Adaptive control method for end-effector attitude of spraying robot oriented to blade gap

Abstract

This invention relates to the field of spraying control technology, specifically to an adaptive attitude control method for the end effector of a spraying robot targeting blade slits. The method includes extracting the cross-sectional profile of the slit using a laser scanner, identifying the dynamic geometric centroid after curvature weighting correction from the intersection of the normals at the narrowest points of the two side walls, and fitting it as a fifth-order non-uniform rational B-spline spatial guide line with third-order continuity; determining a virtual motion plane based on the guide line, and locking the redundant degrees of freedom of the robot's end effector in non-spraying directions using a zero-space mapping algorithm of the Jacobian matrix, thus confining attitude adjustment within the plane; calculating the local curvature of the sensing and shielding side using a virtual mirror compensation model, and combining it with real-time dynamic mapping of the feed speed and the pressure of the proportional air valve. This invention achieves smooth following and locking of the nozzle attitude in complex curves, solves the pain point of sensing and shielding, and ensures the consistency of coating thickness and operational stability within variable cross-section slits.

Description

Technical Field

[0001] This invention relates to the field of spraying control technology, specifically to an adaptive control method for the end-effector posture of a spraying robot oriented towards blade gaps. Background Technology

[0002] In the manufacturing process of aero-engine blades, industrial pump impellers or large turbine machinery, the gaps between blades (especially narrow and deep gaps) often need to be coated with functional coatings to enhance their oxidation resistance, corrosion resistance or wear resistance. Such blade gaps are usually characterized by large spatial curvature, narrow and deep opening, and non-linear geometric shape. Currently, spraying for blade gaps mainly relies on industrial robots with preset trajectories or manual spraying. However, in the field of fluid coating, the nozzle's attitude, spraying distance, and relative feed speed are key factors that determine the uniformity of coating thickness. Existing control methods have a significant technical problem when dealing with gaps: due to the sensor's perception lag and the disconnect from kinematic constraints, it is difficult for the nozzle to spray evenly and accurately in real time in gaps with complex curvature. When the direction or opening of the slit changes abruptly, traditional robot control algorithms cannot detect the curvature change in advance and adjust the end effector posture accordingly. This causes the nozzle to overshoot or deviate in direction at bends, resulting in coating buildup or missed areas in the slit. Furthermore, the sensor occlusion effect caused by narrow and deep slits makes it difficult for the robot to obtain complete slit morphology data, further deteriorating the accuracy of nozzle posture adjustment and severely affecting the quality consistency of spraying on the inner wall of complex blade slits.

[0003] To address this, an adaptive control method for the end-effector posture of a spraying robot targeting blade gaps is proposed. Summary of the Invention

[0004] The purpose of this invention is to provide an adaptive control method for the end-effector posture of a spraying robot oriented towards blade gaps, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: An adaptive control method for the end-effector posture of a spraying robot oriented towards blade gaps, including...

[0006] A laser scanner is used to detect the blade slit wall in real time and identify the width center point of the blade slit. The continuous width center points are fitted as a spatial guide line, and a virtual motion plane is determined based on the spatial guide line. The controller of the robot's end effector corrects the end effector path in real time according to the deviation between the width center point and the current position of the nozzle. At the same time, the redundant degrees of freedom of the robot end effector in the non-spraying direction are locked, and the attitude adjustment is limited to the virtual motion plane. A virtual detection vector is set to extend outward along the nozzle axis. The virtual detection vector is used to perform advanced detection along the extension direction of the spatial guide line to sense the curvature change of the blade gap. When the virtual detection vector deviates from the tangential direction of the spatial guide line, the end effector is adjusted to deflect in the bending direction of the spatial guide line. The physical gap between the slit wall and the nozzle is monitored in real time by a laser scanner, and the real-time opening of the slit at the center point of the width is calculated. The end feed speed is adjusted according to the real-time opening so that the nozzle axis always points to the center point of the width in the low-speed feed state.

[0007] Preferably, the width center point is the dynamic geometric centroid obtained by calculating the intersection of the normals at the narrowest points of the two side walls and weighting it in conjunction with the local curvature of the wall profile. The spatial guide line is a three-dimensional smooth trajectory with curvature continuity generated by fitting a fifth-order non-uniform rational B-spline curve with the dynamic geometric centroid as the sample point; during the fitting process, jump suppression is performed based on the rate of change of the position of the dynamic geometric centroid at adjacent time points.

[0008] Preferably, the real-time correction of the end path specifically involves obtaining the current coordinates of the nozzle end in the virtual motion plane in real time, and calculating the lateral offset vector between it and the nearest reference point on the spatial guide line; mapping the lateral offset vector to the Jacobian transpose matrix of the robot joint space, and calculating the compensation motion of each joint; and driving the nozzle to move closer to the spatial guide line by superimposing the compensation motion on the main running trajectory.

[0009] Preferably, the specific process of limiting the posture adjustment within the virtual motion plane determined by the spatial guide line is as follows: extracting the normal vector and tangent vector of the spatial guide line at the current sampling point in real time, and defining the spatial pose of the virtual motion plane by the normal vector and tangent vector; locking the rotational degree of freedom of the robot end about the spraying center axis and the translational degree of freedom perpendicular to the virtual motion plane by the zero-space mapping algorithm of the Jacobian matrix; While keeping the nozzle axis always within the virtual motion plane, the range of motion of each joint of the robotic arm is dynamically updated according to the direction of the spatial guide line, so that the end effector posture can only be deflected within the virtual motion plane.

[0010] Preferably, the virtual detection vector is a geometric vector established in the robot's kinematic model and projected outward from the nozzle center point along the current axis; the advance detection process includes real-time calculation of the orthogonal projection point of the end of the virtual detection vector on the spatial guide line, and extraction of the local curvature at the projection point; when the virtual detection vector deviates from the tangent direction at the projection point, the controller, according to the magnitude of the angular deviation, drives the end effector of each joint of the robotic arm to rotate around the normal axis of the virtual motion plane, so that the nozzle axis approaches the curvature direction of the spatial guide line in advance before physically reaching the projection point.

[0011] Preferably, the extraction of local curvature at the projection point is used to establish a virtual mirror compensation model based on the symmetry of the gap wall; the visibility of the two walls on both sides of the gap is identified in real time using the laser scanner; when one wall is obscured due to depth limitations, the controller extracts the local geometric features of the visible wall on the other side; the local geometric features are input into the virtual mirror compensation model to generate a symmetrical virtual feature point cloud at the orthogonal projection point, and the virtual feature point cloud is input into the sliding window filter for smoothing; the virtual feature point cloud and the synchronously acquired real feature point cloud are uniformly transformed to the robot base coordinate system using the forward kinematics matrix; the combined point set formed by the transformed virtual feature point cloud and the real feature point cloud is dynamically weighted and fitted to calculate the predicted local curvature that incorporates the constraint trends on both sides of the gap, guiding the end effector to perform attitude deflection compensation.

[0012] Preferably, the real-time detection of the blade slot wall using a laser scanner specifically includes: inputting the raw point cloud data acquired by the laser scanner into a preset sliding window filter, calculating the Euclidean distance deviation between adjacent point cloud coordinates within the window; setting a coordinate change rate threshold, and when the instantaneous change rate of the point cloud coordinates is greater than the threshold, marking it as an abnormal interference point and performing mean-based replacement processing; reading the feedback values ​​of the encoders of each joint of the robot in real time, and transforming the filtered point cloud local coordinate system to the robot base coordinate system through the forward kinematics matrix; A time axis alignment operator is used to perform linear interpolation on the transformed point cloud coordinates based on the ratio of the robot controller's interpolation period to the scanner's sampling period, generating a sequence of gap feature points synchronized with the robot's motion trajectory timestamp.

[0013] Preferably, the step of adjusting the end feed speed according to the real-time opening is to obtain the spray fan parameters of the nozzle in the current posture in real time, and extract the projection area of ​​the spray center axis on the slit wall; input the area of ​​the projection area, the real-time opening and the preset paint flow rate into the preset mapping table of the controller, and match the corresponding end feed speed command. While executing the feed speed command, the controller adjusts the output pressure of the pneumatic proportional valve in the same direction according to the change in the feed speed, and adjusts the opening step of the pneumatic proportional valve; when the virtual detection vector detects that the turning angle of the gap direction exceeds the preset limit, the amplitude of the end feed speed command is reduced, and the detection span of the virtual detection vector is shortened simultaneously.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. By using a laser scanner to extract the dynamic geometric centroid and fitting it into a five-order non-uniform rational B-spline space guide line, this application effectively solves the problem that traditional preset trajectories cannot adapt to the geometric deformation of blade gaps. Due to the adoption of a real-time compensation algorithm based on the Jacobian transpose matrix, the robot can instantly correct the end path according to the deviation of the width center point, ensuring that the nozzle always locks the optimal spraying position of the gap during dynamic movement. This not only improves the physical accuracy of the spraying trajectory, but also fundamentally avoids the phenomenon of uneven local coating thickness or missed spraying caused by path deviation.

[0015] 2. This application introduces a virtual detection vector projected along the nozzle axis to simulate the predictive logic of feedforward control. When the virtual detection vector senses that the curvature of the spatial guide line is about to change, the controller can drive the end effector to deflect the attitude before the nozzle actually reaches the position. This advanced detection mechanism successfully overcomes the attitude lag problem caused by system inertia and signal delay in the robot arm in complex continuous curves, so that the spraying axis can smoothly and continuously follow the gap direction, which greatly improves the attitude following stability under complex curvature conditions.

[0016] 3. To address the technical challenge of perception occlusion caused by single-sided walls in narrow and deep slits, this application establishes a virtual mirror compensation model based on the symmetry of the slit wall, enabling intelligent reconstruction of missing feature point clouds. Even when the scanner's field of view is limited, the system can still use the geometric features of the visible side to infer the global constraint trend and generate a predicted curvature that integrates both sides of the constraint. This mechanism significantly enhances the robot's survivability and control continuity in extremely unstructured environments, ensuring that even in deep slits with incomplete visual information, a high level of coating process consistency can still be maintained. Attached Figure Description

[0017] Figure 1 A flowchart illustrating the adaptive control method for the end-effector posture of a spraying robot oriented towards blade gaps; Figure 2 This is a schematic diagram of the gap dynamic geometric centroid identification process of the present invention; Figure 3 This is a schematic diagram of the virtual motion plane attitude locking and zero-space mapping process of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Example 1: Please see Figure 1 This invention provides an adaptive control method for the end-effector posture of a spraying robot oriented towards blade gaps, the technical solution of which is as follows: An adaptive control method for the end-effector posture of a spraying robot targeting blade gaps includes: A laser scanner is used to detect the blade slit wall in real time and identify the width center point of the blade slit. The continuous width center points are fitted as a spatial guide line, and a virtual motion plane is determined based on the spatial guide line. The controller of the robot's end effector corrects the end effector path in real time according to the deviation between the width center point and the current position of the nozzle. At the same time, the redundant degrees of freedom of the robot end effector in the non-spraying direction are locked, and the attitude adjustment is limited to the virtual motion plane. A virtual detection vector is set to extend outward along the nozzle axis. The virtual detection vector is used to perform advanced detection along the extension direction of the spatial guide line to sense the curvature change of the blade gap. When the virtual detection vector deviates from the tangential direction of the spatial guide line, the end effector is adjusted to deflect in the bending direction of the spatial guide line. The physical gap between the slit wall and the nozzle is monitored in real time by a laser scanner, and the real-time opening of the slit at the center point of the width is calculated. The end feed speed is adjusted according to the real-time opening so that the nozzle axis always points to the center point of the width in the low-speed feed state. The method of using a laser scanner to detect the blade slot wall in real time involves mounting a laser scanner on the end effector of a spraying robot, with its detection beam perpendicularly pointed at the area to be sprayed in the blade slot. As the robot moves the nozzle along a preset trajectory, the laser scanner acquires the original cross-sectional contour data of the two side walls of the blade slot in real time through high-speed sampling. This data contains the spatial coordinate information of the complex curved surfaces on both sides of the slot, providing basic geometric features for subsequent center point identification. The width center point is the dynamic geometric centroid obtained by calculating the intersection of the normals at the narrowest points of the two side walls and then weighting it based on the local curvature of the wall profile. The spatial guide line is a three-dimensional smooth trajectory with curvature continuity generated by fitting a fifth-order non-uniform rational B-spline curve with the dynamic geometric centroid as the sample point; during the fitting process, jump suppression is performed based on the rate of change of the position of the dynamic geometric centroid at adjacent time points.

[0020] Specifically, by analyzing the original contour data of the cross-section in real time, the feature regions closest to each other on both sides of the wall are identified. An algorithm is used to calculate the normals at the narrowest points of these two walls, and the intersection of these two normals is used as a basic reference point. The algorithm is a traversal search algorithm. The controller uses this algorithm to calculate the Euclidean distance between all pairs of points on the left and right walls, identifying the pair of corresponding points with the smallest distance values, which are defined as the narrowest points on the left and right walls, respectively. Centered on the narrowest point on the left, several adjacent sampling points are selected, and local linear fitting is performed using the least squares method to construct the local tangent vector at that point. Within the current cross-section, the vector perpendicular to this tangent vector is calculated based on geometric relationships; this is the normal of the left wall. Similarly, the same method is used to process the narrowest point on the right to reconstruct the normal of the right wall. If the normals of the two walls do not intersect absolutely in space, the midpoint of the common perpendicular segment of the two normals is taken as the basic reference point. To eliminate positioning deviations caused by surface roughness or minor protrusions during blade casting, local curvature correction is introduced. The controller calculates the radius of curvature of the wall profile near the base reference point and presets a dynamic weight allocation logic. When the curvature of a certain side wall changes smoothly, a higher confidence weight is assigned to the geometric features of that side. When a sudden change in curvature is detected (such as the presence of minor pits or weld points), the weight of that side is reduced. The preferred range is set to a standard curvature threshold range of 50mm to 150mm according to the blade design specifications. If the measured radius of curvature of the left wall is within the above threshold range and the fluctuation is less than 10%, it is determined to be a smooth area and assigned a weight coefficient of 0.7. If the radius of curvature of the right wall suddenly drops below 10mm (determined to be a pit or weld point), it is determined to be a sudden change area and its weight coefficient is reduced to 0.3. By weighting the coordinates of the base reference point with the curvature distribution of the two side walls, a precise coordinate that is both located at the physical center of the gap and can filter surface noise is finally output, namely the dynamic geometric centroid. In this way, it is ensured that each sample point can truly reflect the structural center of the gap; After obtaining a series of dynamically generated geometric centroids that are continuously generated over time, these are used as original sample points. A fifth-order non-uniform rational B-spline curve is used for trajectory fitting. The process involves determining the node vector based on the Euclidean distance between the sample points to ensure that the distribution of the curve in the parameter space is consistent with the physical space. The coordinates of the control points are determined by solving a system of linear equations, so that the generated curve satisfies positional continuity at each centroid sampling point. Due to the use of fifth-order polynomial basis functions, the algorithm automatically ensures the mathematical continuity of the first derivative (velocity vector), second derivative (acceleration vector), and third derivative (jerk vector) of the curve at the connection point during the solution process. This ensures the smooth switching of angular velocity and angular acceleration of the robot end effector when passing through the gap and turning point at the physical level. During the fitting process, the controller monitors the rate of change of the position of the dynamic geometric centroid at adjacent time points in real time to suppress jumps. In this embodiment, the preset physical threshold is set as follows: within a 10ms sampling period, the spatial displacement increment of the dynamic geometric centroid must not exceed 2mm, and the velocity jump threshold is set to 200mm / s. If the currently acquired centroid point causes the displacement increment to reach 3mm or the instantaneous velocity exceeds the above threshold, it is immediately identified as a pseudo-signal jump point and an interception mechanism is executed to automatically remove the point. An interpolation compensation operation is performed: the three most recent valid historical sample points before the jump point occurs are retrieved, and the end tangent vector of the current trajectory is calculated using the first-order forward difference. Along the tangent direction, combined with the feed velocity at the previous moment and the 10ms sampling period, a virtual compensation point coordinate is predicted, and the compensation point is temporarily used to replace the abnormal jump point as a temporary sample point to participate in the fitting of the fifth-order spline curve, and finally a smooth spatial guide line with continuity of position, curvature and rate of change in three-dimensional space is generated.

[0021] By combining traversal search for intersection of normals with curvature weighted correction, the robustness of center positioning in complex blade gaps is significantly improved, effectively filtering out geometric deviations caused by casting defects. In conjunction with quintic spline curve fitting and advanced tangent compensation mechanism, the third-order continuity of the trajectory in position, curvature, and rate of change is ensured from a mathematical perspective, eliminating end-effector jitter caused by sensor jumps and ensuring the smooth operation of the robot and the consistency of coating thickness in extremely narrow and deep working conditions.

[0022] See Figure 2 Based on the spatial guide line, the virtual motion plane is determined as follows: during the robot's spraying operation, the controller extracts the normal vector and tangent vector of the spatial guide line at the current sampling point in real time; where the tangent vector represents the instantaneous tangent direction of the gap, while the normal vector is perpendicular to the guide line and points to the opening direction of the gap. These two mutually orthogonal vectors together define the pose of the virtual motion plane in three-dimensional space. The controller at the robot's end effector corrects the end effector path in real time based on the deviation between the width center point and the current position of the nozzle. The specific process of real-time end effector path correction is as follows: the current coordinates of the nozzle end effector in the virtual motion plane are obtained in real time, and the lateral offset vector between it and the nearest reference point on the spatial guide line is calculated; the lateral offset vector is mapped to the Jacobian transpose matrix in the robot joint space, and the compensation motion of each joint is calculated; by superimposing the compensation motion on the main running trajectory, the nozzle is driven to move closer to the spatial guide line.

[0023] As the robot moves the nozzle along the spatial guide line, the controller acquires the current spatial coordinates of the nozzle tip in the virtual motion plane in real time every 10ms as a control cycle. Simultaneously, it uses a search algorithm to find the point on the generated spatial guide line that is closest to the current nozzle position in terms of Euclidean distance, and defines it as the nearest reference point. The controller calculates the spatial difference between the current coordinates of the nozzle and the nearest reference point, extracts the spatial position deviation between the two in the virtual motion plane, and defines it as the lateral offset vector. A Jacobian transpose matrix in joint space is introduced for real-time calculation. The controller obtains the encoder feedback values ​​of each joint of the robot, constructs a real-time Jacobian matrix in combination with the current pose, and transposes it. This transpose matrix is ​​used as a mathematical bridge for the mapping of torque and displacement, and the above-mentioned lateral offset vector is mapped to each rotational joint of the robot. This mapping mechanism avoids complex matrix inversion operations and can project the offset of the spatial coordinate system onto each joint degree of freedom with extremely high computational efficiency, thereby calculating the amount of compensation motion required by each joint to eliminate the lateral offset. After obtaining the compensated motion quantities of each joint, the controller uses trajectory superposition technology to achieve dynamic correction. That is, the system linearly superimposes the preset robot main running trajectory command with the calculated compensated motion quantities of each joint at the servo level. The superimposed composite motion command is sent to the servo drivers of each joint, driving the motors to perform minute attitude corrections. This ensures that while maintaining high-speed feed motion, the nozzle tip can continuously and in real time approach the spatial guide line, ensuring that the nozzle axis is always precisely locked on the ideal trajectory formed by the dynamic geometric centroid in the complex curvature gap. By introducing the Jacobian transpose matrix to replace the traditional complex matrix inversion operation, the computing efficiency of the controller is greatly improved, enabling the robot to complete the closed-loop correction of path deviation within an extremely short period of 10ms, ensuring that the nozzle can respond instantaneously to the dynamic changes in the curvature of the gap. By constructing a virtual motion plane defined by the normal vector and tangent vector in three-dimensional space and performing trajectory superposition, lateral swaying in the non-spraying direction is eliminated from the underlying physical logic. This not only effectively avoids the risk of the nozzle colliding with the wall in narrow and deep gaps, but also ensures that the nozzle axis is always precisely locked on the ideal trajectory formed by the dynamic geometric centroid.

[0024] See Figure 3 The specific process of confining the posture adjustment within the virtual motion plane determined by the spatial guide line is as follows: extracting the normal vector and tangent vector of the spatial guide line at the current sampling point in real time, and defining the spatial pose of the virtual motion plane by the normal vector and tangent vector; locking the rotational degree of freedom of the robot end about the spraying center axis and the translational degree of freedom perpendicular to the virtual motion plane by the zero-space mapping algorithm of the Jacobian matrix. While keeping the nozzle axis always within the virtual motion plane, the range of motion of each joint of the robotic arm is dynamically updated according to the direction of the spatial guide line, so that the end effector posture can only be deflected within the virtual motion plane.

[0025] To strictly limit the nozzle's attitude adjustment within the virtual motion plane, the controller mathematically locks the robot's redundant degrees of freedom using a Jacobian matrix null-space mapping algorithm. This algorithm involves the controller first acquiring the encoder feedback values ​​for each joint of the robot and constructing a global Jacobian matrix reflecting the mapping relationship between the end effector's spatial velocity and joint velocities. Then, using generalized inverse operations or singular value decomposition, the null-space basis vectors of this Jacobian matrix are obtained. These basis vectors define the allowed redundant motion directions within the robot's joint space without changing the nozzle's current position. Finally, the rotational degrees of freedom of the robot's end effector around the spraying center axis and the translational degrees of freedom perpendicular to the virtual motion plane are mapped to these null-space basis vectors, generating compensatory joint corrections. When performing path tracking motion, the joint correction amount is superimposed on the main motion command, so that when each joint of the robot drives the nozzle to move, it can automatically cancel out any motion components that attempt to deviate from the virtual motion plane or generate unnecessary rotation, thereby ensuring that the motion increment in the above redundant directions is always zero. Under the physical constraint of keeping the nozzle axis always within the virtual motion plane, the controller dynamically updates the motion range of each joint of the robotic arm according to the direction of the spatial guide line. By calculating the pose changes of the virtual motion plane in real time, the controller calculates the limit positions of each joint that satisfy the planar constraints and inputs them as safety boundaries into the motion control layer of the controller. This follow-up constraint mechanism ensures that when the nozzle tip travels along the slit, its attitude deflection can only be performed within a two-dimensional plane defined by the normal vector and the tangent vector, eliminating the risk of unnecessary shaking or collision with the sidewall of the nozzle in narrow and deep slits from a physical level. Through the above process, by using the spatial pose definition of the virtual motion plane and the zero-space mapping algorithm, the dimensionality reduction control of the robot's end effector was successfully achieved. This not only ensures that the nozzle axis is always at the optimal spraying angle in the complex curvature gap, but also greatly improves the motion stability of the end effector during dynamic attitude adjustment by locking redundant degrees of freedom.

[0026] A virtual detection vector is defined, extending outward along the nozzle axis. This virtual detection vector is a geometric vector established in the robot's kinematic model and projected outward from the nozzle center point along the current axis. The advance detection process includes real-time calculation of the orthogonal projection point of the end of the virtual detection vector on the spatial guide line, and extraction of the local curvature at the projection point. When the virtual detection vector deviates from the tangent direction at the projection point, the controller, based on the magnitude of the angular deviation, drives the end effector of each joint of the robotic arm to rotate around the normal axis of the virtual motion plane, so that the nozzle axis approaches the curvature direction of the spatial guide line in advance before physically reaching the projection point.

[0027] Specifically, the virtual detection vector is a geometric vector established in the robot's kinematic model, projected outward from the nozzle center point along the current axis. In practice, the controller presets a detection step size value within the virtual motion plane based on the robot's current feed speed; in this embodiment, it is set to 20mm. Since the end-effector redundant degrees of freedom are locked, using forward kinematic calculation, a virtual geometric line segment is extended from the nozzle end-effector center point, always pointing along the nozzle axis within the virtual motion plane. This vector serves as the antenna for attitude prediction, used to perceive the bending trend of the gap in front within the plane in real time. The advanced detection process includes real-time calculation of the orthogonal projection point of the end of the virtual detection vector on the spatial guide line; in each control cycle, the controller uses a traversal search algorithm to find the point with the smallest distance from the end of the detection vector on the spatial guide line as the orthogonal projection point, and accurately locks the target path position that the nozzle is about to reach through the orthogonal projection point. Extract local geometric information at the projection point, including the degree of local curvature at that location and the corresponding ideal tangent vector; the controller compares the angular deviation between the direction of the current virtual detection vector and the tangent direction at the projection point in real time. When a deviation occurs, it is determined that the direction of the gap is about to change, and the specific direction of its curvature toward the guide line is determined. Based on the angular deviation, the controller coordinates the joints of the robotic arm to drive the end effector to rotate around the normal axis of the virtual motion plane for compensation. This deflection action ensures that the nozzle axis moves towards the curved side of the spatial guide line before physically reaching the position. This predictive control of "deflecting first and then arriving" offsets the physical lag caused by mechanical transmission and communication, ensuring that the nozzle always accurately points to the center of the gap during continuous turning.

[0028] The extraction of local curvature at the projection point is used to establish a virtual mirror compensation model based on the symmetry of the gap wall. The laser scanner is used to identify the visibility of the two walls on both sides of the gap in real time. When one wall is obscured due to depth limitations, the controller extracts the local geometric features of the visible wall on the other side. The local geometric features are input into the virtual mirror compensation model to generate a symmetrical virtual feature point cloud at the orthogonal projection point. The virtual feature point cloud is then input into the sliding window filter for smoothing. The forward kinematics matrix is ​​used to transform the virtual feature point cloud and the synchronously acquired real feature point cloud to the robot's base coordinate system. The combined point set formed by the transformed virtual feature point cloud and the real feature point cloud is dynamically weighted and fitted to calculate the predicted local curvature that incorporates the constraint trends on both sides of the gap, guiding the end effector to perform attitude deflection compensation.

[0029] The process of establishing a virtual mirror compensation model based on the symmetry of the slot wall involves the controller pre-setting a set of geometric coordinate mapping operators about the central symmetry plane of the slot according to the geometric design parameters of the blade slot. These geometric coordinate mapping operators define the mirror transformation logic for mapping spatial point coordinates from one side of the slot to the other. A laser scanner, fixed to the end of the nozzle, synchronously emits a laser beam into the slot as the robotic arm feeds. By receiving the reflected signals from the wall, it acquires the spatial coordinates of the slot wall in the local coordinate system in real time. The controller analyzes the received point cloud distribution in real time, identifies the point cloud density by calculating the number of effective echo points per unit area, and synchronously reads the point cloud coordinates within the scanner's range to identify depth information, thereby determining the visibility of both side walls. When the nozzle enters a narrow, deep gap, limiting the sensor's field of view, and causing perception occlusion on one side of the wall due to depth limitations or obstruction (i.e., the number of echo points on that side is lower than a preset density threshold), the controller immediately locks onto and extracts the local geometric features of the other side of the wall that is fully visible, including the normal vector of the visible side wall and the set of spatial position points of that side wall in the current coordinate system. The controller then calls the aforementioned geometric coordinate mapping operator to perform coordinate permutation on the position point set of the visible side according to the mirror rule, thereby generating a symmetrical virtual feature point cloud at the occluded side position corresponding to the orthogonal projection point. To eliminate potential numerical jitter during mirror simulation, the generated virtual feature point cloud is input into a preset sliding window filter for smoothing. In this embodiment, a sliding window with a length of 5 sampling periods is used to calculate a weighted average of the point cloud coordinates, thereby obtaining smooth and continuous simulated features on the occluded side. Using a forward kinematics matrix, the smoothed virtual feature point cloud and the real side feature point cloud synchronously acquired by the sensor are uniformly transformed to the robot's base coordinate system. This process ensures the alignment of virtual and real features in a unified spatial pose, forming a combined point set that fuses both sides of the feature. The process of dynamic weighted fitting and predictive local curvature calculation involves the controller dynamically weighting the combined point set and allocating weights in real time based on the confidence level of the visible side data. In specific implementation, the real feature point cloud is assigned a weight of 0.6, and the virtual mirror feature point cloud is assigned a weight of 0.4. The curvature of the weighted point set is calculated using the least squares method to obtain the predicted local curvature that incorporates the constraint trends on both sides of the gap. The predicted local curvature accurately predicts the actual bending direction of the gap in front under shading conditions and is output to the controller to guide the end effector to perform rotational compensation around the normal axis of the virtual motion plane, ensuring that the nozzle can still achieve accurate attitude deflection correction even when perception is limited. By deeply integrating the virtual mirror compensation model with the dynamic weighted fitting algorithm, the system utilizes the geometric symmetry of the gap to intelligently complete the missing wall information, thus solving the path tracking distortion problem caused by unilateral perception occlusion. Combined with sliding window filtering and bilateral feature weight allocation, the system effectively suppresses data noise and ensures the continuous smoothness of the predicted curvature.

[0030] The real-time detection of the blade slot wall using a laser scanner specifically includes: inputting the raw point cloud data acquired by the laser scanner into a preset sliding window filter, calculating the Euclidean distance deviation between adjacent point cloud coordinates within the window; setting a coordinate change rate threshold, and marking point cloud coordinates as abnormal interference points and performing mean-based replacement processing when the instantaneous change rate of point cloud coordinates exceeds the threshold; and reading the feedback values ​​of the encoders at each joint of the robot in real time, and transforming the filtered point cloud local coordinate system to the robot's base coordinate system using a forward kinematics matrix. Using a time axis alignment operator, linear interpolation is performed on the transformed point cloud coordinates based on the ratio of the robot controller's interpolation period to the scanner's sampling period, generating a sequence of gap feature points synchronized with the robot's motion trajectory timestamp; A laser scanner scans the slit wall at a fixed frequency, and the acquired raw point cloud data is input into a preset sliding window filter in real time. In this embodiment, a sliding window with a length of 5 sampling periods is set. The controller calculates the Euclidean distance deviation between adjacent point cloud coordinates within a sliding window containing a fixed number of sample points. At this time, a coordinate change rate threshold is set to identify random noise or artifacts caused by metal reflection in the data. In this embodiment, the coordinate change rate threshold is set to 5% to 8%. When the instantaneous change rate of the point cloud coordinates is detected to be greater than the threshold, it is marked as an abnormal interference point, and mean substitution processing is performed. That is, the mean of the spatial coordinates of the remaining normal points within the window is used to fill the abnormal position, thereby ensuring that the point cloud data input to subsequent stages has good spatial continuity. The feedback values ​​from the encoders of each joint of the robot are read in real time. These values ​​represent the precise rotation angle or displacement of each joint at the current moment. By constructing a forward kinematics matrix, the local coordinate system of the filtered point cloud is transformed to the robot's base coordinate system. This transformation process converts the relative position information measured by the scanner into the absolute coordinate information of the robot in the global workspace, so that the geometric contour of the gap wall can be aligned and calculated with the posture of the robot's end effector in the same spatial dimension. To eliminate signal asynchrony issues caused by the inconsistency between sensor sampling frequency and robot control frequency, a time axis alignment operator is used for synchronization. The controller determines the time mapping relationship between the robot's interpolation period and the scanner's sampling period. For point cloud data falling between two adjacent interpolation points on the robot, the controller executes a linear interpolation algorithm, generating a sequence of gap feature points that is strictly synchronized with the robot's motion trajectory timestamps by weighted calculation of the coordinates of adjacent time points. This precise alignment in the time dimension ensures that when the robot's end effector performs adaptive actions, the gap feature points it relies on are precisely the physical geometric features corresponding to the current position of the robotic arm, eliminating control errors caused by data lag or lead. Through the combined processing of point cloud filtering, spatial coordinate transformation and time axis alignment, basic data is provided for subsequent identification of the width center point and fitting of the spatial guide line. This multi-dimensional preprocessing mechanism not only effectively suppresses the jitter of point cloud signals under harsh spraying conditions, but also solves the problem of spatial and temporal coupling matching of multi-sensor systems, significantly improving the dynamic reliability of robot end-effector posture adaptive control.

[0031] The physical gap between the slit wall and the nozzle is monitored in real time using a laser scanner, and the real-time opening of the slit at the center point of the width is calculated. The end feed speed is adjusted according to the real-time opening so that the nozzle axis always points to the center point of the width in the low-speed feed state. The adjustment of the end feed speed according to the real-time opening involves obtaining the spray fan parameters of the nozzle in the current posture in real time and extracting the projection area of ​​the spray center axis on the slit wall. The area of ​​the projection area, the real-time opening, and the preset paint flow rate are input into the preset mapping table of the controller to match the corresponding end feed speed command. While executing the feed speed command, the controller adjusts the output pressure of the pneumatic proportional valve in the same direction according to the change in the feed speed, and adjusts the opening step of the pneumatic proportional valve; when the virtual detection vector detects that the turning angle of the gap direction exceeds the preset limit, the amplitude of the end feed speed command is reduced, and the detection span of the virtual detection vector is shortened simultaneously. The specific implementation process of adjusting the end feed speed based on the real-time opening is as follows: the controller acquires the spray fan parameters of the nozzle in its current position in real time. These fan parameters include the nozzle's spray angle and the outer diameter of the atomized flow field. Based on the nozzle's current spatial coordinates and direction, the system extracts the projection area of ​​the spray center axis onto the slit wall. Since the slit wall has a certain curvature and depth, this projection area presents as an irregular geometric surface in three-dimensional space. The controller calculates the area of ​​the projection area in real time, combining this with the real-time slit opening at the center point of the width, as monitored in real time by a laser scanner, and the preset paint flow rate. The controller inputs three variables—the calculated projected area, the real-time gap opening, and the paint flow rate—into a preset mapping table. This mapping table, established based on offline process test data, stores the optimal speed matching relationship under different gap widths and spray coverage conditions. The corresponding end-feed speed command is obtained through table lookup. When the gap opening narrows or the projected area decreases, the controller automatically increases the feed speed command value to prevent excessive paint buildup in localized areas; conversely, when the gap opening widens, the feed speed is correspondingly reduced, thereby ensuring that the amount of paint deposited per unit area in the variable cross-section gap remains constant. While executing the feed speed command, the controller adjusts the output pressure of the air pressure proportional valve in the same direction according to the change in feed speed, calculates the difference between the current feed speed and the feed speed at the previous moment in real time, and adjusts the opening step of the air pressure proportional valve according to the difference. In this embodiment, when the feed speed command increases by 10%, the controller links the air pressure proportional valve to increase the opening by a preset number of steps to increase the atomizing air pressure or fan pressure, ensuring that the shape of the sprayed fan becomes more stable as the speed increases, and avoiding uneven atomization caused by speed fluctuations. When the virtual detection vector detects a turning angle in the gap direction exceeding a preset limit (greater than 30 degrees in this embodiment), it determines that the current location is a sharp turn or a complex curvature change region. At this time, the controller immediately performs a speed reduction intervention, decreasing the amplitude of the end-effector feed speed command to allow sufficient servo response time for the attitude deflection of each joint of the robotic arm. Simultaneously with the speed reduction, the controller also shortens the detection span of the virtual detection vector, i.e., by reducing the forward step size, increasing the sampling frequency for local high curvature regions, making the attitude compensation action more delicate and stable, effectively avoiding deflection overshoot caused by excessively fast feed or an excessively large detection range at sharp turns; Through the above-mentioned process of calculating the projection area, matching the speed with the mapping table, and adjusting the air pressure proportional valve, multi-parameter coupled control of speed, pressure, and flow rate is achieved. This adaptive adjustment mechanism not only solves the problem of uniformity in spraying of variable cross-section gaps, but also solves the problem of motion stability at complex bends by dynamically correlating the detection vector with the feed speed, ensuring that the nozzle always accurately points to the center point of the gap width when in low-speed feed mode.

[0032] This invention significantly improves the operational accuracy of a painting robot in confined spaces by constructing an adaptive control architecture based on spatial guide lines and a virtual motion plane. Utilizing a forward detection mechanism achieved through virtual probe vectors, it possesses the feedforward capability to predict the curvature trend of the slit, fundamentally solving the motion lag problem of the robotic arm at complex curves. Simultaneously, by calculating the slit opening in real time and dynamically compensating for the end feed speed, decoupled matching of spray flow rate and travel speed is achieved, ensuring a high degree of uniformity in coating thickness within variable cross-section slits. This multi-parameter coupled adaptive adjustment mechanism not only effectively avoids the risk of physical collision between the nozzle and narrow, deep walls but also ensures that the nozzle axis is precisely locked at the center of the slit width throughout the entire dynamic process, greatly improving the automation level and process consistency of blade slit spraying.

[0033] Example 2: This embodiment applies the end-effector attitude adaptive control method of the spraying robot for blade slots to the spraying of thermal barrier coatings in narrow and deep slots of aero-engine turbine blades. In the thermal barrier coating spraying operation of a certain type of aero-engine turbine blades, the gaps formed between the blades reach a depth of 80 mm and have complex nonlinear spatial distortion curvature. The laser scanner first performs high-speed sampling on the inner wall of the gap to obtain the original point cloud containing metal reflection interference. The controller uses a sliding window filter to mark abnormal jump points with an instantaneous coordinate change rate exceeding 8% as metal reflection artifacts and removes them. Then, the point cloud is uniformly transformed to the robot's base coordinate system through the forward kinematics matrix to generate a sequence of gap feature points aligned with the robot's motion command time axis. The controller identifies the intersection of the normals at the narrowest points of the two side walls using a traversal search algorithm, and performs weighted correction based on the wall contour curvature to calculate the dynamic geometric centroid. For the surface roughness region caused by blade casting, the weight of that side is automatically reduced (to 0.3 in this embodiment) to filter geometric deviations. Subsequently, a series of centroid points are fitted into a third-order continuous spatial guide line using a fifth-order non-uniform rational B-spline curve. By extracting the normal and tangent vectors of this line at the current sampling point, a virtual motion plane constraining the nozzle motion is defined. To ensure that the nozzle does not collide within the narrow and deep slit, the controller uses a zero-space mapping algorithm of the Jacobian matrix to lock the nozzle's rotation around the axis and its translational degrees of freedom perpendicular to the virtual motion plane. When the robot's end effector deviates from the guide line due to mechanical vibration, its lateral offset vector in the plane is calculated in real time, and the joint compensation amount is calculated using the Jacobian transpose matrix. The nozzle is then driven to approach the guide line through trajectory superposition technology, ensuring that the spraying axis is always locked on the physical center path of the slit. When encountering a turning section where the blade twisting increases dramatically, the controller activates an advanced detection mechanism. The nozzle extends a virtual detection vector 20mm outward along the axial direction, which can be adjusted according to the minimum radius of curvature of the gap and the maximum angular acceleration of the robotic arm, and calculates its orthogonal projection point on the guide line in real time. When a deviation is detected between the vector direction and the tangent direction of the projection point, the robot pre-drives the end effector to pre-deflect around the normal axis of the virtual motion plane. This strategy of deflecting first and then arriving effectively offsets the signal response lag within the 10ms sampling period, allowing the nozzle to smoothly follow the gap direction.

[0034] When the nozzle penetrates deep into the bottom of the gap, causing the sensor signal on one side to be blocked, the controller activates the virtual mirror compensation model. The system extracts the normal vector and position coordinates of the visible side wall and uses a geometric coordinate mapping operator to generate a symmetrical virtual feature point cloud on the blocked side. By dynamically weighting and fitting the real point cloud and the virtual point cloud (set to 0.6:0.4 in this embodiment), the predicted local curvature with fused bilateral constraints is calculated, guiding the robot to maintain accurate posture compensation even with missing information.

[0035] The controller calculates the projected area of ​​the spray fan on the curved wall in real time and dynamically matches the feed speed by combining the real-time opening query mapping table. When the virtual detection vector detects that the gap turning angle exceeds 30°, it automatically reduces the feed speed and shortens the detection span. At the same time, it links the air pressure proportional valve to adjust the opening step to stabilize the atomization pressure. This series of linkages ensures that the coating thickness can still maintain a high degree of uniformity under extreme gap conditions with variable cross-section and high curvature, eliminating the risk of accumulation or missed spraying.

[0036] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An adaptive control method for the end-effector posture of a spraying robot oriented towards blade gaps, characterized in that, include: A laser scanner is used to detect the blade slit wall in real time and identify the width center point of the blade slit. The continuous width center points are fitted as a spatial guide line, and a virtual motion plane is determined based on the spatial guide line. The controller of the robot's end effector corrects the end effector path in real time according to the deviation between the width center point and the current position of the nozzle. At the same time, the redundant degrees of freedom of the robot end effector in the non-spraying direction are locked, and the attitude adjustment is limited to the virtual motion plane. A virtual detection vector is set to extend outward along the nozzle axis. The virtual detection vector is used to perform advanced detection along the extension direction of the spatial guide line to sense the curvature change of the blade gap. When the virtual detection vector deviates from the tangential direction of the spatial guide line, the end effector is adjusted to deflect in the bending direction of the spatial guide line. The physical gap between the slit wall and the nozzle is monitored in real time by a laser scanner, and the real-time opening of the slit at the center point of the width is calculated. The end feed speed is adjusted according to the real-time opening so that the nozzle axis always points to the center point of the width in the low-speed feed state.

2. The adaptive control method for the end-effector posture of a spraying robot oriented towards blade gaps according to claim 1, characterized in that, The width center point is the dynamic geometric centroid obtained by calculating the intersection of the normals at the narrowest points of the two side walls and then weighting it based on the local curvature of the wall profile. The spatial guide line is a three-dimensional smooth trajectory with curvature continuity generated by fitting a fifth-order non-uniform rational B-spline curve with the dynamic geometric centroid as the sample point; during the fitting process, jump suppression is performed based on the rate of change of the position of the dynamic geometric centroid at adjacent time points.

3. The adaptive control method for the end-effector posture of a spraying robot oriented towards blade gaps according to claim 1, characterized in that, The specific process of real-time correction of the end path involves obtaining the current coordinates of the nozzle end in the virtual motion plane in real time, and calculating the lateral offset vector between it and the nearest reference point on the spatial guide line; mapping the lateral offset vector to the Jacobian transpose matrix in the robot joint space, and calculating the compensation motion of each joint; and driving the nozzle to move closer to the spatial guide line by superimposing the compensation motion on the main running trajectory.

4. The adaptive control method for the end-effector posture of a spraying robot oriented towards blade gaps according to claim 1, characterized in that, The specific process of confining the attitude adjustment within the virtual motion plane determined by the spatial guide line is as follows: extract the normal vector and tangent vector of the spatial guide line at the current sampling point in real time, and define the spatial pose of the virtual motion plane by the normal vector and tangent vector; lock the rotational degree of freedom of the robot end about the spraying center axis and the translational degree of freedom perpendicular to the virtual motion plane by the zero-space mapping algorithm of the Jacobian matrix. While keeping the nozzle axis always within the virtual motion plane, the range of motion of each joint of the robotic arm is dynamically updated according to the direction of the spatial guide line, so that the end effector posture can only be deflected within the virtual motion plane.

5. The adaptive control method for the end-effector posture of a spraying robot oriented towards blade gaps according to claim 1, characterized in that, The virtual detection vector is a geometric vector established in the robot's kinematic model and projected outward from the nozzle center point along the current axis. The advance detection process includes real-time calculation of the orthogonal projection point of the end of the virtual detection vector on the spatial guide line and extraction of the local curvature at the projection point. When the virtual detection vector deviates from the tangent direction at the projection point, the controller, based on the magnitude of the deviation, drives the end effector of each joint of the robotic arm to rotate around the normal axis of the virtual motion plane, so that the nozzle axis approaches the curvature direction of the spatial guide line in advance before physically reaching the projection point.

6. The adaptive control method for the end-effector posture of a spraying robot oriented towards blade gaps according to claim 5, characterized in that, The extraction of local curvature at the projection point is used to establish a virtual mirror compensation model based on the symmetry of the gap wall. The laser scanner is used to identify the visibility of the two walls of the gap in real time. When one wall is obscured due to depth limitations, the controller extracts the local geometric features of the visible wall on the other side. These local geometric features are input into the virtual mirror compensation model to generate a symmetrical virtual feature point cloud at the orthogonal projection point. The virtual feature point cloud is then input into a sliding window filter for smoothing. Using the forward kinematics matrix, the virtual feature point cloud and the synchronously acquired real feature point cloud are uniformly transformed to the robot's base coordinate system. A dynamic weighted fitting is performed on the combined point set formed by the transformed virtual and real feature point clouds to calculate the predicted local curvature that incorporates the constraint trends on both sides of the gap, guiding the end effector to perform attitude deflection compensation.

7. The adaptive control method for the end-effector posture of a spraying robot oriented towards blade gaps according to claim 1, characterized in that, The real-time detection of the blade slot wall using a laser scanner specifically includes: inputting the raw point cloud data acquired by the laser scanner into a sliding window filter, calculating the Euclidean distance deviation between adjacent point cloud coordinates within the window; setting a coordinate change rate threshold, and marking point cloud coordinates as abnormal interference points and performing mean-based replacement processing when the instantaneous change rate of point cloud coordinates exceeds the threshold; and reading the feedback values ​​of the encoders at each joint of the robot in real time, and transforming the filtered point cloud local coordinate system to the robot's base coordinate system using a forward kinematics matrix. A time axis alignment operator is used to perform linear interpolation on the transformed point cloud coordinates based on the ratio of the robot controller's interpolation period to the scanner's sampling period, generating a sequence of gap feature points synchronized with the robot's motion trajectory timestamp.

8. The adaptive control method for the end-effector posture of a spraying robot oriented towards blade gaps according to claim 1, characterized in that, The step of adjusting the end feed speed according to the real-time opening is to obtain the spray fan parameters of the nozzle in the current posture in real time, and extract the projection area of ​​the spray center axis on the slit wall; input the area of ​​the projection area, the real-time opening and the preset paint flow rate into the preset mapping table of the controller, and match the corresponding end feed speed command. While executing the feed speed command, the controller adjusts the output pressure of the pneumatic proportional valve in the same direction according to the change in the feed speed, and adjusts the opening step of the pneumatic proportional valve; when the virtual detection vector detects that the turning angle of the gap direction exceeds the preset limit, the amplitude of the end feed speed command is reduced, and the detection span of the virtual detection vector is shortened simultaneously.

Images