• <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 ?

    Twisted Pair Cable Fault Diagnosis via Random Forest Machine Learning

    2022-08-23 02:20:08GhazaliSemanIsaRamliAbidinMustamHaekAbidinandAsrokin
    Computers Materials&Continua 2022年6期

    N.B.Ghazali,F.C.Seman,,K.Isa,K.N.Ramli,Z.Z.Abidin,S.M.Mustam,M.A.Haek,A.N.Z.Abidin and A.Asrokin

    1Faculty of Electrical and Electronic Engineering,Universiti Tun Hussein Onn Malaysia,Batu Pahat,Johor

    2TM Research&Development,Cyberjaya,Malaysia

    Abstract:Applying the fault diagnosis techniques to twisted pair copper cable is beneficial to improve the stability and reliability of internet access in Digital Subscriber Line (DSL) Access Network System.The network performance depends on the occurrence of cable fault along the copper cable.Currently,most of the telecommunication providers monitor the network performance degradation hence troubleshoot the present of the fault by using commercial test gear on-site, which may be resolved using data analytics and machine learning algorithm.This paper presents a fault diagnosis method for twisted pair cable fault detection based on knowledge-based and data-driven machine learning methods.The DSL Access Network is emulated in the laboratory to accommodate VDSL2 Technology with various types of cable fault along the cable distance between 100 m to 1200 m.Firstly, the line operation parameters and loop line testing parameters are collected and used to analyze.Secondly,the feature transformation,a knowledge-based method,is utilized to pre-process the fault data.Then, the random forests algorithms (RFs), a data-driven method, are adopted to train the fault diagnosis classifier and regression algorithm with the processed fault data.Finally, the proposed fault diagnosis method is used to detect and locate the cable fault in the DSL Access Network System.The results show that the cable fault detection has an accuracy of more than 97%, with less minimum absolute error in cable fault localization of less than 11%.The proposed algorithm may assist the telecommunication service provider to initiate automated cable faults identification and troubleshooting in the DSL Access Network System.

    Keywords:Twisted pairs;random forest machine learning;cable fault;DSL

    1 Introduction

    Worldwide fixed access technologies depend on fiber-optic and copper-wires technologies to deliver high-speed internet access to the end customers.Fiber-optic lines offer much higher data rates than copper wires that is becoming more common for bringing broadband Internet access to homes.However,with the latest DSL Technologies,internet access can offer up to 100Mbps within a specific distance.In Malaysia,the network dependencies on the fixed access technologies to Fiber to The Home(FTTH)and DSL is about 40%and 60%,respectively.The fiberisation national plan has increased to 70%,and that allows high-speed internet access using copper wires within less than 1.5 km distance to the fiber-optic backhaul,especially in the suburban area.The offered data speed using copper wires is also subjected to the cable conditions along the access network, and cable fault due to the impedance mismatch becomes the most concern.Therefore, the copper cable fault diagnosis,network monitoring,and qualification on the DSL access network systems are necessary to increase and maintain the network performance.

    The popular copper line fault detection is based on Time Domain Reflectometry (TDR).The commercial test gear is normally attached to the copper line for line fault detection.TDR is a wellknown method to measure the characteristics of the electrical lines,impedance of discontinuities as a function of time or distance,then reflection signal can be translated to determine the faults caused by a splice,cable transition,and mismatched cable connections[1].This method is reliable to detect the fault type and localization;however,test gears are needed with manual intervention to the field site.In [2], a TDR simulator is proposed to assist loop diagnostics on copper twisted pairs based on the transmission line ABCD parameter modelling, which also needs on-site simulation and verification during DSL network maintenance and troubleshooting.

    The loop diagnostics based on the scattering parameters at microwave frequencies allows the backscattering model for twisted pairs is established in[3],where local deterioration of transmission parameters, faults can be precisely localized and instantly addressed by comparing with a reference measurement.The copper cables transmission line characteristics based on the resistance,inductance,conductance, and capacitance per unit length (RLGC) circuit [4] are exploited to investigate the impact of cable bleeding of twisted pair towards the achievable bit rate.In[1],the copper line quality assessment process is proposed to assist the technical operation team and network access planner before deployment of the high-speed broadband services to the subscribers at High Speed Broadband(HSBB),Suburban Broadband(SUBB)or Rural Broadband(RBB)areas.

    Recently,the technologically advanced techniques exploiting fault diagnosis approach based on machine learning.Feature extraction and clustering analysis become the framework of fault detection,classification,and localization,which are mapped based on the electrical signals along the cables[5].In the case where the data clustering is capable of distinguishing ideal and faulty conditions,there is no need to implement an additional fault detection method.The data clustering can be performed based on unsupervised or supervised machine learning, where the algorithm interprets the input of data and discovering the group of data from the featured data.Automated analysis techniques [5,6]are demonstrated to detect,locate,and diagnose the fault using Support Vector Machine[7],Neural Network[8],and Random Forest[9].

    In[10],an intelligent fault diagnosis machine learning method using random forests algorithms for a three-phase power-electronics energy conversion system based on knowledge-based and data-driven methods is developed.The slopes of two current trajectories of three-phase AC currents are adopted to train the fault diagnosis classifier based on a data-driven method (the random forest algorithm),which has the adaptive ability to different loads.In[9],a new Convolutional Neural Network(CNN)using the Random Forest (RF) classifier is proposed for hydrogen sensor fault diagnosis.The datadriven is established where the input 1-D time-domain data of fault signals are converted into 2-D gray matrix images,and later the features extraction by using CNN are processed into the random forest classifier.Analytical-driven based on the time-frequency domain reflectometry signals is proposed[11]to detect the presence and the location of a fault and further differentiates the faulty line within the multi-core control and Instrumentation(C&I)cables.Neural networks and the hierarchical clustering algorithm are used, and the algorithm is verified via experiments with four possible fault scenarios using automotive wires and C&I cables for nuclear power plants.

    In the sequence to the fault localization, the conventional methods can be categorised into impedance focused methods and travelling wave based methods.Depending on the source of data,cable fault location methods may be further divided into single-ended technique and double-ended technique [5].In [12], the performance of different regression methods, which are linear regression and regression trees,are developed on a hybrid power system.Different faults with random distances on the transmission line were simulated and a fault database was created by recording the current and voltage signals of these faults.The same applications are deployed in intelligent fault location[13]based on the integration of Artificial Neural Network(ANN)and fuzzy expert system called Adaptive Network-Based Fuzzy Inference System (ANFIS).The fault location error is very minimum, which are ANFIS network(0.0012),followed by Gaussian process regression(0.0684)and linear regression(0.4092).

    From the literature studies that have been discussed earlier, the fault diagnosis tools on many applications have been established by using machine learning algorithms.However,the copper cable faults detection still relies on the manual detection by using on-site commercial test gear and to date, there is no established fault diagnosis that focuses on the DSL Access Network that has been discussed.The motivation of the work also considers that most of the telecommunication providers still troubleshoot the present of the cable fault by using commercial test gear on-site, which may be resolved using data analytics and machine learning algorithm and catered by the proposed diagnostic tools presented in here.In this paper, data-driven methods assisted with knowledge-driven methods are adopted to interpret the network parameters, which are gathered based on a single line testing in twisted pair cable for fault diagnosis.The DSL access network system is collected and used to analyze for initial observation.Later, based on the knowledge-based method, the featured data are extracted,which consists of line operation parameters(LOP)and loop line testing parameters(LLT).The processing data are trained into the random forest algorithms to determine the copper cable fault type and localization.The accuracy of the algorithm is analyzed, and the decision criteria based on the LOP and LLT are evaluated.

    2 General Background on VDSL 2 Technologies

    A typical VDSL2 network infrastructure consists of central office (CO), Multi-Service Access Network (MSAN), cabinet, distribution point (DP), and customer premises [14], as illustrated in Fig.1.In the VDSL2 copper access networks, fiber cables become the backbone infrastructure that connects the CO and MSAN.The copper cables are laid between MSAN and the customer premises through the DP.The VDSL2 network cards and Plain Old Telephone Service Line(POTS)cards are integrated at MSAN, and this is the important key factor that allows both telephone and internet services to be offered using the same copper cables infrastructure[15].This network topology is usually established at the suburban site where the capital expenditure of the telco needs to reflect the customer’s populations in that particular area.

    Figure 1:Experimental platform of DSL Access Network System

    Very-High-Speed Digital Subscriber Line 2 (VDSL2) is a family of DSL Technologies that allow fast internet connection via legacy copper lines.VDSL 2 co-exist from VDSL and falls under International Telecommunication Union(ITU-T)G.993.2[16].There are several types of parameter profiles in VSDL2 technologies.In this paper,the VDSL2 Profile 17a was exploited,using modulation method of Discrete Multi-Tone(DMT),where data is carried in the 4095 tone spacing of 4.3125 kHz within a spectrum bandwidth of 17.664 MHz.The modulation technique allows a separation between the upstream and downstream data signals.This VDSL2 network configuration can support up to 100 Mbps data rates for distance up to 1.5 km [17].However, the possible offered data rates is also affected by the nature of the copper cables itself where the attenuation degrades rapidly with distances and the presence of external electromagnetic interferences which leads to crosstalk.The most common cable faults caused by the imbalance impedance normally occur at the cable jointing and improper termination line at the customer premises.In the field site,these cable faults can be identified through manual inspection by using commercial test gears[18].

    The cables faults that may occur in the copper loop will modify LLT parameters which consist of primary elements such are resistance and capacitance,and secondary parameters such as current and voltage[19,20].In any case,where the cable faults are not severe,the internet service is still accessible by the subscriber, perhaps at lower data speed.In this case, the VDSL2 network conditions can be assessed from valid LOP parameters,which are the speed rate,attenuation,and signal to noise ratio(SNR)[21].Note that the LOP and LLT will be used as the parameter attributes for the copper fault detection and localization.To date,there are not many papers emphasize the usage of machine learning for cable fault classification in the VDSL2 copper access network.

    3 DSL Experimental Platform and Data Acquisition

    The experimental platform DSL access network system shown in Fig.2 consists of a traffic generator,MSAN,copper cable binder,and modems.The experimental platform is set up to imitate the actual deployment for the DSL access network system in the field site.Note that the cable binder contains ten unshielded twisted pairs(UTP Cat-3)which are connected to the 10 modems to represent the network termination at the customer premises.The cable faults are realized along the cable at the tag block or cable jointing.The raw data are gathered from MSAN using the command prompt Telnet script.These raw data are in text file (.txt format) and later processed into comma separated values file(.csv format)for data preparation.

    Figure 2:Experimental platform of DSL Access Network System

    3.1 Twisted Pair RC Lumped Elements

    The twisted pair can be modelled as the RLGC transmission line model.However,the transmission line model can be simplified as RC lumped components,where other available components are neglected in the electrical schematic due to insignificant changes when different types of cable faults are emulated.This is carried out when the initial data observation was conducted.The twisted pair consists of Tip(A)and Ring(A),and the resistive and capacitance components(RC)[22]as illustrated in Fig.3.Each A and B is connected to the Ground (G), which are represented by component Resistances,RA-B,RA-G,and RB-G;and Capacitances,CA-B,CA-G,and CB-G.Each A and B component has the same component of RC for ideal cable condition but may change when the cable fault occurs along the line[23].In MSAN, these components are categorized under LLT parameters.In the typical case where the internet data is accessible by the modem,the network performances are also provided by MSAN under LOP parameters.

    Figure 3: Twisted pair RC, Tip (A), Ring (B), and Ground (G) component based on RC lumped component elements

    3.2 Cable Fault Realization

    Cable fault imitation in the DSL Access Network can be imitated by understanding the scenario of the physical connection.In normal cases,the networks are in ideal condition without any cable fault occurrence where the access to the internet is at the optimum condition.Tab.1 describes the scenario of the five copper cable faults that occur in the typical DSL Access Network.The cable faults are emulated at the cable jointing, which are conducted on the tag block.The standard practice in the telecommunication industry is to use commercial test gears to troubleshoot the present of the cable fault and its localization.For on-site installation and repair, the EXFO test gear is commonly used by the telco engineer to observe the DSL network performance, therefore here, the same test gear was used to verify the cable fault emulations conducted in the laboratory.This DSL copper cable test gear features the traditional copper measurements,which involve voltage,resistance,capacitance,and time domain reflectometry and also correlated with featured parameters given by MSAN.This testing is very important to make sure that the impairment emulations have the same characteristics as the actual cable fault existing in the field site.

    Table 1: Data gathering for various cable conditions

    During the fault emulation,the group of data are labelled based on the scenario.Each of the cable faults would either degrades the internet speed,but the line condition is still in the‘Activated’mode or disrupts the network,causing the line condition in an‘Activating’mode.However,Partial open fault has triggered the network condition in both activated and activating mode,and this depends on how severe the fault is,especially in a longer cable length.In Activated mode,both LOP and LLT can be used as the dataset for the machine learning algorithm;however,in the activating mode,the internet access is disrupted,and only LLT is valid.

    4 Realization of Fault Diagnosis Using Machine Learning

    The overall flow chart of the fault diagnosis using machine learning algorithm is illustrated in Fig.4.The algorithm consists of importing a dataset which is based on the featured parameters,data pre-processing,training the build algorithm,testing on the developed algorithm,and analysis of fault detection accuracy.

    Figure 4:Flow chart of RF algorithm

    4.1 Selection of Parameters and Pre-Processing

    There are many primary parameters which is categorized as LLT parameters and LOP parameters,that are captured in MSAN during network operation.However,only the most significant parameters are selected for the cable fault detection by using feature important selection.Due to abbreviate the details of the important feature selection will not be discussed here.Out of all LLT and LOP parameters,only six LOP and LLT parameters are selected for cable fault detection and localization as tabulated in Tab.2.

    Table 2: LOP and LLT parameters

    These parameters go through the data-preprocessing stage, which transforms the data into a suitable format for machine learning algorithms.The collected data are prepared, the categorical parameters and any invalid or missing data that may affect the accuracy are handled.The targeted parameters, which are the labelled fault type and localization, are determined.Depending on the presence of cable fault in the network,the line can be either in the activated or activating as explained in Section 3.2.For activating case, the LOP captured by MSAN is only historical data and is no longer valid to represent the current line condition.In some other cases as well, such as when cable fault is present such as short and partial short,the values of capacitance either A-G or B-G are not available.The same trend is applicable to the values of resistance A-G or B-G for open and partial open conditions.These parameters are replaced with a specific value beyond the actual values.The values of the parameters are also standardized within the range of features of the input data set.

    4.2 Machine Learning Algorithm

    Initially,there are several supervised machine learning classification algorithms that were explored using WEKA,which are Artificial Neural Network,Na?ve Bays,Random Forest(RF),and k-Nearest Neighbor and Decision tree.WEKA is an open source software that was developed for data mining tasks using Java languages [24].It contains tools for data preparation, classification, regression,clustering,association rules mining,and visualization which are very useful for initial machine learning exploration in many applications.At first,MSAN data were exported into WEKA software and the outcome showed that RF algorithm provided the highest accuracy which is about 80%.Therefore,RF algorithm is finalized for further development due to its better accuracy in classifying fault based on the electrical characteristics of the MSAN dataset.There are two algorithms developed for fault diagnosis.Firstly, the RF classifier is used to detect the fault type, secondly is the RF regression method,which is used to localize the location of the fault.

    The RF algorithm is an ensemble learning method based on a decision tree.The split train test method of 70:30 is utilized.In a dataset,a training set is implemented to build up the algorithm,while a testing data set is to validate the model built.Out of 2738 datasets,each tree randomly selects 70%of training samples as a sub-training set.There are six features to be selected for an optimal feature from each split.Therefore, each tree can obtain training results according to different sub-training sets.Each input sample is determined separately,and the final classification is determined according to the voting results.Then the deployed RF model is reliable to classify in large datasets.In the case of fault localization, the averaging of the results of each decision tree is made.The accuracy of the model to detect impairment and localization are determined,and for the case,if the accuracy is less than 90%,the algorithm will be re-trained.The RF model is shown in Fig.5.

    Figure 5:Flow chart of RF algorithm

    5 Data Analysis

    The proposed cable fault diagnosis is conducted based on the knowledge-driven and data-driven.The knowledge-driven of the LLT and LOP parameters trending is carried out by observing the collected data.Then the accuracy fault detection and localization algorithm is carried out based on the data-driven.

    5.1 Data Observation

    Even though the fault detection location and localization algorithm are developed based on machine learning,the understanding of the data behavior is very important as it reflects the electrical characteristics of the lines.Fig.6 shows samples of the features selection over the distance.These parameters are shown as it obviously illustrating the changes of LLT and LOP when the faults are emulated.Figs.6a and 6b shows the maximum attainable inversely proportional with cable distance while the signal attenuation increases proportionally with cable distance.Both parameters show a linear correlation with distance despite outliers in the few cases,especially when the cable distance is less than 300 m.The maximum attainable rate of the line reaches more than 100 Mbps at 100 m distance and reduces to 30Mbps at 1000 m.This linear correlation can be represented by y=-47.21 x+87300.In average,the attainable rate for BT condition is slightly 10%to 20%lower than the ideal condition.While for signal attenuation,the average signal attenuation is 5 dB at 100 m and gradually increases to 20 dB at 1000 m.The correlation also can be represented by y=0.0143x+5.6942.Note that both attainable rate and signal attenuation for ideal and BT cases are overlapping and need to depend on other features to differentiate.Fig.6c illustrates the RA-Gfor Ideal case, BT case, Partial Short and Short case,and the parameter is not available for Partial Open and Open case.This is due to short and Partial Short conditions introduce an extremely low impedance at the cable jointing hence reflects the overall RA-G.This significant data trend is expected and used by the machine learning during dataset training.The similar trend is observed for CA-Bwhere the data is not available for cable fault short and partial short as shown in Fig.6d.Open and partial open conditions also produce an average 20%lower CA-Bfrom Ideal and BT conditions.Fig.6e is highlighted to show the trend of the CA-G.Note that the linear correlation,which is represented by y=0.0595x–6.777 for all cable conditions may be used to determine the fault localization in the regression techniques.Note that there are six LOP and LLT parameters are involved in the development of machine learning however, only five parameter selections are shown here due to distinct noticeable data pattern that can be observed to represent each fault type.

    5.2 Performance of Cable Fault Detection Algorithm

    Random forest algorithm is implemented for cable fault classification.The classification models are conducted using the Sklearn pickle model to load the trained dataset and later detect the new testing data.Tab.3 shows the confusion matrix of the cable fault detection with Random Forest Classifier.The confusion matrix illustrates the classification results of all cable conditions in detail,including both classification and misclassification fault types.The top row of the confusion matrix represents the actual label of classification, whereas the first column represents the detected label,and the diagonal value of the confusion matrix up to the seventh column represents the matching or mismatching sample number between the actual and the detected condition.

    Figure 6:Selection of LLT and LOP parameters trend with cable faults;Ideal,Bridge Tap,ShortPartial Short,Openand Partial Open

    Tab.3 shows the classification performance of the VDSL dataset that is done using the train-test split method function in python programming.The VDSL dataset samples are also split into 67%of the training dataset and 33% of the testing dataset.The classification of VDSL fault type gives out 97% using 2738 testing samples of VDSL data which each fault location sample are separated into distances up to 1200 m.The classifications of short and partial short fault types seem to result in higher accuracies; this is due to their quite distinct differences in parameter values such as RA-Gas shown in Fig.6c compared to bridge tap and ideal VDSL condition.The lowest precision is 91%and 93%,which occurs for open conditions and partial open.Despite this,the average accuracy of the proposed method,which is 97%,indicates that the proposed method is capable of cable fault diagnosis.

    5.3 Performance of Cable Fault Localization Algorithm

    The localization of the cable fault that may occur in the DSL Access Network is conducted in a separate machine learning algorithm which is the Random Forest regression algorithm.The regression model is conducted using the Sklearn pickle model to load the trained dataset and later detect the new testing data,which is similar to the detection algorithm.Note that for ideal condition,the length of the distance represents the actual of cable distance between MSAN and cable termination before the modem.While for other fault conditions it will represent the fault location distance from the MSAN.The accuracy of the localization is assessed from the scale-dependent metrics, which are Minimum Absolute Error(MAE),Minimum Absoluter Percentage Error(MAPE),and Root Mean Square Error(RMSE).MAE is a metric used to measure the average magnitude of the absolute errors between the predicted fault location and actual location.While for MAPE, the absolute errors are normalized to the actual location to give a percentage error.RMSE measures the average magnitude of error between the predicted fault location and actual location.Thus,RMSE is the average distance measured vertically from the actual value to the corresponding predicted value on the fit line.In comparison with MAE,the RMSE has a relatively high weight for large errors because the errors are squared before averaging,as shown in Tab.4.

    Table 4: Accuracy of cable fault localization using random forest regression

    Tab.4 also shows the average deviation of location based on MAPE for all cable faults is less than 6.6%.However,for Bridge Tap and Open faults,the accuracy of the cable fault localization is about 11%.The accuracies are measured based on the magnitude of error but do not indicate the direction of error.

    5.4 Comparison with Other Diagnostic Tool on DSL Access Network

    The coppers wires are vulnerable,prone to signal leakage and interference issues,the offered data speed using coppers wires is also subjected to the cable conditions along the access network and cable fault becomes the most concern.Therefore, the copper cable fault diagnosis, network monitoring,and qualification on the DSL access network systems are necessary.To evaluate the abilities of the proposed fault diagnosis method,the proposed method was compared with other diagnosis methods on DSL Access Network as shown in Tab.5.The diagnostic tools that are based on TDR which also called Single Ended Loop Testing(SELT)provides high accuracy in the fault location identification,however it is a destructive testing where a disconnection to the existing network is required during the assessment.The frequency domain concept generally incurs Double Ended Loop Testing(DELT)where the cable tapping is needed on the starting point and the ending point of the network.Obviously,this is not practical for on-site testing.All above stated concepts also require prior domain knowledge during the assessment.The most significant difference between the previous methods and the proposed method are the utlisation on the machine learning method that (1) it can diagnose both offline and online fault cables;(2)it is nonintrusive and nondestructive method;and(3)it needs no prior domain knowledge.

    Table 5: Comparison with other fault diagnosis methods

    Table 5:Continued

    6 Conclusion

    In this paper, the copper cable faults classification and localization with Random Forest were carried out using Python programming.The results showed that the proposed machine learning algorithm able to provide accuracies up to 97% in the fault classification and average 6% difference in fault localization.Based on the existing literature, the accuracy is considered excellent for fault diagnosis.The proposed solution allows the telecommunication service’s provider to troubleshoot the present of the cable fault online instead of relying on the commercial test gear on-site.In the future,the cable fault diagnosis may be extended by considering the possibilities of having multiple faults in the same cable.

    Acknowledgement:The authors wish to acknowledge the funding received from Ministry of Science,Technology and Innovation Malaysia, and Research Management Centre (RMC) of Universiti Tun Hussein Onn Malaysia (UTHM), the research collaborator of the grant which is Telekom Malaysia Research and Development (R&D) Sdn Bhd.Communication of this research is made possible through monetary assistance by Universiti Tun Hussein Onn Malaysia and the UTHM Publisher’s Office via Publication Fund E15216.

    Funding Statement:The authors received the funding from Smart Challenge Fund (SR0218I100)and GPPS Grant VOT H404, from Ministry of Science, Technology and Innovation Malaysia, and Research Management Centre(RMC)of Universiti Tun Hussein Onn Malaysia(UTHM)

    Conflicts of Interest:The authors declare that they have no conflicts of interest to report regarding the present study.

    波野结衣二区三区在线| av黄色大香蕉| 亚洲国产精品国产精品| 午夜精品一区二区三区免费看| 老女人水多毛片| 大又大粗又爽又黄少妇毛片口| 少妇的逼水好多| 中文资源天堂在线| 亚洲,欧美,日韩| 少妇高潮的动态图| .国产精品久久| 免费观看在线日韩| 免费在线观看影片大全网站| 色5月婷婷丁香| 久久精品国产亚洲av涩爱 | 麻豆国产av国片精品| 亚洲无线在线观看| 国产成年人精品一区二区| 久久精品久久久久久噜噜老黄 | 久久这里只有精品中国| 亚洲无线观看免费| 国产欧美日韩精品一区二区| 国产aⅴ精品一区二区三区波| 99久国产av精品| 非洲黑人性xxxx精品又粗又长| 午夜免费激情av| 国模一区二区三区四区视频| 听说在线观看完整版免费高清| 日韩三级伦理在线观看| 少妇被粗大猛烈的视频| 看非洲黑人一级黄片| 中国国产av一级| 99热6这里只有精品| 亚洲在线观看片| 亚洲专区国产一区二区| 久久久久久伊人网av| 丝袜美腿在线中文| 极品教师在线视频| 国产精品一区二区三区四区免费观看 | 听说在线观看完整版免费高清| 简卡轻食公司| 亚洲成av人片在线播放无| 日本免费一区二区三区高清不卡| 国产一区二区亚洲精品在线观看| 欧美xxxx性猛交bbbb| 国产探花在线观看一区二区| 国产三级中文精品| 欧美性感艳星| 免费在线观看成人毛片| 久久亚洲国产成人精品v| 午夜福利高清视频| 18禁在线播放成人免费| 中文字幕免费在线视频6| 久久精品国产亚洲av涩爱 | 久久6这里有精品| 性欧美人与动物交配| 尾随美女入室| 18禁黄网站禁片免费观看直播| 舔av片在线| 久久人人爽人人片av| 国产欧美日韩精品一区二区| 亚洲精品一区av在线观看| 淫秽高清视频在线观看| aaaaa片日本免费| 午夜视频国产福利| 亚洲精品一区av在线观看| 国产精品女同一区二区软件| 久久久久性生活片| 麻豆一二三区av精品| 成人毛片a级毛片在线播放| 丰满的人妻完整版| 99久久精品国产国产毛片| 久久久a久久爽久久v久久| .国产精品久久| 草草在线视频免费看| 午夜老司机福利剧场| 国产精品伦人一区二区| 老师上课跳d突然被开到最大视频| 一级av片app| 99在线视频只有这里精品首页| 狠狠狠狠99中文字幕| av在线播放精品| 91久久精品国产一区二区成人| 日日干狠狠操夜夜爽| 美女黄网站色视频| 国产精品一区二区三区四区久久| 黄色视频,在线免费观看| 色播亚洲综合网| 国产精品久久久久久av不卡| 亚洲精品影视一区二区三区av| 亚洲内射少妇av| 国国产精品蜜臀av免费| 亚洲18禁久久av| 99热网站在线观看| 国产私拍福利视频在线观看| 色哟哟·www| 日日摸夜夜添夜夜添小说| 国产精品爽爽va在线观看网站| 国产真实伦视频高清在线观看| 少妇丰满av| 日本精品一区二区三区蜜桃| 麻豆国产97在线/欧美| 国产精品免费一区二区三区在线| 国产伦精品一区二区三区视频9| 亚洲美女黄片视频| 国产精品无大码| 亚洲精品成人久久久久久| 日本色播在线视频| 搡老岳熟女国产| 久久精品国产亚洲av香蕉五月| 天堂动漫精品| 国产精品精品国产色婷婷| 久久久国产成人免费| 国产精品永久免费网站| 97超碰精品成人国产| 简卡轻食公司| 五月玫瑰六月丁香| 国产精品av视频在线免费观看| 丰满的人妻完整版| 免费大片18禁| 亚洲乱码一区二区免费版| aaaaa片日本免费| 国产精品久久久久久久久免| 国产av一区在线观看免费| 99热这里只有精品一区| 国产aⅴ精品一区二区三区波| 天堂影院成人在线观看| 啦啦啦观看免费观看视频高清| 亚洲精华国产精华液的使用体验 | 精品午夜福利在线看| 国产精品久久久久久久久免| 波多野结衣巨乳人妻| .国产精品久久| 美女免费视频网站| 91av网一区二区| 欧美一区二区国产精品久久精品| 日本爱情动作片www.在线观看 | 小蜜桃在线观看免费完整版高清| 日本黄色视频三级网站网址| 在线观看一区二区三区| 变态另类丝袜制服| 成人午夜高清在线视频| av女优亚洲男人天堂| 欧洲精品卡2卡3卡4卡5卡区| 人人妻人人看人人澡| 全区人妻精品视频| 91午夜精品亚洲一区二区三区| 九九久久精品国产亚洲av麻豆| 俄罗斯特黄特色一大片| 日本欧美国产在线视频| 干丝袜人妻中文字幕| 99精品在免费线老司机午夜| 国产老妇女一区| 午夜福利18| 俄罗斯特黄特色一大片| 亚洲成人久久爱视频| 亚洲三级黄色毛片| 夜夜看夜夜爽夜夜摸| 人人妻人人澡欧美一区二区| 我的女老师完整版在线观看| 嫩草影院精品99| 听说在线观看完整版免费高清| 女人十人毛片免费观看3o分钟| 午夜精品在线福利| 一级av片app| 成人精品一区二区免费| 免费无遮挡裸体视频| 午夜福利在线在线| 日本精品一区二区三区蜜桃| 两个人视频免费观看高清| 麻豆国产av国片精品| 精品人妻一区二区三区麻豆 | 免费看a级黄色片| www.色视频.com| 少妇裸体淫交视频免费看高清| 美女cb高潮喷水在线观看| 午夜激情福利司机影院| 国产综合懂色| 可以在线观看毛片的网站| 一级黄色大片毛片| 99国产极品粉嫩在线观看| 免费电影在线观看免费观看| 两个人视频免费观看高清| 亚洲av免费高清在线观看| 晚上一个人看的免费电影| 美女被艹到高潮喷水动态| 亚洲av五月六月丁香网| 2021天堂中文幕一二区在线观| 国产精品国产三级国产av玫瑰| 嫩草影院精品99| 亚洲欧美清纯卡通| 国产精品久久电影中文字幕| 久久久欧美国产精品| 不卡一级毛片| 午夜a级毛片| 俄罗斯特黄特色一大片| 亚洲国产色片| 一级av片app| 国产欧美日韩精品一区二区| 亚洲国产高清在线一区二区三| 18禁黄网站禁片免费观看直播| a级毛片免费高清观看在线播放| 免费人成视频x8x8入口观看| 人妻少妇偷人精品九色| 尤物成人国产欧美一区二区三区| 国产午夜精品论理片| 日本黄色视频三级网站网址| 无遮挡黄片免费观看| 国产蜜桃级精品一区二区三区| 成人漫画全彩无遮挡| 久久人人爽人人片av| 老女人水多毛片| 黄色日韩在线| 国产乱人视频| 精品乱码久久久久久99久播| 69人妻影院| 精品无人区乱码1区二区| 欧美一级a爱片免费观看看| 一个人看的www免费观看视频| 亚洲av中文av极速乱| 十八禁网站免费在线| 嫩草影院入口| 在线看三级毛片| 国产大屁股一区二区在线视频| 国产精品亚洲美女久久久| 精品熟女少妇av免费看| 久久精品夜夜夜夜夜久久蜜豆| 美女xxoo啪啪120秒动态图| 国产成人aa在线观看| 国产大屁股一区二区在线视频| 久久久久国产精品人妻aⅴ院| 国产精品乱码一区二三区的特点| 国产精品久久久久久久久免| av专区在线播放| 少妇熟女aⅴ在线视频| 国产三级中文精品| 在线免费观看的www视频| 天美传媒精品一区二区| 国产精华一区二区三区| 国产av一区在线观看免费| 成人高潮视频无遮挡免费网站| 两个人的视频大全免费| 一进一出好大好爽视频| 两个人视频免费观看高清| 成年免费大片在线观看| 国产在视频线在精品| 日日摸夜夜添夜夜添av毛片| 国产av在哪里看| 国产一区二区在线av高清观看| 99在线视频只有这里精品首页| 亚洲精品456在线播放app| 赤兔流量卡办理| 欧美区成人在线视频| 一级a爱片免费观看的视频| av在线天堂中文字幕| 亚洲四区av| 午夜福利成人在线免费观看| 99久久精品一区二区三区| 亚洲中文字幕日韩| 午夜a级毛片| 亚洲高清免费不卡视频| a级毛色黄片| 一级毛片aaaaaa免费看小| 国产精品久久久久久久电影| 少妇人妻一区二区三区视频| 国产精品一及| 91久久精品国产一区二区三区| 色播亚洲综合网| 国产成人影院久久av| aaaaa片日本免费| 亚洲不卡免费看| 国产真实伦视频高清在线观看| 亚洲精品日韩在线中文字幕 | 国产成人福利小说| 男人舔奶头视频| 欧美成人免费av一区二区三区| 一进一出抽搐gif免费好疼| 亚洲国产色片| 色在线成人网| 亚洲欧美清纯卡通| 成人午夜高清在线视频| 成年女人永久免费观看视频| 有码 亚洲区| 日韩三级伦理在线观看| 精品欧美国产一区二区三| 日韩欧美 国产精品| 六月丁香七月| 99久久精品热视频| 三级男女做爰猛烈吃奶摸视频| 淫妇啪啪啪对白视频| 亚洲成a人片在线一区二区| 卡戴珊不雅视频在线播放| 男人的好看免费观看在线视频| 婷婷色综合大香蕉| 人妻制服诱惑在线中文字幕| 特大巨黑吊av在线直播| 狂野欧美激情性xxxx在线观看| 亚洲真实伦在线观看| a级一级毛片免费在线观看| 国模一区二区三区四区视频| 欧美日本亚洲视频在线播放| 中国美白少妇内射xxxbb| 国产亚洲精品久久久com| 最新中文字幕久久久久| 日韩一区二区视频免费看| 日韩成人av中文字幕在线观看 | 亚洲专区国产一区二区| 国产成人aa在线观看| 国产午夜福利久久久久久| 亚洲成a人片在线一区二区| 成人特级av手机在线观看| 99视频精品全部免费 在线| 麻豆乱淫一区二区| 国产精品av视频在线免费观看| 联通29元200g的流量卡| 亚洲精品影视一区二区三区av| 亚洲成人av在线免费| 一进一出抽搐动态| 国产精品av视频在线免费观看| 成人一区二区视频在线观看| 日本免费a在线| 在线a可以看的网站| 女的被弄到高潮叫床怎么办| 99久久成人亚洲精品观看| 波多野结衣巨乳人妻| 不卡一级毛片| 国产一区二区三区av在线 | 最新中文字幕久久久久| 久久6这里有精品| 国产淫片久久久久久久久| 日本三级黄在线观看| 亚洲av中文字字幕乱码综合| 国产成人a∨麻豆精品| 免费看av在线观看网站| 又黄又爽又刺激的免费视频.| 国产高清不卡午夜福利| 婷婷六月久久综合丁香| 欧美成人a在线观看| 成年女人永久免费观看视频| 免费高清视频大片| 日本熟妇午夜| 成人午夜高清在线视频| 国产av麻豆久久久久久久| 国产精品无大码| 夜夜爽天天搞| 不卡一级毛片| 乱系列少妇在线播放| 2021天堂中文幕一二区在线观| 久久精品国产亚洲网站| 精品不卡国产一区二区三区| 免费搜索国产男女视频| 亚洲无线观看免费| 精品久久久久久久久久免费视频| 国产综合懂色| 免费观看精品视频网站| 久久久精品大字幕| 精品国内亚洲2022精品成人| 欧美国产日韩亚洲一区| 级片在线观看| av专区在线播放| 成人高潮视频无遮挡免费网站| 免费av观看视频| 欧美高清成人免费视频www| 欧美+日韩+精品| 欧美激情久久久久久爽电影| 国产伦在线观看视频一区| 国产精品不卡视频一区二区| 欧美激情久久久久久爽电影| 欧美不卡视频在线免费观看| 久久精品91蜜桃| 寂寞人妻少妇视频99o| 夜夜看夜夜爽夜夜摸| 蜜臀久久99精品久久宅男| 村上凉子中文字幕在线| 亚洲精品一区av在线观看| 国产免费一级a男人的天堂| 2021天堂中文幕一二区在线观| 亚洲精品日韩av片在线观看| 男女下面进入的视频免费午夜| 我要搜黄色片| 99热这里只有是精品在线观看| 国产亚洲精品av在线| 欧美又色又爽又黄视频| 亚洲国产欧洲综合997久久,| 男人和女人高潮做爰伦理| 午夜视频国产福利| 免费看a级黄色片| 国产一区二区激情短视频| 国产欧美日韩一区二区精品| 精品午夜福利视频在线观看一区| 国产激情偷乱视频一区二区| 国产精品国产高清国产av| ponron亚洲| 一级a爱片免费观看的视频| 最近视频中文字幕2019在线8| 老司机影院成人| 级片在线观看| 日本黄大片高清| 国产一区亚洲一区在线观看| 男女视频在线观看网站免费| 内地一区二区视频在线| 婷婷精品国产亚洲av| 五月伊人婷婷丁香| 亚洲性夜色夜夜综合| 99热6这里只有精品| 国产激情偷乱视频一区二区| 国产男靠女视频免费网站| 久久久久久大精品| 好男人在线观看高清免费视频| 亚洲国产精品国产精品| 熟女电影av网| 男女边吃奶边做爰视频| 亚洲四区av| 十八禁国产超污无遮挡网站| 欧美潮喷喷水| av福利片在线观看| 日韩三级伦理在线观看| 亚洲欧美日韩卡通动漫| 最近的中文字幕免费完整| 久久久久精品国产欧美久久久| 人妻制服诱惑在线中文字幕| 男人舔奶头视频| 欧美精品国产亚洲| 久久久国产成人免费| 国产精品电影一区二区三区| 久久精品人妻少妇| 淫秽高清视频在线观看| 能在线免费观看的黄片| 特大巨黑吊av在线直播| 波野结衣二区三区在线| 婷婷六月久久综合丁香| 日韩高清综合在线| 亚洲在线观看片| 不卡一级毛片| 久久亚洲精品不卡| 久久久久久久午夜电影| 午夜亚洲福利在线播放| 久久精品国产亚洲网站| 级片在线观看| 国产高清不卡午夜福利| 久久久精品欧美日韩精品| 久久6这里有精品| 久久草成人影院| 国产精品国产高清国产av| 自拍偷自拍亚洲精品老妇| 欧美激情在线99| 又黄又爽又刺激的免费视频.| 中文亚洲av片在线观看爽| 精品国产三级普通话版| 一区二区三区四区激情视频 | 1000部很黄的大片| 偷拍熟女少妇极品色| 成人亚洲精品av一区二区| 成人特级黄色片久久久久久久| 男女做爰动态图高潮gif福利片| 久久天躁狠狠躁夜夜2o2o| 天堂网av新在线| 99国产极品粉嫩在线观看| 一夜夜www| 国产精品永久免费网站| .国产精品久久| 国产精品一区二区三区四区久久| 亚洲va在线va天堂va国产| 一本久久中文字幕| 国产色爽女视频免费观看| 此物有八面人人有两片| 欧美在线一区亚洲| 国产一区二区亚洲精品在线观看| aaaaa片日本免费| 亚洲欧美精品自产自拍| 淫秽高清视频在线观看| 干丝袜人妻中文字幕| 又粗又爽又猛毛片免费看| 免费搜索国产男女视频| 老师上课跳d突然被开到最大视频| 国产亚洲精品久久久久久毛片| 欧美一区二区精品小视频在线| 国产精品,欧美在线| 国产精品久久久久久av不卡| 国内久久婷婷六月综合欲色啪| 亚洲精品一卡2卡三卡4卡5卡| www日本黄色视频网| 女人十人毛片免费观看3o分钟| 啦啦啦啦在线视频资源| 搞女人的毛片| 日本 av在线| 一级毛片我不卡| 亚洲av中文字字幕乱码综合| av在线观看视频网站免费| 日日干狠狠操夜夜爽| 91精品国产九色| 黄色日韩在线| 乱人视频在线观看| 欧美中文日本在线观看视频| 国产成人freesex在线 | 亚洲熟妇熟女久久| 欧美色视频一区免费| 97超碰精品成人国产| 成人特级黄色片久久久久久久| 香蕉av资源在线| 久久午夜亚洲精品久久| 成人精品一区二区免费| 精品国产三级普通话版| 欧美成人a在线观看| 精品99又大又爽又粗少妇毛片| 国产免费男女视频| 日韩中字成人| 狠狠狠狠99中文字幕| 午夜精品国产一区二区电影 | 免费不卡的大黄色大毛片视频在线观看 | 自拍偷自拍亚洲精品老妇| 91久久精品国产一区二区三区| 热99re8久久精品国产| 91在线观看av| 精品人妻视频免费看| 夜夜爽天天搞| 成人亚洲精品av一区二区| 男女啪啪激烈高潮av片| 最近中文字幕高清免费大全6| 亚洲欧美中文字幕日韩二区| av黄色大香蕉| 三级经典国产精品| 日韩制服骚丝袜av| 欧美成人一区二区免费高清观看| 午夜日韩欧美国产| 2021天堂中文幕一二区在线观| 男女之事视频高清在线观看| 欧美极品一区二区三区四区| 天天躁夜夜躁狠狠久久av| 色播亚洲综合网| 黄色一级大片看看| 露出奶头的视频| av在线老鸭窝| 少妇丰满av| 99热这里只有是精品在线观看| 国内精品美女久久久久久| 亚洲欧美日韩卡通动漫| 久久久久久久久中文| 天堂√8在线中文| 久久久a久久爽久久v久久| 午夜a级毛片| 久久久国产成人精品二区| 99视频精品全部免费 在线| 舔av片在线| 国产久久久一区二区三区| 人妻夜夜爽99麻豆av| 久久久精品欧美日韩精品| 精品国内亚洲2022精品成人| 日日啪夜夜撸| 97超视频在线观看视频| 夜夜夜夜夜久久久久| 国产精品无大码| 禁无遮挡网站| 欧美激情国产日韩精品一区| 九色成人免费人妻av| 欧美3d第一页| 亚洲成人久久性| 国产高清有码在线观看视频| 欧美成人精品欧美一级黄| 少妇人妻一区二区三区视频| 99久国产av精品国产电影| 国产毛片a区久久久久| 精品久久久久久久末码| 久久久久免费精品人妻一区二区| 一个人免费在线观看电影| 亚洲欧美日韩东京热| 波多野结衣高清作品| 国内精品久久久久精免费| 久久久久久国产a免费观看| 精品久久久噜噜| 男女边吃奶边做爰视频| 久久久a久久爽久久v久久| 亚洲熟妇中文字幕五十中出| 久久久久久久亚洲中文字幕| 国产黄a三级三级三级人| 天堂影院成人在线观看| 久久久久久大精品| 亚洲中文字幕日韩| 国模一区二区三区四区视频| 亚洲成人久久性| 久久久色成人| 精品国产三级普通话版| 十八禁国产超污无遮挡网站| 久久6这里有精品| 国产高清视频在线播放一区| 欧美丝袜亚洲另类| www日本黄色视频网| 日本三级黄在线观看| 成人特级av手机在线观看| 国产成人一区二区在线| 麻豆国产av国片精品| 大香蕉久久网| 高清毛片免费观看视频网站| 日韩av不卡免费在线播放| 亚洲国产精品成人久久小说 | 久久精品国产自在天天线| 久久精品国产鲁丝片午夜精品| 免费看美女性在线毛片视频| 天天一区二区日本电影三级| 在线免费十八禁| 亚洲真实伦在线观看| 人妻丰满熟妇av一区二区三区| 天天躁日日操中文字幕| 成年女人看的毛片在线观看| 亚洲人成网站在线播| 高清日韩中文字幕在线| 国产精品99久久久久久久久| 亚洲精品影视一区二区三区av| 亚洲乱码一区二区免费版| 别揉我奶头 嗯啊视频| 听说在线观看完整版免费高清| 黄色配什么色好看| 51国产日韩欧美| 国产三级在线视频| 可以在线观看毛片的网站|