A laser cleaning system and method

Abstract

The application provides a laser cleaning system and method for controlling a laser cleaning robot to clean a workpiece surface, the system comprising: a dirt identification module, a cleaning path planning module, a control module, and a cleaning module; the dirt identification module is used to identify dirt on the workpiece surface based on a target dirt identification model to determine a target dirt position; the cleaning path planning module is used to obtain a current position of the laser cleaning robot and plan a path according to the current position and the dirt position to obtain a planned path of the laser cleaning robot; the control module is used to obtain a theoretical real-time position and a current real-time position of the laser cleaning robot based on the planned path and adjust the current real-time position of the laser cleaning robot according to the theoretical real-time position; and the cleaning module is used to clean the target dirt position when the current real-time position is the target dirt position. The application realizes low-cost and real-time efficient laser cleaning of the workpiece surface.

Description

Technical Field

[0001] This invention relates to the field of laser cleaning technology, and in particular to a laser cleaning system and method. Background Technology

[0002] With the development of modern industry, the demand for cleaning large structural components such as aircraft, ships, oil tanks, and locomotives is becoming more and more widespread. In particular, there are challenges in cleaning large curved workpieces in applications such as aircraft paint removal, ship rust removal, oil tank cleaning, and mold degreasing.

[0003] Existing cleaning methods for large curved workpieces have the following technical problems: 1. Manual cleaning of large curved workpieces, such as mechanical friction cleaning, chemical corrosion cleaning, and high-impact cleaning, is not only costly, inefficient, and environmentally unfriendly, but also yields unsatisfactory cleaning results. 2. Using three-dimensional laser scanning technology to obtain point cloud information of the workpiece surface and then planning a path for laser cleaning, however, this method requires significant costs and has a long equipment development cycle.

[0004] Therefore, there is an urgent need to propose a laser cleaning system and method to achieve low-cost, real-time, and efficient laser cleaning. Summary of the Invention

[0005] In view of this, it is necessary to provide a laser cleaning system and method to solve the problems of high cleaning cost and inefficient real-time cleaning of workpiece surfaces.

[0006] On one hand, the present invention provides a laser cleaning system for controlling a laser cleaning robot to clean the surface of a workpiece. The laser cleaning system includes: a dirt identification module, a cleaning path planning module, a control module, and a cleaning module.

[0007] The dirt recognition module is used to identify dirt on the surface of the workpiece based on the target dirt recognition model and determine the location of the target dirt.

[0008] The cleaning path planning module is used to obtain the current position of the laser cleaning robot, and to perform path planning based on the current position and the target dirt position to obtain the planned path of the laser cleaning robot.

[0009] The control module is used to obtain the theoretical real-time position and current real-time position of the laser cleaning robot based on the planned path, and adjust the current real-time position of the laser cleaning robot according to the theoretical real-time position.

[0010] The cleaning module is used so that when the current real-time location is the target dirt location, the laser cleaning robot cleans the target dirt location.

[0011] In some possible implementations, the dirt identification module includes: an initial identification model construction submodule, a dataset acquisition submodule, and a training submodule;

[0012] The initial identification model construction submodule is used to construct the initial dirt identification model;

[0013] The dataset acquisition submodule is used to acquire datasets;

[0014] The training submodule is used to train the initial dirt recognition model based on the dataset and a preset loss function to obtain the target dirt recognition model.

[0015] In some possible implementations, the loss function Loss is:

[0016]

[0017] Where, bbox loss is the bounding box loss function, confidence loss is the bounding box confidence loss function, class loss is the object classification loss function, cell number*B is the number of candidate box anchors generated for each grid cell, and I... object To determine whether a candidate box is responsible for the parameters, CIOU is the improved intersection-over-union function, m is the number of samples, and CB is the parameter. focal Here, is the focal loss function, p0 is the probability distribution of the ground truth bounding box confidence, q0 is the probability distribution of the predicted bounding box confidence, CE(p0, q0) is the cross-entropy loss function of the probability distributions of the ground truth bounding box confidence and the predicted bounding box confidence, y is the number of samples, p(c) is the probability distribution of the ground truth object classification, q(c) is the probability distribution of the predicted object classification, c is the number of classifications, and CE(p(c), q(c)) represents the cross-entropy loss function of the probability distributions of the ground truth object classification and the predicted object classification.

[0018] In some possible implementations, the cleaning path planning module includes a map acquisition submodule and a path planning submodule;

[0019] The map acquisition submodule is used to acquire a binary grid map;

[0020] The path planning submodule is used to search for a path and determine the planned path based on the location of the target dirt, the current location of the laser cleaning robot, and the binary grid map using a preset path planning algorithm.

[0021] In some possible implementations, the laser cleaning system further includes a path smoothing module, which is used to smooth the planned path and determine the planned smooth path; the path smoothing module includes a path list acquisition submodule, a line-of-sight path smoothing submodule, a path point refinement path smoothing submodule, and an interpolation path smoothing submodule;

[0022] The path list acquisition submodule is used to obtain the number of paths and all nodes of each path according to the planned path, and to determine whether the number of paths is greater than or equal to a first threshold.

[0023] The line-of-sight path smoothing submodule is used to define a first node, a second node, and a third node when the number of paths in the list is greater than or equal to a first threshold, and to adjust the first node, the second node, and the third node based on a preset adjustment rule.

[0024] The path point refinement and path smoothing submodule is used to determine the minimum turning path and all nodes of the minimum turning path when the number of paths in the path list is less than a first threshold or the third node is the endpoint, and to update all nodes of the minimum turning path according to the optimal control distance.

[0025] The interpolation path smoothing submodule is used to determine the planned smooth path based on all nodes of the minimum turning path through cubic spline interpolation.

[0026] In some possible implementations, the line-of-sight path smoothing submodule includes a first judgment module, a second judgment module, and a third judgment module;

[0027] The first judgment module is used to determine whether the third node is visible from the first node; if it is visible, it proceeds to the second judgment module; if it is not visible, it updates the first node, the second node, and the third node, and then proceeds to the first judgment module to make another judgment.

[0028] The second judgment module is used to obtain the weights of the first node, the second node, and the third node, and to determine whether the weight of the first node going through the second node to the third node is greater than the weight of the first node going directly to the third node; if so, the second node is removed and the weight of the third node is updated, and the process proceeds to the third judgment module; if not, the first node, the second node, and the third node are updated, and the process proceeds to the first judgment module for judgment again.

[0029] The third judgment module is used to determine whether the third node is the destination. If it is the destination, the minimum turning path is determined. If the third node is not the destination, the first node, the second node, and the third node are updated, and the first judgment module is entered again for judgment.

[0030] In some possible implementations, the control module includes a theoretical real-time position acquisition submodule, a current real-time position acquisition submodule, a compensation signal submodule, and a position adjustment submodule;

[0031] The theoretical real-time position acquisition submodule is used to determine the theoretical real-time position of the laser cleaning robot based on the planned path;

[0032] The current real-time position acquisition submodule is used to determine the current real-time position of the laser cleaning robot;

[0033] The compensation signal submodule is used to generate a compensation signal based on the current real-time position and the theoretical real-time position;

[0034] The position adjustment submodule is used to adjust the current real-time position of the laser cleaning robot based on the compensation signal.

[0035] In some possible implementations, the laser cleaning robot includes a laser emitter and a rotating arm for adjusting the laser emitter; the cleaning module includes a rotating arm control submodule and a laser cleaning submodule.

[0036] The rotating arm control submodule is used to control the rotating arm to adjust the laser emitter when the current real-time position is the target dirt position, so that the focus of the laser emitter is aligned with the target dirt position;

[0037] The laser cleaning submodule is used to control the laser cleaning robot to clean the target contaminant location.

[0038] In some possible implementations, the laser cleaning system further includes a detection module for determining whether the target contaminant location has been cleaned. If it has not been cleaned, the system controls the laser cleaning robot to re-clean the target contaminant location.

[0039] On the other hand, the present invention also provides a laser cleaning method, comprising:

[0040] The target contaminant identification model is used to identify contaminants on the workpiece surface and determine the location of target contaminants.

[0041] The current position of the laser cleaning robot is obtained, and path planning is performed based on the current position and the target dirt position to obtain the planned path of the laser cleaning robot;

[0042] The theoretical real-time position and current real-time position of the laser cleaning robot are obtained based on the planned path, and the current real-time position of the laser cleaning robot is adjusted according to the theoretical real-time position.

[0043] When the current real-time location is the target dirt location, the laser cleaning robot is controlled to clean the target dirt location.

[0044] The beneficial effects of the above embodiments are as follows: The laser cleaning system provided by the present invention identifies contaminants on the workpiece surface based on a target contaminant identification model using a contaminant identification module to determine the contaminant location; the cleaning path planning module obtains the current position of the laser cleaning robot and plans the path based on the current position of the laser cleaning robot and the target contaminant location to obtain the planned path of the laser cleaning robot; the control module obtains the theoretical real-time position and the current real-time position of the laser cleaning robot based on the planned path and adjusts the current real-time position according to the theoretical real-time position; when the current real-time position is the target contaminant location, the cleaning module controls the laser cleaning robot to clean the target contaminant location. This achieves automated cleaning of the workpiece without manual intervention, solving the technical problems of high cost and incomplete cleaning caused by manual cleaning in the prior art, reducing cleaning costs and improving cleaning effect. Furthermore, the present invention eliminates the need to use three-dimensional laser scanning technology to obtain point cloud information of the workpiece surface and then determine the target contaminant location from the point cloud information. The target contaminant location can be accurately identified through the target contaminant identification model, thereby improving the identification efficiency of the target contaminant location and thus improving the laser cleaning efficiency, achieving highly efficient cleaning. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 A schematic diagram of an embodiment of the laser cleaning system provided by the present invention;

[0047] Figure 2 A schematic diagram of an embodiment of the dirt identification module provided by the present invention;

[0048] Figure 3 This is a schematic diagram of an embodiment of the cleaning path planning module provided by the present invention;

[0049] Figure 4 A schematic diagram of an embodiment of the path smoothing module provided by the present invention;

[0050] Figure 5 A schematic diagram of an embodiment of the line-of-sight path smoothing submodule provided by the present invention;

[0051] Figure 6 A schematic diagram of an embodiment of the cleaning module provided by the present invention;

[0052] Figure 7 A schematic flowchart of an embodiment of the laser cleaning method provided by the present invention; Detailed Implementation

[0053] 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 a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0054] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. Furthermore, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0055] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in several embodiments of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0056] This invention provides a laser cleaning system and method, which will be described below.

[0057] Figure 1 A schematic diagram of an embodiment of the laser cleaning system provided by the present invention is shown below. Figure 1 As shown, the laser cleaning system 10 includes: a dirt recognition module 100, a cleaning path planning module 200, a control module 300, and a cleaning module 400;

[0058] The dirt recognition module 100 is used to identify dirt on the surface of the workpiece based on the target dirt recognition model and determine the location of the target dirt.

[0059] The cleaning path planning module 200 is used to obtain the current position of the laser cleaning robot and perform path planning based on the current position and the target dirt position to obtain the planned path of the laser cleaning robot.

[0060] The control module 300 is used to obtain the theoretical real-time position and current real-time position of the laser cleaning robot based on the planned path, and adjust the current real-time position of the laser cleaning robot according to the theoretical real-time position.

[0061] The cleaning module 400 is used to control the laser cleaning robot to clean the target dirt location when the current real-time location is the target dirt location.

[0062] Compared with existing technologies, the laser cleaning system 10 provided in this embodiment of the invention uses a contaminant recognition module to identify contaminants on the workpiece surface based on a target contaminant recognition model, determining the location of the target contaminant. A cleaning path planning module acquires the current position of the laser cleaning robot and plans a path based on the current position of the laser cleaning robot and the target contaminant location. A control module acquires the theoretical real-time position and the current real-time position of the laser cleaning robot based on the planned path and adjusts the current real-time position accordingly. When the current real-time position is the target contaminant location, the cleaning module controls the laser cleaning robot to clean the target contaminant location. This achieves automated cleaning of the workpiece without human intervention, solving the technical problems of high cost and incomplete cleaning caused by manual cleaning in existing technologies, reducing cleaning costs and improving cleaning effectiveness. Furthermore, it eliminates the need for three-dimensional laser scanning technology to obtain point cloud information of the workpiece surface and then determine the target contaminant location from the point cloud information; the target contaminant location can be accurately identified using the target contaminant recognition model, thereby improving the identification efficiency of the target contaminant location and thus increasing laser cleaning efficiency, achieving highly efficient cleaning.

[0063] It should be noted that: the image of the cleaning area to be identified is obtained, the image of the cleaning area to be identified is segmented into dirt regions to obtain dirt images, and the dirt images are input into the target dirt recognition model to obtain the location of the target dirt.

[0064] This invention improves the efficiency of target dirt location identification by segmenting the image to be identified, thereby reducing the computational workload of the target dirt identification model.

[0065] In some embodiments of the present invention, such as Figure 2 As shown, Figure 2 This is a schematic diagram of an embodiment of the dirt recognition model provided by the present invention. The dirt recognition module 100 includes: an initial recognition model construction submodule 110, a dataset acquisition submodule 120, and a training submodule 130.

[0066] The initial identification model construction submodule 110 is used to construct the initial dirt identification model;

[0067] The dataset acquisition submodule 120 is used to acquire datasets;

[0068] The training submodule 130 is used to train the initial dirt recognition model based on the dataset and a preset loss function to obtain the target dirt recognition model.

[0069] It should be noted that the initial dirt recognition model can use target detection models such as Faster R-CNN, SSD, and YOLO.

[0070] This invention integrates target region prediction and target category determination into a single neural network model based on the YOLOv4 target detection model. The detection results can be obtained simply by inputting the image into the neural network, which improves the speed and accuracy of identifying the location of target dirt.

[0071] In a specific embodiment of the present invention, the dataset acquisition submodule 120 can be specifically used to acquire sample images of the region to be cleaned, generate a dataset using classification labels based on the sample images of the region to be cleaned, divide the dataset into a training set and a validation set in an 8:2 ratio, and perform image data augmentation on the dataset.

[0072] The training submodule 130 can be specifically used to train the initial target dirt recognition model based on the training set and the preset loss function, adjust the hyperparameters of the model based on the validation set, and conduct a preliminary evaluation of the model's capabilities. If the training results meet the requirements, the target dirt recognition model is obtained. If the training results do not meet the requirements, the K-mean++ algorithm optimization framework is used to adjust the network parameters of the improved YOLOv4 and continue training the model until the training results meet the requirements.

[0073] In some embodiments of the present invention, the loss function Loss is:

[0074]

[0075] Where bbox loss is the bounding box loss function, confidence loss is the bounding box confidence loss function, class loss is the object classification loss function, cell number*B is the number of candidate box anchors generated for each grid cell, and I object The parameters responsible for determining whether a candidate box is responsible for the object are: m is the number of samples, CIOU is the improved intersection-over-union function, and CB... focalHere, is the focal loss function, p0 is the probability distribution of the ground truth bounding box confidence, q0 is the probability distribution of the predicted bounding box confidence, CE(p0, q0) is the cross-entropy loss function of the probability distributions of the ground truth bounding box confidence and the predicted bounding box confidence, y is the number of samples, p(c) is the probability distribution of the ground truth object classification, q(c) is the probability distribution of the predicted object classification, c is the number of classifications, and CE(p(c), q(c)) represents the cross-entropy loss function of the probability distributions of the ground truth object classification and the predicted object classification.

[0076] It should be noted that the improved intersection-union ratio function CIOU is:

[0077]

[0078] Where IOU is the intersection-union ratio function, R1 is the center of the predicted box, R2 is the center of the actual box, c is the diagonal distance between the smallest closed region of the predicted box and the ground truth box, v is a parameter used to measure the proportional consistency between the anchor box and the target box, and α is a parameter used to balance the bounding box confidence loss function.

[0079] The Intersection over Union (IOU) function is:

[0080]

[0081] Where A is the predicted bounding box and B is the ground truth bounding box;

[0082] The parameter α used for the balancing ratio of the bounding box confidence loss function is

[0083]

[0084] The parameter v used to measure the proportional consistency between the anchor box and the target box is:

[0085]

[0086] Among them, w gt h is the width of the actual bounding box. gt w is the height of the ground truth bounding box, w is the width of the predicted bounding box, and h is the height of the predicted bounding box.

[0087] focal loss function FL(P) t )for:

[0088] FL(P t )=-α t (1-P t ) γ ln(P t )

[0089] Among them, P tLet α be the label prediction probability, γ be the focusing parameter (a hyperparameter greater than 0), and α be the hyperparameter. t It is the balancing parameter of the focal loss function;

[0090] The equilibrium parameter α of the focal loss function t for:

[0091]

[0092] β is a hyperparameter that depends on the size of the assumed sample space, and n is the number of samples of a particular type.

[0093] This invention optimizes the target dirt recognition model by using three loss functions: bounding box loss function, bounding box confidence loss function, and target classification loss function, thereby reducing the loss between the true value and the predicted value and enabling the target dirt recognition model to accurately identify the location of dirt.

[0094] In some embodiments of the present invention, such as Figure 3 As shown, the cleaning path planning module 200 includes a map acquisition submodule 210 and a path planning submodule 220;

[0095] Map acquisition submodule 210 is used to acquire binary grid maps;

[0096] The path planning submodule 220 is used to search for a path and determine the planned path based on the location of the target dirt, the current position of the laser cleaning robot, and the binary grid map using a preset path planning algorithm.

[0097] This invention improves the feasibility of a laser cleaning robot by acquiring a binary grid map and searching for a path based on the A* algorithm. This allows the robot to find a collision-free, safe path from the starting point to the end point within a specified area.

[0098] In one specific embodiment of the present invention, the preset path planning algorithm is the A* algorithm. The map acquisition submodule 210 is specifically used to receive fused navigation data and a color environment map from the upper subsystem of the laser cleaning robot, extract environmental map information based on the color environment map, generate a binary grid map through image processing, and identify the position of static obstacles in the image processing step. At the same time, the position of the wall-climbing robot itself is mapped onto the generated binary grid map using the fused navigation data.

[0099] The path planning submodule 220 can be specifically used to calculate the shortest safe path using the A* algorithm based on the location of the target dirt, the current position of the laser cleaning robot, and the binary grid map. This path consists of a series of waypoints, including the task start point and the task end point, to determine the planned path.

[0100] It should be noted that the fused navigation information is obtained through multiple sensors, such as inertial measurement units, vision sensors (monocular and binocular cameras), and lidar. The information includes: the coordinate system used, environmental map information, and the velocity, attitude, position, acceleration, noise, and deviation of objects.

[0101] Because the planned path is jagged, the uneven path will cause the laser cleaning robot to frequently make large turns and vibrate, thus affecting the stability of the laser robot. Therefore, in some embodiments of the present invention, such as Figure 4 As shown, the laser cleaning system 10 also includes a path smoothing module 500, which is used to determine the planned smoothing path according to the planned path using a path smoothing method.

[0102] The embodiments of the present invention can eliminate the technical problem of instability of laser robots caused by unsmooth paths by smoothing the planned path, thereby improving the stability of laser cleaning robots walking on the planned path.

[0103] In some specific embodiments of the present invention, the path smoothing module 500 includes a path list acquisition submodule 510, a line-of-sight path smoothing submodule 520, a path point refinement path smoothing submodule 530, and an interpolation path smoothing submodule 540.

[0104] The path list acquisition submodule 510 is used to obtain the number of paths and all nodes of each path according to the planned path, and to determine whether the number of paths is greater than or equal to the first threshold.

[0105] The line-of-sight path smoothing submodule 520 is used to define the first node, the second node, and the third node when the number of paths in the list is greater than or equal to the first threshold, and to adjust the first node, the second node, and the third node based on preset adjustment rules.

[0106] The path point refinement and path smoothing submodule 530 is used to determine the minimum turning path and all nodes of the minimum turning path when the number of paths in the path list is less than the first threshold or the third node is the destination, and to update all nodes of the minimum turning path according to the optimal control distance.

[0107] The interpolation path smoothing submodule 540 is used to determine the planned smooth path based on all nodes of the minimum turning path through cubic spline interpolation.

[0108] This invention employs three path smoothing methods—line-of-sight path smoothing, path point refinement path smoothing, and interpolation path smoothing—to smooth the planned path, removing unnecessary nodes and making the laser cleaning robot's motion path more rounded rather than straight.

[0109] In a specific embodiment of the present invention, the first threshold can be 4. When the number of paths in the list is greater than or equal to 4, the line-of-sight path smoothing submodule 520 can be specifically used to check all nodes in three consecutive paths based on the nodes of the path points generated by the A* algorithm. The current check point is recorded as the first node. The node after the laser cleaning robot passes the first node is recorded as the second node. The node after the laser cleaning robot passes the second node is recorded as the third node. The first node, the second node, and the third node are updated by judging the first node, the second node, and the third node.

[0110] The path refinement and path smoothing submodule 530 can be specifically used to determine the minimum turning path and all nodes of the minimum turning path when the number of paths in the path list is less than 4 or the third node is the destination. The current checkpoint is denoted as the first node. The next node after the laser cleaning robot passes the first node is denoted as the second node; the next node after the second node is denoted as the third node. The distance from the first node to the second node is calculated. If the distance from the first node to the second node is less than the optimal control distance, the second node is removed. If the next node after the third node is the destination, the updated minimum turning path node is input into the interpolation path smoothing submodule 540. If the next node after the third node is not the destination, the third node is used as the updated second node, and the next node after the third node is used as the updated third node. The judgment is then made again based on the optimal control distance according to the updated nodes.

[0111] The cubic spline interpolation in the interpolation path smoothing submodule 540 specifically involves connecting a set of control points (or path points on the path) using splines composed of third-order polynomials. The interval between two control points is in polynomial function form, and the polynomial function is:

[0112] f k (x)=A 3k (xx k ) 3 +A 2k (xx k ) 2 +A 1k (xx k )+A 0k

[0113] Where k is the exponent of the control point, A 3k A 2k A 1k A 0k These are polynomial coefficients.

[0114] It should be noted that a node is the next possible location for the robot to explore during trajectory planning.

[0115] In some embodiments of the present invention, such as Figure 5 As shown, the line-of-sight path smoothing submodule 520 includes a first judgment module 521, a second judgment module 522, and a third judgment module 523;

[0116] The first judgment module 521 is used to determine whether the third node is visible from the first node; if it is visible, it proceeds to the second judgment module; if it is not visible, it updates the first node, the second node, and the third node, and then proceeds to the first judgment module to make another judgment.

[0117] The second judgment module 522 is used to obtain the weights of the first node, the second node, and the third node, and to determine whether the weight of the first node going through the second node to the third node is greater than the weight of the first node going directly to the third node; if so, the second node is removed and the weight of the third node is updated, and the process proceeds to the third judgment module; if not, the first node, the second node, and the third node are updated, and the process proceeds to the first judgment module for another judgment.

[0118] The third judgment module 523 is used to determine whether the third node is the destination. If it is the destination, the minimum turning path is determined. If the third node is not the destination, the first node, the second node, and the third node are updated, and the first judgment module is entered again for judgment.

[0119] In this embodiment of the invention, the line-of-sight path smoothing module checks each node sequentially, iterating from the first point to the last point, to reduce unnecessary path points and thus reduce the complexity of the planned path.

[0120] In a specific embodiment of the present invention, the first judgment module 521 may be specifically used to, when the third node is not visible from the first node, take the second node as the updated first node, take the third node as the updated second node, take the next node passing through the third node as the updated third node, and make a judgment again using the first judgment module 521 based on the updated node.

[0121] The second judgment module 522 can be specifically used to determine the updated first node and the updated second node when the weight of the first node going through the second node to the third node is less than the weight of the first node going directly to the third node. The second node is then used as the updated first node, the third node is used as the updated second node, the next node after the third node is used as the updated third node, and the first judgment module 511 is used to make a judgment again based on the updated nodes.

[0122] The third judgment module 523 can be specifically used to, when the third node is not the end point, take the third node as the updated second node, take the next node that passes through the third node as the updated third node, and then use the first judgment module 521 to make a judgment again based on the updated node.

[0123] It should be noted that "visible" refers to whether the node is displayed on the binary grid map in the path planning module 200. Some nodes are not displayed, for example, if the exploration location is on an obstacle.

[0124] In some embodiments of the present invention, such as Figure 6 As shown, the control module 300 includes a theoretical real-time position acquisition submodule 310, a current real-time position acquisition submodule 320, a compensation signal submodule 330, and a position adjustment submodule 340.

[0125] The theoretical real-time position acquisition submodule 310 is used to determine the theoretical real-time position of the laser cleaning robot based on the planned path;

[0126] The current real-time position acquisition submodule 320 is used to determine the current real-time position of the laser cleaning robot;

[0127] The compensation signal submodule 330 is used to generate a compensation signal based on the current real-time position and the theoretical real-time position;

[0128] The position adjustment submodule 340 is used to adjust the current real-time position of the laser cleaning robot based on the compensation signal.

[0129] In this embodiment of the invention, the control module 300 adjusts the current real-time position of the laser cleaning robot, enabling the laser cleaning robot to accurately reach the target dirt location according to the planned trajectory, thereby improving the accuracy of reaching the target dirt location.

[0130] In one specific embodiment, the theoretical real-time position acquisition submodule 310 can be specifically used to issue control signals according to the planned path, and determine the theoretical real-time position of the laser cleaning robot based on the response to the control signals.

[0131] The real-time position acquisition submodule 320 can be specifically used to send a monitoring signal and determine the current real-time position of the laser cleaning robot based on the response to the monitoring signal.

[0132] The overall control signal includes control signals, monitoring signals, and compensation signals. The overall control signal τ {k} for:

[0133]

[0134] U PD{k} =K p s{k}

[0135]

[0136] Among them, U PD{k} It is a PD control signal, U s{k} It is a supervisory control signal, U a{k}It is a compensation signal, τ {k} It is the overall control signal, K p It is the positive gain of the PD controller, I s It is a diagonal matrix. It is a control signal during the monitoring period.

[0137] In some embodiments of the present invention, the laser cleaning robot includes a laser emitter and a rotating arm for adjusting the laser emitter; such as Figure 6 As shown, the cleaning module 400 includes a rotary arm control submodule 410 and a laser cleaning submodule 420;

[0138] The rotating arm control submodule 410 is used to control the rotating arm and adjust the laser emitter so that the focus of the laser emitter is aligned with the dirt when the current real-time position is the target dirt position;

[0139] The laser cleaning submodule 420 is used to control the laser cleaning robot to clean the target dirt location.

[0140] In embodiments of the present invention, the focus of the laser emitter is aligned with the target dirt position by controlling the rotating arm. The laser emitter has high energy at its focus, which can more effectively clean the dirt, thereby improving the cleaning effect.

[0141] In some application scenarios, the workpiece is a curved surface, and conventional laser cleaning robots cannot perform real-time cleaning on curved surfaces. Therefore, in some embodiments of the present invention, the laser cleaning robot is a magnetic laser cleaning robot. The magnetic laser cleaning robot can be magnetically attached to the surface of the curved workpiece to achieve real-time cleaning of the curved workpiece.

[0142] In a specific embodiment of the present invention, the path smoothing module 500 can be specifically used to control the rotating arm control submodule 410 to swing the rotating arm left and right when the current real-time position of the laser cleaning robot is the target dirt position, so that the focus of the laser emitter is aligned with the target dirt position.

[0143] The laser cleaning submodule 420, when the laser control switch is turned on, completes the cleaning process, and the laser cleaning robot cleans the target dirt location.

[0144] In some embodiments of the present invention, the laser cleaning system further includes a detection module 600, which is used to determine whether the target location is clean. If it is not clean, the laser cleaning robot is controlled to clean the target location again.

[0145] In one specific embodiment, after laser cleaning, the camera acquires images of the surface to be cleaned and the dirt. After image processing, the cleaned area is inspected online to determine whether there are any areas that have not been thoroughly cleaned. If so, the laser cleaning robot is controlled to move to the area for secondary cleaning, thereby realizing an automated cleaning process.

[0146] This invention also provides a laser cleaning method, applicable to any one of the 10 embodiments of the laser cleaning system described above, such as... Figure 7 As shown, the laser cleaning method includes:

[0147] S701. Based on the target contaminant identification model, identify contaminants on the workpiece surface and determine the location of target contaminants;

[0148] S702. Obtain the current position of the laser cleaning robot, and perform path planning based on the current position and the position of the dirt to obtain the planned path of the laser cleaning robot.

[0149] S703. Obtain the theoretical real-time position and current real-time position of the laser cleaning robot based on the planned path, and adjust the current real-time position of the laser cleaning robot according to the theoretical real-time position.

[0150] S704. When the current real-time location is the target dirt location, control the laser cleaning robot to clean the target dirt location.

[0151] The laser cleaning method provided in the above embodiments can realize the technical solutions described in the above laser cleaning system embodiments. The specific implementation principles of each module or sub-module can be found in the corresponding content of the above laser cleaning system embodiments, which will not be repeated here.

[0152] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware, and the program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0153] The laser cleaning system and method provided by the present invention have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A laser cleaning system for controlling a laser cleaning robot to clean the surface of a workpiece, characterized in that, The laser cleaning system includes a dirt recognition module, a cleaning path planning module, a path smoothing module, a control module, and a cleaning module. The dirt recognition module is used to identify dirt on the surface of the workpiece based on the target dirt recognition model and determine the location of the target dirt. The cleaning path planning module is used to obtain the current position of the laser cleaning robot and perform path planning based on the current position and the target dirt position to obtain the planned path of the laser cleaning robot. The cleaning path planning module includes a map acquisition submodule and a path planning submodule; the map acquisition submodule is used to acquire a binary grid map; The path planning submodule is used to search for a path and determine the planned path based on the location of the target dirt, the current position of the laser cleaning robot, and the binary grid map using a preset path planning algorithm. The path smoothing module is used to smooth the planned path and determine the planned smooth path; the path smoothing module includes a path list acquisition submodule, a line-of-sight path smoothing submodule, a path point refinement path smoothing submodule, and an interpolation path smoothing submodule; the path list acquisition submodule is used to obtain the number of path lists and all nodes of each path according to the planned path, and determine whether the number of path lists is greater than or equal to a first threshold. The line-of-sight path smoothing submodule is used to define a first node, a second node, and a third node when the number of paths in the list is greater than or equal to a first threshold, and to adjust the first node, the second node, and the third node based on a preset adjustment rule; the path point refinement path smoothing submodule is used to determine the minimum turning path and all nodes of the minimum turning path when the number of paths in the list is less than the first threshold or the third node is the endpoint, and to update all nodes of the minimum turning path according to the optimal control distance; the interpolation path smoothing submodule is used to determine the planned smoothing path based on all nodes of the minimum turning path through cubic spline interpolation; The control module is used to obtain the theoretical real-time position and current real-time position of the laser cleaning robot based on the planned path, and adjust the current real-time position of the laser cleaning robot according to the theoretical real-time position. The cleaning module is used to control the laser cleaning robot to clean the target dirt location when the current real-time location is the target dirt location.

2. The laser cleaning system according to claim 1, characterized in that, The dirt identification module includes: an initial identification model construction submodule, a dataset acquisition submodule, and a training submodule; The initial identification model construction submodule is used to construct the initial dirt identification model; The dataset acquisition submodule is used to acquire datasets; The training submodule is used to train the initial dirt recognition model based on the dataset and a preset loss function to obtain the target dirt recognition model.

3. The laser cleaning system according to claim 2, characterized in that, The loss function for: in, The bounding box loss function is... The bounding box confidence loss function is... For the target classification loss function, The number of candidate bounding boxes and anchor boxes generated for each grid. To determine whether a candidate box is responsible for the parameters of that object, For the improved cross-union ratio function, For the sample size, for loss function This represents the probability distribution of the confidence level of the true bounding box. To predict the probability distribution of bounding box confidence, The cross-entropy loss function is the probability distribution of the true bounding box confidence and the probability distribution of the predicted bounding box confidence. For the sample size, The probability distribution for classifying the true target. To predict the probability distribution of the target classification, For the number of categories, The cross-entropy loss function represents the probability distribution of the true target classification and the probability distribution of the predicted target classification.

4. The laser cleaning system according to claim 1, characterized in that, The line-of-sight path smoothing submodule includes a first judgment module, a second judgment module, and a third judgment module; The first determination module is used to determine whether the third node is visible from the first node; if it is visible, then proceed to the second determination module. If not visible, update the first node, the second node, and the third node, and then proceed to the first judgment module for another judgment; The second judgment module is used to obtain the weights of the first node, the second node, and the third node, and to determine whether the weight of the first node from the second node to the third node is greater than the weight of the first node directly from the third node; if so, the second node is removed and the weight of the third node is updated, and the process proceeds to the third judgment module; if not, the first node, the second node, and the third node are updated, and the process proceeds to the first judgment module for judgment again. The third judgment module is used to determine whether the third node is the destination. If it is the destination, the minimum turning path is determined. If the third node is not the destination, the first node, the second node, and the third node are updated, and the first judgment module is entered again for judgment.

5. The laser cleaning system according to claim 1, characterized in that, The control module includes a theoretical real-time position acquisition submodule, a current real-time position acquisition submodule, a compensation signal submodule, and a position adjustment submodule; The theoretical real-time position acquisition submodule is used to determine the theoretical real-time position of the laser cleaning robot based on the planned path; The current real-time position acquisition submodule is used to determine the current real-time position of the laser cleaning robot; The compensation signal submodule is used to generate a compensation signal based on the current real-time position and the theoretical real-time position; The position adjustment submodule is used to adjust the current real-time position of the laser cleaning robot based on the compensation signal.

6. The laser cleaning system according to claim 1, characterized in that, The laser cleaning robot includes a laser emitter and a rotating arm for adjusting the laser emitter; the cleaning module includes a rotating arm control submodule and a laser cleaning submodule. The rotating arm control submodule is used to control the rotating arm to adjust the laser emitter when the current real-time position is the target dirt position, so that the focus of the laser emitter is aligned with the target dirt position; The laser cleaning submodule is used to control the laser cleaning robot to clean the target contaminant location.

7. The laser cleaning system according to claim 1, characterized in that, The laser cleaning system also includes a detection module, which is used to determine whether the target dirt location has been cleaned. If it has not been cleaned, the system controls the laser cleaning robot to clean the target dirt location again.

Images