dimensional quality optimisation of high-speed cnc milling process with dynamic quality...

Upload: robson-bruno-dutra-pereira

Post on 02-Apr-2018

216 views

Category:

Documents


0 download

TRANSCRIPT

  • 7/27/2019 Dimensional Quality Optimisation of High-speed CNC Milling Process With Dynamic Quality Characteristic

    1/12

  • 7/27/2019 Dimensional Quality Optimisation of High-speed CNC Milling Process With Dynamic Quality Characteristic

    2/12

    greater potential in the dry turning and milling of

    ALSi10MG casting alloy compared with PCD and K10.

    Vieira et al. [11] investigated the performance of cutting

    fluids during HS face milling of steels. They reported

    that semi-synthetic cutting fluid exhibited the best

    cooling effects in the HS machining process, followed

    by the emulsion-based mineral fluid, and the 10%concentration of synthetic fluids. The converse observa-

    tion was made on the power consumption. Avila et al.

    [12] investigated the effects of the HS cutting fluids on

    turning hardened AISI 4340 steel using mixed alumina

    (Al2O3+TiC). They found that appropriate use of

    emulsion without mineral oil led to better tool life,

    surface finish and chip control. Su et al. [1315] studied

    the effects of the coated materials on HS milling process

    using Taguchi static approach. Their results showed that

    multi-layer coating construction has better performance

    on the cutting tool wear than single layer at the same

    thickness basis.

    The research for investigating factors effects of HS

    cutting process and improving its performance have

    continuously received much attention worldwide. Of the

    various study methodologies, Taguchi method is the

    preferred approach to undertake the present study as it

    has been proved to be an effective and efficient

    approach. However, most past Taguchi experiments

    concentrated on the optimisation of static or single

    quality characteristic, indicating that only a fixed output

    target instead of a wide range is developed. A new

    parameter design has to be developed whenever a new

    process/product is planned, leading to a waste of time

    and money.Nowadays the international market in the mold and

    die manufacture industry becomes increasingly compe-

    titive. To meet customer requirements for short delivery,

    high quality, and low cost, the high-speed CNC milling

    process must have the capability of versatility, flexibility,

    and robustness. Taguchi dynamic approach can be far

    more powerful in developing a robust process design

    with dynamic quality characteristic. Hence, the main

    objective of this study is an attempt to apply Taguchi

    dynamic approach to optimise the dimensional quality

    of high-speed CNC milling process with dynamic quality

    characteristic.

    2. Equipments and materials

    Feeler QM-22 CNC 3-axis milling machine with the

    maximum spindle revolution 30000 rpm was used

    throughout the experiments coupled with a c 6 mm

    end-milling cutter. Mitutoyo MF Series toolmakers

    microscope and SJ-301 surface profiler were used to

    measure the dimension and surface roughness of the

    machined products, respectively. Tool steels SKD61 and

    SKD11 were selected as the materials with the chemical

    compositions as displayed in Table 1.

    3. Experimental

    3.1. Engineered system and Taguchi robust design

    As shown in Fig. 1, any man-made machine or

    equipment is regarded by Taguchi methods as an

    engineered system with a specific function, requested

    by customers. The system consists mainly of four

    components including control, noise, signal factors,

    and output response. A control factor is a factor that

    can be selected and fixed to a certain level after

    parameter design. However, a noise factor is a factor

    that cannot be controlled, due to either practical or

    economical reasons. Noise entering a system may take

    many forms, and is a true disturbance to the engineered

    system. Signal factor is a factor to change the output

    response which is what the system is designed toproduce. When the output response of an engineered

    system can dynamically vary with the input signal to

    generate a range of responses, it has the so-called

    dynamic quality characteristic [17].

    Engineered system starts to use energy transformation

    to carry out its function when the input signal is

    received. The process of energy transformation is

    governed by the control factors to convert input energy

    into intended output energy by using laws of physics.

    The engineered system reaches its ideal function when

    all of its applied energy is transformed efficiently into

    creating desired output energy. However, noise factorsusually cause variability in the energy transformations

    leading to unintended outputs.

    In every engineered system, there exists some form of

    ideal relationship between its input signal and output

    response. However, correct identification of an ideal

    function is not easy, and ideal function may differ from

    case to case. One of the most common ways of

    expressing a designs ideal function is

    Y fM bM, (1)

    where a linear relationship with a slope b exists between

    Y (=ideal output response) and M (=input signal) as

    ARTICLE IN PRESS

    Table 1

    The chemical compositions of tool steel [16]

    Material Chemical composition (%)

    C Si Mn P S Cr Mo V Cu

    SKD11 1.5 0.4 0.6 0.03 0.03 12 1.0 0.35 0.25

    SKD61 0.37 1.0 0.5 0.03 0.03 5.0 1.25 1.0 0.25

    T. Yih-fong, J. Ming-der / Robotics and Computer-Integrated Manufacturing 21 (2005) 506517 507

  • 7/27/2019 Dimensional Quality Optimisation of High-speed CNC Milling Process With Dynamic Quality Characteristic

    3/12

    shown in Fig. 2(a). However, in reality, energy

    transformation of any system does not happen as

    designed or intended, because there might be noise

    factors disturbing the system. The reality of the system

    function consists of nonlinear effects between the input

    and output that can be mathematically described as

    follows:

    Yr fM; C1; C2; C3;. . .

    ; Cm; N1; N2; N3;. . .

    ; Nn bMferrorM; C1; C2; C3; . . . ; Cm; N1; N2; N3; . . . ; Nn

    YferrorM; C1; C2; C3; . . . ; Cm; N1; N2; N3; . . . ; 2Nn,

    2

    where Yr is the real output response, C1; C2; C3; . . . ; Cmthe control factors, N1; N2; N3; . . . ; Nn the noise factors,and ferror the difference between the ideal function

    and the reality. The real function is demonstrated by

    Fig. 2(b).

    Orthogonal array is well known as one of the most

    important tools used in Taguchi robust design because it

    provides a fair comparison of any factor. This enablessufficient information to be collected by conducting only

    partial series of experiments. Taguchi parameter design

    strategy separates the control factors from both the

    noises and the signals by using inner and outer arrays,

    respectively. Noise factors coupled with signal factor are

    assigned to the outer array for exposing the process to

    varying noise conditions. An experiment is conducted

    for all combinations between the inner and the outer

    arrays. The tendencies of control factors, and how these

    may affect robustness are evaluated. The best combina-

    tion of control factor levels is therefore sought so that

    the system becomes most insensitive to noise factors.

    When such a technology is developed, it has a robust

    function [17].

    Taguchi robust design seeks to attain the ideal state of

    an engineered system, referred to as the designs ideal

    function. Proper level settings of control factors will not

    only make the design robust against noise factors, but

    also can adjust the output response to the desired target.Therefore, ferror stably gets close to its minimum, and

    then Yr approaches to Y bM:

    3.2. Proposed ideal function model for CNC milling

    system

    Fig. 3 is the application of Taguchis concept about an

    engineering system in the high-speed CNC machining

    process. The high-speed CNC milling system is essen-

    tially designed to perform material removal for produ-

    cing high-dimensional quality of products requested by

    the customers. The input signal, reflecting the intent of

    the customers, to the CNC milling system represents

    the factor informing system of exactly what job to do. It

    can be perceived as the programmed geometrical

    dimension of product to be machined for driving

    the machining function to transform the solid model

    on the PC to the work piece. If there is no energy loss

    due to noise factors to create symptoms of poor

    function, such as tool wear, vibration, noise, lubricant

    aging, etc., the geometrical shape of solid model on the

    PC will be accurately duplicated on the work piece.

    Hence, the ideal function of the CNC milling systemcan be defined in terms of transformability. The ideal

    function Y bM is modelled as Y being the desired

    product dimension, M the programmed dimension,

    and b 1:

    3.3. Robustness evaluation using signal/noise ratio

    The signal-to-noise (S/N) ratio originated in the

    communication field. Taguchi methods expanded its

    function into quality engineering area including staticand dynamic characteristics. As for the dynamic-type

    S/N ratio, it is written based on the ideal function of a

    product or process that is related with input/output

    energy transformation. For any engineered system, as

    the input signal, control factors, and noise factors come

    together to perform its designed function, their com-

    bined impacts on the output response can be evaluated

    using the dynamic-type S/Nratio to measure the quality

    of energy transformation. As shown in Fig. 3, the output

    response of high-speed CNC milling system consists

    of the so-called useful and harmful outputs. For a

    dynamic quality characteristic application, the concept

    ARTICLE IN PRESS

    Control Factors

    Engineered

    SystemsOutput Response

    Noise Factors

    Signal Factors

    Fig. 1. An engineered system model used by Taguchi methods.

    Fig. 2. (a) The systems ideal function and (b) its reality.

    T. Yih-fong, J. Ming-der / Robotics and Computer-Integrated Manufacturing 21 (2005) 506517508

  • 7/27/2019 Dimensional Quality Optimisation of High-speed CNC Milling Process With Dynamic Quality Characteristic

    4/12

    of S/N ratio is [17]

    Z Useful output

    Harmful output

    10 logLinear relationship between M and Y

    Variability around the linear relationship

    10 logb2

    s2 , 3

    where b is the slope of best-fit line between the measured

    values and the inputs, and s2 the variability around the

    best-fit line.

    The S/N ratio measures the level of system perfor-

    mance and the effects of noise factors on performance so

    that it can be employed as an indicator of the

    consistency of a given system. The higher this ratio,

    the more the system is doing what is intended to do

    regardless of noise factors and the system is more robust

    against noise.

    3.4. Control factors and their levels

    An L18 is selected for arranging the overall experi-

    mental tests. Eight dominant process conditions

    of the high-speed CNC milling process are identi-

    fied as the control factors, which are listed in Table 2

    together with their alternative levels. It is noted

    that most of the factors has three levels except factor

    A, which has two. All levels are selected for proper

    reasons.

    3.5. Noise factors

    Generally, there are a lot of noise factors associated

    with the high-speed CNC milling process, such as,

    material lot, machine setting variability, material hard-

    ness variability, lubricant condition, tool wear, and

    manufacturing variability of milling machine, etc. Due

    to hard control, Taguchi methods suggested the use of

    the compounding strategy to arrange them to be two

    extreme conditions. For the simplification of experi-

    mentation, only material hardness variability is chosen

    as the noise factor. The identified two noise conditions

    of N1 and N2 are displayed in Table 3. It is designed so

    ARTICLE IN PRESS

    Fig. 3. The schematic of an engineering system as a high-speed CNC milling system.

    Table 2

    Control factors and their levels

    Control

    factors

    Level

    1 2 3

    A Milling type Downward

    mill

    Upward

    mill

    B Cutting speed

    (m/min)

    150 225 300

    C Feed per tooth

    (mm/tooth)

    0.03 0.04 0.05

    D Film material TiCN TiN TiAlN

    E Tool material K10 (Co 8%) K20 (Co

    10%)

    K30 (Co

    12%)

    F Number of tooth 2 3 4

    G Rake angle 41 71 101

    H Helix angle 301 351 451

    T. Yih-fong, J. Ming-der / Robotics and Computer-Integrated Manufacturing 21 (2005) 506517 509

  • 7/27/2019 Dimensional Quality Optimisation of High-speed CNC Milling Process With Dynamic Quality Characteristic

    5/12

    because that from the standpoint of energy the softer

    material is easier to be machined, and vice versa.

    3.6. Test piece and signal arrangements

    How to vary the signals experimentally is also one of

    the most important steps in the Taguchi robust design.Since some of the research objectives are to develop

    versatility and flexibility into the high-speed CNC mill-

    ing process, the optimised technology development must

    be applicable to a family of products and future pro-

    ducts. Therefore, it is most effective to specify the signal

    levels so that the range covers all usage conditions.

    In view of this goal, test piece with a range of

    geometrical characteristics is designed for the study. Allof the test workpieces is finish machined with a constant

    depth of cut of 0.1 mm. The test piece is easier to

    measure dimensions and would allow them to evaluate

    multiple levels of the signal easily. Fig. 4 illustrates

    the test piece we designed for the study. As shown in

    Fig. 4(a), the test piece has three typical of geometrical

    characteristics to be machined, including rectangular,

    circle, and triangle on each of three layers with a volume

    ratio of 1:2:3.

    ARTICLE IN PRESS

    Table 3

    Noise factors arranged as two extreme conditions

    Noise factors Material Hardness (HRC)

    N1 (positive side extreme condition) SKD11 2123

    N2 (negative side extreme condition) SKD61 1719

    Fig. 4. The designed test piece for experimentations: (a) solid modeling, (b) top view with the programmed vertices, diameters, and circle center, (c)

    side view.

    T. Yih-fong, J. Ming-der / Robotics and Computer-Integrated Manufacturing 21 (2005) 506517510

  • 7/27/2019 Dimensional Quality Optimisation of High-speed CNC Milling Process With Dynamic Quality Characteristic

    6/12

    As illustrated in Fig. 4(b), the zero point is fixed on

    the (0,0) point on the southwest corner. All locations on

    the rectangular and triangles are defined by positive X

    and Y coordinates relative to the fixed origin. The

    diameters of three circles are defined in the same

    fashion, but in relation to their fixed center point (Pi;i=2527). The programmed dimensions for positioning

    the various vertices, diameters, and the common circle

    center are listed in Table 4. There are a total of 51 of the

    programmed dimensions used as the input signals in an

    incremental way.

    4. Experimental results and analysis

    4.1. General effects of control factors

    Fig. 5 plots the entire experimental results as the

    input/output relationship that shows slight scattering

    phenomena due to noise effects. Table 5 displays a

    complete experimental layout, resultant data, and

    performance evaluation. The linearity effects of each

    experimental arrangement are evaluated by computing

    their dynamic S/N ratios as shown in Table 5. The

    slopes of the best-fit line for the defined ideal function

    model are calculated using simple linear regression

    analysis and listed in Table 5. Another importantmeaning of the combination and S/N ratio and slope

    here is that it represents the input/output transform-

    ability of high-speed CNC milling system under the

    defined experimental conditions. That is, the larger S/N

    ratio coupled with near-one slope, the higher both the

    linear effects and the transformability between input and

    output. Those S/N data can be further translated into

    the effect each control factor has on S/N by computing

    their average values as listed in Table 6. Fig. 6 is its

    response graph.

    Table 6 suggests that the best levels for each control

    factors are A1; B1; C3; D3; E2; F3; G1; and H1 due to

    their maximum S/N ratios. The maxmin value is

    equal to the range of S/N ratio variance due to the

    change in the level setting. The larger the range,

    the more powerful impact the control factor has on

    the dimensional precision. The ranking in Table 6

    demonstrates that factor A (milling type), factor B

    (cutting speed), and factor F (number of tooth) have

    relatively strong impacts on the dimensional precision,

    while C, D, E, G, and H have relatively weak impacts.

    Factors A, B, and F should be strictly controlled for

    high-dimensional precision during the high-speed CNC

    milling process.

    Factor B (cutting speed) has been found to be the

    most important factor governing process robustness

    (dimensional quality) because of the maximum changein S/Nratios. It is due to cutting speed directly affecting

    tool wear. This is why the lowest speed level B1 results

    in the best dimensional quality. Factor A (milling type)

    is identified as the second due to its significant influence

    in vibration. A1 is downward milling type, causing

    smaller vibration. Factor F (number of tooth) is the

    third. F3 is the best level setting due to smaller cutting

    force.

    The slope of the best-fit line in the study technically

    denotes the overcutting or undercutting effects of

    the machined products. When the slope is greater

    than 1, the machined product has the undercuttingeffects, the larger the more serious, and vice versa.

    It is noted in Table 5 that all of the arranged

    experimental conditions in L18 array led to under-

    cutting effects. Table 7 shows the average effects

    each control factors have on slope and Fig. 7 is its

    response graph. The same observation about under-

    cutting effects is made for all of the control factors.

    The ranking in Table 7 displays that control factors A,

    B, and F have stronger cutting effects on the dimen-

    sional accuracy of machined products. Of the control

    factors, factor A is the most important due to its largest

    impact.

    ARTICLE IN PRESS

    Table 4

    Input signals, and their corresponding positions and programmed dimensions

    Input signal M1 M2 M3 M4 M5 M6 M7 M8 M9 M10 M11 M12 M13 M14Position P1X P1Y P6Y P7X P19Y P2X P8X P2Y P5Y P20Y P3X P9X P3Y P6Y

    Programmed dimension (mm) 2.4 2.4 2.4 2.4 2.4 5.55 5.55 6.4 6.4 6.4 9.06 9.06 12.4 12.4

    Input signal M15 M16 M17 M18 M19 M20 M21 M22 M23 M24 M25 M26 M27 M28

    Position P19Y P24 P13Y P14Y P15Y P25Y P26Y P27Y P23 P6X P12X P5X P11X P23Programmed dimension (mm) 12.4 19.2258 24.9 24.9 24.9 24.9 24.9 24.9 27.1894 29.04 29.04 32.55 32.55 33.3

    Input signal M29 M30 M31 M32 M33 M34 M35 M36 M37 M38 M39 M40 M41 M42

    Position P4X P10X P9Y P12Y P18Y P8Y P11Y P17Y P7Y P10Y P16Y P25X P26X P27XProgrammed dimension (mm) 35.7 35.7 37.4 37.4 37.4 43.4 43.4 43.4 47.4 47.4 47.4 58.425 58.425 58.425

    Input signal M43 M44 M45 M46 M47 M48 M49 M50 M51

    Position P13X P14X P15X P21X P21X P17X P20X P18X P19XProg rammed dimensi on ( mm) 81. 15 84. 3 87. 81 1 07 .79 10 7. 79 1 11 .3 1 11 .3 114 .4 5 11 4.4 5

    T. Yih-fong, J. Ming-der / Robotics and Computer-Integrated Manufacturing 21 (2005) 506517 511

  • 7/27/2019 Dimensional Quality Optimisation of High-speed CNC Milling Process With Dynamic Quality Characteristic

    7/12

    4.2. Analysis of variance (ANOVA)

    The analysis of variance (ANOVA) on the experi-

    mental results is performed to evaluate the source of

    variation during the high-speed CNC milling process.

    From the analysis, it is easy to identify which factors are

    the most important in terms of quality characteristic.

    Consequently, those important factors have to be

    carefully monitored during the process for a consistently

    high-quality product.

    Table 8 shows the ANOVA that is done on

    the S/N ratios. It is therefore easy to see that

    factors A, B, and F are the most important in terms

    of dimensional quality. These three factors account

    for about 65 percent of the experimental variance.

    The observation agrees well with those reflected in

    ARTICLE IN PRESS

    Table 5

    Experimental layout, resultant data, and performance evaluation

    NO Control factors M1 (2.4mm) M2 (2.4mm) M50 (114.45mm) M51 (114.45mm) Performance evaluation

    A B C D E F G H N 1 N2 N1 N2 N1 N2 N1 N2 S/N (db) Slope (b)

    1 1 1 1 1 1 1 1 1 2.3392 2.3474 2.3547 2.3565 114.5024 114.4947 114.4930 114.4514 26.3816 1.000001

    2 1 1 2 2 2 2 2 2 2.3426 2.3960 2.3956 2.3908 114.4463 114.5182 114.4510 114.4645 30.0520 1.000051

    3 1 1 3 3 3 3 3 3 2.4030 2.3968 2.3848 2.4046 114.5107 114.4803 114.5012 114.4508 31.5543 1.000270

    4 1 2 1 1 2 2 3 3 2.2512 2.3144 2.2703 2.3849 114.5619 114.5108 114.5201 114.4604 21.4463 1.000260

    5 1 2 2 2 3 3 1 1 2.3269 2.3904 2.3280 2.3733 114.5227 114.4910 114.5659 114.4735 26.1627 1.000324

    6 1 2 3 3 1 1 2 2 2.2854 2.3493 2.3074 2.3755 114.6139 114.6169 114.6715 114.6003 21.1510 1.000929

    7 1 3 1 2 1 3 2 3 2.1593 2.3501 2.2139 2.3792 114.6374 114.4911 114.6078 114.4749 19.6770 1.000283

    8 1 3 2 3 2 1 3 1 2.3809 2.3657 2.3372 2.3376 114.5573 114.4899 114.5087 114.4664 24.1879 1.000105

    9 1 3 3 1 3 2 1 2 2.3296 2.3652 2.3396 2.3929 114.5438 114.4695 114.5461 114.4281 23.6981 1.000300

    10 2 1 1 3 3 2 2 1 2.3458 2.3863 2.3265 2.3925 114.5476 114.4933 114.5326 114.4704 24.0331 1.000273

    11 2 1 2 1 1 3 3 2 2.3134 2.3033 2.3096 2.3656 114.5829 114.4428 114.6165 114.5200 20.7488 1.000213

    12 2 1 3 2 2 1 1 3 2.3379 2.3653 2.3447 2.3524 114.6115 114.5321 114.6151 114.5331 22.4899 1.000859

    13 2 2 1 2 3 1 3 2 2.1212 2.3235 2.0641 2.3146 114.8699 114.5506 114.8776 114.5712 14.0266 1.001444

    14 2 2 2 3 1 2 1 3 2.3485 2.3645 2.3459 2.3810 114.5181 114.4941 114.5665 114.4937 21.7157 1.000340

    15 2 2 3 1 2 3 2 1 2.3382 2.3878 2.3158 2.3966 114.5993 114.4978 114.6720 114.4842 24.3425 1.000624

    16 2 3 1 3 2 3 1 2 2.3256 2.3925 2.3223 2.3645 114.5560 114.5319 114.5801 114.5310 25.4485 1.000453

    17 2 3 2 1 3 1 2 3 2.0989 2.2817 2.1220 2.3633 114.7854 114.5193 114.7802 114.4915 13.8235 1.001144

    18 2 3 3 2 1 2 3 1 2.1808 2.3856 2.1357 2.2879 114.8085 114.5074 114.7783 114.5056 16.4607 1.001239

    Table 6

    S/N ratio response table

    Level A B C D E F G H

    1 24.9234 25.8766 21.8355 21.7401 21.0225 20.3434 24.3161 23.5947

    2 20.3433 21.4741 22.7818 21.4781 24.6612 22.9010 22.1798 22.5208

    3 20.5493 23.2828 24.6817 22.2164 24.6556 21.4041 21.7845

    Maxmin 4.5802 5.3274 1.4472 3.2036 3.6387 4.3122 2.9120 1.8103Ranking 2 1 8 5 4 3 6 7

    0

    10

    20

    30

    40

    50

    60

    70

    80

    90

    100

    110

    120

    0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100 105 110 115 120

    Programmed dimension(mm)

    Product

    dimension(mm)

    Fig. 5. The complete results plotted as the input/output relationship.

    T. Yih-fong, J. Ming-der / Robotics and Computer-Integrated Manufacturing 21 (2005) 506517512

  • 7/27/2019 Dimensional Quality Optimisation of High-speed CNC Milling Process With Dynamic Quality Characteristic

    8/12

    Table 6 and Fig. 6. Additionally, error contributing

    about 3.4% to total variance indicates that experi-

    mentation is fairly successful and reliable. It is also

    found that ANOVA has almost the same function as

    Table 6 except for the provision of more detailed

    information.

    4.3. High-speed CNC milling process optimisation

    The goal for study is to seek a robust process design

    for which the high-speed CNC milling process can

    always produce a variety of products with high-

    dimensional quality. To meet this goal, Taguchi

    proposed a two-step optimisation strategy in obtaining

    the best process conditions. The two-step optimisation

    strategy reduces the variations in the product dimension

    in the first step and then adjusts the slope of the bestfitting line to the desired level. The optimisation is:

    Step 1: reduce dimensional variability. To produce the

    strongest linear relation between signal factor (M=pro-

    grammed dimension) and output response (Y=product

    dimension), it is necessary to reduce dimensional

    variability. As such, the optimal levels for each control

    factors are those levels that maximise S/Nratios leading

    to a robust machining performance, i.e. high-dimen-

    sional precision. The selected levels are A1; B1; C3; D3;E2; F3; G1; and H1:

    Step 2: adjust slope to the desired level. According to

    the dynamic formula Y bM; it appears to select the

    ARTICLE IN PRESS

    26.6333

    A1 A2 B1 B2 B3 C1 C2 C3 D1 D2 D3 E1 E2 E3 F1 F2 F3 G1 G2 G3 H1 H2 H3

    Controlfactors and their levels

    S/Nratio(db)

    22.6333

    18.6333

    Fig. 6. S/N ratio response graph.

    Table 7

    Slope response table

    Level A B C D E F G H

    1 1.000280 1.000278 1.000452 1.000424 1.000501 1.000747 1.000379 1.000428

    2 1.000732 1.000653 1.000363 1.000700 1.000392 1.000410 1.000551 1.000565

    3 1.000587 1.000703 1.000395 1.000626 1.000361 1.000588 1.000526

    Max-min 0.000452 0.000376 0.000341 0.000305 0.000234 0.000386 0.000209 0.000137Ranking 1 3 4 5 6 2 7 8

    1.000206

    1.000506

    1.000806

    A1 A2 B1 B2 B3 C1 C2 C3 D1 D2 D3 E1 E2 E3 F1 F2 F3 G1 G2 G3 H1 H2 H3

    Control factors and their levels

    Slope

    Fig. 7. Slope response graph.

    Table 8

    Analysis of variance

    Control factors S F V r (%)

    A 94.4012 1 94.4012 24.5815

    B 97.2363 2 48.6181 25.3198

    C 6.4818 2 3.2409 1.6878

    D 37.9696 2 18.9848 9.8871

    E 41.2854 2 20.6427 10.7505

    F 56.4299 2 28.2149 14.6940

    G 27.2900 2 13.6450 7.1062

    H 9.9452 2 4.9726 2.5897

    Error 12.9939 2 6.4970 3.3835

    Total 384.0333 100

    T. Yih-fong, J. Ming-der / Robotics and Computer-Integrated Manufacturing 21 (2005) 506517 513

  • 7/27/2019 Dimensional Quality Optimisation of High-speed CNC Milling Process With Dynamic Quality Characteristic

    9/12

    slope (b) closer to 1 for higher machining accuracy.

    However, it must be noted that the adjustment of the

    slope is best not to affect the process robustness. Based

    on this principle, since all other control factors working

    at the optimal level have already achieved both highest

    S/N ratios and smallest slopes, the slope for control

    factor C (feed per tooth) is the best choice foradjustment. It is therefore predicted that the change

    from C3 to C2 leads to a highest machining accuracy

    without significantly affecting process robustness. Thus,

    the predicted optimal combination is selected as A1; B1;C2; D3; E2; F3; G1; and H1:

    4.4. Optimal performance forecasting

    Taguchi methods require a confirmation run to check

    for both the validity of experimentation and, also the

    reproducibility of the experimental results. It is im-

    portant to predict the performance under the optimal

    conditions before running the confirmation. The pre-

    dictions of S/N ratio and slope for the optimal

    conditions can be calculated by means of the additively

    law as follows:

    For the optimal conditions: A1; B1; C2; D3; E2; F3; G1;H1

    Zopt A1 B1 C2 D3 E2 F3 G1 H1 7TZ

    24:9234 25:8766 23:2828 24:6817 24:6612

    24:6556 24:3161 23:5947 7 22:6333

    37:5588 db,

    bopt A1 B1 C2 D3 E2 F3 G1 H1 7Tb

    1:000280 1:000278 1:000363 1:000395

    1:000392 1:000361 1:000379 1:000428

    7 1:000506

    0:999334.

    For the initial conditions: A1; B1; C1; D1; E1; F1; G1; H1

    Zinitial A1 B1 C1 D1 E1 F1 G1 H1 7TZ

    24:9234 25:8766 21:8355 21:7401 21:0225

    20:3434 24:3161 23:5947 7 22:6333

    25:2190 db,

    binitial A1 B1 C1 D1 E1 F1 G1 H1 7Tb

    1:000280 1:000278 1:000452 1:000424

    1:000501 1:000747 1:000379 1:000428

    7 1:000506

    0:999946,

    where TZ and Tb represented the average effects of the

    overall control factors.

    As a result, 12.3398 db db gain in S/N ratio is

    predicted.

    4.5. Confirmation run

    Fig. 8 is plotted from confirmation results for the

    optimal conditions. Table 9 is the comparison of the

    prediction and the confirmation between the initial and

    the optimal conditions. It is noted that the confirmed

    S/N ratio 37.2933 db for the optimal conditions is still

    the best when compared to the entire results in Table 5.

    The actual gain is 10.9117 db that is very close to the

    predicted 12.3398 db. This indicates the best combina-

    tion of the control factors level setting is robust enough

    against noise effects and results in high reproducibility.

    Accordingly, as demonstrated in Fig. 8, there is a very

    strong linear relationship between the input pro-

    grammed dimension and the product dimension.

    Furthermore, the % variability range improved can be

    ARTICLE IN PRESS

    y = 1.000101 x

    0

    10

    20

    30

    40

    50

    60

    70

    80

    90

    100

    110

    120

    0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100 105 110 115 120

    Programmed dimension (mm)

    Productdimension(m

    m)

    Fig. 8. Plot of the confirmation results as input/output relationship.

    T. Yih-fong, J. Ming-der / Robotics and Computer-Integrated Manufacturing 21 (2005) 506517514

  • 7/27/2019 Dimensional Quality Optimisation of High-speed CNC Milling Process With Dynamic Quality Characteristic

    10/12

    calculated by the equation

    % variabilityrange improved 1

    2

    gain=6

    % variability rangeinitial. 4

    So the gain of 10.9117 db is equivalent to reducing thevariation range by 10.5(10.9117/6)=71.6508 which in-

    dicates that the process robustness has improved 3.53

    times.

    It is also noted in Fig. 8 that using simple regression

    analysis of the confirmation run data generates the best-

    fit line equation Y=1.000101X. This is in practice the

    best functional relationship between input signal and

    output response of the high-speed CNC milling system

    under the optimal conditions. It can be estimated

    through the equation Y=1.000101M that the under-

    cutting of the machined product dimension is

    1:000101 M M=M 100% 0:0101%.

    If there is still a need to adjust the product dimension for

    suitable dimensional allowance for meeting post-proces-

    sing and customer requirements, the input programmed

    dimension can be modified through the following

    mathematical relationship:

    adjusted 1

    1:000101Yfinal, (5)

    where Madjusted is the adjusted input programmeddimension, and Yfinal is the desired final product

    dimension.

    4.6. Cutter flank wear and work piece surface roughness

    analysis

    Table 10 shows the cutter flank wear and the work

    piece surface roughness analysis comparison between

    the raw experimental runs and the confirmation trial.

    Flank wear values are the average of the data mea-

    sured on each tooth of the milling cutter. Roughness

    values are the average of 18 data measured on thesurface of the machined product. It is clear that

    the average flank wear values are among the range

    of 132.43411887.0565mm for N1 condition and of

    60.9085375.2826mm for N2 condition. The average

    surface roughness values are found among the range of

    0.25330.4833mm for N1 and of 0.18330.3617mm for

    N2 condition. The results indicate that for both the

    cutter flank wear and the work piece surface roughness,

    N2 is appreciably better than N1 because of their lower

    values plus with narrower ranges. This is due to that

    material used in N1 is harder, causing more cutting tool

    wear and then worse surface quality.

    ARTICLE IN PRESS

    Table 9

    The comparison between the prediction and the confirmation

    Level combination Prediction Confirmation

    S/N (db) Slope (b) S/N (db) Slope (b)

    Initial A1B1C1D1E1F1G1H1 25.2190 0.999946 26.3816 1.000001

    Optimum A1B1C2D3E2F3G1H1 37.5588 0.999334 37.2933 1.000101

    Gain 12.3398 10.9117

    Table 10

    Cutter flank wear and work piece surface roughness results

    NO Control factors Average flank wear (mm) Average roughness: Ra (mm)

    A B C D E F G H N 1 N2 N1 N2

    1 1 1 1 1 1 1 1 1 323.8422 215.8492 0.2533 0.1844

    2 1 1 2 2 2 2 2 2 417.3984 140.2617 0.4028 0.1911

    3 1 1 3 3 3 3 3 3 136.6434 133.7832 0.3983 0.3617

    4 1 2 1 1 2 2 3 3 275.0565 194.1669 0.2556 0.1833

    5 1 2 2 2 3 3 1 1 477.6723 158.6058 0.3600 0.2461

    6 1 2 3 3 1 1 2 2 729.1758 60.9085 0.2856 0.2361

    7 1 3 1 2 1 3 2 3 945.8786 177.0516 0.3867 0.3189

    8 1 3 2 3 2 1 3 1 537.9706 86.3831 0.2756 0.2811

    9 1 3 3 1 3 2 1 2 1283.2963 167.0189 0.4394 0.3344

    10 2 1 1 3 3 2 2 1 132.4341 97.5158 0.3178 0.2767

    11 2 1 2 1 1 3 3 2 323.2417 154.5529 0.2694 0.2733

    12 2 1 3 2 2 1 1 3 725.3799 273.2197 0.3778 0.2739

    13 2 2 1 2 3 1 3 2 1470.6875 375.2826 0.4833 0.2250

    14 2 2 2 3 1 2 1 3 229.2672 338.1873 0.3289 0.3039

    15 2 2 3 1 2 3 2 1 202.1113 67.3019 0.3667 0.3106

    16 2 3 1 3 2 3 1 2 620.2814 119.4116 0.3428 0.3261

    17 2 3 2 1 3 1 2 3 747.5027 117.9605 0.3806 0.2378

    18 2 3 3 2 1 2 3 1 1887.0565 124.0422 0.3472 0.3578

    Confirmed 1 1 2 3 2 3 1 1 180.6815 79.9970 0.3111 0.2244

    T. Yih-fong, J. Ming-der / Robotics and Computer-Integrated Manufacturing 21 (2005) 506517 515

  • 7/27/2019 Dimensional Quality Optimisation of High-speed CNC Milling Process With Dynamic Quality Characteristic

    11/12

    It is noted in Figs. 9(a), (c), (e), and (g) that the

    optimal conditions results in better form shape at the

    corners of the machined products. It is also observed in

    Figs. 9(b), (d), (f), and (h) that flank wear has been

    significantly improved at the optimal conditions which

    has 180.6815mm for N1 condition, and 79.9970 mm for

    N2 condition. Compared to the entire results in the L18

    array, they are better than most of the data. Similar

    observation is made on the surface roughness analysis.

    It seems to point out that the optimised processparameters developed in terms of dimension vs. dimen-

    sion also works in both the reduction in tool wear and

    the improvement in surface roughness.

    4.7. Conclusion remarks

    The study using Taguchi dynamic experiment coupled

    with the ideal function model of programmed dimension

    vs. product dimension has been very successful in

    developing a robust, versatile, high-dimensional quality

    of high-speed CNC milling technology. Based on the

    experimental results, conclusions can be drawn as

    follows:

    1. The optimised control factors are: A1 (milling type),

    B1 (cutting speed), C2 (feed per tooth), D3 (film

    material), E2 (tool material), F3 (number of tooth),

    G1 (rake angle), and H1 (helix angle).

    2. The most important factors identified by S=N andANOVA analysis affecting the process robustness are

    in a decreasing manner: factor B (cutting speed),factor A (milling type), and F (number of tooth).

    They account for about 65% of total variance.

    3. Actual gain 10.9117 db is very close to the predicted

    12.3398 db. It shows very good reproducibility and

    confirms the success of the experiment.

    4. Dimensional variability after process optimization

    has been significantly improved to be 28.3492% of

    the initial conditions, leading to 3.55903 times

    improvement in the process robustness.

    5. The optimal function between input signal and

    output response can be linearly described as

    Y=1.000101M.

    ARTICLE IN PRESS

    Fig. 9. OM photographs of the form shape of the machined products and the flank wear of the cutting tools. (a), (b), (c), and (d) are for the initialconditions and (e), (f), (g), (h) for the optimal conditions.

    T. Yih-fong, J. Ming-der / Robotics and Computer-Integrated Manufacturing 21 (2005) 506517516

  • 7/27/2019 Dimensional Quality Optimisation of High-speed CNC Milling Process With Dynamic Quality Characteristic

    12/12

    References

    [1] Fallbohmer P, Rodriguez CA, Ozel T, Altan T. High-speed

    machining of cast iron and alloy steels for die and mold

    manufacturing. J. Mater. Process. Technol. 2000;98:10415.

    [2] Ning Y, Rahman M, Wong YS. Investigation of chip formation

    in high speed end milling. J. Mater. Process. Technol. 2001;113:

    3607.[3] He N, Lee TC, Lau WS, Chan SK. Assessment of deformation of

    a shear localized chip in high speed machining. J. Mater. Process.

    Technol. 2002;129:1014.

    [4] Gu J, Barber G, Tung S, Gu RJ. Tool life and mechanism of

    uncoated and coated milling inserts. Wear 1999;225:27384.

    [5] Jawaid A, Koksal S, Sharif S. Cutting performance and wear

    characteristics of PVD coated and uncoated carbide tools in face

    milling Inconel 718 aerospace alloy. J. Mater. Process. Technol.

    2001;116:29.

    [6] Smith S, Dvorak D. Tool path strategies for high speed milling

    aluminum workpieces with thin webs. Mechatronics 1998;8:

    291300.

    [7] Dolinsek S, Sustarsic B, Kopac J. Wear mechanism of cutting

    tools in high-speed cutting processes. Wear 2001;250:34956.

    [8] Diniz AE, Micaroni R. Cutting condition for finish turningprocess aiming: the use of dry cutting. Int. J. Mach. Tools Manuf.

    2002;42:899904.

    [9] Gunter KL, Sutherland JW. An experimental investigation

    into the effect of process conditions on the mass concentra-

    tion of cutting fluid mist in turning. J. Cleaner Product. 1999;7:

    34150.

    [10] Lahres M, Jorgensen G. Properties and dry cutting perfor-

    mance of diamond-coated tools. Surf. Coat. Technol. 1997;96:

    198204.

    [11] Vieira JM, Machado AR, Ezugwu EO. Performance of cuttingfluids during face milling of steels. J. Mater. Process. Technol.

    2001;116:24451.

    [12] Avila RF, Abrao AM. The effect of cutting fluids on the

    machining of hardened AISI 4340 steel. J. Mater. Process.

    Technol. 2001;119:216.

    [13] Su YL, Yao SH, Wei CS, Wu CT. Analyses and design

    of a WC milling cutter with TiCN coating. Wear 1998;215:

    5966.

    [14] Su YL, Yao SH, Wei CS, Wu CT, Kao WH. Design of a titanium

    nitride-coated WC milling cutter. J. Mater. Process. Technol.

    1999;86:2336.

    [15] Su YL, Kao WH. Optimum multiplayer TiNTiCN coatings for

    wear resistance and actual application. Wear 1998;223:11930.

    [16] Huang JC. Metal Handbook, Machinery, 2nd ed. Taiwan: Taipei;

    2001.[17] Robust Design Using Taguchi Methods, Workshop Manual, 3rd

    Version, 2000, ASI.

    ARTICLE IN PRESS

    T. Yih-fong, J. Ming-der / Robotics and Computer-Integrated Manufacturing 21 (2005) 506517 517