Scholarly article on topic 'Process planning optimization on turning machine tool using a hybrid genetic algorithm with local search approach'

Process planning optimization on turning machine tool using a hybrid genetic algorithm with local search approach Academic research paper on "Mechanical engineering"

Share paper
Academic journal
Advances in Mechanical Engineering
OECD Field of science

Academic research paper on topic "Process planning optimization on turning machine tool using a hybrid genetic algorithm with local search approach"

Advances in


Research Article Engineering

Process planning optimization on turning machine tool using a hybrid genetic algorithm with local search approach

Yuliang Su1, Xuening Chu1, Zaifang Zhang2 and Dongping Chen1


A turning machine tool is a kind of new type of machine tool that is equipped with more than one spindle and turret. The distinctive simultaneous and parallel processing abilities of turning machine tool increase the complexity of process planning. The operations would not only be sequenced and satisfy precedence constraints, but also should be scheduled with multiple objectives such as minimizing machining cost, maximizing utilization of turning machine tool, and so on. To solve this problem, a hybrid genetic algorithm was proposed to generate optimal process plans based on a mixed 0-1 integer programming model. An operation precedence graph is used to represent precedence constraints and help generate a feasible initial population of hybrid genetic algorithm. Encoding strategy based on data structure was developed to represent process plans digitally in order to form the solution space. In addition, a local search approach for optimizing the assignments of available turrets would be added to incorporate scheduling with process planning. A real-world case is used to prove that the proposed approach could avoid infeasible solutions and effectively generate a global optimal process plan.


Process planning, turning machine tool, genetic algorithm, operation sequencing, local search approach

Date received: 19 November 2014; accepted: 2 March 2015 Academic Editor: Pietro Scandura

Advances in Mechanical Engineering 1-14

© The Author(s) 2015

DOI: 10.1177/1687814015581241



The purpose of computer-aided process planning (CAPP) is to transform a part design specification from a computer-aided design (CAD) model into a set of manufacturing instructions used by the computer-aided manufacturing (CAM) system. CAPP is usually composed of several consecutive activities: (1) machining feature recognition, which recognizes manufacturing information such as features of slots, pockets, and so on from the geometry of a part in a CAD system described by low-level entities, such as points, curves, surfaces, and their relationships; (2) machining operation types selection, which is used to finish a feature of

the designed part based on the manufacturing capability of available machine tools in workshop; and (3) process optimization, which sequences machining operations so that the least machining cost of the part

' School of Mechanical Engineering, Shanghai Jiao Tong University, Shanghai, P.R. China

2School of Mechatronic Engineering and Automation, Shanghai University, Shanghai, P.R. China

Corresponding author:

Xuening Chu, School of Mechanical Engineering, Shanghai Jiao Tong University, Shanghai 200240, P.R. China. Email:

Creative Commons CC-BY: This article is distributed under the terms of the Creative Commons Attribution 3.0 License ( which permits any use, reproduction and distribution of the work without further permission provided the original work is attributed as specified on the SAGE and Open Access pages ( openaccess.htm).

can be obtained. The manufacturing capability of machine tool has a significant influence on process planning.

The demand of producing parts with high efficiency and accuracy leads to a new generation of multifunction computer numerical control (CNC) machine tool, named turning machine tool.1 Turning machine tool could perform several operations at the same time with multiple turrets on its multiple spindles. A spindle with three-jaw chuck of turning machine tool is called part holding location (PHL), and turret equipped with servo motors is called power turret (PT).2 Turning machine tool became popular recently, because it could perform both milling and turning operations. The multiple functions require fewer setups, which reduces workpiece transferring and setup time significantly. Fewer setups could eliminate or reduce setup errors, thus increase the product accuracy. Another advantage is that the turning machine tool is equipped with new manufacturing capabilities, simultaneous machining mode, and parallel machining mode. In the first machining mode, a feature or an operation could be performed by two PTs simultaneously. In the second mode, PTs perform different operations. By contract, using traditional machine tools, only one operation could be performed at a time. Consequently, a turning machine tool could be considered as a replacement for two traditional machine tools (a milling and a lathe machine tool). It is valuable for a small company with limited working area in the workshop.

The process planning for turning machine tool was a hot topic in recent years. Levin and Dutta3 developed a prototype CAPP system for turning machine tool. Waiyagan and Bohez4 classified the manufacturing parts into three main groups, rotational, prismatic, and prisronal parts (mill-turn part) containing both rotational and prismatic features. Kim et al.5 developed a feature-based geometric reasoning system to represent the mill-turn part digitally. Chiu et al.6 formulated the problem as operation sequencing problem on parallel machine tools and proposed a genetic algorithm (GA) to solve it, but failed to make use of simultaneous machining mode. Norman and Bean7 considered operation sequencing problem for turning machine tool as scheduling problem with a fixed operation sequence. Chung and Suh8 developed a branch-and-bound approach-based heuristic algorithm to generate process plans for mill-turn parts. They also developed a rule-based expert system to perform process planning.

This careful study reveals that precedence constraints would be taken into consideration as well as scheduling problem, since performing one operation should not only satisfy the precedence constraints, but also wait for available time of PT or PHL of the turning machine tool. Lack of consideration of precedence constraints may result in infeasible machining sequence, while lack

of consideration of scheduling may result in conflicts of PTs and unproductive waiting time. However, the existing approaches still have difficulties in handling the complex precedence constraints, not to mention the integration of process planning and scheduling.9 Precedence constraints could be handled with two main kinds of approaches. The first one repairs infeasible solutions after operation sequencing with additional modification algorithm, which may result in low effi-ciency.10,11 The second one tries to avoid infeasible solutions at the very beginning using specially designed encoding strategy such as topological storing-based representation method.12,13

These above approaches did not pay enough attention to scheduling problem, while integrated process planning and scheduling (IPPS) has attracted more and more attention in the last two decades. Saygin14 prepared a few candidate process plans for scheduling and performed re-scheduling procedure when alternative machine tools were used. Zhang et al.15 generated the optimal schedule by iteratively selecting alternative process plans until the predefined optimization objective was achieved. Zhang and Yan16 formulated IPPS as a nonlinear mixed integer program model and used a hybrid GA (HGA) to solve it. Mohammadi et al.17 used a hybrid simulated annealing (SA) algorithm to search the optimal or near-optimal solution in the combined solution space of process planning and scheduling.

In summary, the usage of alternative process plans could not cope with the changes of manufacturing recourse in time during scheduling.14,15 The trend is to perform process planning and scheduling simultane-ously.16,17 However, it is difficult for these optimization algorithms to handle complicated precedence relationships among operations efficiently.12 These algorithms are approximate approaches that merely attempt to generate the near-optimal solutions within acceptable computational cost, as well as ant colony optimization (ACO), tabu search (TS) algorithms.9,18 20 Furthermore, the involved scheduling problem could lead to an increase in complexity.

To overcome this problem, this article proposes a HGA and local search approach for process planning on turning machine tool, which regards sequencing and scheduling as a whole. According to the reviews of CAPP, GA has a significant advantage over other algorithms for precedence constrained operation sequencing problem (PCOSP).21,22 GA is designed to generate the global solution, though it would be trapped by local optimum sometimes. Hence, a local search approach is added to improve the convergence of GA. Due to its parallel processing capability, GA could optimize the selection of PHLs and cutting tools for each operation, sequence of operations, and assignments of PTs simul-taneously.23 25 Considering the complexity of this problem, a data-structure-based encoding strategy is

Figure!. Turning machining tool.

introduced to represent all process information for operation sequencing and scheduling. The precedence constraints are derived from the analysis of geometrical relationship constraints, datum reference precedence constraints, fixed-order constraints of machining operations, and good manufacturing practice. Precedence constraints would be represented in the form of an operation precedence graph (OPG) that is a directed acyclic graph.26 The vertices in OPG represent the operations, while the directed edges represent precedence constraints. To improve the decision-making accuracy and efficiency, a local search approach is utilized to optimize manufacturing parameters (PHLs, PTs, and cutting tools). The assignments of PHLs and PTs could be regarded as scheduling that has already been integrated with process planning. The output could be used as the input and working introduction of CAM. The proposed approach is proven through a

case study of process planning for a mill-turn part on a turning machine tool equipped with two PHLs and two PTs.

Problem formulation

Turning machine tool has much more significant flexibility and versatility than traditional machine tool, due to multiple PHLs and PTs.2 The PHL refers to the spindle with three-jaw chuck that could hold a part by clamping one of the external cylinder surfaces. As shown in Figure 1, the most commonly used turning machine tool is equipped with dual spindles, a main spindle, and an auxiliary spindle. Stock could be loaded on the main spindle (PHL1 in Figure 1), while finished workpiece would be unloaded on the auxiliary spindle (PHL2 in Figure 1). Workpiece handling system guarantees that a workpiece could be transferred from one

PHL to the other automatically, so that turning machine tool could save setup time and reduce setup errors.27 PT is a kind of turret with several cutting heads, some of which are equipped with electromotor and additional Y axis for milling, drilling, and other cutting tool rotating operation types. A part with milling and turning features is called mill-turn part. The results of feature recognition for mill-turn parts would be regarded as input information of process planning. Feature recognition is considered as a completed job based on the work of Li and Shah.1

The distinguishing features of turning machine tool are the two different machining modes: parallel machining mode and simultaneous machining mode. In the parallel machining mode, two PTs could perform different features, respectively. It could enhance efficiency and reduce machining cost. The simultaneous machining mode occurs when two PTs perform the same feature cooperatively. When the simultaneous machining mode is applied, as shown in Figure 2, the two cutting forces caused by PTs have opposite directions. The opposite cutting forces would lead to a balance between them, and reduce vibration during machining. Hence, turning machine tool has significant improvements in surface roughness and machining accuracy. In practice, the parallel machining mode and simultaneous machining mode would be applied together in one process plan.

When process planning and scheduling are performed concurrently, the parameters that require optimizing includes selection of machining operation types, cutting tools, operation sequencing, as well as scheduling operations on PTs and PHLs. A mixed 0-1 integer programming (MIP) model is the commonly used mathematical tool to formulate this problem, and some assumptions are given as follows:

1. Operation cannot be interrupted after it begins.

2. The number of operations in parallel cannot exceed the number of PTs and PHLs.

3. Operations of a part using different volume

removing methods could not be performed


Milling and tuning operations cannot be performed simultaneously. Turning operation means that a work-piece turns about a horizontal axis against a fixed cutting tool, while milling operation means that a fixed workpiece is shaped by a rotating tool. They are totally different machining methods and would interrupt each other when they are being performed on a workpiece simultaneously. Based on the assumptions mentioned above, a MIP model is developed to formulate the process planning problem on turning machine tool. Some decision variables for the MIP model are defined as follows:


C = {C1, C2, ..., C}: setofPHLs, i = 1,2,., / P = {P1, P2, ., Pj}: set ofPTs, j = 1,2,., J T ={T1, T2, ., TN}: set of cutting tools, n = 1,2,., N

O = {O1, 02, ., Ok}: set of operations, k = 1, 2, ., K

Oj: the operation Oj performed on C, by PTj

Oj. Tool: the selected cutting tool for Oj

Oj. PHL: the selected PHL (C,) for Oj

ij!: the machining time of operation Oj using cutting

tool Tn

SCI: the PHL (C,) changing time

TCI: the cutting tool changing time

sj the starting time of Oj; sj = + n if it does not

fjj: the finishing time of Oj; f =0 if it has not been completed

[sjfl where sj 'f: the machining period of Oj; it means nothing if sj > fjj

Oj[sjf]: the state of Ojj; when Oj is been performed on C, by PTj in time period [sj,f], Ojj[sj,f] equals 1, otherwise Oj[sj,f7] equals 0

Pjsjfj]: the state of PTf, it equals 1, when PTj is busy, and it equals 0 when PTj is idle Cisjff]: the state of C,; it equals 1, when C, is busy, and it equals 0 when PHL C, is idle

With the above decision variables, the MIP model for mill-turn machine with / PHLs and J PTs is formulated as follows.

Objective function

minfj = mini ^ (TC/(Oj.TWIlOj_j.TW)

\k = 2

+ SC/ (Oj.PHL||Oj_1.PHL)) + ¿tjOjlsj,./^]))

k = 1 /

Figure 2. Simultaneous machining mode.

where i = 1,2, ..., I; j = 1,2, ..., J; and k = 1,2, ..., K. f£ is the finishing time of operation Ok on PHL Ci, and fl is the finishing time of a workpiece. tl is the machining time of operation Ok. An operation may have different machining time when it is performed by different cutting tools. The value of tk could refer to a cost matrix such as Table 3.


Parts should be machined following an operation sequence that is generated based on precedence constraints. Constraints are derived from manufacturing principles and good manufacturing practice, and ensure that the tolerance requirement and machining accuracy could be achieved

EcKf^i (3)

jf] = Pj\4fij]Ci[4,fj] (4)

EEoM,ff]<J (5)

j = 1 k = 1

oMJi] + Oj\sj ,ff]< 1 (6)

where Ok 2 M, 4 2 L, 4 <4 f, or 4' <4 f M

means machining method set including milling and drilling, while L means turning

4 =fl 1 + TCI (Oj.Tool\\Oj_1.TooD

+ SCI (Oj.PHL\\Ok_1 .PHL), 1 < k < K

fj = 4 + tkjkfj (8)

Equation (1) is the objective function, which is to minimize the total completion time of a part. Constraint (2) means each PT could perform one operation at most. Constraint (3) means that each PHL could hold one part at most. Constraint (4) means that an operation could start only if there are any available PTs. Constraint (5) is the constraint for the parallel machining mode. It means that the number of operations in parallel could not exceed the number of PTs. Constraint (6) means no multiple operations could be performed simultaneously by the same PT. Constraint (7) illustrates the setup changing cost and cutting tool changing cost. Constraint (8) means the completion time of an operation is the sum of its starting time and the machining time.

A HGA approach

GA is a kind of heuristic algorithm, and it imitates the biological evolution process and drives its power from principles of natural selection and survival of fittest. Although there are many variants of GAs, they typically consist of five steps: encoding, initialization, evaluation (using fitness function), selection, and reproduction (consist of crossover and mutation). A local search approach is added to optimize the assignments of PHLs and PTs, as well as the selection cutting tools.

The procedure of HGA is illustrated in Figure 3. A judiciously designed encoding strategy helps transcoding candidate solutions (process plans) to chromosomes. A well-designed representation strategy could improve the performance of GA and be computationally inexpensive.28 Initialization produces the first generation based on encoding strategy. The evaluation procedure calculates fitness values of all the chromosomes using fitness function based on the mathematical model given in section ''Problem formulation." HGA weeds out bad individuals by using a properly designed selection strategy, so that HGA trends to good solutions. The reproduction consists of crossover and mutation, which keep the searching moving on.29 In addition, the local search approach is used to help GA optimizing the selection of PHLs, PTs, and cutting tools. These procedures of HGA will not stop until

termination condition occurs. If HGA could not get any better solution after numerous iterations, then the current optimal solution is regarded as the global optimum of HGA.

Encoding and initialization

An individual in a population is named chromosome, and chromosome evolves via successive iterations called generations. The process plans would be encoded to chromosome following precedence constraints. Precedence constraints are represented as a directed acyclic OPG.26 The OPG of a given part consists of

two sets: a vertex (operation) set V and an edge (precedence constraint) set E. Each edge is defined by an ordered pair of vertices. The OPG is one of the main inputs of process planning, and a sample OPG is given in Figure 4.

A new encoding strategy is proposed to handle the precedence constraints. The new strategy focuses on two aspects: (1) chromosomes and genes contain all manufacturing information of each operation and (2) make sure that all the precedence constraints could be obtained. In order to illustrate the new encoding strategy, an example is shown in Figure 5. The data structure of genes and chromosome is illustrated in pseudo code as follows.

• Structure of chromosome.

• struct chromosome{

• string ID;//chromosome ID

• struct_gene[] genes;//genes of chromosome, regarded as the operations sequence of Ok

• double makespan;//totally complete time, regarded as max fjij

• double start_time[];// start time sj of each operation in sequence

• double finish_time[];// finish timefkj of each operation in sequence

Stage 1 Instert the first edge

Stage 2 Instert all the left edges

Stage 3 Instert all the left vertices

(1) Randomly select edge eis from the OPG

(2) Remove edge eis from the OPG

(3) Remove the endpoint vi, vs and all edges connected with them

(4) Insert Vertices V1 and V5 into the sequence

(1) Randomly select edge e23 from the OPG

(2) Remove edge e23 from the OPG

(3) Remove the endpoints V2, vi and all edges connected with them

(4) Randomly select one of the endpoints from the selected edge

(5) Insert the selected Vertex V2 into one of the candidate positions by random

(6) Insert the left vertex v? into the sequence follow the precedence constraints

(7) Procedure is repeated until no edges are left in the OPG

(1) If no edges left in the OPG, then select one of the vertex by random

(2)Insert the selected vertex V4 into one of the candidate positions by random

(3) Remove the selected vertex

(4) Insert the selected vertex V6 into one of the candidate positions by random

(5) Insert the selected vertex V6 into one of the candidate positions by random

(6) Procedure is repeated until no vertices are left in the OPG

Figure 5. Encoding strategy.

Structure of gene. struct gene{

string gene_id;//operation ID Ok

string[] tool_set;//a set of candidate cutting tools

for Ok

string[] PHL_set;//a set of candidate PHLs for

string[] method_set;//a set of candidate machining methods for Ok

bool machining mode_flag; //true or false, means whether the operation could be performed by two PTs simultanesouly string PT;//selected PT, regarded as j in OlJk string PHL;// selected PHL, regarded as i in Ok string machining_method; //selected machining method of Ok, which consist of two types, Ok 2 M or Ol 2 L

string tool;//selected tool of Ok, regarded as Of.Tool

double machining_time;//machining time tl for

Evaluation and selection

In the evaluation procedure, the fitness function plays a key role in GA. A well-designed fitness function can help finding the optimal solution and reducing the computational time. The fitness function is used to decide whether a chromosome could be passed to the next generation. It usually originates from the object function of the mathematical model. We use equation (1) as the fitness function in this article.

The object of selection is to choose two chromosomes as parents for crossover. There are some common selection methods such as roulette wheel selection, stochastic universal sampling, local selection, truncation selection, and tournament selection. This article uses tournament selection that runs several tournaments among a few chromosomes chosen at random, since tournament selection is efficient to code, works on parallel architectures, and allows the selection pressure to be easily adjusted. The one with the best fitness is the winner of a tournament and would be selected for crossover. Table 1 illustrates that the selection pressure is determined by the tournament size. If the tournament size is larger, chromosomes with worse fitness have a smaller chance to be selected.


The genetic operators consist of crossover and mutation that ensures the offspring inherit some characteristics from their parents.13 The average fitness will increase, since the chromosomes with better fitness values have higher probabilities and are more likely to participate in the crossover and mutation procedure. The traditional crossover method is designed based on a binary encoding strategy that directly combines two segments of genes from parents to generate a new one. It would result in precedence constraints violation, not to mention reduplication of genes in one chromosome.30 To avoid such a problem, order crossover (OX) is used in this article. Figure 6 illustrates an example of OX.


Mutation promotes the act of search as well as crossover but much more indistinctively. Mutation acts on the current population that has just been produced by crossover. Chromosomes are selected with a mutation ratio that is designed by experience. Then a fragment of genes is chosen at random and replaced by a new order of these genes. In order to follow the precedence constraints, the alternative fragment of genes is generated based on encoding strategy. An instance of mutation is given in Figure 7.

Local search approach

To implement GA in scheduling, a local search approach is utilized to optimize every newly generated

Figure 6. Order crossover.

Table 1. Parameters of tournament selection.

Tournament size 1 2 3 5 10 30

Selection pressure 0 0.56 0.85 1.15 1.53 2.04

Figure 7. Mutation.

chromosome. The local search approach is used to move the solution to a local optimum through reassign-ments of PHLs, PTs, and cutting tools. The reassign-ments are based on greedy rule that means the manufacturing parameters of each operation are selected as long as it reduces the changing time of PHLs, PTs, and cutting tools. The local search approach aims at improving the performance of GA. The pseudo code of optimizing PTs by the local search approach is given as follows and optimizations of the other manufacturing parameters are similar:

• Local search approach for optimizing the assignments of PTs.

• Begin

• For i = 2 to K do

• If the PT selected by Ok-1 has available time Pjsjfj] == 0 for operation Ok

• Then Ok would use the same PT as Ok-1

• Recalculate the start time and finish time of Ok and Ok-1

• Set Pj{sjfj]=1

• Update the states of the other PTs

• Else the PT selected by Ok-1 does not have available time Pjsj'fj7] = = 1

• If exists available PT for Ok

• Then Ok would use the same PTl

• Set Pl[sf]=1

• Update the states of the other PTs

• Else none of PTs are available for Ok

• Find the PT that finish its job firstly and use this PT for Ok

• Recalculate the start time and finish time of Ok + 1. .to Ok

• Update the states of the PTs

• End if

• End if

• End for

• End

Figure 8. Example part.

Case study

Case description

To illustrate the performance of HGA, a test part is given in Figure 8. This part has 33 features and 43 operations, whose detailed information is illustrated in Table 2. This part is a typical mill-turn part. Some features could only be machined by rotating workpiece against a fixed cutting tool on the lathe machine tool, while some features could only be machined by rotating cutting tools. The features that could either be lathed or milled are defined as mill-turn features.

The details of operations are shown in Table 2, such as operation IDs, related feature information, candidate machining methods, and so on. All these information would be encoded based on the approach proposed in section "Encoding and initialization.''

The machining time of each operation is listed in Table 3, and an operation has different machining time when it is performed by different cutting tools. In Table 3, 24 cutting tools including 18 different types are equipped on PT1 and PT2. Each PT could use four turning tools and eight milling tools at most. The types of cutting tools are listed as follows. T1 is an external turning tool. T2 is a cylinder end-face turning tool. T3 is a flat internal cylinder turning tool. T4 is a facing tool. T5 is a flat milling tool. T6 is a ball-end milling tool. The cutting tools mentioned above are equipped on both PTs for simultaneous machining. Drilling tools include T7, T10, T13, T16, and T17. Reaming tools include T8 and T14. Boring tools include T9 and T15. The tapping tool is T12. End milling tools include T12 and T18. Some operations such as O16, O18, O20, and O22 could only be performed by one cutting tool, while some operations such as O5, O6, O24, O25, and so on

Table 2. Operations list.

Feature Feature Operation Machining operation type (volume removing method) Setup

description ID -

Workpiece Cutting tool

rotating rotating

Fl Cylinder 1 Ol Lathing PHLl phl2

F2 Cylinder 2 O2 Lathing PHLl phl2

F3 Cylinder 3 O3 Lathing PHLl

F4 Cylinder 4 O4 Lathing phl2

F5 End face 1 O5 Lathing Milling PHLl

F6 End face 2 O6 Lathing Milling phl2

F7 Hole 1 07 08 09 Lathing Drilling Reaming Boring PHLl PHLl PHLl

F8 hole 2 010 011 Drilling Drilling PHLl PHLl

F9 Hole 3 Ol2 013 014 Lathing Drilling Reaming Boring phl2 phl2 phl2

Fl0 Hole 4 015 016 Drilling Tapping PHLl PHLl phl2 phl2

Fll Hole 5 017 018 Drilling Tapping PHLl PHLl phl2 phl2

Fl2 Hole 6 019 020 Drilling Tapping PHLl PHLl phl2 phl2

Fl3 Hole 7 021 022 Drilling Tapping PHLl PHLl phl2 phl2

Fl4 Cavity 1 O23 M lling PHLl

Fl5 Plane 1 O24 Lathing M lling phl2 phl2

Fl6 Plane 2 O25 M lling PHLl phl2

Fl7 Plane 3 O26 Lathing M lling PHLl phl2

Fl8 Plane 4 O27 M lling PHLl phl2

Fl9 Plane 5 O28 M lling PHLl phl2

F20 Plane 6 O29 M lling PHLl phl2

F2l Plane 7 O30 M lling PHLl phl2

F22 Plane 8 O3l M lling PHLl phl2

F23 Plane 9 O32 M lling PHLl phl2

F24 Plane 10 O33 M lling PHLl phl2

F25 Plane 11 O34 M lling PHLl phl2

F26 Chamfer 1 O35 M lling PHLl phl2

F27 Chamfer 2 O36 M lling PHLl phl2

F28 Chamfer 3 O37 M lling PHLl phl2

F29 Chamfer 4 O38 M lling PHLl phl2

F30 Arc surface 1 O39 M lling PHLl phl2

F3l Arc surface 2 O40 M lling PHLl phl2

F32 Arc surface 3 O4l M lling PHLl phl2

F33 Arc surface 4 O42 M lling PHLl phl2

PHL: part holding location.

could be performed by more than one cutting tools. The preparing time is determined by the machine tool types. In this case, PHL changing time SCI is set to 50, and cutting tool changing time is set to 5.

Before process planning, precedence constraints that usually be represented as a directed OPG would be generated as input. Figure 9 illustrates the OPG of the test part in Figure 8. There are only two operation types, one could be performed in simultaneous machining mode and the others could not. The parallel machining mode would occur naturally when PTs perform two different operations. In this case, the operations O1,

O2, O3, and O4 (derived from external cylinder feature F1, F2, F3, and F4) belong to simultaneous machining

Determination of parameters for HGA

The proposed HGA is programmed using MATLAB, and run on IBM-comfortable PC (OS is windows 7) with 2.2-GHz dual-core processor and 8 GB RAM. The main parameters include the number of population POP, the crossover probability Pc, the mutation probability Pm, and the tournament size TOUR.

Figure 9. The OPG of the test part.

Figure 10. Parameter determination for HGA: (a) results with different POP, (b) results with different Pc and Pm, and (c) results with different TOUR.

Experiments in Figure 10 are used to determine these parameters. After tests in Figure 10(a), the population size POP is set to 150. Figure 10(b) illustrates that HGA could achieve a better performance when Pc = 0.8 and Pm = 0.1. Referring to TOUR, Figure 10(c) shows that TOUR = 2 is better than others.

Computational results

The convergence curves of HGA in Figure 11 validate the feasibility and the effectiveness. The precedence

constraints were handled efficiently, and no single one was violated. Figure 11 illustrates that HGA with local search approach could generate better solutions than GA without local search approach. According to the convergence curves, HGA has better robustness as well.

One of the generated optimal process plans for the turning machine tool is shown in Figure 12. The optimal process plan has lots of operations in parallel machining mode, which could increase the manufacturing efficiency significantly and reduce total machining time. Using traditional machine tools, the total machining time would be doubled at least. Because using

Figure 11. Comparison of GA and HGA.



PT1 PT2 - Both]-

Q Operation could only be performed by one PT O EH Operations are performed by turning

□ Operation could be performed by two PTs simultaneously O D Operations are performed by nrtirtmg crtting tools

PHL changing

Figure 12. Optimal process plan.

traditional machine tools requires much more machine changes and setups. Furthermore, the utilization of the parallel machining mode could promote the machining efficiency as well as the simultaneous machining mode.

The solution in Figure 12 is an applicable process plan. The operations O1, O2, O3, and O4 could be machining by two PTs simultaneously. However, O1 did not use simultaneous machining mode, because it could save more machining time when the operations O1 and O7 are performed using parallel machining mode. When workpiece is held on PHL2, PT2 perform much more operations than PT1, because the operations O12, O13, and O14 are time-consuming jobs and take almost as much time as 18 operations performed by PT2. In the optimal process plan, only one setup changing is required, which reduces the preparing time as well.

To do some further study, it is assumed that two PHLs are equipped with different cutting tools. In that case, the simultaneous machining mode would not be

used anymore. Based on the above assumptions, all cutting tools of PT1 in Table 3 are available, while cutting tools T1, T2, T3, T4, T5, and T6 are removed from PT2 to ensure that no same cutting tools are equipped. Taking the test part in Figure 8 for example, the optimal result is illustrated in Figure 13 after calculation. Compared with the result in Figure 12, PT2 spends much more time waiting or being idle when it could not assist in machining. In conclusion, the closer the machining capabilities of two PTs are, the more machining cost could be saved. It is reasonable that two PTs should be equipped with same cutting tools as much as possible.

If there are no restrictions on tool magazine capacity, both PTs could use any cutting tools in Table 3. In that case, the turning machine tool could provide the maximum machining capability. In addition, the machining efficiency could also be affected by the geometric design of the part. The test part in Figure 8

Table 3. Cost matrix.

Operation ID PTI PT2

28 20 25 30

22 66 200 22 66 200 20 60 180 20 60 180

30 30 45

35 50 35

5 15 45 5 15 45

50 150 50 150

10 25 75 10 10 25 75

12 36 12 36

12 36 12 36

12 36 12 36

12 36 12 36

15 45 15 45

15 45 15 45

15 45 15 45

15 45 15 45

5 15 5 15

5 15 5 15

5 15 5 15

5 15 5 15

60 60 60 60 60 60 60 60

PT: power turret.

Figure 13. Optimal process plan without using simultaneous machining mode.

consists of many symmetric geometric features, so that the parallel machining mode could easily be engaged.

Conclusion and future work

The turning machine tool increases the efficiency of metal removing and manufacturing accuracy significantly. In order to make full use of turning machine tool, a HGA was developed for process planning integrated with scheduling, and the experiment validates the effectiveness and efficiency. Conclusions are summarized as follows:

1. The new encoding strategy could handle the complicated precedence constraints efficiently and make sure that none of the precedence constraints are violated.

2. The local search approach cooperates with GA and improves the performance and robustness of GA. HGA is proved to be better than GA without local search approach by experiments.

There are some unsolved problems, and the future work will focus on (1) manufacturing feature recognition and representation for mill-turn part, (2) approaches of representing, managing, and utilizing human expertise to implement precedence constraints; and (3) some other heuristic algorithms or approaches would be compared with HGA.

Declaration of conflicting interests

The authors declare that there is no conflict of interests regarding the publication of this article.


This article is supported by the National Natural Science Foundation, China (No. 51475290, 51075261), Research Fund for the Doctoral Program of Higher Education of

China (No. 20120073110096) Shanghai Science and

Technology Innovation Action Plan (No.11DZ1120800).


1. Li S and Shah JJ. Recognition of user-defined turning features for mill/turn parts. J Comput Inf Sci Eng 2007; 7: 225.

2. Moriwaki T. Multi-functional machine tool. CIRP Ann ManufTechnol 2008; 57: 736-749.

3. Levin JB and Dutta D. PMPS: a prototype CAPP system for parallel machining. J Manuf Sci E Trans ASME 1996; 118: 406-414.

4. Waiyagan K and Bohez ELJ. Intelligent feature based process planning for five-axis mill-turn parts. Comput Ind 2009; 60: 296-316.

5. Kim YS, Kim Y, Pariente F, et al. Geometric reasoning for mill-turn machining process planning. Comput Ind Eng 1997; 33: 501-504.

6. Chiu N-C, Fang S-C and Lee Y-S. Sequencing parallel machining operations by genetic algorithms. Comput Ind Eng 1999; 36: 259-280.

7. Norman BA and Bean JC. Scheduling operations on parallel machine tools. IIE Trans 2000; 32: 449-460.

8. Chung DH and Suh SH. ISO 14649-based nonlinear process planning implementation for complex machining. Comput Aided Des 2008; 40: 521-536.

9. Nallakumarasamy G, Srinivasan PSS, Venkatesh Raja K, et al. Optimization of operation sequencing in CAPP using simulated annealing technique (SAT). Int J Adv ManufTechnol 2011; 54: 721-728.

10. Li WD, Ong SK and Nee AYC. Hybrid genetic algorithm and simulated annealing approach for the optimization of process plans for prismatic parts. Int J Prod Res 2002; 40: 1899-1922.

11. Yun YS, Chung HS and Moon C. Hybrid genetic algorithm approach for precedence-constrained sequencing problem. Comput Ind Eng 2013; 65: 137-147.

12. Yun Y and Moon C. Genetic algorithm approach for precedence-constrained sequencing problems. J Intell Manuf 2011; 22(3): 379-388.

13. Moon C, Kim J, Choi G, et al. An efficient genetic algorithm for the traveling salesman problem with precedence constraints. Eur J Oper Res 2002; 140: 606-617.

14. Saygin C. Integrating flexible process plans with scheduling in flexible manufacturing systems. Int J Adv Manuf Technol 1999; 15: 268-280.

15. Zhang YF, Saravanan AN and Fuh JYH. Integration of process planning and scheduling by exploring the flexibility of process planning. Int J Prod Res 2003; 41: 611-628.

16. Zhang XD and Yan HS. Integrated optimization of production planning and scheduling for a kind of job-shop. Int J Adv Manuf Technol 2005; 26: 876-886.

17. Mohammadi G, Karampourhaghghi A and Samaei F. A multi-objective optimisation model to integrating flexible process planning and scheduling based on hybrid multi-objective simulated annealing. Int J Prod Res 2011; 50: 5063-5076.

18. Lee DH, Kiritsis D and Xirouchakis P. Search heuristics for operation sequencing in process planning. Int J Prod Res 2001; 39(16): 3771-3788.

19. Li Y and Gong SH. Dynamic ant colony optimisation for TSP. Int J Adv Manuf Technol 2003; 22: 528-533.

20. Salehi M and Bahreininejad A. Optimization process planning using hybrid genetic algorithm and intelligent search for job shop machining. J Intell Manuf 2011; 22: 643-652.

21. Xu X, Wang L and Newman ST. Computer-aided process planning—a critical review of recent developments and future trends. Int J Comp Integr Manuf 2010; 24: 1-31.

22. Yusof Y and Latif K. Survey on computer-aided process planning. Int J Adv Manuf Technol 2014; 75: 77-89.

23. Cai N, Wang L and Feng HY. GA-based adaptive setup planning toward process planning and scheduling integration. Int J Prod Res 2009; 47: 2745-2766.

24. Chaube A, Benyoucef L and Tiwari MK. An adapted NSGA-2 algorithm based dynamic process plan generation for a reconfigurable manufacturing system. J Intell Manuf 2012; 23: 1141-1155.

25. Yip-Hoi D and Dutta D. A genetic algorithm application for sequencing operations in process planning for parallel machining. IIE Trans 1996; 28: 55-68.

26. Sun X, Chu X, Su Y, et al. A new directed graph approach for automated setup planning in CAPP. Int J Prod Res 2010; 48: 6583-6612.

27. Tseng YJ and Liu CC. Concurrent analysis of machining sequences and fixturing set-ups for minimizing set-up changes for machining mill-turn parts. Int J Prod Res 2001; 39: 4197-4214.

28. Chen C-F, Wu M-C, Li Y-H, et al. A comparison of two chromosome representation schemes used in solving a family-based scheduling problem. Robot Cim Int Manuf 2013;29:21-30.

29. Novkovic S and Sverko D. The minimal deceptive problem revisited: the role of "genetic waste.'' Comput Oper Res 1998; 25:895-911.

30. Qiao L, Wang XY and Wang SC. A GA-based approach to machining operation sequencing for prismatic parts. Int J Prod Res 2000; 38: 3283-3303.