the main control module.

[0063] The deep learning model embedded in the main control module has a built-in weld inflection point feature database trained on a large number of samples. This database contains visual feature templates, trajectory morphology feature templates, and feature thresholds for typical inflection points such as straight seam-circumferential seam inflection points and variable curvature inflection points under different tank types, weld specifications, and welding processes. Combined with the weld 3D visualization model generated by the weld feature detection device and the weld area image data collected by the visual detection device, the location of weld inflection points can be predicted in advance. This provides accurate and real-time decision-making basis and parameter support for adjusting the robot's walking direction and ensuring the continuity of detection at inflection points.

[0064] The curvature detection device is a curvature detection sensor used to detect the curvature radius data of the weld in real time and transmit the data to the main control module. The main control module combines the curvature radius and weld trajectory data to dynamically generate speed compensation commands and attitude adjustment commands to ensure probe fit and detection accuracy when detecting large curvature welds.

[0065] The main control module uses the Kalman filter algorithm to perform centralized fusion calculation on the real-time data transmitted by each detection device according to the preset weighting allocation coefficient, thereby improving the accuracy of weld detection in complex detection environments. It realizes the collaborative operation of the entire process from fuzzy positioning to precise positioning of the weld, and finally completes the precise detection of the weld and the acquisition of cross-sectional data, providing accurate data support for the pre-planning and real-time correction of the robot's walking path.

[0066] The sensing and detection module 3 also includes a fuselage displacement detection device and a roughness detection device.

[0067] In this embodiment, the body displacement detection device uses a grating ruler photoelectric displacement sensor to collect real-time walking displacement, travel trajectory and position deviation data of the robot when it is working on the surface of the tank. In addition to this embodiment, the body displacement detection device can also use a magnetostrictive displacement sensor, a laser rangefinder displacement sensor or other existing displacement sensors to accurately obtain the robot's spatial position information.

[0068] The main control module incorporates a high-precision closed-loop control algorithm, which is integrated with the fuselage displacement detection device, the torque sensor at the walking leg joint, and the speed sensor on the auxiliary walking device to achieve high-precision closed-loop adjustment of the walking module 2, improving the motion precision of the walking module 2. Figure 5 As shown, the core process is as follows: S1. The sensor collects data according to the sampling period: The torque sensor collects the torque value T of the force on the inner side of the walking leg joint bearing in real time. 测 The displacement sensor collects the real-time position coordinates (x, y) of its installation point. 测 y 测 , z 测 The system outputs the raw three-dimensional displacement data of the robot's movement. The speed sensors collect the rotational speed N at the axle end of the auxiliary walking device along the circumferential (x-axis), axial (y-axis), and normal (z-axis) directions of the robot's operation on the tank wall. x测 N y测 N z测 .

[0069] S2. The main control module receives and preprocesses the data collected by each sensor: The moving average filtering algorithm is used to analyze the torque value T. 测 The torque value T is obtained by smoothing the torque. The real-time displacement coordinate data (x) at the installation point of the displacement sensor 测 y 测 , z 测 The displacement coordinates are converted into real-time displacement coordinates at the robot's reference point, and error compensation is performed to obtain the calibrated displacement coordinates (x, y, z) at the reference point. Rotational speed data N for the three coordinate axes x测 N y测 N z测 After filtering and noise reduction, the calibrated three-axis velocity V is calculated based on parameters such as the wheel diameter and reduction ratio of the auxiliary walking device. x V y V z ; The preprocessed data is synchronized and aligned with timestamps to ensure the correlation of torque, displacement, and speed data in the same time dimension.

[0070] S3. The main control module calls the system's preset rated parameter threshold library and performs multi-dimensional threshold comparison and deviation calculation with the processed real-time data: Calculate the joint torque load rate according to the following formula (1). ···········(1) In the formula, The torque load rate is denoted by T, where T is the torque value at the walking leg joint collected in real time by the torque sensor, and T0 is the torque threshold built into the system. When the force exceeds the comprehensive safety factor of the components at the walking leg joint, it is determined that the joint is under abnormal force. At this time, the main control module controls the walking drive device to reduce the output torque and sends an alarm signal to remind the staff to troubleshoot the fault. While determining the joint torque load rate, the main control module retrieves the target three-dimensional displacement coordinates (x0, y0, z0) of the robot's work path planning and calculates the actual displacement deviation of the three coordinate axes according to the following formulas (2)-(4). ·········· ·········(2) ·········· ·········(3) ·········· ·········(4) In the formula, , , Let (x, y, z) represent the actual displacement deviation of the robot along the three coordinate axes, (x, y, z) represent the calibrated displacement coordinates at the reference point, and (x0, y0, z0) represent the displacement coordinates planned for the work path. like , , If the displacement deviation does not exceed the system's preset threshold, it is determined that the robot's trajectory conforms to the preset path during this time period. At this point, return to step S1 to collect data for the next time period. when , , If any of the values ​​exceeds the system's preset displacement deviation threshold, it is determined that the displacement deviation exceeds the limit, and the speed deviation linkage judgment is activated.

[0071] S4. Based on the linear velocity data of the three-axis coordinates, calculate the three-axis velocity fluctuation, deviation ratio and influence weight, and output the core adjustment axis of velocity-displacement linkage and related deviation information, providing a precise basis for subsequent motion adjustment.

[0072] S401. Retrieve the walking speed corresponding to the current position in the robot's work path planning, and decompose it into the three independent speeds V at the wheel axles of the auxiliary walking device. xr V yr V zr The axial velocity fluctuation is calculated according to the following formulas (5)-(7). ·········· ········(5) ·········· ·········(6) ·········· ·········(7) In the formula, , , V represents the velocity fluctuation of the split axis. x V y V z V is the velocity of the axis. xr V yr V zr For the axis speed in the work path planning; If V > V on a certain axis r The corresponding shaft has an overspeed deviation, marked as V. + If V < V on a certain axis r The corresponding shaft has a low-speed deviation, marked as V. - .

[0073] S402. Retrieve the actual displacement deviation of the split axis calculated in step S3. , , Sort by numerical value, mark the axis with the largest deviation value as the primary deviation axis, and the other two axes as secondary deviation axes; combine the robot motion trajectory time series data to determine the trend of the three-axis displacement deviation. If the deviation fluctuates slightly or continues to decrease, it is a steady-state deviation axis, indicating that the system is running stably or is self-correcting and no intervention is needed. If the deviation continues to increase, it is an increasing deviation axis and requires key intervention.

[0074] S403. Calculate the basic proportion coefficient of triaxial velocity deviation according to the following formulas (8)-(10). ·········· ········(8) ·········· ········(9) ·········· ·········(10) In the formula, , , This is the basic percentage coefficient for triaxial velocity deviation. , , These are the velocity fluctuations of the three axes, respectively.

[0075] S404. Differentiated assignment based on displacement deviation characteristics: Introduce displacement deviation quantification correction coefficients. For the primary and secondary deviation axes marked in S402, the displacement deviation quantification correction coefficients are respectively... and ,and > Introducing displacement deviation trend correction coefficients, the displacement deviation trend correction coefficients for the increasing deviation axis and steady-state deviation axis marked in S402 are respectively... and 1, and If the value is greater than 1, the final correction factor for each axis is the sum of the above rules. For example, if the X-axis is both the principal deviation axis and the incremental deviation axis, then the displacement deviation correction factor for the X-axis is... = × ; Furthermore, the initial influence weights of the three axes are calculated according to the following formulas (11)-(13), and normalized according to formulas (14)-(16) to obtain the final influence weights of the three axes. ·········· ········· (11) ·········· ········ (12) ·········· ·········(13) ·········· ·········(14) ·········· ·········(15) ·········· ········ (16) In the formula, , , The initial influence weights for the three axes are... , , The final influence weights for the three axes are... , , This is the basic percentage coefficient for the speed deviation of the three axes. , , This is the correction factor for the displacement deviation of the three axes.

[0076] S405. Linkage Judgment Result Output and Data Transmission: Output core information on triaxial velocity deviation, including the velocity fluctuation of the three axes. , , and the direction of velocity deviation of the three axes (V) + / V - ); Output the final influence weights of the three axes. , , To clarify the proportion of displacement deviations on each axis caused by velocity fluctuations on the corresponding axis; Output core adjustment axis mark: Mark the axis with the greatest final impact weight as the speed-displacement linkage core adjustment axis, and set the priority for subsequent displacement-speed dual-loop PID control; All judgment results are transmitted to the main control module in real time, and are integrated with torque deviation and displacement deviation data to participate in the calculation of the three-axis adjustment of the walking drive motor 204.

[0077] S5. Based on the data output from steps S3 and S4, the data processing unit of the main control module calculates the adjustment amount: the outer ring displacement deviation along the three axes. , , Calculate the triaxial speed correction value, and for the inner loop, use the triaxial speed correction value and the triaxial speed fluctuation. , , As input, calculate the triaxial speed adjustment. , , .

[0078] S6. The motion control unit of the main control module receives the data transmitted by the data processing unit, generates motion adjustment commands, adjusts the output torque and speed of the walking drive motor 204, and drives each joint of the walking leg to complete motion correction. After the walking drive motor 204 is adjusted, each sensor immediately collects new real-time data synchronously, repeats steps S1-S4, judges the deviation of torque, displacement and speed parameters, and continuously iterates and adjusts until all parameters meet the accuracy and safety requirements.

[0079] In this embodiment, the roughness detection device is a laser roughness sensor. Based on the principle of optical triangulation, it emits a laser beam towards the tank wall through a transmitting module, receives the reflected light points, calculates the displacement deviation, and generates roughness data characterizing the micro-undulations of the tank wall through analog-to-digital conversion. The main control module has a built-in adaptive gait adjustment algorithm, such as... Figure 6 As shown, based on the roughness data of the tank wall fed back by the roughness detection device, the system automatically switches between three gaits: steady, climbing, and traversing. The specific steps are as follows: R1, the roughness detection device collects data such as road surface roughness, road surface slope, and obstacle height, and transmits the data to the data processing unit.

[0080] R2, the data processing unit receives the data, processes it, and extracts key parameters such as road surface smoothness, slope, and obstacle height.

[0081] R3. Refer to the preset gait parameter library to match the corresponding working conditions and gait strategies: If the road surface is determined to be flat, a flat and steady strategy is executed, and the three walking legs move alternately at a constant speed, repeating steps L1-L4 to complete the robot's movement. If the road surface slope is detected to be greater than the set slope threshold, a climbing gait strategy is executed to increase the knee joint lifting angle and enhance the adsorption force of the main adsorption module. If the height of an obstacle is detected to be greater than the set height threshold, a straddle gait strategy is executed, in which the knee joint of a single walking leg extends to its maximum range, while the other two walking legs are locked in place.

[0082] R4, the motion control unit receives the gait strategy, generates action commands, and controls the walking drive motor 204 to complete the relevant actions.

[0083] R5, the fuselage displacement detection device, torque sensor and speed sensor collect motion execution data in real time. The motion control unit first performs preliminary PID calculations to correct small deviations; if the deviation exceeds the threshold, the deviation data is uploaded to the data processing unit.

[0084] R6. The data processing unit performs PID parameter self-tuning based on high computing power, generates corrected gait parameter commands, and sends them back to the motion control unit. The motion control unit regenerates drive signals according to the corrected commands, drives the servo motor to adjust the motion, until the actual motion state is consistent with the command requirements.

[0085] Based on the aforementioned weld seam positioning and recognition, the high-precision closed-loop control algorithm of the walking module 2, and the adaptive gait adjustment algorithm, the main control module can perform coordinated operations of pre-planning and real-time correction of the walking path during the robot's operation. Figure 7 As shown, the specific steps are as follows: P1. 3D Model Import and Path Pre-planning: The main control module imports the 3D model of the tank to be inspected and the weld trajectory model obtained from weld positioning and identification in advance, and completes path pre-planning based on the step-by-step direction of the weld. In step P1 above, path pre-planning includes nodes such as the theoretical coordinates of the path, the travel speed reference, the inflection point position, and the key inspection area of ​​the weld. At the same time, a path deviation threshold is preset to provide a judgment standard for subsequent real-time correction. All nodes must be accurately mapped to the unified three-dimensional coordinate system provided by the three-dimensional coordinate calibration device to ensure that the path coordinates are consistent with the actual spatial position of the tank wall.

[0086] P2. Real-time acquisition and preprocessing of multi-source data: The sensor module 3 starts all-round data acquisition. After all acquired data is transmitted to the main control module, it undergoes noise reduction, calibration and synchronization preprocessing to eliminate the influence of individual sensor errors, environmental interference and data acquisition timing differences. Then, it is fused centrally through the Kalman filter algorithm to improve data reliability and consistency. In step P2 above, the data acquisition performed by the sensing and detection module 3 includes: the visual detection device acquiring image data of the weld area; the body displacement detection device acquiring real-time walking displacement, travel trajectory and position deviation data of the robot; the posture perception and compensation device acquiring tilt angle, offset and motion posture data of the body 1; and the torque sensor at the walking leg joint and the speed sensor on the auxiliary walking device synchronously acquiring motor load and operating status data.

[0087] P3. Path Deviation Identification and Quantization Based on Deep Learning: The deep learning model embedded in the main control module extracts features from the preprocessed fused data and determines the robot's actual coordinates in three-dimensional space based on the extracted real-time features. By comparing the actual coordinates with the theoretical coordinates of the pre-planned path, the coordinate difference in three-dimensional space is calculated, thus quantifying the path deviation. At the same time, combined with the weld cross-sectional dimension data collected by the weld feature detection device, it is determined whether there is a deviation between the positioning center and the weld center that exceeds the standard. The quantified deviation value is used as the input condition for subsequent control algorithms. In step P3 above, the features extracted from the fused data include: the contour features of the tank wall surface, the features of the inner corner points of the tank wall (marking points, bolt holes and other dimensionally stable points), the features of the center line of the walking path, and the features of the weld feature points (marking points, inflection points), etc.

[0088] P4. Fuzzy PID algorithm drives precise correction of action execution: The main control module receives the quantized path deviation value and motor load data, and automatically and dynamically adjusts the parameters according to the preset "load change and PID parameter mapping relationship" to ensure smooth motor operation; the servo motor precisely executes control commands to adjust the robot's speed and posture. In step P4 above, the control commands generated by the main control module include: in path tracking control, the algorithm calculates the speed difference between the left and right wheels in real time according to the type of deviation (straight line, curve, broken line weld deviation or variable cross-section weld deviation), and generates speed compensation commands and attitude adjustment commands; for inflection points and large curvature weld areas, the speed difference parameters and attitude adjustment amplitude are optimized by combining the curvature radius data collected by the curvature detection sensor to ensure probe fit and detection accuracy; when a change in the weld cross-sectional size is detected that causes the deviation to exceed the standard, the robot positioning center is automatically calibrated to keep it precisely aligned with the weld center.

[0089] During the path correction process, the dynamic fall protection redundancy control algorithm and the high-precision closed-loop control algorithm work together to ensure the safe and stable operation of fuselage 1 while improving the precision of the correction action.

[0090] In the actual process of weld inspection, in order to improve work efficiency, multiple devices are often required to conduct inspections together. Since the devices are equipped with automatic obstacle avoidance function, they will automatically detour and return to the original path when an obstacle is detected. In order to avoid multi-machine collisions during the detour, the main control module has a built-in multi-machine cooperative walking scheduling algorithm to maintain the walking order of multiple devices.

[0091] Specifically, the device features a QR code identification label on its side, which, together with the wireless communication unit, enables real-time data interaction and collaborative movement scheduling between multiple devices. The wireless communication unit includes a wireless network integrated on the main control module and a signal transceiver installed on the top of the device. The signal transceiver is connected to the internal wireless network module via a shielded cable, supporting self-organizing network communication among multiple devices.

[0092] During operation, the transceiver of a single device converts the electrical signal output by the wireless network into a radio electromagnetic wave signal and transmits it outward. At the same time, it receives wireless signals transmitted by other devices in the vicinity, converts them into electrical signals, and transmits them to the wireless network. When a new device needs to establish a communication connection, it completes identity authentication using its own QR code and then accesses the network.

[0093] The visual detection device or ranging device in the sensing and detection module is used to assist in detecting obstacles in the working environment. When an obstacle (such as weld spatter or bolt) is detected, the device automatically detours and continuously sends its real-time position information (such as coordinates and heading angle) to other robots via wireless communication. Other robots dynamically adjust their walking speed or micro-path based on this information to avoid collisions during the detour.

[0094] After a single device completes obstacle avoidance and detour, it will return to the original planned path through a path correction algorithm. At this time, the device sends a "return path" status command to the main control module and other devices through the wireless communication unit. The main control module coordinates the return timing and speed of each device according to the real-time position of all devices to ensure that the return process of a single device does not interfere with the normal movement of other devices and avoids omissions or overlaps in the work area due to the path deviation of a single device.

[0095] Through the coordinated operation of the walking module 2, the sensing and detection module 3 and the main control module, the robot has acquired the ability to accurately locate welds, plan paths and walk stably, providing a solid foundation for weld inspection. To further achieve accurate identification, location and quantitative analysis of weld defects, the detection and scanning module 4 integrates multiple core technologies, forming a complete inspection system from basic scanning to precise detection, and from routine working conditions to special scene adaptation.

[0096] The detection scanning module 4 includes a multi-joint folding detection arm, on which a bracket drive motor is connected. The bracket drive motor drives the detection arm to move in a "Z-axis lifting + circumferential rotation" scheme to form a spiral three-dimensional scanning path, which can fully cover the heat-affected zone on both sides of the weld. The bracket drive motor adopts a lightweight hollow shaft servo motor, which effectively reduces the overall weight of the equipment, while having high-precision rotation performance, which can ensure the uniformity of the scanning trajectory and achieve comprehensive scanning coverage without omissions.

[0097] The inspection arm has multiple array-type defect detection devices linearly distributed on it, which are used to detect defects on the weld surface, subsurface, and internal surfaces. Each defect detection device selects an appropriate detection sensor according to the different defect detection requirements. Auxiliary detection sensors can also be selected according to the material characteristics of the actual inspection scenario to further expand the detection range. Multiple defect detection devices perform scanning work simultaneously, and the detection data is integrated and processed by a dedicated data fusion algorithm, which effectively improves the detection efficiency while ensuring the timeliness and stability of data integration.

[0098] In this embodiment, the defect detection device includes a phased array ultrasonic probe and a high-frequency array eddy current probe. The phased array ultrasonic probe emits controllable sound waves through a multi-element crystal, receives reflected signals, and generates a weld cross-section image, which can accurately identify internal defects such as cracks and incomplete penetration. The high-frequency array eddy current probe is specifically designed for surface defect detection. By adjusting the excitation frequency, it determines the size and shape of surface defects based on impedance changes, adapting to the detection requirements of conductive material welds. The two detection methods work simultaneously, achieving synchronous detection of internal and surface defects. The detection areas effectively overlap, ensuring comprehensive defect detection without blind spots. The data is cross-validated, effectively improving the defect detection rate and avoiding the limitations of a single detection technology.

[0099] In addition to this embodiment, a magnetic particle sensor can be selected according to the characteristics of the material being tested. By combining the magnetized magnetic field with visual imaging, auxiliary detection of surface cracks in ferromagnetic materials can be carried out, further expanding the detection application scenarios.

[0100] The main control module can control the focusing depth of the phased array ultrasonic probe. In the initial scan stage, a conventional focusing depth is used to achieve full coverage of a large detection area and quickly identify potential defects. When a defect is detected, the focusing depth of the phased array ultrasonic probe is automatically adjusted to quickly narrow the scanning range and achieve precise scanning of the defect area. During the secondary focusing scan, eddy current sensors are used to conduct specialized detection on surface defects. Through the fusion of multi-dimensional detection data, quantitative analysis of key parameters such as defect length, depth, and width is achieved.

[0101] To adapt to various complex weld inspection needs and ensure consistent inspection accuracy and efficiency in different inspection scenarios, the main control module dynamically and adaptively optimizes key variables during the scanning process, including the adjustment of the scanning step distance and the scanning frequency.

[0102] Specifically, based on the weld width data collected by the weld feature detection device in the sensing and detection module, the main control module automatically adjusts the scanning step distance. When a narrow weld is detected, the scanning step distance is automatically reduced to ensure the density and accuracy of the detection data; when a wide weld is detected, the scanning step distance is appropriately increased to balance detection efficiency and coverage integrity.

[0103] Based on the robot's walking speed data collected by the rotation speed sensor on the auxiliary walking device, the main control module automatically adjusts the scanning frequency to ensure that the number of scanning points per unit length meets the detection accuracy requirements, avoids deviations in detection data due to changes in walking speed, and further ensures the stability and accuracy of the detection process.

[0104] To enhance the robot's environmental adaptability and operational flexibility, and ensure stable and efficient testing in various special scenarios, the testing arm is wrapped with a dedicated water-cooled heat dissipation sleeve. This sleeve utilizes circulating cooling water for efficient heat dissipation and is equipped with water level monitoring and alarm functions to prevent heat dissipation failure due to water shortage. Both the robot body shell and the probe shell are made of high-temperature resistant materials, effectively resisting the impact of high-temperature environments on the equipment components. Furthermore, a temperature and humidity sensor is installed on the side of the robot body to collect real-time environmental temperature and humidity data. When environmental parameters exceed the adaptation threshold, the system automatically adjusts the core testing parameters to ensure the accuracy of testing data under high temperature and high humidity conditions, guaranteeing the continuous and stable operation of the testing work.

[0105] The detection arm has a two-section hinged structure, which can flexibly switch between unfolded and folded states. When unfolded, it ensures the scanning range of conventional detection. When folded, the size of the equipment is greatly reduced, which can smoothly enter narrow working spaces. The folding and unfolding actions of the detection arm are driven by a pneumatic cylinder to achieve rapid switching. During the switching process, the adsorption stability of the machine body 1 is not affected, ensuring that the scanning area can still be fully covered in narrow spaces without any blind spots.

[0106] The detection arm is equipped with a displacement sensor to detect the deviation between the scanning path and the preset weld trajectory in real time. Based on the deviation data, the main controller drives the servo motor to adjust the position of the probe frame through a mechanical transmission structure, so as to realize the rapid compensation and correction of the path deviation and ensure that the scanning path is always accurately aligned with the center of the weld, effectively avoiding the distortion of detection data caused by path deviation.

[0107] To achieve remote, safe operation and visual monitoring, this intelligent inspection robot integrates a remote control algorithm and a 5G / Wi-Fi dual-mode communication module. It enables remote control command transmission, ensures data security through AES-256 encryption, and uses a remote control interface as a visualization platform to display real-time device status and inspection data. It supports functions such as 3D reconstruction of the scan trajectory and defect location marking, is compatible with multiple terminals, and features a remote emergency stop function. Furthermore, it adopts an ergonomic design, with curved gripping grooves on both sides of the body for anti-slip treatment. The operation buttons are conveniently located near the gripping area for one-handed operation, and danger zones are marked with warning colors. The emergency stop button supports both physical and remote operation, balancing operational safety and comfort.

[0108] In terms of human-computer interaction, the top of the device is equipped with a touch screen display with a simple and efficient UI interface, which can display relevant data and various status information of the device in real time. The core function operation logic is simple, and the interface data is synchronized with the remote visualization platform. The device also integrates a voice recognition module, which supports Chinese voice commands. It can perform operations such as starting and stopping the device, switching modes, and querying data through voice. The query results are fed back through both voice and screen, which improves the convenience of human-computer interaction and operation efficiency.

[0109] From the perspective of operational continuity and hardware safety, the intelligent inspection robot body 1 is equipped with a dual independent battery compartment structure, supporting hot-swappable battery replacement for quick replacement without stopping the machine, ensuring the continuity of inspection operations. At the same time, the whole machine is equipped with a dedicated charging management module, which can flexibly switch between fast charging and slow charging modes to adapt to the energy replenishment needs of different operating scenarios. The charging management module has a built-in overload protection mechanism. When the motor operating current is detected to exceed 120% of the rated value, the power supply circuit is immediately and automatically cut off and an audible and visual alarm is triggered, thus avoiding the risk of equipment failure caused by overload operation from the hardware level.

[0110] The battery compartment and main control unit are both integrated at the bottom of the body 1, which effectively lowers the robot's center of gravity and enhances the structural stability and anti-tipping ability of the equipment. The overall appearance adopts a streamlined design, which can effectively reduce the interference of air resistance during movement and further improve the stability of walking posture.

[0111] To balance structural compactness with flexibility in adapting to different scenarios, this intelligent inspection robot adopts a modular integrated design and stacked spatial layout, achieving a highly compact overall structure. While optimizing the utilization of internal space, it scientifically plans multiple heat dissipation channels to ensure the heat dissipation efficiency of core components. The core functional modules adopt a quick-release structural design, enabling rapid disassembly and replacement for maintenance, significantly reducing equipment maintenance costs and downtime. At the same time, the body 1 has a reserved standardized expandable interface panel, supporting the flexible installation and expansion of various functional modules. It can complete the functional configuration adjustment according to the needs of different inspection scenarios and adapt to diverse inspection operation requirements.

[0112] The main body, outer shell, and key load-bearing components of the fuselage are made of aerospace-grade alloys and composite materials suitable for the operating conditions. Combined with a lightweight hollow structure design, high-strength lightweight component selection, and high-density integrated packaging, the overall lightweight design goal is achieved while ensuring the structural strength and load-bearing capacity of the equipment, taking into account both the structural reliability and portability of the equipment. The equipment adopts a fluororubber sealing ring, pressure balance valve, and waterproof and explosion-proof joint to construct a full-dimensional anti-corrosion and sealing protection structure. The gaps, interfaces, and compartments of the fuselage are sealed in all directions to effectively isolate the intrusion of water vapor and corrosive media in complex environments, ensuring the stable operation of the equipment in complex temperature and humidity environments such as high and low temperatures, high humidity, and multiple corrosive gases.

[0113] In addition, all key mating parts of the equipment are precision machined using high-precision CNC machining technology, strictly controlling the fitting accuracy of the kinematic pairs and the overall motion accuracy of the equipment; high-elasticity impact-resistant buffer structures are added to key impact-prone locations such as the bottom of the machine body, the end of the legs, and the motor connection, which greatly improves the vibration and impact resistance of the equipment, effectively mitigating the vibration and impact loads on the equipment during walking, obstacle crossing, and operation, and ensuring the operational reliability of core components and motion mechanisms.

[0114] For those skilled in the art, any changes, modifications, substitutions, and variations made to the embodiments without departing from the principles and spirit of the present invention, based on the teachings of the present invention, still fall within the protection scope of the present invention.

Claims

1. An industrial intelligent inspection robot, comprising a body (1), a walking module (2) installed at the bottom of the body (1), and a rotating platform (5) disposed at the center of the body (1), wherein a sensing and detection module (3) and a detection and scanning module (4) are integrated on the rotating platform (5), characterized in that, The main control module is installed inside the body (1). The main control module is connected to the walking module (2), the sensing and detection module (3), and the detection and scanning module (4) respectively. It is used to receive the data collected by each module and issue control commands to realize the functions of weld positioning, path planning, walking control and defect detection.

2. The industrial intelligent inspection robot according to claim 1, characterized in that, The main control module adopts a dual-main-control architecture, including a data processing unit and a motion control unit. The data processing unit is used for multi-source data fusion, algorithm calculation and global decision scheduling, while the motion control unit is used to parse control commands and generate motion drive signals.

3. The industrial intelligent inspection robot according to claim 2, characterized in that, The walking module (2) includes multiple walking legs. The walking legs are designed based on the bionic structure of human legs, including an upper support leg (201), a lower support leg (202), and a support foot (203). The posture is adaptively adjusted through the hinged cooperation of the hip joint, knee joint, and ankle joint. The bottom of the support foot (203) integrates an auxiliary walking device, a main adsorption device, and an emergency adsorption device. The main adsorption device and the emergency adsorption device have the same structure and adopt an electromagnetic-vacuum dual-mode switching system. The three walking legs adopt a staggered control strategy. When one walking leg is powered off, the other two remain in an adsorption state. The joints of the walking legs are equipped with torque sensors, and the supporting foot (203) is equipped with an electromagnetic brake locking device. Together with the emergency adsorption device, it works in conjunction with the dynamic anti-fall redundant control unit of the main control module to achieve anti-fall protection.

4. The industrial intelligent inspection robot according to claim 2, characterized in that, The rotating platform (5) includes a platform base connected to the body (1). The platform base has a built-in torque sensor to collect the eccentric torque and rotational resistance torque of the platform in real time. A differential rotation mechanism is provided under the platform base. The output axis of the differential rotation mechanism extends upward and is connected to the sensing and detection module (3) and the detection and scanning module (4). An angle sensor is provided at the connection point, and the main control module has a built-in high-precision dynamic balance control algorithm.

5. The industrial intelligent inspection robot according to claim 2, characterized in that, The sensing and detection module (3) is equipped with a ranging device, a visual positioning device, a weld feature point recognition device, an attitude perception compensation device, and a tank wall three-dimensional coordinate calibration device; the fuzzy positioning and precise positioning of the weld are achieved through a multi-anchor point linkage positioning scheme, and the weld inflection point is predicted by a deep learning model to ensure the continuity of detection.

6. The industrial intelligent inspection robot according to claim 2, characterized in that, The detection scanning module (4) includes a detection arm and an array-type defect detection device. The detection arm can realize Z-axis lifting and circumferential rotation to form a spiral three-dimensional scanning path. The defect detection device adopts a dual-mode synchronous detection method to simultaneously collect internal and external defects and cross-sectional parameters of the weld, and realize the overlapping and cross-verification of the detection area.

7. The industrial intelligent inspection robot according to claim 3, characterized in that, The main control module has a built-in high-precision closed-loop control algorithm and an adaptive gait adjustment algorithm. The high-precision closed-loop control algorithm integrates torque, displacement and speed sensor data to achieve precise adjustment of the walking module (2). The adaptive gait adjustment algorithm automatically switches the walking gait according to the tank wall working condition data.

8. The industrial intelligent inspection robot according to any one of claims 1-7, characterized in that, The machine body (1) is equipped with a wireless communication module (6) and a QR code identification mark, which supports self-organizing network communication and identity authentication of multiple devices. The main control module has a built-in multi-machine collaborative walking scheduling algorithm to realize the sharing of location information of multiple devices, dynamic adjustment of paths and coordination of return timing, so as to avoid collisions and omissions or overlaps in the work area.

9. The industrial intelligent inspection robot according to claim 8, characterized in that, The main control module can dynamically adjust the scanning step distance based on the weld width data collected by the sensing module (3) and optimize the scanning frequency based on the walking speed data of the walking module (2), taking into account both detection accuracy and work efficiency.

10. The industrial intelligent inspection robot according to claim 1, characterized in that, The robot supports 5G / Wi-Fi dual-mode remote communication and encrypted control, and has functions such as remote emergency stop, inspection data visualization, and defect marking.

Images