Micro-vibration motor armature defect detection method and system
By combining Darknet53 and Feature Map Pyramid Network (FPN) with a detector, and utilizing auxiliary features such as tin segments, boundaries, and intersections, efficient identification and accurate localization of tin stripping defects in the armature of a micro-vibration motor are achieved, solving the problem of detecting minute defects in existing technologies.
Patent Information
- Application Number
- CN202310302872.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-03-26
- Publication Date
- 2025-11-25
- Estimated Expiration
- 2043-03-26
AI Technical Summary
Existing technologies struggle to efficiently identify and accurately locate solder stripping defects in the armature of miniature vibration motors. Traditional target detection algorithms cannot detect defects based on quantization, and image classification algorithms cannot distinguish specific defect types.
Multi-scale features are extracted using the Darknet53 backbone network, and feature fusion is performed by combining it with the Feature Map Pyramid Network (FPN). The features are then classified and located using a detector. Auxiliary features such as solder segments, boundaries, and intersections are used, along with complex combinational logic conditions, to identify solder stripping defects.
It achieves accurate identification and location of solder stripping defects in the armature of a micro-vibration motor, with an identification rate of 95%, solving the problem of difficulty in detecting minute defects in traditional methods.
Smart Images
Figure CN116579979B_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application belongs to the field of microelectronics technology, and relates to micro vibration motor defect detection, in particular to a micro vibration motor armature defect detection method and system. BACKGROUND
[0002] Micro vibration motor is a kind of micro component used to provide vibration and tactile feedback in electronic devices. With the increasing application of electronic products in daily life, micro vibration motor has great market demand and commercial value. Armature is the core component of micro vibration motor. Therefore, the welding surface detection of armature is very important for enterprises manufacturing micro vibration. The defect detection of micro motor has been proved to be more challenging due to its small size and complex characteristics.
[0003] In industrial work systems, micro motor is usually checked manually before assembly. In order to ensure the accuracy of defect detection, it is necessary to train a number of workers to use microscopes to view and check surface defects, which brings huge cost, including labor cost and time cost. In view of this, there is a need for an efficient and cost-effective technology to detect defects of micro motor parts.
[0004] Traditional detection techniques such as stator fault detection program [J. Penman, H. G. Sedding, B. A. Lloyd and W. T. Fink, Detection and location of interturn short circuits in the stator windings of operating motors in IEEE Transactions on Energy Conversion, vol. 9, no. 4, pp. 652-658, Dec. 1994, doi: 10.1109 / 60.368345] and partial discharge test [G. Stone and J. Kapler, Stator winding monitoring in IEEE Industry Applications Magazine, vol. 4, no. 5, pp. 15-20, Sept.-Oct. 1998, doi: 10.1109 / 2943.715501] are used to analyze abnormal magnetic flux or current. However, problems such as structural defects that produce similar effects are not handled by the above methods. Structural defects of armature are mostly manifested on the surface. For these defects, since the fault signal is displayed after assembly, some defects exist in the form of potential damage.
[0005] With the development of image classification methods based on deep neural networks, they are applied to the defect detection industry and play a good role. However, the image classification algorithm can only evaluate whether the micro motor exists defects, but cannot distinguish which defect it is, and cannot accurately locate the defect position. The target detection algorithm based on CNNs is successfully used for fault detection, such as the two-stage R-CNN series and the one-stage YOLO series, which can label the defect position with anchor box and realize accurate positioning. However, the target detection algorithm is based on the shape features of the target object to distinguish the fault type, which is only suitable for detecting defects with large feature differences. The target detection algorithm cannot detect defects based on quantification, such as "tin scraping", one of the micro motor defects, which has little difference with the micro motor without defects. The traditional target detection algorithm cannot achieve accurate identification. SUMMARY
[0006] The purpose of the present application is to provide a micro vibration motor armature defect detection method and system for identifying and detecting micro defects such as virtual welding, tin beads and tin scraping of the micro vibration motor armature.
[0007] To achieve the above purpose, the present application adopts the following technology.
[0008] The micro vibration motor armature defect detection method provided by the present application comprises the following steps:
[0009] S1, based on the collected image, using a backbone network to extract multi-scale features, and using a feature pyramid network (FPN) to fuse the multi-scale features;
[0010] S2, using a detector to classify and locate the fused features, and simultaneously detecting and identifying tin segments, boundary lines and intersection points auxiliary features;
[0011] S3, according to the auxiliary features, satisfying one or more of the following complex combination logic conditions, indicating that the micro vibration motor armature has tin scraping defects, otherwise indicating that the micro vibration motor armature does not have tin scraping defects:
[0012]
[0013] Wherein, Length(XD) is the length of the tin segment, Distance|JC-XD| is the length from the center point of the intersection anchor box to the upper end point of the tin segment; Height(XD) and Height(FJX) are the upper end point height of the tin segment and the boundary line height, respectively; Tin segment represents the tinned part of the armature copper wire, boundary line represents the horizontal line that divides the armature into upper and lower parts, and intersection point represents the intersection of the two copper wires.
[0014] In the step S1, the backbone network is Darknet53, and the image collected can be extracted by the Darknet53 to obtain shallow features, intermediate features and deep features, which are respectively 8, 16 and 32 times down-sampling features of the original image, namely multi-scale features. In order to extract high-resolution and low-resolution features at the same time, a feature pyramid network (FPN) is applied to combine the multi-scale information in the network. The FPN branch includes a bottom-up path, a top-down path and a lateral connection. The deep features extracted by the Darknet53 are first up-sampled to restore the feature map to the same spatial size as the intermediate layer; then the two layers are connected to combine their feature maps, and the fusion features of the intermediate and deep feature maps are obtained. The same principle is applied to the shallow layer again, and the fusion feature map of the combined intermediate and deep features is up-sampled and spliced with the shallow layer to obtain the final feature map. Through this operation, the shallow layer containing the bottom layer features and rich detail information (such as the edge of the object) can also contain rich strong semantic information.
[0015] In the step S2, the fused feature map obtained in the step S1 is used to perform defect classification prediction and positioning by the detector. The defect classification includes virtual welding and solder ball. Since the system is trained for multi-target classification and target positioning, in order to alleviate the conflict between the classification task and the positioning task, a decoupling head network structure is applied to the fused feature map in the step S1. The basic method is to divide the feature map into two parts, one part is used for the classification task, and the other part is further used for the positioning task. In the present application, the decoupling head network structure of the detector includes a classification module for identifying the micro-vibration motor armature virtual welding and solder ball defects and a positioning module for anchoring the identified defects; further, the positioning module of the detector simultaneously detects the solder segment, the boundary line and the intersection auxiliary features. The classification module includes five two-dimensional convolution layers, ReLU activation functions are respectively arranged after the first four two-dimensional convolution layers, a Sigmoid activation function is arranged after the last two-dimensional convolution layer, and the number of output channels of the feature map in the last two-dimensional convolution layer is set to the product of the number of anchor boxes and the number of parameters. The positioning module includes five two-dimensional convolution layers, ReLU activation functions are respectively arranged after the first four two-dimensional convolution layers, and the number of output channels of the feature map of the last convolution layer is set to the number of anchor boxes multiplied by 4.
[0016] In the step S3, the purpose is to identify the tin missing defect. For tin missing, the judgment basis is mainly the length of tin. In the case of small length change, the deep learning network is difficult to have good performance. Through research, the accurate judgment of tin missing defect can be realized through the complex combination logic condition given by experts. The present application mainly judges whether it is tin missing through three auxiliary features of tin segment, boundary line and intersection point. As mentioned above, the tin segment represents the tinned part of the armature copper wire, Length(XD) is the length of the tin segment, which refers to the diagonal length of the anchor frame containing the tin segment, and Height(XD) is the height of the upper end point of the tin segment. The boundary line represents the horizontal line that divides the armature into upper and lower parts, and Height(FJX) is the height of the boundary line. The intersection point represents the intersection of two copper wires, and Distance|JC-XD| is the length from the anchor frame center point of the intersection point to the upper end point of the tin segment. Through the above three auxiliary features, accurate identification of tin missing defect can be realized based on quantitative analysis.
[0017] The present embodiment also provides a micro-vibration motor armature defect detection system, which comprises:
[0018] The backbone network is used for multi-scale feature extraction of the collected image.
[0019] The feature pyramid network FPN is used for fusing the multi-scale features.
[0020] The detector is used for classifying and positioning the fused features, and simultaneously detecting and identifying the tin segment, boundary line and intersection point auxiliary features.
[0021] The tin missing defect identifier is used for judging whether one or more of the complex combination logic conditions is satisfied according to the auxiliary features. If yes, it indicates that the micro-vibration motor armature has tin missing defect, otherwise, it indicates that the micro-vibration motor armature does not have tin missing defect.
[0022] The output module is used for outputting and displaying the identification results of the detector and the tin missing identifier.
[0023] The detector is mainly used for identifying the false welding and tin bead defects, and simultaneously indicating the three auxiliary features. The detector comprises a classification module and a positioning module. The classification module is used for identifying the false welding and tin bead defects. The positioning module is used for anchoring the positions of the false welding and tin bead, and simultaneously identifying the tin segment, boundary line and intersection point.
[0024] The tin missing defect identifier is mainly used for identifying the complex combination logic condition given by the above formula (1).
[0025] Compared with the prior art, the present application has the following beneficial effects:
[0026] The present application can identify virtual welding and tin bead defects and positioning based on target detection algorithm. The present application can realize effective detection of tin picking defects with large identification difficulty by detecting tin section, boundary line and intersection point, and combining complex combination logic conditions. BRIEF DESCRIPTION OF DRAWINGS
[0027] Figure 1 For micro-vibration motor armature defect type; wherein (a) is virtual welding, (b) is tin bead, and (c) is tin picking.
[0028] Figure 2 For micro-vibration motor armature auxiliary feature display; wherein (a) is tin section, (b) is boundary line, and (c) is intersection point.
[0029] Figure 3 For the flowchart of the micro-vibration motor armature defect detection method of the present application.
[0030] Figure 4 For micro-vibration motor defect recognition result; wherein (a) shows the virtual welding defect anchor frame position and intersection point anchor frame position information of the first micro-vibration motor sample; (b) shows the tin section anchor frame position, boundary line anchor frame position and intersection point anchor frame position information of the second micro-vibration motor sample; (c) shows the tin bead defect anchor frame position, tin section anchor frame position, boundary line anchor frame position and intersection point anchor frame position information of the third micro-vibration motor sample; (d) shows the tin section anchor frame position, boundary line anchor frame position and intersection point anchor frame position information of the fourth micro-vibration motor sample; DETAILED DESCRIPTION
[0031] The technical solutions of the embodiments of the present application will be described clearly and completely in combination with the drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor belong to the present application.
[0032] EMBODIMENT
[0033] This embodiment explains the micro-vibration motor armature defect detection method provided by the present application in detail based on 500 micro-vibration motor armature images. Among them, 400 constitute a training set, and 100 constitute a test set.
[0034] First, analyze the defects of the 500 images and label the defects contained therein, including defect type, defect position information and three auxiliary features. The defect type is shown in Table 1 and Figure 1 As shown. Each motor may contain multiple defects at the same time, for example, when the defect is a tin bead, there may also be a tin picking defect at the same time.
[0035] Table 1. Types, causes, and effects of armature defects
[0036]
[0037] For desoldering, this invention is mainly based on three auxiliary features: solder segments, dividing lines, and intersections. For example... Figure 2 As shown, the tin segment represents the tin-plated portion of the armature copper wire, the dividing line represents the horizontal outline that divides the armature into upper and lower parts, and the intersection point represents the point where the two copper wires meet. A coordinate system is established with the direction parallel to the horizontal outline line as the X-axis and the direction perpendicular to the horizontal outline line as the Y-axis. During the detection process, the tin segment is anchored by selecting a rectangle using the tin segment as the diagonal, and the tin segment length (Length(XD)) is the diagonal length of the anchor frame containing the tin segment. The vertical height (Height(XD)) of the upper end point relative to the lower end point of the tin segment can be obtained by the difference between the Y-axis coordinates of the upper and lower ends of the tin segment. The dividing line is anchored by the horizontal center line of the selection rectangle, and the height (Height(FJX)) of the dividing line is the critical value for determining whether it is tin removal. When Height(XD) is greater than this critical value, it is tin removal; otherwise, it is not. Height(FJX) is obtained by the difference between the Y-axis coordinate of the line containing the dividing line and the Y-axis coordinate of the lower end point of the tin segment. A rectangular intersection anchor frame is generated with the intersection of the two copper wires as the center. Distance|JC-XD| is the length from the center of the intersection anchor frame to the upper end of the tin segment along the length direction of the tin segment. There are no requirements for the size of the intersection anchor frame, as long as it can frame the intersection.
[0038] Based on the above analysis, such as Figure 3 As shown in the figure, the armature defect detection method of the micro-vibration motor provided in this embodiment includes the following steps:
[0039] S1 extracts multi-scale features from the acquired images using a backbone network; and uses a feature map pyramid network (FPN) to fuse the multi-scale features.
[0040] In this embodiment, the backbone network adopted is Darknet53, through which shallow layer features, intermediate layer features and deep layer features can be extracted from the collected images, which are respectively the features after 8, 16, 32 times of down-sampling of the original image, i.e. multi-scale features; the Darknet53 network structure and specific operation can be referred to J. Redmon and A. Farhadi, "YOLOV3: An incremental improvement", arXiv: 1804.02767, 2018. Then a feature pyramid network FPN is applied to merge the multi-scale information in the network. The FPN branch contains a bottom-up path, a top-down path and a lateral connection. The deep layer features extracted by the Darknet53 are first up-sampled to restore the feature map to the same spatial size as the intermediate layer; then the two layers are connected to merge their feature maps, obtaining the fusion features of the intermediate layer and the deep layer feature map. The same principle is applied again to the shallow layer, and the fused feature map of the merged intermediate layer and deep layer features is up-sampled and spliced with the shallow layer to obtain the final feature map. Through this operation, the shallow layer containing the bottom layer features and rich detail information (such as the edges of the object) can also contain rich strong semantic information. Each stage of the FPN outputs a feature map.
[0041] S2 classifies and locates the fused features using the detector, and simultaneously detects and identifies tin segments, boundary lines and intersection point auxiliary features.
[0042] The obtained feature map fused in step S1 is subjected to defect classification prediction and positioning by the detector. Since the system has trained multi-target classification and target positioning, in order to alleviate the conflict between the classification task and the positioning task, a decoupling head network structure is applied on each feature map. The basic method is to divide each feature map into two parts, one part is used for the classification task, and the other part is further used for the positioning task. The decoupling head network structure of the detector includes a classification module and a positioning module, and the network modules of the classification module and the positioning module are as follows:
[0043] Table 1 Network structure of classification module and positioning module
[0044]
[0045]
[0046] Among them, Conv2d represents two-dimensional convolution, and the parameters in the parentheses are the input channel number of the feature map, the output channel number of the feature map, the convolution kernel size and the padding number. In this application, the number of anchor boxes generated finally num_anchors is set to 9, and the number of parameters that can be predicted num_para is set to 5. In addition, the activation function used is RelU and Sigmoid.
[0047] In this embodiment, the inputs of the classification module and the positioning module are both the final fusion feature map obtained in step S1; the output of the classification module is the defect category (such as fake welding, tin bead); the positioning module of the detector detects the anchor box of the fake welding / tin bead defect, and simultaneously detects the tin segment, the boundary line and the intersection auxiliary features to obtain the anchor box of the tin segment, the boundary line and the intersection.
[0048] The anchor box of the tin segment is obtained by selecting a rectangular box as an anchor with the tin segment as a diagonal line. The anchor box of the boundary line is obtained by selecting a rectangular box as an anchor with the boundary line as the horizontal center line. The intersection anchor box is generated with the intersection of the two copper wires as the center, and there is no requirement for the size of the intersection anchor box, as long as it can frame the intersection.
[0049] S3, according to the auxiliary features, if one or more of the following conditions are met, it indicates that the micro-vibration motor armature has a tin pulling defect, otherwise it indicates that the micro-vibration motor armature does not have a tin pulling defect:
[0050]
[0051] wherein Length(XD) is the length of the tin segment, Distance|JC-XD| is the length from the center point of the intersection anchor box to the upper end point of the tin segment; Height(XD) and Height(FJX) are the upper end point of the tin segment and the boundary line respectively; tin segment represents the diagonal line of the anchor box containing the tin segment, boundary line represents the horizontal center line dividing the armature into upper and lower parts, and intersection represents the intersection of the two copper wires.
[0052] The embodiment further provides a micro-vibration motor armature defect detection system, which comprises:
[0053] a backbone network for multi-scale feature extraction on the collected images. The backbone network here is Darknet53.
[0054] a feature pyramid network FPN for fusing the multi-scale features.
[0055] a detector for classifying and positioning the fused features, while detecting and identifying tin segment, boundary line and intersection auxiliary features.
[0056] The detector comprises a classification module and a positioning module. The classification module is used for identifying fake welding and tin bead defects. The positioning module is used for anchoring the positions of fake welding and tin bead, and simultaneously identifying tin segment, boundary line and intersection. The network structure of the classification module and the positioning module is shown in Table 1.
[0057] The tin scraping defect recognizer is used to determine whether one or more of the complex combination logic conditions is met according to the auxiliary features, and if yes, it indicates that the micro-vibration motor armature has a tin scraping defect, otherwise, it indicates that the micro-vibration motor armature does not have a tin scraping defect. The tin scraping defect recognizer is mainly used to identify the complex combination logic condition given by the above formula (1).
[0058] In the application process, first, the micro-vibration motor armature defect detection system is trained by using the training set, and the mean absolute error (MNE) is used as the loss function, then the system learning parameters are optimized by the Adam optimization algorithm until the loss value meets the set requirement. Then, the trained micro-vibration motor armature defect detection system is used to test the test set according to the above steps S1-S3. Through the micro-vibration motor armature defect detection method and system provided by the present application, the identification accuracy of the defects including tin scraping can reach 95%, and accurate positioning can be realized.
[0059] The output module is used to output and display the identification results of the detector and the tin scraping recognizer. The output results mainly include the anchor frame of the identified false welding / solder ball defect, or / and the anchor frame of the tin scraping defect (including the anchor frame of the tin segment, the anchor frame of the boundary line, and the anchor frame of the intersection point).
[0060] Figure 4 The identification results of four micro-vibration motor sample images in the test set are given, and it can be seen from the figure that the micro-vibration motor armature defect detection method and system provided by the present application can realize effective identification and accurate positioning of false welding, solder ball and tin scraping defects.
[0061] Those skilled in the art will appreciate that the embodiments described herein are presented for the purpose of helping the reader understand the principles of the present application, and should be understood as not limiting the scope of protection of the present application to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations according to the technical inspiration disclosed in the present application without departing from the spirit of the present application, and these modifications and combinations are still within the scope of protection of the present application.
Claims
1. A micro-rotational vibration motor armature defect detection method, characterized in that, The method comprises the following steps: S1, based on the collected image, a multi-scale feature is extracted by using a backbone network; And the multi-scale feature is fused by using a feature pyramid network FPN; S2, the fused feature is classified and positioned by using a detector, and tin segment, boundary line and intersection point auxiliary features are detected and identified; S3, according to the auxiliary features, one or more of the following complex combination logic conditions are met, indicating that the micro-vibration motor armature has a tin pulling defect, otherwise indicating that the micro-vibration motor armature does not have a tin pulling defect: Wherein, Length(XD) is the length of the tin segment, Distance|JC-XD| is the length from the center point of the intersection point anchor frame to the upper end point of the tin segment; Height(XD) and Height(FJX) are the upper end point height of the tin segment and the boundary line height respectively; Tin segment represents the tin plated part of the armature copper wire, boundary line represents the horizontal line that divides the armature into upper and lower parts, and intersection point represents the intersection of the two copper wires.
2. The micro- resonator motor armature defect detection method of claim 1, wherein, In step S1, the backbone network used is Darknet53.
3. The micro- resonant vibration motor armature defect detection method according to claim 1 or 2, characterized in that, In step S2, the detector includes a classification module for identifying micro-vibration motor armature virtual welding and tin bead defects, and a positioning module for anchoring the identified defects.
4. The micro- resonant vibration motor armature defect detection method of claim 3, wherein, The classification module includes five two-dimensional convolution layers, ReLU activation functions are set after the first four two-dimensional convolution layers, and a Sigmoid activation function is set after the last two-dimensional convolution layer, and the number of feature map output channels in the last two-dimensional convolution layer is set as the product of the number of anchor frames and the number of parameters.
5. The micro- resonant vibration motor armature defect detection method of claim 3, wherein, The positioning module includes five two-dimensional convolution layers, ReLU activation functions are set after the first four two-dimensional convolution layers, and the number of feature map output channels of the last convolution layer is set as the number of anchor frames multiplied by 4.
6. The micro- resonant vibration motor armature defect detection method of claim 3, wherein, The positioning module of the detector simultaneously detects and identifies tin segment, boundary line and intersection point auxiliary features.
7. A micro- resonator motor armature defect detection system, characterized by, It comprises: a backbone network for multi-scale feature extraction of the collected image; a feature pyramid network FPN for fusing the multi-scale features; a detector for classifying and positioning the fused features, and simultaneously detecting and identifying tin segment, boundary line and intersection point auxiliary features; a tin pulling defect identifier, which is used by a user to determine whether one or more of the complex combination logic conditions are met, and if so, it indicates that the micro-vibration motor armature has a tin pulling defect, otherwise it indicates that the micro-vibration motor armature does not have a tin pulling defect; the complex combination logic condition is: Wherein, Length(XD) is the length of the tin segment, Distance|JC-XD| is the length from the center point of the intersection point anchor frame to the upper end point of the tin segment; Height(XD) and Height(FJX) are the upper end point height of the tin segment and the boundary line height respectively; Tin segment represents the tin plated part of the armature copper wire, boundary line represents the horizontal line that divides the armature into upper and lower parts, and intersection point represents the intersection of the two copper wires. an output module for outputting and displaying the identification results of the detector and the tin pulling identifier.
8. The micro- resonant vibration motor armature defect detection system of claim 7, wherein, The detector comprises a classification module and a positioning module; the classification module is used for identifying the false soldering and the tin bead defects; the positioning module is used for anchoring the positions of the false soldering and the tin bead, and simultaneously marking the tin section, the boundary line and the intersection point.