Method for motor rotor cabling defect detection using deep learning

Abstract

The application relates to a motor rotor parallel-wiring defect detection method using deep learning, and belongs to the technical field of deep learning. The method comprises the following steps: acquiring image acquisition components to perform image acquisition on each interval, obtaining at least two target images corresponding to each interval, inputting each target image into an identification model based on deep learning, obtaining a target region where at least one target interval in the target image is located, a target frame to which each target interval belongs, and a defect classification result of the target frame, and determining that the motor rotor has a parallel-wiring defect in the case that at least one target image has a defect classification result indicating that the distance between the two enameled wires is less than a preset distance. The problem that the detection effect of the rotor parallel-wiring defect is poor due to the fact that the commutator is small, the distance between the two enameled wires cannot be grasped accurately by manual operation, and the manual detection efficiency is low can be solved.

Description

TECHNICAL FIELD

[0001] The application relates to a method for detecting winding defects of a motor rotor by using deep learning, and belongs to the technical field of deep learning. BACKGROUND

[0002] The motor rotor is an important component of the motor. The motor rotor has a commutator at one end, the commutator is uniformly provided with protruding hooks on the periphery, an interval is formed between adjacent two hooks, the enameled wire of the motor rotor has an insulating coating, and the enameled wire is wound into the interval on one side of each hook and then wound out of the interval on the other side of the hook. At this time, there are two enameled wires in each groove. When the motor is working, the rotor drives the commutator to rotate at high speed. At this time, if the distance between the two enameled wires in the groove is too close, the enameled wires will rub against each other continuously, causing cracks or even explosion of the insulating coating. When the insulating layer explodes, the enameled wire will leak current, causing the motor to short circuit. Therefore, it is necessary to detect the winding defects of the motor rotor.

[0003] In the traditional method for detecting winding defects of the rotor, the two enameled wires in the interval of the commutator are usually identified manually to confirm whether there is a defect that the distance between the enameled wires is too close.

[0004] However, the small size of the commutator makes it difficult for manual detection to accurately grasp the distance between the enameled wires, and the manual detection efficiency is low, thus resulting in poor detection effect of the winding defects of the rotor. SUMMARY

[0005] The application provides a method for detecting winding defects of a motor rotor by using deep learning, which can solve the problem that the small size of the commutator makes it difficult for manual detection to accurately grasp the distance between the enameled wires, and the manual detection efficiency is low, thus resulting in poor detection effect of the winding defects of the rotor. The application provides the following technical scheme:

[0006] In a first aspect, a method for detecting winding defects of a motor rotor by using deep learning is provided. The motor rotor to be detected is placed on a detection position of a defect detection table, and an image acquisition assembly located on the defect detection table is adapted to acquire images of the detection position. One end of the motor rotor has a commutator, the commutator is uniformly provided with at least two protruding hooks on the periphery, an interval is formed between adjacent two hooks, and the enameled wire of the motor rotor is wound into the interval on one side of each hook and then wound out of the interval on the other side of the hook. The method comprises the following steps:

[0007] Acquiring images of each interval by the image acquisition assembly to obtain at least two target images corresponding to each interval; wherein each target image includes image data of at least one interval and two enameled wires in the interval;

[0008] input each target image into a recognition model based on deep learning, to obtain a target region where at least one target interval in the target image is located, a target box to which each target interval belongs, and a defect classification result of the target box; the target interval is an interval in the at least one interval that meets the recognition standard of the model, and the position of the target box is determined based on the positions of the two enameled wires in the target interval;

[0009] In the at least two target images, if the defect classification result corresponding to at least one target image indicates that the distance between the two enameled wires is less than a preset distance, it is determined that the motor rotor has the parallel defect.

[0010] Optionally, the training process of the recognition model comprises:

[0011] acquiring image acquisition of each interval by the image acquisition assembly, to obtain at least two training images corresponding to each interval;

[0012] adding a target region label, a target box label, and a defect classification label corresponding to each target box to each training image by using an image processing software, to obtain a defect classification training set;

[0013] training a pre-created neural network model by using the defect classification training set, to obtain the recognition model, and the neural network model is established based on YOLOv5.

[0014] Optionally, the training of the pre-created neural network model by using the defect classification training set to obtain the recognition model comprises:

[0015] In the training process, a deep learning visualization tool is used to monitor the training process of the recognition model, to obtain training values corresponding to the training process, and the training values comprise recall rate, precision rate, and average precision mean;

[0016] In the case where the training values are all greater than or equal to corresponding preset standard values, the recognition model obtained by this training is output.

[0017] Optionally, before the input of each target image into the recognition model based on deep learning, to obtain a target region where at least one target interval in the target image is located, a target box to which each target interval belongs, and a defect classification result of the target box, the method further comprises:

[0018] In the case where the model format of the recognition model does not match the processing format of an image processing library, the model format is converted into the processing format; the image processing library comprises an open source computer vision library OpenCV or HALCON.

[0019] Optionally, in the case that the defect classification result corresponding to at least one of the at least two target images indicates that the distance between the two enameled wires is less than the preset distance, it is determined that the motor rotor has the overlapping defect.

[0020] The first confidence of each target frame in the target area output by the recognition model is obtained.

[0021] It is determined whether the first confidence of the target frame is less than a preset confidence.

[0022] The target frame and the defect classification result corresponding to the target frame, whose first confidence is less than the preset confidence, are deleted.

[0023] In the case that the defect classification result corresponding to at least one of the at least one screened target frame indicates that the distance between the two enameled wires is less than the preset distance, it is determined that the motor rotor has the overlapping defect.

[0024] Optionally, in the case that the defect classification result corresponding to at least one of the at least two target images indicates that the distance between the two enameled wires is less than the preset distance, it is determined that the motor rotor has the overlapping defect, comprising:

[0025] The first confidence of each target frame in the target area output by the recognition model is obtained.

[0026] The reference target frame with the highest first confidence in the target area is determined.

[0027] The intersection-over-union between the other target frames in the target area and the reference target frame, except for the reference target frame, is calculated.

[0028] The other target frames with the intersection-over-union greater than or equal to an intersection-over-union threshold are deleted.

[0029] In the case that the defect classification result corresponding to at least one of the at least one deleted target frame indicates that the distance between the two enameled wires is less than the first preset distance, it is determined that the motor rotor has the overlapping defect.

[0030] Optionally, the defect detection station is further provided with an illumination assembly, and the illumination assembly is adapted to provide auxiliary illumination for the intervals.

[0031] Before the image acquisition assembly acquires images of each interval to obtain at least two target images corresponding to each interval, the method further comprises:

[0032] The illumination assembly is controlled to start.

[0033] The image acquisition assembly is used to acquire images of each interval after illumination.

[0034] Optionally, the defect detection station further comprises a distance measuring assembly adapted to measure the distance between the distance measuring assembly and the commutator.

[0035] Before the step of obtaining at least two target images corresponding to each interval by the image acquisition assembly, the method further comprises:

[0036] obtaining the distance value collected by the distance measuring assembly;

[0037] determining whether the image acquisition assembly is opposite to the interval by using the distance value; collecting a first distance value by the distance measuring assembly in the case that the image acquisition assembly is opposite to the interval, and collecting a second distance value by the distance measuring assembly in the case that the image acquisition assembly is opposite to the hook;

[0038] in the case that the image acquisition assembly is opposite to the interval, collecting images of the interval by using the image acquisition assembly.

[0039] Optionally, the defect detection station comprises a rotating driving member adapted to drive the detection position to rotate, and the rotating shaft of the detection position is the axis of the motor rotor when the motor rotor is placed on the detection position.

[0040] Before the step of obtaining the distance value collected by the distance measuring assembly, the method further comprises: controlling the rotating driving member to drive the detection position to rotate continuously, and controlling the distance measuring assembly to collect the distance value every preset time interval, so as to trigger the step of obtaining the distance value collected by the distance measuring assembly; determining whether the image acquisition assembly is opposite to the interval by using the distance value; and in the case that the image acquisition assembly is opposite to the interval, collecting images of the interval by using the image acquisition assembly.

[0041] Alternatively,

[0042] The method further comprises: in the case that the image acquisition assembly is opposite to the hook, controlling the rotating driving member to drive the detection position to rotate by a preset angle; and controlling the distance measuring assembly to collect the distance value, so as to trigger the step of obtaining the distance value collected by the distance measuring assembly again; determining whether the image acquisition assembly is opposite to the interval by using the distance value; and in the case that the image acquisition assembly is opposite to the interval, collecting images of the interval by using the image acquisition assembly. The preset angle is less than the included angle formed by the image acquisition assembly and the adjacent two hooks.

[0043] Optionally, after determining that the motor rotor has the parallel defect in the case that at least one target image corresponding to the at least two target images indicates that the distance between the two enameled wires is less than a preset distance threshold, the method further comprises:

[0044] sending a gripping instruction to the gripping device to move the motor rotor from the detection position to a target position in response to the gripping instruction, the target position being used to store the motor rotor with the doubling defect

[0045] The beneficial effects of the present application at least include: obtaining at least two target images corresponding to each interval by acquiring image acquisition components to image each interval, wherein each target image includes image data of at least one interval and two enameled wires in the interval, inputting each target image into an identification model based on deep learning to obtain a target region where at least one target interval in the target image is located, a target frame to which each target interval belongs, and a defect classification result of the target frame, the target interval being an interval in the at least one interval that meets the identification standard of the model, the position of the target frame being determined based on the positions of the two enameled wires in the target interval, and in at least two target images, there is a case that the defect classification result corresponding to at least one target image indicates that the distance between the two enameled wires is less than a preset distance, and it is determined that the motor rotor has a doubling defect; the problem of poor detection effect of the doubling defect of the rotor due to the fact that the commutator is small and the distance between the two enameled wires is not grasped accurately by manual detection, and the low efficiency of manual detection can be solved; the efficiency of using the identification model obtained by deep learning to identify the target image is high, and the model is more accurate in grasping whether the distance between the two enameled wires in the target image is less than the preset distance, so the detection effect of the doubling defect of the motor rotor can be improved.

[0046] In addition, since only the target frame exists in the target region, and the target frame only exists in the target region, the neural network model is trained using the target region label, the target frame label, and the defect classification label corresponding to each target frame, which can verify each label with each other and improve the accuracy of the identification after training.

[0047] In addition, in the case that the training value is greater than or equal to the corresponding preset standard value, the identification model obtained by this training is output, which can ensure the performance of the identification model obtained after training, thereby improving the identification effect of the identification model.

[0048] In addition, converting the model format into the processing format of the image processing library can load the identification model using the image processing library; and using the image processing library to load the converted identification model can assist the image processing of the target image by the image processing library, thereby improving the processing speed of the identification model.

[0049] In addition, since the lower the first confidence of the target frame, the higher the false detection rate, by deleting the target frame and the defect classification result corresponding to the target frame whose first confidence is less than the preset confidence, the false detection rate can be reduced, thereby improving the accuracy of the identification model.

[0050] In addition, the first reference target frame with the highest confidence in the target region is determined, the intersection-over-union between the other target frames in the target region and the reference target frame is calculated, and the other target frames with the intersection-over-union greater than or equal to the intersection-over-union threshold are deleted, so as to ensure that each target interval corresponds to a target frame and obtain a target frame most matched with the target interval, thereby improving the recognition effect of the recognition model.

[0051] In addition, the image acquisition component is used to acquire images of each interval after illumination, so as to avoid the influence of ambient light on the recognition model and improve the applicability of the recognition model.

[0052] In addition, the distance value is used to determine whether the image acquisition component is opposite to the interval, and the image acquisition component is used to acquire images of the interval in the case that the image acquisition component is opposite to the interval, so as to avoid the case that the interval exists in the edge of the target image and the recognition model cannot recognize the interval, thereby improving the recognition effect of the recognition model.

[0053] The above description is only a summary of the technical scheme of the present application. In order to more clearly understand the technical means of the present application and can be implemented according to the content of the description, the following will be described in detail with the preferred embodiments of the present application and with the help of the accompanying drawings.

DRAWINGS

[0054] Figure 1 is a schematic diagram of a system for detecting defects of parallel wires of a motor rotor using deep learning provided by an embodiment of the present application;

[0055] Figure 2 is a schematic diagram of a motor rotor provided by an embodiment of the present application;

[0056] Figure 3 is a schematic diagram of two enameled wires in an interval provided by an embodiment of the present application;

[0057] Figure 4 is a schematic diagram of an interval provided by an embodiment of the present application;

[0058] Figure 5 is a flowchart of a method for detecting defects of parallel wires of a motor rotor using deep learning provided by an embodiment of the present application;

[0059] Figure 6 is a schematic diagram of adding a target frame label and a defect classification label corresponding to the target frame using labelme software provided by an embodiment of the present application;

[0060] Figure 7 is a curve diagram of training values output by a model training process provided by an embodiment of the present application;

[0061] Figure 8 is a result image output by the target image processed by the OpenCV loaded recognition model according to an embodiment of the present application;

[0062] Figure 9 is a schematic diagram of the detection result of the first station according to an embodiment of the present application;

[0063] Figure 10 is a schematic diagram of the detection result of the second station according to an embodiment of the present application.

DETAILED DESCRIPTION

[0064] The specific embodiments of the present application are described in detail below with reference to the accompanying drawings and examples. The following examples are used to illustrate the present application, but are not used to limit the scope of the present application.

[0065] Figure 1 is a schematic diagram of a system for motor rotor wire defect detection using deep learning according to an embodiment of the present application. The motor refers to an electromagnetic device that realizes electric energy conversion or transmission according to the electromagnetic induction law. The motor is composed of a motor rotor 2 and a motor stator.

[0066] As shown in Figure 1 , the system for motor rotor 2 wire defect detection using deep learning at least includes a defect detection table 1 and an image acquisition assembly 12.

[0067] The defect detection table 1 is suitable for carrying the motor rotor 2.

[0068] The motor rotor 2 is a rotating part in the motor, and one end of the motor rotor 2 usually has a commutator, and the commutator is uniformly and circumferentially provided with at least two hooks, and a gap is formed between adjacent two hooks.

[0069] Among them, the gap on the commutator is also called a hook groove. The gap between the adjacent two hooks can be a concave groove, or a part of the circumferential surface of the commutator clamped between the two hooks, and the present embodiment does not limit the implementation manner of the gap.

[0070] As shown in Figure 2 , one side of the motor rotor 2 has a commutator 20, and the commutator 20 is uniformly and circumferentially provided with protruding hooks 201 close to one side of the rotor body 21.

[0071] The enameled wire of the motor rotor 2 is wound into the gap on one side of each hook and wound out from the gap on the other side of the hook. At this time, there are two enameled wires in each gap.

[0072] The enameled wire refers to the winding wire wound on the rotor body of the motor rotor 2 for current communication. The surface of the enameled wire has an insulating coating to avoid mutual conduction between the enameled wires after being electrified, causing short circuit of the motor.

[0073] As shown in Figure 3 The enameled wire 31 is wound from the interval on the left side of the hook 32 and then wound from the interval on the right side of the hook 32. At this time, there are two enameled wires 31 in each interval.

[0074] The defect detection platform 1 is also provided with an image acquisition assembly 12. The image acquisition assembly 12 is adapted to acquire images of the intervals to obtain at least two target images corresponding to each interval. Each target image includes image data of at least one interval and two enameled wires in the interval.

[0075] The side of the commutator with the hook is close to the rotor body. As shown in Figure 4 Since the plane where the interval 41 is located is inclined to the center axis 42 of the commutator 20, the interval 41 cannot be seen from the other side of the commutator 20 along the direction parallel to the center axis 42. Therefore, if you want to obtain the image data of all intervals and two enameled wires in the intervals on the commutator, you need to use the image acquisition assembly 12 to acquire at least two target images.

[0076] Since the plane where the interval is located is inclined to the center axis, the rotor body will block the interval. Therefore, the position of the image acquisition assembly 12 needs to be adjusted.

[0077] Specifically, the image acquisition assembly 12 is installed at a position corresponding to the position on the defect detection platform 1 on the right side of the center axis, so that the image acquisition assembly 12 is opposite to the plane where the interval is located. At this time, the interval will not be blocked by the rotor body.

[0078] In this embodiment, the image acquisition assembly 12 also has an identification function. The image acquisition assembly 12 identifies the enameled wire in the target image and obtains a defect classification result corresponding to the target image.

[0079] The defect classification result is used to indicate whether the distance between the two enameled wires in the same interval is less than a preset distance.

[0080] The clamping device is a device for moving the motor rotor 2 to different positions in a system for detecting the parallel wire defects of the motor rotor 2 using deep learning.

[0081] The clamping device is in communication with the image acquisition assembly 12, so as to move the motor rotor 2 to different positions according to the defect classification result transmitted by the image acquisition assembly 12.

[0082] The image acquisition component 12 has a large acquisition range, and the large bearing area of the defect detection platform 1 leads to a small resolution of the target image acquired by the image acquisition component 12, so that the image acquisition component 12 cannot accurately identify the interval and the enameled wire in the interval.

[0083] Optionally, the defect detection platform 1 further has a preset detection position 11. The motor rotor 2 to be detected is placed on the detection position 11 of the defect detection platform 1, and the image acquisition component 12 is used to acquire images of the detection position 11 to obtain a target image.

[0084] Optionally, the defect detection platform 1 further comprises a rotating driving component, and the rotating driving component is adapted to drive the detection position 11 to rotate, and the rotating shaft of the detection position 11 is the axis of the motor rotor 2 placed on the detection position 11.

[0085] The axis of the motor rotor 2 is the same as the axis of the center shaft of the commutator, that is, the rotating shaft of the detection position 11 is the same as the center shaft of the commutator.

[0086] In the embodiment, the rotating driving component has two working modes:

[0087] Firstly, the rotating driving component continuously rotates.

[0088] Secondly, the rotating driving component rotates by a preset acquisition angle every preset time interval. The preset time interval and the preset acquisition angle can be pre-stored in the device or input by the user, and the embodiment does not limit the implementation of the preset time interval and the preset acquisition angle.

[0089] Optionally, the defect detection platform 1 further comprises an illuminating component 13. The illuminating component 13 is adapted to provide auxiliary illumination for the interval to reduce the influence of ambient light on the target image acquired by the image acquisition component 12.

[0090] When the rotating driving component drives the detection position 11 to rotate and the image acquisition component 12 acquires images of the interval, the hook may be opposite to the image acquisition component 12, and in this case, the image data of the interval and the enameled wire in the interval in the acquired target image may not meet the identification standard, so that the interval and the enameled wire in the interval cannot be identified.

[0091] The identification standard not met refers to that the image data of the two enameled wires in the interval has noise affecting identification, image distortion occurs, or the interval region is not completely acquired.

[0092] Optionally, the defect detection platform 1 further comprises a distance measuring component 120. The distance measuring component 120 is adapted to measure the distance between the distance measuring component 120 and the commutator.

[0093] The ranging assembly 120 includes, but is not limited to, a fiber sensor, a radar sensor, and the like.

[0094] The ranging assembly 120 is located in the same plane as the image acquisition assembly 12, so the distance between the ranging assembly 120 and the commutator can be approximated as the distance between the image acquisition assembly 12 and the commutator.

[0095] Due to the height difference between the interval and the hook on the commutator, the distance between the hook and the ranging assembly 120 is less than the distance between the interval and the ranging assembly 120, so using the ranging assembly 120 to acquire the distance between the ranging assembly 120 and the commutator can determine whether the interval is opposite to the image acquisition assembly 12.

[0096] When the motor starts to work, the motor rotor 2 rotates at high speed. If the distance between the two enameled wires in the interval is less than the preset distance, i.e., the motor rotor 2 has a parallel defect, the two enameled wires will rub against each other during the high-speed rotation of the motor rotor 2, causing the insulation layer on the enameled wire to crack or even burst, resulting in a short circuit of the motor.

[0097] In actual implementation, the preset distance can be set according to different needs, so the preset distances used by different users are the same or different.

[0098] In the traditional rotor parallel defect detection method, the two enameled wires in the commutator interval are usually identified manually to confirm whether there is a defect of too close distance between the enameled wires.

[0099] However, the small size of the commutator makes it difficult for manual detection to accurately grasp the distance between the enameled wires, and manual detection is low in efficiency, thus resulting in poor detection effect of the rotor parallel defect.

[0100] To solve the above problems, in the embodiment, the image acquisition assembly 12 is configured to: acquire images of each interval by the image acquisition assembly 12 to obtain at least two target images corresponding to each interval; input each target image into an identification model based on deep learning to obtain a target region where at least one target interval in the target image is located, a target box to which each target interval belongs, and a defect classification result of the target box; and determine that the motor rotor 2 has a parallel defect in a case where at least one target image has a defect classification result indicating that the distance between the two enameled wires is less than a preset distance.

[0101] The target interval is an interval in the at least one interval that meets the identification standard of the model, and the position of the target box is determined based on the position of the two enameled wires in the target interval.

[0102] In this embodiment, by acquiring image collection components to collect images of each interval, at least two target images corresponding to each interval are obtained, wherein each target image includes image data of at least one interval and two enameled wires in the interval, each target image is input into an identification model based on deep learning to obtain a target area where at least one target interval is located, a target frame to which each target interval belongs, and a defect classification result of the target frame, the target interval is an interval in the at least one interval that meets the identification standard of the model, the position of the target frame is determined based on the positions of the two enameled wires in the target interval, and in at least two target images, there is at least one target image corresponding to a defect classification result indicating that the distance between the two enameled wires is less than a preset distance, it is determined that the motor rotor has the parallel line defect; the problem that the detection effect of the rotor parallel line defect is poor due to the fact that the commutator is small and the distance between the two enameled wires is not grasped accurately by manual detection, and the low efficiency of manual detection can be solved; the efficiency of using the identification model obtained through deep learning to identify the target image is high, and the model grasps whether the distance between the two enameled wires in the target image is less than the preset distance accurately, so that the detection effect of the motor rotor parallel line defect can be improved.

[0103] Next, the method for detecting the parallel line defect of the motor rotor using deep learning provided by the present application will be described in detail. The following embodiments are described by taking the case of the processor in the electronic device connected in communication with the image collection component as an example, or other devices connected in communication with the image collection component, such as: a user terminal or a server, etc., wherein the user terminal includes but is not limited to: a mobile phone, a tablet computer, a wearable device, or an industrial control computer, etc. Figure 1

[0104] The communication connection mode can be wired communication or wireless communication, and the wireless communication mode can be short-range communication or wireless communication, etc., and the present embodiment does not limit the communication mode between the mobile device and other devices.

[0105] Specifically, the communication connection mode includes Modbus protocol or Controller Area Network (CAN) protocol, etc.

[0106] In actual implementation, the method can also be applied to the image collection component, and the present embodiment does not limit the implementation mode of other devices and the implementation mode of the user terminal.

[0107] Figure 5 is a flowchart of the method for detecting the parallel line defect of the motor rotor using deep learning provided by an embodiment of the present application. The method includes at least the following steps: ​

[0108] In step 501, the image acquisition component acquires images of each interval to obtain at least two target images corresponding to each interval.

[0109] Illustratively, the number of target images is related to the number of target intervals acquired each time on the target image and the number of intervals on the motor rotor.

[0110] For example, the motor rotor has 26 intervals, and each target image acquires 2 target intervals, so there are 13 target images.

[0111] Among them, the target interval is at least one interval that meets the identification criteria of the model.

[0112] Because when the image acquisition component acquires images of the motor rotor, ambient light will cause uncontrollable noise in the target image, affecting the recognition effect of the recognition model.

[0113] Optionally, before acquiring the at least two target images corresponding to each interval by the image acquisition component, the method further comprises: controlling the lighting component to start; and using the image acquisition component to acquire images of each interval after lighting.

[0114] By using the lighting component to assist lighting, it is ensured that the ambient light in the target image is consistent, thereby excluding the interference of ambient light on the recognition model and improving the recognition effect of the recognition model.

[0115] Because the acquisition range of the image acquisition component is limited, and the image acquisition component acquires the circumferential side of the motor rotor, when the hook is opposite to the image acquisition component, the interval may be at the edge of the target image, resulting in noise, distortion, or possible incomplete acquisition of the interval and the image data in the interval; and because the image acquisition component continues to acquire the next target image after a certain preset acquisition time interval, in the case of misalignment between the first target image and the interval, the subsequent acquired target images may all be misaligned with the interval, resulting in the inability to detect the end-to-end defects of the motor rotor corresponding to the target image. Therefore, before the image acquisition component acquires the target image, it is necessary to determine whether the image acquisition component is opposite to the interval.

[0116] Optionally, before acquiring the at least two target images corresponding to each interval by the image acquisition component, the method further comprises: acquiring a distance value collected by the distance measuring component; using the distance value to determine whether the image acquisition component is opposite to the interval; the first distance value collected by the distance measuring component when the image acquisition component is opposite to the interval is greater than the second distance value collected by the distance measuring component when the image acquisition component is opposite to the hook; and in the case that the image acquisition component is opposite to the interval, using the image acquisition component to acquire images of the interval.

[0117] Since the ranging assembly and the image acquisition assembly are located in the same plane, the distance between the ranging assembly and the commutator can be approximated to the distance between the image acquisition assembly and the commutator.

[0118] Since there is a height difference between the interval and the hook on the commutator, the second distance value between the hook and the ranging assembly is smaller than the first distance value between the interval and the ranging assembly, so using the ranging assembly to collect the distance value between the ranging assembly and the commutator can determine whether the image acquisition assembly is opposite to the interval.

[0119] Since the rotary drive assembly has different working modes, using the ranging assembly to determine whether the image acquisition assembly is opposite to the interval to collect the target image also includes but is not limited to the following two ways.

[0120] First, before obtaining the distance value collected by the ranging assembly, it further includes: controlling the rotary drive to drive the detection position to rotate continuously, and controlling the ranging assembly to collect distance values every preset time interval, so as to trigger the execution of obtaining the distance value collected by the ranging assembly; using the distance value to determine whether the image acquisition assembly is opposite to the interval; in the case that the image acquisition assembly is opposite to the interval, using the image acquisition assembly to collect images of the interval.

[0121] Second, before obtaining the distance value collected by the ranging assembly, it further includes: in the case that the image acquisition assembly is opposite to the hook, controlling the rotary drive to drive the detection position to rotate by a preset angle; and controlling the ranging assembly to collect distance values, so as to trigger the execution of obtaining the distance value collected by the ranging assembly again; using the distance value to determine whether the image acquisition assembly is opposite to the interval; in the case that the image acquisition assembly is opposite to the interval, using the image acquisition assembly to collect images of the interval.

[0122] Wherein, the preset angle is smaller than the included angle formed by the adjacent two hooks and the image acquisition assembly, which can avoid the case that other hooks are opposite to the image acquisition assembly after the motor rotor rotates.

[0123] In the case that the image acquisition assembly is opposite to the interval, the target image is collected, which can determine that the interval is within the collection range of the image acquisition assembly, so as to avoid the target interval appearing at the edge of the target image, thereby improving the recognition effect of the recognition model.

[0124] Step 502, input each target image into the recognition model based on deep learning to obtain a target region where at least one target interval in the target image is located, a target box to which each target interval belongs, and a defect classification result of the target box.

[0125] Wherein, the target region refers to a region containing all target intervals in the target image, and the position of the target box is determined based on the position of the two enameled wires in the target interval.

[0126] Before using the recognition model, a pre-created neural network model needs to be trained by deep learning to obtain the recognition model.

[0127] Optionally, the training process of the recognition model comprises: acquiring image collection of each interval by the image collection component to obtain at least two training images corresponding to each interval; adding target region labels, target frame labels, and defect classification labels corresponding to each target frame to each training image by using image processing software to obtain a defect classification training set; training the pre-created neural network model by using the defect classification training set to obtain the recognition model, and the neural network model is established based on YOLOv5.

[0128] The image processing software comprises but is not limited to labelme or labelimg, etc.

[0129] For example, as shown in FIG. 6, labelme is used to add target frame labels 61 and defect classification labels 61 corresponding to the target frame 60 to the training image. Figure 6

[0130] At this time, the defect classification training set obtained after adding labels to the training image by using labelme is in “json” format, while the neural network model created based on YOLOv5 can only process data in “txt” format; therefore, it is still necessary to convert the format of the defect classification training set to a defect classification training set in “txt” format by using a Python script, so as to train the neural network model created based on YOLOv5 by using the converted defect classification training set.

[0131] Among them, the defect classification label 61 “ng” means that the distance between the two enameled wires in the target frame 60 is less than the preset distance; and the defect classification label 63 “ok” means that the distance between the two enameled wires in the target frame 62 is greater than the preset distance.

[0132] Since there is only a target frame in the target region, and the target frame only exists in the target region, training the model by using the target region label, the target frame label and the defect classification label corresponding to the target frame at the same time can make each label verify each other, and improve the accuracy of the recognition model obtained after training.

[0133] ​The pre-created neural network model can be a region-based convolutional neural network (R-CNN), a spatial pyramid pooling network (SPPNet), or a you only look once (YOLO) series.

[0134] The YOLO series includes YOLOv3, YOLOv4, YOLOv5, etc.

[0135] Compared with other neural network models, YOLOv5 can perform mosaic enhancement on the training image, so that the detection effect of small target objects is improved.

[0136] Mosaic enhancement refers to splicing four images in a random scaling, random cropping, and random distribution manner. Random scaling can improve the detection effect of small target objects.

[0137] The interval of the motor rotor and the two enameled wires in the interval are small target objects, so using YOLOv5 can improve the detection effect of the target interval.

[0138] During the training of the neural network model, the user only knows the input defect classification training set and the output defect classification training result, and cannot know the training process. When the model output result does not meet the user's expectation, the user needs to spend a lot of time to find the reason why the model deviates from the expectation, thereby affecting the efficiency of the model training.

[0139] Optionally, the pre-created neural network model is trained using the defect classification training set to obtain an identification model, including: in the training process, using a deep learning visualization tool to monitor the training process of the identification model to obtain training values corresponding to the training process, the training values including recall rate, precision rate, and average precision mean; in the case that the training values are all greater than or equal to corresponding preset standard values, outputting the identification model obtained by the present training.

[0140] The deep learning visualization tool includes but is not limited to a Weights & Biases (wandb) visualization tool or a TensorBoard tool, etc.

[0141] For example, wandb can record the changes of various indicators and the settings of hyperparameters in the model training process, and can also visualize the comparison of the output results, so as to help the user analyze the problems existing in the model training process and improve the efficiency of the model training.

[0142] Illustratively, monitoring the training process of the identification model using a deep learning visualization tool obtains training values corresponding to the training process, including: using a deep learning visualization tool to obtain a confusion matrix of the model, and obtaining training values corresponding to the training process through the confusion matrix.

[0143] The confusion matrix is an analysis table that summarizes the prediction results of a model in data analysis and machine learning, and summarizes the records in the training data set according to the two standards of the true class and the classification judgment made by the classification model in the form of a matrix.

[0144] Taking the confusion matrix corresponding to a binary classification model as an example, as shown in Table 1:

[0145] Table 1:

[0146]

[0147] Among them, TP refers to the number of samples predicted as positive samples among the samples actually positive, that is, the number of samples predicted correctly among the positive samples; FN refers to the number of samples predicted as negative samples among the samples actually positive, that is, the number of samples predicted incorrectly among the positive samples; TN refers to the number of samples predicted as negative samples among the samples actually negative, that is, the number of samples predicted correctly among the negative samples; FP refers to the number of samples predicted as positive samples among the samples actually negative, that is, the number of samples predicted incorrectly among the negative samples.

[0148] Recall, also known as recall rate, refers to the probability of being predicted as a positive sample among the samples actually positive. Its formula is as follows:

[0149]

[0150] Precision, also known as precision rate, refers to the probability of being actually positive among all samples predicted as positive, that is, the probability of being predicted correctly among the results predicted as positive. Its formula is as follows:

[0151]

[0152] In practical application, recall rate and precision rate correspond one-to-one and are negatively correlated.

[0153] The average precision mean (mean average precision, mAP) is the average of the average precision of all classes.

[0154] Among them, the average precision (average precision, AP) is the average of the precision corresponding to each recall in each class.

[0155] For example, the defect classification training set is input into a neural network model for training. Based on the curve diagram obtained by the visualization tool wanbd, as shown in Figure 7 The curve diagram is established with recall as the horizontal axis and precision as the vertical axis. The positive sample is the target box corresponding to the defect classification label "ok". The negative sample is the target box corresponding to the defect classification label "ng". When the Intersection over Union (iou) value is greater than 0.5 and the defect classification labels are the same, it is determined that the recognized defect classification result is correct.

[0156] The iou refers to the ratio of the intersection of the two target boxes to the union of the two target boxes. In other embodiments, the iou value can also be 0.7, and the present embodiment does not limit the implementation of iou.

[0157] As shown in Figure 7 The category "ok" refers to the sample recognized as a positive sample in the actual positive sample, and the corresponding AP is 0.988. The category "ng" refers to the sample recognized as a negative sample in the actual negative sample, and the corresponding AP is 0.883. The mAP of all categories is 0.910.

[0158] The recognition model needs to recognize the target image, and the typical image processing library can assist the recognition model to process the target image, thereby improving the processing speed of the recognition model on the target image. Therefore, the image processing library can be used to load the recognition model before the image is processed using the recognition model.

[0159] Optionally, before the target image is input into the recognition model based on deep learning to obtain the target region where at least one target interval in the target image is located, the target box to which each target interval belongs, and the defect classification result of the target box, the method further includes: converting the model format into the processing format in a case where the model format of the recognition model does not match the processing format of the image processing library; and the image processing library includes: an open source computer vision library OpenCV or HALCON.

[0160] Specifically, the OpenCV is used to load the recognition model. When the recognition model is trained by YOLOv5, the format of the recognition model is ".py", and the OpenCV cannot load the ".py" model. At this time, the export.py script is used to convert the recognition model into the processing format of the OpenCV: Onnx format. At this time, the OpenCV can load the converted recognition model, and the CUDA module of the OpenCV is used to accelerate the data processing process of the converted recognition model, thereby improving the data processing speed of the converted recognition model.

[0161] For example, the recognition result obtained by using the OpenCV to load the converted recognition model is as shown inFigure 8 As shown, a total of 155 target intervals and two enameled wires within each interval were detected. Among them, 67 defect classification results indicated that the distance between the two enameled wires was less than a preset distance. At this time, two target bounding boxes 82 exist within the target region 81 of the target image loaded in OpenCV.

[0162] Step 503: If, in at least one of the at least two target images, the defect classification result indicates that the distance between the two enameled wires is less than a preset distance, then it is determined that the motor rotor has a parallel wire defect.

[0163] Each motor rotor corresponds to at least two target images. When the defect classification result of one of the target images indicates that the distance between the two enameled wires is less than a preset distance, it indicates that the motor rotor has a parallel wire defect. At this time, the location of the parallel wire defect on the motor rotor can be recorded.

[0164] Optionally, if at least one of the target images indicates that the distance between the two enameled wires is less than a preset distance, determining that the motor rotor has a parallel fault includes: numbering the at least two target images according to the acquisition time; determining that the motor rotor has a parallel fault if the defect classification result of at least one target image indicates that the distance between the two enameled wires is less than a preset distance, and outputting the number of the target image and the defect coordinates of the target box corresponding to the defect classification result.

[0165] The defect coordinates can be the coordinates of the center point of the target bounding box, the width of the target bounding box, and the height of the target bounding box; the defect coordinates can also be the coordinates of the four vertices of the target bounding box. This embodiment does not limit the implementation method of the defect coordinates.

[0166] Since the target image is captured by capturing images of the intervals located around the motor rotor, there may be intervals with poor shooting angles at the edges of the target image. However, if these intervals meet the model's recognition criteria, the recognition model will still recognize them. In this case, the model's false detection rate is high.

[0167] Optionally, if at least one of the target images corresponds to a defect classification result indicating that the distance between the two enameled wires is less than a preset distance, it is determined that the motor rotor has a parallel fault, including: obtaining the first confidence score of each target box within the target area output by the recognition model; determining whether the first confidence score of the target box is less than a preset confidence score; deleting the target box and the defect classification result corresponding to the target box whose first confidence score is less than the preset confidence score; and determining that the motor rotor has a parallel fault if at least one of the filtered target boxes corresponds to a defect classification result indicating that the distance between the two enameled wires is less than a preset distance.

[0168] The identification model also outputs a first confidence corresponding to the target box when outputting the target box. The higher the first confidence of the target box, the lower the false detection rate, and the first confidence of the target box located at the edge of the target image is usually low. Therefore, screening the target box through the first confidence can reduce the false detection rate and improve the identification effect of the model.

[0169] Since the same target interval in the same target image can correspond to multiple target boxes that overlap with each other, and only one target box needs to be obtained for one target interval, the target boxes can be screened.

[0170] Optionally, in a case where the defect classification result corresponding to at least one of the at least two target images indicates that the distance between the two enameled wires is less than the preset distance, it is determined that the motor rotor has the overlapping defect, comprising: obtaining the first confidence of each target box in the target area output by the identification model; determining a reference target box with the highest first confidence in the target area; calculating the intersection-over-union between the reference target box and other target boxes in the target area except the reference target box; deleting other target boxes with an intersection-over-union greater than or equal to an intersection-over-union threshold; and in a case where the defect classification result corresponding to at least one of the deleted target boxes indicates that the distance between the two enameled wires is less than the preset distance, determining that the motor rotor has the overlapping defect.

[0171] Specifically, the other target boxes with an intersection-over-union less than or equal to the intersection-over-union threshold with the reference target box with the highest first confidence can be deleted by using a Non-Maximum Suppression (NMS) algorithm.

[0172] The intersection-over-union threshold can be pre-stored in the image acquisition component or input by a user, and the embodiment does not limit the acquisition method of the intersection-over-union threshold.

[0173] Since the target intervals do not overlap with each other, the target boxes corresponding to the target intervals also do not have a large overlapping area. Screening through the first confidence and the intersection-over-union can obtain the target box that best matches the target interval, thereby improving the identification effect of the identification model.

[0174] After detecting the overlapping defect of the motor rotor, the motor rotor with the overlapping defect needs to be separated out so as to analyze the defect cause or repeatedly inspect the motor rotor with the overlapping defect in the future.

[0175] Optionally, after determining that the motor rotor has the overlapping defect in a case where the defect classification result corresponding to at least one of the at least two target images indicates that the distance between the two enameled wires is less than the preset distance threshold, the method further comprises: sending a clamping instruction to the clamping device to move the motor rotor from the detection position to the target position in response to the clamping instruction.

[0176] The target position is used to store the motor rotor with the lapping defect.

[0177] Illustratively, the method of using deep learning to detect the lapping defect of the motor rotor can also be applied to the visual defect detection process of the motor rotor, which includes three different motor rotor defect detection stations and one classification and placement station.

[0178] The first station is used to complete the dynamic balance of the motor rotor, the hook defect (flat hook, skew hook, and broken hook), the core (dynamic balance, long and short), and the slot wedge length defect detection.

[0179] For example, the detection result of the first station is shown in FIG. 6, in which a total of 152 motor rotors are detected, and there are three unqualified products that should be rejected, and the corresponding identification on the software is “NG”. Figure 9

[0180] The second station is used to realize the commutator surface damage detection, the fine turning length, the slot paint, and the slot burr detection. When the second station detection is performed, the motor rotor needs to be powered.

[0181] For example, the detection result of the second station is shown in FIG. 7, in which a total of 152 motor rotors are detected, and there are 67 unqualified products that should be rejected, and the corresponding identification is “NG”; there are 11 motor rotors that can be qualified after repair, and the corresponding identification is “RW”; there are 9 hook defects in the target image, and at this time, the hook defect rejection is closed, that is, only the motor rotor with the hook defect is detected, but the motor rotor with the hook defect is not rejected. Figure 10 The third station is used for lapping defect detection of the motor rotor.

[0182] When the motor rotor sequentially passes through the three stations, it enters the last classification and placement station. At this time, the clamping device moves the motor rotor without defects to the storage area corresponding to the qualified product, and stamps the rotor body surface of the motor rotor, and places the qualified motor rotor; moves the motor rotor with defects to the storage area corresponding to the unqualified product.

[0183] In actual implementation, the unqualified motor rotor can also be moved to the storage area corresponding to the unqualified product by the clamping device after each station detection is completed, and the implementation mode of the unqualified motor rotor is not limited by the embodiment.

[0184]

[0185] ​​To sum up, the method for detecting winding defects of a motor rotor using deep learning provided in this embodiment can obtain at least two target images corresponding to each interval by acquiring images of each interval by using an image acquisition component, wherein each target image includes image data of at least one interval and two enameled wires in the interval, input each target image into an identification model based on deep learning to obtain a target area where at least one target interval is located, a target frame to which each target interval belongs, and a defect classification result of the target frame, the target interval is an interval in the at least one interval that meets the identification criteria of the model, the position of the target frame is determined based on the positions of the two enameled wires in the target interval, and in at least two target images, there is at least one target image corresponding to a defect classification result indicating that the distance between the two enameled wires is less than a preset distance, and it is determined that the motor rotor has a winding defect; the problem that the detection effect of the winding defect of the rotor is poor due to the fact that the commutator is small and the distance between the two enameled wires is not grasped accurately by manual detection and the efficiency of manual detection is low can be solved; the efficiency of using the identification model obtained through deep learning to identify the target image is high, and the model can accurately grasp whether the distance between the two enameled wires in the target image is less than the preset distance, so the detection effect of the winding defect of the motor rotor can be improved.

[0186] In addition, since only the target frame exists in the target area, and the target frame only exists in the target area, the neural network model can be trained using the target area label, the target frame label, and the defect classification label corresponding to each target frame, so that each label can be verified with each other, and the accuracy of the identification obtained after training can be improved.

[0187] In addition, in the case where the training value is greater than or equal to the corresponding preset standard value, the identification model obtained by this training is output, which can ensure the performance of the identification model obtained after training, thereby improving the identification effect of the identification model.

[0188] In addition, converting the model format into the processing format of the image processing library can load the identification model using the image processing library; and loading the converted identification model using the image processing library can assist the image processing of the target image by the identification model through the image processing library, thereby improving the processing speed of the identification model.

[0189] In addition, since the lower the first confidence of the target frame, the higher the false detection rate, by deleting the target frame and the defect classification result corresponding to the target frame whose first confidence is less than the preset confidence, the false detection rate can be reduced, thereby improving the accuracy of the identification model.

[0190] In addition, the first reference target frame with the highest confidence in the target region is determined, the intersection-over-union between the reference target frame and other target frames in the target region is calculated, and the other target frames with the intersection-over-union greater than or equal to the intersection-over-union threshold are deleted, so that each target interval corresponds to a target frame, the target frame most matched with the target interval is obtained, and the recognition effect of the recognition model is improved.

[0191] In addition, the image acquisition component is used to acquire images of each interval after illumination, so that the influence of ambient light on the recognition model is avoided, and the applicability of the recognition model is improved.

[0192] In addition, the distance value is used to determine whether the image acquisition component is opposite to the interval, and the image acquisition component is used to acquire images of the interval in the case that the image acquisition component is opposite to the interval, so that the case that the recognition model cannot recognize the interval due to the interval existing in the edge of the target image is avoided, and the recognition effect of the recognition model is improved.

[0193] Optionally, the present application also provides a computer readable storage medium, wherein a program is stored in the computer readable storage medium, the program is loaded and executed by a processor to implement the method for detecting motor rotor parallel line defects using deep learning.

[0194] Optionally, the present application also provides a computer product, which comprises a computer readable storage medium, wherein a program is stored in the computer readable storage medium, the program is loaded and executed by a processor to implement the method for detecting motor rotor parallel line defects using deep learning.

[0195] The technical features of the above-mentioned embodiments can be combined arbitrarily. To make the description concise, all possible combinations of the technical features in the above-mentioned embodiments are not described, however, as long as the combinations of the technical features do not exist contradictory, they should be considered as the scope of the present application.

[0196] The above-mentioned embodiments only express several implementation manners of the present application, the description is more specific and detailed, but it should not be understood as the limitation of the patent scope of the present application. It should be pointed out that, for ordinary skilled in the art, without departing from the concept of the present application, several modifications and improvements can be made, which belong to the protection scope of the present application. Therefore, the patent protection scope of the present application should be subject to the appended claims.

Claims

1. A method for motor rotor end winding defect detection using deep learning, characterized in that, A motor rotor to be detected is placed on a detection position of a defect detection table, and an image acquisition assembly of the defect detection table is adapted to acquire images of the detection position; one end of the motor rotor has a commutator, and at least two hooks in the form of protrusions are uniformly arranged around the commutator, and a space is formed between adjacent two hooks, and the enameled wire of the motor rotor is wound into the space on one side of each hook and then wound out of the space on the other side of the hook; the method comprises: acquiring images of each space by the image acquisition assembly to obtain at least two target images corresponding to each space; each target image comprises image data of at least one space and two enameled wires in the space; inputting each target image into an identification model based on deep learning to obtain a target region where at least one target space in the target image is located, a target frame to which each target space belongs, and a defect classification result of the target frame; the target space is a space that meets the identification standard of the model in the at least one space, and the position of the target frame is determined based on the positions of the two enameled wires in the target space; in a case where the defect classification result corresponding to at least one target image in the at least two target images indicates that the distance between the two enameled wires is less than a preset distance, it is determined that the motor rotor has a parallel defect; the defect detection table is further provided with a distance measuring assembly adapted to measure the distance between the distance measuring assembly and the commutator; before the acquiring images of each space by the image acquisition assembly to obtain at least two target images corresponding to each space, the method further comprises: acquiring a distance value collected by the distance measuring assembly; determining whether the image acquisition assembly is opposite to the space by using the distance value; in a case where the image acquisition assembly is opposite to the space, a first distance value collected by the distance measuring assembly is greater than a second distance value collected by the distance measuring assembly in a case where the image acquisition assembly is opposite to the hook; in a case where the image acquisition assembly is opposite to the space, acquiring images of the space by using the image acquisition assembly.

2. The method of claim 1, wherein, the training process of the identification model comprises: acquiring images of each space by the image acquisition assembly to obtain at least two training images corresponding to each space; adding a target region label, a target frame label, and a defect classification label corresponding to each target frame to each training image by using an image processing software to obtain a defect classification training set; training a pre-created neural network model by using the defect classification training set to obtain the identification model, and the neural network model is established based on YOLOv5.

3. The method of claim 2, wherein, the training of the pre-created neural network model by using the defect classification training set to obtain the identification model comprises: in the training process, monitoring the training process of the identification model by using a deep learning visualization tool to obtain training values corresponding to the training process, and the training values comprise recall rate, precision rate, and average precision mean value; in a case where the training values are all greater than or equal to corresponding preset standard values, outputting the identification model obtained by this training.

4. The method of claim 1, wherein, Before the inputting each target image into the recognition model based on deep learning, obtaining a target region in which at least one target interval in the target image is located, a target frame to which each target interval belongs, and a defect classification result of the target frame, the method further comprises the following steps of: In a case where the model format of the recognition model does not match a processing format of an image processing library, converting the model format into the processing format; the image processing library comprises an open source computer vision library OpenCV or HALCON.

5. The method of claim 1, wherein, In a case where the defect classification result corresponding to at least one target image in the at least two target images indicates that the distance between the two enameled wires is less than the preset distance, determining that the motor rotor has the parallel line defect, comprising: obtaining a first confidence of each target frame in the target region output by the recognition model; determining whether the first confidence of the target frame is less than a preset confidence; deleting the target frame and the defect classification result corresponding to the target frame whose first confidence is less than the preset confidence; in a case where the defect classification result corresponding to at least one target frame after screening indicates that the distance between the two enameled wires is less than the preset distance, determining that the motor rotor has the parallel line defect.

6. The method of claim 1, wherein, In a case where the defect classification result corresponding to at least one target image in the at least two target images indicates that the distance between the two enameled wires is less than the preset distance, determining that the motor rotor has the parallel line defect, comprising: obtaining a first confidence of each target frame in the target region output by the recognition model; determining a reference target frame with the highest first confidence in the target region; calculating an intersection-over-union between other target frames in the target region except the reference target frame and the reference target frame; deleting other target frames with an intersection-over-union greater than or equal to an intersection-over-union threshold; in a case where the defect classification result corresponding to at least one target frame after deletion indicates that the distance between the two enameled wires is less than the preset distance, determining that the motor rotor has the parallel line defect.

7. The method of claim 1, wherein, The defect detection station is further provided with an illumination assembly adapted to provide auxiliary illumination for the intervals; Before the image acquisition assembly acquires images of each interval to obtain at least two target images corresponding to each interval, the method further comprises the following steps of: controlling the illumination assembly to start; acquiring images of each interval after illumination by using the image acquisition assembly.

8. The method of claim 7, wherein, The defect detection station comprises a rotary driving member adapted to drive the detection position to rotate, and the rotation axis of the detection position is the axis of the motor rotor when the motor rotor is placed on the detection position. Before the distance value collected by the distance measuring assembly is acquired, the method further comprises the following steps of: controlling the rotary driving member to drive the detection position to rotate continuously, and controlling the distance measuring assembly to collect the distance value every preset time interval, so as to trigger the step of acquiring the distance value collected by the distance measuring assembly; using the distance value to determine whether the image acquisition assembly is opposite to the interval; in a case where the image acquisition assembly is opposite to the interval, using the image acquisition assembly to acquire images of the interval; or, The method further comprises: when the image acquisition component is opposite to the hook, controlling the rotation driver to drive the detection position to rotate by a preset angle; and controlling the distance measurement component to collect the distance value, so as to trigger the collection of the distance value of the distance measurement component again; using the distance value to determine whether the image acquisition component is opposite to a gap; when the image acquisition component is opposite to the gap, using the image acquisition component to collect an image of the gap; and wherein the preset angle is less than an included angle formed by two adjacent hooks and the image acquisition component.

9. The method of claim 1, wherein, After determining that the motor rotor has the overlapping defect, the method further comprises: sending a clamping instruction to the clamping device, so that the clamping device moves the motor rotor from the detection position to a target position in response to the clamping instruction, and the target position is used to store the motor rotor with the overlapping defect.

Images