CONTROLLED-DEPTH LASER CUTTING OF ALUMINUM SHEET FOR LAMINATED-OB JECT MANUFACTURING

B Y

Pun Pang (Matthew) Shiu

A thesis submitted to the Department of Mechanical Engineering in conformity with the requirements for the &gree of Master of Science (Engineering)

Queen's University Kingston, Ontario, Canada

September 200 1

Copyright Q hin Pang Shiu, 2001

National Libraiy 1*1 o f C a d a

Acquisitions and Acquisi i et Bibiiiraphic Services senrices bibiiiraphiques

The auîhor has granted a non- L'auteur a accordé une licence non exclusive licence aiiowing the exchsive permettant à la National h i of Canada to BibIiothèque nationale du Canada de reproduce, loan, disîriiute or seil reprodime, prêter, disîriiuer ou copies of this thesis in microform, vendre des copies de cette thése sous paper or electronic formats. ia forme de microfichelfilm, de

reproduction sur papier ou sur format électronique.

The author retains ownershrp of the L ' a m conserve la propriété du copyright in this thesis. Neither the droit d'auteur qyi protège cette thèse. thesis nor substantial extracts fiom it Ni la thése ni des extraits substantiels may be printed or otherwise de celle-ci ne doivent être imprimés reproduced without the author's ou autrement reproduits sans son permission. autorisation.

Abstract

Controlled-depth laser cutting is an important step in the Laminated Object

Manufacturing from aluminurn sheet, a novel Rapid Prototyping process being currently

developed. The precise depth control is required in order to cut through the top sheet

without damage to the stack of laminations below which have been bonded with a

polymeric adhesive.

An extensive series of experiments has been conducted to identify suitable process

parameters to achieve this goal. Laser cutting tests with a Nd:YLF 10-W Q-switched

laser were ciïiïried out on soiid aluminum plate and on laminated-sheet specimens. The

primary process parameters varied were laser power and scan speed while the primary

obsewed laser-cut grwve characteristic was the cut &ph. An novel metallographic

examination procedure which allows extraction of the cut profile image has been

developed. Specimen examination showed that the cut depth increased with pwer and

decreased with speed. The general trend observed was confirmed by the existing

theoretical models.

In addition to the cut &pth, other geomemc characteristics of the cut were observed A

groove profile shape parameter was inwduced as a way of concisely describing the

profile shape in addition to its basic height and width dimensions.

An empiricai model of the laser grooving process has been developed based on the

Artificiai Neural Networks. The model formulation involved identifying training and test

subsets h m the experimental data, uaining and testing various proposed network

structures and identifying the most suitable ones. Two-Iayer networks with two or three

neurons in hid&n Iayer were able to model the relationship with good accuracy.

Acknowledgements 1 wish to express my appreciation to my supervisor, Prof. Gene Zak, for his valuable

guidance and advice while 1 conducted my research and during the preparation of this

thesis.

1 am grateful to Wanda W. Clapp for providing engineering knowledge and advice in

metallographical studies, Sarah S. Hinchcliffe for image capturing, Paul Noland for SEM

image capturing, and to my lab mates, Mingling Chen, Wendy Wang, Jasmine Wang for

their shining ideas, advice, and help, in the times 1 needed help in research and in

personal matters.

1 am sincerely thankful for the financial assistance by Dr. Zak, through a CAMM

research grant, and to the Department of Mechanical Engineering and School of Graduate

Studies at Queen's University.

Finally, 1 would like to thank my mother, Yan, father, Kwok-Cheung, brothers, Wai,

Hang, Yu, and sisters, Wemby and Shinny, for providing me with continuous support

during ail these years.

iii

Table of Contents . . Absfra~t .......................m........m...n.. m no u ... ... AckmwIedgementS ....n......n.....Uii..... m.m."~.........m..e.......ae.......~..e......mm..~m.n~*m..~

.............. Table of Contenis .. ....... ..................... ..................................... ..............O......... iv ... LM of Figures ..m..............~..~....~m............~....*.~m..~".m.~~..mm....a....a~~~~~mm......~....mammm..oe....

... List of Tables ................................................................................................................ x u i

Nomenclature ........................... .... ............................................................................... xiv

Chapter 1 Introduction ............................................................................................. 1.1

....................................................... 1.1 Rapid Prototyping. Manufacturing. and TooIing 1.1

............................... 1.2 Commercial Laminated-Object Manufacturing (LOM) Process 1.2

....................................................................................................... 1.3 Literature Review 1.4

1.3.1 Laser Cutting .................................................................................................. 1.4

............................................................................................... 1.3.2 Laser Grooving 1.5

1.4 Problem Identification .............................................................................................. 1.6

1.5 Research Objectives and Methodology ...................................................................... 1.7

1.6 Thesis Outline ............................................................................................................ 1.8

........................................................................... Chapter 2 Laser Materid Processing 2.1

........................................................ 2.1 Reasons for Using a Laser in LûM-AL Rricess 2.1

.......................................................................................... 2.2 Laser-Materiai Intemction 2.1

.............................................................................................. 23 Material-Related Issues 2.4

..................... 2.3.1 Metai materiai properties relevant to Iaser material processing 2.4

2.3.2 Surface finish ................................................................................................. 2.5

...................................................................................... 2.4 Laser Processing Parameters 2.6

............................................................................... 2.4.1 Power and power density 2.6

.............................................................................................. 2.4.2 Temporai Mode 2.7

2.4.3 Spatial Mode ................................................................................................. 2.7

.............................................................................................. 2.4.4 Pulse fiequency 2.8 . * 2.4.5 Polanzauon ..................................................................................................... 2.9

2.5 Beam-Delivery S ystem ............................................................................................ 2.10

2.5.1 Mirmrs .......................................................................................................... 2.10

.................................................................................................... 2.5.2 Focus Iens 2.10

2.5.3 Bearn expander ............................................................................................ 2.12

2.6 Theoretical Analysis of Laser Cutting ..................................................................... 2.12

2.7 Summary .................................................................................................................. 2-14

Chapter 3 Laser-Cutting Experiments: Preparation and Data-Extraction Metbods

.... ..................................................... ............... ................................................... 3 . 1

3.1 Specimen Preparation ................................................................................................. 3.1

3.2 Experimental Apparatus ............................................................................................. 3.4

3.3 Procedure .................................................................................................................... 3.6 3.3.1 Preiiminary tests ............................................................................................. 3.6

3.3.2 Primary-mess-Parameter Tests .................................................................. 3.7

3.3.3 Seconm-Process-Parameter Tests ........................................................... 3.8

Gas Assist ....................................................................................................... 3.9

................................................................................ Lmer pulse fiequency 3.10

Defocused Cutting ........................................................................................ 3.11

............................................................................................. Multiple passes 3.11 . . ............................................................................................ 3.4 Specimen Examnation 3.12

3.4.1 Scanning Electron Microscope (SEM) .................................................... 3.12

3.4.2 Profilometry (Vertical Scanning Interferometry) ........................................ 3.13

.............................................................................................. 3.4.3 Metallography 3.15

3.5 Summary ................................................................................................................. .3.17

Chapter 4 Laser-Cutting Experiments: Results and Analysis ................, .. .... ....- 4.1

4.1 Observeci features of iaser-cut p v e sections .......................................................... 4.1

...................................................................... 4.2 Extracting laser-cut groove-shape data 4.3

............................................................................................ 4.2.1 Shape modehg 4 3

4.2.2 Shape Data Collection ................................................................................... 4-6

4.3 Assessment of Measurement Repeatabitity ........................................................... 4-7

.......................................... 4.4 Preliminary Tests: Through Cutting of AIuminum Sheet 4.8

4.5 Primary-mess-Parameter Tests .......................................................................... 4.10

4.5.1 Solid-Plate (SP) Specirnens ......................................................................... 4-10

4.5.2 Laminated-sheet (LS) Specimens ................................................................. 4.14

4.5.3 Cornparison of results for SP and LS specimens ......................................... 4.17

................................................. 4.5.4 Discussion of recast-formation mechanim 4.18

4.6 Secondary-Process-Parameters Tests ....................................................................... 4.19

.................................................................................... 4.6.1 Gas-Assisted Cutting 4.20

............................................................................................... m e n Assist 4.20

........................................................................................... Nitrogen Assist 4 . 2 2

Summclry ................................................................................................. 4 . 2 4

4.6.2 Defocused cutting ......................................................................................... 4.24

4.6.3 Laser-pulse-frequency variation ................................................................. 4.26

4.6.4 Multiple-pass cutting .................................................................................... 4.37

............................................................................. Case A: Increased Speed 4.27

..................... Case B: Reduced power ... .............................................. 4.29

4.7 Cornparison with Laser Machining Mode1 ............................................................. 4.31

4.8 Summary ............................................................................................................. 4.32

.......... Cbpter 5 Modeüng of Laser-Cut Profde Using Artificial Neural Networks 5.1

..................................................................................... 5.1 Neural Network Architecture 5.2

..................................................................................... 5.1.1 Biological inspiration 5.2

5.1.2 Neuron ........................................................................................................ 5 . 3

5.1.3 Transfer function ................................................................................... . . 5 . 4

5.1.4 A layer of neurons ......................................................................................... 5.4

5.1.5 Multiple layers of neurons ........................................................................... 5.4

5.1.6 Optimal network structure .............................................................................. 5.6

5.1.8 Leaming d e .................................................................................................. 5.7

5.1.9 Leaming rate .............................................................................................. 5 . 7

5.1.10 Convergence ................................................................................................ .5.7

.............................................................................................. 5.1.1 1 Generaiization 5.8

5.2 Neural Network Implementation ................................................................................ 5.8

....................... 5.2.1 Motivation for Neural Network application to laser grooving 5.8

5.2.2 Defining Input/Output parameters for Artificial Neural Network use ........... 5.9

5.2.3 Neural Network Training and ImpIementation .......................................... 5.11

5.3 Neurai Network Tests and Resdts .......................................................................... 5.13

................................................................................................ 5.3.1 Network tests 5.13

........................................................ 5.3.2 Shape-to-Process (S-P) Network Tests 5.13

5.3.3 Ftocess-to-Shape (P-S) Network Tests .................................................. 5.14

.................................................................................................................. 5.4 Summary 5.17

Chapter 6 Conclusions and Recornrnendations.~ ..... ................................................. 6.1

6.1 Conclusions ............................................................................................................... 6.1

6.2 Recommendations and Future Work .......................................................................... 6.3

Appendix A . Chernical etching procedure for preparation of laminated-sheet

specimens .......................................... ". .... ..... ..- ........ ................................................ A.l

Appendix B . Cutthg orientation of specimens .......................................................... A 3

...... ,..*...*......................*.............. Appndix C Detailed Metallograpbic Proceduce A.4

.............................. Appendix D Deîail Procedure for Ssmpling Laser-Cut R o f k A.8

vii

List of Figures Figure 1.1 The commercial LOM process .................................................................... 1.3

Figure 1.2 M a t d removal mechanism for LOM4L ................................................ 1.7

Figure 2.1 Laser cutting process .................................................................................... 2.2

Figure 2.2 Mechanism of laser (through) cutting ............................................................. 2.3

Figure 2.3 Mechanism of laser grooving ................................................................... 2.4

Figure 2.4 Reflectivity vs . laser light wavelength (ANSVAR'S C7.2: 1998) .................. 2.5

Figure 2.5. Power density requirernents for various Laser Material Processing tasks

mohanty et al.. 20011 .................................................................................................. 2.6

Figure 2.6 Peak power and pulse duration for Continuou. Pulsed, and Q-switched

........................................................................................................ temponi modes . 2 . 7

Figure 2.7 (a) Gaussian beam or TEMOO spatial mode. (b) m l 2 spatial mode [Luxon.

20011 ............................................................................................................................... 2.8

Figure 2.8. Pulse-to-pulse overlap .................................................................................... 2.9

figure 2.9 Schematic of a focusad laser beam ......................................................... 2.11

....................................................................................... Figure 2.10. A beam expander 2.12

Figure 2.1 1 Location of the emsion front of laser p v h g ........................................... 213

Figure 3.1 Dimension of the cut location on specimens. the aluminum solid plate (SP)

................. (left) and the laminated sheei (LS) specimen (right) (aii dimensions in mm) 3.2

Figure 3.2 The three-slot spccimen (dimensions in mm) ............................................... 3.3

Figure 3.3 Three-dot (Foreground) and the-laminated sheet specimens prior to the

cutting experiments ..................................................................................................... 3.4

Figure 3.4 The exprimentai setup ................................................................................... 3.5

........................................ Figure 3.5 Close-up view of the X-Y platform and focus Iens 3.6

Figure 3.6 Geometry of specimens and line scans (same line pattern was used for al1

specimens except for three-slot preliminary tests) .......................................................... 3.8

.................................................................................... Figure 3.7 Nozzle for assist gases 3.9

Figure 3.8 (a) SEM image and (b) rnetallographic image .............................................. 3.13

Figure 3.9 Image captured by WYKO profilometry ...................................................... 3.14

Figure 3.10 Cut cross-section data h m WYKO profilometry ................................... 3.14

Figure 3.11 (a) Specimen segment encased in epoxy after the first rnounting stage; top

and front view shown; (b) Two specimen segments placed in a cylindricd mould in the

second mounting stage . top view .................................... ,. ................................. 3.16

Figure 3.12 Photo shows the remaining pieces of (la) soiid-plate and (lb) Iaminated-

sheet specimens after segment removd for examination; (2) specimen after first

.......................................... mounting stage; (3) specimen after second mounting stage 3.16

Figure 3.13 The finished polished specirnen by metallographical method ................... 3.17

Figwe 4.1 Definition of measured laser cut features: (a) sheet cut. 12 W . 2ûûû d m i n .

.................................................................................. (b) sheet cut. 7 W. 2000 mmlmin 4 . 2

Figure 4.2 Effect of profile shape parameter B .......................................................... 4.3

Figure 4.3 Laserat groove profile with p and q parameters defined ............................. 4.4

Figrire 4.4 An exemplary plot of sampled profile curve points used to estimate groove

profile shape B . "Left" and "right" refer to the corresponding halves OC the curve's two

......................................................................................................................... branches.. 4.6

Figure 4.5 SampIed cut profile points together with power function shape mudel fitted to

................................................................................................................. them (B = 2.4). 4.7

Figure 4.6 Profiles h m preliminary tests: (a) 8000 &min scan speed, (b) 4000

............................................................................................................................ mmfmin 4.8

Figure 4.7 Cut &pth for preliminary tests ................................................................. 4.9

Figure 4.8 Kerf width for preliminary tests ...................................................................... 4.9

Figure 4.9 Examples of groove cross-sections for solid-plate specimens: (a) tecaskfilled

p v e (l2W. 2000 d m i n ) ; (b) clean groove (12W. 4500 mdmin) ........................ 4.10

Figure 4.10 Cut depth (experirnent and model) for solid-plate specimens .................... 4.11

Figure 4.1 1 Kerf width for solid-plate specimens .......................................................... 4.12

Figure 4.12 Profile shape parameter B for solid-plate specimens .................................. 4-13

Figure 4.13 Recast depth for solid-plate specimens ....................................................... 4.13

Figure 4.14 Recast area fraction for sotid-plate specimens ........................................... 4.14

Figure 4.15 ExampIes of groove cross-sections for laminated-sheet specimens: (a)

throughcut sheet (l2W, 1500 &min); (b) recast-free groove (12W. 4500 mm/min).4.15

Figure 4.16 Cut depths for laminated-sheet specimens .............................................. 4.16

Figure 4.17 Kerf width for larninated-sheet specimens ................................................. 4.16

Figure 4.18 Recast area fraction for laminated-sheet specimens ................... .... ...... 4 . 17

............................. Figure 4.19 Ratio of laminated sheet depths over solid plate depths 4.18

..................................................................... Figure 4.20. Recast formation mechanism 4.19

.......... Figure 4.21 (a) no O2 gas assist. (b) Ot gas-assist (for both 1500 &min. 7W) 4.20

Figwe 4.22 Cut depth for oxygen-assisted cutting experiments ....................... .. ........ 4.21

Figure 4.23 Kerf width for oxygen-assisted cutting experiments .................................. 4.21

............................... Figure 4.24 Recast depth for oxygen-assisted cutting expenments 4.22

Figure 4.25 (a) no N2 gas assist (a) Nz gas assist (for both 1500 mm/min. 7W) ........... 4.23

Figure 4.26 Cut depth for nitrogen-assisteci cutting experiments .................................. 4.23

Figure 4.27 Kerf width for nitrogen-assisted cutting experiments ................................ 4.24

Figure 4.28 Cut deptti for defocused-cutting experiments . ........................................... 4.25

........................................ Figure 4.29 Recast &pth for defocusedcutting experiments 4.26

Figure 4.30 Cut depth fot pulse-frequency-variation experiments ..................... ........... 4.27

Figure 4.31 Groove profiles comparing (a) single-pas cut (2500 rnmlmin. 7W) and (b)

two-pass cut (5000 mmlmin. 7W) ................................................................................ 4.28

Figure 4.32 Cut depths for single-pas and two.pass. speeddoubled tests ................... 4.28

F i p 4.33 Kerf width for single-pass and two.pass. speed-doubled cuts .................... 4.29

Figure 4.34 Groove profiles comparing (a) single-pass cut (3000 d m i n . 7W) and (b)

two-pass cut (3000 mm/rnin. 3.5W) ............................................................................... 4.30

..... Figure 4.35 Cut depths for singie-pass and two-pas reduced-energy (3.5 W) tests 4.30

. ....................................................*... Figure 5.1 Biologicai Neumns (Hagan. et ai 1995) 5.2

Figure 5.2 Schematic of a Multiple-input neumn ......................................................... 5.3

Figure 5.3 Schematic of a rnultiple-input neumn. two-layer network* ............................ 5.6

Figure 5.4 An example of a bottle curved contour surface .............................................. 5.9

Figure 5.5 An example of the multiple p a s cuaing ................................................... 5.9

Figure 5.6 A shematic diagram to npresent gmuping parameters that related to laser

................................................................................................................ cutting pmcess 5.10

............... Figure 5.7 An exemplary training e m r history graph for 2-3-2 S-P nenvork 5.12

Figure 5.8 Average % mors for the training set in S-P (2-x-2) networks ..................... 5.14

Figure 5.9 Average % emrs for the test set in S-P (2-x-2) networks ............................ 5.15

........................ Figure 5.10 Average 46 emrs for minhg set in P S (2-x-2) nehvorlrs 5.15

Figure 5.1 1 Average % erron for the test set in P S (2-x-2) networks .......................... 5.16

Figure 5.12 Kerf half-width predictions by P S 2-3-2 network.(holIow&Ie data points

are from the testing set) .................................................................................................. 5.17

Figure 5.13 Cut depth predictions by 2-3-2 PS network (hollow-circle data points are

h m the testing set) .................................................................................................. 5.17

xii

List of Tables Table 3.1 Sealing temperature and pressure ..................................................................... 3.3

Table 3.2 Common laser settings (used in aü experiments. uniess noted otherwise) ...... 3.7

Table 3.3 Experimentai settings for primary process parameter tests .............................. 3.8

Table 3.4 Experimental settings for gas-assisted tests ................................................. 3.10

Table 3.5 Different h l s e frequency .............................................................................. 3.10

Table 3.6 Defocused Tests ............................................................................................. 3.11

Table 3.7 Experimentai setîings for multiple-pass tests ................................................. 3.12

..................................... Table 4.1 Measurement Repeatability Results .... ................. 4.7

.......................................................... Table 5.1 Transfer Functions in a Neural Network 5.6

........................ Table 5.3 Experimentai Observation of Laser Cut Profile Parameten* 5.13

................................. Table 5.4 Testing increased number of neurons S-P network test 5.14

Nomenclature

1

i-j-t network

material density

laser kam wavelength

area of the focused laser beam at the workpiece surface

absorptivity

neuron bias value

laser-cut groove profile shape parameter

specific heat

continuous wave laser output

diameter of the focused spot

diameter of the unfocusecl beam

activation function or transfer function

focal length of the focusing Iens.

power density

A two-Iayer neural network which is defined by i inputs, j fleurons

in the hidden layer, and r outputs

the latent heat

summation of the parameters, pi, pt, p3.--.pj, mdtiplied by the

weighis, wl, w2, ~3...-wj.

kerf haIf-width of the laser-cut profile

laser output power

Neural Network input parameter

cut &pth of the Iaser-cut profile

p v e depth predicted by a mathematical mode1

transverse electromagnetic mode defineci energy dismbution

within îhe laser barn in the plane perpendicular to the direction of

propagation

general representation of output parameter for a Neural Network

m m temperature

laser scanning speed

Chapter 1. Introduction 1.1 Rapid Prototyping, Manufacturing, and Tooling

Product development cycle starts with design and continues with creation of non-

functionai and functionai prototypes, cuiminating with the manufacniring pmess

development New product development represents a major investrnent for a Company;

reduction of product development time would be of great benefit. By significantly

speeding up the fabrication of prototype models, Rapid Pmtotyping (RP) technologies

have enabled companies to achieve significant savings in product development time and

expense (ïhilmany, 200 1)

Generally, RP technologies buifd parts through a layer-by-Iayer additive process based on

a part mode1 created with Computer Aided Design (CAD) software. However, most RP

techniques are limited in their material selection to polymers or papers. While attempts

have been made to employ these polymer-based RP technologies to produce short-run

production tooling, their success has been limited.

RP technologies can be classified by material into t h e categories: liquid-based, powder-

based, and solid-based Stereolithography (SL), developed by 3D Systems, is the most

well-known example of a liquid-based technology and is the first commercialized RP

process. It empIoys Iiquid photocurable resin which is selecrively solidified in thin layers

by a W laser (Jacobs, 1992). There have been several attempts to use SL for fabrication

of low-volume production twls (Tsang and Bennett, 1995) and patterns for investment

casting (Jacobs, 1995). However, due to material Iimitations (low strength, poor themal

conductivity) these have met with oniy iimited success.

An example of powder-based RP technology is Selective Laser Sintering (SLS), a

commercial process implemented by DTM hc. The pmess can use a variety of

materiais in powder form, such as thermoplastics, ceramics and metds (a high-melting-

point metal is coated by a low-meiting-point one) (Boureii et ai., 1994). Thin Iayers of

powder are selectivdy scanned by a CO2 laser to fuse the particles together by local

heating. Metal-based objects can be made into short-nm tooling by secondary processes

(infiltration with molten metal) required to eliminate the porosity (Barlow et al., 19%).

Limitations are the smail size of these objects, expense and time complication of the

wondary pmcess steps. Size limitations are due to larger thermal stresses leadhg to

crac king.

Laminated-Object Manufacturing (LOM) can be classified as a solid-based RP

kchnology (Chua and Leong, 1997, Ready, 2001, and Jacobs, 1996). It builds objects by

sequential lamination of thin-sheet materials. For each layer, a heated d e r bonds the

thermoplastic-adhesive-lined sheet to the previously built stack, and then a laser cuts the

desired 2-D contour. The advantages of this approach are, first, that a phase

transformation is avoided (and together with it the problems of shrinkage and thermal

stress buiId up), second, that the process is easily scaied up to build Iarger parts, and,

third, that it is faster for building bulky objects since there is no need to "scan" the

interior of the 2-D contours, unlike in the SL and SLS.

However, the commercial process has been Iimited in material selection to (mainly) paper

and phtic sheet. There has been research carried out where the LOM method has been

applied to composite prepreg composite sfteets (KIosterman et al., 1997) and to sIipcast

ceramic strips (Rodrigues et ai., 2000). The latter wouId be then further processed

through binder burnout and sintenng to make ceramic objects.

Currently research is king carried out at the Rapid Laminated Tooling laboratory in the

Centre for Automotive Materiai and Manufactunng to develop an RP process which

would allow fabrication of Iaminated objects from aiuminum, a LOM-AL process. Two

main elements of any LOM process are sheet bonding and sheet cutting. This thesis will

deai with the investigation of laser cutting of aluminum sheet within the context of the

LOM-AL process developmen~

1.2 Commercial Laminated-Object Manufacturing (LOM) Procesis

LOM proces was developed by Heiïsys, Inc., with the h t system shipping in 1991

(Chua, 1997). A commercial LOM system consists of a CO2 laser, a heated d e r , a

moving optical head, and a Z motion system for the part-carrying platfonn (Figure i.1).

Part fabrication in LOM follows the steps of prepmessing, building, and postprocessing.

Preprocessing srarts with creating a desired mode1 in CAD systern, then slicing this

mode1 into cross-sections to obtain the 2 9 contours of layers. Building process consists

of:

(1) Transfemng the new sheet material ont0 the top of the previously built up stack.

(2) Bonding the new sheet to the stack by applying heat and pressure with a heated d e r

moving across the newly placed sheet. Bonding is accomplished by melting the

thennoplastic adhesive backing.

(3) Cutting the new sheet with the laser beam. The beam traces two types of patterns on

the sheet: first, contours of the part's cross-section and, second, a special hatching

pattern for the regions surmunding the part and in the interior which are to be

removed during pos~~ocessing.

Postprocessing first involves "break out" of the part itseif from the "block" of

laminations obtained at the end cf the building process. Additionally, surface coating

may be appiied to enhance part appeamce and longevity or several subparts may be

glued together to form an object of size exceeding the system's work envelope.

MOVlNG OPnCAL HEAD 7 ,- MIRROR

LASER

LAVER COHCOUR

HEATED ROUER

SHEET MATER W

TAK€-UP R O U

Figure 1.1 The w m m e ~ LOM proeess,

1.3 Literature Review

Since the invention of the Iaser, diverse applications in industry were found due to both

its ability to generate high power density and its ability to confine this power in a very

small beam spot. The applications include welding, cutting, grooving, drilling,

machining, and surface treatments (Ready, 2001). The pxinciple advantage of the laser

cutting for rapid prototype manufacturing is the ability to cut materiais under autornated

computer control without physicd contact. The generai category of appkcations of

relevance to this thesis can be identified as "laser machining." Among the variety of

Iaser machining processes, drilIing, cutting, and grooving are the most relevant,

Laser drilling involves making through or blind holes; laser cutting reiers to making

through cuts in materiais in foil, sheet, or plate form; laser grooving produces finitedepth

grooves on the material's surface. A brief review of research in these areas foilows.

13.1 Laser Cutting

Chryssolouris (1991) developed a generalized laser-machining model for cutting based

on the heat conduction theory at the erosion front (the material surface absorbing laser

beam energy). The model was able to estimate the cutting speed for specific thickness.

Mode1 derived by the above approach for laser grooving resulted in identicd

mathematical expression.

Kaekrnick et ai. (1999) investigated a mathematical model of laser cutting based on a

simplified 3-D geomeay of the pulsed laser cut, The investigation focused on finding the

correlation between kerf width and pulse width. Experiments on mild-steel sheet with

oxygen-assisted cutting showed that the optimal cutting conditions couid be found when

the kerf width appeared to first increase and then slightly decrease after the critical speed

was reached

Di Pietro et al. (2000) investigated quality irnprovements in laser cutting. Their work

uses a model-based optirnization strategy based on numerical description of the process

to manipulate Iaser beam power in order to stabilize the cutting h n t temperature and

thus to optimize the cutting quaIity. Improvements recorded were reduced kerf widening

effects and reduced power losses.

De Graaf et al. (2000) investigated laser cutting of aiuminm-synthetic laminates

consisting of up to six aitemating Iayers of aluminum (025 mm thick) and synthetic. The

synthetic materiais were epoxy-embedded glass or aramid libers, or polypropylene. The

laminated material total thickness was 1 to 3 mm. Traditionai machining was found to

delaminate the lower layer, therefore, laser cutting was employed. Their results showed

that the material could be cut at the same speeds as the homogeneous aiuminum alIoys.

Walczyk (1998) studied bevei cutting of steel laminations to be used for profiled-edge

laminated tooling in order to eliminate the stepping at the edges. Cutting methods, such

as Nd:YAG laser, abrasive water jet, and machining with endmiIl, were applied to

produce bevel cuts in steel at different angles up to 80 degrees. Laser cutting was found

to be the best approach due to non-existent toolwear, absence of cutting force, and narrow

kerf width.

An assisting gas is used in some laser rnachining applications to increase cutting rates,

change surface characteristics, or simply to protect the focusing optics, Kristensen et al.

(1994) investigated CO2 laser cutting of aiuminum and its alIoys. Applying 02 gas assist

increased the cutting rate by 50%. Faerber and Broden (1999) found use of Oz gas in

cutting of aluminum ailoys produced rougher cut surface than N2.

133 Laser Grooving

Cho and ChryssoIouris (1995) investigated both laser grooving and cutting processes.

Their analysis is based on the assumption of complete moIten materiai removai. The

mode1 produces a closed-fonn solution to approximate gniove depth. The solution was

vaüdated by experiments with aiuminum oxide with an off-axis jet aiding in removal of

the molten material. Results showed that cut depih estimation was in good agreement

with experimental data

Lallemand et al. (2000) exploced the effect of gas (air or onygen) deiivery-tube angIe on

the groove shape in (NdYAG) laser grooving of met& (aluminum, copper, titanium).

Taguchi methods were used to &nt@ the most important parameters influencing the

p v e size. Significant sensitivity of the p v e shape to the gas tube position was

found.

Sheng et ai. investigate an acoustic sensing technique for the in-process monitoring of the

erosion front condition in laser drilling (Sheng et al. 1994a) and grooving (Sheng et al.

1994b). DriU depths were successfuily controlled with this methoci. Theoretical analysis

was developed relating resonant frequency and the hoie and p v e geometry. Mode1

estimated the groove depths within 25% of the experiment for aqlic, aluminum oxide

and steel.

1.4 Problem ldentif ication

In the LOM-AL process, it is expected that thin sheets of aluminum (0.1-0.3 mm thick)

will be bonded to each other by layers of polyrnexic adhesive (0.03 -0.05 mm thick). The

laser mut be able to cut through the topmost sheet whiIe avoiding signifiant damage to

the next lamination (Figure 1.2). Laser cutting of these aiuminum-adhesive laminates

poses several unique challenges not found in the reporteci laser cutting and grooving

applications.

In one respect our process is similar to blind cutting, since the vaporized and molten

matenal cannot escape through the bottom of the CUL However, blind cutting is normally

performed within homogeneous material, whereas in our case the laser must cut through

one sheet of aluminum. Also, uniike in a laser-cutting process, very fine control of the cut

depth is required in our case to prevent significant damage to the underlying sheets.

Since the kerf material cannot escape through the cut bottom, assisting gas can no longer

be used to aid in material removd. If used at aii, its role is reduced to that of Iens

protection. Additionaüy, the primary materiai-removai p e s s must be vaporization,

since otherwise unacceptably poor cut quality wiIi result.

1.5 Research Objectives and Methodology

The goal of my research is to investigate the relationship between the process parameters

and the laser cut groove dimensions, in order to provide guidance for the laser-cutting

task in LOM-AL process.

To achieve the above goal, it is proposed that a series of experiments be conducted where

primary laser-machining parameters wiIl be varied. The effect of these variations wiU be

observed for the grooves cut in both solid aluminum plates and in laminated sheets. A

key element of the thesis will be development of procedure to exûact groove-shape data

h m the specimens. An empirical mode1 relating process parameters and the groove

shape will be developed.

Incotning Laser Beam

Focusing

Cuttmg Nozzie

Cut Edge - Adhesive - bwer layer

Molien Iayer

Figure 1.2 Material nmoval mechanism for LOM-AL.

t .6 Thesis Outline

Chapter 2 provides background knowledge related to laser cutting and grooving.

Chapter 3 discusses the experimental equipment and data extraction methodology.

Chapter 4 outlines the experimental results, and Chapter 5 presents their analysis based

on the artificial neural networks. In Chapter 6, conclusions are drawn h m the anaiysis

and recomrnendations are made for future work.

Chapter 2. Laser Material Processing 2.1 Reasons for Using a Laser in LOM-AL Process

Since laser's invention in 1960, laser material pmcessing (UP) ltas become wideIy

accepted in industry due to laser's ability to supply inertia-free energy and allow non-

contact macerial processing. Laser beam provides a hi&-precision, rapid, flexible twl

for weIding, cutting, grooving, driiiing, surface matment of materials. and micro

machining.

Benefits of using a laser for LûM-AL process are explained beluw. A laser has high

beam intensity at a Iocalized area on the workpiece which results in low distortion of the

cut component compared wiîh d t i o n a l machining. There is no direct contact of tools,

which Ieaves the rninimai possibility of contamination. Cutting can be done with very

thin materiaIs as thin as 0.025 mm. Cut edges are relatively smooth and approximateiy

perpendicular to the surface. Because of the n m w kerf widths and heat-affected zones,

patterns can be cut in a small area without affixting precision. The process can be

interfaced with other equipment and can be highly automated Difficult materials cm be

cut, such as hard materials (ceramics) and very soft materials Iike rubber and ioam.

2.2 Laser-Material Interaction

Lasec cutting is accomplished by using a Iens to focus a laser beam on the surface of a

workpiece (Figure 2. L and Figure 22), in order to produce high power density (W/cm2) at

the focal point to either melt or evaporate the material. Laser cutting c m be separated

into the following categories (Petring, 2001): (1) laser fusion cutting, (2) laser oxidation

cutting, (3) laser vaporization cutting, and (4) mixed processes. Laser fusion cutting

converts the kerf (width of the cut) material primarily into the molten state to break up

the material. Laser oxidation cutting pdominantly uses oxygen to initiate exothermal

proçess in the material. Laser vaporïzation cutting delivers suffident energy to mostiy

vaporize the kerf materid, while mixed cutting process can be any combination of the

other categories.

in metals, typical power densities produced by COz and continuous wave (0 Nd:YAG

lasers produce more melting than vaporization. Therefore, an assisting gas used coaxially

or off-axiaiIy with the focused beam becomes a significant mechanism to blow away the

moIten and vaporized material from the kerf while pmtecting the tens from the debris.

The type of gas used depends on the type of rnawrial to be CUL

Pulsed mode lasers are mare usehl in cutting processes. Nd:YAG pulsed lasers generate

enough peak power to vaporize a significant portion of the materid in the focus spot.

This vapor provides much of îhe mechanicd force. With gas pressure not required for

removal of molten material, the roIe of assisting gas becomes less important, its main

function now king the protection of the lens.

Figure 21 Laser cutthg process.

Focusing L m

Vaporized Matcrial or P-

Cross Section of the workpiece

Cut ~ d ~ : L i i d Material Ejected

Moltai Layer a

Figure 2.2 Mechanism of laser (through) cutting.

Laser grooving process, (Figure 2.3), while similar to laser cutting, has some notable

differences. Since the assisting gas cm no longer eject the material h m the kerf bottom,

the material removal process now depends mostly on vaporization. Therefore, the laser

power density required is much higher than in laser cutthg and the assisting gas, once

again, is mainiy employed to protect the optics.

Focusing Lens 1

Cun

Cut Edgc

mg Direction

I

Liquid atma al \ Mohm Layer

Figure 23 Mechanism of laser grwvhg.

2.3 Material-Related Issues

23.1 Metal materiai properties relevant ta laser material processing

The wide range of mateds which the laser can process can be divided into two

categories: metals and non-metais. Given the focus of this thesis on metal cutting, only

the former will be discussed.

Metals commonly processeci by lasers mcIude carbon steels, sîainiess steeIs, aluminum,

copper and its aiioys, and nickeI-based deys- Generally, cornpareci to non-metals,

metais have high thermal diffusivity, melting temperature, vaporization temperature and

opticai reflectivity. Thus, processing metals generally is more dificuit than non-metais.

1t requires high laser power cf-ty to achieve effective materiai removai despite the

unfavorable material propeaies.

2.3.2 Surface W h

Surface reflectivity of a material affects the initial coupling of laser energy during

processing. Figure 2.4 shows reflectivity vs. Iight wavelength at room temperature for

several common metals (ANSUAWS 0 2 1998). Note that most metals are highly

reflective at the wavelengths of the lasers commonly used in LMP. For COz lasers (10.6

p), al1 metals shown are over 90% reflective. For Nd:YAG lasers (1.06 pm), the range

is from 60% for carbon steel to 95% for polished silver. For aluminum and Nd:YAG

laser combination, the reflectivity is about 75%. While the reflectivity affects the initial

coupling of laser energy, it is known that, as the materiai temperature inmases (and as it

changes phase h m solid to Liquid), the refiectivity decreases (Migliore, 1996).

Therefore, metal cutting requires hi& energy density to quickly raise the surface

temperature to the point where reflectivity falls and laser energy coupling improves. An

alternative way to improve energy coupling is to rnodifj the materiai surface, for example

by increasing its roughness or anodizing (for duminum).

Figure 2.4 Reflectivity vs, k r üght wavelength (ANSVAWS C7.2: 1998).

2.4 Laser Processing Panmeters

24.1 Power and power density

Laser output power or rate of energy delivery is expressed in SI units of power, Watt.

Laser beam power per unit area at the surface is commonly hown as the beum p w e r

density or irradiance and measured in w/cm2. A suitable power density must be

delivered to the material surface in order to accompiish a particular laser material-

processing task. Figure 2.5 indicates typicai power densities for different processes.

When through-cutting the materiai, it only needs to be partially vaporized since the assist

gas can eject the material through the kerf. Therefore, relatively lower power density is

sufficient (e-g.. 10'' - 10" w/cm2). Welding, alloying and cladding require even lower

power densities since by their nature the pmcesses require material melting. On the other

hanci, laser machining (and grooving) has higher power density requirments (10~' -

10'~ wlcm2) in order to vapnize the material and thus pmduce clean cuts.

Figure 2.5. Power deasity requirements for various Laser Matenal Processing tasks (Mohanty et al., U101).

2.43 Temporal Mode

Lasers, depending on their design and type, operate in three different temporai modes:

continuous, pulsed and Q-switched. Given the same average power, each mode will

produce a different peak power value. The continuous wave (CW) mode naturally

produces the lowest peak power (Figure 2.6). Since in material processing (and

especially in laser rnachining) high power densities are required, to achieve the desired

results in this mode would require lasers of very high power. Pulsed power delivery can

be achieved by either pulsed mode or Q-switch mode operation.

Given the same laser power, shorter pulses will result in higher peak powers. Q-switched

operation typicdy produces the shorter pulse durations (25-100 ns). Shortening the

pulse duration also has other benefits in LMP. For example, short interaction time rneans

that the heat-affected zone will be smaller.

-.-- Conîinuous Mode - Pulsed Mode .----- Q-Switch Mode

Figure 2.6 Peak power and pulse duration for Continuous, Pulsed, and Q-switched

temporal modes.

2.43 Spatial Mode

Different spatiai modes have different imdiance (power per unit ma) distribution across

the beam diameter, Gaussian mode, called TEhb (Figure 2.7(a)), has the highest beam

intensity concentrated at the center. Its intensity profile is described by:

where I,, is the intensity at the center of the beam, r is the radiai coordinate, and a, is the

Gaussian beam waist. For comparison, TEM12 mode's intensity distribution is shown in

Figure 2.7(b).

The T- mode aiso has minimum divergence, which provides srnailest focd spot size,

For the above reasons, this mode can provide the highest power density compared to

other modes, and therefore is best suited for laser machi~ng applications (Zayhowski,

2001).

(a) (b)

Figure 2.7 (a) Gaussian beam or TE& spatial mode, (b) TEMu spatial mode (Luxon, 2001).

24.4 Puise frequency

When a laser oprates in Q-switched mode, there wiii be a relationship between the puise

frequency and the acceptable maximum speed for quasiçontinuous cutting. If the speed

is too hi@, eventuaily, instead of producing a continuous cut, the laser will make a series

of hoIes. In order to maintain continuous line cut, there has to be a certain minimum

overlap between individuai pulses- The relationship between this speed and the pulse rate

can be estimated as foiiows. For convenience, let us define the overIap ratio (Figure 2.8)

as a = s,/d where so is the overlap distance and d is the beam spot diameter pduced by

each pulse. We negIect the translation of the beam during the pulse time since for Q-

switched laser the pulse time is extremely short. For example, for a LOO us pulse and a

retatively high translation speed of 10,000 mmtmin, the beam will travel only 0.017 W.

Figure 2.8. Rilse-to-pulse overlap.

From Figure 2.8, we can express the between-pulse travel distance as:

AIso, given scan velocity V, laser pulse frequency f, and correspondîng pulse pend T,

we can write:

Equating Eqs. (2.2) and (2.3), we get the expression relating scan velocity, frequency,

and d e p of puIse overlap:

For example, to reach overlap ofa = 0.3 givenJtl0 kHz and d = 15 pm, the scan speed

must ûe 6300 mmimin.

2.45 Poiarization

If the Iaser beam is iineariy potarized, cutting performance in metals will be affected by

the relative orientation of the beam polarization and the direction of beam translation.

Therefore, when consistent kerf width is desireci inespective of the cutting direction,

circularly poIarized beams are used. Circular polarization is achieved by the insertion of

a circdar poIarizer into the beam path (Chryssolouris,l991).

2.5 Beam-Delivery System

Proper design of a beam-delivery system is tequired in order to transfer the beam fiom

the laser to the material surface with minima1 losses and in order to achieve the required

high power density at the focd spot. WhiIe design of the beam-delivery system is not a

part of this thesis, the following background information is provided because of its

importance to achieving good performance in a laser cutting process.

A typical bearn-deiivery system employed in precision laser cutting applications consists

of a beam expander, mirrors, and a focus lens. The beam travels from the laser and is

normally reflected several times before reaching the focus lens. The beam is also

normally expanded (hm 5 CO 10 times) to reduce the focal spot diameter, as will be

explained below.

25.1 Mirrom

Mirrors are used to change the laser-beam direction. They must have coating which

maximizes reflectivity at a specific laser wavelength. They must also be of s ~ c i e n t

diameter to prevent power losses due to part of the beam not king reflected. This is

especially significant after the beam expansion step.

2.52 Focus lem

Selection of highquality and correctiy specified focal lem is essential for obtaining

tightly focused beam spot. It is the most efficient way of achieving high power density.

The theorericd minimum focus spor dimneter, d, can be calculated as follows:

where is the laser beam wavelength, D is the diameter of the unfocused beam, f is the

focal length of the focusing lem.

Incldmt beam widlh

I-"-I

Focusing ûptics

Focus Spot Diamda (Spot Size)

Figure 2.9 Schematic of a focused laser beam.

While on one hand we may want to increase the focai length to put the lens further away

from the material ejected during the cutting, on the oîher han& this would increase the

focai spot diameter. Since the power density changes as a square of the diameter, it will

have a negative effect on the power density.

Another important characteristic of the beam-delivery system is the depth of focus. It is

the distance range in the direction of beam propagation within which the power density is

greater than 90% of its peak value at the FocaI point, Figure 2.9. Depth of focus, 2, is

given b y

if the depth of focus is too smaii, there may be a problem with maintabhg the material

surface within its limits due to surface irregularities or due to misaligrnent between the

materials surface and the plane of translation of the beam (if moving-system is used) or

of the workpiece. On the other hmd, increasing the depth of focus by increasing the

focal length wili increase the spot diameter. T'hm, a trade-off between these two

requirements is required.

2.53 Beam expander

From Eq. (2.5) we can see that increasing the incident beam diameter D can decrease the

focal spot diameter. A beam expander is used for this purpose (Figure 2.10). It consists

of two lenses: the expanding lens to increase the beam diameter and another lem to

collimate the expanded beam to a constant beam diameter. The degree of expansion may

be h m two to ten times that of the incident beam diameter Di. it is not advisable to

design a beam expander with the expansioa ratio more than ten times as the optical

quality of the beam deteriorates dramaticatly beyond this point.

2.6 Theoretical Analysis of Laser Cutting

Chryssolouris [1991] has developed general laser machinhg processes model. This

model is developed based on heat transfa's balance equation at the constant erosion fiont

(the constant laser beam effective spot on the work piece, see Figure 2.11) and

calculations of material state, such as the energy required to convert the materiai state

h m soiid to iiquid and vapor.

Figure 2.10. A beam expander.

Figure 2.11 Location of the erosion front of laser grooving.

The generalized equation of laser grooving predicts the gmve depth S by:

w here:

p is the materiai density

a is material absorptivity (hm O to 1)

C, is specific heat

dis the laser focused spot diamem

L is the Iatent heat

P is power

T, is the m m temperature

Ts is the materiai vaporization or melting temperature, depending on the assurned

mechanism of materiai removal

Vis the scanning speed

The rdationship implies that the grooving depth S is directly proportional to the input

power P, and inversely proportional to the scanning speed V.

A number of assumptions have been made in order to develop the mode1 in Eq. (2.7):

1) The beam is moving at a constant velocity V.

2) The material is isotropie with constant properties, such as themal

conductivity k, specific heat C,, and density p.

3) The material is opaque; the absorptivity is assumed to be constant, as the

absorptivity behavior in evaporation phase is not well known.

4) Change of phase of the material from solid to vapor occurs in one step at a

single evaporation temperature.

5) The evaporated material or plasma does not interfere with the laser b a r n

reaching the erosion front,

6) Multiple reflections of laser beam within the p v e are neglected.

7) Laser beam spot size is constant on the surface of materiai.

2.7 Summary

This chapter provided the background information required for understanding of the laser

materiai processing. It described the advantages of using a laser for the LOM-AL

process, such as ability to cut thin material cleanly without physicd contact. Given the

properties of metals, it was emphasized how important it is to achieve high power density

for successN cutting of metals. Laser operating parameters such as power, spatial and

temporal modes, and pulse frequency were explaïneci. The importance of correct design

of the bem-ddivery system was emphasized given the trade-off existing between

achieving minimum focai spot and acceptable depth of focused. Finaliy, a simple closed-

fonn expression for predicting the law-cut p v e depth has been presented.

The following chapter will &al with the main focus of this thesis, the experiments

designed to determine appropriate operating parameters for laser cutting of aluminum

sheet in LOM-AL process.

Chapter 3. Laser-Cutting Experiments: Preparation and Data-Extraction Methods

To accomplish the main objective of this research, i.e. determination of suitable process

parameters for controlled-depth laser cutting of duminurn sheet, a number of experiments

were conducted The two prirnary controllable process parameters are laser power and

scan speed. Secondary parameters include type of gas assist and gas pressure, nuber of

passes made by laser beam, laser pulse hquency, and focal point height with respect to

material surface,

The primary observed parameter is the groove &ph. The primary interest in p v e

depth is due to the need to cut the sheet laminations precisely to the depth of one sheet

without significant damage to the underl ying stack. Secondary observed parme tes are

the kerf widih, groove cross-section profile shape, and the amount of recast material

found in the groove.

A second objective of the experiments was to determine the effect of changing the

rnaterial king cut h m a (reIatively) tfück solid aluminum plate (3.2 mm thick) CO a stack

of thin laminated sheets (sheet thickness of 0.12 mm). Laser p v i n g is normdly

performed on relatively thick materid and thus theoretical models would norrnally also

assume thick substrates. On the other hanci, the material to be cut in our process consists

of a sandwich of polymeric adhesive and aiuminum sheec. Aluminum is an excellent heat

conductor (tfiermai conductivity 210 Wlm C) while a typicd adhesive acts as an

insuiator, with very iow conductivity (0.2-0.5 Wlm C). Thus, the solid plate represents a

known "contrai" case while the Iaminated sheet spechens approximate the special

material combination to be encountered in the laminated fabrication pcocess.

3.1 Specimen Preparation

Two types of specimens were useci: soiid plates and Iaminaied sheets bonded to pIates

(Figure 3.1). Henceforth, ihey wiii be identifieci as SP and LS specimens, respectively.

The piates were prepared by cutting 2"x1/8" (50.8 mmx32 mm) aiuminum aiioy bar

stock (ASTM B221 6061 T6511) into 95 mm lengths. These plates were used either

directly or as supports for three larninated sheets bonded to them. The latter were cut

h m a mUed aluminum sheet stock (6061,0.05" or 0.12 mm gauge) into 90 mm by 20

mm strips. Figure 3.1 shows the overall specirnen dimensions as well as the location and

dimensions of the cut lines on the specimens. The cut Lines can be identified in the

figure as four groups of seven parallel vertical lines. Additionally, a speciai specimen

was prepared as an aid in determinhg the approximate range of suitable scan speeds for

sheet cutting. This specimen (Figure 32) consisted of the alurninwn plate with t h e

slots, with a single alurninurn sheet bonded to it so as to cover the slots. Figure 3.3

shows the laminated-sheet and the three-slot specimens pnor to the experiments.

Figure 3.1 Dimensions of the cut locations on specimeas, the alurninum soiid plate (SP) (left) and the laminated sheet (LS) specimen (right) (di dimensions in mm).

To fabricate the laminated-sheet specimens, attention was given to surface preparation to

assure good adhesion. Specimens were prepared by following steps:

(1) To improve adhesive, the aluminum bar stock for the plates was sanded first with 60

coarse, aluminum oxide, C-weight paper, foUowed by 220-Mt, Sic, B-weight paper.

(2) The aluminum sheet was cut into 90 mm by 20 mm strips by using cutter guided by

a steel d e r .

Figure 3 3 Tbe tb~eeslot speciman (dimensions in mm).

(3) To ensure consistent surface quaiity between specimens and to remove

contamination such as fingerprints and human oil, the sheets and plates were

chemicalIy etched with caustic solution before bonding sheets ont0 the support

plates. See Appendix A for detailed procedure.

(4) The sheet strips were bonded to the plates using a film adhesive, scotchTM 467 MP

h m 3M. This film consists of 200 MP high-performance acryiic adhesive Iayer (50

pm thick) attached to a polycoated Kraft paper üner. After attaching each new sheet

layer with the adhesive film @y applying finger pressure), the adhesive was cured by

applying heat and pressure uing a heat d e r apparatus (THEUERTM Mode1 PC

h m H. W. Theiler, Inc. Petaluma, Ca). The operating conditions are shown in

ThIe 3-1.

Table 3.1 S d o g temperature and pressure

Upper die temp set

Lower die temp set

150°C

1WC

Figure 3 3 Three-dot (foreground) and tke-laminated sheet specimens prior to the cutthg experiments.

3.2 Experimental Apparatus

The experimentai setup consisted of a laser light source, a beam delivery system, and a

motion systemf . Figure 3.4 shows an overail view of the senip, while Figure 3.5 shows

the close-up of the specimen, the focus lens, and the gas-assist nozzle. The platform

carrying the specimen moved in X-Y plane (with accuracy of Il pm) while the optical

system remained stationary. The Iight source was a GSI Lumonics Sigma-400 diode-

pumped Nd:YLF laser operating at a wavelength of 1054 nm (near infrared), capable of

up to 12W average power and (Q-switched) puise frequency up to 20 kHz.

The optical path h m the laser to the specimen surface consisted of five-time beam

expanderlcoIlimator, three mirmrs, and a 55-mm apochromat triplet lem. Given the light

wavelength, initial beam diameter (1.3 mm), beam expansion ratio, and the focal length,

the theoretical focal spot diameter is estimateci by Eq. (2.5) to be 11.3 p. The depth of

focus is estimated using Eq. (2.6) to be only 112 ~ c m , which implies that care must be

taken to make sure the specïmen surface is parailel with the plane of X-Y motion and that

the beam is properly focused. Assuming actual focal spot diameter to be approximately

15 pn (larger diameter is expected due to optical aberrations of the non-ideal focusing

lens), the power density at the specimen's surface during a single pulse can be estimated

to range from 2.3 to 6.8~10' w/crni? for laser power outputs of 4W to 12W and pulse

irequency of 10 kHz used in the experiments. Therefore, our experimental setup was

capable of producing sufficient power &nsity to vaporize the aluminum.

Figure 3A The qerlmcntai setup.

55-mm Focus Lens

, Specimen were king

cut

Platfonn of X-Y Motion System

- - - - - - - - - - -- - - ~

Figure 3.5 Close-up view of the X-Y plathm and focus fens.

3.3 Procedure

The experiments can be separated into three parts:

(1) Preüminary tests on three-slot specimens.

(2) Primary-process-parameter tests comparing sotid-plate and laminated-sheet

specimens.

(3) Secondary-pmess-parameter tests on laminated-sheet specimens.

33.1 Pdiminary tests

Since it was not known a M o n what p e s s parameter combination (power and speed)

wiii result in cutting approximately to the depth of one sheet, a p r e b a r y test was

conducteci before proceeding with the main set of experiments. Because this test was

conducted during the same session as the rest of the ex-nts, a way of determinhg

when the cut-through occuned had to be found which reIied on unaided observation. By

cutting the sheet attached to the th - s lo t specimen, we were able to observe at what

speed the cut through occwred by Iooking ihrough the slots, and observing for which

scan iine the iight could pass through the sheet. 12 scan lines were made in total, with the

speeds ranging from 1000 to 12000 mmlmin, incremented by either 500 or

1000 mm/min. Four lines were scanned for each sbt. Details of line locations are given

in Appendix B. A middle power setting of 7 W was used. The cut through was detected

at 3000 mrnlmin based on which a more refined range of scan speeds was set for the

following experiments.

3.3.2 Primary-Pmcess-Parameter Tests

Al1 of the following tests were conducted for both SP and LS specimens. For each

specimen, four groups of seven paraliel 3.7-cm lines were scanned, with two groups

scanned for each power setting (Figure 3.6). The scan directions in the two groups were

reversed with the intention to ascertain absence of any directional effects. (Due CO

experimentai emr, both sets of lines for 4 and 7W on SP specirnens were drawn in the

same direction. Since no significant difference between directions was observed for the

two higher powers of SP set, this ermr is not expected to affect the results.) Within each

group, the scan speed increased from right to left, as viewed in the figure, from 1500 to

4500 in 500 mm/min increments. Four power sertings were used: 4,7,10, and 12W. No

gas assist was used in these tests.

Table 3 3 Common laser settings (used in di experiments, unless noteâ otherwise).

--- - - -

Temporai Mode

Spatial Mode

~ u t y Cycle

Q-Switch

Tm10 82.4 96

Figure 3.6 Geometiy of specimens and line sains (same üne pattern was used for all specimens except for three-slot pceüminary tesîs).

Table 3 3 Experimental settings for primary proces parameter tests

3.3.3 Secondnry-ProcesParameter Tests

The secondary pmess parameters tested were:

Powers (W)

Scan S p d s (mdmin)

(1) Gas assist: type and pressure

(2) Laser puise frequency

(3) Defocused cutting

(4) MdtipIe passes

4,7, 10, 12

15ûû,2ûûû,25ûû,3000,3500,4000,4500

Al1 the experiments in this part were conducted on laminated-sheet @mens. Same cut

paths and speed ranges as for the primary-process-parameter tests, uniess noted

otherwise.

GusAssist

Oxygen is often used in Iaser cutting to increase the cutting rate. On the other han4

nitmgen is used as an inert shieIding gas. In both cases, it was also expected that the

assist gases would affect the rate of matenal removal h m the kerf. A range of gas

pressure settings (Table 3.4) was used (as controlled by the supply cyiinder valve and set

by the valve's dia1 indicator). Note ttiat in order to employ gas assist, a nozzle was

attached to the facusing lem. The node is of a conical design which provides the gas

flow CO-axially with the direction of Iaser beam propagation (Figure 3.7). Based on the

previous expcrience with this equiprnen& addition of the node may rcduce slightly the

Iaser power delivered at the surface. Therefore, zero-pressure (Le. no gas assist) tests

with the nozzle attached were included as a conml.

Figure 3.7 Nozzle for ossist gases.

' Verbai commuuication h m iht techician assisihg wiîh the expcrimnis at the Laser Micromaching

Fdty.

Table 3.4 Experimental settings for gas-assisted tests.

Laser pulse fiequency

Higher puise Erequencies were expected to produce smoother cuts. The laser is rated up to

a maximum frequency of 20 kHz. However, for the laser type used in the experirnents,

increasing the frequency means that the energy delivered per pulse drops. Therefore,

only marginally higher maximum frequency of 12 lcHz was attempted here. Lower

Frequencies, on the other hand, gave higher energy per pulse. Table 3.5 lis& the

frequencies and the actual energies per pulse noted during the experiments. The duty

cycle of less ihan 100% was used to prolong the life of the laser's crystal r d The laser

operated on the 2-ms period duing which the pulses were on for the percent of tirne

recorded as the duty cycle in the table.

Table 3 5 Dierent Pulse frequency

l Puise frequency tests I Puise muency (IdIz)

12

10

8

6

Power

Watts 7.0

7.1

7.0

6.8

Pulse Energy

PJ 582.6

590.8

874.1

1139.9

Duty Cycle

% 79.7

79.7

85.5

88.7

in some laser material processing applications, better resu1t.s are achieved when the focal

plane of the lens is moved away b r n klow the material surface. Also, by varying the

surface height, we could ascertain the process Iimitations due to the deph of focw. Thus,

cuts were ma& at the normal range of speeds h m 1500 to 4500 d m i n , with the fmal

plane moved bdow incrementally h m the sheet's top to its bottom surface. Table 3.6

shows the setting used

Table 3.6 Defocused Tests.

M w d Tests 1 Focusing distance (mm)

Puise Frequency

Multiple passes

O (facused), 0.04,0.08,0.12

10 kHz

Power

Duty cycle -

By reducing the depth of cut during a single pass, it is expected that better cut quality cm

be obtaind These tests intended to determine whether passing the laser beam over the

s a m path more than once can improve the cut quality. Two sets of tests, each comparing

single-pass to two-pass cuts were perforrned (Table 3.7). In the first set , the control case

(single pass, 7 W, speeds ranging from 1500 to 4500 mm/min) was compared against a

two-pas test. However, to achieve simiiar over ail cut depth in both cases, the cutting

speeds in the two-pas test were doubIed (3ûûû-9000 mm/min). The second set

compared the above control case against a different two-pass case (3.5 W, 1500-4500

mdmin). Here; the idea was to achieve reduced cut deph in each p a s by reducing the

laser power.

7 W

82.4 %

Table 3.7 Experimentai settinp for multiple-pass tests.

Power No. of Sm-

3.4 Specimen Examination

Several data collection methods were iried before a suitable one was found The

difficulty of data co[lection is due to the smali size of kerf width (typically about 40 ym),

fragility of the larninated sheet specimens, and presence of debris and mast material

within the cut groove. Three methods investigated were a scanning etecmn rnicroscopy

(SEM), profilometry, and metaiIography.

3.4.1 Seanning Electron Microscope (SEN)

Scanning electron microscopes use electrons bombarding the surface of specimen in a

vacuum chamber to obîain surface image similar to that obtained from optical

microscope, When used for kerf examination, the benefits of SEM are that it is a non-

destructive technique and that the data cm be obtained comparativeIy quickly. A general

major benefit of SEM vs. opticd microscope is a much greater depth of focus for SEM

(given the same magnification). This factor is particularly significant for examining Iaser-

cut grooves, which present a surface with significant height vaciability. SEM was used to

examine the surfaces of laser-£ut specimens (Figure 3.8(a)). The images provide useful

information about the ken quality in terms of its uniformity and the amount of recast

material ejected from the kerf. However, even smaii amount of recast near the top of the

groove obscured the groove bottom and w d s fmm view, making judgment of the cut

quality and depth impossible. Figure 3.8(b) shows a metallographicaI section of the

SEM-imaged kerf. Clearly, the image conveys signincantly more informacion about the

cut profiles*

Figure 3.8 (a) SEM image ami (b) metallographic image.

3.4.2 Protiiomeîry (Vertical Seanniag InterPeronietry)

The WYKO profilometry (a commm*ai process) is a non-contact optical meihod for

exarnining surface contour of specimens. It employs vertical scanning interferorneter

technique in which a white light beam is used to examine the surface profile. The beam

passes through a beam spiitter to reflect half of the incident beam to the reference surface.

Another haIf of the beam is reflected h m the sampIe surface to the beam spiitter. The

two beams recombine at the beam spiitter to fom interference fringes. These altemating

Iight-and-dark bands can be observeci and counted to calculate the height of the sample

surface. See an example in Figure 3.9 of an image obtained h m WYKO profilometry

h m one of the eariy experimental trials (individual pits were produceci by laser beam

due to the combination of a 50% duty cycle and fast scan speeds.).

The benefits of profilometry are th@ it is nondesmctive and that data on kerf profile can

be obtained quickly. However, the Limitations are that (1) similar to SEM, the groove is

obscured if it is f W with &bris and recast; (2) aeady vertical groove walls and uneven

surface scatter the white iight and thus prevmt ttit intederence fringes h m fonning. For

example, the black regions occupying large part of the image in Figure 3.9 represent

areas where the &vice could not obtain valid data Incompleteness of the groove cross-

section pfile can be also seen in îhe sectional representation of the depth &îa h m

Figure 3.8 shown in Eigure 3.10. The section is Iocated dong the horizontal cross-hair in

Figure 3.9.

0.0 30.0 6û.O 90.0 120.0 152.9

Figure 3.9 Image captured by WYKO pmfdometry.

* Cut . Depth

Figure 3.10 Cut cross-section data from WYKO pmfilometry.

Metaliographic examination normally involves placing the specimen to be examined in a

mould so the specimen can be encased in epoxy (hard polymeric matenai). Afier

hardening, the moulded piece containing the specimen is cut and the cut surface is

polished to a mirror-like finish. images of this cross-section can then be taken. While

metallography is a destructive measurement method, it was the only method found to

reliably produce the desired data.

Note that unlike a more common case, where the specimen to be examined is a

mechanicaily homogeneous solid, our laminated-sheet specimens consist of a six-Iayer

sandwich, with three layers of duminurn sheet and three Iayers of polymeric adhesive, ail

bnded to a 3.2-mm aiuminum plate. The relative softness of adhesive layers requires

extra care when attempting to perfom metailographic sectioning. If the aluminum sheet

is not held rigidly by the epoxy, shifting may occur, distorting the original shape of the

cut groove profile.

Thus, to make sure that the sheet is completeIy irnmobilized, a special two-stage

mounting process was employed. See Appendix C for detailed description of the

procedure. In the fmt stage, one 25-mm segment was cut away from each end of the

specimen. Figure 3.6 shows the cut locations by dashed lines. Each segment was piaced

in a rectanguia. rubber mould (83. x 53W x 30H mm) (Figure 3.1 l(a)) and set in epoxy

(Araidite GYS02). Once hardened (after 24 hours), the block was further cut into two

pieces, each containing one line p u p (cut lines indicated by clashes in Figure 3.1 l(a)).

Before the second stage, the pieces were reshaped by grinding in order to fit two of them

into a one-inch diameter cylindrical mould (Figure 3.11 (b)). Photos of specimens at

various stages are shown in Figure 3.12.

Figure 3.11 (a) Specimen segment e n d in epoxy dter the first mounting stage; top and front view shown; (b) Two specbn segments piaced in a cylindrical mould

in the second munting stage - top view.

After demodding of the second stage, polishing was performed in three steps: (1) rough

grinding with 240-grit SIC abrasive paper to remove sawing damage; (2) intemediate

polish (to 3 p) with 600-grit paper, and (3) final colloidal-silica wet (0.5 pm) polish

(Figure 3.13). GrayYscale images (256-level, 975 x 975 pixels) of polished sections were

captured through an optical microscope and used subsequentiy to measure the groove

crosssection features. These images achieved 0.25 pmlpixel resolution.

Figure 3.12 Photo shows the renrnining p h of (la) solid-plrite aud (lb) hminated- sheet specimens after segment n m o d for examination; (2) specimen after first

mounting stage; (3) specimen after second mounting stage.

Figure 3.13 The hished poüshed spechn by metallographical method.

3.5 Summary

Chapter 3 presented (1) tfie procedure for specimen preparation, (2) the experimental

equipment used, (3) procedure and (4) îhe metallographic method for obtaining the cross-

section profiles of the laser CU&. Next chapter wilI present the experimental mulis,

discussion, and the cornparison of the existing theoretical mode1 with the exprimenial

results*

4. Laser-Cutting Experiments: Results and Analysis

hages of laser-cut p v e cross-sections aiiowed coilection of detailed and accurate data

reportai fierein about the p v e geometry and cut quality- The chapter will Fmt define

the observeci parameters of the p v e s and then will present the data collected. The

tesults are arranged in order starting with the preliminary tests, foilowed by those

focusing on the primary pmess parmeters (laser power and scan speed), and finally the

result from the secondary-pmcess-parameter tests.

4.1 Observed ïeatures of lasereut gtoove sections

The data ptesented here were coiiected from 256-level gray-sale images captured

through an optical microscope. Tci aid with measurement of the cut features, Scion

image image-pmcessing soFtware (Scion Corp.) was used. Two sample section images

are shown in Figure 4.1. Featwe measurements wen made by sening the image-to-world

conversion factor using the scaie rnarker and by utiiizing the software's ability to

automatically record mouse click coordinates and calculate distances between these

points. For area measurements, the software calculated atea of a region enclosed by a

Whand-drawn curve.

Three geometrïc and two quality features were observecl. The geometric features are: cut

depth, ked width, and groove profile shape. The quality features are recast depth and

recast area fraction. The Figure 4.1 defines some of these features. Cur depth is the

de@ of materiai removed by the laser, nieasured from the material's top surface. Kerf

width is the width of the laser cut measured at the materiai's top surface. When the

groove is recast-filied, the boundary between the bulk materiai and recast could usually

be visuaiiy identifie& This boundiuy was then projected to the intersection with the top

surface. Thus, when not sharply defined, the kerf width was measured as the distance

betuleen the two pjected intersections of the cut's boundaryaarv

The third geometric featufe, g m v e proifle shupe pmmneterl characterizes the shape of

the groove section p f i k m e . The measurement appiies only to partial (not h u g h )

cuts. This measurement is based on the proposition introduced herein that the shape of

the groove section profile can be well appmximated by a power function of the general

fom:

where x and y are the coordinates of the profile curve. The parameter measured is B. the

exponent in Eq. (4.1). Details of profile shape data extraction are presented in Section

4.2 beIow.

Cut Width

Figure 4.1 Definition of moasured iaser cut feahws: (a) sheet cut, 12 W, 2000 mmlmin, (b) sheet cut, 7 W, 2ûûû d m i n .

Tu quanti@ the quality of the cut, we measured the arnount of recast found in the groove.

These measurements were done in two ways. Recast &pth is defined as the height of the

recast "bridge" joining the two walls of the cut groove (Figure 4.l(a)). if there was no

bridge forme& the recast depth was recorded as zero. Recast area jhcrion is defined as

t4e percent fraction of the groove cross-section area occupied by recast-

--

4.2 Extncting lasercut grwverhape data

It is proposed herein that the shape of the groove profile can be well approxirnated by a

power function in Eq. (4.1). Such shape data can be utilized in laser machining

applications where the shape of the groove walls may be of significance. Note that

Eq. (4.1) is not valid for xcû for some vaIues of B (Figure 4.2). Therefore, the actuai

equation used is:

However, for notational simplicity, the fom in Eq. (4.1) will be used henceforth.

- 1 . b -1.000 -0.500 O. 0.500 1 .O00 1.: W

Unit Korf WUth ktk

Figare 4 3 Effect of pmfiie shape parameter B.

Figure 4.2 shows the effect of varying the exponent B in equation (4.2) on the curve

shape. Oniy shapes for B > 1 are physicaiiy possible given the Iaser machinhg process

limitations. When B is less than or qua1 to one, a sharp peak (a first-âerivative

discontinuity) exists at O.

We would like to rewrite the general function of Equation (4.1) in terms of the cut

profile's overall dimensions:

p - half-width of the cut at the material's surface (half-width as opposed to full width was

chosen for computational convenience);

q - depth of the cut profile.

Figure 4 3 Laserat groove profile with p and q panmeten detined.

Let the origin of the curve coordinate system &-YC be located verticaiiy at the surface of

the cut matenal and horizontally at the plane positioned midway between the groove's

edges (Le., at the profile's plane of symmetry) Figure 4.3. Given the above definition of

the cwe's origin, its equation can be expressed in terms ofp and q as:

Parameters p and q are directly reIated CO kerf width and cut depth, respectively. To

estimate B, the points sampled fmm the cuwe's profile will be fitted to the power curve in

Eq. (4.3). To simpliQ this process and to visuaily isolate the curve's shape characteristic

from its overall dimensions (height and width), we will transfomi the raw data obtained

h m the image as follows.

k t us assume that the coordinates of the sampled points on the curve profile have been

transformed h m the image coordinates O[t, YI) (expressed in pixels) to the coordinate

system &Y,. Thus, given (x,, yJ pairs in this latter cwrdinate system, we can

nonnaiize x and y by dividing by p and q, respectively. This step makes al1 curves have

the sarne unit height and half-width.

x=L and Y=- Y , P 4

Substituting forx, and y, in Eq. (43 , we get:

Now B can be caicuiated by a Iinear fit to the Iogarithms of X and (Y+l):

Then, B is the slope of the line obtained by a least-squares fit to the Iog-log plot of the

data. Note that we must use the absolute value of X as the argument of log function in

Eq. (4.6) in order for expression to be valid for negatîve X. Therefore, through the above

derivation, we have shown how a lasef-cut groove profile geometry cm be represented by

three parameters: p, q, and B.

To obtain the pfiIe shape parameter 8, a set of &ta point cwrdinates were sampled

dong the profile curve. Mer sampting 30 to 50 points, (1) the x and y values were

normalized by p and q, respectively, and (2) the power parameter B was esàmated by a

linear fit to the log-log plot of the coordinates. The points for each cut profile were

sampled hm times for each image, and B was calculateci each time and averaged. See

Appendix D for &tails of data collection.

Figure 4.4 shows an example of a typical plot for the sampled cuve points together with

the Iinear fits. The same points are plotted in Figure 4.5 in x-y cwrdinates together with

the modeI's fitted curve (B = 2.4).

1 x tek 1 O right I

I - - -Linear (leit) - - - Linear (right) 1

y (Right) = 3.5079~ + 0.04

Figure 4.4 An exemplary plot of sampled profile cuve points used to estimate groove profile sbape B. "Left" and "Rght" d e r to the correspondhg halves of the

cu171ets two branches,

Figure 4.5 Sampled cut profde points together with power function shape mode1

fitted to them (B = 2.4).

4.3 Assessrnent of Measurement Repeatability

One cut profile (for speed 3500 -n and power 10 W) was selected to evaluate the

repeatability of our method of measuring the cut profile parameters (kerf half-width p, cut

depth q, and stiape B). To evaluate repeatability, this profile was measured six times.

The resuits are shown in Table 4.1. They indicate that the uncertainty in the p v e ' s

dimensions (based on one standard deviation) is about 48, while for the shape parameter

B, it is 8.6%.

Table 4.1 Messurement Repeatabüity Results. -

Cut h f ü e Parameter

Kerf ùaif-width p (CM)

Cut depth q (pu)

Shape B

Average

24.7

142.7

3.5

Standard Deviation

1 .O

5.8

0.3

4.4 Preliminary Tests: Through Cutting of Aluminum Sheet

While the main reason for performing these tests was to find the nanower range of scan

speeds for fuaher testing, these specimens were aiso exarnined via metdlographic

procedure. They provide us with data for the situation experienced in through-cutting,

where sheet is not attached to a substrate. Figure 4.6 shows examples of profiles

observed for higher (a) and Iower (b) speed ranges.

I

Figure 4.6 Profiies from prelllninary tests: (a) 8000 d m i n scan speed, (b) 4000 mmlmin.

Plotting cut depth vs. scan speed (Figure 4.7) shows îhat above certain threshold speed

(approximately 6000 rnmhin) the cut depth remains constant oust under 40 w). Kerf

width (Figure 4.8) increases with speed h m 15 to 40 ~ i m , with the rate of increase

falling above 6000 mmlmin,

Figure 4.7 Cut depth for preiiminary tests.

Figure 4 8 Ker€ width for preüminary tests.

1

100; L

To help with explanation of the change in groove shape vs, speed relationship above

6000 mmlmin, it is instructive to caIcuIate pulse-to-pulse spatial separation. Given the

Iaser pulse frequency of 10 kHz and the speed, at 6ûûû d m i n , the pulse-to-pulse

distance is 10 p. It increases to 20 pn at 12000 mmlmin. Considering that the

minimum kerf width observed was 15 pn and the theoreticai minimum focal spot

3000 5000 7000 9000 11000

Sein Speeds (m mlmln)

- I 6 Exp. Conditions

" 7W,10kHz E 80- - $ - 60- E Y

Q 40- 4

8 - ; 20-- U

O 1

I

diameter is 11.3 pm, it is clear that the acnial spot diameter will be somewhere between

11 and 15 p. Therefore, above 6000 mm/min the lack of overlap between pulses

changes the operating mode from quasi-continuous to discrete-pulse.

Based on the preliminary tests, the middle of the speed range for M e r testing was set at

3000 mmlmin, and the range of speed variation to I 1500 mm/min.

4.5 Primary-Process-Parameter Tests

In performing these tests we sought to isolate the effect of primary process parameters,

laser power and scan speed For this reason, no gas assist was used. Also, a cornparison

between cutting in soiid plate and in laminated sheets was made.

4.5.1 Soiid-Plate (SP) Specimens

Figure 4.9 shows two examples of groove cross-sections, one with significant recast

present and another showing a successfuI (clean) groove. Both figures show upward flow

pattern which developed as the vaporized and molten metal escaped vertically. Recast

occurs when molten and vaporized metal resolidifies upon coming in contact with the

relatively cool waiis of the groove. Under certain process conditions, there is enough

resoiidification to fonn a "bridge" a m the groove.

(a) (6) F i 4 3 Examples of gmove cross-sections for soiid-plate specUnCm: (a) recast-

tilïed W v e (12W, 2ûûû d m i n ) ; (b) c1ea11 pave (12W, 4Sûû mmlmin).

Figure 4.10 shows cut depths averaged h m two observations, each made for a different

scan line on îhe same specimen. (For each speed and power combination, two scan Lines

were made). As expected, the cut depth in SP specimens decreases with scan speed and

increases with power. A theoretical mode1 curves also show wiii be explaineci in Section

4.7.

Scln -rd (mmlmln)

Figure 4.10 Cut depth (expriment and madel) for solid-plate specimens.

The kerf width (Egure 4.11) increases progressively with power and speed, h m a

minimum of about 12 pm at 4 W to a maximum 42 pm at 12 W.

O 1000 2000 3000 4000 5000

Cutting Speeds (mdmin)

Figure 4.11 Kerf widtb for soiid-plate specimens.

The groove profile shape parameter B decreases with increasing speed and is not affected

significantly by the laser power (Etgure 4.12): the range of variations due to power is well

within two standard deviations of the measurement enor ( W . 3 , as reported in Section

4.3). Higher B values mean the profile shape becornes more "square" and less

"triangulai' (Figure 4.2) with the decrease in cutting speed.

Groove quality was evaluated by measurïng the recast depth (Figure 4.13) and area

fraction (Figure 4.14). Botfi parameters follow a similar trend of decrease with speed.

The recast @th can dso be seen to depend on power, lower powers producing smailer

values. However, when using normalized data (recast area fraction), the relationship

appears to be independent of power, Without normaiizing, Iower powers produce less

deep cuts and therefore smaiter recast depths.

Figure 4.12 Profde shape parameter B for soüd-plate specimens.

Figure 4.13 Recast depth for solid-plate specimens.

0.0 4 1 1 0 0 0 1 5 0 0 2 0 0 0 2 5 0 0 3 0 0 0 3 5 0 0 4 0 0 0 4 5 0 0 5 0 0 0

S a n Speed (mrnlmin)

Figure 4.14 Recast area fraction for solid-plate specimens.

45.2 Laminateci-sheet (LS) Spechns

For TS specimens, some combinations of speed and power resulted in successful through

cutting of the aiurninurn sheet (Figure 4.15(a)) while others produced grooves, similar to

the SP specimens (Figure 4.15(b)).

The trends for the cut depth values (Figure 4.16) are similar to those for the SP

specimens, except for the upper limit of 120 prn (sheet thickness). The kerf widths for

the LS specimens, on the other han& exhibit different mnds compareci to SP specimens

(Figure 4.17). Fmt, there is less significant effect of speed. Second, while the width for

powers from 7 to 12 W increases with power similarly to the SP specimens, at 4 W, the

kerf width for LS specimens nearly quai those at 12 W. It is possible that the different

heat-transfer-related characteristics caused by the laminateci structure of LS specimens

have the most significant effect at the lowest power.

The cut quality is of paaicular significance for the LS specimens because these

specimens are meant to simulate processing environment of LOM-AL. Recast area

fraction displays some of the m d s of the SP specimens, but aiso has some different

features (Figure 4.18). In a trend simiiar to that for the SP specimens, for speeds above

2000 mm/min, the recast area decreases with speed. One important difference, however,

is a signifiant decrease in recast formation at 1500 &min for the hvo highest powers

(10 and 12W). These are also the powers and speeds at which a through-cut was

observed (see Figure 4.15(a)). It is conjectured that at these higher powers and lower

speeds, sufficient energy is detivered through the cut groove to the adhesive below to

increase the pressure from vaporized material to the point where the materiai is removed

h m the p v e before it is able to resoiidify.

Figure 4.15 Examples of p v e cnws.Sections for laminated-sheet specimens: (a) through-cut sheet (12W, 1500 d m i n ) ; (ô) recast-free gmove (12W, 4500

mmlrnin).

Scan Speed (mm/min)

Figure 4.16 Cut depths for larninated-sheet specimens.

Scrn Sperd (mmlmin)

Figure 4.17 Kerf width for iaminated-sheet specimens.

Figure 4.18 Recast ares tmction for laminated-sheet specimens.

4.5.3 Cornparison of m l t s for SP and LS specimem

Figure 4-19 shows the ratio of cut deprhs for laminateci sheet over those for soiid plate.

For example, values of this ratio l e s than one wodd indicate that, for equivaient power

and speed parameters, shaiiower grooves were produced in the laminated-sheet

specimens. It is interesting to observe that there is a consistent decline in this ratio with

speed for the lowest power (4W), from 0.79 to 0.44, but, for al1 other (higher) powers, the

ratio increases linearly with speed reaching nearly unity at the highest speeds. Also, at

the lowest speeds (1500 and 2000 mmfmin), laser power does not appear to affect the

ratio of cut depths. Unexpected results for cut &phs at 4 W for LS specimens conf ï ï

earlier unexpected observations for kerf widths (Figure 4.17)

Figure 4.19 Ratio of laminated sheet depths over soiid plate deptbs.

4.5.4 Discussion of recast-formation mechanisni

To understand the pmcess of recast formation observed in our experiments, we will

describe the sequence of events which happens, according to our understanding, when the

hi&-intensity laser beam interacts with the metal surface. Note first that the duration of

each pulse is very short, about 100 ns, while the time between pulses is much longer

(1000 times longer or 100 ps at 10 kHz). As the energy is absorbed, the local

temperature rises very quickly, with minimum of heat king Iost due to conduction,

convection or radiation because of the very short interaction time. As more energy is

delivered to the surface, metal quicWy melts and then vaponzes (or sublimates directly to

vapour). Given that no assist gas was used in these experimenis, the only mechanism for

molten metd removal from the gmve is the pressure of the vaporized material itseif.

As the material phase changes to vapour, its volume increases very rapidly (Figure 4.20).

This hi@-temperiinire plasma (which is an ionized macerial in gaseous form) becomes

the major mechanicd force removing the material h m the cut. As the hole depth

increases, the plasma has to travel m e r dong the channel finmed by the hoie walis. As

the gas makes contact with the waiis on the way out, it cools rapidly since there is a large

difference between the gas temperature and the walls. Assuming the waUs are near the

melting point, it would be about 1800°C. Thus, re-condensation and re-solidification

OCcUi:

The resolidified layer continues to grow as the hole becomes deeper. If the laser were

stationary then this growth could onIy extend until the beam impacts ont0 the newly

forming "bridge" structure. However, in our case the beam is moving. The movement

occurring over the duration of the pulse itself is very small. At the maximum speed of

4500 &min, and assuming pulse duration of 150 ns, it is approximately 10 nm, which

is a very smaii fraction of 10-20 pn beam spot diameter. However, subsequent pulses

will overlap to some extent, depending on the scan speed Thus, since we are actually

cutting a groove, not a hole, the escaping plasma can go not only directly up, but also

back dong the p v e , and, given the right conditions, complete the formation of a recast

bridge over the cut.

Recast / materiai

Figure 4Ai. Recast formation mechanism.

4.6 Secondary-Process-Parameters Tests

These tests were performed to observe the effect of process parameters believed to be less

significant for attainment of conmlIed-depth laser cutting of aiuminwn sheet. These

included (1) adding assisting gas (2) cutting with defocused beam (3) varying laser pulse

frequency. Also reported in this section are the results of multiple-pass cutting. This

approach was expected to produce betterquality cuts at the expense of extra processine

the.

Gas-assisted cutting was perfonned by attaching a nozzle over the focus lens and feeding

a supply of compressed gas (Oz or Nd thmugh the nozzle. Since the nozzle may interfere

with beam delivery, conml tests with zero gas pressure were added

Adding oxygen assist increased the cut depth, but at the expense of the cut quality (Figure

4.21). Greater amount of more dense recast can be seen in the section images. Gas

pressure variation ( h m 50 to 120 kPa) does not appear to have signifiant influence on

cut depth (Figure 4.22). (Note that cut depth resuIts for gas-assisted tests at 1500

mmhin are not show since a cut-through was observed). No effect on kerf width was

observed (Figure 4.23). Increasing gas pressure does appeaf to produce significant recast

formation ever at higher speeds (Figure 4.24).

Figure 431 (a) no gas assist, (b) gas-assist (for boîh 1500 mdmia, 7W).

120

100

' 80 i O

E w 60

d - 40

G +7w, 100 kPa, 0 2

20

O O 1 O00 2000 3000 4000 5000

Cutting Spmed (mmlmin)

Figure 4.22 Cut deptb for oxygen-assisted eutting experiments.

Figure 4.23 Ked width for oxygen-Pssisted cutting experiments.

O 1 O00 2000 3000 4000 5000

Cutting Spred (mmlmin)

Figure 4.24 Recast depth For oxygen-assisteri cutting experirnents.

Nitrogen Assist

Nitrogen, similarly ta oxygen, produced deeper cut depths and more recast (Figure 4.25),

and depth was found to be not affected by gas pressure changes (Figure 4.26). Kerf

widths are not affected by pressure as well (Figure 4.27)- but they appear to be reduced at

higher speeds. While recast disappem for speeds 2000 mrnlmin and higher at zero gas

pressure, with the Nz added, this happens for al1 gas pressures only at 3000 mmhnin.

Figure 4.25 (a) no Nt gas assist (a) N2 gas assist (for both 1500 rndmin, 7W)

Figure 4.26 Cut depth for nitmgen-assisted cutting experiments.

O 1 O00 2000 3000 4000 5000

Cutting Speeds (mrnlmin)

Figure 4.27 Kerf width for uitrogenlassisted cutting expetiments.

Adding gas-assist increased cut depth but at the expense of reduced cut quality. We can

conjecture that gas inmases the cooling rate of the kerf walls and possibly presents a

M e r obstacle to the vapour escaping h m the kerf and in this way increases the rate of

the resotidification process. Given the importance of good cut quality in LOM-AL

pmess, the use of gas assist is not recommended (or at least should be applied at a

minimum pressure affording the lens protection).

Defocused cutting is achieved by shifting the specimen surface above or beIow the focal

plane of the Iens. In our case, the specimen surface was raised so that the beam focused

below the surface of the sheet. The experiments intendeci to observe if improved cutting

quality cwld be obtained by such means. Figure 428 shows that deviating h m focal

plane within the range of 0.08 mm does not affect the cut depth significantiy. Note that

we bave estimated that the theoretical depth of focus is 1 12 pu h m Eq. (2.6). Howevw,

when the beam was focused tu 0.12 mm below the materid surface, the cut depth was

affected significantiy, with about 30% decrease on average. Therefore, we can estimate

the range of maximum acceptable variation in focusing the laser beam ta be within M.8

mm.

+O.Ci4 defocused 1 I -0.08 delocused

-%+ 0.1 2 detocusad

Figure 4 3 Cut depth for defocusednitting eq~riments.

Kerf width was not observeci ti, have signifiant diffemnce within the tested focus range

(and is not shown here). Figure 4.29 shows that defacusing the beam spot h m the

matenal surface does reduce the recast formation to some de-. However, by shifting

the focal pIane h m the materiai surface, we aIso decrease the margin of error for the

surface height position.

120

1 O0 -0.08 defocused

i -0.12 defocused

1 I 'O

2 0

... \ - - - - - - I 1 O00 2000 3000 4000 5000

Sean rpood (mm lm in)

Figure 4.29 Recast depth for defocusedcutting experiments.

4.6.3 Laser-pulse-frequency variation

Higher pulse frequencies produce greater pulse-to-pulse overlap and thus were expected

to better approximate continuous-wave laser operation, giving improved cut quality. On

the other hand, lower puise frequencies produce greater energy per puise, possibly

resulting in p a t e r cut depth. Figure 4.30 shows that the cut depth does not Vary

significantly for frequencies between 8 kHi and 12 kHz. However, a slightly higher cut

depth was observed for 6 kHz at sspeeds below 3000 mmlrnin. The kerf width and recast

were not affected by the frequency variation (and are not show here).

4 1 2 kHz 100 - \ +10 kHz

Figure 4.30 Cut depth for pufse=fre~uency.variation experiments.

4.6.4 Multiple-pass cutting

Passing over the same path more than once to achieve the same cut depth in two passes as

in one requires reduction in power density in two-pas cut vs. the single pass. This

reduction cm be achieved either by reducing the power or inmasing the scan speed.

Case A: Increased Speed

As was observed in singie-pass tests, increasing the scan speed generally reduces recast

formation and produces better quaiity cuis. Thus, by doubiing the speeds for the two-

pass tests, we produced as expected notabiy better cut quality (Figure 4.31). The two-

pass grooves have nearly verticaI ketf walIs and very Little debris inside and outside the

kerf. However, the cut depth achieved by two passes is about 20-30 8 Iower than for the

single-pass cut with the same total energy delivered (Figure 4.32). Kerf widths were

generally greater for the two-pass cuts when cornparing equalenergy points (Figure

4.33). At highest speeds, the kerf width for both approaches level off at 41-43 p.

Figure 4 3 1 Groove profdes comparing (a) sin&-pass cut (2500 mmlmin, TW} and (b) two-pass cut (5000 mm/min, 7W).

1 + 4 - ~ ~ ~ ~ ~ ~ 1 ~ ~ - 1 O0 + 1 pass, 7W

20

O ! r

O 2000 4000 6000 8000 1 O000

Cutting Speed (mmlmin)

Figure 4.32 Cut depths for single-pas riad two=pass, speed-doubled tes&

5

O 4

O 2000 4000 6000 8000 1 O000

Cutting Speeds

Figure 4.33 Kerf width for single-pass and two-pas, speeddoubled CU&

Case B: Reduced power

Lowering the power for two-pass tests while keeping scan speeds the same produced

shalIower and narrower grooves (Figure 4.34(b)). The cut depth also showed pater

variability with speed, while not increasing with decreased speed as expected (Figure

4.35). The cut depth ranged between 50 and 75 jun for ail speeds. Kerf width was also

roughly constant at about 12-14 p. It appears that, once the power drops below certain

Ievel, cutting performance deteriorates. Thus, for mutti-pas cutting, acceptable

operating conditions are bounded at one end by the minimum pwer and on the other by

the maximum speed possible, given either the iimitations of translation device or the

finite pulse frequency.

Two-pass tests showed that it is possible to achieve betterquaiity cuts by this appach .

Therefore, if single-pass cutting cannot produce acceptable cut quality, two-pass cutting

offers an alternative approach.

Figure 4.34 Gmve protides cornparhg (a) single-pass cut (3000 mmlmin, 7W) and (b) two-pass cut (3000 d m i n , 3 m .

- - I - - - -Cut through

1500 2000 2500 3000 3500 4000 4500

Scin Sprad (m m lm In)

Figure 4.35 Cut depths for siagbpass and t w o = p reâuceâ-energy (3.5 W) tests.

4.7 Cornparison with Laser Machining Model

In order to compare ou- laser cutting observations with the existing theory, we can fit our

data to the laser-machining mode1 of Chryssolouris (1991) presented in Section 2.6. The

equation is reintroduced below for ease of reference:

The known parameter values (aluminum dloy material properties) are p = 2700 kg/m3,

C, = 896 JkgPC, latent heats of fusion 395 Id/kg and vaporization, 10.5 MJkg,

vaporization temperature Ts =2467 OC.

Two parameters, focal spot diameter, d, and absorptivity, a, are not accurately known.

The theoretical lower bound for d is 11.3 pu, (Eq- (2.5)) for our optical system

characteristics. Our kerf width measurements for lower powers were for example, for

4W, 20 um, and for 3.5 W, 12-15 p. Assuming that at lower powers the kerf width is

approximateIy equal to the beam diameter, we wiil Iet d = 15 pm. While absorptivity at

m m temperature for aluminurn is known (0.22) (ANSUAWS C7.2: 1998). It is not

known how this value changes at elevated temperatures. At temperatures encountered in

laser rnachining, it is only known that the absorptivity increases. AdditionalIy, the

extremeIy fast nature of the interaction between the laser pulse and material surface and

the unknown factor of interference as a resuit of plasma formation, mean that the actual

absorptivity value likely changes rapidly.

Therefore, we can simply use the value of "a" as a means of "fitting" the mode1 in

Eq. (2.7) to our observations. Dashed curves in Figure 4.10 and Figure 4.16, which show

cut depths for SP and LS specimens, were obtained by using Eq. (2.7) with a = 0.24. For

SP specimens, an excellent fit was obtained for al1 but the lowest power setting where the

mode1 consistently underestimated the cut depth. On the other hancl, for the LS

specimens good fit was obtained for all power settings.

4.8 Summary

Through primary-process-parameter tests, we established a consistent reiationship

between (geometric and quality) characteristics of the laser-cut grooves and the process

parameters (scan speed and laser power). We have also introduced a novel approach to

characterizing laser-cut groove shape by a singIe parameter (power function exponent).

The groove profile shape was shown to be a function of scan speed only. A suitable

combination of speed and power was fond which resulted in complete cut through of the

top aluminurn sheet without significant damage to lower lamina.

Secondary process-parameter-tests showed that gas assist is not beneficial in our process

application because it exacerbates recast formation problem without bringing any

significant benefits (other than the physical protection of the lens h m debris).

Defocused cutting and pulse frequency variation did not offer any tangible benefits.

Defocused cutting tests however did demonstrate the acceptable focus variation range to

be H.08 mm.

Multiple-pass cutting tests demonstrated ability of this approach to produce very clean-

cut grooves, with nearly vertical kerf walls, stable kerf width, and generdly consistent

performance.

5. Modeling of Laser-Cut Profile Using Artificial Neural Networks

M s e shap control of the lasercut grwve profiIe mu t be maintained in order to

machine components accurately by a laser. Several analyticai and numerical m&ls

relaiing the process parameters to laser-cut W v e shape have been proposed munting et

ai., 1975, Chryssolouris, 1991, Modest, 19%). However, due to complex rature of the

interaction between the laser beam and material, the proposed modeis are jtill ümited in

their accurately and applicability.

On the other hand, artificiai neural networks (NN) have k e n shown to be abIe to mode1

any cornplex non-linea. function in many applications. These include aircraft autopilot

and automobile guidance systerns, object discrimination, sonar and radar signa1

processing, process control, welding quality andysis (Dernuth, 98).

Number of researchers have applied NN in laser material pmessing modeling and

control. Lee (1998) teports using NN in Iaser cIeaning prwess to recognize the acoustic

frequency spectrum pattern to enabb the control system to maintain optimal cleaning

paramerers. Beyer (1994) rrained NN by plasma signais to detect full penetration point in

laser welding with high degree of certainty.

Gong (1997) employs NN in a control system which optimizes focal point position in a

Corlaser-based welding system. The NN was mined with photodiode signds under

variety of process conditions, and good focal point positionhg accuracy was obtained.

ControUed-depth laser cutting is another complex process which is difficdt to model

accurately. Given the limitations of the currient analytical models and given that there

have been no hown applicaiions of NN in this area, we propose herein that the NN be

used to model the laseratting process. The purpose of such a model would be to relate

pmess parameters with the laserat groove shape.

This chapter wiIl describe the basic principles of artificiai neuraI network, the

mathematical modeting of laser cut profile, the implementation of neural network, and

the analysis of its results.

5.1 Neural Networ k Architecture

The artificial n e d network's operation is inspired by the brain bidogy. Therefore, we

will briefly describe those characteristics of brain function that are related to the

development of artificial neural network.

5.1.1 Biologicai inspiration

The biological brain has a large number of highiy connected elements called neurons.

They have three major components, (1) the dendrites, (2) the ce11 body, and (3) the axon,

see Figure 5.1.

(1) The dendrites are tree-iike receptive networks of nerve fibers that c q signds

into the ce11 body.

(2) The ce11 body effectively sums and thresholds these incoming signais.

(3) The axon is a single long fiber that carries the signal h m the ce11 body out to

other neurons.

Figure 5.1 Biologicai Nemas mgan, et al. i995).

The contact point between a cell's axon and a dendrite of the other ceii is called synapse.

It is the arrangement of neurons and the strengths of the individual synapses that estabIish

the function of the neural network

Artificial neural networks do not reach the complexity of the brain. However, there are

two major similarities between bioIogicai and artificial neural networks: (1) the basic

building blocks of both networks are simple computational mechanisms that are highly

interconnected, (2) the connections between neurons determine the function of the

networks.

51.2 Neuron

The fundamental processing element of an artificial neurai network is the neuron. A

general neuron, Figure 5.2, has many inputs with one output. Each input to the general

neuron is associated with a weight w, and a bias temi b is associated with each neuron.

(

inputs( output

\

Figure 5.2 S c h e d c d a Multipleinput neuron.

The neuton output is calculated as a function of the input parameters. This function is

known as the neuron activation function or the transfer function, J With reference to

Figure 5.2, the input-output reiationship of a neuron is given by the following equation:

The relationship n can be also written in matnx fom as

Now the neuron output can be written as

Both wj and b are adjustable parameters of the neuron. They are updated after each

training cycle, which is referred to as an epoch.

5.13 Transfer funetion

The transfer function may be a linear or a nonlinear function of n. The choice of specific

transfer function depends on the particular problem. Table 5.1 shows different possible

types of transfer functions.

One-neuron network with many inputs is not sutficient to solve most cornplex function

probIems. In that case, increasing nurnber of neurons operating in parailel is required

These neurons operating in parailel are calIed a layer. Note that now each input connects

to each neuron in the hidden layer (Figure 5.3). Neurons within the same Iayers use the

same type of transfer function but have varying bias values. An increased number of links

(weights) resulting h m the higher number of neurons in a Iayer ailows better

approximation of non-Iinear functions.

5.1.5 Multiple layen of neurons

Adding more layers to a single-layer of NN M e r extends its power. A two-layer

network having a sigmoid fVst layer and a iinear second layer cm be trained to

appximate most functions arbitrariiy weii (Hagan et al. 1995).

Adding biases to network further improves its modeiing performance. The reason is the

biases give the networks an extra variable in neurons so that the network can solve hiber

order non-tinear function.

Table 5.1 Tramfer Funcaoas in a Neural Network.

Hard Limit

I Linear

Symmetric Saturaring

Liner

Hyperbolic Tangent

Sigrnoid

Positive Linear a=O ne0 a = n O I n

a = 1 neumn with max n

a = n al1 other neurons

Tfie layer whose output is the network output is calleci the output Iayer. The other layers

in the network are called hidden layers. WhiIe some authors refer to the network in

Ftgure 5.3 as a three-layer network (counting the input, hidden, and output layers), we

will follow the more common practice, and will refer to such networks as "mm-layer"

ones.

s 4, i i

Hidden Layer Output Layer

figure 53 Schematic of a multiple-input neuron, hvo-layer network.

To identify the structure of a muiti-layer network, we will use the following notation,

where the superscript is the indicator of the layer sequence.

For example, a two-layer network with 3 inputs, 3 nemns in fmt layer @idden layer),

and 2 neurons (outputs) in second layer is identified as 3 - 3 - 2 network

5.1.6 Optimal network structure

Finding r suitable network structure means finding the number of Iayers and number of

neurons per layer appropriate for the problem at hand. One approach is to use hiU-

ciimbing searches that selectively rnodify an existing network structure. There are two

ways to do this: start h m a large network and make it srnalier, or vice versa A more

advanced approach is to use a genetic aigorithm to search the space of network strucnires

@usseIl, S. and Norvig, P., 1995). However, this approach is very CPU-intensive since

the search space is very large.

in general, there are no determitlistic rnethods for finding NN structure for optimum

performance (Hagan et d. L995), and most often one has to resort to trial and error

approach (Lee et aI. 1998).

A leamhg rule or training algorithm is a procedure for modifying the weights and biases

of a network. By using the training algorith, the network c m be trained to perform

certain tasks. It is especially usefuI to use supervised training.

In the case of supervised training, the leaming rule is pmvided with a set of examples

(training set), such as

Where p~ is an input to the network, and t~ is the corresponding correct (target) output.

As the input is apptied tc, the network, the network outputs are compared with targets

(emr = target - network output). The learning rule is in here used to adjust the weights

and biases of the network so that the network outputs are closer to the targets.

The adjusunent race for weights and biases is constrained by a value cailed learning rate.

The higher the learning rate, the larger changes to the weights and biases in each epoch;

higher vaiues shorten the time required for the network io attain preset man square m r

goal, and consequently, result in faster search termination. However, high leaming rate

vaIues c m Iead to unstable soIution of the probIem, which means the convergence may

not be the global minimum. On the other hand, bw Iearning rate value increases the

number of epochs, with correspoading increase in computation expense. For our

problem, we do not find it necessary to use higher leaniing rate to reduce the computation

rime due to relatively srnaII data set, which ~ s u l t s in a short training process.

5.1 9 Convergence

Convergence is reached when Ieaniing aigorithm h d s a solution with a minimum mean

square m r . In many cases, however, the convergence may be reached but the minimum

e m r found is not a global minimum. For example, complex functions may have many

local minima, which could lead to network training algorithm to stop before it reached

the giobal minimum. Therefore, training the network with different initiai conditions cm

ensure a global optimum solution has been obtained.

5.1.10 Generalization

In most cases the multi-layer network is trained with a limited number of examples

(training sets) to approximate the described behavior of a network. It is very critical that

the network actuaiiy able to generalize the function. Generalization is the ability of the

neural network to approximate the function not only for the provided examples but aiso

for the inputs in between the examples. Therefore, if the input vaiues in between the

training set values are provided to the network, the output values shouId still have

acceptable network outputs. Thus, if one is not careful, the network stmcnue may be not

well generalized; such a case is refemd to as over-fitting. This behavior is the direct

result of employing more hidden layers and neurons than required by the complexity of

the functionai relationship being modeled The nile of thumb is not to use a larger

network if a smaller network achieves the same performance (Hagan et al., 1995).

5.2 Neural Network lmplementation

5.2.1 Motivation for Neural Network appiication to laser grooving

Examination of the cut profile data in Chapter 4 indicates that there is a predictable

relationship between certain laser power and speed combinations and kerf width and cut

depth. If we wanted to make a contoured surface (such as that in Figure 5.4) by using

laser machining, we need to be able to set laser pmessing parameters so as to obtain the

desired cut shape (Figure 5.5). Thus, we would like to design a neural network, which

wouId enable us to achieve the above goal by modeling the laser machining process.

Figure 5.4 An example of a bottle c w e d contour surface.

Cut Shape Requiremen ts

d Shape Machming Pattern S a p e

Figure 5.5 An example of the multiple pass cuttiag.

5.2.2 Defining hput/Output parameters for ArtSncial N d Network use

An artificiai neural network is trained by examples; an example means ihat, given certain

conditions, a particuiar outcome results. In our case, examples consist of two paired

parameter groups: one group definhg the laser-cut profile shape, the shpe group, and

another defining the corresponding ptocess parameters, the process group. A set of these

group pairs used to train the network is called a training set.

As shown in Chapter 4, we can compIetely define the geometry of the cut profile by three

parameters: kerf haif-width, p, cut depth, q, and profile shape exponent B. While both, p

and q, are affected by the two primary process parameters, laser power P and scan speed

V, B has been shown to be a tùnction of V, only. Therefore, we cannot

independently set ail three profile shape parameters since setting p and q defines P and V,

and B = f (V) only.

Based on the above training sets, two complimentary networks have been designed

implemented, and tested. The first network (2-x-2) bas shape group as input and pmess

group as output, and wili be calIed shape-to-pmess or S-P network. The second network

(also 2-x-2 type) has the above inputs and outputs reverse4 and so is called process-to-

shape or P S network.

Training Example Data

Shape Group

t'- Y

Rocess Group 7

Figure 5.6 A schematic diagram to represent grouping parameters tbat relatai to laser catting pnicess,

5.2.3 N e u d Network Training anà Impkmentation

To train our networks, we use the backppagation algorithum (Demuth et ai. 1998) in

which, after the fmt epoch, the network computes a set of outputs which are compared to

the expected value (or target). Then these m r s are fed back to the network (back-emr-

propagation) for further iteration to reduce its ecrors. The procedure is repeated until the

preset goal is reached. The network was trained in a batch mode or supervised learning

mode, which means that the weights and biases were oniy updated after al1 the inputs and

targets have been presented; in other words, al1 the training examples were provided to

the network at once.

When seiecting the data to be used for training of the networks, we need as large and as

complete a set as possible. In our case, the largest number of data points (14) was

collected h m the solid-plate laser-cuning tests (Table 5.2). This set of observations was

further subdivided into the training set (9 points) and a tesring set (5 points). The testing

set is used to evaluate the performance of the network obtained based on the training set.

The testing set points were chosen so that they alternate with training set points.

The networks were developed, trained, and implemented using MATLAB Neural

Network Tool. The input-output pairs of data were normaiized to -1 and 1 range. The

nonnalization increases the network's ability to solve large number of example sets

effectively. The learning rate was set to 0.05, which was found to give a stable solution

from the network for al1 runs. The number of epochs is limited to 1600 and the training

is halted when 1x10" mean square emrs is reached The 1600 epaçhs lirnit was chosen

to prevent unwanted early stop iteration for the networks. The mean square e m r vdue

(IO-') is not significant, since it is significantiy below our expected observation errer- A

typical m r history during the neurai network training process is shown in Figure 5.7.

Table 5 3 Experimental Observation of Laser Cut h t ü e Parameters*.

Laser Pnicess Parameters Laser Cut Profile Parameters

*Note: italics & boldindicate test set data, remaining data belong to training set.

Pedomiance is O.[XXZlanB. Goal is 1 â905 10' . 1

S-V (mmlmin)

3000 3!W 4000 4500 2500 3000 3500 4000 4500 2500 3040 3500 m 4500

221 Epochs

Power P (Watt)

7 7 7 7 1 O 10 10 10 10 12 12 12 12 12

Figure 5.7 An exemplary training error history graph for 2-3-2 S-P network.

8

2.89 3.96 2.86 2.26 4.78 2.91 3.04 2.62 2.62 3.56 3.46 2.81 2.90 2.39

P

23 25 29 31 26 30 34 34 34 29 29 34 34 35

q

121 98 87 80 181 144 125 103 93 1 88 161 146 117 1 02

5.3 Neural Network Tests and Results

5.3.1 Network tests

The purpose of network testing is to find a suitable stmcture of the network (i.e. number

of neurons and layers). This section presents results of network testing for S-P and P S

networks. For each network, a set of different structures is tried, trained, and evaluated.

Based on evaluation, an appropriate structure for each network (S-P and P-S) is proposed.

When testing a network, its performance is evaluated in terms of percent emr as

Network output - Experimental Values Percentage Error = x 100% (5.6)

Ekperimental Values

The percent error can be calculated for two sets of data: training set and test set. By

observing variation of these emrs for different network structures, we cm select

appropriate network which achieves good generalization and avoids under- or overfitting.

5.3.2 Sbape-to-Procesi (S-P) Network Tests

ûnly twalayer network structures (2-x-2) have been tested. Based on their satisfactory

performance, it was not found necessary to increase nurnber of hidden layers any further.

Testing was canied out with number of hidden-layer neurons (x) increasing from 1 to 10

(Table 5.3).

Table 5.3 Testing increased number of neurons S-P neîwork test.

Calculating emr for the training set inputs resulted in the plot in Figure 5.8. The average

e m declines to nearly zero for 4 hidden-1ayer neurons ( 2 4 2 network). Once the emr

falls to zero, we know that we are ovetntting our data because we know that there is

(non-zero) experimental error in our observations. Zero e m r for the training implies that

om network c m "fit" exactly to the training set data.

On the other hanci, calculating average emr for the testing set inputs, see (Eigure 5.9), we

observe a different trend. The e m r falls to a minimum of about 3% for 2 neurons in

hidden layer (2-2-2), and then rises quickly. This indicates that the 2-2-2 network is best

able to generalize based on the training set.

- -

Figure 5.8 Average % emrs for the training set in S-P (2-x-2) networks.

5.3.3 Procoss-to-Shape (P-S) Network Tests

The inverse versions of the S-P networks listed in Table 5.3 were tested in order to find

suitable structure for the P-S network. Calculation of e m r for the training set inputs

produced the same resuIt of near-zero e m r for four and higher number of hidden-layer

neurons (Figure 5.10). For the test set, the emr also went through a minimum which

occurred this tirne at 4 neurons (Figure 5.1 1). We know that zero error for training set at

4 neurons means overfitting occurs at this point. Therefore, a suitable structure wiU have

3 or fewer hidden-Iayer neurons.

20.00 18.00

t 16.00 14.00 1200 g 10.00 +% emrs P

c 8.00 6.00 4.00 I 200 i 0.00 i

No. of namm in hiddrn layw (NomMnid NN ouipub)

Figure 5.9 Average % errors for the test set in S-P (2-x-2) networks.

Given that the emr is only slighdy higher (4 9% as 3.5 96) for 2 neurons vs. 3, we codd

choose the simpler network (2-2-2), or opt for reduced error of 2-3-2 network. We can

verify the performance of our selected P-S network (2-3-2) by cornparhg the

experimental observations with the network predictions. Figure 5.12 shows such

cornparison for the kerf half-width, p. and Figure 5.13 for the cut depth, q. Both pIots

demonstrate excellent fit.

No. of N.uiori. in NN Lvyrrr pi- NN Odpub)

Figure 5.10 Average % errors for aoinùag set in PIS (2-x-2) networlis.

0.50 -1 1 2 3 4 5 6 7 8 9 1 0

No. of Nouons in NN byrcr (Nor~îraliwd NN Output8)

I

Figure 5.1 1 Average % emrs for tbe test set in P-S (2-x-2) networks.

Figure 5.12 Kert haif-width predictions by P-S 2-3-2 network (hollow-eirele data points are from the testhg set).

50 -I I

2500 3000 3500 4000 4500

Scrnnlng Speeds (mmhln)

Figure 5.13 Cut depth ptedictiom by 2-3-2 P-S network (hoiiow-circle data points am from the testhg set).

5.4 Summary

Two-layer (one hidden layer) neurai networks were used to mode1 the ~Iationship

berween the laser-cut groove section geornetry and the process parameters. Both forward

and inverse relationships were modeled successfully. Optimum structure for these

networks was found to be 2-2-2 for S-P and either 2-2-2 or 2-3-2 for P S networks.

By applying the neural network modeling to our problem, we have demonstrated the

feasibility of this approach and have shown the proper procedure to be foiiowed in

seeking the most appropriate network structure. The main benefit of using NN here is in

their inherent fiexibility. if more pmcess parameters or cut characteristics need to be

added to the modei, they can be easily accommodated

Appendix A . Chernical etching procedure for preparation of laminated-sheet specimens.

Table A.l Cheinid etching solutions: 153 Finishing Lab in Alcan Reseairh Center

1 Tank 17, in Rrn 153 as above 1 Caustic Etch, Long life type 1 I

Temperature 1

Step 1: Etching Foils and Support Plates

Use a metai basket (supplied by the Iab). This basket has wire slots in order to place the

60 OC

100 to 150 gpL

I

t

sheets vertically in these slots. This was to prevent sheets from floating during the

NaOH

35 to 45 gpL

etching and aiso to avoid one side of a sheet king etched longer than the other si&,

Gluconate- sand 0 2 Etch Additive

resulting in inconsistency of the surface quality. The procedure is similar for the support

plates. The sheets and support plates were placed into the chernical etching solution for

approximately 30 seconds for ail specimens.

Step 2: Rinsing

Use water to rinse the sheets and support plates after etching. Exact time for nnsing was

not important, but the basket should be moved around to ailow water jets to clean the

surface h m al1 angIes. Procedure was perfonned in Water Rinsing Tank (Tank 16).

Step 3: Second cleaning step

Use Dessmut solution to farther clean the surface. This solution was used to remove any

residuai matter left on the surfaces as a resuIt of etching: nnse for about 5 seconds. Use

Tank 15, "Dessmut Solution Tank".

Step 4: Second nnse

Repeat water rinse at Tank 16.

Step 5: Drying

Bring the sheets and plates to air dryer to eiiminate any residual water on the sheet and

plate surfaces. After the air dryer, place the sheets and plates on paper towers to dry.

Take care not to contaminate the surfaces with fingerpints or dust.

Appendix 6 . Cutting orientation of spedmens.

The tines were cut in sequence going h m right to left with progressively increasing

speeds. In Figure B.1 arrows indicate the cutting direction; line spacing and dimensions

are also given.

Figure B. 1 The cut line pattern and nomencIanire for thriee-dot spimens

Pattern 2: AU other spechens

Patteni 2 was used for ail the remaining tests, Figure B.2. There were four groupings of

seven lines. The direction and sequence of cuts were the same for the top two groupings.

The cutting direction was reversed for the bottom two groupings, but the speed sequence

rernains the same, Each grouping was completed in tum, starting with the top right, then

top left, bottom right, bottom left.

I

Figure B.2 The cutting directions and grouping order for experiments.

Appendix C . Detailad Metallographic Proceâure

Detailed procedure for performing metallographic examination of laser-eut specimens is

given below. The procedure begins after the laser-cutting experimmts have been

completed.

I Cut specimens into appropriately sized segments

1) Cut off from both ends of the specirnen approximately 1" segments to be put into

the rubber mould. The rubber mould is about 5.3 cm by 8.5 cm (width and length)

and 3 cm in depth. Wash the mould with Cotton and cleaning agent then use air

dryer to dry it. This rubber mould is used for fust part of securing the Iaminated

sheets into a singie part. As the laminated sheets have very weak stress resistance,

by using normal blade saw sectioning excessive damage zone would be created.

Therefore, after the specimens were mounted in the mould, the damage zone can be

reduced to about one to two millimeters.

2) Rub some silicon paste around the mould to act as a lubricant to ease demouIding.

II Mix the epoxy m i n (Araldite)

3) Use a cup to hold the epoxy resin, Araldite GY 502. (a) Calibrate the empty cup to

be zero in the measurement counter. Open the epoxy resin tank to let the m i n flow

into the cup. As the viscosity of m i n is high, after the flow is turned off, wait some

time until al1 the m i n has fallen into the cup. Measute the weight of the min. Now

the liquid hardener for solidifjing the resin shouid be added. Before that. set the

weight counter to the total amount of hardener plus resin weight.

The total amount of araldite (hardener and min) is calculated by this method;

Total araldite weight = Resin Weight + Hardener Weight

where Hardener Weight = 20 % of the measured Resin Weight.

It is better to add a Iittie more hardener to ensure that the resin cures compIetely and

achieves its full hardness.

4) Use a container filled with hot water to about 80 % of its volume. This is intended

to help mixing the hardener and the min faster. Then, with one hand holding the

cup, and with the other hand holding a wood stick, submerge the cup tiII the Ievel of

water and araldite in the cup are equai (Be careful, not to ailow any water to get into

the araldite. If you do, this sample is spoiled and should be placed into the fue

hood A new rnixing is required.) Then use a wooden stick, provided in the lab, to

stir in a steady slow circular motion. Note that some white color will be observed in

the &dite during the stining. Stir the araldite until aiI the white color disappears.

III Pour resin into the mould with the specimen

Place the rubber moulds that we had prepared with silicon paste on a flat table.

Place some spacer at the bottom of the rubber mould; the purpose of this is to

ensure the specimen is surrounded by resin.

Pour the araldite slow1 y into the rubber rnould up to 114 of the rubber mould height.

Slow pouring is required to minimize the formation of air bubbles.

Place your sectioned specimen into the mould Due to high viscosity of araldite,

submerge the specimen at an angle. The angled placement can produce a gradient

pressure to push the air bubbles to the side and remove them from the bottom of the

specimen.

M e r letting the specirnen sink to the bottom, pour the araldite into the rubber

mould to fil1 up to about 80 8 of the mould height, or until there is enough araldite

to hold the specimen tight.

To remove the trapped air bubbles h m the prepared specimens, place the modd in

a vacuum chamber. Transport the rubber mouid to the vacuum chamber, slide the

chamber top to side of the chamber. T m the vacuum valve off. T m on the air

pump until the pressure reaches 25 psi. Let the specimen stay for 4 minutes. After 4

minutes, open the vaive a bit to leak some air in until the pressure drops to 10 psi.

Close the valve for another 4 minutes, then tum off the air pump and open the

vaive. Slide the top, take out the specimens and place in the fVe hoocl, As this type

of araldite is heat-cured (the reaction between epoxy and hardener) by the

exothennic reaction with the hardener. Wait for about 24 h o u to complete the

curing process.

IV Second Stage of the Metallognrphy

10) Use a blade saw to cut the specimen into pieces of half-inch width and 0.25-inch

height Use Nnder to grind off sharp edges and also to grind the specimen into the

size so it can fit into the one-inch cylindncal mould. Wash and rinse specimens, and

dry with air dryer. Take out the one-inch cyiindrical mould and use silicon paste as

for the mbber mould. Try putting the specimen inside the mould and pour half of

araldite in first. Wait until araldite gets into most of the gaps. Then, follow

procedures of the Step 8 and Step 9 above.

11) Next day, take out the bottom of the one-inch mould and push the araldite-specimen

out from the mould Wash the mouid with soap and dry it. Put the specimens into

oven for another 4 to 6 hrs to M e r cure the araldite, at a temperature of (70°C).

12) Take out the specimens and [et them cool duwn to m m temperature. Note that the

specimen is soft after taking it out from the oven. Therefore, it is very important to

let the specimens cm1 down to room temperature before polishing.

13)Grind off the edges. Use four wheels for polishing work. Fit wheel is used for

removing the damage zone h m the blade saw and the excess araldite on the m s

section to be exarnined. Second wheel is used for polishing the surface down UJ

15 m. Third wheel is used for poiishing the surface down to 3 p. Fourth

wheel is used for polishing the surface down to down to 0.5 W. The steps in this

part are: (a) Grind the edges. Use k t wheel with 240 grit polish paper. And use

600 grit for better surface quality- (b) Drill a hole at the back of the specimen.

Use next wheel, spray the required poiishing liquid, Diamond Compound

(Norpar), on the wheel, and mount the specimen on the wheel machine using the

dnlled hole. Set timer to poüsh for 3 minutes. Clean specimen and airdry it.

Check the specimen with a microscope for any uneven polishing. If so, re-polish

before proceeding to next wheeI. (c) Third wheel is used with Microid Diamond

Compound to polish specimens dom to 3 pm. The procedure is the same as for

the previous wheel. (d)Fourth wheel is used with coiioidal siIica to polish the

specimen down to 0.5 p. This wheel rquires only 2 minutes of polish rime.

After this wheel, the polish work is completeci Once again, if the quaiity is not

satisfactory, re-polishing is required. Then use the digital capturing microscope

to obtain images.

Appendix D ûetail Procedura for SMipling Later-Cut Profiles

Details of this procedure are presented below:

1. Open image file and calibrate the image using the real-world distance noted on

the scde marker shown on the image.

2. Record the cut width by measuring the distance on the image between the groove

edges on the materiai surface.

3. Record the cut &pth by measunng the vertical distance between the point half

way benveen the p v e edges at the materiai's top surface and the lowest point

on the cut p v e .

4, Record locations of between 30 to 50 points dong the cut profile. The spacing

between points was kepc approximately constant, with deeper cuts (150 vs.

50 pm for shallow cuts) quiring more points. Save the point locations by

exporting them as a tabdelimited file.

5. Open the file in a spreadsheet application (MS Excel) and transform the point

coordinates from the image b u e to the desired coordinate frame defined in Fig.

5. The origin of image coocdinate frame is located in the upper-teft corner of the

image, with the X axis pointing to the right and Y axis downward. Therefore, an

offset distance in X and Y must be calculated to transform the point coordinates.

This distance is calculated by finding the point midway between the two points

where the cut profile meets the top material surface. After subtracting the offset

distance, the Y coordinates are inverted. The coordinates are then nonnalized by

dividing by p and q, and then 1 is added to the Y &ta to make it positive as

quired by Eq. (5.6).

6. Calculate logarithms of the Y coordinates and of the absolute values of X

coordinates (since the "logn function is ody valid for positive arguments).

Perform a separate Linear fit for Ieft and right branches of the cuve to obtain its

slope, and average the two values to obtain B.

Sample cut profile data

Table D.l A typical cut profile data is processeci as the five steps.

offset Cut profile Log (absolute (Normalized Cut ProfiIe))

Unprocessed Cut profile

Normalized Cut Profile X(nom) = x/p, Y(norm) = ylq +

Table D.2 The essential masurernmts.

Table D.3 The find step of validating p, q and B vatues to match experimental cut profile

and modeled cut profiIe.

ANSIIAWS C7.2: 1998, Rerommeded Practices far Laser Beam Weiding, Cutting,

aad DrilUag, Amencan Welding Society.

Barlow, J.W., JJ. Bearnan, and B. Balasubmanian, 19%. "A Rapid Mould-Making

System: Material Properties and Design Considerations," Rapid Protoryping Journal,

Vol. 2, NO. 3, pp. 4-15.

Beyer, E., D. Maischner, and Ch. Kratzsch, 1994, "A Neural Necwork to Analyze Plasma

Fiuctuations with the A h to Determine the Degree of Full Penetration in Laser

Welding", I W O , pp.5 1-57.

Boureil, D. L, R. H. Crawford, R L. Marcus, J. J. Beaman, and 1. W. Barlow, 1994,

"Selective Laser Sintering of Metais," ASME [nt Mechicd Eng Congress and

Erposirion, Chicago, ii, PED-Vol. 68-2, pp. 519-528.

Bunting, K. A. and G. Cornfield, 'Toward a General Theory of Cutting: A relationship

Between the Incident Power Density and the Cut Speed", Trmûctions of ASME-Journal

of Heat Trmfer, 1975, pp. 1 16-122.

Cho, W. C. and George Chryssolouis, 1995, "Analysis of the laser grooving and cutting

processes", J. Phys. D: Appl. Phys. V01.28,1995, pp. 873-878.

ChryssoIouris, G., 1991, Laser Machining-Theory and Pracîice, Springer-Verlag.

Chua Chee Kai and Long Kah Fai, 1997, '%apid Prototyping in Singapore: 1988 to

1997, Rapid Protutyping Journal, Vol. 3, No. 3, pp. 116-1 19.

Chua Chee Kai and Long Kah Fai, 1997, Rapid Prototyping-Pcinciple and

Appiications in ManufaEhinag, John Wiley & Sons, Inc..

De Graal, R. F. and J. Meijer, 2000, "Laser cucting of meta1 Iaminates: anaiysis and

experimental validation", Joumd of M0teriuZ.s Processing Technology, Vol. 103, pp. 23-

28.

Demuth, Howard and Mark Beale, 1998, Neurai Network Taolbox-For use with

MATLAB, Version 3, The Math Wotks hc..

Faerber, M., G. Broden, 1999, ''Laser Beam Cutting of Aluminum Alloys", Proceedings

of the SheMetP99, Sept., pp.43 1-436.

Gong, Hui and Flemming O. Olsen, 1997, "Automatic Optimization of Focal Point

Position in CO2 Laser Welding with Neural Network in a Focus Control System",

Section C-ICALEû, pp.67-74.

Hagan, M. T., Howard B. Demuth and Mark Bealc, 1995. Neural Network Design, PWS

hblishing Company.

Jacobs, P. F., 1992, Rapid Prototyping and Manufacturing: Fundamentals of

Stereoüthography, SME, Dearborn, Michigan.

Jacobs, P. F., 1995, "QuickCast 1-1 and Rapid Tooling," SU?h fnternational Conference

on Rapid Prototyping, pp. 237-257.

Jacobs, P. F., 1996, Stereolithography and other RP&M Technologies-from rapid

prototyping to rapid twl ig , ASME Press.

Kaebernick, H., D. Bicleanu, and M. Brandt, 19W, "Theoreticai and Experimental

Investigation of Pulsed Laser Cutting", Annals of the CIRP, Vol. 481111999, pp. 163-166.

Klosterman, D., B. Priore, and R. Chartoff, 1997, "Laminated Object Manufacturing of

Polyrner Maûix Composites," Pruceeding of the Seventh lntenrational Conference on

Rapid Prototyping, San Francisco, CA, pp. 283-292.

fistensen, T. and F. Olsen, 1994, "Investigation of Cutting of Al aIloys with pulsed and

CW CO' Lasersv*, The Welding in the World, VOL 33, No. 5,1994, pp. 9-15.

Laliemanci, G., G. Jacrot, E. Cicala, and D. F. Grevey, 2000, "Grooving by N&YAG

Laser Treament", J o u d of Materials Processing Techology, Vol. 99, pp. 32-37.

Le, J. M., K.. G. Watkins, W. M. Steen, 1998, '?nvestigation of intelligent Power

Monitoring in the Laser Cleaning Pmcess", Section C-ICALEO, pp. 18-23.

Lwon, J. T., 2001, "Optics and optical system", in LIA Handbook of Laser Material

Rctcess, ed. J. F. Ready, Laser Institute of Amenca, Magnolia Publishing, Inc., 2001,

pp.95-%.

Migliore, L., 1996, Laser Materials Rocessing, Marcel Dekker, inc.

Modest, M. F., 1996, "Three-dimensional, transient mode1 for laser machining of

ablating/decomposing materials", Int. J. Heat Mass Tramfer, Vol. 39, No. 2, pp.221-234.

Mohanty, P. S. and 1. Mazumder, 2001, "Energy Transport in Laser-irradiated

Materiais", in LIA Handbook of Laser Material Process, ed. J. F. Ready, Laser institute

of America, Magnolia Publishing, inc., pp. 184-200.

Nantel, M., Y. Liao, B. Bruner, P. Heman, R. Marjoribanks, R. J. D. Miller, K. Chen, 1.

Konovalov, S. Ness, A. Oettl, B. J. Siwick, 1999, "Laser Micro processing at Photonics

Research Ontario", ICALE0'99, Laser institute of America Proceedings, Vol. 88, and pp.

186 - 194.

Petring, D., 2001, "Cutting Process", in LIA Handbook of Laser Material Process, ed J.

F. Ready, Laser institute of America, Magnolia Publishing, Inc., 2001, pp.425-428.

Pietro, P. D., Y. L. Yao, and A. Jeromin, 2000, "Quality optimization for laser machining

under transient conditions". Journal of Maerials Processing Technology, Vol. 97,2000,

pp-158-167.

Ready, J. F., 2001, 'ZIA Handbook of Laser Materials Processing", Laser Institute of

America Magnolia Publishing, Inc.

Rodngues, S. J., RP, Chartoff, D A Klosterman, M. Agarwala, and N. Hecht, 2000,

"Soiid f t ee fm fabrication of functional siiïcon nitride ceramics by laminated object

manufacturing," Solid Freefom Fabricahon Symposium, Austin, Texas, CD-ROM.

Russeii, S., and P. Norvig, 1995, Artiîidal Inte-A Modem Approach, Prentice

haii Series in Artificiai intelligence.

Sheng, P., G. Chryssolouris, 1994% "Investigation of acoustic sensing for laser

machining process-Part 1: Laser driiiing", Jountd of Material Processing Technology,

Vol. 43, pp. 125-144,

Sheng, P., G. Chryssolouris, 1994b, "Investigation of acoustic sensing for laser

machining process-Part 2: Laser grooving and cucting", Jountal of Matenal Processing

Technology, Vol. 43, pp. 145-163.

Thilmany, J., 2001, "Printing in t h e dimensionsT', Mechical Engineering, Vol. 123,

No. 5, May, pp. 58-60.

Tsang, H., and G. Bennett, 1995, "Rapid tooling - direct use of SLA moulds for

invesunent casting," First National Conference on Rapid Proroiyping and Tooling

Research, pp. 237-247.

Waiczyk, D. F., 1998, "Bevel Cutting Methods and Cutting Trajectory Conml for Steel

Laminations Used in Tooling", Solid Freeform Fabrication Symposium, Proceedings, pp.

121-129.

Zayhowski, J. 1.. 200 1, "Laser configuration", in LIA Handbook of Laser Materiai

Process, ed. J. F. Ready, Laser institute of Amenca, MagnoIia Publishing, Inc., 2001,

pp.9-16.