Method and apparatus for detecting defects in a wind turbine blade layup process, electronic device and medium

Abstract

The application discloses a wind power blade layering process defect detection method and device, electronic equipment and medium, relates to the technical field of wind power generation, and the defect detection method comprises the following steps: collecting a to-be-detected image in a layering process after first data sent reaches a preset numerical threshold value, and generating second data; sending the second data to a server; receiving target defect data sent by the server; and respectively performing early warning and laser marking based on the target defect data. In the process of identifying the on-site image by the mature identification model, when the defect size exceeds the preset defect threshold value, by analyzing the center coordinates of the rectangular frame and the actual length of the image single pixel point mapped to the on-site, the actual coordinates of the defect center can be determined, the server sends instruction information to the laser holder, the laser holder can alarm, and the laser is projected to the actual coordinate position of the defect center, so that accurate defect positioning and alarm are realized.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of wind power generation, and particularly relates to a wind power blade layering process defect detection method and device, electronic equipment and a medium. BACKGROUND

[0002] Visual defect detection has been carried out in the detection field of the wind power blade industry. The defect detection applied in the related technology is all directed to blade finished products or wind field hanging blades, and image collection, analysis and the like are carried out by borrowing unmanned aerial vehicles, mobile robots and the like. For the wind power blade industry, the application in the blade manufacturing process scene has not been carried out at present, and the related visual defect detection technology focuses on the development of machine equipment and the optimization of image splicing positioning algorithm, and can only identify cracks, lightning strikes, carbonization and other appearance damage defects, and cannot meet the needs of the factory for monitoring and identifying the manufacturing end personnel behavior and operation process defects, and it is difficult to prevent and correct defects from the source. SUMMARY

[0003] The wind power blade layering process defect detection method and device, electronic equipment and medium provided by the embodiments of the application solve the problem that the wind power blade factory cannot monitor and identify the manufacturing end personnel behavior and operation process defects, and it is difficult to prevent and correct defects from the source.

[0004] In a first aspect, the embodiments of the application provide a wind power blade layering process defect detection method for a terminal device, which comprises: after the first data sent reaches a preset numerical threshold value, collecting a to-be-tested image in the layering process to generate second data; sending the second data to a server, so that the server identifies the second data based on a mature discrimination model, and determines target defect data corresponding to the second data based on the mature discrimination model in the case that the mature discrimination model identifies a specific semantic action; and receiving the target defect data sent by the server, and respectively performing early warning and laser marking based on the target defect data.

[0005] According to the first aspect of the embodiments of the application, before the to-be-tested image in the layering process is collected to generate the second data after the first data sent reaches the preset numerical threshold value, the method further comprises: collecting a defect image in the specific semantic action and the layering process to generate the first data; and sending the first data to the server.

[0006] According to a first aspect of the embodiments of the present application, before the first data is generated by collecting the defect images in the specific semantic action and the laying process, the method further comprises: determining target width information according to a focal length, a working distance and a target height of an industrial area array camera to be installed, the industrial area array camera being used to collect the defect images and the to-be-measured images; and installing the industrial area array camera at predetermined intervals along an edge of a main mold of a wind turbine blade, based on the target width information, wherein in a root region of the wind turbine blade, two groups of the industrial area array cameras are symmetrically arranged on two sides of a laying work area, and in a mid-region and a tip region of the wind turbine blade, one group of the industrial area cameras is arranged on one side of the laying work area.

[0007] According to the first aspect of the embodiments of the present application, before the target defect data is received from the server and the pre-warning and the laser marking are performed based on the target defect data, the method further comprises: installing a laser holder along an artificial channel on two sides of a laying work area of a wind turbine blade, based on the target width information, the laser holder being correspondingly arranged with the industrial area array camera and being arranged at intervals, and the laser holder being used to perform laser marking based on the target defect data to determine a defect position in the laying process of the wind turbine blade.

[0008] According to the first aspect of the embodiments of the present application, before the target defect data is received from the server and the pre-warning and the laser marking are performed based on the target defect data, the method further comprises: projecting two end points on the mold of the wind turbine blade using the laser holder; determining a target section covering a line connecting the two end points and measuring a length L of the line connecting the two end points; and using the industrial area array camera to shoot an image covering the target section to determine third data, wherein the third data comprises the length L of the target section and the image covering the target section; and sending the third data to the server, so that the server determines a number of pixel points of the target section in the third data based on pixel coordinates of the two end points in the third data.

[0009] In a second aspect, the embodiments of the present application provide a wind turbine blade laying process defect detection method for a server, the method comprising: receiving second data sent by a terminal device, identifying the second data based on a mature discrimination model, and in a case where the mature discrimination model identifies a specific semantic action, determining target defect data corresponding to the second data based on the mature discrimination model, wherein the target defect data comprises defect position coordinate information and size information; and sending the target defect data to the terminal device, so that the terminal device performs pre-warning and laser marking based on the target defect data.

[0010] According to a second aspect of the embodiments of the present application, before the target defect data corresponding to the second data is determined based on the mature discrimination model in the case that the specific semantic action is identified by the mature discrimination model, the method further comprises: receiving first data sent by the terminal device, the first data comprising the specific semantic action and a defect image in the laying process; preprocessing the first data to determine labeled defect data; and determining the mature discrimination model based on the specific semantic action and the labeled defect data through training and tuning.

[0011] According to a second aspect of the embodiments of the present application, before the target defect data is sent to the terminal device for the terminal device to respectively perform early warning and laser marking based on the target defect data, the method further comprises: receiving third data sent by the terminal device, and determining the number of pixel points of a target section in the third data based on the pixel coordinates of two end points in the third data; determining an actual length d of a single pixel point according to the length L of the target section and the number of pixel points of the target section; selecting a target defect image from the defect information according to a preset defect threshold, and determining a target rectangular frame corresponding to the target defect image and a center point coordinate of the target rectangular frame, wherein the preset defect threshold comprises a preset length-width ratio threshold and a preset area threshold; calculating a to-be-labeled coordinate based on the center point coordinate of the target rectangular frame and the actual length d of the single pixel point, and determining the target defect data.

[0012] According to a second aspect of the embodiments of the present application, preprocessing the first data to determine labeled defect data comprises: performing image cleaning on the first data to remove each of repeated images, low-quality images and non-standard images, and determining first intermediate data; performing flipping, brightness adjustment and noise processing on the first intermediate data to determine second intermediate data; and labeling the second intermediate data by using an image labeling tool to determine labeled defect data, wherein the labeled defect data comprises the type of the defect and the rectangular frame coordinate of the defect.

[0013] According to a second aspect of the embodiments of the present application, the mature discrimination model is determined based on the specific semantic action and the labeled defect data through training and tuning, comprising: dividing the specific semantic action and the labeled defect data into a training set and a test set according to a predetermined proportion; training an initial discrimination model by using a YOLOV5 algorithm in the training set, and generating visualized data, wherein the visualized data comprises training loss, validation loss, a confusion matrix, precision and recall.

[0014] According to a second aspect of the embodiments of this application, after training an initial discrimination model using YOLOv5 based on a training set and generating visualization data, the method further includes: testing a test set based on the initial discrimination model to determine the accuracy and detection rate; adjusting the hyperparameters of the initial discrimination model using a controlled variable method based on the accuracy and detection rate of the test set to optimize the initial discrimination model and improve the detection rate, thereby determining a mature discrimination model, wherein the hyperparameters include the learning rate and batch size.

[0015] Thirdly, embodiments of this application provide an automatic defect detection device for wind turbine blade layup process, used in terminal equipment. The device includes: a data acquisition module, used to acquire a test image during the layup process and generate second data after the first data sent reaches a preset numerical threshold; a first sending module, used to send the second data to a server for the server to identify the second data based on a mature identification model, and to determine the target defect data corresponding to the second data based on the mature identification model when the mature identification model identifies a specific semantic action; and a receiving module, used to receive the target defect data sent by the server and to perform early warning and laser marking based on the target defect data.

[0016] Fourthly, embodiments of this application provide an automatic defect detection device for wind turbine blade layup process, used in a server. The device includes: an identification module, used to receive second data sent by a terminal device, identify the second data based on a mature identification model, and determine the target defect data corresponding to the second data based on the mature identification model when the mature identification model identifies a specific semantic action; and a second sending module, used to send the target defect data to the terminal device so that the terminal device can perform early warning and laser marking based on the target defect data.

[0017] Fifthly, embodiments of this application provide an electronic device, which includes a transceiver, a processor, a memory, and a program stored in the memory and executable on the processor. When the program is executed by the processor, it implements the wind turbine blade layup process defect detection method described in the first or second aspect above.

[0018] In a sixth aspect, embodiments of this application provide a readable storage medium on which a program or instructions are stored, which, when executed by a processor, implement the steps of the methods in the first and second aspects described above.

[0019] The wind power blade layering process defect detection method, device, electronic equipment and medium provided by the embodiment of the application, by collecting specific semantic actions and defect images in the wind power blade layering process to train and determine a mature discrimination model, applying the mature discrimination model to actual layering operations, discriminating the on-site images collected by the industrial area array camera and determining the defect position and defect size, in the process of discriminating the on-site images by the mature discrimination model, the model will automatically mark a rectangular frame in the image defect area, by comparing the detected defect size with the preset defect threshold, the server can determine whether to remind the operator to process the defect in time. In the case that the defect size exceeds the preset defect threshold, by analyzing the center coordinates of the rectangular frame and the actual length of the mapping of a single pixel point of the image to the on-site, the actual coordinates of the defect center can be determined, the server sends instruction information to the laser holder, the laser holder can alarm and project laser to the actual coordinates of the defect center, realizing accurate defect positioning and alarming, so that the operator can correct the defect in time in the wind power blade layering process, and the specific semantic action of the operator can be recognized as the starting signal of detecting the layering defect, reducing unnecessary alarm frequency, and filling the blank of intelligent defect detection for the wind power blade layering process. BRIEF DESCRIPTION OF DRAWINGS

[0020] The features, advantages, and technical effects of the exemplary embodiments of the application will be described below with reference to the accompanying drawings.

[0021] Figure 1 is a flowchart of a wind power blade layering process defect detection method provided by the embodiment of the application;

[0022] Figure 2 is a flowchart of another wind power blade layering process defect detection method provided by the embodiment of the application;

[0023] Figure 3 is a flowchart of a method for determining the actual size of the mapping of a single pixel point;

[0024] Figure 4 is a flowchart of target defect data determination provided by the embodiment of the application;

[0025] Figure 5 is a structural diagram of a wind power blade layering process defect detection device provided by the embodiment of the application;

[0026] Figure 6 is a structural diagram of another wind power blade layering process defect detection device provided by the embodiment of the application;

[0027] Figure 7 is a structural diagram of another wind power blade layering process defect detection device provided by the embodiment of the application;

[0028] Figure 8 is a structural schematic diagram of still another wind power blade layering process defect detection device provided by an embodiment of the present application;

[0029] Figure 9 is a structural schematic diagram of an electronic device provided by an embodiment of the present application;

[0030] Figure 10 is a layout schematic diagram of an industrial area array camera in a wind power blade layering process provided by an embodiment of the present application;

[0031] Figure 11 is a connection schematic diagram of a terminal device and a server provided by an embodiment of the present application;

[0032] Figure 12 is a layout schematic diagram of an industrial area array camera and a laser holder provided by an embodiment of the present application.

[0033] Reference signs:

[0034] A, root area; B, middle area; C, tip area;

[0035] SS, leeward surface; PS, windward surface;

[0036] a, industrial area array camera; b, alarm system; c, laser holder; g, support. DETAILED DESCRIPTION

[0037] The features and exemplary embodiments of various aspects of the present application will be described in detail below, in order to make the purposes, technical solutions and advantages of the present application more clear and apparent, the present application will be further described in detail below in combination with the drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain the present application, but not to limit the present application. The present application can be implemented without some of these specific details by those skilled in the art. The following description of the embodiments is only to provide a better understanding of the present application by showing examples of the present application.

[0038] It is to be noted that, in the present document, relational terms such as first and second and the like can be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises... a" does not, without more constraints, exclude the presence of additional identical elements in the process, method, article, or apparatus that comprises the element.

[0039] The wind power blade is large in size, and when each part of the wind power blade (for example, a blade main beam, a skin, and the like) is manufactured, a corresponding mold needs to be laid with layers of fibers as a structural cloth layer of a component, and the corresponding component is formed by resin infusion on the structural cloth layer. The production and manufacturing of the wind power blade are personnel-intensive operations, and the process is complicated, and various defects are prone to occur. Different degrees and different categories of defects in the laying layer will affect the quality and performance of the blade, and the quality requirement of the blade production is extremely high, and the identification and detection of defects in the first process of laying are crucial. At the present stage, experienced inspectors mainly supervise and remind in time, and the workload is large and easy to miss detection.

[0040] To solve the problems in the prior art, the embodiment of the present application provides a wind power blade laying process defect detection method, device, electronic equipment and medium.

[0041] First, a wind power blade laying process defect detection method provided by the embodiment of the present application is introduced.

[0042] The flowchart of the wind power blade laying process defect detection method provided by the embodiment of the present application, which is used for a terminal device and can include the following steps S110-S130.

[0043] S110, after the first data sent reaches a preset numerical threshold value, collecting a to-be-tested image in a laying process to generate second data.

[0044] In the embodiment of the present application, the preset numerical threshold value is the number of first data required for the YOLOV5 algorithm training model sent by the terminal device side to the server, and after the model training is completed on the server side, the terminal device in the wind power blade laying field can collect the to-be-tested image and generate the second data.

[0045] Based on this, in order to reduce the defects (such as: cloth layer wrinkle, cloth layer folding, cloth layer loose yarn, foreign matter, core material block, etc.) in the wind turbine blade layer operation, the terminal equipment captures images and transmits them to the server side for identification, which can realize automatic monitoring of the defects of the blade layer operation.

[0046] S120, send the second data to the server for the server to identify the second data based on the mature identification model, and in the case that the mature identification model identifies the specific semantic action, determine the target defect data corresponding to the second data based on the mature identification model.

[0047] In the embodiment of the application, when the terminal equipment side collects the specific semantic action given by the blade layer site operation personnel, the mature identification model identifies the specific semantic action, and then starts the defect identification program to identify, locate, analyze and record the second data.

[0048] Based on this, the algorithm model on the server side has the characteristics of high intelligence, and the model identifies the specific semantic action (such as: stop rolling cloth action behavior) and then detects the defect, instead of detecting the defect all the time during the cloth laying process, which can reduce unnecessary defect alarm frequency.

[0049] S130, receive the target defect data sent by the server, and perform pre-warning and laser marking based on the target defect data.

[0050] In the embodiment of the application, the target defect data contains coordinate information of the actual position of the defect. Based on this, the terminal equipment in the blade layer site can automatically alarm and mark the defect position after receiving the target defect data, to remind the operation personnel in the layering operation site to timely rectify the defect position.

[0051] The above is a specific implementation of the wind turbine blade layer process defect detection method for the terminal equipment provided in the embodiment of the application. The terminal equipment first collects the first data and sends it to the server, so that the server side establishes a mature identification model according to the received first data. In the case that the first data collected by the terminal equipment reaches a preset numerical threshold, the terminal equipment collects and sends the second data to the server. The server side can automatically identify whether the second data has a defect, information of the defect, etc. according to the mature identification model, and send the identification result to the terminal equipment side to remind the operation personnel to timely rectify the defect position in the wind turbine blade layer process.

[0052] In some embodiments, based on the aforementioned wind turbine blade lamination process defect detection method, the application further provides an implementation manner of the wind turbine blade lamination process defect detection method, which further comprises the following steps before S110: (1) collecting defect images in the specific semantic action and the lamination process to generate first data; and (2) sending the first data to the server.

[0053] In the embodiments of the application, in order to ensure that the model trained on the server side can start to determine the target defect data after recognizing the specific semantic action of the operator, in the process of training the model, in addition to collecting defect images, the terminal device also needs to collect the specific semantic action of the operator and send it to the server to train the model through the YOLOV5 algorithm.

[0054] Based on this, the model trained by the server side based on the first data can recognize the defect information in the image, and can also recognize whether the specific semantic action is included in the image. When the model recognizes that the specific semantic action is included in an image, the model starts the program of defect recognition and marking, which can improve the necessity of defect reminder information pushing, and the program of model defect recognition can effectively interact with the behavior information of the operator in the work site.

[0055] In some embodiments, based on the aforementioned wind turbine blade lamination process defect detection method, the application further provides another implementation manner of the wind turbine blade lamination process defect detection method, which can further comprise the following steps before collecting defect images in the specific semantic action and the lamination process to generate first data: (1) determining target width information according to the focal length, working distance and target height of the industrial area array camera a to be installed, the industrial area array camera a being used to collect defect images and test images; and (2) installing the industrial area array camera a along the edge of the main mold of the wind turbine blade at a predetermined interval based on the target width information.

[0056] It can be understood that, as shown in Figure 10 FIG. 1 is a schematic view of the planar surface of the wind turbine blade lamination, in which SS is the leeward surface of the wind turbine blade, and PS is the windward surface of the wind turbine blade. The chordwise width of the root region A of the wind turbine blade is large, and therefore, in order to ensure that the industrial area array camera a can collect complete images of the root region A, the industrial area array camera a is provided with two groups and symmetrically arranged on both sides of the root region A in the lamination work area. Since the chordwise width of the mid-region B and the tip region C of the wind turbine blade is small, one group of industrial area array cameras a is arranged on one side of the mid-region B and the tip region C, which can ensure that the industrial area array camera a can collect complete images of the mid-region B and the tip region C.

[0057] In the embodiments of the present application, the industrial area array camera a is a terminal tool for collecting images of the blade laying site, and the industrial area array camera a can collect the entire surface of the blade mold. The industrial area array camera a is installed about 1 meter beside the worker channel, as shown in Figure 12 Based on this, the target width information can be determined based on the camera focal length, working distance, and target height, so as to determine the field of view size, and the installation of the industrial area array camera a is performed according to the target width information.

[0058] In some embodiments, based on the foregoing wind power blade laying process defect detection method, the present application also provides another implementation manner of the wind power blade laying process defect detection method, which can further include the following steps before receiving the target defect data sent by the server, and performing pre-warning and laser marking based on the target defect data: based on the target width information, installing the laser holder c along the artificial channel on both sides of the wind power blade laying work area, please refer to Figure 12 The laser holder c is arranged in pairs and at intervals with the industrial area array camera a, and the laser holder c is used for laser marking based on the target defect data to determine the defect position in the wind power blade laying process.

[0059] In the embodiments of the present application, please refer to Figure 12 The industrial area array camera a and the laser holder c are installed at intervals on the bracket d, and each industrial area array camera a is provided with a laser holder c on one side. It can be understood that the blade root area A, the blade middle area B, and the blade tip area C of the wind power blade are each provided with a plurality of laser holders c. Based on this, after the server side identifies and calculates the actual coordinate information of the defect, the coordinate information is sent to the corresponding laser holder c, so that the corresponding laser holder c projects a laser point towards the coordinate position to remind the operator of the position of the defect.

[0060] In one example, please refer to Figure 11 The industrial area array camera a and the alarm system b are arranged in pairs and at intervals, the alarm system b includes a laser holder c capable of automatically projecting a laser point according to the coordinate information returned by the server, and an alarm capable of alarming according to instructions, and the alarm and the laser holder c cooperate with each other. When the server returns the target defect data, the alarm alarms, and at the same time, the laser holder c performs laser positioning indication on the defect position.

[0061] In another example, in addition to the laser positioning indication on the defect position, the laser holder c can also automatically complete the defect record and trace report. After one laying work is completed, the laser holder c can automatically send the defect record report to the server side, so that the quality department performs data analysis.

[0062] In some embodiments, it can be understood that in the case that the model trained on the server side can identify the position information of the defects in the image, it is also necessary to establish the correspondence between the image coordinates and the actual coordinates. To this end, after obtaining the defect coordinates in the image, the image defect coordinates need to be weighted using the conversion scale of the position between the image and the actual.

[0063] Based on the foregoing wind power blade layering process defect detection method, the application further provides another implementation manner of the wind power blade layering process defect detection method. Before receiving the target defect data sent by the server, and based on the target defect data, the implementation manner performs pre-warning and laser marking, as shown in the following steps: Figure 3

[0064] S310, using the laser holder c to arbitrarily project two endpoints on the wind power blade mold.

[0065] S320, determining a target section covering the line connecting the two endpoints, and measuring the length L of the line connecting the two endpoints.

[0066] S330, using the industrial area array camera a to shoot an image covering the target section, and determining third data, wherein the third data includes the length L of the target section and the image covering the target section.

[0067] S340, sending the third data to the server, so that the server determines the number of pixel points of the target section in the third data based on the pixel coordinates of the two endpoints in the third data.

[0068] In the embodiment of the application, to calculate the conversion scale between the position coordinates in the image shot by the industrial area array camera a installed on the layering site and the actual coordinates, two points are projected on the blade mold. The length L of the target section composed of the two points as endpoints can be obtained by measurement. From the image covering the target section shot by the industrial area array camera a, the pixel coordinates of the two endpoints can be obtained, so that the total number of pixel points in the target section can be calculated. From the total number of pixel points and the length L of the target section, the actual size of a single pixel point in the image shot by the industrial area array camera a mapped on the layering site can be determined, that is, the size conversion scale between the image and the actual is determined.

[0069] Based on this, after the mature identification model identifies and calculates the center coordinates of the defects in an image, the center coordinates of the defects in the image are weighted using the foregoing size conversion scale between the image and the actual, so that the actual coordinate information of the defects in the image corresponding to the wind power blade layering site can be determined, so that the laser holder c can position the defect position to remind the workers on the layering site.

[0070] The embodiment of the application further provides a flowchart of a wind power blade layering process defect detection method, as shown in the following figure:​Figure 2 As shown, the wind turbine blade layering process defect detection method is used for a server, and can include the following steps S210-S220.

[0071] S210, receiving second data sent by a terminal device, identifying the second data based on a mature discrimination model, and determining target defect data corresponding to the second data based on the mature discrimination model in a case where the mature discrimination model identifies a specific semantic action, wherein the target defect data includes defect position coordinate information and size information.

[0072] In an embodiment of the present application, the server side has completed training for various defects in blade layering when receiving the second data, and can identify information contained in the second data according to the mature discrimination model. The identification process of the mature discrimination model includes two aspects. On the one hand, the model needs to detect whether the second data contains a specific semantic action, which is a start instruction given by the layering site to start defect detection. On the other hand, the model starts defect identification on the second data after detecting the specific semantic action from the second data. Based on this, in a case where there is a defect in an image corresponding to one second data, the model can automatically give the position coordinate information and size information of the defect, i.e., target defect information.

[0073] S220, sending the target defect data to the terminal device for the terminal device to respectively perform early warning and laser marking based on the target defect data.

[0074] In an embodiment of the present application, after the mature discrimination model identifies the size and position information of the defect, it can obtain the approximate center coordinate information of the defect center position in the image, and perform the aforementioned weighted processing of the size conversion scale on the approximate center coordinate information in the image, thereby obtaining the actual coordinate information of the defect.

[0075] Based on this, the target defect information sent by the server to the terminal device contains the actual coordinate position of the corresponding defect in the layering operation site, so as to enable the laser holder c to position the defect. The laser holder c can quickly position and turn to the line laser to aim at the defect according to the coordinates, and continue for 3 seconds. If there are multiple defects in the field of view of the industrial area camera a, the laser holder c traverses each defect according to the coordinate size in order until there is no defect in the field of view.

[0076] In some embodiments, before S210, the wind turbine blade layering process defect detection method can further include the following steps:

[0077] (1) receiving first data sent by a terminal device, the first data including a specific semantic action and a defect image in a lamination process; (2) preprocessing the first data to determine labeled defect data; and (3) determining a mature discrimination model through training and tuning based on the specific semantic action and the labeled defect data.

[0078] In the embodiments of the present application, the server side needs to first preprocess the first data collected and sent by the front end to obtain an ideal image that is relatively clear and convenient for computer to recognize features before training the model based on the YOLOV5 algorithm. In the model training process, the model needs to identify the specific semantic action given by the operator in addition to the ability to identify defects.

[0079] It can be understood that the blade lamination process mainly includes operations such as cloth laying, cloth rolling, and glue spraying. In order to make the cloth layer as flat and wrinkle-free as possible, the cloth rolling operation is often needed. In the current production mode, if there are still defects such as wrinkles and folds after the operator rolls the cloth, the supervisor needs to remind the operator to modify in time. In order to make the algorithm model more humanized and intelligent, the model identifies the operator's stop rolling action behavior and then performs defect detection, rather than detecting all the time during the cloth laying process, which greatly reduces the unnecessary defect alarm frequency.

[0080] Therefore, the mature discrimination model trained by the server side based on the YOLOV5 algorithm will first identify whether the specific semantic action for starting defect detection given by the operator in the lamination site is included in the image when formally put into application.

[0081] In some embodiments, before S220, as shown in Figure 4 The wind power blade lamination process defect detection method can further include the following steps:

[0082] S410, receiving third data sent by a terminal device, and determining the number of pixel points of a target section in the third data based on the pixel coordinates of two end points in the third data.

[0083] In the embodiments of the present application, when calculating the number of pixel points of the target section, first determine the horizontal coordinate difference and the vertical coordinate difference of the two pixel coordinates, then square the horizontal coordinate difference and the vertical coordinate difference respectively, and sum them to obtain an intermediate value. Based on this, square the intermediate value to obtain the number of pixel points of the target section.

[0084] S420, determining the actual length d mapped by a single pixel point according to the length L of the target section and the number of pixel points of the target section.

[0085] In the embodiments of the present application, according to the total length L of the target section and the corresponding number of pixel points, the actual size corresponding to a single pixel point can be determined.

[0086] S430, screening a target defect image from the defect information according to a preset defect threshold, and determining a target rectangular frame corresponding to the target defect image and a center point coordinate of the target rectangular frame, wherein the preset defect threshold comprises a preset aspect ratio threshold and a preset area threshold.

[0087] In the embodiments of the present application, the preset defect threshold is a threshold at which the defect type and defect size do not seriously affect the quality of the ply and are within an acceptable range. After the mature identification model screens the target defect image that exceeds the predetermined defect threshold from the defect information contained in the second data, the corresponding defect position is automatically marked. Based on this, the model adds a rectangular frame that can completely cover the defect on the image, and estimates the center coordinates of the defect on the image according to the center coordinates of the rectangular frame.

[0088] S440, calculating a to-be-identified coordinate based on the center point coordinate of the target rectangular frame and the actual length d of a single pixel point mapping, and determining target defect data.

[0089] In the embodiments of the present application, the actual length d of a single pixel point mapping is the aforementioned size conversion scale between the image and the actual size. After obtaining the center coordinates of the defect on the image, the actual length d of a single pixel point mapping is weighted to obtain the to-be-identified coordinate.

[0090] In some embodiments, the first data is preprocessed to determine the labeled defect data, which can specifically include the following steps: (1) image cleaning is performed on the first data to remove each of repeated images, low-quality images and non-standard images, to determine first intermediate data; (2) the first intermediate data is flipped, brightness-adjusted and noise-processed to determine second intermediate data; (3) the second intermediate data is labeled by using an image labeling tool to determine the labeled defect data, wherein the labeled defect data includes the type of the defect and the rectangular frame coordinates of the defect.

[0091] In the embodiments of the present application, a large number of defect images and operation worker specific action images (such as stopping rolling action, etc.) are collected by using the industrial area array camera a during the leaf laying process, and image cleaning operation is performed to clean repeated, low-quality and non-standard images; then the original images are modified and processed by flipping, rotating, darkening, brightening, adding pepper and salt noise, adding Gaussian noise and the like to determine the second intermediate data, thereby providing better visual effect while increasing the data amount.

[0092] Based on this, the second intermediate data is labeled by using the labelimg image labeling tool to obtain a file containing picture labeling information, wherein the labeling information includes the type of each defect and the rectangular frame coordinates of the defect.

[0093] In some embodiments, a mature identification model is determined by training and tuning based on specific semantic actions and labeled defect data. Specifically, the following steps may be included: (1) dividing the specific semantic actions and labeled defect data into a training set and a test set according to a predetermined ratio; (2) training an initial identification model using the YOLOV5 algorithm based on the training set and generating visualization data.

[0094] In this embodiment, the YOLOv5 algorithm is used to train the model. During the training process, a large amount of information is generated, including visualization results such as training loss, validation loss, precision, recall, and mAP, as well as statistical images of the dataset such as PR curves, confusion matrices, and test results. Based on this, an initial discrimination model is obtained.

[0095] In some embodiments, after training an initial discrimination model using YOLOv5 based on the training set and generating visualization data, the method may further include the following steps: (1) testing the test set based on the initial discrimination model to determine the accuracy and detection rate; (2) adjusting the hyperparameters of the initial discrimination model using the control variable method according to the accuracy and detection rate of the test set to optimize the initial discrimination model to improve the detection rate and determine the mature discrimination model, wherein the hyperparameters include the learning rate and batch size.

[0096] In this embodiment, after the server-side obtains an initial identification model through training, it uses this initial identification model to detect defects on the test set, observing the accuracy and detection rate of the initial identification model for defect identification. Based on this, by continuously adjusting the hyperparameters in the YOLOv5 algorithm training, the initial identification model can be improved, and a mature identification model that can be put into practical application can be determined.

[0097] This application embodiment also provides a defect detection device 500 for the wind turbine blade layup process. This device 500 is used in terminal equipment, such as... Figure 5 As shown, the device 500 may include the following modules:

[0098] The acquisition module 510 is used to acquire the image to be tested during the layup process and generate second data after the first data sent reaches a preset numerical threshold. The acquisition module 510 can be used to execute S110 described above.

[0099] The first sending module 520 is used to send second data to the server, so that the server can identify the second data based on a mature identification model. If the mature identification model identifies a specific semantic action, the server can determine the target defect data corresponding to the second data based on the mature identification model. The first sending module 520 can be used to execute S120 described above.

[0100] The receiving module 530 receives target defect data sent by the server, and performs early warning and laser marking based on the target defect data, respectively. The receiving module 530 can be used to perform S130 described above.

[0101] According to embodiments of the present application, any multiple modules of the acquisition module 510, the first sending module 520 and the receiving module 530 can be combined in one module, or any one of them can be split into multiple modules. Alternatively, at least part of the function of one or more of these modules can be combined with at least part of the function of other modules, and implemented in one module.

[0102] In some embodiments, referring to Figure 7 The wind power blade layering process defect detection device 500 for the terminal device can further include an acquisition unit 540, which is specifically configured to collect the to-be-tested images in the layering process after the first data sent reaches a preset numerical threshold, collect the defect images in the layering process and the specific semantic actions before generating the first data, and generate the second data; and send the first data to the server.

[0103] In some embodiments, referring to Figure 7 The defect detection device 500 can further include a first installation unit 550, which is specifically configured to determine target width information according to the focal length, working distance and target height of the industrial area array camera to be installed before collecting the defect images in the layering process and generating the first data, the industrial area array camera being used to collect the defect images and the to-be-tested images; and install the industrial area array camera along the edge of the wind power blade master mold at a predetermined interval based on the target width information.

[0104] In some embodiments, referring to Figure 7 The defect detection device 500 can further include a second installation unit 560, which is specifically configured to install a laser holder along the artificial passageway on both sides of the wind power blade layering work area based on the target width information before receiving the target defect data sent by the server and performing early warning and laser marking based on the target defect data, respectively, the laser holder corresponding to and being arranged at intervals with the industrial area array camera, and the laser holder being used to perform laser marking based on the target defect data to determine the defect position in the wind power blade layering process.

[0105] In some embodiments, referring to Figure 7The defect detection apparatus 500 can further include a section sampling unit 570, which is specifically configured to receive the target defect data sent by the server, use a laser holder to project two endpoints on the wind turbine blade mold at will before prewarning and laser marking are performed based on the target defect data, determine a target section covering the line connecting the two endpoints, and measure the length L of the line connecting the two endpoints; use an industrial area array camera to capture an image covering the target section, and determine third data, wherein the third data includes the length L of the target section and the image covering the target section; and send the third data to the server, so that the server determines the number of pixel points of the target section in the third data based on the pixel coordinates of the two endpoints in the third data.

[0106] Figure 7 Each module / unit in the apparatus shown has the function of implementing each step in the foregoing method for detecting defects in the wind turbine blade layering process of a terminal device, and can achieve the corresponding technical effects. For brevity, no further description is given herein.

[0107] The embodiments of the present application also provide a defect detection apparatus 600 for a wind turbine blade layering process, which is used in a server, as shown in the apparatus 600 can include the following modules: Figure 6

[0108] The identification module 610 is configured to receive the second data sent by the terminal device, identify the second data based on the mature identification model, and determine the target defect data corresponding to the second data based on the mature identification model in the case where the mature identification model identifies a specific semantic action. The identification module 610 can be used to perform S210 described above.

[0109] The second sending module 620 is configured to send the target defect data to the terminal device, so that the terminal device performs prewarning and laser marking based on the target defect data. The second sending module 620 can be used to perform S220 described above.

[0110] According to the embodiments of the present application, the identification module 610 and the second sending module 620 can be combined in one module, or any one of them can be split into multiple modules. Alternatively, at least part of the function of one or more of these modules can be combined with at least part of the function of other modules, and implemented in one module.

[0111] In some embodiments, please refer to Figure 8 ​The wind power blade layering process defect detection device 600 for the server can further include a modeling unit 630, which is specifically configured to receive the second data sent by the terminal device, identify the second data based on the mature discriminant model, receive the first data sent by the terminal device before the mature discriminant model identifies the specific semantic action, the first data including the specific semantic action and the defect image in the layering process, preprocess the first data to determine the labeled defect data, determine the mature discriminant model based on the specific semantic action and the labeled defect data by training and optimization.

[0112] In one example, referring to Figure 8 The modeling unit 630 can further include a preprocessing module 640, which is specifically configured to perform image cleaning on the first data to remove each of duplicate images, low-quality images, and non-standard images to determine first intermediate data, perform flipping, brightness adjustment, and noise processing on the first intermediate data to determine second intermediate data, and label the second intermediate data using an image labeling tool to determine the labeled defect data, wherein the labeled defect data includes the type of defect and the rectangular box coordinates of the defect.

[0113] In one example, referring to Figure 8 The modeling unit 630 can further include a first modeling module 650, which is specifically configured to divide the specific semantic action and the labeled defect data into a training set and a test set according to a predetermined ratio, train an initial discriminant model using a YOLOV5 algorithm based on the training set, and generate visual data, wherein the visual data includes training loss, validation loss, confusion matrix, precision, and recall.

[0114] In one example, referring to Figure 8 The modeling unit 630 can further include a second modeling module 660, which is specifically configured to test the test set based on the initial discriminant model after training the initial discriminant model using YOLOV5 based on the training set and generating visual data, determine the accuracy and the detection rate, adjust the hyperparameters of the initial discriminant model using a control variable method according to the accuracy and the detection rate of the test set to optimize the initial discriminant model to improve the detection rate, and determine the mature discriminant model, wherein the hyperparameters include a learning rate and a batch capacity.

[0115] In some embodiments, referring to Figure 8The defect detection apparatus 600 can further include a section analysis unit 670, which is specifically configured to receive third data sent by the terminal device before sending target defect data to the terminal device for the terminal device to respectively perform early warning and laser marking based on the target defect data, and determine the number of pixel points of a target section in the third data based on the pixel coordinates of the two end points in the third data; determine an actual length d of a single pixel point mapping according to the length L of the target section and the number of pixel points of the target section; select a target defect image from the defect information according to a preset defect threshold, and determine a target rectangular frame corresponding to the target defect image and a center point coordinate of the target rectangular frame, wherein the preset defect threshold includes a preset aspect ratio threshold and a preset area threshold; calculate a to-be-marked coordinate based on the center point coordinate of the target rectangular frame and the actual length d of the single pixel point mapping, and determine the target defect data.

[0116] Figure 8 Each module / unit in the apparatus has the function of implementing each step in the foregoing method for detecting defects in a wind turbine blade lamination process for a server, and can achieve the corresponding technical effects. For brevity, no further description is given here.

[0117] In some embodiments, the present application provides an electronic device, a structural schematic diagram of which is shown in Figure 9

[0118] The electronic device can include a processor 750 and a memory 760 storing computer program instructions.

[0119] Specifically, the processor 750 described above can include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or can be configured as one or more integrated circuits that implement embodiments of the present application.

[0120] The memory 760 can include a mass storage for data or instructions. By way of example and not limitation, the memory 760 can include a hard disk drive (HDD), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a universal serial bus (USB) drive or a combination of two or more of these. Where appropriate, the memory 760 can include removable or non-removable (or fixed) media. Where appropriate, the memory 760 can be internal or external to the integrated gateway disaster recovery device. In a particular embodiment, the memory 760 is non-volatile solid-state memory.

[0121] ​Memory 760 can include read-only memory (ROM), random-access memory (RAM), magnetic disk storage mediums, optical storage mediums, flash memory devices, electrical, optical, or other physical / tangible memory storage devices. Accordingly, generally, memory 760 includes one or more tangible (non-transitory) computer-readable storage media (e.g., a memory device) encoded with software that, when executed, e.g., by the one or more processors, is configured to perform the operations described above in any of the methods of gathering user behavior data in the embodiments described above.

[0122] Processor 750 implements any of the methods of detecting defects in a wind turbine blade layup process described above by reading and executing computer program instructions stored in memory 760.

[0123] In one example, the electronic device can further include bus 710 and bus interface 740. As shown, processor 750, memory 760 are connected through bus 710 and complete communication between each other. Figure 10

[0124] Bus 710 includes hardware, software, or both that couples components of the online data traffic billing device to each other in a form, non-limiting example. Bus 710 can include, without limitation, an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an InfiniBand™ interconnect, a Low Pin Count (LPC) bus, a memory bus, a Micro Channel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association local (VLB) bus, or another suitable bus or a combination of two or more of these. Where suitable, bus 710 can include one or more buses. Although a particular bus is described and illustrated in this embodiment, the present application contemplates any suitable bus or interconnect.

[0125] Bus interface 740 provides an interface between bus 710 and transceiver 720, which provides a means for communicating with various other apparatus over a transmission medium. Data processed by processor 750 is transmitted over a wireless medium via antenna 730, and further, antenna 730 receives data and transmits the data to processor 750.

[0126] In one example, the electronic device is composed of a terminal device and a server, and the terminal device and the server exchange information through a communication module. Please refer to Figure 11 ​, the power supply and communication of the laser holder c in the industrial area array camera a and the alarm system b are connected to the POE switch through the cable of the self-equipment, the POE switch collects the cable and converts the electrical signal into the optical signal through the photoelectric module, the optical signal is transmitted to the office where the server is located through the optical fiber, and then the optical signal is converted into the electrical signal through the photoelectric module to supply power to the equipment. Among them, the communication equipment such as the POE switch is placed in the customized vertical rod box, which can effectively prevent the interference of dust and other factors on the site.

[0127] In addition, in combination with the wind turbine blade layering process defect detection method in the above-mentioned embodiments, the embodiments of the present application can provide a computer storage medium for implementation. The computer storage medium stores computer program instructions; the computer program instructions are executed by the processor to implement any one of the wind turbine blade layering process defect detection methods in the above-mentioned embodiments.

[0128] In summary, the wind turbine blade layering process defect detection method provided by the embodiments of the present application realizes accurate defect positioning and alarm by using an algorithm model to control the line laser pointing of the laser holder c, automatically completes defect recording and traceability report, and can define the detection opportunity to identify the specific semantic action of the operator as the start signal of detecting the layering defect, intelligently detects the defect according to the specific semantic action, and fills the gap of automatic detection of defects in the blade layering process.

[0129] Specifically, relying on the algorithm model of YOLOV5, the model is fully trained and optimized by collecting a large number of images on site, and the unreasonable state of the production material such as glass fiber cloth wrinkle, glass fiber cloth folding, glass fiber cloth loose yarn, missing block in the core material laying process, and foreign matter in the operation process are identified, judged, and fed back to achieve the purpose of recording and correcting the behavior of the on-site operator. The detection efficiency can reach seconds, and 24-hour uninterrupted duty is realized, reducing the missed detection of the inspector, effectively relieving the pressure of detecting defects, greatly improving the timeliness of detection, improving the labor efficiency to a certain extent, and having a guiding significance for realizing automation in the blade layering process. It provides technical reserves for realizing automatic production of blades.

[0130] In addition, the wind turbine blade layering process defect detection method, device, electronic equipment and medium provided by the present application can be applied to most of the operation steps in the wind turbine blade manufacturing process, such as pouring, bonding, post-processing and other processes. It only needs to collect defect images of the corresponding process to train a new model, which can complete the generalization of the model and identify defects in the wind turbine blade manufacturing process. For example, the wind turbine blade layering process defect detection device can be deployed on the mold of the shell and the web component to realize the detection function of the process defects by adjusting the number of industrial area array cameras a. In addition to relying on multiple industrial area array cameras a for fixed-point monitoring, the image acquisition operation can also use a drone equipped with 2-4 high-definition cameras to fly along the surface of the blade mold to collect images.

[0131] It is to be understood that the application is not limited to the particular configurations and processes described hereinabove and shown in the figures. For the sake of brevity, detailed descriptions of known methods and apparatuses are omitted so as not to obscure the description of the present application. In the above-described embodiments, several specific steps are described and illustrated as examples. However, the methods process of the present application is not limited to the specific steps described and illustrated, as various modifications, alterations, and permutations can be made with the benefit of this disclosure without departing from the spirit and scope of the present application, and the order of steps can so modified.

[0132] The functional blocks shown in the above structural block diagrams can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application specific integrated circuits (ASICs), appropriate firmware, plug-ins, functional cards, and the like. When implemented in software, the elements of the present application are program or code segments that are used to perform the required tasks. The program or code segments can be stored in a machine-readable medium or transmitted through a data signal carried in a carrier wave over a transmission medium or communication link. A "machine-readable medium" includes any medium that can store or transfer information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, and the like. The code segments can be downloaded via computer networks such as the Internet, intranets, and the like.

[0133] It is also to be understood that the example embodiments described in this application are based on a series of steps or apparatuses to describe some methods or systems. However, the present application is not limited to the order of the above steps, that is, the steps can be performed in the order mentioned in the embodiments, or in an order different from the embodiments, or several steps can be performed simultaneously.

[0134] The computer program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other processing device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other processing device to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks.

[0135] The above is merely specific implementation of the present application, and those skilled in the art can clearly understand the specific working process of the system, module and unit described above for the convenience and brevity of description, which can refer to the corresponding process in the foregoing method embodiments, and will not be described herein. It should be understood that the protection scope of the present application is not limited to this, and any person skilled in the art can easily think of various equivalent modifications or replacements within the technical range disclosed in the present application, and these modifications or replacements should be covered in the protection scope of the present application.

Claims

1. A method of wind turbine blade layup process defect detection, characterized in that, The method for a terminal device comprises: Collecting defect images in a specific semantic action and a laying process to generate first data, the specific semantic action being a stop rolling action behavior; Sending the first data to a server for the server to train a mature discrimination model based on the first data; After the first data sent reaches a preset numerical threshold, collecting a to-be-tested image in the laying process to generate second data; Sending the second data to the server for the server to identify the second data based on the mature discrimination model, and determining target defect data corresponding to the second data based on the mature discrimination model in the case that the mature discrimination model identifies a specific semantic action; Receiving target defect data sent by the server and respectively performing early warning and laser marking based on the target defect data.

2. A wind turbine blade layup process defect detection method according to claim 1, characterised in that, Before the collecting of the defect images in the specific semantic action and the laying process to generate the first data, the method further comprises: Determining target width information according to a focal length, a working distance and a target height of an industrial area array camera to be installed, the industrial area array camera being used to collect the defect images and the to-be-tested images; Based on the target width information, installing the industrial area array camera along the edge of a wind turbine blade main mold at a predetermined interval, wherein in a wind turbine blade root area, the industrial area array camera is provided with two groups and is symmetrically arranged on both sides of a laying work area, and in a wind turbine blade mid and tip area, the industrial area array camera is provided with one group and is arranged on one side of the laying work area.

3. A wind turbine blade layup process defect detection method according to claim 2, characterised in that, Before the receiving of the target defect data sent by the server and the performing of early warning and laser marking based on the target defect data, the method further comprises: Based on the target width information, installing a laser holder along an artificial channel on both sides of a wind turbine blade laying work area, the laser holder corresponding to the industrial area array camera one by one and being arranged at intervals, the laser holder being used to perform laser marking based on the target defect data to determine a defect position in a wind turbine blade laying process.

4. The wind turbine blade ply process defect detection method of claim 3, wherein, Before the receiving of the target defect data sent by the server and the performing of early warning and laser marking based on the target defect data, the method further comprises: Using the laser holder to project two endpoints on a wind turbine blade mold at will; Determining a target section covering the line connecting the two endpoints and measuring the length L of the line connecting the two endpoints; Using the industrial area array camera to shoot an image capable of covering the target section to determine third data, wherein the third data comprises the length L of the target section and the image covering the target section; Sending the third data to a server for the server to determine the number of pixel points of the target section in the third data based on the pixel coordinates of the two endpoints in the third data.

5. A method of wind turbine blade layup process defect detection, characterized in that, The method for a server comprises: Receiving first data sent by a terminal device, the first data comprising defect images in a specific semantic action and a laying process, the specific semantic action being a stop rolling action behavior; Performing preprocessing on the first data to determine labeled defect data; determining a mature identification model through training and optimization based on the specific semantic action and the labeled defect data; receiving second data sent by the terminal device, identifying the second data based on the mature identification model, and determining target defect data corresponding to the second data based on the mature identification model in a case where the mature identification model identifies a specific semantic action, wherein the target defect data includes defect position coordinate information and size information; sending the target defect data to the terminal device for the terminal device to perform early warning and laser marking based on the target defect data.

6. The wind turbine blade ply process defect detection method of claim 5, wherein, Before the target defect data is sent to the terminal device for the terminal device to perform early warning and laser marking based on the target defect data, the method further includes: receiving third data sent by the terminal device, and determining the number of pixel points of a target section in the third data based on pixel coordinates of two end points in the third data; determining an actual length d of a single pixel point mapping according to a length L of the target section and the number of pixel points of the target section; selecting a target defect image from the defect information according to a preset defect threshold, and determining a target rectangular frame corresponding to the target defect image and a center point coordinate of the target rectangular frame, wherein the preset defect threshold includes a preset aspect ratio threshold and a preset area threshold; calculating a to-be-identified coordinate based on the center point coordinate of the target rectangular frame and the actual length d of the single pixel point mapping, and determining target defect data.

7. A wind turbine blade ply process defect detection method according to claim 6, characterised in that, The preprocessing of the first data to determine labeled defect data includes: performing image cleaning on the first data to remove each of duplicate images, low-quality images, and non-standard images, and determining first-level intermediate data; performing flipping, brightness adjustment, and noise processing on the first-level intermediate data to determine second-level intermediate data; labeling the second-level intermediate data using an image labeling tool to determine labeled defect data, wherein the labeled defect data includes the type of defect and the rectangular frame coordinate of the defect.

8. The wind turbine blade ply process defect detection method of claim 6, wherein, The method of determining a mature identification model through training and optimization based on the specific semantic action and the labeled defect data includes: dividing the specific semantic action and the labeled defect data into a training set and a test set according to a predetermined proportion; training an initial identification model using a YOLOV5 algorithm based on the training set, and generating visual data, wherein the visual data includes training loss, validation loss, confusion matrix, precision, and recall.

9. The wind turbine blade ply process defect detection method of claim 8, wherein, After the initial identification model is trained using YOLOV5 based on the training set and the visual data is generated, the method further includes: testing the test set based on the initial identification model to determine accuracy and detection rate; adjusting hyperparameters of the initial identification model using a control variable method according to the accuracy and the detection rate of the test set to optimize the initial identification model to improve the detection rate, and determining a mature identification model, wherein the hyperparameters include a learning rate and a batch capacity.

10. A wind turbine blade layup process defect detection apparatus, characterized by, The device for a terminal device includes: The collection unit is used for collecting a specific semantic action and a defect image in a laying process to generate first data, the specific semantic action being a stop rolling action behavior; the first data is sent to a server, so that the server trains a mature identification model based on the first data; The collection module is used for collecting a to-be-tested image in a laying process to generate second data after the first data sent reaches a preset numerical threshold; The first sending module is used for sending the second data to the server, so that the server identifies the second data based on the mature identification model, and determines target defect data corresponding to the second data based on the mature identification model in a case where the mature identification model identifies a specific semantic action; The receiving module receives target defect data sent by the server, and performs early warning and laser marking based on the target defect data.

11. A wind turbine blade layup process defect detection apparatus, characterized by, For the server, the device comprises: The modeling unit is used for receiving first data sent by a terminal device, the first data including a specific semantic action and a defect image in a laying process, the specific semantic action being a stop rolling action behavior; the first data is preprocessed to determine labeled defect data; a mature identification model is determined through training and optimization based on the specific semantic action and the labeled defect data; The identification module is used for receiving second data sent by the terminal device, identifying the second data based on the mature identification model, and determining target defect data corresponding to the second data based on the mature identification model in a case where the mature identification model identifies a specific semantic action; The second sending module is used for sending target defect data to the terminal device, so that the terminal device performs early warning and laser marking based on the target defect data.

12. An electronic device, comprising: Comprise: A transceiver, a processor, a memory, and a program stored on the memory and executable on the processor, the program being executed by the processor to implement the wind power blade laying process defect detection method according to any one of claims 1 to 9.

13. A readable storage medium, characterized by, The readable storage medium stores a program or instructions, the program or instructions being executed by the processor to implement the steps of the method according to any one of claims 1 to 9.

Images