Davim J. Paulo (editor). Machining. Fundamentals and Recent Advances

Подождите немного. Документ загружается.

Advanced (Non-traditional) Machining Processes 323

the jet of MAPs is another important parameter of this process. However, this

MAF technology needs to be brought out of the research laboratories so that it can

be employed on the shop floor for industrial use. Results have been reported

[26,

27] in the literature regarding finishing of stainless-steel rollers using MAF to

obtain a final surface roughness R

value of 7.6 nm from an initial surface rough-

ness value of 220 nm.

11.5.3 Magnetic Float Polishing (MFP)

Various processes are available for finishing flat, complex-shaped and cylindrical

surfaces but none of these processes qualifies for finishing spherically shaped

workpieces, say, balls. A new process called magnetic float polishing (MFP) is

applicable mainly for spherical balls. This process was developed for gentle fin-

ishing of very hard materials like ceramics, which develop defects during grind-

ing, leading to fatigue failure. To achieve a low level of controlled forces, a mag-

netic field is used to support an abrasive slurry when finishing ceramic products,

namely ceramic balls and rollers. This process is known as magnetic float polish-

ing (MFP). This technique is based on the ferro-magnetic behaviour of the mag-

netic fluid that can levitate a non-magnetic float and suspended abrasive particles

in it through the magnetic field. The levitation force applied by the abrasives is

proportional to the magnetic field gradient and is extremely small and highly

controllable. This is a good method for super-finishing of brittle materials with

flat and spherical shapes [5,

27,

30,

31].

Figure 11.23. Schematic diagram of the magnetic float polishing apparatus

Spindle

Ceramic

balls

Drive

Shaft

Al Base

Magnetic

Fluid &

Abrasives

Float

324 V.K. Jain

The setup consists of a magnetic fluid containing fine abrasive grains and ex-

tremely fine ferromagnetic particles in a carrier fluid (water or kerosene) (Fig-

ure 11.23). On the application of the magnetic field, the ferro-fluid is attracted

downwards to the area of higher magnetic field. At the same time, a buoyancy

force is exerted on the non-magnetic material, pushing it upward to the area of

lower magnetic field. The abrasive grains, the ceramic balls, and the acrylic float

inside the chamber, all being non-magnetic materials, are levitated by this mag-

netic buoyant force. The drive shaft is fed down to contact the balls and to press

them down to reach the desired force level. The balls are polished by the relative

motion between the balls and the abrasive particles.

11.6 Micromachining

Micromachining is sometimes misunderstood, as if it were performed only on

microscale parts. Micromachining can be performed on both macro- and micro-

components. Micromachining encompasses all those processes in which material

is removed at the micron level. Micromachining processes are classified mainly

into two classes: bulk micromachining, where the objective is to shape and size

a component, usually a miniaturized component; This also includes drill-

ing/machining of microholes and microcavities. Here, material removal takes

place in the micron domain (say, microchips), hence leading to the term micro-

machining. The second class includes surface micromachining, where the objec-

tive is to improve the surface finish to the micron or sub-micron (nanoscale)

range. Surface roughness values obtained by some of these processes have been

reported to be as low as the size of or a fraction of an atom. Table 11.1 shows the

capabilities of some of the micro/nanofinishing processes. To be more precise,

sub-micromachining and sub-microfinishing are also known as nanomachining

and nanofinishing, respectively. The majority of AMPs are capable of performing

nanomachining/nanofinishing operations.

Normally, mechanical-type advanced machining processes (MAMP) are used

for macromachining, so their process parameters should be scaled down so they

can be used for micromachining of both macro- and microproducts. For example,

ultrasonic micromachining and abrasive water jet micromachining are two such

processes. However, almost all advanced finishing processes (AFM, MAF, MRF,

MRAFF, etc.) are microfinishing and nanofinishing processes. In these processes,

material is removed in the form of chips of micron/nanosize. These processes are

capable of producing a surface finish in the sub-micron range (Table 11.1). There

is another class of processes, known as mechanical-type advanced finishing proc-

esses, in which material is removed in the form of atoms or molecules (such as

elastic emission machining and ion beam machining). Such processes are capable

of producing surface finish Ra values smaller than the size of an atom (Ta-

ble 11.1), that is, in the sub-nano range.

In the case of thermal AMP, material is removed by melting and/or vaporiza-

tion. Some of these processes have been modified to machine at the micro- or

nanoscale. If EDM is performed with lower machining parameters it can easily be

used for microdrilling or microshaping of miniature components. The same is true

Advanced (Non-traditional) Machining Processes 325

with laser beam micromachining. However, electron beam machining performs

operations at the micro/nanolevel depending upon the values of the parameters

selected. However, the use of PAC does not seem feasible for micromachining.

ECM is mainly used for bulk material removal specially for macrocomponents,

however it has been appropriately scaled down for performing electro-chemical

micromachining (ECMM) operations on both macro- and microproducts. Photo-

chemical machining and chemo-mechanical polishing are already popular micro-

machining processes for macroproducts as well as for microproducts.

11.7 Finished Surface Characteristics

The performance of advanced fine abrasive finishing processes is evaluated by

their achievable surface finish, form accuracy and resulting surface integrity.

These fine abrasive finishing processes have been shown to yield accuracies in the

nanometre range. The order of surface damage ranges from micrometres to nano-

metres in these advanced finishing processes. This makes their measurement

a formidable task. Table 11.1 summarises the attainable surface finish of advanced

fine finishing processes.

References

[1] Jain VK (2002) Advanced Machining Processes. Allied, New Delhi.

[2] Benedict GF (1987) Nontraditional Manufacturing Processes. Marcel Dekker, New

York, NY.

[3] McGeough JA (1988) Advanced Methods of Machining. Chapman & Hall, London.

Table 11.1. Surface finish obtainable from some advanced finishing processes

No. Finishing process Workpiece Ra (nm)

1. Grinding – 25 – 6250

2. Honing11

– 25 – 1500

3. Lapping

– 13 – 750

Abrasive flow machining (AFM) with SiC

abrasives

Hardened steel 50

Magnetic abrasive finishing (MAF) with

diamond abrasives

Stainless steel rods 7.6

Magnetic float polishing (MFP) with CeO

abrasives

4.0

Magneto-rheological finishing (MRF) with

CeO

Flat BK7 glass 0.8

Elastic emission machining (EEM) with

ZrO

abrasives

Silicon <0.5

9. Ion beam machining (IBM) Cemented carbide 0.1

326 V.K. Jain

[4] McKeown PA (1987) The role of precision engineering in manufacturing of the

future. Ann CIRP 36(2): 495–501.

[5] Taniguchi N (1983) Current status in, and future trends of, ultraprecision machining.

Ann CIRP 32(2): 573–582.

[6] Hashish M (1989) A model for abrasive water jet (AWJ) machining. Trans ASME J

Eng Mater Technol 111: 154–162.

[7] Hemanth P, Naqash PV, Ramesh Babu NR (2005) A simplified model for predicting

the penetration of an abrasive water jet in a piercing operation. Int J Manuf Technol

Manage 7(2–4): 366–380.

[8] Shankar P, Jain VK, Sundararajan T (1997) Analysis of spark profiles during EDM

process. Mach Sci Technol 1(2): 195–217.

[9] Kobayashi K (1995) The present and future developments of EDM and ECM. Proc.

XI ISEM, 29–47.

[10] Koshy P, Jain VK, Lal GK (1996) Mechanism of material removal in electric dis-

charge diamond grinding. Int J Mach Tools Manuf 36(10): 1173–1185.

[11] Murthy VSR, Philip PK (1987) Pulse train analysis in ultrasonic assisted EDM. Int J

Mach Tools Manuf 27(4): 469–477.

[12] Pandey PC and Shan HS (1980), Modern Machining Processes, published by Tata

McGraw Hill, New Delhi.

[13] Chryssolouris G (1990) Laser Machining – Theory and Practice. Springer-Verlag,

New York, NY.

[14] Bilgi DS, Jain VK, Shekhar R, Mehrotra S (2004) Electrochemical deep hole drilling

in super alloy for turbine application. J Mater Proces Technol 149: 445–452.

[15] Gedam A, Noble CF (1971) An assessment of the influence of some wheel variables

in peripheral electrochemical grinding. Int J Mach Tool Des Res 11: 1–12.

[16] Dubey AK, Jain NK, Shan HS (2006) Precision micro-finishing by electrochemical

honing, Proceedings of ICOMAST 2006 held at Melaka, Malaysia.

[17] Adhikari S, Jain VK (2007) Some new observations in electrochemical spark ma-

chining (ECSM) of quartz. J Mater Process Technol

doi: 01016/jmatprotec.2007.08.071.

[18] Gautam N, Jain VK (1998) Experimental investigations into ECSD process using

various tool kinematics. Int J Mach Tools Manuf 38 (1–2): 15–27.

[19] Bhattacharyya B, Doloi BN, Sorkhel SK (1999) Experimental investigations into

electrochemical discharge machining (ECDM) of non-conductive ceramic materials.

J Mater Process Technol 95: 145–154.

[20] Jain VK, Chak SK (2000) Electrochemical spark trepanning of alumina and quartz.

Mach Sci Technol 4:277–290.

[21] Jain VK, Dixit PM, Pandey PM (1999) On the analysis of the electrochemical spark

machining process. Int J Mach Tools Manuf 39: 165–186.

[22]

Jain VK, Rao PS, Choudhury SK, Rajurkar KP (1991) Experimental investigations

into traveling wire electro-chemical spark machining (TW-ECSM) of composites.

J Eng Ind 113: 75– 84.

[23] Gorana VK, Jain VK, Lal GK (2004) Experimental Investigation into cutting forces

and active grain density during abrasive flow machining. Int J Mach Tools Manuf 44:

201–211.

[24] Gorana VK, Jain VK, Lal GK (2006) Prediction of surface roughness during abrasive

flow machining. Int J Adv Manuf Technol 31(1/2): 258–267.

[25] Jain VK, Ranganatha C, Murlidhar K (2001) Evaluation of rheological properties of

medium for AFM process. Mach Sci Technol 5(2): 151–170.

[26] Fox M, Agarwal K, Shinmura T, Komanduri R (1994) Magnetic abrasive finishing of

rollers.Ann CIRP 43: 181–184.

Advanced (Non-traditional) Machining Processes 327

[27] Komanduri R (1996) On material removal mechanism in finishing of advanced ce-

ramics and glasses. Ann CIRP 45(1): 509–514.

[28] Singh DK, Jain VK, Raghuram V (2004) Parametric study of magnetic abrasive

finishing process. J Mater Process Technol 149: 22–29.

[29] Jain VK, Kumar P, Behra PK, Jayswal SC (2001a) Effect of working gap and

circumferential speed on the performance of magnetic abrasive finishing process.

Wear 250: 384–390.

[30] Tani Y, Kawata K (1984) Development of high efficient fine finishing process using

magnetic fluid. Ann CIRP 33(1): 377–381.

[31] Umehara N (1994) Magnetic fluid grinding – A new technique for finish advanced

ceramics. Ann CIRP 43(1): 85–188.

[32] Kim JD (1997) Development of magnetic abrasive jet machining system for internal

polishing of circular tubes. J Mater Process Technol 71: 384–393.

Intelligent Machining: Computational Methods

and Optimization

Sankha Deb and U.S. Dixit

Department of Mechanical Engineering, Indian Institute of Technology Guwahati,

Guwahati-781 039, Assam, India.

E-mail: sankha.deb@iitg.ernet.in, uday@iitg.ernet.in

This chapter provides an introduction to intelligent machining. The various com-

putational techniques to achieve the goal of intelligent machining are described.

First, a description of neural networks and fuzzy set theory is presented. These are

soft computing techniques. Afterwards the application of the finite element

method to the machining processes is briefly mentioned. Finally, the optimization

of machining processes is described.

12.1 Intelligent Machining

Machining processes are inherently complex, nonlinear, multivariate and often

subjected to various unknown external disturbances. A machining process is usually

performed by a skilled operator, who uses his decision-making capabilities based on

the intuition and rules of thumb gained from experience. To develop a fully auto-

mated machining system with intelligent functions like those of expert operators is

a very difficult problem. A fully automated system requires the capabilities for

automatic control, monitoring and diagnostics of the machining processes. Such

a system can be compared with a human operator. A number of sensors provide

feedback to the system, in a way the sensory organs provide feedback to a human.

Like the brain of a human, an automated system is equipped with a computer for

processing the feedback information from sensors in real time and taking the appro-

priate decision to ensure optimal operating conditions. For the decision making, an

automated system must have a model of the machining process. The model can be

based on the physics of the process. The physics of the process is understood by

a researcher and is converted in mathematical form, usually in the form of differen-

tial equations. The differential equations of the process are solved by a suitable

technique such as the finite element method (FEM). In many instances, the physics

of the process is not known properly. However, if there is a sufficient amount of

330 S. Deb and U.S. Dixit

data describing the behaviour of the process, a model can be developed based on the

data, which is called modelling based on the data.

Modelling based on the physics is accomplished by the computational tools that

may be called hard computing methods. On the other hand, modelling based on

the data is accomplished by soft computing tools such as neural networks and

fuzzy sets. Soft computing tools try to generate approximate solutions of the prob-

lem in the presence of uncertain or imprecise physics and/or the process variables.

Soft computing differs from conventional (hard) computing in that, unlike hard

computing, it is tolerant of imprecision, uncertainty, partial truth and approxima-

tion. As for modelling, the process optimization can also be carried out by using

either conventional (hard) optimization algorithm or heuristic soft algorithm such

as genetic algorithms (GAs). Similarly, the control of the process may be based on

either conventional (hard) control techniques or soft control techniques such as

fuzzy logic. A hybrid of hard and soft techniques may also be used.

Human beings themselves adapt to changes in the environment. In the same

way, an adaptive control mechanism is an integral part of an intelligent machine.

An adaptively controlled machine is able to adapt to the dynamic changes of the

system caused by the variability of machining process due to changes in the cut-

ting conditions such as the hardness of the work material, tool wear, deflection of

the tool and the workpiece, and so on. The following are the main objectives of an

adaptive control system:

• to adjust the machining parameters such as cutting speeds and feed rates

and/or the motion of the cutter to optimize the machining process by

maximizing some performance criteria based on the cost or the production;

• to satisfy various constraints against variations due to external factors and

respond to such variations in the process in real time;

• to automatically improve the performance of the machining process

through its learning capability.

An adaptive control system that can fulfil the third objective may be said to pos-

sess intelligence as one of the meanings of the word ‘intelligence’ is the capacity

to acquire and apply knowledge.

For monitoring of machining processes and fault diagnostics, the intelligent

machining system must be equipped with the knowledge of how to recognise

failures, how to localize them and how to relate the faults and their effects to the

operating state. Moreover, state classification and process intervention have to be

completed in real time to avoid additional damage when a deviant state has been

detected. There are various signals (force, torque, temperature, mechanical vibra-

tion, acoustic emission, etc.) which correlate with the condition of the machining

process. The control and monitoring algorithms should be based on the simultane-

ous measurement and processing of different signals.

Figure 12.1 shows a scheme of an intelligent machining centre. The intelligent

machining centre functions as follows. The design model of the component to be

machined is provided to the computer. Using suitable software, the tool path is

generated and the spindle speed and feed rate are decided. This information is fed

to the machine controller, which provides signals to the drive motors. The adap-

Intelligent Machining: Computational Methods and Optimization 331

tive controller adjusts the feed rate, spindle speed and tool path, according to

changes in the cutting conditions. The four different types of adaptive control

system are: adaptive control optimization (ACO), adaptive control constraint

(ACC), geometric adaptive control (GAC) and vibration adaptive control (VAC).

The purpose of ACO is to search for the optimum values of feed rate and spindle

speed that maximize some performance criteria such as minimum cost of machin-

ing or maximum production rate. ACC selects the feasible solution of the feed rate

and spindle speed to satisfy the constraints in manufacturing by an algorithm that

is rather simpler than ACO. GAC tries to obtain a highly accurate surface finish by

adjusting the tool offset against the deflection of tools and work material caused

by temperature rise and/or cutting force. The function of VAC is to avoid the vi-

bration or chatter of tools, mainly due to regenerative oscillation, by adjusting the

spindle speed or the resonance frequency of machine tools. Feedback about the

process can be obtained by means of a number of different kinds of sensors, al-

though the figure shows feedback only in the form of temperature, vibration and

force signals.

An intelligent machining system should also possess a memory, like an expert

operator. The knowledge available in machining data handbooks and that acquired

from experience can be stored in the memory of the intelligent machining system.

The system should have the capability to acquire data efficiently, store useful data

by filtering out the noise and retrieve it efficiently. At the same time, it is desirable

that the system is able to communicate with humans in some way, such that its

decisions and actions become transparent to the users concerned. This implies that

the acquired or available knowledge can be stored in the form of an expert system.

In this chapter, the computational techniques and optimization methods that can

be used to develop intelligent machining systems are briefly described. Sec-

tion 12.2 describes the application of neural networks in modelling the machining

processes. Section 12.3 describes fuzzy-set-based modelling. Section 12.4 de-

scribes the hybrid system composed of neural networks and fuzzy set theory, ei-

ther in the form of the mixture or the compound. Section 12.5 briefly discusses the

Figure 12.1. An intelligent horizontal machining centre

332 S. Deb and U.S. Dixit

application of FEM to the study of machining processes. Section 12.6 describes

the optimization techniques useful for optimizing the machining processes. Both

conventional (hard) and non-conventional (soft) techniques are touched upon.

Finally, Section 12.7 discusses the challenges to be met for developing a truly

intelligent machine tool.

12.2 Neural Network Modelling

It is well known that computers can perform a number of jobs at a much faster rate

compared to human beings. However, there are many other tasks which a human

being can perform in a faster and better way. One such task is recognising a face.

A human being can recogonise a face quickly, because the information about vari-

ous attributes of the face is processed parallely in the brain. The brain contains a

neural network consisting of a number of interconnected information-processing

units called neurons. A schematic drawing of a neural network showing only two

neurons and their connections is shown in Figure 12.2. A neuron consists of a cell

body called the soma, a number of fibres called dendrites and a single long fibre

called the axon. The dendrites receive the electrical signals from the axons of

other neurons. The axon transmits the electrical signal from one neuron to other

neurons via the dendrites. The connection between the axon and the dendrite is

called a synapse. At the synapses, the electrical signals are modulated by different

amounts. The synapses release chemical substances that cause changes in the

electrical potential of the soma. When the potential reaches its threshold, an elec-

trical pulse called the action potential is sent down through the axon.

Artificial neural networks are a humble attempt to model biological neural net-

works. Figure 12.3 shows a schematic diagram of an artificial neuron receiving

signals (x

, i

1 to n)

from a number of neurons and emitting signals (o

, I

1 to k)

to be transmitted to a number of neurons. In an artificial neural network, the sig-

nals are in the form of numerical values rather than electricity. The action of the

synapses is simulated by multiplying each input value by a suitable weight w.

Figure 12.2. A schematic drawing of a biological neural network

Intelligent Machining: Computational Methods and Optimization 333

A biological neuron fires an output signal only when the total strength of the input

signals exceeds a certain threshold. This phenomenon is modelled in an artificial

neuron by calculating the weighted sum of the inputs to represent the total strength

of the input signals, and applying a suitable activation (threshold) function to the

sum to determine its output.

Neural networks can be classified based on their topology (architecture) and the

method of training. The most common neural network architectures are: (1) feed-

forward neural networks, (2) feedback neural networks, and (3) self-organizing

neural networks. Feedforward neural networks are the most popular and widely

used ones. There are two types of neural networks in this category: multi-layer

perceptron (MLP) neural network and radial basis function (RBF) neural net-

works. Figure 12.4 shows a feedforward architecture of a typical neural network

consisting of three layers. Each layer contains a number of neurons, depicted by

circles in the figure. Each layer has full interconnection to the next layer but no

connection within a layer. The first layer of the network is known as the input

layer whose neurons take on the values corresponding to different variables repre-

senting the input pattern. The second layer is known as a hidden layer because its

outputs are used internally and not used as the final output of the network. In MLP

neural network, there may be more than two hidden layers. In RBF neural net-

work, only one hidden layer is present. In Figure 12.4, only one hidden layer is

shown. The final layer of network is known as the output layer. The values of the

neurons of the output layer constitute the response of the neural network to an

input pattern presented at the input layer. In MLP neural networks, the number of

hidden layer in a network is an important design parameter. In both MLP and RBF

neural networks the number of neurons in a hidden layer has to be chosen judi-

ciously. The general rule is to design a neural network model which uses fewer

parameters (weights and biases).

In recurrent or feedback neural networks, the outputs of some neurons are also

fed back to some neurons in layers before them. Thus, signals can flow in both

forward and backward directions, as shown in Figure 12.5. A recurrent network is

said to have a dynamic memory. The output of such networks at a given instant

reflects the current input as well as the previous inputs and outputs. An example of

Figure 12.3. A typical artificial neuron