A curved surface adaptive polishing method, device, equipment and readable storage medium
By combining a robotic polishing system with multimodal sensors to adjust polishing parameters in real time, the problems of poor adaptability of traditional curved surface polishing systems and inconsistency of manual polishing are solved, achieving efficient and stable curved surface polishing results.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- LINK TOUCH(BEIJING)TECH CO LTD
- Filing Date
- 2026-07-07
- Publication Date
- 2026-08-04
AI Technical Summary
Traditional surface grinding systems struggle to adapt to real-time curvature changes and shape deviations of complex surfaces, leading to grinding quality defects. Furthermore, manual grinding is labor-intensive and yields inconsistent results.
A robotic polishing system is adopted, which combines a 3D vision sensor, a 6D force sensor and an acoustic emission sensor to acquire visual, force and auditory features in real time. These features are fused through a cross-modal attention mechanism to dynamically adjust the target value of the normal force, the feed rate and the polishing trajectory to achieve adaptive polishing.
It achieves adaptability and consistency in robotic surface grinding, improves grinding quality, reduces labor intensity, and reduces quality defects.
Smart Images
Figure CN122508763A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of intelligent processing technology, and in particular to a method, apparatus, equipment and readable storage medium for adaptive surface grinding. Background Technology
[0002] Surface grinding is a key process in the manufacturing of complex curved surface components such as aerospace blades, wind turbine blades, automobile bodies, and ship propellers, which directly affects the surface quality, aerodynamic performance, and fatigue life of the components.
[0003] For a long time, surface grinding has mainly relied on manual labor. However, workers need to hold grinding tools for a long time to repeatedly work on complex surfaces, which is extremely physically demanding. Moreover, the grinding results of different workers or the same worker at different times vary greatly due to the influence of workers' experience, skill level and fatigue. This makes it difficult to meet the strict requirements of high-end equipment such as aerospace for surface roughness, contour accuracy and consistency.
[0004] Although some robotic grinding systems have emerged in the existing technology, traditional robotic grinding mostly adopts teaching programming or offline programming methods, which makes it difficult to adapt to the curvature changes of large and complex curved surfaces and the shape and position deviations of workpieces in real time. This can easily lead to quality defects such as misalignment, penetration, overheating, or chatter. Therefore, how to achieve adaptive grinding of curved surfaces has become a research direction. Summary of the Invention
[0005] In view of this, this application provides a method, apparatus, device and readable storage medium for adaptive grinding of curved surfaces, so as to realize adaptive grinding of curved surfaces.
[0006] To achieve the above objectives, the following solution is proposed: An adaptive surface polishing method includes: The workpiece to be polished is scanned to obtain a sparse point cloud of the workpiece and an initial curve model is constructed. Based on the initial curve model, an initial grinding trajectory is planned in the surface tangent space. The initial grinding trajectory includes the position information of the path points and the tool axis vector. Based on the initial grinding trajectory and the preset initial grinding parameters, the robot is controlled to grind the workpiece to be ground. The initial grinding parameters include the target value of the normal force and the feed rate. Visual, force, and auditory features during the polishing process are acquired in real time, and fused features are obtained based on these features. Based on the fusion features, the target value of the normal force, the feed rate and the grinding trajectory are adjusted in real time, and the process is returned to execute the steps of the robot to grind the workpiece until the grinding is completed.
[0007] Optionally, the process of obtaining fused features based on the visual, force, and auditory features includes: A cross-modal attention mechanism is adopted, using visual features as queries and force and auditory features as keys and values. Dynamic feature fusion between heterogeneous modalities is achieved through attention weights to obtain fused features.
[0008] Optionally, the method for acquiring the visual features includes: Acquire sparse point cloud data in real time from a 3D vision sensor; Feature extraction is performed on the sparse point cloud data, and the surface geometric feature vector is output as a visual feature.
[0009] Optionally, the method for acquiring the auditory features includes: Acquire high-frequency information about the contact between the grinding wheel and the workpiece to be ground, collected by an acoustic emission sensor; The high-frequency information is processed to obtain a time-spectrum diagram; The time-spectrum image is input into a pre-trained grinding state recognition model to obtain acoustic features. The grinding state recognition model is configured to process the time-spectrum image to obtain acoustic features and determine the grinding state based on the acoustic features.
[0010] Optionally, the method for obtaining the force sensory features includes: Acquire force data from a six-dimensional force sensor; The force data is subjected to wavelet transform to obtain force data with high-frequency noise removed; Based on the force data with high-frequency noise removed and the time spectrum diagram, the temporal joint modeling of force perception and hearing is performed to dynamically predict the temporal sequence of force perception and obtain force perception features.
[0011] Optionally, after polishing, the following may also be included: Visual acquisition is performed on the grinding area of the workpiece to be ground to obtain point cloud data of the grinding area; The residual height was calculated by comparing the point cloud data of the polished area with the CAD model. When the residual height exceeds the preset tolerance range, a replacement path is generated for the non-compliant area.
[0012] Optionally, after obtaining the fusion features, the method further includes: Based on the fusion features, the anomaly confidence level is determined; When the abnormal confidence level exceeds a preset threshold, emergency stop protection is triggered.
[0013] A curved surface adaptive grinding device, comprising: The curve model construction module is used to scan the workpiece to be polished, obtain the sparse point cloud of the workpiece to be polished, and construct the initial curve model. The initial trajectory planning module is used to plan an initial grinding trajectory in the surface tangent space based on the initial curve model. The initial grinding trajectory includes the position information of the path points and the tool axis vector. The workpiece grinding module is used to control the robot to grind the workpiece based on the initial grinding trajectory and the preset initial grinding parameters, wherein the initial grinding parameters include the target value of the normal force and the feed speed. The multimodal data fusion module is used to acquire visual features, force features, and auditory features during the polishing process in real time, and to obtain fused features based on the visual features, force features, and auditory features. The adaptive parameter adjustment module is used to adjust the target value of the normal force, the feed rate and the grinding trajectory in real time based on the fused features, and return to execute the steps of the robot to grind the workpiece until the grinding is completed.
[0014] An adaptive surface polishing device includes: a memory and a processor; The memory is used to store programs; The processor is used to execute the program to implement the various steps of the aforementioned adaptive surface polishing method.
[0015] A readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the aforementioned adaptive surface polishing method.
[0016] As can be seen from the above technical solution, the adaptive surface grinding method provided in this application includes: scanning the workpiece to be ground to obtain a sparse point cloud of the workpiece and constructing an initial curve model; planning an initial grinding trajectory in the tangential space of the surface based on the initial curve model, the initial grinding trajectory including the position information of path points and the tool axis vector; controlling the robot to grind the workpiece based on the initial grinding trajectory and pre-set initial grinding parameters, the initial grinding parameters including the target value of the normal force and the feed rate; acquiring visual features, force features, and auditory features during the grinding process in real time; obtaining fused features based on the visual features, force features, and auditory features; adjusting the target value of the normal force, the feed rate, and the grinding trajectory in real time based on the fused features; and returning to execute the step of controlling the robot to grind the workpiece until grinding is completed. This application achieves adaptive surface grinding by acquiring visual features, force features, and auditory features in real time and adjusting the target value of the normal force, the feed rate, and the grinding trajectory during the grinding process. Attached Figure Description
[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.
[0018] Figure 1 A flowchart of a surface adaptive polishing method provided in this application embodiment; Figure 2 A flowchart of a fusion feature processing method is provided for an embodiment of this application; Figure 3 This is a schematic diagram of a curved surface adaptive grinding device provided in an embodiment of this application; Figure 4 This is a hardware structure block diagram of a surface adaptive grinding device provided in an embodiment of this application. Detailed Implementation
[0019] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0020] Surface grinding is a critical process in the manufacturing of complex components such as aerospace blades, wind turbine blades, automobile bodies, and ship propellers. Traditional manual grinding suffers from problems such as high labor intensity, health hazards from dust, and poor processing consistency. Based on this, this application utilizes a robot to replace manual labor to complete the grinding work. The grinding robot of this application may include a robot body; a multimodal perception module installed at the end of the robot body, which may include a three-dimensional vision sensor, a six-dimensional force sensor, and an acoustic emission sensor; an end-effector adaptive grinding actuator for performing the grinding work, which may include an active force control flange, a high-speed electric spindle, and a quick-change grinding disc interface; and a controller for performing cross-modal attention fusion of visual, force, and auditory data and adjusting the grinding trajectory and force control parameters in real time based on the fusion results.
[0021] The robot body can be a six-axis industrial robot or an AGV + robotic arm composite mobile collaborative robot.
[0022] In the multimodal perception module, the 3D vision sensor can be a line laser profilometer or a structured light 3D camera with a sampling frequency ≥50Hz to acquire high-precision 3D point cloud data of the surface in real time. It is installed at the end of the robot body and can be fixedly offset from the center of the grinding disc by 30-50mm using an L-shaped bracket. The six-dimensional force sensor can be installed between the robot end and the grinding tool to collect contact force and torque information in real time. The acoustic emission sensor has a sampling frequency ≥1 MHz and monitors high-frequency vibration / acoustic emission signals, such as the typical frequency band 100kHz–1MHz. The installation position can be close to the grinding spindle or a key position of the workpiece. In addition, an inertial measurement unit can be installed inside the force control flange to compensate for end-effector attitude deviations during the robot's dynamic motion.
[0023] In the end-effector adaptive grinding actuator, the active force control flange can be equipped with a voice coil motor or a proportional servo valve to achieve millisecond-level active adjustment of the normal contact force; the quick-change grinding disc interface can support quick replacement of sandpaper with different grit sizes.
[0024] The robot end effector flange can be connected in sequence to: a six-dimensional force sensor, an active force control flange, a high-speed electric spindle, and a grinding disc. A three-dimensional vision sensor can be mounted on the side of the spindle using an L-shaped bracket, maintaining a fixed offset from the center of the grinding disc, such as 30-50mm. An acoustic emission sensor can be attached to the spindle housing, and an inertial measurement unit can be built into the force control flange. The aforementioned connection method is an optional connection method between the robot end effector flange and the six-dimensional force sensor, the active force control flange, and the high-speed electric spindle-grinding disc.
[0025] The controller may include a multimodal data acquisition card supporting protocols such as GenCP, EtherCAT, and IEPE, a GPU-accelerated edge computing unit, a multimodal fusion processing unit, and a robot motion controller. The multimodal fusion processing unit can achieve 3D point cloud segmentation and surface reconstruction, force-potential-sound fusion based on time-series data, real-time planning of grinding trajectories, and adaptive adjustment of impedance control parameters.
[0026] This application provides a surface adaptive polishing method, applied in the aforementioned controller, for controlling a robot to complete the polishing work. Figure 1 A flowchart of a surface adaptive polishing method provided in this application embodiment is shown. The method may include the following steps: Step S100: Scan the workpiece to be polished to obtain the sparse point cloud of the workpiece and construct an initial curve model.
[0027] Specifically, the robot drives a 3D vision sensor to quickly scan the workpiece to be polished. The scanning speed can be set to less than or equal to 200 mm / s. After scanning, sparse point cloud is obtained, and an initial surface model S0 is established.
[0028] Step S101: Based on the initial curve model, plan the initial grinding trajectory in the surface tangent space.
[0029] Specifically, based on the initial curve model, an improved A* algorithm and B-spline curves can be used to plan the grinding trajectory within the tangent space of the surface. The initial grinding trajectory includes the position information (x, y, z) of the path points and the tool axis vector (n). x ,n y ,n z This ensures that the grinding tool always conforms to the normal direction of the curved surface.
[0030] Step S102: Based on the initial grinding trajectory and the preset initial grinding parameters, control the robot to grind the workpiece to be ground.
[0031] Specifically, the initial grinding parameters include the target value of the normal force and the feed rate. During the grinding process, an impedance control strategy can be used for force-position coordinated control. Among them, M d B d K d These are the target inertia, damping, and stiffness matrices, respectively; F ext External contact force; The desired trajectory includes position coordinates (x, y, z) and attitude angles (Rx, Ry, Rz); For the desired speed; For the desired acceleration; The actual trajectory includes position coordinates (x, y, z) and attitude angles (Rx, Ry, Rz); This refers to the actual speed; This is the actual acceleration; K is adjusted in real time. d Parameters are used to achieve smooth polishing.
[0032] Step S103: Acquire visual, force, and auditory features during the polishing process in real time, and obtain fused features based on the visual, force, and auditory features.
[0033] Specifically, the visual, force, and auditory data collected during the polishing process are spatiotemporally aligned and fused with depth features.
[0034] Step S104: Based on the fusion features, adjust the target value of the normal force, the feed rate, and the grinding trajectory in real time.
[0035] Specifically, the fused features can be processed through a fully connected layer or a multilayer perceptron, and adaptively adjusted according to the anomaly type, outputting the target value of the normal force and the correction amount of the feed rate. Anomaly types can include: sudden increase in curvature, slippage, tool skipping, over-grinding, and scorching. Specifically, when visual features indicate a large change in point cloud density, force features indicate a decrease in normal force, or auditory features indicate a high-frequency whistling sound, a sudden increase in curvature can be considered an anomaly, in which case the feed rate can be reduced and stiffness increased. When visual features indicate discontinuous wear marks, force features indicate pulsating tangential force, or auditory features indicate low-frequency vibration, slippage or tool skipping can be considered an anomaly, in which case the normal force can be increased and the spindle speed reduced. When visual features indicate abnormal color and texture, force features indicate the normal force exceeds a preset threshold, or auditory features indicate the disappearance of high-frequency components, over-grinding or scorching can be considered an anomaly, in which case an emergency stop, an alarm, and the generation of a re-polishing trajectory can be executed.
[0036] Based on the adjusted target value of normal force, feed speed and grinding trajectory, the actuator attitude, force control parameters and process parameters are adjusted in real time using an impedance control strategy. The robot then returns to the step of grinding the workpiece to be ground until the grinding is completed.
[0037] The above embodiments provide a surface adaptive grinding method, which may include: scanning the workpiece to be ground to obtain a sparse point cloud of the workpiece and constructing an initial curve model; planning an initial grinding trajectory in the tangential space of the surface based on the initial curve model, the initial grinding trajectory including the position information of path points and the tool axis vector; controlling a robot to grind the workpiece based on the initial grinding trajectory and pre-set initial grinding parameters, the initial grinding parameters including the target value of the normal force and the feed rate; acquiring visual, force, and auditory features in real time during the grinding process; obtaining fused features based on the visual, force, and auditory features; adjusting the target value of the normal force, the feed rate, and the grinding trajectory in real time based on the fused features; and returning to execute the step of controlling the robot to grind the workpiece until grinding is completed. This application achieves surface adaptive grinding by acquiring visual, force, and auditory features in real time and adjusting the target value of the normal force, the feed rate, and the grinding trajectory during the grinding process.
[0038] In some embodiments of this application, step S103, obtaining fused features based on visual features, force features, and auditory features, may include: using a cross-modal attention mechanism, using visual features as queries, force features and auditory features as keys and values, and achieving dynamic feature fusion between heterogeneous modalities through attention weights to obtain fused features.
[0039] Specifically, the calculation method for the cross-modal attention mechanism is as follows: Wherein, visual features serve as the query Q; force features serve as the key K; auditory features serve as the value V; d k W represents the dimension of key K; Q To query the parameter matrix of the linear projection of Q; F v For visual features; W K W is the linear projection parameter matrix of key K; V F is the linear projection parameter matrix for the value V; f Force perception characteristics; F a For auditory features, among which, W Q W K W V It can be obtained through pre-training.
[0040] After features are extracted independently for each modality, they can interact in the feature space. For feature extraction, visual perception can use CNN / ViT, auditory perception can use ResNet / LSTM to process spectrograms, and kinetic perception can use MLP to process vectors. The fusion mechanism can use one modality as the query and other modalities as keys / values, dynamically calculating weights. For example, when visual perception is occluded, the weights for auditory and kinetic perception are automatically increased. During processing, multimodal features can be tokenized, and a self-attention mechanism can capture long-range dependencies and nonlinear associations, allowing the solution to balance accuracy and flexibility and adapt to scene changes.
[0041] The above method achieves information interaction and deep coupling between heterogeneous modalities by calculating attention weights. Vision queries force and auditory information, realizing a fusion logic of visual guidance, force, and sound assistance. During the fusion process, visual features always act as the query, responsible for providing geometric guidance, such as curvature and normal, while force and auditory assistance act as the key and value, providing contact details and state monitoring. Here, K and V are linear mappings between force and auditory features. The attention mechanism enables visual geometric information to effectively guide the extraction of force and auditory features, achieving end-to-end mapping from physical quantities to control quantities. By calculating attention weights through a multi-head attention mechanism, vision-guided fusion of force and auditory information is achieved, ultimately yielding a fused feature. This fused feature fully integrates multimodal complementary information, providing a reliable foundation for subsequent control decisions. Impedance parameters are adaptively adjusted based on the multimodal fusion results to achieve compliant fit and smoothing.
[0042] In some embodiments of this application, visual data can be acquired using a three-dimensional vision sensor mounted on the end effector of the robot body. Based on this, the methods for acquiring visual features may include: S11. Acquire sparse point cloud data collected in real time by the 3D vision sensor.
[0043] S12. Extract features from sparse point cloud data and output surface geometric feature vectors as visual features.
[0044] Specifically, visual information acquisition can be performed using an improved PointNet++ encoder to extract features from 3D point cloud data, thereby obtaining a visual geometric feature vector F that includes local curvature, normal vector, and Gaussian curvature. v : in, , The coordinates of adjacent point clouds are given; N(i) represents the local neighborhood; the final output is a visual feature vector. .
[0045] In some embodiments of this application, sound data can be collected using an acoustic emission sensor mounted on the end effector of the robot body. Based on this, the method for acquiring auditory features may include: S21. Acquire high-frequency information about the contact between the grinding wheel and the workpiece to be ground, collected by the acoustic emission sensor.
[0046] Specifically, the signal s(t) collected by the acoustic emission sensor contains high-frequency information about the contact between the grinding wheel and the workpiece to be ground, and the frequency band can be 100kHz-1MHz.
[0047] S22. Process the high-frequency information to obtain the time-frequency spectrum.
[0048] Specifically, the time-frequency spectrum can be obtained using the following formula: in, The center time position of the window function; For window functions (such as Hanning window, Gaussian window); For time; It is the imaginary unit.
[0049] S23. Input the time-spectrum graph into the pre-trained grinding state recognition model to obtain acoustic features.
[0050] Specifically, the grinding state recognition model is configured to process the time-spectrum graph to obtain acoustic features F. a Among them, the extracted acoustic features Based on acoustic features, the system can determine the polishing state, which can include normal, slippage, flutter, and overheating. The polishing state recognition model can employ a CNN model, such as ResNet-18.
[0051] In some embodiments of this application, a six-dimensional force sensor mounted on the end effector of the robot body can be used to collect force data. Based on this, the method of obtaining force perception features may include: S31. Acquire force data collected by the six-dimensional force sensor.
[0052] Specifically, the data collected by the six-dimensional force sensor includes normal force Fn, tangential force Ft, torque M, etc.
[0053] S32. Perform wavelet transform on the force data to obtain force data with high-frequency noise removed.
[0054] S33. Based on force data and time spectrum diagrams with high-frequency noise removed, the temporal joint modeling of force perception and hearing is performed to dynamically predict the temporal sequence of force perception and obtain force perception features.
[0055] Specifically, due to the low sampling frequency of 3D vision sensors and the need for point cloud preprocessing after acquisition, a significant computational delay is introduced. In contrast, 6D force sensors can provide near real-time acquisition of low-dimensional force signals, including normal force, tangential force, and torque components. The sampling is dense and the processing delay is small. Therefore, directly fusing the delayed visual features with the current force features will lead to spatiotemporal misalignment, affecting the fusion accuracy and control stability, especially in complex curved surface regions with drastic curvature changes.
[0056] To avoid the aforementioned problems, an LSTM encoder can be used to dynamically predict the force characteristics of the current or next moment based on historical force signal sequences and combined with temporal features of the spectrogram. This compensates for the sampling and processing delays of the 3D vision sensor, achieving spatiotemporal alignment of multimodal data. The predicted F... f As a more reliable force feature, visual features dominate geometric queries, and LSTM predicts the force feature F. f Provides real-time contact details, significantly improving the quality of fused features.
[0057] Meanwhile, when dynamically predicting force-sensory time sequences, the LSTM encoder can further integrate spectrogram features extracted from the auditory branch to achieve joint force-sound time-series modeling, thereby improving the prediction accuracy of grinding states and the compensation effect for visual sampling delay, improving the prediction accuracy of grinding states such as flutter and slippage, and better compensating for visual delay.
[0058] refer to Figure 2 As shown, Figure 2This application provides a flowchart of a fusion feature processing method. Figure 2 Point cloud data acquired by a 3D vision sensor is used as a query input to the fusion layer after feature extraction. Force data acquired by a 6D force sensor is combined with the time-spectrum graph converted from sound data acquired by an acoustic emission sensor. After passing through an LSTM encoder, force features that are concurrent with the visual features are predicted and used as key inputs to the fusion layer. The time-spectrum graph converted from sound data acquired by the acoustic emission sensor is used as a value input to the fusion layer. Through processing by the fusion layer, fused features are obtained and input to a multilayer perceptron. Finally, the target value of normal force and the correction value of feed rate are obtained. The target value of normal force, feed rate and grinding trajectory are adjusted in real time. At the same time, the multilayer perceptron can also output the anomaly confidence level.
[0059] In some embodiments of this application, after grinding is completed, the grinding effect can be checked and evaluated. Based on this, the adaptive grinding method may further include: visually acquiring the grinding area of the workpiece to be ground, obtaining point cloud data of the grinding area, comparing the point cloud data of the grinding area with the CAD model to calculate the residual height, and generating a polishing path for the unqualified area when the residual height exceeds the preset tolerance range.
[0060] Specifically, after polishing is completed, switch to visual inspection mode, collect point cloud data and compare it with CAD reference model, calculate the deviation of residual height or material removal amount, and set the preset tolerance range to deviation > 10%. When the residual height exceeds the preset tolerance range, generate a polishing trajectory for the unqualified area and perform secondary polishing.
[0061] In some embodiments of this application, after obtaining the fusion features, the fusion features obtained by fusing visual features, force features and acoustic features can be used to further determine the abnormal situation of the current polishing and determine whether polishing needs to be stopped. Based on this, the method can also determine the abnormality confidence based on the fusion features. When the abnormality confidence exceeds a preset threshold, emergency stop protection is triggered.
[0062] Specifically, the possibility of a problem can be judged based on the fusion features. Different weights can be set for different data. For example, abnormal surface texture, normal force exceeding the threshold, and disappearance of high-frequency components can be judged as abnormal conditions such as over-grinding and burning. The weights can be increased. Finally, based on the current actual data, the abnormal confidence value is calculated. When it exceeds the preset threshold, emergency stop protection can be triggered and an alarm message can be issued.
[0063] The adaptive surface grinding apparatus provided in the embodiments of this application is described below. The adaptive surface grinding apparatus described below can be referred to in correspondence with the adaptive surface grinding method described above.
[0064] Figure 3This application provides a schematic diagram of a surface adaptive grinding device, which may include: The curve model construction module 10 is used to scan the workpiece to be polished, obtain the sparse point cloud of the workpiece to be polished, and construct the initial curve model. The initial trajectory planning module 20 is used to plan the initial grinding trajectory in the surface tangent space based on the initial curve model. The initial grinding trajectory includes the position information of the path points and the tool axis vector. The workpiece grinding module 30 is used to control the robot to grind the workpiece based on the initial grinding trajectory and the preset initial grinding parameters, including the target value of the normal force and the feed speed. The multimodal data fusion module 40 is used to acquire visual features, force features and auditory features in real time during the polishing process, and obtain fused features based on the visual features, force features and auditory features; The adaptive parameter adjustment module 50 is used to adjust the target value of the normal force, the feed rate and the grinding trajectory in real time based on the fusion features, and return to execute the steps of the robot to grind the workpiece until the grinding is completed.
[0065] The above embodiment provides a surface adaptive grinding device, which may include: a curve model construction module 10, used to scan the workpiece to be ground, obtain the sparse point cloud of the workpiece to be ground, and construct an initial curve model; an initial trajectory planning module 20, used to plan an initial grinding trajectory in the surface tangent space based on the initial curve model, the initial grinding trajectory including the position information of path points and the tool axis vector; a workpiece grinding module 30, used to control the robot to grind the workpiece to be ground based on the initial grinding trajectory and preset initial grinding parameters, the initial grinding parameters including the target value of normal force and the feed rate; a multimodal data fusion module 40, used to acquire visual features, force features and auditory features during the grinding process in real time, and obtain fused features based on the visual features, force features and auditory features; and an adaptive parameter adjustment module 50, used to adjust the target value of normal force, feed rate and grinding trajectory in real time based on the fused features, and return to execute the steps of controlling the robot to grind the workpiece to be ground until the grinding is completed. This application achieves adaptive grinding of curved surfaces by acquiring visual, force, and auditory features in real time and adjusting the target value of the normal force, feed speed, and grinding trajectory during the grinding process.
[0066] Optionally, the multimodal data fusion module 40 performs a process of obtaining fused features based on visual features, force features, and auditory features, which may include: A cross-modal attention mechanism is adopted, using visual features as queries and force and auditory features as keys and values. Dynamic feature fusion between heterogeneous modalities is achieved through attention weights to obtain fused features.
[0067] Optionally, the multimodal data fusion module 40 may perform the process of acquiring visual features, including: Acquire sparse point cloud data in real time from a 3D vision sensor; Feature extraction is performed on sparse point cloud data, and surface geometric feature vectors are output as visual features.
[0068] Optionally, the multimodal data fusion module 40 may perform the process of acquiring auditory features, including: Acquire high-frequency information about the contact between the grinding wheel and the workpiece to be ground, collected by an acoustic emission sensor; The high-frequency information is processed to obtain the time-spectrum diagram; The time-spectrum image is input into the pre-trained grinding state recognition model to obtain acoustic features. The grinding state recognition model is configured to process the time-spectrum image to obtain acoustic features and determine the grinding state based on the acoustic features.
[0069] Optionally, the multimodal data fusion module 40 may perform the process of acquiring force sensory features, including: Acquire force data from a six-dimensional force sensor; Wavelet transform is performed on the force data to obtain force data with high-frequency noise removed; Based on force data with high-frequency noise removed and time-spectrum graphs, a joint temporal model of force perception and hearing is performed to dynamically predict the temporal sequence of force perception and obtain force perception features.
[0070] Optionally, the surface adaptive grinding device may also include: The evaluation and acquisition module is used to perform visual acquisition of the grinding area of the workpiece to be ground, and to obtain point cloud data of the grinding area. The comparison module is used to calculate the residual height by comparing the point cloud data of the polished area with the CAD model; The supplementary throwing module is used to generate supplementary throwing paths for non-conforming areas when the residual height exceeds the preset tolerance range.
[0071] Optionally, the surface adaptive grinding device may also include: The anomaly confidence determination module is used to determine the anomaly confidence based on fused features; The anomaly detection module is used to trigger emergency stop protection when the anomaly confidence level exceeds a preset threshold.
[0072] This application also provides a surface adaptive grinding device. Figure 4 The hardware structure block diagram of the surface adaptive grinding equipment is shown. (Refer to...) Figure 4The hardware structure of the surface adaptive grinding equipment may include: at least one processor 1, at least one communication interface 2, at least one memory 3 and at least one communication bus 4; In this embodiment of the application, the number of processor 1, communication interface 2, memory 3, and communication bus 4 is at least one, and processor 1, communication interface 2, and memory 3 communicate with each other through communication bus 4; Processor 1 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention. Memory 3 may include high-speed RAM, and may also include non-volatile memory, such as at least one disk storage device; The memory stores a program, which the processor can call. The program is used to implement the various processing steps in the aforementioned adaptive surface polishing method.
[0073] This application embodiment also provides a storage medium that can store a program suitable for execution by a processor, the program being used to implement various processing flows in the aforementioned adaptive surface polishing method.
[0074] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.
[0075] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The various embodiments can be combined with each other, and the same or similar parts can be referred to each other.
[0076] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A surface adaptive grinding method, characterized in that, include: The workpiece to be polished is scanned to obtain a sparse point cloud of the workpiece and an initial curve model is constructed. Based on the initial curve model, an initial grinding trajectory is planned in the surface tangent space. The initial grinding trajectory includes the position information of the path points and the tool axis vector. Based on the initial grinding trajectory and the preset initial grinding parameters, the robot is controlled to grind the workpiece to be ground. The initial grinding parameters include the target value of the normal force and the feed rate. Visual, force, and auditory features during the polishing process are acquired in real time, and fused features are obtained based on these features. Based on the fusion features, the target value of the normal force, the feed rate and the grinding trajectory are adjusted in real time, and the process is returned to execute the steps of the robot to grind the workpiece until the grinding is completed.
2. The method according to claim 1, characterized in that, The fused features obtained based on the visual, force, and auditory features include: A cross-modal attention mechanism is adopted, using visual features as queries and force and auditory features as keys and values. Dynamic feature fusion between heterogeneous modalities is achieved through attention weights to obtain fused features.
3. The method according to claim 1, characterized in that, The methods for acquiring the visual features include: Acquire sparse point cloud data in real time from a 3D vision sensor; Feature extraction is performed on the sparse point cloud data, and the surface geometric feature vector is output as a visual feature.
4. The method according to claim 1, characterized in that, The methods for acquiring the auditory features include: Acquire high-frequency information about the contact between the grinding wheel and the workpiece to be ground, collected by an acoustic emission sensor; The high-frequency information is processed to obtain a time-spectrum diagram; The time-spectrum image is input into a pre-trained grinding state recognition model to obtain acoustic features. The grinding state recognition model is configured to process the time-spectrum image to obtain acoustic features and determine the grinding state based on the acoustic features.
5. The method according to claim 4, characterized in that, The methods for acquiring the force sensory characteristics include: Acquire force data from a six-dimensional force sensor; The force data is subjected to wavelet transform to obtain force data with high-frequency noise removed; Based on the force data with high-frequency noise removed and the time spectrum diagram, the temporal joint modeling of force perception and hearing is performed to dynamically predict the temporal sequence of force perception and obtain force perception features.
6. The method according to any one of claims 1-5, characterized in that, After polishing is complete, it also includes: Visual acquisition is performed on the grinding area of the workpiece to be ground to obtain point cloud data of the grinding area; The residual height was calculated by comparing the point cloud data of the polished area with the CAD model. When the residual height exceeds the preset tolerance range, a replacement path is generated for the non-compliant area.
7. The method according to any one of claims 1-5, characterized in that, After obtaining the fusion features, the process also includes: Based on the fusion features, the anomaly confidence level is determined; When the abnormal confidence level exceeds a preset threshold, emergency stop protection is triggered.
8. A curved surface adaptive grinding device, characterized in that, include: The curve model construction module is used to scan the workpiece to be polished, obtain the sparse point cloud of the workpiece to be polished, and construct the initial curve model. The initial trajectory planning module is used to plan an initial grinding trajectory in the surface tangent space based on the initial curve model. The initial grinding trajectory includes the position information of the path points and the tool axis vector. The workpiece grinding module is used to control the robot to grind the workpiece based on the initial grinding trajectory and the preset initial grinding parameters, wherein the initial grinding parameters include the target value of the normal force and the feed speed. The multimodal data fusion module is used to acquire visual features, force features, and auditory features during the polishing process in real time, and to obtain fused features based on the visual features, force features, and auditory features. The adaptive parameter adjustment module is used to adjust the target value of the normal force, the feed rate and the grinding trajectory in real time based on the fused features, and return to execute the steps of the robot to grind the workpiece until the grinding is completed.
9. A surface adaptive grinding device, characterized in that, include: Memory and processor; The memory is used to store programs; The processor is used to execute the program to implement the various steps of the surface adaptive polishing method as described in any one of claims 1-7.
10. A readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements each step of the surface adaptive polishing method as described in any one of claims 1-7.