• <tr id="yyy80"></tr>
  • <sup id="yyy80"></sup>
  • <tfoot id="yyy80"><noscript id="yyy80"></noscript></tfoot>
  • 99热精品在线国产_美女午夜性视频免费_国产精品国产高清国产av_av欧美777_自拍偷自拍亚洲精品老妇_亚洲熟女精品中文字幕_www日本黄色视频网_国产精品野战在线观看 ?

    Design of Online Vision Detection System for Stator Winding Coil

    2023-12-28 08:49:08LIYanLIRuiXUYang

    LI Yan(李 艷), LI Rui(李 芮), XU Yang(徐 洋)

    College of Mechanical Engineering, Donghua University, Shanghai 201620, China

    Abstract:The quality of the stator winding coil directly affects the performance of the motor. A dual-camera online machine vision detection method to detect whether the coil leads and winding regions were qualified was designed. A vision detection platform was designed to capture individual winding images, and an image processing algorithm was used for image pre-processing, template matching and positioning of the coil lead area to set up a coordinate system. After eliminating image noise by Blob analysis, the improved Canny algorithm was used to detect the location of the coil lead paint stripped region, and the time was reduced by about half compared to the Canny algorithm. The coil winding region was trained with the ShuffleNet V2-YOLOv5s model for the dataset, and the detect file was converted to the Open Neural Network Exchange (ONNX) model for the detection of winding cross features with an average accuracy of 99.0%. The software interface of the detection system was designed to perform qualified discrimination tests on the workpieces, and the detection data were recorded and statistically analyzed. The results showed that the stator winding coil qualified discrimination accuracy reached 96.2%, and the average detection time of a single workpiece was about 300 ms, while YOLOv5s took less than 30 ms.

    Key words:machine vision; online detection; V2-YOLOv5s model; Canny algorithm; stator winding coil

    0 Introduction

    The stator, an important component of basic electrical equipment such as generators and motors[1], consists of multiple stator winding coils welded together in bulk. The individual stator winding coil is wound first and a section of the lead is left for paint stripping after the winding is completed. Deformation of the winding region, fewer turns of winding, and paint stripping treatment of the coil lead region have a great impact on the safety of the winding[2]. Research on lead surface defect detection is mainly focused on continuity detection, lacquer tumor detection, winding pass detection, and wire diameter detection[3-4]. The detection of paint stripped region is rarely mentioned. At present, factories still use manual detection for stator winding coils, relying on the rich experience and subjective judgement of the workers, and there is no accurate detection standard. So it is necessary to study an online detection system for detecting the conformity of stator winding coils.

    Machine vision has the characteristics of nondestructive detection, which can accurately and efficiently detect part defects, and has been widely used in industrial automation production[5-7]. Machine vision detection systems can also be used on industrial assembly lines for measuring the bore and flat dimensions of mechanical parts[8]. With the rapid development of computer technology, machine deep learning based on convolutional neural networks is widely used in various fields for target detection, with its high accuracy and scene adaptive capability[9-10]. Ni[11]proposed a deep learning-based stator coil defect detection algorithm, proposed a migration learning method for small samples, and verified its effectiveness in the case of insufficient samples. However, this method has high requirements for hardware, needs too many network parameters and the model takes a long time during iteration. Ma[12]designed a three-core power cord machine vision recognition system for automated production and detection to improve the accuracy of three-core power cord color recognition and the accuracy of copper wire riveting defect detection. However, the hardware imaging system using a specially designed opaque dark box with a domed light source has certain application limitations and certain requirements for the space site of the factory production line. Zhu[13]used two line array cameras symmetrically installed on the electric locomotive to capture the image of the contact wire and applied an edge detection algorithm to find the coordinates of the boundary points of the wear trauma surface. In this way, the measurement of the geometric parameter values of the wire was realized, but the camera distortion and the matching of the points in the left and right images still need to be studied. Xuetal.[14]proposed an object recognition method based on multi-visual cross-attention for a bright light interference and complex background interference environment without the need for a massive bright light and complex background interference training dataset, but the real-time performance of target tracking needs to be improved. Yuetal.[15]proposed a deep learning model named YOLOv4-FPM based on the YOLOv4 model, to realize real-time detection for bridge cracks by unmanned aerial vehicle (UAV).

    Currently, machine vision detection of the winding coil is less used because of the fine wire diameters, the reflectivity of materials, and the easy bending of the coil, which can easily produce shadows in the light and cause false detection. Furthermore, there is no specific criterion for judging the qualification of stator winding coil. To improve the detection accuracy and efficiency, in this paper, an online vision detection system was designed for real-time and effective detection and qualification judgement of stator winding coil using image processing and the YOLOv5 algorithm. The visualization software interface was designed to display the detection data and judgement results in real time, and statistical analysis of the data was performed.

    1 Overall Platform Design of Online Vision Detection System

    The diagram of the paint stripped region on the stator winding coil lead is shown in Fig.1. There are qualified regions, unqualified defect types of paint stripped starting position, and the region where paint has not been stripped. The middle part of the lead paint requires to be stripped, the starting position of the paint stripped region should meet the standard and have a certain length. The diagram of the coil winding region is shown in Fig.2. The unqualified winding coil has fewer turns of winding or deformation winding. The qualified winding coil has fixed physical characteristics of two winding wires crossing each other, while the unqualified winding does not.

    Fig.1 Diagram of paint stripped region on stator winding coil lead: (a) qualified region; (b) unqualified starting position; (c) unstripped region

    Fig.2 Diagram of coil winding region: (a) qualified winding; (b) fewer turns of winding; (c) deformation winding

    Based on the above analysis, this project designed an online vision detection platform to detect the presence or absence of paint stripped from the upper and lower leads by image processing algorithms, measured the length, and gave the starting position of the paint stripped region. For the coil winding region, the ShuffleNet V2-YOLOv5s model was used to discriminate the winding wires cross feature. The judgement was based on the length of the stripped regionlbeing 10-20 mm and the horizontal axis coordinate valued<0 at the starting position of the stripped region. The workpiece passed when the winding wires cross feature was identified as “Accept”.

    1.1 Hardware platform design for the dual-camera system

    Stator winding coil online vision detection system using a dual-camera hardware platform mainly consists of vision acquisition devices, which are controlled by a programmable logic controller (PLC), and two industrial cameras used for simultaneous image acquisition; conveyor devices; light sources; industrial personal computers (IPC),etc.The hardware structure of the detection system is shown in Fig.3. The cameras are selected from Hikvision’s 6-megapixel 1/1.8-inch CMOS Gigabit Ethernet industrial array scan camera. The specific model number is MV-CA060-11GM, with a 75 mm fixed-focus lens for camera 1 and a 25 mm fixed-focus lens for camera 2. The two cameras, the IPC and the PLC are connected using a Gigabit Ethernet switch expansion interface. The cameras transmit images to the IPC detection system via GigE cable.

    Fig.3 Diagram of the hardware structure of the detection system

    The conveyor device consists of PLC and conveyor belts. The conveyor belts send the workpiece to be tested by the photoelectric switch, the PLC collects the photoelectric signal and controls the camera to take pictures, and the system judges whether the current workpiece is qualified or not and sends the result to the PLC. If the workpiece is qualified, the conveyor belt continues to run; if it is not qualified, the alarm is raised and the workpiece is taken away by the robotic arm.

    The light source control system includes two LED lights and a switch. Two white LED ring lights provide shadow-free light, the top light source has an outer diameter of 220 mm and an inner diameter of 170 mm, and the side light source has an outer diameter of 180 mm and an inner diameter of 110 mm.

    When building the hardware system, the light source, the workpiece, and the camera lens are kept perpendicular to the center, with a white background plate behind to avoid environmental interference. The distance between the light sources and the camera and the stator winding coil to be tested should meet the requirements of the production site. The hardware structure location diagram of the detection system is shown in Fig.4.

    Fig.4 Hardware structure location diagram of the detection system

    2 Research on Dual Region Algorithm for Online Vision Detection System

    2.1 Lead paint stripped region detection algorithm

    For the lead paint stripped region, firstly, the image is preprocessed to reduce and eliminate the noise interference caused by reflections and shadows as well as ambient light, and to enhance the contrast of the image. Secondly, feature-based template matching is used to locate the image and set up a reference coordinate system. Finally, Blob analysis is performed on the image to separate the object from the background using threshold segmentation and grayscale morphological operation. An improved Canny algorithm is used for lead edge detection.

    2.1.1Imagepreprocessing

    The illumination in the production environment varies due to weather and other reasons, and there are also noise disturbances e.g. light spots inside the image. Firstly, the image is adjusted to grayscale, and further expanded with a 3 × 3 square eight-neighborhood dilation of the structural elements and a 3 × 1 vertical three-element dilation, respectively. Secondly, the pixels are remapped using a grayscale transformation function to ensure that the grayscale values are similar for successive phases. Finally, median filtering is performed to further eliminate noise disturbances. The effect of image preprocessing is shown in Fig.5.

    Fig.5 Effect of image preprocessing

    2.1.2Templatematchingandpositioning

    Template matching uses the sliding window principle to match the template on the target image[16]. The stator winding workpiece recess region is selected as the template by shielding both sides of the conveyor belt part in the image, which has obvious features and a relatively simple geometry. The template matching, region of interest (ROI) is shown in Fig.6(a), and the partial enlargement of the ROI is shown in Fig.6(b). The center of the image recess is found on the result of template matching for positioning, and the direction of axes x and y of the coordinate system after positioning is shown in Fig.6(c). If the template matching fails, the PLC controls the robotic arm to pick up the workpiece.

    Fig.6 Effects of template matching and positioning: (a) template matching ROI; (b) partial enlargement of ROI; (c) coordinate system after positioning

    2.1.3Blobanalysis

    Blob analysis uses threshold segmentation to segment the target and background in the image, the connected region analysis for the target and the grayscale morphology algorithm to eliminate shadows within the reflective region. Due to the effect of light in different environments resulting in different gray values of the image, adaptive dynamic thresholding of the image is performed using the Otsu algorithm. The image after using threshold segmentation is shown in Fig.7(a), and the segmented image is processed for connected region analysis. The holes with smaller areas are filled and the gray-scale morphological erosion algorithm is used to enhance the feature values of the target, and the processing effect is shown in Fig.7(b).

    Fig.7 Effect of Blob analysis on lead region processing : (a) image segmentation; (b) connected region analysis

    2.1.4ImprovedCannyalgorithmforedgedetectionofleadpaintstrippedregion

    The Sobel algorithm and Canny algorithm with better accuracy are used for edge detection of the lead paint stripped ROI, and the detection results are shown in Fig.8.

    Fig.8 Detection effect of different algorithms : (a) Sobel algorithm; (b) Canny algorithm

    The Canny algorithm is effective but time-consuming and does not meet the requirements for online detection. Since the lead detection of the workpiece requires only edge detection in the horizontal direction, there is no need to calculate the magnitude and angle in the vertical direction, and there is no need to query by angle for non-maximum suppression[17]. The simplified Canny algorithm calculates the second-order derivative only for the horizontal direction of the image lead region.

    Since multiple edge pairs may be detected in the lead paint stripped region, which interferes with the detection of actual edges[18], the conventional Canny algorithm uses non-maximum suppression and high and low thresholding methods. The effect is susceptible to noise interference and requires repeated comparisons. The Canny algorithm is improved to use the expected length and contrast model to determine the final edge and reduce the algorithm time.

    Firstly, the edge pairs are scored according to the difference between the expected length and the inspected length. According to the process parameters, the length of the paint stripped region is mostly 13-19 mm, and the desired lengthL0of the upper and lower stripping line is taken as its middle value of 16 mm. A high scorenfor the length of 0.75L0-1.25L0(12-20 mm) is given as

    (1)

    (2)

    whereLis the actual inspection length. The edge pairs that score in the bottom 50% are excluded.

    Next, the contrast model is used to find the final edge pairs, and the contrast thresholdTof the upper and lower leads is found:

    (3)

    In the original image grayscale histogramf(x,y), the images with contrast less than the thresholdTare excluded to obtain the new grayscale histogramg(x,y).The highest contrastCmaxis found out, and its corresponding edge is the final edge.Cmaxis shown as

    (4)

    Detection results of coil lead paint stripped region under the improved Canny algorithm are shown in Fig.9. For 10 groups of workpieces, each group does 50 detections. A comparison of the detection time of the three algorithms is shown in Table 1. As can be seen from Table 1, the improved Canny algorithm reduces the time by more than half compared to the previous one, significantly improving the detection efficiency of the system.

    Fig.9 Detection results of coil lead paint stripped region under improved Canny algorithm : (a) upper lead; (b) lower lead

    Table 1 Comparison of the detection time of three algorithms

    2.2 Coil region winding wires cross feature detection algorithm

    For real-time detection of coil winding wires cross feature, a single-stage deep learning target detection algorithm of the YOLO series is used to extract features and predict the classification and localization of targets by convolutional neural networks. The object detection task is treated as a regression problem, and the two phases of candidate region and detection are combined into one[19]to improve detection efficiency and reduce time.

    YOLO divides the input image intoS×Sgrid cells, and if the center of an object is in that grid, then the grid predicts the objects in it[20]. The prediction yields the output feature dimensionS×Sand the depth dimensionB×(5 +C), whereSis the number of grids after image division,Bis the number of bounding boxes to be predicted for each grid andCis the number of categories of bounding boxes. Then the redundant windows are removed by non-maximum suppression for the previous inference. Each bounding box predicts five values for the center position coordinatesxandyof the box, the widthwand heighthof the box, and the targetivity score.

    Among them, YOLOv5 uses an adaptive anchor frame that can be selected from YOLOv5s, YOLOv5m, and YOLOv5l weight types, and its model selection and training speed are superior compared to the previous ones[21]. Using the YOLOv5 algorithm for the detection of winding wires cross features, the YOLOv5 backbone feature extraction network uses a C3 structure, which brings a large number of parameters and a slow detection speed, so the application is limited. For industrial assembly lines, a low latency accurate and fast model with light weight is needed to make the detection of parts as fast as possible. In this paper, the backbone feature extraction network of the YOLOv5-5.0 version is replaced by the lighter ShuffleNet V2 network[22], the comparison of YOLOv5-5.0 and ShuffleNet V2-YOLOv5 is shown in Table 2. It can be seen that there is a significant reduction in the number of parameters for each part module, and there is one less layer of network modules after the lightweight.

    Original images of 910 stator winding coils are enhanced by translating and rotating to obtain 1 800 images (1 000 of them are qualified and 800 are unqualified). The dataset is produced, and the qualified workpieces are marked as “Accept”, and the unqualified workpieces are marked as “Reject”. The deep learning environment is configured using the Python machine learning library PyTorch 1.9.0, the Python 3.7 programming language, CUDA 11.1 computing architecture, and the computer graphics card model NVIDIA GeForce RTX 3050.

    YOLOv5s, YOLOv5l, and ShuffleNet V2-YOLOv5s weight models are selected for dataset training. The input image size is 640 pixel× 640 pixel, the ratio of the training sets and test sets is 8∶2, the parameter epochs are set to 200 generations, and the batch size is 4. A comparison of parameters after training of the dataset under different weight models is shown in Table 3. The floating point operation speed (GFLOPs) of the lightweight model is significantly reduced to 2.6, which makes the training time shorter. The training time for this dataset is about 16 h for ShuffleNet V2-YOLOv5s, 21 h for YOLOv5s, and 27 h for YOLOv5l.

    Table 2 Comparison of module parameters between YOLOv5-5.0 network and ShuffleNet V2-YOLOv5 network

    Table 3 Comparison of parameters after training of the dataset under different weight models

    The evaluation metrics of the training results under different weight models are shown in Fig.10. From Fig.10, it can be seen that the ShuffleNet V2-YOLOv5s model does not differ much from the other two models in terms of the performance of mAP, precision and recall on this dataset, but performs well in terms of obj_loss.

    Fig.10 Evaluation metrics for training results of models with different weights : (a) mAP@0.5; (b) precision; (c) recall; (d) box_loss; (e) cls_loss; (f) obj_loss

    ShuffleNet V2-YOLOv5s model is used to train the dataset, and the generated best.pt file is selected as the weight file for the new prediction detection for the stator winding coil workpieces on the assembly line. Eight representative coil workpieces (one qualified workpiece and seven unqualified workpieces with different defects) are selected for the prediction result presentation, as shown in Fig.11. The category of workpieces is correctly determined, and the confidence of each workpiece category is above 0.90, which meets the detection requirements.

    Fig.11 Prediction results for typical workpieces : (a) qualified workpiece; (b)-(h) unqualified workpieces

    3 Experiment and Analysis of Online Vision Detection System

    In this study, the software interface of the detection system based on.Net Framework is developed in Visual Studio 2015 environment on Win10 using C# language. It includes camera control, image processing setup, data analysis, YOLOv5 inspection, and remote communication modules. The main interface displays the processed images, measured data values and workpiece discrimination results, and completes the control and debugging of the detection system, as shown in Fig.12. The introduction of each module is as follows.

    Fig.12 The main interface of the system software

    1)Camera control module. It includes the camera hardware and software trigger mode. The software trigger mode is divided into single acquisition and continuous acquisition, the continuous acquisition controls the camera to take pictures simultaneously by setting the interval time. The system sets four time intervals, 500 ms, 750 ms, 1 s and 3 s.

    2)Image processing setup module. The main function of this module is to select algorithms, cameras, or image files used for lead region detection, as well as real-time editing of files.

    3)Data analysis module. The main function of this module is the real-time display of lead detection data, and the clearing and exporting of the detection data. After the detection is finished, it is saved as a.csv file, and the mean and variance of the data are calculated,etc.The average detection time of the workpiece, the number of acquisitions, the number of passes, and the pass rate are given.

    4)YOLOv5 detection module. The main function of this module is to display detection images and results of the detection of winding cross features in the coil region.

    5)Remote communication module. The main function of this module is to send the results to PLC through TCP/IP using Socket after the detection is finished. The system sends the results of “Accept”, “Reject”, or “Retry to PLC” in the form of byte arrays, and PLC completes the subsequent control operation according to the results.

    An experimental platform is built to test the online vision detection system for stator winding coil. Three detection aspects, namely template matching, coil lead region, and coil winding cross feature, are classified and counted for the system’s accuracy test. The 1 000 collected pictures under different light intensities in daily production are tested. Statistics of qualified results for stator winding coil workpieces are shown in Table 4. The display interface of data results for the coil lead region is shown in Fig.13.

    Fig.13 Display interface of data results for coil lead region

    Table 4 Statistics of qualified results for stator winding coil workpieces

    ShuffleNet V2-YOLOv5s model is used to train the dataset and the generated best.pt file is selected as the new weight for the detect file and converted to Open Neural Network Exchange (ONNX) file format. The ONNX deep learning model is invoked in the detection system to detect the stator winding coil workpieces on the assembly line. For eight representative groups of coil workpieces (one group of qualified workpieces and seven groups of unqualified workpieces with different defects), 100 tests are performed for each group of winding cross features, and a total of three tests are done. The average time for workpiece detection under the ShuffleNet V2-YOLOv5s model is shown in Table 5. From Table 5, it can be seen that it has a very fast speed for detection, and the average detection time per image does not exceed 30 ms.

    Table 5 Average time for workpiece detection under ShuffleNet V2-YOLOv5s model

    Ten groups of workpieces of the stator winding coils of the same batch model motor are detected under different light intensities, with 700 sample images for each group. The detection results are shown in Table 6. As can be seen from Table 6, the workpiece discrimination results are judged by the combination of image processing and YOLOv5 results, the discrimination results are consistent with the actual results, and the test system has good stability. The average detection time for a workpiece to complete image acquisition, image processing, and communication does not exceed 300 ms, meeting the requirement that the average detection time of each workpiece does not exceed 500 ms. After testing, the accuracy of the workpiece qualified discrimination is 96.2%, and the data acquisition, processing, and communication meet the design requirements.

    Table 6 Detection results of 10 groups of workpieces under different light intensities

    4 Conclusions

    The quality of the stator winding coil has an important impact on motor safety. In this study, by analyzing the types and specific characteristics of defects in the leads and winding region of stator winding coils, the online vision detection platform was designed. The image processing and ShuffleNet V2-YOLOv5s model were used to detect whether the lead region of the winding workpiece was paint stripped, whether the starting position of the leads stripped was qualified, whether the turns of coil winding were too small or whether the coil winding was deformed. By using the image processing and ShuffleNet V2-YOLOv5s model, the real-time detection requirement was achieved. The human-machine interface was designed to realize the functions of automatic detection and identification, qualified information display, and data statistical calculation. The accuracy of qualified discrimination and detection time of the workpiece were tested. The test results showed that the accuracy of qualified discrimination reached 96.2%, and the average time of individual workpiece detection was about 300 ms, among which the detection time of YOLOv5 for individual workpiece winding cross features was less than 30 ms, which met the requirement of detection on the industrial assembly line. The detection system has positive application value in actual production applications, effectively improving the detection accuracy of stator winding coil, and has certain application value and research significance for coil qualification detection.

    国产国语露脸激情在线看| 国产无遮挡羞羞视频在线观看| 悠悠久久av| 精品一品国产午夜福利视频| 亚洲国产欧美一区二区综合| 大片电影免费在线观看免费| 99久国产av精品国产电影| 在线亚洲精品国产二区图片欧美| 黄色视频不卡| 两性夫妻黄色片| 亚洲成国产人片在线观看| 午夜福利视频精品| 如何舔出高潮| 国产一区二区激情短视频 | 美女午夜性视频免费| 国产探花极品一区二区| 亚洲一卡2卡3卡4卡5卡精品中文| 亚洲,一卡二卡三卡| 激情视频va一区二区三区| 欧美黑人精品巨大| 午夜日本视频在线| 国产一区二区 视频在线| 97人妻天天添夜夜摸| 国产成人系列免费观看| 久久99一区二区三区| 亚洲免费av在线视频| 99九九在线精品视频| 国产欧美日韩一区二区三区在线| 久久久国产精品麻豆| 免费观看a级毛片全部| 亚洲精品国产色婷婷电影| 最近手机中文字幕大全| 日韩 欧美 亚洲 中文字幕| 国产有黄有色有爽视频| 色精品久久人妻99蜜桃| 国产精品免费大片| 97人妻天天添夜夜摸| 巨乳人妻的诱惑在线观看| 精品国产露脸久久av麻豆| 国产精品无大码| 久久久久视频综合| 亚洲国产精品成人久久小说| 另类亚洲欧美激情| 国产一区二区三区av在线| 久久久久人妻精品一区果冻| 久久人人爽人人片av| 一区二区日韩欧美中文字幕| 亚洲一区二区三区欧美精品| 国产成人av激情在线播放| 韩国高清视频一区二区三区| 午夜福利,免费看| 亚洲精品aⅴ在线观看| 亚洲国产欧美在线一区| av女优亚洲男人天堂| 亚洲成人免费av在线播放| 性高湖久久久久久久久免费观看| 精品久久久久久电影网| 中文字幕精品免费在线观看视频| 久久久国产欧美日韩av| 电影成人av| 十分钟在线观看高清视频www| 亚洲婷婷狠狠爱综合网| videos熟女内射| 99精国产麻豆久久婷婷| xxx大片免费视频| 不卡av一区二区三区| 免费黄频网站在线观看国产| 欧美精品一区二区大全| 在线观看国产h片| 国产精品一二三区在线看| 1024香蕉在线观看| 成人手机av| 综合色丁香网| 伊人久久国产一区二区| 大话2 男鬼变身卡| 夫妻午夜视频| 亚洲国产欧美日韩在线播放| 国产精品.久久久| 日韩精品有码人妻一区| 一二三四中文在线观看免费高清| 涩涩av久久男人的天堂| 国产av码专区亚洲av| 菩萨蛮人人尽说江南好唐韦庄| 精品少妇内射三级| 美女午夜性视频免费| 男的添女的下面高潮视频| av一本久久久久| 日韩中文字幕欧美一区二区 | 国产女主播在线喷水免费视频网站| 一边亲一边摸免费视频| 精品国产超薄肉色丝袜足j| 一本一本久久a久久精品综合妖精| 久久性视频一级片| 黄色怎么调成土黄色| 丝袜在线中文字幕| 国产av精品麻豆| 日韩中文字幕视频在线看片| 亚洲精品第二区| 亚洲国产精品999| 香蕉国产在线看| 久久青草综合色| 18禁动态无遮挡网站| 伦理电影大哥的女人| 一区二区av电影网| 国产老妇伦熟女老妇高清| 成年动漫av网址| 一级片免费观看大全| 在线观看国产h片| 老汉色∧v一级毛片| 国产一区二区激情短视频 | 日本色播在线视频| 精品久久久精品久久久| 赤兔流量卡办理| 中文欧美无线码| 啦啦啦视频在线资源免费观看| 18禁观看日本| 成人毛片60女人毛片免费| 色综合欧美亚洲国产小说| 老司机在亚洲福利影院| 久久精品国产亚洲av涩爱| 香蕉国产在线看| 男男h啪啪无遮挡| 久久久国产欧美日韩av| 日韩av不卡免费在线播放| 久久人人爽av亚洲精品天堂| 亚洲精品国产av蜜桃| 日韩熟女老妇一区二区性免费视频| 精品国产一区二区久久| 欧美日韩精品网址| 啦啦啦 在线观看视频| 精品国产一区二区久久| a 毛片基地| 欧美变态另类bdsm刘玥| 婷婷色综合大香蕉| 日本欧美国产在线视频| 国产伦理片在线播放av一区| 亚洲欧洲国产日韩| 午夜激情av网站| 国产毛片在线视频| 国产一区二区在线观看av| 精品久久久精品久久久| 亚洲av成人不卡在线观看播放网 | 一本一本久久a久久精品综合妖精| av福利片在线| 成年人午夜在线观看视频| 日韩欧美一区视频在线观看| 波多野结衣一区麻豆| 丝袜美足系列| 欧美变态另类bdsm刘玥| 精品一区二区免费观看| 久久人人爽人人片av| 国产在线免费精品| 色吧在线观看| 美女视频免费永久观看网站| 国产成人精品久久二区二区91 | 久久国产精品男人的天堂亚洲| 久久久亚洲精品成人影院| 久久久久久久久免费视频了| 久久国产精品男人的天堂亚洲| 18在线观看网站| 国产99久久九九免费精品| 精品亚洲乱码少妇综合久久| 亚洲精品日韩在线中文字幕| 9191精品国产免费久久| 亚洲欧美成人精品一区二区| 一本色道久久久久久精品综合| 少妇人妻精品综合一区二区| 免费高清在线观看日韩| 中文字幕最新亚洲高清| 97在线人人人人妻| 91aial.com中文字幕在线观看| netflix在线观看网站| 国产精品久久久久成人av| 成人午夜精彩视频在线观看| 亚洲一卡2卡3卡4卡5卡精品中文| 国产精品99久久99久久久不卡 | 精品久久久精品久久久| 欧美黑人欧美精品刺激| 一区在线观看完整版| 女人爽到高潮嗷嗷叫在线视频| av视频免费观看在线观看| 精品国产乱码久久久久久小说| 亚洲欧美清纯卡通| 精品少妇黑人巨大在线播放| 自拍欧美九色日韩亚洲蝌蚪91| 国产精品一二三区在线看| 大香蕉久久网| 国产男女内射视频| 久久精品国产a三级三级三级| 午夜福利网站1000一区二区三区| av国产精品久久久久影院| 丁香六月天网| 波多野结衣一区麻豆| av不卡在线播放| 免费在线观看完整版高清| 99久久精品国产亚洲精品| 9热在线视频观看99| 99re6热这里在线精品视频| e午夜精品久久久久久久| 免费黄色在线免费观看| 欧美xxⅹ黑人| 国产免费视频播放在线视频| 18禁国产床啪视频网站| 日韩中文字幕视频在线看片| 韩国精品一区二区三区| 超色免费av| 一本久久精品| 国产欧美亚洲国产| 一级片'在线观看视频| 国产成人欧美| 亚洲av中文av极速乱| 国产成人a∨麻豆精品| 在线观看免费视频网站a站| 免费观看a级毛片全部| 色婷婷久久久亚洲欧美| 午夜福利,免费看| 亚洲成国产人片在线观看| 国产无遮挡羞羞视频在线观看| 国产伦人伦偷精品视频| 91精品国产国语对白视频| 在线观看免费视频网站a站| bbb黄色大片| 国产精品一区二区精品视频观看| 国产av码专区亚洲av| 亚洲情色 制服丝袜| videos熟女内射| 成人手机av| 国产精品嫩草影院av在线观看| 不卡视频在线观看欧美| 男女国产视频网站| 国产精品久久久av美女十八| 亚洲av成人精品一二三区| a 毛片基地| 免费高清在线观看日韩| 国产av国产精品国产| 午夜免费观看性视频| 精品久久久久久电影网| 国产无遮挡羞羞视频在线观看| 久久精品国产亚洲av高清一级| 黄色怎么调成土黄色| 国产淫语在线视频| 久久久久视频综合| 高清在线视频一区二区三区| 日本午夜av视频| av女优亚洲男人天堂| 国产免费现黄频在线看| 人人妻人人添人人爽欧美一区卜| 精品国产一区二区三区四区第35| 国产一区二区在线观看av| 国产亚洲一区二区精品| 国产成人精品久久二区二区91 | 深夜精品福利| 国产精品 欧美亚洲| 日日啪夜夜爽| 在线观看免费午夜福利视频| 丰满饥渴人妻一区二区三| 欧美xxⅹ黑人| 婷婷色综合www| 久久精品国产亚洲av涩爱| 欧美中文综合在线视频| 精品亚洲成a人片在线观看| 国产黄频视频在线观看| 久久人人爽人人片av| 免费女性裸体啪啪无遮挡网站| 一区二区三区四区激情视频| 尾随美女入室| www.熟女人妻精品国产| 亚洲国产看品久久| 最近最新中文字幕免费大全7| 亚洲精品中文字幕在线视频| 丝袜人妻中文字幕| a 毛片基地| 国产精品av久久久久免费| 久久99热这里只频精品6学生| 国产淫语在线视频| 一级毛片 在线播放| 国产精品久久久久久精品古装| 欧美精品人与动牲交sv欧美| 99久国产av精品国产电影| 久久久国产精品麻豆| 黑人欧美特级aaaaaa片| 1024香蕉在线观看| 国产人伦9x9x在线观看| 秋霞伦理黄片| 亚洲综合色网址| 亚洲国产精品国产精品| 国产欧美日韩综合在线一区二区| 婷婷色av中文字幕| 免费人妻精品一区二区三区视频| 国产老妇伦熟女老妇高清| 黑丝袜美女国产一区| 久久青草综合色| 国产淫语在线视频| 国产成人免费观看mmmm| 久久久久久人人人人人| 亚洲人成电影观看| a 毛片基地| 午夜福利免费观看在线| 国产人伦9x9x在线观看| 少妇精品久久久久久久| 一区二区av电影网| 老汉色∧v一级毛片| 国产成人精品无人区| 一区二区三区四区激情视频| 尾随美女入室| 熟女少妇亚洲综合色aaa.| 亚洲专区中文字幕在线 | 国产精品国产三级国产专区5o| 久久久国产精品麻豆| 久久久久人妻精品一区果冻| h视频一区二区三区| 人人妻人人澡人人看| 久久久久国产精品人妻一区二区| 亚洲av日韩在线播放| 日日摸夜夜添夜夜爱| 日韩电影二区| 国产成人免费观看mmmm| 天天躁夜夜躁狠狠躁躁| 街头女战士在线观看网站| 国产在视频线精品| 午夜福利在线免费观看网站| 午夜免费鲁丝| 黄片小视频在线播放| 国产一区亚洲一区在线观看| 嫩草影视91久久| av在线观看视频网站免费| 欧美 亚洲 国产 日韩一| 欧美人与善性xxx| 中文精品一卡2卡3卡4更新| 天天躁夜夜躁狠狠躁躁| 街头女战士在线观看网站| 一边摸一边抽搐一进一出视频| 老司机影院毛片| 91精品伊人久久大香线蕉| 久久鲁丝午夜福利片| 天堂8中文在线网| 免费看不卡的av| 免费av中文字幕在线| 国产色婷婷99| 啦啦啦在线观看免费高清www| 国产亚洲av片在线观看秒播厂| 久久久精品国产亚洲av高清涩受| 亚洲成人国产一区在线观看 | 国产亚洲av高清不卡| 天天躁夜夜躁狠狠久久av| 老司机影院成人| 在线天堂最新版资源| 亚洲婷婷狠狠爱综合网| 精品国产露脸久久av麻豆| 久久精品国产综合久久久| 美女高潮到喷水免费观看| av在线播放精品| 日本黄色日本黄色录像| 亚洲,一卡二卡三卡| 捣出白浆h1v1| 在线观看免费高清a一片| 久久久久精品人妻al黑| 黄色怎么调成土黄色| 亚洲在久久综合| 观看美女的网站| 一本一本久久a久久精品综合妖精| 在线观看国产h片| kizo精华| 天堂8中文在线网| 欧美人与善性xxx| 人人妻人人添人人爽欧美一区卜| 啦啦啦 在线观看视频| 久久久久精品国产欧美久久久 | 在现免费观看毛片| 国产伦人伦偷精品视频| 亚洲一区中文字幕在线| 母亲3免费完整高清在线观看| 美女国产高潮福利片在线看| 一级,二级,三级黄色视频| 亚洲欧美一区二区三区国产| 午夜日本视频在线| 免费看av在线观看网站| 日韩伦理黄色片| 熟妇人妻不卡中文字幕| 极品人妻少妇av视频| 亚洲伊人色综图| 日日撸夜夜添| 搡老岳熟女国产| 大码成人一级视频| 伊人久久国产一区二区| av又黄又爽大尺度在线免费看| 国产日韩欧美在线精品| 国产精品欧美亚洲77777| 久久精品aⅴ一区二区三区四区| √禁漫天堂资源中文www| 两个人免费观看高清视频| 一级毛片电影观看| 熟女av电影| 国产成人av激情在线播放| 老汉色av国产亚洲站长工具| 日日摸夜夜添夜夜爱| 看十八女毛片水多多多| 婷婷色综合www| 最近最新中文字幕大全免费视频 | 考比视频在线观看| 人人澡人人妻人| 99re6热这里在线精品视频| 久久精品久久精品一区二区三区| 女性生殖器流出的白浆| 午夜福利影视在线免费观看| 欧美精品一区二区大全| 青春草国产在线视频| 国产精品偷伦视频观看了| 婷婷色av中文字幕| 亚洲av成人精品一二三区| 精品人妻一区二区三区麻豆| 日韩中文字幕视频在线看片| 80岁老熟妇乱子伦牲交| 久久热在线av| 看免费成人av毛片| 好男人视频免费观看在线| 精品国产露脸久久av麻豆| 2021少妇久久久久久久久久久| 亚洲一码二码三码区别大吗| 国产亚洲av高清不卡| 国产精品无大码| 亚洲国产成人一精品久久久| 国产成人免费观看mmmm| 久久婷婷青草| 人人澡人人妻人| 精品酒店卫生间| 色播在线永久视频| 久久99一区二区三区| 一二三四在线观看免费中文在| 亚洲av国产av综合av卡| 蜜桃国产av成人99| 一区二区av电影网| 精品一区二区三卡| 欧美精品av麻豆av| 国产欧美亚洲国产| 一本大道久久a久久精品| 亚洲精品aⅴ在线观看| 一级黄片播放器| 成人黄色视频免费在线看| 中文字幕制服av| 操美女的视频在线观看| 午夜福利乱码中文字幕| 综合色丁香网| 99精国产麻豆久久婷婷| 国产 精品1| 十八禁人妻一区二区| 国产精品久久久久久精品古装| 人人妻人人澡人人看| 午夜福利视频在线观看免费| 国产激情久久老熟女| 香蕉丝袜av| 久久精品久久精品一区二区三区| 大陆偷拍与自拍| 在线 av 中文字幕| 91国产中文字幕| 男人舔女人的私密视频| 国产成人欧美在线观看 | 久久久亚洲精品成人影院| 性色av一级| 日韩不卡一区二区三区视频在线| 欧美av亚洲av综合av国产av | av有码第一页| 不卡av一区二区三区| 午夜日本视频在线| 久久久久久久国产电影| 日韩中文字幕视频在线看片| 宅男免费午夜| 久久 成人 亚洲| av.在线天堂| 97精品久久久久久久久久精品| 制服诱惑二区| 男女午夜视频在线观看| www.熟女人妻精品国产| 久久鲁丝午夜福利片| 曰老女人黄片| 成人免费观看视频高清| 超碰成人久久| 欧美日韩综合久久久久久| 成年人午夜在线观看视频| 国产精品偷伦视频观看了| 自线自在国产av| 国产男人的电影天堂91| 亚洲成av片中文字幕在线观看| 好男人视频免费观看在线| 精品福利永久在线观看| 妹子高潮喷水视频| 日本vs欧美在线观看视频| 日韩电影二区| 久久青草综合色| 免费观看人在逋| 少妇人妻精品综合一区二区| 人人妻人人澡人人看| 日韩视频在线欧美| 久久久久久久久免费视频了| 国产av码专区亚洲av| 少妇人妻精品综合一区二区| 日韩不卡一区二区三区视频在线| 一区福利在线观看| 高清不卡的av网站| 午夜福利网站1000一区二区三区| 国产一级毛片在线| 精品久久蜜臀av无| 中文字幕人妻丝袜一区二区 | 十八禁网站网址无遮挡| 2021少妇久久久久久久久久久| 汤姆久久久久久久影院中文字幕| 久久人人97超碰香蕉20202| av网站在线播放免费| 亚洲成av片中文字幕在线观看| 一级毛片我不卡| 国产黄色免费在线视频| 亚洲美女搞黄在线观看| 久久久国产一区二区| 老汉色av国产亚洲站长工具| 日本欧美国产在线视频| 欧美日韩视频精品一区| 亚洲欧美色中文字幕在线| 9色porny在线观看| 中文字幕人妻丝袜制服| 韩国高清视频一区二区三区| 日韩大片免费观看网站| 蜜桃在线观看..| 9191精品国产免费久久| 最新的欧美精品一区二区| 亚洲精品一区蜜桃| 久久久精品94久久精品| 日韩熟女老妇一区二区性免费视频| av女优亚洲男人天堂| 少妇被粗大猛烈的视频| 免费黄网站久久成人精品| 99久久精品国产亚洲精品| 免费看不卡的av| 啦啦啦在线免费观看视频4| 色94色欧美一区二区| 毛片一级片免费看久久久久| 日韩一本色道免费dvd| www.自偷自拍.com| 男女国产视频网站| 黑丝袜美女国产一区| 久久人人爽人人片av| 性少妇av在线| 亚洲成人国产一区在线观看 | tube8黄色片| 欧美国产精品一级二级三级| 五月开心婷婷网| 国产黄色视频一区二区在线观看| 日韩精品有码人妻一区| 久久毛片免费看一区二区三区| 99久久99久久久精品蜜桃| 视频在线观看一区二区三区| 亚洲伊人色综图| 在线观看国产h片| 精品国产露脸久久av麻豆| 久久久精品区二区三区| 久久久久国产一级毛片高清牌| 老汉色av国产亚洲站长工具| 国产精品久久久久久精品电影小说| 日韩伦理黄色片| 国产av一区二区精品久久| 国产免费一区二区三区四区乱码| 永久免费av网站大全| 999久久久国产精品视频| 黑人猛操日本美女一级片| 老司机影院毛片| 一区二区三区激情视频| 在线观看三级黄色| 五月开心婷婷网| 亚洲五月色婷婷综合| 国产97色在线日韩免费| 十八禁网站网址无遮挡| 亚洲欧美中文字幕日韩二区| 日韩精品有码人妻一区| 蜜桃在线观看..| 我要看黄色一级片免费的| 日本av免费视频播放| 日韩电影二区| 亚洲婷婷狠狠爱综合网| 大香蕉久久成人网| 欧美97在线视频| 中文字幕av电影在线播放| 麻豆av在线久日| 国产亚洲av片在线观看秒播厂| 天美传媒精品一区二区| 成人毛片60女人毛片免费| www.自偷自拍.com| 国产一区二区 视频在线| 最近最新中文字幕大全免费视频 | 久久女婷五月综合色啪小说| 久久人人爽人人片av| 少妇的丰满在线观看| 日本黄色日本黄色录像| 亚洲欧美精品综合一区二区三区| e午夜精品久久久久久久| 精品人妻熟女毛片av久久网站| 97人妻天天添夜夜摸| 日日撸夜夜添| 久久鲁丝午夜福利片| 久热爱精品视频在线9| 国产麻豆69| 99re6热这里在线精品视频| 黄色视频在线播放观看不卡| 少妇人妻 视频| 国产成人一区二区在线| 天天影视国产精品| 丁香六月欧美| 免费高清在线观看视频在线观看| 久久国产精品男人的天堂亚洲| 毛片一级片免费看久久久久| 成人手机av| 欧美黑人精品巨大| 亚洲欧美清纯卡通| 日韩精品有码人妻一区| 欧美黄色片欧美黄色片| 黄色视频在线播放观看不卡| 日韩大片免费观看网站|