• <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 ?

    Optimization of CNC Turning Machining Parameters Based on Bp-DWMOPSO Algorithm

    2023-12-12 15:49:40JiangLiJiutaoZhaoQinhuiLiuLaizhengZhuJinyiGuoandWeijiuZhang
    Computers Materials&Continua 2023年10期

    Jiang Li,Jiutao Zhao,Qinhui Liu,Laizheng Zhu,Jinyi Guo and Weijiu Zhang

    College of Mechanical and Electrical Engineering,Harbin Engineering University,Harbin,150001,China

    ABSTRACT Cutting parameters have a significant impact on the machining effect.In order to reduce the machining time and improve the machining quality,this paper proposes an optimization algorithm based on Bp neural network-Improved Multi-Objective Particle Swarm(Bp-DWMOPSO).Firstly,this paper analyzes the existing problems in the traditional multi-objective particle swarm algorithm.Secondly,the Bp neural network model and the dynamic weight multi-objective particle swarm algorithm model are established.Finally,the Bp-DWMOPSO algorithm is designed based on the established models.In order to verify the effectiveness of the algorithm,this paper obtains the required data through equal probability orthogonal experiments on a typical Computer Numerical Control(CNC)turning machining case and uses the Bp-DWMOPSO algorithm for optimization.The experimental results show that the Cutting speed is 69.4 mm/min,the Feed speed is 0.05 mm/r,and the Depth of cut is 0.5 mm.The results show that the Bp-DWMOPSO algorithm can find the cutting parameters with a higher material removal rate and lower spindle load while ensuring the machining quality.This method provides a new idea for the optimization of turning machining parameters.

    KEYWORDS Machining parameters;Bp neural network;Multiple Objective Particle Swarm Optimization;Bp-DWMOPSO algorithm

    1 Introduction

    With the development of computer technology,optimization algorithms are gradually applied in various fields.Frequently-used data modelling methods mainly include the Response Surface Methodology [1,2],Back Propagation Neural Network (BPNN) [3,4].Support Vector Regression[5,6] and Gradient Boosted Regression Tree [7].In order to better solve the optimization problem of CNC turning machining parameters,scholars at home and abroad have conducted a lot of research.Wang et al.proposed a multi-objective optimization method for CNC turning machining parameters based on the Response Surface Methodology and Artificial Bee Colony Algorithm,which has better distribution and convergence of optimization algorithms [8].Wang et al.established a mathematical model of machining cost and CNC cutting machining efficiency,which was solved by the Hybrid Multi-Objective Particle Swarm Optimization(HMOPSO).They used the Analytical Hierarchy Process for the final decision on the optimal combination of cutting parameters[9].Meanwhile,Deng et al.explored the applicability of deep reinforcement learning in machining parameter optimization problems and proposed a deep reinforcement learning-based optimization method for CNC Milling Process Parameters[10].For materials such as aluminium 6063,Osoriopinzon et al.[11]constructed optimization functions using the Response Surface Methodology and Artificial neural networks to solve multi-objective optimization problems such as cutting forces,microstructure refinement and material removal rate,and finally used particle swarm algorithms for optimization solutions.Van[12]used BPNN to construct an optimization model of machining parameters with cutting forces,vibration and energy consumption and realized the multi-objective optimization by multi-objective particle swarm algorithm,which provides an effective solution for the parameter optimization of highspeed milling.Li et al.[13]used the response surface methodology to model the relationship between cutting forces and machining parameters for CNC machining.Based on this,they constructed a multiobjective optimization model which considered cutting force,R-value and surface roughness,and used an improved teaching optimization algorithm to solve the model.He et al.[14]simultaneously considered cutting force,machining time and energy consumption in the carbon steel machining process,established a correlation relation model between each objective and machining parameters through theoretical analysis and empirical formulas.They obtained the Pareto front of the problem by using a decomposition-based multi-objective evolutionary algorithm to find the optimal solution.In order to make the Particle Swarm Optimization (PSO) in the late iteration still have a chance to jump out of the local optimal solution,Wang et al.[15]used methods that can deal with the stopping and receding state particles and random fluctuating inertia weight.Geng et al.[16] designed a PSO algorithm based on an orthogonal experiment mechanism to improve the convergence speed of the algorithm.Wang et al.[17] studied the problem of falling material or unsmooth deep drawing of needle tooth molds in the machining process.They constructed a Bp neural network by using the relationship between the cutting-edge parameters,shear strength and cut-off displacement,and used the experimental results to verify the correctness and reliability of the predicted optimal tooth mold cutting-edge parameters.For the problem of excessive local deformation of the parts in the machining process of the annular thin-walled part,Han et al.[18]proposed an optimization method for milling parameters of annular thin-walled parts with an improved particle swarm algorithm.

    In summary,the research direction of optimization of CNC turning machining parameters using optimization algorithms mainly focuses on the algorithm’s global search ability,stability and convergence speed,etc.In related research,there is a lack of research on the machining reliability of the optimized machining parameters,such as whether the optimized parameters can meet the machining accuracy and surface roughness(Ra).Therefore,this paper proposes a Bp Neural Network-Improved Multi-Objective Particle Swarm Algorithm.The difference between it and other algorithms is that the algorithm fully combines the advantages of the Bp neural network’s learning ability,strong generalization ability and particle swarm algorithm’s strong global search ability,which can improve the machining efficiency under the premise of ensuring the machining quality.Consequently,this algorithm is more suitable for the optimization of CNC turning machining parameters.

    The research objective of this paper is to establish the relationship model between machining accuracy,surface roughness and machining parameters during the CNC turning machining process by using the powerful approximation ability and learning ability of the Bp neural network.Then,the improved multi-objective particle swarm algorithm is used to optimize the multi-objective function in order to obtain a set of optimal machining parameter combinations.And the obtained parameters can not only meet the requirements of machining accuracy and surface roughness in the machining process but also consider the optimization of other machining performance indicators,thus improving productivity,reducing costs,and ensuring that the quality and performance of machined parts meet the requirements.

    The research in this paper has three main points: (1) this paper analyzes the problems of the traditional multi-objective particle swarm algorithm and introduces the Bp neural network algorithm and the Dynamic Weight Multi-Objective Particle Swarm Algorithm;(2)this paper proposes an optimization method of CNC turning machining parameters based on the improved particle algorithm;(3)a case study is used to verify the effectiveness of the method.

    2 Algorithm Analysis

    2.1 Selection of Optimization Algorithm

    Multi-Objective Particle Swarm Optimization(MOPSO)is a heuristic optimization algorithm for solving multi-objective optimization problems.The basic idea of the algorithm is to find the optimal solution set by maintaining a population of particles and iteratively updating them continuously.

    However,the algorithm cannot deal with classification and regression problems.When it is applied to solve the CNC turning parameter optimization problem,the resulting machining parameters cannot be directly applied in actual production machining,so it is necessary to judge whether the resulting machining accuracy and surface quality meet the requirements.The Bp neural network (algorithm)has strong adaptability and generalization ability,which is suitable for processing classification and regression problems.Therefore,this paper combines Bp neural network and MOPSO to solve the problem better and introduces the dynamic weighting strategy to improve the performance of the algorithm.

    2.2 Bp Neural Network

    Bp neural network,also known as Back Propagation Neural Network,is a frequently-used artificial neural network structure for solving classification and regression problems.The network consists of the input,the hidden and the output layers,where the hidden layer can be multiple layers[19].The structure of a Bp neural network is shown in Fig.1.

    Figure 1:Bp neural network structure chart

    The training process of the Bp neural network consists of two steps: forward propagation and backward propagation.In the structure shown in Fig.1,each layer contains multiple neurons(nodes),and each neuron has an activation function.When the model is working,these neurons are first transmitted forward in the order of the input,hidden and output layers(in the direction of the solid line).After that,the error is transmitted in the reverse direction from the output layer to the hidden layer within the model,that is,the transmission represented by the dashed line in Fig.1.The function of the transmission is to transmit the error between the actual output and the desired output and to continuously adjust the weights(w)and the bias(b)in the formula in order to achieve the reduction of the error.The specific formulas are as follows:

    where:yidenotes the output of thejthneuron;xidenotes theithinput feature;wjidenotes the connection weight between theithinput feature and thejth neuron;bjdenotes the bias of thejthneuron;?()denotes the activation function.

    where:δjdenotes the error of thejthneuron;netjdenotes the input of thejthneuron;?′()denotes the derivative of the activation function;wkjdenotes the connection weight between thejthneuron and thekth neuron;δkdenotes the error of thekth neuron.

    where:Δwjidenotes the update of the connection weight between theithinput feature and thejthneuron;θdenotes the learning rate;δjdenotes the error of thejthneuron;andxidenotes the value of theith input feature.

    2.3 Dynamic Weighted Multi-Objective Particle Swarm Algorithm

    2.3.1 Principle of Dynamic Weighted Multi-Objective Particle Swarm Algorithm

    As a swarm intelligence algorithm,the Dynamic Weighted Multi-Objective Particle Swarm Optimization(DWMOPSO)has the advantages of fast convergence and strong optimality search[20].It searches for the optimal solution based on the Pareto superiority relation.The particle updates itself by tracking two “extremes”: the first extremum,called the individual extremum point,is the best solution found by the particle itself,which is denoted by Pbest,and the other extremum,called the global optimal solution,is the current optimal solution found by the whole population,which is denoted by Gbest.In MOPSO,each particle corresponds to its own Gbest,while the single-objective particle swarm algorithm shares one global extreme point for the whole particle swarm.Meanwhile,DWMOPSO improves the algorithm’s performance by dynamically adjusting the inertia weights.The principle of DWMOPSO is as follows:

    1) Initialization: Initialize a particle swarm,including N particles.Each particle has a random position and velocity.The position and velocity is a vector,and the dimension of the vector is equal to the number of independent variables of the function.Each particle should record its current best position Pbest and best fitness pbest_val,as well as the current particle in the current particle in the Pareto optimal set.

    2)Update the position and velocity:Each particle updates its velocity and position,according to the following equations:

    where:wis the inertia weight;c1andc2is the acceleration constant;randis a random number in the range[0,1];Pbestiis the current best position of the particlei;Gbestiis the local optimal solution.

    3) Dynamic weighting strategy: The dynamic weight controls the speed of particle movement and the probability of jumping out of the local optimal solution.In DWMOPSO,the inertia weight decreases as the number of iterations increases.As shown in Eq.(6):

    where:weandwsare the upper and lower bounds of the dynamic weight;iteris the current number of iterations;niteris the maximum number of iterations.

    2.3.2 Steps of the Dynamic Weight Multi-Objective Particle Swarm Algorithm

    N randomly selected particles constitute a particle swarm,and each particle is a multidimensional vector.The flow chart of the DWMOPSO algorithm is shown in Fig.2.

    Step 1:Initialize the algorithm and set the initialization parameters such as population number,the maximum number of iterations,velocity and displacement of particles,etc.

    Step 2:Calculate the fitness value of each particle,and update individual optimal Pbest and global optimal Gbest according to the fitness of the particle.

    Step 3: Update the velocity and position of the particle according to the individual optimal solution,global optimal solution,velocity,position and inertia weight of the current particle at the current position.

    Step 4:Calculate the fitness value fitness of each particle and update the individual optimal Pbest and the global optimal Gbest.

    Step 5:Based on the current iteration number iter,the total iteration number niter,the initial value of inertia weightwsand the final value of inertia weightwe,calculate the current inertia weightw.

    Step 6:Update the rep set;determine whether the rep set overflows;get the current non-dominated solution.

    Step 7: If the number of iterations iter reaches the maximum number of iterations niter,the algorithm ends and outputs the final non-dominated solution set;otherwise,increase the number of iterations iter and return to Step 2.

    3 Bp-DWMOPSO Algorithm

    3.1 Introduction of Bp-DWMOPSO Algorithm

    The Improved Multi-Objective Particle Swarm Algorithm (Bp-DWMOPSO) proposed in this paper is an improved algorithm based on the Bp neural network model and the Dynamic Weight Multi-Objective Particle Swarm Algorithm Model.The algorithm has strong generalization ability and interpretability and can achieve good results when applied to the optimization of CNC turning machining parameters.

    Figure 2:Flow chart of the dynamic weighted multi-objective particle swarm algorithm

    3.2 Construction of Bp-DWMOPSO Algorithm

    The structure diagram of the Bp-DWMOPSO algorithm is shown in Fig.3.

    Figure 3:Structure chart of the Bp-DWMOPSO algorithm

    As seen in Fig.3,the Bp-DWMOPSO algorithm solves the problem in the following steps:

    Step 1: According to the objective function and constraint function,the dynamic weight multiobjective particle algorithm solves the problem and derives the Pareto solution set.

    Step 2:The solutions in the Pareto solution set are substituted into the Bp neural network model to derive the predicted values.

    Step 3: After substituting the predicted values into the judgment function for comparison,the solutions that meet the requirements are output;otherwise they are removed.

    The solution set screening formula is as follows:

    where:mpidenotes the solution in the Pareto solution set;yidenotes the output of thejth neuron;xidenotes theith input feature;wjidenotes the connection weight between theith input feature and thejth neuron;bjdenotes the bias of thejth neuron;?(x)denotes the activation function;f(x)denotes the judgment function,and whenf (x)=1,output the solution;otherwise remove the solution.

    3.2.1 Data Acquisition Methods

    The construction of the Bp-DWMOPSO algorithm requires the acquisition of training data.There are various methods of data acquisition,such as orthogonal experiments,single-factor experiments,and multi-factor experiments.According to the characteristics of CNC turning machining parameters,this paper uses equal probability orthogonal experiments to obtain the required data,and this experimental method is a multi-factor multi-level experimental design method.The advantages of this experimental method:(1)It saves experimental cost and time and significantly reduces the number of experiments;(2)It has a balance between the levels of factors,thus reducing the influence of random errors;(3) It can reveal the mutual influence between different factors and discover the main and secondary influencing factors.

    3.2.2 Data Preprocessing Methods

    Data preprocessing is one of the key steps in optimization model building to ensure that the input data is suitable for the training and learning process of the algorithm.In this paper,the following steps are used to preprocess the data,and the flow of data preprocessing is shown in Fig.4.

    Figure 4:Flow chart of data preprocessing

    Step 1:Data Cleaning:Before starting to build the optimization model,the raw data needs to be cleaned.This includes dealing with missing values,abnormal values and noisy data.This paper uses equal probability orthogonal experiments to collect data.This experimental method can effectively avoid the appearance of data such as missing values,abnormal values and noise data,so the collected data do not need to be cleaned and processed.

    Step 2:Feature Selection:Feature selection is to choose the most relevant and valuable features from the raw data in order to reduce the data dimension and avoid overfitting.This paper uses the Pearson Correlation Coefficient Method for feature selection.The Pearson Correlation Coefficient expression formula is as follows:

    Step 3: Data Normalization: In the process of building the optimization model,the range and distribution of the input data may have an impact on the training effect of the model.Therefore,the input data are generally normalized to have similar scales and distributions.This paper uses the maximum-minimum normalization(Min-Max Scaling)method,and the expression is as follows:

    where:Xscaleis the normalized data;Xis the sample data value;Xminis the sample data minimum;Xmaxis the sample data maximum.

    Step 4:Data Partition:Divide the data set into the training set,validation set and testing set.And the purpose of dividing the data set is to conduct model training,parameter tuning and performance evaluation.

    3.2.3 Characteristic Parameters of the Model

    The parameters are one of the core components of the model,and they are critical to the performance and effectiveness of the algorithm.Different parameter values can lead to different algorithm performance.Therefore,the correct selection and adjustment of parameters have an important impact on the performance and effect of the algorithm.For the Bp-DWMOPSO algorithm proposed in this paper,the parameters shown in Table 1 are set as the core parameters of the algorithm to improve the performance of the algorithm,such as accuracy,stability and generalization ability.

    Table 1:Algorithm core parameters table

    3.2.4 Selection of Activation Function and Optimization Algorithm

    In small sample prediction,overfitting is easy to occur due to the small number of training samples.Therefore,it is necessary to select an appropriate activation function to avoid overfitting and improve the model’s generalization ability.ReLU (Rectified Linear Unit) activation function is one of the most frequently-used activation functions in neural networks,which has the advantages of high computational efficiency,fast convergence and solving the gradient disappearance problem.This activation function is good at solving the problem of small sample prediction.Its function formula is as follows:

    The training process of the model is implemented in Python programming language,and the training algorithm is the Adaptive Moment Estimation Optimization Algorithm.It is a frequentlyused Stochastic Gradient Descent Optimization Algorithm that combines the advantages of the Momentum Method and RMSProp algorithm,with better adaptivity,stability and convergence speed.The updated formula of the adaptive moment estimation optimization algorithm is as follows:

    3.2.5 Model Evaluation Criteria

    In this paper,mean square error(MSE),mean absolute percentage error(MAPE)and coefficient of determination(R2)are used as the prediction accuracy evaluation indexes of the Bp neural network model.The smaller the value of MSE and MAPE,the closer the prediction value is to the true value,and the closer R2 is to 1.0,the stronger the prediction accuracy of the model.The expressions of the three indexes are

    where:p(i)andy(i)are respectively the predicted and the measured values of theith test sample;y is the average of the measured value;m is the number of test samples.

    3.2.6 Selection of the Number of Hidden Layers and the Number of Neurons

    The number of hidden layers has a significant impact on the performance of the model.Generally speaking,increasing the number of hidden layers can make the network have stronger nonlinear representation ability and fit complex data better,thus improving the accuracy and generalization ability of the model.However,increasing the number of hidden layers also increases the complexity of the network and may lead to overfitting problems.The number of hidden layer neurons in Bp neural networks is related to the number of input parameters.Based on experience,the number of hidden layer neurons is generally greater than or equal to twice the number of input parameters.Test the fitting performance of Bp neural networks under different hidden layer numbers and compare their military errors in the training and test sets.In the testing process,the mean square error of the test set is used as the first criterion,and the mean square error of the training set is used as the second criterion to determine the most suitable number of hidden layers.

    Similarly,on the premise of determining the number of hidden layers,the most suitable number of neurons is determined by testing the fitting effect of the Bp neural network with different numbers of neurons and comparing its mean square error in the training set and the test set,using the mean square error in the test set as the first criterion and the mean square error in the training set as the second criterion.

    3.2.7 Mathematical Description of the Multi-Objective Optimization Problem

    In general,the mathematical expression of the multi-objective optimization problem is

    where:f (xi)is the optimization objective;xiis the decision variable,xi∈Rn(Rnis the search space);g(xi)is the constraint.

    In multi-objective optimization,the optimization objectives are often not optimal simultaneously,because improving one optimization objective usually leads to a decrease in the values of the other objectives.A solution in the solution space is said to be a non-inferior solution if it is not dominated by other solutions [21].The ultimate goal of a multi-objective optimization problem is to find the set of Pareto solutions,that is,a set of complementarily dominated optimal solutions.For a multiobjective optimization problem withmobjective functions,for any two solutionsxiandxj,the solutionxidominates the solutionxjif the following two conditions hold.

    1)For allk∈[1,2,...,m],there isfk(xi)≤fk(xj);

    2)There exists at least onek∈[1,2,...,m],such thatfk(xi)

    3.2.8 Objective Function

    In CNC turning machining,spindle load (F) and material removal rate (Q) are two important indicators of machining quality and efficiency.Generally speaking,the higher the material removal rate,the higher the machining efficiency.However,a high material removal rate can also lead to a high tool spindle load,which will affect machining quality and machine life.Therefore,multiple indicators,such as spindle load and material removal rate,need to be integrated during the CNC turning process to develop a reasonable machining strategy to obtain ideal machining results and economic benefits.

    Spindle load (F) is the force on the spindle during machining,mainly consisting of the cutting force and the axial force.The change of spindle load(F)not only reflects the size of the cutting force but also reflects the tool wear,machine condition and other information during the machining process,which has an important reference value.Its empirical formula[22]is as follows:

    where:Fcis the cutting force;Kcis the cutting force coefficient;the material selected in this paper is 45 steel and check the table to get Kcis 400 N/(mm?kgf);f is the cutting width;vcis the cutting speed;vfis the feed per revolution.;apis the depth of cut.

    The material removal rate (Q) is the volume of material cut by the cutting edge per unit time.The material removal rate(Q)is an important indicator of machining efficiency and economy,and its empirical formula[22]is as follows:

    where:vcis the cutting speed;apis the depth of cut;vfis the feed per revolution.

    3.2.9 Multi-Objective Optimization Model

    In the actual turning process,in order to improve the machining efficiency of the machine tool,the larger the material removal rate(Q)is,the better,while taking into account the machine tool loss,tool wear and the stability of the machining system,the smaller the spindle load(F)is,the better.The established multi-objective optimization model is shown in Eqs.(21)–(24):

    where:(vcmin,vcmax),(vf min,vf max),(apmin,apmax)are respectively the bounded range intervals of cutting speed,feed per revolution and backlash.

    4 Case Study of Optimization of CNC Turning Parameters

    4.1 Experimental Data Collection

    This experiment selects CAK50135 machine tool as the processing equipment,45 steel as the processing material,stainless steel specialized triangular CNC cylindrical turning blade as the processing tool and a liquid concentration of about 10%of the water-soluble coolant for cooling.This experiment takes?28 mm×100 mm blank as the experimental material and aims to machine it as the workpiece axis shown in Fig.5,and the experimental equipment is shown in Fig.6.

    Figure 5:Dimensional chart of the machining workpiece shaft

    Figure 6:Experimental processing equipment

    This paper uses the digital micrometer,whose error accuracy is±0.002 mm and resolution is 0.001 mm,produced by SHRN to measure the machining accuracy of the workpiece axis.Meanwhile,the JD220 roughness measuring instrument,produced by Beijing Jitai Keji Equipment Ltd.China,is used to measure the surface quality of the workpiece.This instrument has high precision and accuracy,and the error of its indicated value does not exceed 10%,and the indicated value accuracy is 0.01 um.The measurement equipment is shown in Fig.7.

    Figure 7:Experimental measurement equipment

    In actual production,the craftsman or operator usually selects the turning machining parameters based on the technical parameters of the machine tool and the range of machining parameters recommended by the turning tool manufacturer.In this paper,the above method will be used to select the turning machining parameters for the recommended machining parameters,and the recommended machining parameters are shown in Table 2.

    Table 2:Recommended process parameters

    The values of the process parameters corresponding to each level in the equal probability orthogonal experimental design used in this paper are shown in Table 3.

    Table 3:Process parameter levels

    In this paper,a 3-factor 8-level equal-probability orthogonal experimental design with 32 experiments was used,that is,an L32(8∧3)orthogonal table was used.The process parameters,dimensional accuracy and surface roughness Ra for each group of experiments are shown in Table 4.

    Table 4:Orthogonal experimental results

    4.2 Determination of Model Structure and Parameters

    4.2.1 Design of Model Structure

    This paper establishes a neural network model NN1with workpiece machining accuracy as the output and a neural network model NN2with surface roughness Ra as the output.The input variables of both models are three key process parameters of turning machining: cutting speed (vc),feed per revolution (f)and depth of cut(ap).The model structure diagram is shown in Fig.8.The 32 sets of data are randomly divided into two parts according to the ratio of 3:1,of which 22 sets of data are used as the training set data of the model and ten sets of data are used as the test set data of the model.

    Figure 8:Two single-output neural network models,NN1 and NN2

    4.2.2 Determination of the Number of the Hidden Layers

    In order to determine the most suitable number of hidden layers,the Bp neural network models with different numbers of hidden layers are solved in this paper,and the results are shown in Figs.9 and 10.From Figs.9 and 10,it can be seen that the MSE of the test set is the lowest when the number of hidden layers is respectively 4 and 1 in the NN1and NN2network models.Although the MSE on the training set is lower when the number of hidden layers is 5 and 6,the MSE on the test set is higher,which indicates that too many hidden layers will make the model solution more complicated and thus,the phenomenon of overfitting will occur,leading to the weak generalization ability of the model.Therefore,in this paper,the number of hidden layers in the NN1and NN2network models is set to 4 and 1.

    Figure 9:The influence of different hidden layers on MSE in NN1 network model

    Figure 10:The influence of different hidden layers on MSE in NN2 network model

    4.2.3 Number of the Hidden Layer Neurons

    In Section 4.2.2 of this paper,the number of hidden layer layers in the NN1and NN2network models has been determined.However,in order to find a more suitable number of neurons for the hidden layer network,this paper has tested the effect of different numbers of neurons on the performance of the two network models.As can be seen from Figs.11 and 12,in the NN1model,when the number of neurons is 7,the MSE of the training and validation sets is the lowest.In the NN2model,the MSE of both the training and validation sets is lowest when the number of neurons is 6.Therefore,the number of neurons in the hidden layer of the NN1model is taken as 7,and the number of neurons in the hidden layer of the NN2model is taken as 6.

    Figure 11:The effect of different number of neurons on MSE in NN1 network model

    Figure 12:The effect of different number of neurons on MSE in NN2 network model

    4.2.4 Model Prediction Results and Analysis

    The regression performance of the models determines the accuracy of the prediction results.From the fitted curves Figs.13 and 14 as well as Table 5,it can be seen that although there are some errors in the prediction results of the two models,the MSE and MAPE are low,while the R2 is high,which indicates that both models show good regression performance with high confidence.

    Table 5:Model evaluation results

    4.2.5 Model Prediction Results and Analysis

    The forecasting problem studied in this paper is a regression problem.In order to evaluate the accuracy of the prediction models,three model performance indicators—Mean Square Error(MSE),Mean Absolute Percentage Error(MAPE),and Determination Coefficient(R2),are used in this paper.The results of the model evaluation are shown in Table 5.From Table 5,it can be observed that although there are some errors in the prediction results of the two models,the Mean Square Error(MSE),Mean Absolute Percentage Error (MAPE) are both relatively small,and the determination coefficient (R2) is relatively high,which indicates that the two models have good performance in regression performance and have high confidence level.

    Figure 13:NN1 network model training set and test set fitting results

    Figure 14:NN2 network model training set and test set fitting results

    4.2.6 Multi-Objective Function

    In this case,the machining parameters in Table 1 are used as constraints.Meanwhile,to facilitate the observation of the graph,the optimization objective is transformed from maximizing the material removal rate to minimizing the reciprocal of the material removal rate(1/Q),and the multi-objective optimization model for this case is obtained after substitution into Eqs.(21)–(24) as shown in Eqs.(25)–(28).

    4.3 Optimization Results of the Bp-DWMOPSO Algorithm

    The population initialization is set to 100;the maximum number of iterations is 50[9];the initial number of iterations is 0;the initial inertia weight is 0.4;and the final inertia weight is 0.9[23].The CNC turning machining parameter optimization problem is solved using the Bp-DWMOPSO algorithm,and the solved Pareto solution is shown in Fig.14.

    The traditional multi-objective particle swarm algorithm only solves the Pareto solution set according to the objective function requirements,as shown in Fig.15(a).However,the Bp-DWMOPSO algorithm can satisfy both the objective function requirements and other performance indexes to solve the Pareto solution set,as shown in Fig.15(b).

    Figure 15:Comparison of Pareto solution sets for different algorithms

    Although the number of solutions of the traditional multi-objective particle swarm algorithm is very large,many of the solutions do not necessarily satisfy the machining accuracy and surface roughness (Ra) requirements.The Pareto solution set obtained by the Bp-DWMOPSO algorithm not only meets the requirements of the objective function but also can meet the requirements of machining accuracy and surface roughness (Ra).And the results obtained are more in line with the actual machining and production requirements.

    4.4 Decision Analysis by Analytical Hierarchy Process

    This paper uses hierarchical analysis to select the optimal combination of cutting parameters among the six sets of machining parameter solutions obtained by the Bp-DWMOPSO method to obtain the optimal solution among the conflicting objectives of cutting maximum productivity and minimum production cost.The hierarchical analysis method uses level-by-level refinement and hierarchical comparison to determine the weights and finally synthesizes them according to the hierarchical structure to form the weights of each factor for the total objective[24].

    In this paper,the six sets of solutions in Fig.14 are used as the solution layer;the values obtained from the two objective functions are used as the criterion layer;the results of the identified optimal parameters are used as the objective layer.The pairwise comparison matrix of the solution layer to the criterion layer is:

    The maximum eigenvalues are divided into 6.37 and 6.44.Take the weight vector W2=[0.6,0.4]T from the criterion layer to the target layer,whose consistency index CI is 0,respectively,and pass the consistency test.The eigenvectors corresponding to the maximum eigenvalues of F and 1/Q are found and normalized to obtain the weight vector W1from the scheme layer to the criterion layer as:

    Finally,according to the total ranking w of the hierarchy,the 6th group has the largest weight,and the obtained optimal parameters are shown in Table 6.

    Table 6:Decision results

    It can be seen that the selection of machining parameters has an important influence on the maximum production efficiency and the cost of CNC machining,and reasonable parameter selection has an important guiding significance for enterprise production.

    5 Conclusion

    As can be seen from the results of the turning machining case in Section 4,the Bp-DWMOPSO algorithm proposed in this paper fully combines the advantages of the Bp neural network’s learning ability,strong generalization ability and the particle swarm algorithm’s strong global search ability,which has achieved encouraging results.In this study,a reliable optimization model is established by collecting the data through equal probability orthogonal experiments and processing the data through the Bp-DWMOPSO algorithm.Eventually,the machining parameters are successfully optimized so that the requirements of machining accuracy and surface roughness can be met during the CNC turning machining process while taking into account the optimization of other key machining performance indexes.This not only significantly improves productivity and reduces cost but also ensures the quality and performance of the machined parts.

    The results of this research show that with the help of the Bp neural network and the improved multi-objective particle swarm algorithm,more excellent results can be achieved in the field of CNC turning machining.This method can not only be widely used in the existing machining process but also provides a useful reference for the research of other similar multi-objective optimization problems.

    Acknowledgement:Thank you to Shaofei Li and Yang Wang for their technical and equipment support.

    Funding Statement:The authors received no specific funding for this study.

    Author Contributions:Study conception and design: Jiang Li,Jiutao Zhao;data collection: Jiutao Zhao,Laizheng Zhu;analysis and interpretation of results:Qinhui Liu,Jiutao Zhao;draft manuscript preparation:Jinyi Guo,Weijiu Zhang.All authors reviewed the results and approved the final version of the manuscript.

    Availability of Data and Materials:The data in this article are all from on-site actual processing,and the data results refer to Tables 3 and 4 in the article.If you have any other questions,you can send an email to 1245885036@qq.com Email for communication and discussion.

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

    高清av免费在线| 特级一级黄色大片| 日韩强制内射视频| 久久久欧美国产精品| 欧美国产精品一级二级三级 | 久久99热这里只有精品18| 女人久久www免费人成看片| 又黄又爽又刺激的免费视频.| 在线免费观看不下载黄p国产| 亚洲精品视频女| 日韩一区二区三区影片| 亚洲成人精品中文字幕电影| 亚洲精品中文字幕在线视频 | 亚洲熟女精品中文字幕| 欧美 日韩 精品 国产| 又大又黄又爽视频免费| 精品人妻一区二区三区麻豆| 亚洲一级一片aⅴ在线观看| 视频中文字幕在线观看| 久久这里有精品视频免费| 亚洲精品国产成人久久av| 中国美白少妇内射xxxbb| 亚洲av一区综合| 精品酒店卫生间| 一级毛片我不卡| 熟女电影av网| 中文乱码字字幕精品一区二区三区| 永久免费av网站大全| 亚洲精品第二区| 街头女战士在线观看网站| 国产精品久久久久久久久免| 最新中文字幕久久久久| 新久久久久国产一级毛片| 大码成人一级视频| 国产色婷婷99| 国产精品爽爽va在线观看网站| 一个人看视频在线观看www免费| 国产高清有码在线观看视频| 成人亚洲精品一区在线观看 | 交换朋友夫妻互换小说| 免费黄色在线免费观看| 中文欧美无线码| 欧美成人精品欧美一级黄| 久久久成人免费电影| 午夜精品一区二区三区免费看| 丰满乱子伦码专区| 日本熟妇午夜| 一区二区三区乱码不卡18| 寂寞人妻少妇视频99o| 少妇人妻精品综合一区二区| 亚洲天堂av无毛| 国产亚洲91精品色在线| 天堂网av新在线| 噜噜噜噜噜久久久久久91| 伊人久久精品亚洲午夜| 最后的刺客免费高清国语| 中文天堂在线官网| 女人被狂操c到高潮| 国产精品秋霞免费鲁丝片| 成人综合一区亚洲| 在线观看国产h片| 网址你懂的国产日韩在线| 国产片特级美女逼逼视频| 欧美亚洲 丝袜 人妻 在线| 高清毛片免费看| 国产男女内射视频| 国产成人freesex在线| 亚洲精品国产av成人精品| 欧美亚洲 丝袜 人妻 在线| 亚洲不卡免费看| 夫妻午夜视频| 国产毛片a区久久久久| 国产亚洲av嫩草精品影院| 婷婷色麻豆天堂久久| 九九爱精品视频在线观看| 自拍欧美九色日韩亚洲蝌蚪91 | 最近的中文字幕免费完整| 久久韩国三级中文字幕| a级毛片免费高清观看在线播放| 少妇的逼好多水| 黄色日韩在线| 亚洲电影在线观看av| 国产中年淑女户外野战色| 国产精品久久久久久精品电影| 看黄色毛片网站| 欧美日韩国产mv在线观看视频 | 成人亚洲精品一区在线观看 | 免费播放大片免费观看视频在线观看| 97人妻精品一区二区三区麻豆| 永久网站在线| 免费大片黄手机在线观看| 欧美潮喷喷水| 老师上课跳d突然被开到最大视频| 老师上课跳d突然被开到最大视频| 久久99热这里只频精品6学生| 中文字幕av成人在线电影| av黄色大香蕉| 久久97久久精品| 神马国产精品三级电影在线观看| 涩涩av久久男人的天堂| 久久人人爽av亚洲精品天堂 | 亚洲人成网站在线观看播放| 能在线免费看毛片的网站| av播播在线观看一区| 欧美97在线视频| 美女内射精品一级片tv| 汤姆久久久久久久影院中文字幕| 国产大屁股一区二区在线视频| 天天一区二区日本电影三级| 成人毛片60女人毛片免费| 国产成人免费观看mmmm| 欧美一区二区亚洲| 成人毛片a级毛片在线播放| 久久精品国产亚洲av天美| 一级毛片久久久久久久久女| 1000部很黄的大片| 免费黄网站久久成人精品| 搡老乐熟女国产| 国产免费一级a男人的天堂| 亚洲精品一区蜜桃| 国产黄片视频在线免费观看| 亚洲激情五月婷婷啪啪| 国产免费又黄又爽又色| 欧美xxxx黑人xx丫x性爽| 97人妻精品一区二区三区麻豆| 波多野结衣巨乳人妻| 热99国产精品久久久久久7| www.色视频.com| 免费大片黄手机在线观看| 国产精品一二三区在线看| 国产国拍精品亚洲av在线观看| av天堂中文字幕网| 欧美日韩亚洲高清精品| 青青草视频在线视频观看| 亚洲精品亚洲一区二区| 国产探花在线观看一区二区| 国产免费一区二区三区四区乱码| 久久国产乱子免费精品| 国产精品国产av在线观看| 国产精品久久久久久精品电影小说 | 日本欧美国产在线视频| 久久久久久伊人网av| 99久久九九国产精品国产免费| 超碰97精品在线观看| 亚洲最大成人av| 日韩av不卡免费在线播放| 欧美一级a爱片免费观看看| 精品久久久精品久久久| 国产伦精品一区二区三区视频9| 久久久久国产网址| 欧美 日韩 精品 国产| 少妇裸体淫交视频免费看高清| freevideosex欧美| 日韩欧美一区视频在线观看 | 国产精品麻豆人妻色哟哟久久| 美女主播在线视频| 一级a做视频免费观看| 舔av片在线| freevideosex欧美| 97超碰精品成人国产| 国产中年淑女户外野战色| 夫妻午夜视频| 亚洲精品视频女| 91精品一卡2卡3卡4卡| 亚洲av成人精品一区久久| 免费观看性生交大片5| 亚洲欧美日韩无卡精品| 国产成人freesex在线| 午夜精品国产一区二区电影 | 最后的刺客免费高清国语| 免费电影在线观看免费观看| 日本免费在线观看一区| 免费黄频网站在线观看国产| 日韩免费高清中文字幕av| 亚洲av免费在线观看| 肉色欧美久久久久久久蜜桃 | 汤姆久久久久久久影院中文字幕| 国产精品不卡视频一区二区| 色5月婷婷丁香| 搞女人的毛片| 岛国毛片在线播放| 夫妻性生交免费视频一级片| 日本免费在线观看一区| 身体一侧抽搐| 国产成人一区二区在线| 别揉我奶头 嗯啊视频| 亚洲av免费高清在线观看| 男女国产视频网站| 日韩欧美精品v在线| 亚洲av二区三区四区| 国产精品麻豆人妻色哟哟久久| 国产精品99久久99久久久不卡 | 又黄又爽又刺激的免费视频.| 日韩欧美一区视频在线观看 | 男女国产视频网站| 观看美女的网站| 亚洲精品影视一区二区三区av| 午夜老司机福利剧场| 高清av免费在线| 街头女战士在线观看网站| 在线a可以看的网站| 精品一区二区三卡| 国产69精品久久久久777片| av卡一久久| 国产av不卡久久| 自拍偷自拍亚洲精品老妇| 白带黄色成豆腐渣| 中文字幕人妻熟人妻熟丝袜美| 简卡轻食公司| 嫩草影院新地址| 日韩大片免费观看网站| 亚洲精品乱码久久久久久按摩| 特级一级黄色大片| videos熟女内射| 51国产日韩欧美| 国产成人a∨麻豆精品| 久久人人爽人人爽人人片va| 欧美日韩在线观看h| 精品国产露脸久久av麻豆| 99久久精品一区二区三区| 国产一级毛片在线| 男人和女人高潮做爰伦理| 综合色av麻豆| 九九久久精品国产亚洲av麻豆| 国产精品一区二区性色av| 欧美+日韩+精品| 又爽又黄a免费视频| 久久人人爽人人片av| 午夜福利视频精品| 国产一区二区亚洲精品在线观看| 97人妻精品一区二区三区麻豆| 亚洲国产高清在线一区二区三| 18+在线观看网站| 直男gayav资源| 99热6这里只有精品| 人妻 亚洲 视频| 欧美极品一区二区三区四区| 亚洲欧美中文字幕日韩二区| 天美传媒精品一区二区| 亚洲国产欧美在线一区| 深夜a级毛片| 欧美xxⅹ黑人| 九草在线视频观看| 久久这里有精品视频免费| 欧美高清性xxxxhd video| 日本一二三区视频观看| 少妇裸体淫交视频免费看高清| 各种免费的搞黄视频| 少妇人妻一区二区三区视频| 国产精品一区二区性色av| 黄色一级大片看看| 小蜜桃在线观看免费完整版高清| 99久久人妻综合| a级毛色黄片| 国产亚洲精品久久久com| 青春草亚洲视频在线观看| av免费在线看不卡| 中文欧美无线码| 男的添女的下面高潮视频| 免费av不卡在线播放| 麻豆成人av视频| 国产免费又黄又爽又色| 欧美高清性xxxxhd video| 亚洲精品成人av观看孕妇| 亚洲va在线va天堂va国产| 日本av手机在线免费观看| 久久久国产一区二区| 少妇人妻 视频| 免费看a级黄色片| 国产精品一区二区性色av| 老司机影院毛片| 久久精品国产亚洲av涩爱| av专区在线播放| 久久久a久久爽久久v久久| 18禁动态无遮挡网站| 久久精品国产亚洲网站| 老司机影院成人| 美女内射精品一级片tv| 久久影院123| 国产一区二区亚洲精品在线观看| 中国三级夫妇交换| 国产精品av视频在线免费观看| 91aial.com中文字幕在线观看| 精品人妻偷拍中文字幕| 成年女人看的毛片在线观看| 看十八女毛片水多多多| 欧美人与善性xxx| 国产精品人妻久久久久久| 精品人妻视频免费看| 精华霜和精华液先用哪个| 国产极品天堂在线| 亚洲精品aⅴ在线观看| 日本免费在线观看一区| 黄色日韩在线| 成人免费观看视频高清| 麻豆成人av视频| 日韩国内少妇激情av| av在线观看视频网站免费| 爱豆传媒免费全集在线观看| 91在线精品国自产拍蜜月| 亚洲成人久久爱视频| 中文字幕人妻熟人妻熟丝袜美| 高清视频免费观看一区二区| 在线精品无人区一区二区三 | 欧美高清性xxxxhd video| 蜜桃久久精品国产亚洲av| 国产91av在线免费观看| 欧美日韩亚洲高清精品| 18禁动态无遮挡网站| 18禁动态无遮挡网站| 久久久久久久国产电影| 久久久久久久大尺度免费视频| 国产精品伦人一区二区| 美女视频免费永久观看网站| 成人二区视频| kizo精华| 日本wwww免费看| 精品人妻熟女av久视频| 国产一区二区三区综合在线观看 | 日韩av在线免费看完整版不卡| 综合色丁香网| 一级毛片aaaaaa免费看小| 亚洲美女视频黄频| 精品一区二区三区视频在线| 久久鲁丝午夜福利片| 色哟哟·www| av线在线观看网站| 日韩av在线免费看完整版不卡| 亚洲aⅴ乱码一区二区在线播放| 免费黄频网站在线观看国产| 免费看a级黄色片| 18禁裸乳无遮挡免费网站照片| 国产又色又爽无遮挡免| 亚洲欧美日韩无卡精品| 国产亚洲精品久久久com| 少妇高潮的动态图| 亚洲最大成人av| 狠狠精品人妻久久久久久综合| 肉色欧美久久久久久久蜜桃 | 日韩一区二区视频免费看| av播播在线观看一区| 国产精品久久久久久av不卡| 成年女人在线观看亚洲视频 | 蜜桃亚洲精品一区二区三区| 寂寞人妻少妇视频99o| 校园人妻丝袜中文字幕| 欧美一级a爱片免费观看看| 久久精品久久久久久噜噜老黄| 下体分泌物呈黄色| 亚洲经典国产精华液单| 亚洲最大成人av| 久久精品熟女亚洲av麻豆精品| 搡女人真爽免费视频火全软件| 免费在线观看成人毛片| 欧美xxⅹ黑人| 69av精品久久久久久| a级一级毛片免费在线观看| 国内揄拍国产精品人妻在线| 三级国产精品片| 亚洲av日韩在线播放| 久久97久久精品| 久久久欧美国产精品| av免费观看日本| 青春草国产在线视频| 看十八女毛片水多多多| 久久人人爽av亚洲精品天堂 | 国产伦在线观看视频一区| 国产黄a三级三级三级人| 熟妇人妻不卡中文字幕| freevideosex欧美| 久久久色成人| 亚洲性久久影院| xxx大片免费视频| 亚洲精品456在线播放app| 久久热精品热| 国产亚洲精品久久久com| 成人高潮视频无遮挡免费网站| 亚洲伊人久久精品综合| 国产成人aa在线观看| 成人免费观看视频高清| 2018国产大陆天天弄谢| 中文字幕亚洲精品专区| 1000部很黄的大片| 欧美激情在线99| 精品视频人人做人人爽| 又大又黄又爽视频免费| 久久国内精品自在自线图片| 国产精品人妻久久久久久| 我的女老师完整版在线观看| 国产91av在线免费观看| 精品人妻视频免费看| 国产亚洲精品久久久com| 人妻制服诱惑在线中文字幕| 亚洲国产欧美人成| 亚洲人成网站高清观看| 搡女人真爽免费视频火全软件| 啦啦啦啦在线视频资源| 国产女主播在线喷水免费视频网站| 女人十人毛片免费观看3o分钟| 可以在线观看毛片的网站| 国产成人91sexporn| 久热久热在线精品观看| 夜夜看夜夜爽夜夜摸| 国产精品三级大全| 伊人久久国产一区二区| 交换朋友夫妻互换小说| 性插视频无遮挡在线免费观看| 国产精品人妻久久久久久| 一本色道久久久久久精品综合| 91精品国产九色| 三级国产精品欧美在线观看| 日韩欧美精品免费久久| 波多野结衣巨乳人妻| 精品国产乱码久久久久久小说| 水蜜桃什么品种好| 三级经典国产精品| 国产精品三级大全| 亚洲久久久久久中文字幕| 麻豆久久精品国产亚洲av| 女人十人毛片免费观看3o分钟| 夜夜爽夜夜爽视频| 日韩中字成人| 最近的中文字幕免费完整| 国产探花极品一区二区| 国产一区二区亚洲精品在线观看| 国产欧美日韩一区二区三区在线 | 七月丁香在线播放| 午夜激情福利司机影院| 国产熟女欧美一区二区| 最近中文字幕高清免费大全6| kizo精华| 亚洲精品久久午夜乱码| 国产老妇女一区| 日本欧美国产在线视频| 汤姆久久久久久久影院中文字幕| 中文字幕制服av| 国产成人免费观看mmmm| 日韩一区二区视频免费看| 亚洲av电影在线观看一区二区三区 | 久久99热这里只有精品18| av在线老鸭窝| 国产亚洲最大av| 人人妻人人澡人人爽人人夜夜| 亚洲婷婷狠狠爱综合网| 免费观看av网站的网址| 国产 精品1| 国产精品偷伦视频观看了| 中文字幕亚洲精品专区| 爱豆传媒免费全集在线观看| 国产成人午夜福利电影在线观看| 国精品久久久久久国模美| 国产高清不卡午夜福利| 精品人妻熟女av久视频| 日本午夜av视频| 国产在线一区二区三区精| 亚州av有码| 国产又色又爽无遮挡免| 少妇猛男粗大的猛烈进出视频 | 免费在线观看成人毛片| 久久久久精品久久久久真实原创| 国产日韩欧美在线精品| 噜噜噜噜噜久久久久久91| 人妻制服诱惑在线中文字幕| 禁无遮挡网站| 一区二区三区精品91| 亚洲久久久久久中文字幕| 黑人高潮一二区| 免费av观看视频| 一二三四中文在线观看免费高清| 成年女人看的毛片在线观看| a级毛色黄片| 日韩精品有码人妻一区| 日韩电影二区| 日本av手机在线免费观看| 人体艺术视频欧美日本| 欧美xxxx性猛交bbbb| 欧美日韩精品成人综合77777| 老女人水多毛片| 成人特级av手机在线观看| 人妻制服诱惑在线中文字幕| 热99国产精品久久久久久7| 国产精品国产三级专区第一集| 精品99又大又爽又粗少妇毛片| 一本—道久久a久久精品蜜桃钙片 精品乱码久久久久久99久播 | 国产精品av视频在线免费观看| 黄色欧美视频在线观看| 国产综合精华液| 人人妻人人澡人人爽人人夜夜| 成年人午夜在线观看视频| av在线观看视频网站免费| 亚洲图色成人| 久久久久久久午夜电影| 成年版毛片免费区| 黄色视频在线播放观看不卡| 26uuu在线亚洲综合色| 亚洲人成网站高清观看| 国产综合精华液| 欧美区成人在线视频| 97精品久久久久久久久久精品| 日韩中字成人| 联通29元200g的流量卡| 最近中文字幕高清免费大全6| 亚洲电影在线观看av| 少妇熟女欧美另类| 久久久久久伊人网av| 国产男女内射视频| 99re6热这里在线精品视频| 一本久久精品| 国产黄频视频在线观看| 最近最新中文字幕大全电影3| 激情 狠狠 欧美| 老司机影院毛片| 久久精品夜色国产| 久久99热这里只有精品18| 80岁老熟妇乱子伦牲交| 在线观看免费高清a一片| 人妻夜夜爽99麻豆av| 又爽又黄无遮挡网站| 一级毛片 在线播放| kizo精华| 18禁裸乳无遮挡免费网站照片| 日日啪夜夜撸| 中文在线观看免费www的网站| 国产高清国产精品国产三级 | 我的女老师完整版在线观看| 日韩成人伦理影院| 美女cb高潮喷水在线观看| 欧美亚洲 丝袜 人妻 在线| 尤物成人国产欧美一区二区三区| 搡老乐熟女国产| 三级男女做爰猛烈吃奶摸视频| 日韩 亚洲 欧美在线| 亚洲欧洲国产日韩| 赤兔流量卡办理| 18禁在线无遮挡免费观看视频| 亚洲精品亚洲一区二区| 97在线人人人人妻| 深爱激情五月婷婷| 欧美日韩一区二区视频在线观看视频在线 | 久久精品熟女亚洲av麻豆精品| 亚洲,一卡二卡三卡| 免费av不卡在线播放| 美女cb高潮喷水在线观看| av网站免费在线观看视频| 成年av动漫网址| 五月玫瑰六月丁香| 亚洲人成网站在线观看播放| 在线观看美女被高潮喷水网站| 亚洲成色77777| 一级黄片播放器| 高清日韩中文字幕在线| 亚洲,欧美,日韩| 日本欧美国产在线视频| 少妇被粗大猛烈的视频| 一级毛片aaaaaa免费看小| 欧美日韩视频高清一区二区三区二| 国产av码专区亚洲av| 干丝袜人妻中文字幕| 国产亚洲91精品色在线| 伦理电影大哥的女人| 国产黄色视频一区二区在线观看| 少妇人妻久久综合中文| 亚洲欧美成人精品一区二区| 高清欧美精品videossex| 亚洲熟女精品中文字幕| 成年女人在线观看亚洲视频 | 国产爱豆传媒在线观看| 亚洲精品一区蜜桃| 久久人人爽av亚洲精品天堂 | 男女边吃奶边做爰视频| 老司机影院毛片| 美女高潮的动态| 国产成人精品婷婷| 成人亚洲精品一区在线观看 | 国产乱人偷精品视频| 成人漫画全彩无遮挡| 国内精品美女久久久久久| 国产欧美日韩一区二区三区在线 | 看十八女毛片水多多多| 韩国av在线不卡| 爱豆传媒免费全集在线观看| 免费大片黄手机在线观看| 久久久久国产精品人妻一区二区| 成人国产av品久久久| 激情五月婷婷亚洲| 三级男女做爰猛烈吃奶摸视频| 色视频www国产| 国产国拍精品亚洲av在线观看| 亚洲经典国产精华液单| 建设人人有责人人尽责人人享有的 | av国产免费在线观看| 99久久精品一区二区三区| 一级毛片我不卡| 亚洲成人中文字幕在线播放| 午夜视频国产福利| 一本久久精品| 精品国产乱码久久久久久小说| 国产又色又爽无遮挡免| 99久国产av精品国产电影| 在线观看一区二区三区| 久久亚洲国产成人精品v| 在线 av 中文字幕| 精品久久久久久电影网| 伦精品一区二区三区| 下体分泌物呈黄色| 一级毛片 在线播放| 黄色配什么色好看| 少妇人妻精品综合一区二区| 国产成年人精品一区二区| 国产白丝娇喘喷水9色精品| a级毛色黄片| 国产黄a三级三级三级人| 99视频精品全部免费 在线| 欧美 日韩 精品 国产| 亚洲最大成人av|