• <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 ?

    A Novel MILP Model Based on the Topology of a Network Graph for Process Planning in an Intelligent Manufacturing System

    2021-11-26 03:45:40QihaoLiuXinyuLiLiangGao
    Engineering 2021年6期

    Qihao Liu, Xinyu Li*, Liang Gao

    State Key Laboratory of Digital Manufacturing Equipment and Technology, School of Mechanical Science and Engineering, Huazhong University of Science and Technology,Wuhan 430074, China

    Keywords:

    ABSTRACT Intelligent process planning(PP)is one of the most important components in an intelligent manufacturing system and acts as a bridge between product designing and practical manufacturing. PP is a nondeterministic polynomial-time(NP)-hard problem and,as existing mathematical models are not formulated in linear forms, they cannot be solved well to achieve exact solutions for PP problems. This paper proposes a novel mixed-integer linear programming(MILP)mathematical model by considering the network topology structure and the OR nodes that represent a type of OR logic inside the network. Precedence relationships between operations are discussed by raising three types of precedence relationship matrices. Furthermore, the proposed model can be programmed in commonly-used mathematical programming solvers, such as CPLEX, Gurobi, and so forth, to search for optimal solutions for most open problems. To verify the effectiveness and generality of the proposed model, five groups of numerical experiments are conducted on well-known benchmarks. The results show that the proposed model can solve PP problems effectively and can obtain better solutions than those obtained by the state-ofthe-art algorithms.

    1. Introduction

    Intelligent manufacturing involves intelligent manufacturing technology and intelligent manufacturing systems [1–3]. Intelligent process planning(PP)is one of the most important intelligent manufacturing technologies and plays a crucial role in intelligent manufacturing systems[4,5].PP can effectively shorten production cycle and improve product quality, and further, reduce the consumption for resources and energy[6].Hence,PP has gained a considerable amount of attention in industrial applications.Nevertheless, due to the extensive internal flexibility within PP,it is very difficult to solve for an optimal process plan in practical production scenarios [7].

    Solving a traditional PP problem usually involves three steps:process selection, resource allocation, and operation sequencing[5]. The first step is to fix the process selection, afterwards, the length of a process route can be determined because the number of operations varies from different process approaches. Next, the allocation of resources, including machines and tools, should be configured for all the operations. Sequencing the operations is the last and most critical step for obtaining a feasible process route that obeys the precedence constraints[8].For example,milling and grinding must be arranged before fine milling and fine grinding operations,and tapping for a hole thread must be carried out after the operation of drilling a hole [9].

    PP problems have been proved to be nondeterministic polynomial-time (NP)-hard [6,10] meaning that it is difficult to solve by only relying on the traditional gradient descent methods,graph theory methods, or simulation-based methods. Therefore,most researchers attempt to introduce meta-heuristics [11] to study PP problems.The main research approaches for PP problems include genetic algorithm (GA) [7,12], tabu search (TS) [13], particle swarm optimization (PSO) [8], ant colony optimization (ACO)[14], and honey-bees mating optimization (HBMO) [10], which aim to find high-quality solutions within less computational cost,and attach more importance to the efficiency, rather than the optimality.

    In this background, optimal solutions for many open PP problems[8,15] have not been found yet.One important reason is that recent studies have paid more attention to improving the performance of intelligent algorithms than to the modification of current mathematical models, especially mixed-integer linear programming (MILP) mathematical models [14]. An effective MILP model can be solved by a mathematical programming solver, such as CPLEX,Gurobi,and so forth,and the optimal solution of a PP problem may be stably obtained under certain conditions.However,the solving effect of an MILP model is highly related to the problem scale. Once the scale becomes larger, the solving effect of the model deteriorates rapidly, and the model may even fail to find a feasible solution within a long computation time [16]. Therefore,current research work for PP does not focus on MILP models. In fact, the research status of MILP models is in a state of ‘‘compromise” because almost all reported PP models are feature-based[8,14,17]. These models convert an original network graph into a tabular form to simplify process representations.This kind of conversion, to some extent, changes the original process information of the network.(Discussions on this situation are presented in Section 3.3.) In addition, although there are other modeling methods that can avoid the abovementioned situation,they require complicated preprocessing, such as the generation of combinations of all possible processing operations,in order to eliminate and substitute OR nodes that represent a type of OR logic inside the network[18].This preprocessing procedure is complicated, in particular, the presence of too many OR nodes or a complex topological structure will greatly increase the complexity and computational time of a PP problem.

    In order to fill in the gap of recently reported research work, a novel MILP model is proposed based on the PP network topological structure. The main contributions of this paper are as follows:

    (1) This paper proposes a new OR-node-based MILP model for PP problems with no need for any feature conversion or other preprocessing.

    (2)The precedence relationships of the operations are discussed in detail, and three types of precedence relationship matrices are presented to illustrate the precedence relationship constraints.

    (3) Coded in the general algebraic modeling system (GAMS)/CPLEX solver,this MILP model successfully finds new optimal solutions for open problems in the literature [8,15,16].

    The rest of this paper is organized as follows. Section 2 introduces the related work, while Section 3 proposes the model and discusses the related analysis mentioned above in detail.Section 4 presents several groups of comparative experiments to verify the advantage of the proposed model. Section 5 provides conclusions and outlines some future work.

    2. Related work

    The related work on PP can be divided into two categories:algorithms and mathematical models. Xu et al. [6] and Leo Kumar [4]have both provided good reviews of PP problems. Intelligent algorithms such as GA, simulated annealing (SA) algorithms, TS algorithms, and PSO algorithms have shown sufficient advantages and have been widely applied in PP problems[6].To solve PP problems with complex prismatic parts, Li et al. [19] proposed a good hybrid algorithm comprising a GA and a SA algorithm. The local search capability of the hybrid algorithm was enhanced by a strategy of searching and selecting solutions based on the Hamming distance between alternative routes. Hua et al. [20] proposed a synthesis algorithm based on a GA to search for the global or near-global optimal solution in PP problems.Li et al.[15]proposed an effective genetic programming (GP) algorithm, which performed genetic operations on OR-node branches in the network graph. Taking into account manufacturing resources such as tool selection and feed direction, Shin et al. [21] proposed a symbiotic evolutionary algorithm to optimize three objectives, including balancing the machine workload, minimizing part movements,and minimizing tool changes. Wang et al. [22] put forward a PSO combining two local search strategies to solve the PP problem.Subsequently, Li et al. [8] presented a modified PSO to solve PP problems considering the transmission time between machines.Liu et al. [14] combined ACO with the constraint matrix and the state matrix of the problem,and applied it to solve the PP problem of two prismatic parts.

    Although some achievements have been made in applying intelligent algorithms to solve PP problems,the solution quality can be further improved in most of the existing PP instances,since metaheuristic algorithms cannot guarantee the optimality of the solutions.Besides,as a method of describing problems,a mathematical model can help researchers to understand and comprehend problems more deeply and thoroughly [23]. Therefore, research on PP mathematical models is very meaningful.Floudas and Lin[24]analyzed several mixed-integer programming (MIP) models of PP problems on the methods of time representation, and then proposed several effective optimization approaches to improve the computational efficiency of the model.In view of the complementarity of PP and scheduling, Li et al. [25] established a mathematical model to integrate a PP problem and shop scheduling problem.Xia et al. [26] proposed a feature-based mathematical model for a reconfigurable PP problem. Similarly, based on features, Jin and Zhang [16] established MILP models for PP considering the transmission time between machines.

    To the best of our knowledge,mathematical models for PP problems are always feature-based [8,14,17]. Although this kind of modeling method can describe most kinds of jobs,it is still inevitable that some direct precedence constraints must be added,which do not exist in the original network [16]. Therefore, the network graph method is more capable of describing different manufacturing flexibilities than a feature-based method.This paper proposes a novel MILP model directly based on the topology of the network graph; it can describe all types of manufacturing flexibilities of PP without adding or omitting any constraints.By solving the proposed model, new optimal solutions of some famous benchmarks in previous publications are successfully obtained.

    3. The proposed MILP model for PP

    3.1. Problem description

    The types of flexibilities in PP include process flexibility,machine selection flexibility, and operation sequencing flexibility.There are several approaches to describe PP problems,such as Petri nets [27], feature tables [28], AND/OR graphs, and networks[29,30]. In Fig. 1, the manufacturing flexibilities are represented in the form of a network graph. This network graph is composed of five types of nodes: The starting node, which is virtual, represents the start of a part’s production; the ending node, which is also virtual, indicates the end of a part’s production; intermediate nodes represent operations; and the OR node, combined with the fifth type of node,the JOIN node,represents the process flexibilities[15]. An intermediate node contains three pieces of information:the operation number in the solid circle, the alternative machine number in { }, and the corresponding processing time in [ ]. For example, the intermediate node 6 indicates that operation 6 can be processed on any machine out of the three alternative machines 3, 7, and 13, and the required processing time is 44, 48, and 49,respectively.The time unit in the paper is omitted just like original data. Arrows connecting pairs of nodes in the network indicate precedence relationship constraints between operations [21]. For example, the arrow between operations 2 and 3 declares that operation 2 must be processed before operation 3. There are no fixed precedence constraints between operations not connected by an arrow. Only one link connected to the OR node will be selected. The operations contained in the link with other operations compose a feasible operation combination for the part [29].Take Fig.1 as an example:If operations 2 →3 are selected between OR node 1 and JOIN node 1,and if operation 7 is selected between OR node 2 and JOIN node 2,then one of the complete feasible process routes is 1 →2 →3 →5 →6 →7 →9 →10.

    Fig. 1. A flexible process plan network. OR1, OR2, JOIN1, and JOIN2 represent OR node 1, OR node 2, JOIN node 1, and JOIN node 2, respectively.

    3.2. Precedence relationship between operations

    According to the definitions of binary variables, there are three modeling methods for PP problems [31]:

    For the large amount of variables and constraints, the binary variable θjtbased on time periods is rarely mentioned in PP problem modeling.In the existing literature,the definition ρjtdescribes the operation sequence by the position of operation j [16,32,33],which is widely used in the current literature [24,34,35]. In this paper, the proposed MILP model for PP is established based on the third method, qjj?, for the first time.

    As shown in Fig. 2, a feasible operation sequence of this part is 1 →2 →3 →4 →5 →6 →7 →8 →9.Operation 1 is in front of all the other eight operations; thus, according to the definition, q1j?=1,where j?=2, ..., 9. Operation 2 is processed before all the other seven operations, so q2j?=1, where j?=3, ..., 9. The number of‘‘1”s in the matrix Q=[qjj?] can be obtained as (n–1)n/2, where n stands for the total number of operations,which is also the length of the operation sequence. Fig. 3 shows the precedence relationship transformation from a sequence to a Q matrix.

    For a sequence that has been ordered, every two operations have a sequential relationship. Therefore, the priority relation can be represented by an n×n matrix. The Q matrix contains all the precedence relationships of an operation sequence. The Q matrix can be regarded as a completely expressing matrix (CEM) for the precedence relationships of an operation sequence. Through the observation and analysis, the characteristic constraints of CEM Q can be concluded as follows:

    (1) The diagonal elements of CEM Q are equal to 0:

    (2) The sum of the two elements symmetrical about the diagonal is equal to 1:

    (3) The sum of the elements in any two different columns is not equal:

    Fig. 2. A process plan network.

    Fig. 3. The transformation of the precedence relationship.

    Eq. (1) indicates that precedence relationships only exist between different operations. In Eq. (2), qjj?=1 makes qj?j=0,whereas qjj?=0 makes qj?j=1, because there is only one precedence relationship between the two operations. Eq. (3) is established because the sum of the column elements corresponds to the position that is unique in the operation sequence. The Q matrix contains all the precedence relationships between operations.According to the corresponding Q matrix, it can be quickly and easily determined whether a sequence satisfies the precedence constraints in a network.

    The notation sjj?is defined to represent the precedence constraints in Fig. 2:

    The corresponding constraint matrix S=[sjj?] is shown in Fig. 4,where each‘‘1”in the S matrix corresponds to an arrow in the network, representing a precedence constraint. If a sequence satisfies all the precedence constraints,its CEM Q is supposed to contain all the precedence values shown in matrix S,which can be formulated as follows:

    Matrix S can be generated from the network graph, but the matrix Q is unknown because the operation sequence has not been determined yet. Therefore, Eq. (4) can be regarded as the constraint.

    Fig. 4. Precedence constraints in a network and the corresponding matrix S.

    The number of ‘‘1”s in CEM Q is (n–1)n/2. However, the least number of ‘‘1”s required to determine an operation sequence in a matrix is (n–1). For example, to determine the sequence 1 →2 →3 →4 →5 →6 →7 →8 →9, it only needs to set eight variables as‘‘1,”which are q12,q23,q34,q45,q56,q67,q78,and q89.From this point,it is helpful to define a notation vjj?whose corresponding matrix V=[vjj?]contains the least number of variables equal to 1.The matrix V can be regarded as an exactly expressing matrix (EEM) of the precedence relationships of the sequence,which contains the exact and least number of ‘‘1”s. The EEM V of the above operation sequence is shown in Fig. 5.

    The EEM matrix can also be directly called the precedence matrix, since it just contains the precedence relationship of the two directly adjacent operations. According to the direct precedence relationships, it is simple to obtain an operation sequence by sequentially identifying the elements of the EEM V. The EEM matrix has several characteristic constraints as follows:

    (1) The number of variables that equal to 1 in the EEM V is(n–1):

    (2) Each row or column of the EEM V has at most one element equal to 1:

    Fig. 5. The EEM V of the sequence.

    (3) The relationship between matrix Q and matrix V is expressed as follows:

    Matrix V is a simplified representation of matrix Q, and both of them can determine a unique operation sequence. However,matrix S cannot determine a unique operation sequence due to its incomplete representation of the precedence relationships of a determined sequence. Therefore, matrix S is named the partly expressing matrix (PEM) in this paper. More than one sequence can satisfy the precedence constraints expressed by an S matrix.CEM Q, EEM V, and PEM S are shown in Fig. 6.

    3.3. A discussion on feature-based and network-based process representations

    In the current literature, all the mathematical models for PP problems are feature-based. For example, an instance adopted in the literature [8,16] is shown in feature tabular form in Fig. 7(a).In fact, this instance is derived from the 18th example in Ref.[29], as shown in Fig. 7(b). Therefore, Figs. 7(a) and (b) illustrate the same instance expressed in two different representation forms.

    In the feature table[8,16],features F2,F5,and F9have alternative operations or operation sets, corresponding to the three OR nodes OR1, OR2, and OR3 in the network graph. However, in the tabular form,the operation sets under the same feature are constrained by direct precedence relationships that do not exist in the corresponding network graph. For example, if O4–O5is chosen for feature F2,then O5must be processed directly after O4. However, O5does not have to be processed immediately after O4according to the original network graph Fig. 7(b). As a result, the solution space might be changed,leading to the failure to obtain the optimal solution.Table 1 shows two optimal solutions respectively obtained by solving a feature-based model and a network-based model with the GAMS/CPLEX solver. The letter M with number subscript in the bracket means the allocated machine, and it is the same in the follow-up tables.

    Observed from Table 1,the main distinction between these two sequences is that the sequence obtained by the network-based method does not have the direct precedence constraint of O4–O5.In addition, the production time of 356 obtained by the networkbased method is superior to the value of 357 obtained by the feature-based method. Therefore, the method of building a model directly on the basis of the network is superior to the featurebased method.

    Fig. 6. Three types of precedence relationship matrices: (a) CEM Q, (b) EEM V, and (c) PEM S.

    Fig. 7. Two representation forms of the same instance: (a) feature tabular form and (b) network graph.

    Table 1 Two optimal solutions obtained by using the feature-based model and the network-based model.

    The process flexibility of the feature-based representation is implemented through alternative operations or operation sets for features. For the network-based representation, the process flexibility is expressed by selecting the links of OR nodes.The links refer to the arrow connected to the OR nodes.Each link of the OR nodes corresponds to a choice for the process flexibility.In Fig.8,according to the order from the top to bottom and the left to right,the OR nodes and their links are numbered as shown.If link 1 of OR node 1 is selected, then operations 2, 3, and 4 will be selected. Additionally, whether link 1 or 2 of OR node 2 is selected, operations 6 and 7 will not be selected. The reason is that, in addition to OR node 2,operations 5,6,7,and 8 are controlled by link 2 of OR node 1. The selection performed by the OR node is valid only if the link at which this OR node lies is selected.

    A binary parameter wjrlis introduced to describe the control function of OR nodes on the operations. The definition of the parameter wjrlis stated as follows:

    For the example in Fig. 8, the corresponding values of wjrlare shown in Table 2.

    Hence,the controlling function of the OR node can be concluded as:Operation j will be selected only under the condition that all the controlling links of operation j are selected.On the contrary,operation j will not be selected as long as one of its controlling links is not selected. A binary variable urlis introduced to describe the choice of the links, and another binary variable xjis used to describe the operation choosing state. The definitions of urland xjare given as follows:

    The model established in Section 3.4 is based on these two variables, urland xj, and on a parameter, wjrl.

    3.4. Mathematical model for PP

    The proposed MILP model is precedence-based and OR-nodebased. Most of the process optimization objective functions are time related [8] or cost related [14,18]. In this paper, with the objective of minimizing the production time,the transmission time between machines is taken into account in the model. The sets,subscripts, parameters, and variables of the model are introduced below (Table 3).

    The total production time as the objective can be formulated as follows:

    On the right side,the first part of Eq.(9)refers to the total transmission time,and the second is the total processing time.The constraints of the model are displayed as follows:

    (1) OR-node controlling constraints:

    Fig. 8. An example of a controlling discussion.

    The constraint in Eq. (10) indicates the ‘‘unselected” condition:Operation j will not be selected as long as one of all operation j’s controlling links is not selected. The constraint in Eq. (11) is the‘‘selected” condition. The constraint in Eq. (12) means that only one link of an OR node can be chosen.

    (2) Precedence constraints:

    Eqs.(13)–(18)refer to the constraints in the precedence relationships.Unlike Eqs.(1)–(4),the operation selecting condition is added to this group of constraints.Eq.(13)corresponds to Eq.(1).Eqs.(14)and(15)correspond to Eq.(2),on the condition that operation j and j? are selected. Under the same condition, the constraint in Eq. (16)corresponds to Eq.(3).As for Eq.(17),the meaning is that the precedence relationships qjj?and qj?jshould be set as 0 if operation j is unselected. The constraint in Eq. (18) guarantees that the selected operation sequence obeys the precedence constraints.

    (3) EEM V and CEM Q constraints:

    Since a matrix V contains the precedence relationships between the two adjacent operations, the transmission time can easily be calculated. The constraints in Eqs. (19)–(21) describe the property of matrix V. Because matrix V is derived from matrix Q, Eq. (22) is presented.

    (4) Machine selection constraint:

    Eq. (23) means that there is only one machine that can be assigned for the selected operation, and there is no need to assign any machine for an unselected operation.

    Eqs.(24)and(25)formulate the transmission time of operation j from the current processing machine to the next machine.

    4. Experiments and discussions

    To verify the proposed model, five groups of comparative experiments are carried out based on famous benchmarks. All the experiments are directly compared with the results of other reported methods. On a personal computer (PC) with 3.7 GHz and 16 GB random-access memory(RAM),the proposed model is coded in the GAMS, and the solver CPLEX is used to solve the PP problems. In this paper, the parameter Gap (%) is also introduced to evaluate the proposed model and the computation results. The Gap value represents the relative tolerance of the obtained solution;its definition is (BF–BP)/BP,where BF is the current bestsolution of the objective function, and BP is the lower bound. The smaller the Gap value is, the closer the current solution is to the optimal solution. The computation time of GAMS/CPLEX is set as 3600 s. If the optimal solution is not found within the time limit,the computation will be terminated and the best known solution will be output. The transmission time between machines adopted by the cases in experiments 1,2,4,and 5 is shown in Table 4[8,18].

    Table 2 The values of parameter wjrl.

    Table 3 Definitions of the sets, subscripts, parameters, and variables of the mathematical model for PP.

    4.1. Experiment 1

    The three cases in experiment 1 are adopted from Jin and Zhang[16], where a dynamic programming (DP)-like heuristic algorithm is applied to solve the PP problem. The results obtained by the proposed MILP model and the DP-like heuristic are presented in Table 5. The production time of case 1 obtained by the MILP method is 357, which is better than the 360 provided by the DP-like heuristic. Furthermore, all three optimal solutions to the cases are found by the MILP model.

    4.2. Experiment 2

    The four cases in experiment 2 are from different publications:Case 1 comes from Zhang and Nee[36],and cases 2–4 come from Li and McMahon[37].The detailed information of these cases can be found in the corresponding papers. Computational results from a modified PSO algorithm are given in the work of Li et al. [8]. The modified PSO algorithm is one of the state-of-the-art algorithms that are used to solve combinatorial optimization problems. The comparison between PSO and MILP is given in Table 6. It can be observed that the MILP model obtains the better solution in case 1. Furthermore, the optimal solutions for cases 2–4 are found by the MILP model method within short computational time (less than one second).

    4.3. Experiment 3

    The two cases in experiment 3 are adopted from Li et al. [15],and the machine transmission time matrix is shown in Table 7.These two cases employed the same part from Li et al. [15]. The only distinction between the two cases is that machine 2 is assumed to be broken down in case 2. The comparative results are listed in Table 8. Both the GP algorithm and the MILP model are able to find the optimal solutions of the two instances.

    4.4. Experiment 4

    The 17 cases in experiment 4 are adopted from the well-known Kim dataset [29], which consists of 18 parts, and the comparative calculation results are from Ref. [8]. Because there are some problems regarding the data of part 4 from Ref.[8],it is not selected here.The results obtained by the modified PSO algorithm,the simple GA,and the simple SA are presented in Table 9. Within a reasonable computational time,the proposed MILP model can find 13 optimal solutions out of 17 cases. For the cases that the optimal solutions are not found, the MILP model still obtains better solutions compared to the other algorithms,such as cases 3,6,12,and 15.

    4.5. Experiment 5

    The 11 cases in experiment 5 are adopted from another famous Shin benchmark [21], and the comparative calculation results are from Li et al. [8]. Because there are some problems regarding the data of some parts in Ref. [8], 11 parts (omitting parts 9, 10, 12,13,15,17,and 18)are selected in this experiment group.The solutions obtained by the modified PSO algorithm, the simple GA, and the simple SA are presented in Table 10.Within a reasonable computational time, the proposed MILP model can find nine optimal solutions out of 11 instances. For the cases that the optimal solutions are not found, the MILP model still obtains better solutions,such as case 3.

    Table 4 The transmission time matrix [8,18].

    Table 5 Comparative results of experiment 1.

    Table 6 Comparative results of experiment 2.

    Table 7 Transmission time between the machines.

    4.6. Discussion

    The proposed MILP model obtained 28 optimal solutions out of 37 instances within acceptable calculation time. The solutions found by the MILP model are better than those obtained by the high-performance heuristic [16] and meta-heuristic algorithms [8,15]. Experiments 4 and 5 were carried out on two widely-used benchmarks [21,29], and the better results suggest the superiority of the proposed model.

    As shown in Table 11, the proposed model contains four types of subscripts, that is, operation, machine, OR node, and links, while the model reported in Ref. [16] contains six types of subscripts, that is, feature, operation set, operation, machine,position, and place. Fewer subscripts in the proposed model make the computation more effective than that of the model developed by Ref. [16]. Furthermore, the OR-node-based modeling method makes the proposed model more universal for solving different types of PP problems. This is why most optimalsolutions can be obtained for both of the Kim [29] and Shin[21] benchmarks.

    Table 8 Comparative results of experiment 3.

    5. Conclusions and future work

    Considering the topology of network graph,this paper proposed a new MILP mathematical model based on OR nodes. Firstly, for precedence relationships between operations, three precedence matrices were introduced.Secondly,for better generality,the notations wjrl, url, and xjwere introduced to describe the controlling function of the OR nodes. Finally, the proposed MILP model were coded in the mathematical programming solver CPLEX and tested on public benchmarks. The extensive comparative results verified the correctness and superiority of the proposed model.

    In this work,an OR-node-based modeling method was proposed for the first time,demonstrating a new perspective for PP problems and their extension research.The analyses of the three precedence matrices in the paper also revealed the essence of the operation sequencing sub-problem,which is beneficial for further comprehension of the PP problem.However,there are still some limitations in this PP model research. Optimal solutions cannot be found for the minorityof instances,and the computationalefficiency is not always satisfactory,which implies that the proposed approaches can be further improved. Some simplification and speed-up strategies are urgently required for the further research work.

    Table 9 Comparative results of experiment 4.

    Table 10 Comparative results of experiment 5.

    Table 11 The subscripts of the models.

    Acknowledgements

    This work is supported in part by the National Natural Science Foundation of China (51825502 and 51775216), and in part by the Program for Huazhong University of Science and Technology(HUST) Academic Frontier Youth Team (2017QYTD04).

    Compliance with ethics guidelines

    Qihao Liu, Xinyu Li, and Liang Gao declare that they have no conflict of interest or financial conflicts to disclose.

    激情视频va一区二区三区| 波多野结衣av一区二区av| 黄色丝袜av网址大全| а√天堂www在线а√下载 | 人人妻人人爽人人添夜夜欢视频| 久久久精品区二区三区| 黄色毛片三级朝国网站| av福利片在线| 极品教师在线免费播放| 80岁老熟妇乱子伦牲交| 亚洲第一av免费看| 黄色毛片三级朝国网站| 91国产中文字幕| 露出奶头的视频| 少妇粗大呻吟视频| 久久久久久亚洲精品国产蜜桃av| 中文字幕人妻丝袜一区二区| 韩国精品一区二区三区| 国产精品98久久久久久宅男小说| 性少妇av在线| 国产成人精品久久二区二区91| 亚洲熟妇中文字幕五十中出 | 在线观看免费日韩欧美大片| 男女午夜视频在线观看| 亚洲av成人一区二区三| 久久久久久久久免费视频了| 久久久国产精品麻豆| 悠悠久久av| netflix在线观看网站| 美女 人体艺术 gogo| 久久久久久久精品吃奶| 亚洲av电影在线进入| 国产国语露脸激情在线看| 又黄又粗又硬又大视频| √禁漫天堂资源中文www| 少妇被粗大的猛进出69影院| 久久久久国产一级毛片高清牌| 人妻 亚洲 视频| 日韩中文字幕欧美一区二区| 人人妻人人添人人爽欧美一区卜| 波多野结衣一区麻豆| 90打野战视频偷拍视频| avwww免费| 亚洲全国av大片| 亚洲欧美日韩另类电影网站| 一区二区三区精品91| 高清视频免费观看一区二区| www日本在线高清视频| 成年女人毛片免费观看观看9 | 免费看十八禁软件| 亚洲伊人色综图| 中文字幕高清在线视频| 国精品久久久久久国模美| 男人操女人黄网站| 一区二区日韩欧美中文字幕| 久久精品亚洲精品国产色婷小说| 黑人欧美特级aaaaaa片| 一级,二级,三级黄色视频| 亚洲视频免费观看视频| 人妻 亚洲 视频| 亚洲专区中文字幕在线| 久久久久久亚洲精品国产蜜桃av| 国产蜜桃级精品一区二区三区 | www.自偷自拍.com| 黑人巨大精品欧美一区二区mp4| 如日韩欧美国产精品一区二区三区| 19禁男女啪啪无遮挡网站| 久久亚洲精品不卡| 女同久久另类99精品国产91| 91精品国产国语对白视频| 亚洲在线自拍视频| 久久香蕉国产精品| 午夜福利视频在线观看免费| 又黄又粗又硬又大视频| 丰满迷人的少妇在线观看| 中文字幕色久视频| 丝袜美腿诱惑在线| 妹子高潮喷水视频| 亚洲成人手机| 中文字幕av电影在线播放| 欧美日韩视频精品一区| 啦啦啦免费观看视频1| 999精品在线视频| 欧美人与性动交α欧美软件| 亚洲色图综合在线观看| 欧美成人免费av一区二区三区 | 国产欧美日韩精品亚洲av| 国产免费男女视频| 老司机午夜福利在线观看视频| 极品教师在线免费播放| 国产高清激情床上av| 国产在线一区二区三区精| 欧美成人午夜精品| 9191精品国产免费久久| 久久国产精品人妻蜜桃| 中出人妻视频一区二区| 一级作爱视频免费观看| 久久久久久久精品吃奶| 国产精品av久久久久免费| 国产精品二区激情视频| 久久久国产欧美日韩av| 涩涩av久久男人的天堂| av不卡在线播放| 两个人看的免费小视频| 欧美日韩成人在线一区二区| 中国美女看黄片| 桃红色精品国产亚洲av| 天天操日日干夜夜撸| 精品人妻熟女毛片av久久网站| 欧美黑人欧美精品刺激| 大香蕉久久网| 精品乱码久久久久久99久播| 一区二区三区激情视频| 亚洲成国产人片在线观看| 18在线观看网站| 中国美女看黄片| 久久精品国产亚洲av高清一级| 精品一品国产午夜福利视频| 老熟妇仑乱视频hdxx| 一边摸一边抽搐一进一出视频| 欧美 亚洲 国产 日韩一| 久久人妻av系列| 91麻豆av在线| 女人被狂操c到高潮| av福利片在线| 后天国语完整版免费观看| 国产欧美日韩综合在线一区二区| 亚洲欧美激情综合另类| 欧美日韩福利视频一区二区| 欧美精品高潮呻吟av久久| 999精品在线视频| 一区二区三区激情视频| 一级黄色大片毛片| 色老头精品视频在线观看| 久久青草综合色| 亚洲精品久久午夜乱码| 人人澡人人妻人| 精品一区二区三区av网在线观看| 久久草成人影院| 久久草成人影院| 18禁裸乳无遮挡动漫免费视频| 欧美精品一区二区免费开放| 国产黄色免费在线视频| cao死你这个sao货| 99精国产麻豆久久婷婷| 国产不卡av网站在线观看| 国产精品香港三级国产av潘金莲| 欧美黑人精品巨大| 美女午夜性视频免费| 国产精品久久久av美女十八| 中文亚洲av片在线观看爽 | 看片在线看免费视频| 久久天躁狠狠躁夜夜2o2o| 中文字幕人妻丝袜一区二区| 宅男免费午夜| 日日爽夜夜爽网站| 久久九九热精品免费| 一二三四社区在线视频社区8| 亚洲精品中文字幕在线视频| 不卡av一区二区三区| 看黄色毛片网站| 亚洲免费av在线视频| 人成视频在线观看免费观看| 亚洲,欧美精品.| 国产在线一区二区三区精| a级毛片黄视频| 婷婷丁香在线五月| 国产av精品麻豆| 欧美亚洲日本最大视频资源| √禁漫天堂资源中文www| 亚洲一区二区三区欧美精品| 欧美日韩中文字幕国产精品一区二区三区 | av国产精品久久久久影院| 国产成人欧美在线观看 | 免费不卡黄色视频| 一边摸一边抽搐一进一小说 | 日韩欧美一区视频在线观看| 久久精品亚洲熟妇少妇任你| 丰满人妻熟妇乱又伦精品不卡| 男人操女人黄网站| av超薄肉色丝袜交足视频| 亚洲精品美女久久av网站| 91成人精品电影| 妹子高潮喷水视频| 激情在线观看视频在线高清 | 成人精品一区二区免费| 亚洲第一欧美日韩一区二区三区| 亚洲五月天丁香| 黄色视频,在线免费观看| 国产91精品成人一区二区三区| 精品亚洲成a人片在线观看| 热re99久久国产66热| 久99久视频精品免费| 一区福利在线观看| 99国产精品一区二区三区| 淫妇啪啪啪对白视频| 亚洲精品乱久久久久久| 午夜老司机福利片| 如日韩欧美国产精品一区二区三区| 午夜福利在线免费观看网站| 久久精品国产清高在天天线| 亚洲视频免费观看视频| av在线播放免费不卡| 国产精品一区二区在线不卡| 精品免费久久久久久久清纯 | 国产av精品麻豆| 国产男女超爽视频在线观看| 亚洲成人国产一区在线观看| 日本五十路高清| 超碰成人久久| 大码成人一级视频| 99久久精品国产亚洲精品| 操出白浆在线播放| 久久精品亚洲av国产电影网| 俄罗斯特黄特色一大片| 国产精品.久久久| 国产淫语在线视频| 亚洲欧美色中文字幕在线| 丝瓜视频免费看黄片| 曰老女人黄片| 久久精品国产99精品国产亚洲性色 | 国产精品偷伦视频观看了| 亚洲精品国产精品久久久不卡| 国产精品免费一区二区三区在线 | 女人高潮潮喷娇喘18禁视频| 丝瓜视频免费看黄片| 老司机在亚洲福利影院| 亚洲精品av麻豆狂野| 中文欧美无线码| 夜夜爽天天搞| 大陆偷拍与自拍| 午夜影院日韩av| 三上悠亚av全集在线观看| 国产精品亚洲一级av第二区| 一本综合久久免费| 黄色毛片三级朝国网站| 两个人看的免费小视频| 亚洲第一欧美日韩一区二区三区| 精品久久久久久电影网| aaaaa片日本免费| 欧美亚洲 丝袜 人妻 在线| 90打野战视频偷拍视频| 久久中文字幕人妻熟女| videosex国产| 国产蜜桃级精品一区二区三区 | 国产精品香港三级国产av潘金莲| 一a级毛片在线观看| 欧美亚洲日本最大视频资源| 一本大道久久a久久精品| 高潮久久久久久久久久久不卡| 黄色丝袜av网址大全| 久久青草综合色| 王馨瑶露胸无遮挡在线观看| 国产高清国产精品国产三级| 国产精品秋霞免费鲁丝片| 国产av精品麻豆| 99re6热这里在线精品视频| 老汉色av国产亚洲站长工具| 18禁国产床啪视频网站| 亚洲国产精品合色在线| 国产1区2区3区精品| 欧美性长视频在线观看| 大码成人一级视频| 成年女人毛片免费观看观看9 | 两个人免费观看高清视频| 欧美av亚洲av综合av国产av| 国产精品影院久久| 亚洲美女黄片视频| 国产xxxxx性猛交| 不卡一级毛片| 国产精品 欧美亚洲| 国产国语露脸激情在线看| 日本a在线网址| 亚洲精华国产精华精| 精品久久久久久,| 999精品在线视频| 亚洲国产欧美日韩在线播放| 999久久久精品免费观看国产| 又紧又爽又黄一区二区| 激情视频va一区二区三区| 国产高清视频在线播放一区| 女性被躁到高潮视频| 亚洲成人手机| 国产在视频线精品| 久久 成人 亚洲| 啦啦啦在线免费观看视频4| 黄片播放在线免费| 成人影院久久| 国产在线精品亚洲第一网站| 两个人看的免费小视频| www.精华液| 欧美精品av麻豆av| www.熟女人妻精品国产| 国产精品久久电影中文字幕 | 国产精品二区激情视频| 黄片小视频在线播放| 久久久精品免费免费高清| 午夜福利一区二区在线看| 人妻 亚洲 视频| avwww免费| 国产不卡一卡二| 悠悠久久av| 自线自在国产av| 亚洲专区字幕在线| 超碰成人久久| 午夜91福利影院| 精品高清国产在线一区| 欧美性长视频在线观看| 又黄又粗又硬又大视频| 韩国av一区二区三区四区| 欧美人与性动交α欧美软件| 亚洲熟女毛片儿| 亚洲熟女毛片儿| 中文字幕高清在线视频| 人人澡人人妻人| 久久热在线av| 国产精品亚洲av一区麻豆| 国产精品免费大片| 99久久精品国产亚洲精品| 欧美成人免费av一区二区三区 | 一二三四社区在线视频社区8| 大陆偷拍与自拍| 18在线观看网站| 狠狠狠狠99中文字幕| 动漫黄色视频在线观看| 国产精品乱码一区二三区的特点 | 久久久久国内视频| 18禁美女被吸乳视频| 99国产综合亚洲精品| 成年动漫av网址| 日韩欧美三级三区| 少妇被粗大的猛进出69影院| 日韩精品免费视频一区二区三区| 精品人妻在线不人妻| www.999成人在线观看| 大陆偷拍与自拍| 夜夜爽天天搞| 亚洲色图 男人天堂 中文字幕| 久久精品91无色码中文字幕| 欧美精品人与动牲交sv欧美| 国产精华一区二区三区| 亚洲精品自拍成人| 在线观看日韩欧美| av超薄肉色丝袜交足视频| 桃红色精品国产亚洲av| 香蕉国产在线看| 女警被强在线播放| 女人被狂操c到高潮| av网站在线播放免费| 国产精品永久免费网站| 淫妇啪啪啪对白视频| 国产成人精品久久二区二区免费| 色综合婷婷激情| 亚洲va日本ⅴa欧美va伊人久久| a级毛片黄视频| av免费在线观看网站| 身体一侧抽搐| 国产真人三级小视频在线观看| 无遮挡黄片免费观看| 日本欧美视频一区| 久久天堂一区二区三区四区| 国产精品二区激情视频| 叶爱在线成人免费视频播放| 麻豆av在线久日| 久久久精品国产亚洲av高清涩受| 亚洲综合色网址| 午夜成年电影在线免费观看| 波多野结衣一区麻豆| 亚洲精品国产一区二区精华液| 一级片'在线观看视频| 看免费av毛片| 黄色怎么调成土黄色| 国产1区2区3区精品| 国产在线一区二区三区精| 亚洲精品成人av观看孕妇| 高清av免费在线| 精品久久久久久久久久免费视频 | 免费看十八禁软件| 国产欧美日韩一区二区精品| 这个男人来自地球电影免费观看| 女警被强在线播放| 欧美人与性动交α欧美精品济南到| 在线观看免费午夜福利视频| 18在线观看网站| 亚洲精品乱久久久久久| 50天的宝宝边吃奶边哭怎么回事| 欧美激情久久久久久爽电影 | 久久狼人影院| 一级a爱片免费观看的视频| 久久久水蜜桃国产精品网| 亚洲性夜色夜夜综合| 亚洲一区二区三区欧美精品| 一进一出抽搐动态| 久久ye,这里只有精品| 欧美 亚洲 国产 日韩一| www日本在线高清视频| 亚洲国产精品合色在线| 亚洲精品国产区一区二| 国产深夜福利视频在线观看| 黄色女人牲交| 一二三四社区在线视频社区8| 亚洲国产精品一区二区三区在线| 欧美精品一区二区免费开放| 精品无人区乱码1区二区| 国产aⅴ精品一区二区三区波| 久久精品亚洲熟妇少妇任你| 自线自在国产av| 一进一出好大好爽视频| 亚洲成av片中文字幕在线观看| 天天添夜夜摸| 久久亚洲真实| 超色免费av| 亚洲欧美激情综合另类| 丰满迷人的少妇在线观看| 精品高清国产在线一区| 757午夜福利合集在线观看| 天堂俺去俺来也www色官网| 欧美 日韩 精品 国产| 99re在线观看精品视频| 一二三四社区在线视频社区8| 国内毛片毛片毛片毛片毛片| 极品人妻少妇av视频| 黄片小视频在线播放| 亚洲av熟女| 欧美人与性动交α欧美精品济南到| 久久久久精品国产欧美久久久| 成人永久免费在线观看视频| 午夜视频精品福利| 国产精品久久久人人做人人爽| 少妇裸体淫交视频免费看高清 | 亚洲欧美一区二区三区久久| 欧美人与性动交α欧美软件| 美女视频免费永久观看网站| 午夜福利一区二区在线看| 国产成人精品无人区| 极品人妻少妇av视频| xxx96com| 欧美日韩av久久| 色在线成人网| 久久国产精品大桥未久av| 这个男人来自地球电影免费观看| 久久久国产成人精品二区 | 精品欧美一区二区三区在线| 女人爽到高潮嗷嗷叫在线视频| 18禁观看日本| 国产午夜精品久久久久久| 中文欧美无线码| 国内久久婷婷六月综合欲色啪| xxx96com| 国产精品亚洲一级av第二区| 国产视频一区二区在线看| 精品久久久久久,| 极品少妇高潮喷水抽搐| 亚洲人成77777在线视频| 美女 人体艺术 gogo| 国产99久久九九免费精品| 欧美激情久久久久久爽电影 | 国产一区有黄有色的免费视频| 亚洲中文av在线| 欧美日韩乱码在线| 免费观看人在逋| 日韩有码中文字幕| 我的亚洲天堂| 欧美成人午夜精品| 国产精品一区二区在线观看99| 人人妻人人澡人人看| 国产精品电影一区二区三区 | 亚洲全国av大片| 女性生殖器流出的白浆| 丝袜人妻中文字幕| 午夜成年电影在线免费观看| 下体分泌物呈黄色| av在线播放免费不卡| 久久99一区二区三区| 女人精品久久久久毛片| netflix在线观看网站| 国产亚洲一区二区精品| 亚洲久久久国产精品| tocl精华| 少妇的丰满在线观看| 色老头精品视频在线观看| 午夜福利乱码中文字幕| 国产三级黄色录像| 一区在线观看完整版| 成人影院久久| 中文亚洲av片在线观看爽 | 亚洲国产毛片av蜜桃av| 91av网站免费观看| 久久久久国产一级毛片高清牌| 99久久综合精品五月天人人| 不卡av一区二区三区| 国产精品香港三级国产av潘金莲| 久久久久国内视频| 在线观看免费午夜福利视频| 操出白浆在线播放| 日本一区二区免费在线视频| 淫妇啪啪啪对白视频| 99国产精品一区二区三区| 自线自在国产av| 天堂√8在线中文| 别揉我奶头~嗯~啊~动态视频| 国产99久久九九免费精品| 老熟妇乱子伦视频在线观看| 久久精品熟女亚洲av麻豆精品| а√天堂www在线а√下载 | 最新美女视频免费是黄的| 色在线成人网| 日韩有码中文字幕| 国产精品亚洲一级av第二区| 男女之事视频高清在线观看| xxx96com| 午夜福利乱码中文字幕| 夜夜夜夜夜久久久久| 欧美 日韩 精品 国产| 久久精品成人免费网站| 中文字幕精品免费在线观看视频| 99在线人妻在线中文字幕 | 精品人妻1区二区| 黄片大片在线免费观看| 黄色丝袜av网址大全| 黑人巨大精品欧美一区二区mp4| 巨乳人妻的诱惑在线观看| 狠狠婷婷综合久久久久久88av| 99久久精品国产亚洲精品| 正在播放国产对白刺激| 又黄又粗又硬又大视频| 精品人妻1区二区| 亚洲人成77777在线视频| 99久久精品国产亚洲精品| 美国免费a级毛片| 91字幕亚洲| 亚洲精品粉嫩美女一区| 69精品国产乱码久久久| 欧美最黄视频在线播放免费 | 亚洲av片天天在线观看| 天堂俺去俺来也www色官网| 亚洲精品美女久久av网站| 精品国产一区二区久久| 淫妇啪啪啪对白视频| 亚洲精品国产区一区二| 岛国毛片在线播放| 三级毛片av免费| 欧美性长视频在线观看| 在线观看免费视频网站a站| 亚洲国产欧美一区二区综合| 成人影院久久| 黄色 视频免费看| 亚洲中文av在线| 色婷婷av一区二区三区视频| 在线观看免费高清a一片| 精品国产超薄肉色丝袜足j| 日韩欧美一区视频在线观看| 大型av网站在线播放| 国产片内射在线| 成年版毛片免费区| 80岁老熟妇乱子伦牲交| 国产色视频综合| 91精品国产国语对白视频| av线在线观看网站| 99国产精品免费福利视频| 久久这里只有精品19| 在线十欧美十亚洲十日本专区| 欧美av亚洲av综合av国产av| 亚洲精品自拍成人| 看片在线看免费视频| 欧美不卡视频在线免费观看 | 亚洲五月色婷婷综合| 国产主播在线观看一区二区| 一级片免费观看大全| 久久久久久久久免费视频了| 日韩制服丝袜自拍偷拍| 中文字幕人妻丝袜一区二区| 老司机午夜十八禁免费视频| 91在线观看av| 久久久精品区二区三区| 久久久国产精品麻豆| 老司机午夜福利在线观看视频| 韩国av一区二区三区四区| 国产精品自产拍在线观看55亚洲 | a级毛片在线看网站| 日日夜夜操网爽| 国产男靠女视频免费网站| 在线观看午夜福利视频| 免费观看精品视频网站| 天天影视国产精品| 1024视频免费在线观看| 亚洲一卡2卡3卡4卡5卡精品中文| 真人做人爱边吃奶动态| 亚洲中文字幕日韩| 亚洲人成77777在线视频| www.精华液| 午夜精品国产一区二区电影| 少妇裸体淫交视频免费看高清 | 法律面前人人平等表现在哪些方面| 午夜免费鲁丝| 午夜精品国产一区二区电影| 久久99一区二区三区| 一级毛片女人18水好多| 99国产极品粉嫩在线观看| 国产精品久久久久久人妻精品电影| xxxhd国产人妻xxx| 日韩三级视频一区二区三区| 婷婷丁香在线五月| 两性午夜刺激爽爽歪歪视频在线观看 | 精品国产一区二区久久| 又黄又爽又免费观看的视频| 男女床上黄色一级片免费看| av片东京热男人的天堂| 午夜激情av网站| 国产一区二区激情短视频| 狂野欧美激情性xxxx| 超碰97精品在线观看| 精品国产一区二区三区久久久樱花| 欧美精品av麻豆av| 亚洲欧美精品综合一区二区三区|