U-rib weld defect detection mass center self-balancing chassis system and control method

Abstract

According to the mass center self-balancing chassis system for U-rib weld defect detection and the control method, the problem that when detection equipment moves on a U-rib curved surface, the precision is reduced due to mass center deviation and view field inclination can be solved. The system integrates a Mecanum wheel omni-directional moving unit, four groups of electric telescopic supporting legs, a two-axis holder and an ECU (electronic control unit), and has the innovation point that dynamic mass center balance and posture self-correction are realized through rigid-flexible combined supporting leg design and multi-modal sensing fusion. The ECU adopts an improved Canny algorithm to process the image, extracts a U rib edge point set and fits a reference center; in combination with pressure and tilt angle sensor data, a real-time mass center is solved through a moment balance formula, and then an improved self-adaptive PID algorithm is used for driving the supporting legs to stretch out and draw back, so that the mass center returns to a reference center; the two-axis holder compensates the view field angle according to the inclination angle data to ensure that the view field angle is perpendicular to the weld joint. Through cooperation of a mechanical structure and an algorithm, the system can control centroid offset to be smaller than or equal to 5 mm and view field inclination to be smaller than or equal to 1 degree, the detection stability and precision are remarkably improved, and the algorithm is easy to deploy.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of automatic control, and simultaneously relates to a parameter setting method of a pressure detection and control system, and is especially suitable for the optimization and debugging technology of PID parameters in a pressure sensor closed-loop control system based on a critical proportional method. BACKGROUND

[0002] Currently, the defect detection of the U-rib weld of the bridge steel box girder mainly relies on automatic detection equipment (such as a detection trolley equipped with a camera and an infrared thermal imager), but the particularity of the U-rib structure (the cross section is a curved surface in the shape of “U”, the two side walls are inclined, and the welds are distributed at the connection between the curved surface and the plane) brings significant challenges to the stable operation of the detection equipment.

[0003] The centroid offset leads to poor stability. The U-rib curved surface has changes in curvature and height differences, and when the detection trolley moves longitudinally on the curved surface, transversely across the rib (U-rib curved surface transverse movement), or turns, the vehicle body is prone to tilt, causing the overall centroid to deviate from the support center. The chassis of the existing detection equipment is mostly rigid structure, lacking dynamic centroid adjustment capability, and relying only on the driving balance of the Mecanum wheel, which is prone to skidding, rollover or increased vibration, seriously affecting the motion stability.

[0004] The detection field of view is inclined, reducing the accuracy. The camera and thermal imager of the traditional detection equipment are mostly fixedly installed, and the inclination of the U-rib curved surface will directly cause the detection field of view to be not perpendicular to the weld, resulting in image distortion and distortion of defect features (such as blurred weld edges and misjudgment of the area of thermal defects). Although some equipment is equipped with a single-axis gimbal, it cannot compensate for the inclination in the pitch and roll directions at the same time, and it is difficult to ensure the perpendicularity of the field of view.

[0005] The existing centroid adjustment algorithm is difficult to engineer. The existing centroid adjustment scheme mostly uses intelligent algorithms (such as genetic algorithm, neural network), which have high theoretical accuracy, but the calculation is time-consuming and requires high hardware computing power, making it difficult to run in real time in an industrial-grade embedded system. At the same time, the edge detection relies on the traditional Canny algorithm, which is prone to edge breakage or false edges in the U-rib curved surface reflection and uneven lighting scenes, resulting in large reference center positioning errors and affecting the accuracy of the centroid adjustment reference.

[0006] The equipment coordination is insufficient. In the existing system, the centroid adjustment (mechanical structure) and defect detection (vision / thermal imaging) are mostly independent modules, lacking data linkage: the detection field angle of the detection equipment is not corrected synchronously when the centroid is offset, resulting in mismatch between the detection data and the spatial position, and low defect positioning accuracy.

[0007] In summary, for the U-rib curved surface detection scene, an integrated chassis system is proposed, which is simple in structure, efficient in control, and can realize dynamic balance of the centroid and real-time compensation of the field of view, to solve the problems of poor stability, low accuracy and difficult engineering in the prior art. SUMMARY

[0008] Problems to be Solved by the Invention

[0009] The present application aims to solve the problems of poor stability and detection accuracy caused by field of view tilt in U-rib weld detection when the device moves on a curved surface due to centroid offset; at the same time, overcome the defects of edge breakage or many false edges of traditional Canny algorithm in U-rib reflection and uneven illumination scene, and the problems of regulation lag caused by fixed PID parameters and lack of device coordination, and improve the detection stability and accuracy.

[0010] Means for Solving the Problems

[0011] The control process of the U-rib weld defect detection centroid self-balancing system is as follows: after system initialization, the process is executed in a loop, pressure sensor support force, tilt sensor data and U-rib image are collected; the improved Canny algorithm processes the image, extracts the edge and fits the reference center; the real-time centroid and offset are calculated combined with the pressure and tilt data; the adaptive PID algorithm outputs the extension amount of the support leg and the angle of the gimbal according to the offset; the support leg and the gimbal are driven to move, and the centroid offset and field of view tilt are compensated; the sensors provide real-time feedback, and the adjustment is repeated until the detection is completed. In order to optimize the control process, the present application proposes innovations in mechanism, method and the like from the following aspects, which are as follows:

[0012] In a first aspect, the present application proposes a self-adaptive contact type support leg mechanism, characterized in that a built-in spoke type pressure sensor and a spring-damping assembly are provided, which not only ensures reliable contact with the U-rib curved surface, but also buffers the impact of the curved surface, and directly collects support force data to provide high-precision input for centroid calculation.

[0013] The mechanism comprises:

[0014] Step 1: initial installation and readiness, the ball surface contact wheel with pressure sensor is connected to the end of the electric telescopic support leg, the spring-damping assembly is arranged between the sensor and the support leg, after the system is started, the support leg is reset, the sensor is self-checked and calibrated to ensure that the initial state of force value collection is accurate;

[0015] Step 2: curved surface fitting and force sensing, when the trolley moves to the U-rib curved surface, the support leg is self-adaptively telescoped according to the curved surface radius, the ball surface contact wheel is fitted to the curved surface, the spring absorbs the impact of contact, and the pressure sensor collects the support force in real time and transmits the signal to the control unit as the basis for centroid calculation;

[0016] Step 3: action adjustment, the control unit drives the support leg to telescope according to the centroid offset instruction, adjusts the contact position of the contact wheel and the curved surface, and in the adjustment process, the spring-damping assembly absorbs vibration, the contact wheel tilts slightly with the curved surface to keep fitting, and the stable collection of force value is ensured;

[0017] Step 4: Protection and feedback loop. If the sensor detects abnormal support force, the control unit triggers the corresponding outrigger to retract and relieve the force. After adjustment, feedback is provided in real time to form a closed loop, ensuring that the outrigger always provides stable support.

[0018] Secondly, this application provides a multimodal edge detection algorithm for use in a defect detection system. The defect detection system includes a vehicle-mounted system that can achieve two degrees of freedom of rotation and pitch via a two-axis servo gimbal. A visual camera and an infrared thermal imager are integrated side by side on the gimbal. An improved Canny algorithm is proposed, which dynamically adjusts the Gaussian filter kernel through local variance, fuses multi-scale Sobel edge responses, and combines dynamic dual thresholds with region growing to effectively suppress U-rib reflections and uneven illumination interference, thereby improving edge continuity and the accuracy of the reference center fitting.

[0019] The method includes:

[0020] Step 1: Adaptive Gaussian filtering preprocessing, dynamically adjusting the standard deviation of the filter kernel based on the gray-level variance of local image regions, using the following formula:

[0021]

[0022] Where σ0 is the baseline standard deviation, α is the adjustment coefficient, and Var(I) is the local variance, achieving noise reduction while preserving edge details;

[0023] Step 2: Multi-scale gradient calculation. The gradient is calculated using the Sobel operator at different scales (e.g., 3×3, 5×5, 7×7), and the gradient magnitude is... direction Multi-scale edge graph fusion by weighting weight w i Determined by the proportion of edge intensity at each scale;

[0024] Step 3: Direction-sensitive non-maximum suppression. Discretize the gradient direction into 8 intervals. For each pixel, compare the gradient magnitude of neighboring pixels along the gradient direction, retain local maxima, and suppress non-edge pixels.

[0025] Step 4: Dynamic dual threshold segmentation, using the Otsu algorithm to calculate the global threshold T. global Dynamically adjust high and low thresholds T H =β·T global T L =γ·T global (γ=0.4β,β∈[1.2,1.5]), higher than T H The edge is strong, between T L With T H Furthermore, edges connected to strong edges are weak edges;

[0026] Step 5: Region growing edge optimization, with strong edges as seeds, by searching for connecting weak edges, removing isolated noise points, and enhancing edge continuity.

[0027] In the third aspect, in the U-rib curved surface weld detection, the detection trolley needs to adapt to the curved surface undulation, the posture adjustment (infrared thermal imager and visual camera rotation) of the two-axis servo holder and the electric contraction of the chassis universal joint coupling continuously change the mass center distribution of the system. The traditional static mass center model cannot adapt to the dynamic working condition, which easily leads to balance instability and detection field deviation. Therefore, the application provides a dynamic mass center calculation model, which constructs a mass center calculation formula containing posture correction based on a pressure sensor network and inclination data, introduces Kalman filtering to fuse multiple source data, and realizes real-time elimination of high-frequency noise and accurate calculation of mass center coordinates under curved surface motion (error ≤5mm).

[0028] The model comprises:

[0029] Step 1: Multi-source data acquisition and calibration, four groups of universal joint legs collect original force values f i (i=1, 2, 3, 4), and calibrated F i = kf i (k is the force-voltage conversion coefficient), to obtain the support reaction force F i ; on-site verticality compensation, the roll angle α and the pitch angle β of the two-axis servo holder are calculated according to the inclination sensor data θ p , θ r , and the formula is:

[0030]

[0031] Where h is the height of the holder, motion smoothing processing is performed on the holder rotation instruction to avoid impact vibration. The visual camera extracts the U-rib reference center coordinates (x p , y p ), the leg installation coordinates (x i , y i , z i ), the holder module mass m c , and the zero position mass center (x c0 , y c0 , z c0 ).

[0032] Step 2: Mass center calculation of the pressure sensor network, in the absolute coordinate system O-XYZ, the support forces of the four groups of legs are F1, F2, F3, and F4, and the installation coordinates are (x i , y i , z i )(i=1, 2, 3, 4), then the real-time mass center coordinates (X c , Y cZ c ) The calculation formula is:

[0033]

[0034] Considering the influence of the inclination angle θ p (pitch angle) and θ r (roll angle) on the centroid projection, the revised formula is:

[0035] X′ c = X′ c cos θ p -Z c sin θ p Y′ c = Y c cos θ r -Z c sin θ r

[0036] Then, according to the pressure sensor data F i and the inclination angle sensor data θ p , θ r , the real-time centroid coordinates (X′ c , Y′ c ) can be calculated;

[0037] Step 3: Correction of gimbal attitude to centroid (rotation matrix). The gimbal rotation will change the position of its centroid in the chassis coordinate system. The Euler angle rotation matrix is used to correct the gimbal centroid position:

[0038]

[0039] Real-time centroid coordinates after gimbal adjustment:

[0040] Step 4: Improve the centroid calculation formula according to step 2, use rigid body centroid mechanics solution (force-position fusion), system total gravity balance: (M is the total mass, g is the acceleration of gravity); Centroid coordinate formula (based on spring damping and installation position of outrigger):

[0041]

[0042] Step 5: Error compensation of visual feedback (Kalman filter). The theoretical center of U rib (x0, y0) is detected by vision, and the offset is calculated:

[0043] Δx = x p -x0, Δy = y p -y0

[0044] Calculate the mass center offset ΔX caused by the inertia force according to the motion state (speed v, acceleration a) of the Mecanum wheel inertia , ΔY inertia , the formula is Where F inertia = ma, h is the mass center height, and w is the chassis height

[0045] Total offset: ΔX = Δx + ΔX inertia , ΔY = Δy + ΔY inertia ;

[0046] Un-denoised mass center deviation:

[0047] Fusion corrected mass center: (x′ c , y′ c ) = K·(x c , y c ) + (1-K)·(x c - ΔX, y c - ΔY) (K is the filtering gain, and the balance of the mechanical model and the credibility of visual observation.)

[0048] The universal joint coupling provides F i through the pressure sensor, and the change of the supporting force directly affects the calculation of the mass center; the two-axis servo head, through the data transmitted by the tilt angle sensor, calculates the attitude angle, which is used to correct the mass center offset caused by the attitude transformation of the head itself; the visual camera provides the center reference of the upper wall of the U rib, and compensates the cumulative error of the mechanical model through geometric deviation.

[0049] In a fourth aspect, the application proposes a fuzzy adaptive PID control strategy, which combines ZN parameter setting and fuzzy rule base, dynamically adjusts PID parameters according to mass center offset and rate of change, combines integral separation and segmented adjustment strategy, and filters the error signal through Kalman filtering, dynamic response optimization, dynamic adjustment of PID parameters according to system response time, realization of leg cooperative control; balance between fast response and steady-state accuracy, adapt to U rib curved surface working condition change.

[0050] The control strategy includes:

[0051] Step 1: parameter self-tuning mechanism

[0052] Initialize parameters based on Ziegler-Nichols method:

[0053] K p = 0.6K cr , K d = 0.075K cr T cr

[0054] where K cr is the critical gain, T cr is the critical oscillation period, the parameters are adjusted in real time by fuzzy control rules, and the rule base includes:

[0055] IF |e(t)| is LARGE AND is LARGE THEN ΔK p = +Δ1, ΔK i = 0, ΔK d = +Δ2

[0056] IF |e(t)| is SMALL AND is SMALL THEN ΔK p = -Δ3, ΔK i = +Δ4, ΔK d = -Δ5

[0057] Step 2: Integral separation strategy:

[0058] When |e(t)|>β, the integral term is canceled to avoid integral saturation:

[0059]

[0060] Step 3: Anti-interference filtering: Kalman filtering is performed on the error signal and centroid deviation e(t), and the formula is:

[0061]

[0062] where K f is the filtering gain;

[0063] Step 4: Calculate the leg extension amount ΔL i based on the adaptive PID algorithm, and the formula is:

[0064]

[0065] Parameters K p , K i , K d are adjusted according to the system response dynamics;

[0066] Step 5: Leg coordination control, decompose the overall centroid adjustment into coordinated actions of each leg, and map through the Jacobian matrix:

[0067]

[0068] where J is the leg kinematics Jacobian matrix, and the transformed formula is:

[0069]

[0070] Simplify as: ΔL i = k x,i · ΔX + k y,i · ΔY + k p,i · θ p + k r,i · θ r , where k x,i , k y,i , k p,i , k r,i is a mechanical structure parameter (determined by the leg geometry);

[0071] Step 6: The segmented adjustment strategy is based on the filtered centroid deviation Segmented control, (threshold), disable the integral term, and only use proportional and derivative fast adjustment; when enable the integral term to eliminate steady-state error; at the same time, combine the system response time feedback to dynamically correct the PID parameters, balance fast response and steady-state accuracy;

[0072] Step 7: Dynamic response optimization refers to adjusting control parameters (such as PID parameters) in real time, so that the system can quickly track the target centroid when the centroid deviates, reduce overshoot and adjustment time, and suppress oscillation. Combined with fuzzy rules and response time feedback, a dynamic balance is achieved between fast response (such as large deviation) and steady-state accuracy (such as fine adjustment), adapting to the real-time changes of U-rib curved surface working conditions. According to the system response time T r adjust the PID parameters, the formula is:

[0073]

[0074] where T target is the target response time;

[0075] In a fifth aspect, the application provides a multi-actuator cooperative adjustment mechanism, which establishes a linkage model of leg extension and pan-tilt rotation, maps the centroid deviation to each leg adjustment amount through the Jacobian matrix, synchronously drives the pan-tilt to compensate for the field tilt, forms a "centroid balance-field vertical" double-loop cooperative control, and improves the overall stability of the system.

[0076] The cooperative adjustment mechanism comprises:

[0077] Step 1: Data acquisition and state perception, the vision camera and infrared thermal imager on the two-axis servo pan-tilt collect U-rib images to obtain the weld position and contour; the pressure sensors at the four sets of universal joint couplings of the base collect support forces, and the tilt angle sensors detect the attitude of the chassis. All data are transmitted to the control unit in real time.

[0078] Step 2: Data processing and decision generation, the control unit processes the image to obtain the weld reference center, combines the pressure and inclination data to calculate the centroid offset; according to the offset, generate two-axis gimbal angle adjustment instructions and universal joint extension amount instructions to ensure that the two action parameters match;

[0079] Step 3: Actuator cooperative action, motor drives universal joint coupling main shaft rotation, realizes the telescopic leg to adjust the support position; at the same time, the two-axis servo gimbal rotates according to the instruction, keeps the camera and thermal imager field of view aligned with the weld; the two synchronous actions compensate for the centroid offset and field of view tilt.

[0080] Step 4: Feedback adjustment and closed-loop control, the sensor real-time feedback of the adjusted support force, attitude and image information, the control unit compares the target state, calculates the offset again and fine-tunes the actuator action, forms a closed loop, until the system centroid is stable and the field of view is accurately aligned.

[0081] In a sixth aspect, the application provides a U-rib weld defect detection centroid self-balancing system:

[0082] A controllable autonomous motion trolley, a two-axis servo gimbal is installed on the trolley, an infrared thermal imager and a visual camera are integrated on the gimbal, the visual camera locates the U-rib weld area and extracts the edge profile; the infrared thermal imager synchronously captures the weld temperature field and identifies the temperature abnormal area; the two data fusion accurately locates the defect position and type; in order to prevent the dark environment inside the U-rib from affecting the detection accuracy, a fill light is integrated on the trolley for lighting the dark environment of the U-rib; the trolley chassis adopts a modular frame design, the main body is a high-strength aluminum alloy profile splicing structure, which takes into account lightweight and rigidity; four groups of electric telescopic legs are symmetrically arranged at the four corners of the chassis, each group of legs takes the coupling main shaft as the core transmission component, is connected through the universal joint and the Mecanum wheel to form an omnidirectional moving support unit. The driving logic of the electric telescopic leg is: the servo motor directly connects the coupling main shaft through the reducer, when the motor rotates in forward and reverse directions, the main shaft converts the linear extension and retraction movement of the extension rod through the screw pair (or crank slider structure); at the same time, the connecting rod mechanism hinged at the end of the main shaft swings synchronously with the rotation of the main shaft, driving the wheels to adaptively adjust the angle within ±15°, ensuring the fit with the U-rib curved surface; the spring-damping assembly is nested in the middle of the leg, and the pressure sensor is located above the spring and the damping, which can collect the support force in real time; the inclination sensor is integrated at the connection between the connecting rod mechanism and the chassis to monitor the leg swing angle. The sensor signals are connected to the central controller through the internal wiring slot to provide data support for centroid balance adjustment; the two-axis servo gimbal mounting position is reserved in the middle of the chassis, and the leg motor control line and sensor signal line are integrated through the cable hole to ensure that the structure is compact and the transmission and detection systems operate cooperatively.

[0083] In a seventh aspect, the protectable points of the present application include: a structural design of an adaptive contact leg mechanism and a force sensing buffer scheme; an adaptive filtering and dynamic threshold algorithm for multi-modal edge detection; a dynamic centroid calculation model with attitude correction; an adaptive PID control strategy combining ZN setting and fuzzy rules; and a linkage mechanism for coordinated adjustment of multiple actuators. All of the above technical solutions are novel and creative and constitute independent protection ranges. BRIEF DESCRIPTION OF DRAWINGS

[0084] Figure 1 A schematic diagram of each module of the U-rib defect detection centroid self-balancing system of the present application;

[0085] Figure 2 A schematic diagram of the U-rib defect detection centroid self-balancing device of the present application;

[0086] Figure 3 A schematic diagram of the U-rib defect detection centroid self-balancing chassis structure of the present application;

[0087] Figure 4 A schematic diagram of the electric telescopic leg structure of the present application;

[0088] Figure 5 A schematic diagram of the U-rib defect detection method flow of the present application;

[0089] Figure 6 A schematic diagram of the U-rib defect detection centroid self-balancing system control flow of the present application. DETAILED DESCRIPTION

[0090] The embodiments of the technical solutions of the present application will be described in detail below with reference to the accompanying drawings. The settings of these embodiments are intended to more intuitively and deeply analyze the technical features of the present application, to help those skilled in the art to understand and apply it. Unless otherwise specified, all technical terms and scientific terms involved in the present application adopt the conventional meanings generally recognized in the technical field to which the present application belongs. If a specific term needs special explanation, it will be explicitly stated in the corresponding embodiment to ensure that there is no ambiguity in understanding the technical solutions. In the process of specific description, the markers, connection relationships and work flows of the components in the drawings will be combined to gradually analyze the constitution, principles and implementation modes of the embodiments, so that the technical solutions of the present application are clear and identifiable from the overall architecture to the local details.

[0091] Specific implementation mode one: Figure 1 A schematic diagram of each module of the U-rib defect detection centroid self-balancing system of the present application, the U-rib defect detection centroid self-balancing system of the present application mainly includes a detection trolley main body, a sensing module, a control module and an execution module;

[0092] Detection trolley body: aluminum alloy profile is used to build a rectangular chassis, four groups of electric telescopic legs are symmetrically arranged at four corners, two-axis servo holder is fixed in the middle, vision camera and infrared thermal imager are integrated on the holder;

[0093] Sensing module: spoke type pressure sensor and inclination sensor are built in the spring-damping assembly of each group of legs; IMU module is carried on the holder for attitude supplementary detection;

[0094] Control module: based on STM32 microcontroller, integrated 4-way motor drive module, CAN bus communication interface and Ethernet module (for image transmission).

[0095] Execution module: DC servo motor is used for electric telescopic legs, through the rotation of a crank linkage mechanism, the legs are driven to extend and retract; two-axis holder uses a stepping motor to realize pitch and roll adjustment.

[0096] Specific implementation method two Figure 2 It is a schematic view of the U-rib weld defect detection centroid self-balancing device; 1: infrared thermal imager; 2: vision camera; 3: pitch adjustment seat of two-axis servo holder; 4: aluminum alloy bearing chassis; 5: light supplement lamp; 6: roll rotation driving unit; 7: Mecanum wheel rim; 8: Mecanum wheel assembly; 9: inclination sensor; 10: universal joint coupling; 11: pressure sensor;

[0097] The infrared thermal imager (1) collects the temperature field of the weld, identifies thermal defects such as burn-through and non-fusion, and the visual camera (2) captures the visual image of the weld, cooperates with the improved Canny algorithm to extract the edge and the reference center of the upper wall of the U rib, and the pitch adjusting seat (3) of the two-axis servo gimbal is built-in harmonic reducer, drives the gimbal horizontal axis to pitch, and adapts the detection view angle of the curved surface; the aluminum alloy bearing chassis (4) is a hollow frame structure, integrates the wiring groove and the module installation interface, and guarantees the rigidity of the system; the light supplement lamp (5) outputs uniform light to compensate for the shadow of the U rib curved surface and improve the precision of the weld edge extraction of the visual camera; the roll rotation driving power supply (6) drives the gimbal to roll around the numerical axis, and cooperates with the pitch to realize two-axis attitude compensation; the Mecanum wheel (7) is a polyurethane rubber-coated wheel, which adapts to the complex path of the U rib curved surface through omnidirectional movement; the Mecanum wheel assembly (8) adopts a polyurethane rubber-coated wheel, and is built-in a hub motor, which is connected to the electric telescopic leg through a universal joint coupling; by using the omnidirectional movement characteristic, the telescopic leg is adapted to the ups and downs of the U rib curved surface, and the planar movement ability of the detection trolley is provided. The inclination sensor (9) collects the relative inclination of the leg and the chassis in real time, and corrects the attitude error of the centroid calculation; the universal joint coupling (10) integrates a screw transmission mechanism, a motor rotates a coupling main shaft, changes the angle between the telescopic rod and the ground, realizes the telescopic movement of the telescopic rod, and drives the wheels to swing up and down; the pressure sensor (11) is connected in series with the spring-damping assembly, detects the supporting force of the leg, and provides force feedback data for the centroid model; the cooperation logic can be divided into: detection end, attitude end, movement end, and bearing end.

[0098] The detection end: the infrared thermal imager (1) and the visual camera (2) work synchronously, the light supplement lamp (5) eliminates the shadow, and the “temperature abnormal area + visual profile” data is outputted to locate the weld defects.

[0099] The attitude end: the gimbal pitch / roll mechanism (3, 6) dynamically adjusts the detection view angle to be perpendicular to the weld plane according to the feedback of the inclination sensor (9);

[0100] The movement end: the Mecanum wheel assembly (8) drives the Mecanum wheel (7) to move, and the universal joint coupling (10) cooperates with the pressure sensor (11) to compensate for the centroid deviation in real time.

[0101] The bearing end: the chassis (4) integrates all modules, and guarantees the stability of the cooperation of multiple actuators through the rigid structure.

[0102] The components are associated through the levels of “detection-attitude-movement-bearing”, which supports the whole-process adaptive control of the U rib curved surface weld defect detection.

[0103] Specific implementation method three: Figure 3Figure 1 is a schematic diagram of a U-rib weld defect detection centroid self-balancing chassis structure of the present application; 1: Mecanum wheel; 2: spring-damping mechanism; 3: wheel hub motor; 4: telescopic leg; 5: inclination sensor; 6: motor driving universal joint main shaft;

[0104] The Mecanum wheel (1) relies on the omni-directional movement characteristics, adapts to the complex path (such as curve, variable-curvature walking) of the U-rib curved surface, and through the wheel body posture self-adaptation to fit the curved surface, guarantees the flexibility and stability of movement; the spring-damping mechanism (2) absorbs road impact (such as U-rib weld high-low difference, protrusion), buffers vibration; integrates pressure and inclination sensors, real-time collects the force size and posture angle of the supporting leg, provides core feedback for dynamic centroid calculation; the wheel hub motor (3) is divided into two motors, the left motor directly drives the Mecanum wheel, accurately controls the wheel speed and steering, supports the power output of omni-directional movement, and the right motor mainly rotates the main shaft of the universal joint coupling through rotary driving, changes the angle of the telescopic leg and the ground, realizes the extension and contraction of the telescopic leg, and drives the up-down swing of the wheel; the telescopic leg (4) adjusts the length of the supporting leg through linear extension and contraction, changes the supporting height and posture of the chassis; cooperates with the universal joint, adapts to the ups and downs of the U-rib curved surface, and actively compensates for the centroid deviation (such as elongation at the weld protrusion and shortening at the weld depression); the inclination sensor (5) real-time collects the relative inclination angle of the supporting leg and the chassis, and corrects the posture error of the centroid calculation; the motor (6) driving the main shaft of the universal joint rotates to drive the main shaft of the universal joint, converts the rotary motion of the motor into the linear extension and contraction of the telescopic leg (such as crank linkage transmission logic), dynamically adjusts the length of the supporting leg, and realizes balance control in cooperation with the centroid model; the cooperative logic can be divided into: power-movement, posture-balance, and full-link closed loop;

[0105] Power-movement: the wheel hub motor (3) provides planar movement capability to the Mecanum wheel (1), and the spring-damping mechanism (2) buffers impact and feeds back the state;

[0106] Posture-balance: the universal joint motor (6) adjusts the extension and contraction of the telescopic leg (4) by rotating the main shaft, adjusts the length of the supporting leg, and combines the sensing data of the spring-damping mechanism (2) to real-time correct the centroid of the chassis;

[0107] Full-link closed loop: movement, sensing, and posture adjustment form a closed loop of "path planning-movement execution-posture compensation-centroid stability", which adapts to the complex detection conditions of the U-rib curved surface.

[0108] Specific implementation method four Figure 4 Figure 1 is a schematic diagram of a U-rib weld defect detection centroid self-balancing chassis structure of the present application; 1: Mecanum wheel; 2: spring-damping mechanism; 3: wheel hub motor; 4: telescopic leg; 5: inclination sensor; 6: motor driving universal joint main shaft; 7: pressure sensor;

[0109] The Mecanum wheel (1) relies on the omni-directional movement characteristics, adapts to the complex path (such as curve, variable-curvature walking) of the U-rib curved surface, and is self-adapted to the curved surface through the wheel body posture, thereby guaranteeing the flexibility and stability of movement; the spring damping mechanism (2) absorbs the road impact (such as the height difference of the U-rib weld and the protrusion), and buffers the vibration; the integrated pressure and inclination sensor collects the force and the posture angle of the support leg in real time, and provides the core feedback for the dynamic centroid calculation; the wheel hub motor (3) is divided into two motors, the left motor directly drives the Mecanum wheel, accurately controls the wheel speed and steering, supports the power output of the omni-directional movement, and the right motor mainly rotates the main shaft of the universal joint coupling through the rotary drive, changes the angle between the telescopic rod and the ground, realizes the extension and contraction of the telescopic rod, and drives the up-down swing of the wheel; the telescopic rod (4) adjusts the length of the support leg through linear extension, changes the support height and posture of the chassis; cooperates with the universal joint, adapts to the ups and downs of the U-rib curved surface, and actively compensates the centroid deviation (such as the elongation at the protruding position of the weld and the shortening at the recessed position); the inclination sensor (5) collects the relative inclination of the support leg and the chassis in real time, and corrects the posture error of the centroid calculation; the motor (6) driving the main shaft of the universal joint rotates the main shaft of the universal joint, converts the rotary motion of the motor into the linear extension and contraction of the telescopic rod, dynamically adjusts the length of the support leg, and realizes balance control in cooperation with the centroid model; the pressure sensor (7) is integrated on the spring damping, is used for detecting the support force of the support leg, cooperates with the inclination sensor (5), constructs the "force + angle" double feedback, and provides the key force data for the centroid model; the cooperative logic is similar to the third embodiment.

[0110] The application proposes a fuzzy adaptive PID control strategy, which fuses ZN parameter setting and fuzzy rule base, dynamically adjusts PID parameters according to the centroid deviation and the rate of change, combines integral separation and segmented adjustment strategy, simultaneously performs Kalman filtering on the error signal to filter out sensor measurement noise and system process noise, dynamically responds and optimizes, dynamically adjusts PID parameters according to the system response time, realizes the collaborative control of the support leg, balances between fast response and steady-state accuracy, and adapts to the working condition change of the U-rib curved surface. The fuzzy adaptive PID control strategy is realized as follows:

[0111] Step 1: parameter self-setting mechanism:

[0112] Initialize parameters based on Ziegler-Nichols method:

[0113] K p = 0.6K cr , K d = 0.075K cr T cr

[0114] Wherein, K cr is the critical gain, T crFor critical oscillation period, parameters are adjusted in real time by fuzzy control rules, rule base includes:

[0115] IF |e(t)| is LARGE AND is LARGE THEN ΔK p = +Δ1, ΔK i = 0, ΔK d = +Δ2

[0116] IF |e(t)| is SMALL AND is SMALL THEN ΔK p = -Δ3, ΔK i = +Δ4, ΔK d = -Δ5

[0117] Step 2: integral separation strategy:

[0118] When |e(t)|>β, cancel the integral term to avoid integral saturation:

[0119]

[0120] Step 3: Anti-interference filter: Kalman filter is used to denoise the error signal and centroid deviation e(t), and the formula is:

[0121]

[0122] Where K f is the filter gain;

[0123] Step 4: Calculate the leg stretching amount ΔL i based on adaptive PID algorithm, the formula is:

[0124]

[0125] Parameters K p , K i , K d are adjusted according to system response dynamics;

[0126] Step 5: leg coordination control, the overall centroid adjustment is decomposed into coordinated actions of each leg, and is mapped through Jacobian matrix:

[0127]

[0128] Where J is the leg kinematics Jacobian matrix, and the transformed formula is:

[0129]

[0130] Simplified as: ΔL i= k x,i • ΔX + k y,i • ΔY + k p,i • θ p + k r,i • θ r where k x,i , k y,i , k p,i , k r,i are mechanical structure parameters (determined by the geometry of the legs);

[0131] Step 6: The segmented adjustment strategy adjusts the centroid deviation after filtering Segmented control, (threshold), disable the integral term, only use proportional and derivative fast adjustment; when enable the integral term to eliminate steady-state error; at the same time, combine the system response time feedback to dynamically correct the PID parameters, balance fast response and steady-state accuracy;

[0132] Step 7: Dynamic response optimization refers to adjusting control parameters (such as PID parameters) in real time, so that the system can quickly track the target centroid when the centroid deviates, reduce overshoot and adjustment time, and suppress oscillation. Combined with fuzzy rules and response time feedback, dynamically balance between fast response (such as large deviation) and steady-state accuracy (such as fine adjustment), adapt to real-time changes in U-rib curved surface working conditions. According to the system response time T r adjust the PID parameters, the formula is:

[0133]

[0134] where T target is the target response time;

[0135] Specific implementation method five: Figure 5The flowchart of the U-rib weld defect detection method of the present application; according to the schematic diagram, the detection method can be divided into several modules; mainly including device initialization module, mainly including trolley positioning, sensor calibration (pressure zero point / camera distortion) and parameter preset (target centroid / defect threshold); multi-source data acquisition module, mainly including visual camera (weld image), infrared thermal imager (temperature field), pressure / tilt angle sensor (leg force / attitude), IMU (gimbal attitude); data preprocessing module, mainly including image denoising (Gaussian filter), sensor data filtering (Kalman filter), temperature field normalization; dynamic balance control module, mainly including centroid offset calculation, fuzzy PID control, closed loop of leg extension / gimbal attitude adjustment, reflecting the fusion of "force-attitude-position"; weld defect recognition module, mainly including visual feature extraction (improved Canny edge detection), infrared anomaly recognition (temperature gradient analysis), feature fusion (defect criterion library), result output module, mainly including labeling defect coordinates (X, Y, Z), defect type (unfused / pinholed / cracked) detection report;

[0136] The specific implementation of the U-rib weld defect detection method includes the following steps:

[0137] Step 1: Device deployment and initialization (detection preparation stage)

[0138] Trolley positioning: place the detection trolley at the starting end of the U-rib, move it omnidirectionally through the Mecanum wheel, align the visual camera optical axis with the starting point of the weld, turn on the fill light to eliminate the shadow on the curved surface.

[0139] Sensor calibration:

[0140] Pressure sensor zeroing: record the initial force value F0 when the leg is naturally drooping, and correct the subsequent measurement value to F i =F 实测 -F0;

[0141] Camera calibration: obtain the intrinsic matrix K and distortion coefficient through the chessboard calibration board to eliminate the influence of lens distortion on weld edge extraction;

[0142] Step 2: Multi-source data synchronous acquisition (real-time detection stage)

[0143] Visual and infrared data: the two-axis gimbal drives the camera and infrared thermal imager to synchronously collect, the visual camera outputs the RGB image of the weld area, focusing on the weld edge; the infrared thermal imager outputs the temperature field data, recording the temperature distribution of the weld area;

[0144] Sensor data: the tilt angle sensor outputs the roll angle θ r and pitch angle θ p of the leg and chassis in real time; the pressure sensor collects four groups of leg support force F i, synchronous transmission to the control module;

[0145] Step 3: data preprocessing and dynamic centroid calculation

[0146] Data filtering: image preprocessing, Gaussian filtering of RGB images to remove noise, and median filtering of infrared temperature field to eliminate isolated outliers; sensor data, Kalman filter is used to correct F i and θ r , θ p , suppress vibration interference, after Kalman filtering of error signal and centroid deviation e(t), the centroid deviation formula is:

[0147]

[0148] Where K f is the filter gain;

[0149] Centroid and attitude correction:

[0150] Centroid calculation of pressure sensor network, in the absolute coordinate system O-XYZ, the support force of the four groups of legs is F1, F2, F3, F4, the installation coordinates are (x i , y i , z i )(i=1,2,3,4), then the real-time centroid coordinate (X c , Y c , Z c ) calculation formula is:

[0151]

[0152] Considering the influence of inclination θ p (pitch angle) and θ r (roll angle) on the centroid projection, the corrected formula is:

[0153] X′ c = X c cosθ p -Z c sinθ p Y′ c = Y c cosθ r -Z c sinθ r

[0154] Then the real-time centroid coordinates (X′ i , Y′ p ) can be calculated according to the pressure sensor data F r and the inclination sensor data θ c , θ c ;

[0155] Correction of gimbal attitude to centroid (rotation matrix), gimbal rotation will change its own centroid position in the chassis coordinate system, through the Euler angle rotation matrix to correct the gimbal centroid position:

[0156]

[0157] Real-time centroid coordinates after gimbal adjustment:

[0158] According to step 2, the centroid calculation formula is improved, and the rigid body centroid mechanics is solved (force-position fusion), and the system total gravity balance is realized: (M is the total mass, g is the acceleration of gravity); Centroid coordinate formula (based on spring damping and installation position of outrigger):

[0159]

[0160] Error compensation of visual feedback (Kalman filter), visual detection U rib theoretical center (x0, y0), calculate the offset:

[0161] Δx = x p -z0, Δy = y p -y0

[0162] According to the motion state (speed v, acceleration a) of Mecanum wheel, the centroid offset ΔX inertia , ΔY inertia caused by inertia force is calculated, and the formula is Where, F inertia = ma, h is the centroid height, and w is the chassis height;

[0163] Total offset: ΔX = Δx + ΔX inertia , ΔY = Δy + ΔY inertia ;

[0164] Un-denoised centroid deviation:

[0165] Fusion corrected centroid: (x′ c , y′ c ) = K·(x c , y c ) + (1-K)·(x c - ΔX, y c - ΔY) (K is the filter gain, balance the credibility of mechanical model and visual observation.)

[0166] Step 4: Outrigger and gimbal cooperative adjustment (balance control)

[0167] Deviation calculation: calculate the offset of real-time centroid and target centroid and offset rate

[0168] Fuzzy PID regulation: when e(t) > 5mm, disable integral term, by proportion K p and derivative K d Drive leg fast extension and retraction (crank connecting rod driven by motor, extension amount ΔL i assigned by Jacobian matrix); when e(t) ≤ 5mm, adopt integral term to eliminate steady-state error, synchronous adjustment of gimbal pitch / roll angle, ensure weld seam centered in field of view;

[0169] Step 5: Weld defect recognition and judgment

[0170] Visual feature extraction: extract edges from preprocessed weld images using improved Canny algorithm, mainly including the following steps:

[0171] Step 1: adaptive Gaussian filter preprocessing, dynamically adjust filter kernel standard deviation according to local region gray variance, formula:

[0172]

[0173] Where σ0 is the reference standard deviation, α is the adjustment coefficient, Var(I) is the local variance, and noise reduction while preserving edge details is achieved;

[0174] Step 2: multi-scale gradient calculation, use different scales (such as 3x3, 5x5, 7x7) Sobel operator to calculate gradient, gradient amplitude

[0175] direction Fusion of multi-scale edge map by weight Weight w i Determined by the proportion of edge intensity of each scale;

[0176] Step 3: direction-sensitive non-maximum suppression, discretize gradient direction into 8 intervals, compare gradient amplitude of neighborhood pixels along gradient direction for each pixel, retain local maximum, suppress non-edge pixels;

[0177] Step 4: dynamic double-threshold segmentation, use Otsu algorithm to calculate global threshold T global , dynamically adjust high and low thresholds T H = β·T global , T L = γ·T global (γ = 0.4β, β ∈ [1.2, 1.5]), higher than T H is strong edge, between T L and T H and connected with strong edge is weak edge;

[0178] Step 5: Region growing edge optimization, with strong edges as seeds, by searching for connecting weak edges, removing isolated noise points, and enhancing edge continuity.

[0179] After the above steps, if there is a continuous interruption or protrusion in the edge, it is marked as "suspected crack / un-fusion";

[0180] Infrared feature analysis: calculate the temperature gradient of the weld area If the local temperature is higher than the normal area by more than 5 degrees, it is marked as "suspected porosity / burn-through";

[0181] Feature fusion: compare the suspected defect positions marked by vision and infrared, combine the defect criterion library (preset 3 types and 8 defect features), output the final defect type and coordinates.

[0182] Step 6: Closed-loop feedback and detection completion, every 100mm of weld detected, the control module adjusts the trolley moving speed according to the defect recognition results (defect area speed reduced to 5mm / s, normal area speed maintained at 10mm / s); after detection is completed, output the detection report containing defect position, type, size, and generate a U-rib weld three-dimensional defect distribution map; the trolley automatically returns to the starting point, the legs reset (extension amount 0mm), and the sensor enters sleep mode.

[0183] Specific implementation method six: Figure 6 The control flowchart of the U-rib weld defect detection centroid self-balancing system of the present application; after the system is powered on, the sensor array and actuator complete initialization and self-checking first, the sensor array high-frequency parallelly collects multi-source data such as leg force, inclination and weld image, and encapsulates and transmits them to the electronic control unit (ECU) according to the time stamp (since this is a schematic diagram, the ECU components are not drawn in the three-dimensional diagram); the ECU fuses data to calculate the real-time centroid, compares it with the theoretical target centroid to get the offset, and then generates leg extension and pan-tilt rotation instructions through PID algorithm, servo motors drive leg extension and feedback actual amount to form a closed loop, the leg extension speed can reach 5-10mm / s, the response time is ≤200ms, through spring-damping components and fuzzy PID closed-loop control, stable regulation without overshoot is realized within ±80mm extension stroke, ensuring the dynamic balance accuracy during U-rib curved surface detection; after the pan-tilt receives the instructions and rotates, the angle is verified; the ECU synchronously processes image recognition of weld defects, responds in stages when abnormal, resets the legs and pan-tilt after the task is completed, and the control terminal exports the report, realizing the whole process of "perception-decision-execution-feedback", dynamic detection accuracy and equipment stability, and adapting to complex curved surface detection.

Claims

1. The application discloses a centroid self-balancing chassis system for U-rib weld defect detection and a control method, and aims to solve the problem of detection precision reduction caused by centroid deviation and field of view inclination of detection equipment in U-rib curved surface movement. The device comprises: a detection trolley body driven by a Mecanum wheel to realize omnidirectional movement, the size and structure design of which are adapted to a U-rib curved surface environment, and can avoid obstacles such as weld seam periphery protrusions; the chassis is equipped with four groups of electric telescopic legs (with built-in spring-damping components, pressure sensors and inclination sensors), a two-axis holder carrying an industrial camera and an infrared thermal imager, and an electronic control unit integrating multiple sensors. The method comprises: The motion position of the trolley body is acquired through a vehicle-mounted sensor, and the base position of the trolley body is determined in combination with the installation datum of the centroid adaptive adjustment mechanism; Based on the kinematic relationship between the legs and the two-axis holder, a position conversion model of the base and the end of the detection equipment is established; In an absolute coordinate system taking the U-rib working space as a datum, the U-rib edge is extracted through an improved Canny algorithm and the datum center is fitted, the real-time centroid is calculated in combination with the support force collected by the pressure sensor and the inclination sensor data, and then the improved adaptive PID algorithm is used to drive the legs to extend and retract and the holder to rotate, so that the centroid returns to the datum center and the detection field is perpendicular to the weld seam. Through the cooperation of the mechanical structure and the control algorithm, the centroid offset is ≤5 mm and the field inclination is ≤1°, and the stability and precision of the U-rib weld defect detection are significantly improved, and the device is suitable for the automatic nondestructive detection scene of the U-rib of a bridge steel box girder.

2. A centroid self-balancing chassis system for U-rib weld defect detection, characterized by, It comprises: The Mecanum wheel omnidirectional movement unit realizes the longitudinal, transverse and rotational movement of the chassis on the U-rib curved surface; Four sets of electric telescopic legs: each set of legs is equipped with a servo motor, an encoder and a pressure sensor for real-time acquisition of support force F i ; The two-axis holder carries an industrial camera and an infrared thermal imager, and adjusts the detection field angle through the azimuth axis and the elevation axis; The electronic control unit performs the following operations: (1) Based on the improved Canny algorithm processing industrial camera collected U rib image, extract edge point set {(x j ,y j )} and through the least square method fitting U rib upper wall reference center coordinates (x p ,y p ); (2) According to the pressure sensor data F i and the tilt sensor data θ p , θ r , and the rigid body centroid mechanics solution, the attitude transformation of the gimbal, and the inertial force calculation of the Mecanum wheel, the real-time centroid coordinates (x′ c ,y′ c ) are calculated, and the formula is: (x′ c ,y′ c ) = K - (x c ,y c ) + (1 - K) - (x c - ΔX,y c - ΔY) (3) Calculate the leg extension amount AL based on the adaptive PID algorithm i , the formula is: where the error e(t) = [ΔX, ΔY] T = [x p -x′ c ,y p -y′ c ] T , the parameter K p , K i , K d is adjusted dynamically according to the system response. (4) cooperatively controls the legs to extend and retract and the holder to rotate, so that the centroid offset |ΔX|, |ΔY| is ≤5 mm and the field inclination is ≤1°.

3. The method of claim 2, wherein, The improved Canny algorithm comprises: (1) adaptive Gaussian filtering: dynamically adjusts the standard deviation σ of the Gaussian kernel according to the local variance of the image, and the formula is: Wherein, σ0 is the reference standard deviation, α is the adjustment coefficient, and Var(I) is the local image variance; (2) multi-scale edge detection: generate multiple sets of edge maps E1, E2, E3 through Gaussian kernels of different scales, and weighted fusion: Wherein, σ0 is the reference standard deviation, α is the adjustment coefficient, and Var(I) is the local image variance. (3) Adaptive dual threshold: global threshold T is calculated based on Otsu algorithm. global And dynamically adjust the high and low thresholds according to the local gradient distribution: T H =β·T global ,T L =γ·T global (γ = 0.4β, β ∈ [1.2, 1.5]) 4. The system of claim 2, wherein, The adaptive PID algorithm comprises: (1) parameter self-tuning mechanism: Initialize parameters based on the Ziegler-Nichols method: K p = 0.6K cr , K d = 0.075K cr T cr where K cr is the critical gain, T cr is the critical oscillation period, the parameters are adjusted in real time by fuzzy control rules, the rule base including: ① ② (2) integral separation strategy: When |e(t)|>β, the integral term is cancelled to avoid integral saturation: (3) anti-interference filtering: Kalman filtering is performed on the error signal e(t), and the formula is: where K f is a filter gain.

5. The system of claim 2, wherein, The electronic control unit (ECU) further performs the following steps: (1) Pre-compensation algorithm: Calculate the mass center offset ΔX inertia , ΔY inertia caused by the inertial force according to the motion state (speed v, acceleration a) of the Mecanum wheel, the formula is where F inertia = ma, h is the mass center height, and w is the chassis height. (2) leg cooperative control: decomposes the overall centroid adjustment into coordinated actions of each leg, and maps through the Jacobian matrix: Wherein, J is the kinematic Jacobian matrix of the leg.

6. The system of claim 2: wherein, The control method of the two-axis holder comprises: (1) On-site verticality compensation: according to the inclination sensor data θ p , θ r , calculate the pan-tilt rotation angle α, β, the formula is: Wherein, h is the height of the holder. (2) motion smoothing processing: S-curve acceleration and deceleration planning is performed on the holder rotation command to avoid impact vibration.

7. A method of detecting U-rib weld defects based on the system of claim 2, characterized by, It comprises: (1) initialization step: the chassis enters the U-rib detection area, the legs are reset to the middle position, and the sensor is self-checked; (2) Data acquisition step: Collecting pressure data F i , inclination data θ p , θ r and U rib image; (3) positioning and calculation step: 1) improved Canny algorithm to extract U-rib edge point set, fitting the reference center (x p ,y p ) 2) calculate real-time centroid (x' c ,y' c ) and offset ΔX, ΔY; (4) Adjustment step: 1) Adaptive PID algorithm calculates the leg extension amount AL i and the pan-tilt rotation angle; 2) Drive the legs and gimbal to compensate for the offset of the center of mass and the tilt of the field of view; (5) Detection step: trigger the industrial camera and infrared thermal imager to collect weld data; (6) Loop step: repeat steps (2)-(5) until the full area detection is completed.

8. The method of claim 7, wherein, The adjustment step further comprises: (1) Subsection adjustment strategy: Segmented control strategy according to filtered centroid deviation Segmented control, (threshold), disable integral term, only use proportional and derivative fast regulation; when enable the integral term to eliminate steady-state error; while combining system response time feedback dynamic correction of PID parameters, balance fast response and steady-state accuracy. (2) Dynamic response optimization: according to the system response time T r Adjust the PID parameters, the formula is: where T target is the target response time.

Images