Lopez de Lacalle L.N., Lamikiz A. Machine Tools for High Performance Machining

Подождите немного. Документ загружается.

5 Advanced Controls for New Machining Processes 183

ally regarded as constant (MISO, multiple input – single output). However, re-

search is currently being done into other input variables that are “easier” to meas-

ure, including cutting torque and the instantaneous power consumed by the spindle

speed motor. A diagram for a self-tuning fuzzy controller for a milling process is

shown in Fig. 5.9 [10], where T

is the torque reference value and T

is the torque

measured from an open architecture CNC. As can be seen, a MIMO (multiple

input-multiple output) architecture has been adopted here, with two manipulated

variables, Δs and Δf (percentages of the values programmed in the CNC). The

whole control scheme, including a self-tuning mechanism and the fuzzy controller,

is shown in Fig. 5.9. The output-scaling factor (GC) multiplied by the crisp control

action (generated at each sampling instant) provides the final actions that will be

applied to the CNC:

f(k)

f(k

–

GC Δf(k)

s(k)

s(k

–

GC Δs(k) (5.18)

5.4.2 The Supervisory Control of the Machining Process:

Merit Variables

As mentioned above, the aim of supervisory control is to ensure the optimum

operational management of the process based on one or more merit variables and

the corresponding established objective function. At this level account must be

taken of all the factors that have not been specifically dealt with and compensated

for in the design of the two lower levels. At this level the merit variables pro-

grammed for each individual case must be used also. Machining strategies can

even be programmed that may change during the machining process for a particu-

lar part. An optimum operation such as this may depend on the tasks to be com-

pleted, such as, for example, rough, semirough or final finishing. In any case, this

higher level should supply the lower levels, in real time, with the cutting condi-

tions and process state variables, such as the cutting forces or torque values that

will in turn establish the references of the relevant controllers located at the proc-

ess level. The following can be used as merit variables: 1) maximising productiv-

ity; 2) minimising tool wear rates; 3) avoiding regenerative chatter; 4) minimising

cost; 5) ensuring surface quality (roughness and surface integrity); and 6) ensuring

dimensional quality. Depending on the machining strategy used, as represented by

a certain objective function, each of these variables must be weighted to a greater

or lesser degree using multi-objective optimisation techniques.

After account has been taken of maximising productivity at the process level,

the most significant functions include minimisation of tool wear rates, the avoid-

ance of regenerative chatter and the guarantee of surface quality. Remember,

nowadays people do not just buy machines, or even optimised processes. They

buy production, that is to say, parts; and ideas such as “right the first time” there-

fore become fundamentally important.

184 J. Ramón Alique and R. Haber

Fig. 5.10 Stability lobes diagram: depth of cut/spindle speed (white: stable zone, grey: chatter

zone). Example of thin-floor machining (1

mm thick) using an axial depth of cut 0.5

mm, per-

formed tests and diagrams at the University of the Basque Country

Merit variables should be controlled at this supervisory level, though the impor-

tance of estimating and predicting some variables must always be borne in mind.

Measuring and predicting these variables will be discussed in Sect. 5.5. This sec-

tion will deal only with some of the ideas concerning automatic control systems.

Current specialist literature uses two kinds of methods to address the issue of

chatter avoidance; this machining problem is described in Sect. 3.5.3. The first is

based on stability lobes diagram (off-line), and the second, on cutting speed modi-

fication (on-line).

Stability lobe diagrams propose two areas on the plane formed by depth of cut

vs spindle speed (Fig. 5.10). The upper area is unstable, while the lower area is

stable. A number of methods have been investigated for identifying the stability

lobe diagrams for a particular machine and a particular tool. In general, the inves-

tigated methods display a lack of precision. There is therefore a tendency to de-

velop analytical and experimental methods in which the transfer functions of exist-

ing systems with multiple degrees of freedom are identified by non-destructive

structural dynamic tests (a piezoelectric hammer). That way the natural frequency,

the stiffness of the mechanical system and the damping factor for each mode of

vibration can be found through experimentation [32].

The most commonly used method for suppressing chatter automatically is to

adjust the spindle speed set point. One heavily used technique is to vary the spin-

dle speed, usually in a sinusoidal manner. While such methodologies may be

promising, their main disadvantage is that there is no theory to support them. No-

body knows how to produce optimum adaptive modulation.

5 Advanced Controls for New Machining Processes 185

Another important function to implement at this supervisory level is monitoring

of the surface roughness process (including surface integrity), with the imposition

of cutting conditions that ensure the desired surface roughness in the final ma-

chined part. An automatic surface roughness control system could be implemented

by using a sensor unit that can predict surface roughness during the process. To do

so, it would be necessary to analyse the factors that affect surface roughness Ra

and modify those values during the process in order to obtain the desired Ra val-

ues. As indicated in Sect. 5.5.3, there is an extensive series of factors that affect

Ra. Of said factors, feed and cutting speed control would seem to be the most

viable industrial solution. Other variables, for example, depth of cut, require the

part program to be rewritten, so they rule out automatic Ra control. By controlling

the emergence of regenerative chatter, feed rate and cutting speed, an effective

automatic control process could be developed for the surface roughness of ma-

chined parts.

5.5 The Sensory System for Machining Processes

As already mentioned, if an “ideal machining unit” is ever achieved, a machining

sensor system will be a key element for identifying how the machine, process and

cutting tool are all behaving. Such sensors plus the post-processing units form

what is identified in Fig. 5.8 as the monitoring units for the machine, the process

and the tool.

The various monitoring strategies are classified as “pre-process”, “in-process”

or “post-process”, depending on the time scale at which monitoring is performed,

and they are divided into off-line and on-line methodologies. Some strategies are

classified on the basis of the ultimate purpose of the monitoring process; these

include tool wear monitoring (TWM), tool condition monitoring (TCM), the

monitoring and the control of machining, surface quality prediction, tool wear

detection (TWD), tool failure detection (TFD), real-time prediction, machine

health monitoring, chatter stability prediction, heat generation and temperature

prediction.

The following points can be made about off-line and on-line methodologies:

Off-line monitoring

• Pre-process. This type of monitoring is based more on an analytical study that

attempts to predict the different parameters that will influence the process. It

may be supported by an analysis of previously monitored results. Basically,

what is necessary for this kind of monitoring is the appropriate IT tools to run

process simulations. However, the initial set-up requires prior measurement (di-

rect measurement) of all the elements or parts whose dimensions are known to

vary. The direct measurement of these elements or parts is usually performed

using instruments such as electronic or optical microscopes, optical cameras,

roughness meters, hardness meters, micrometers, Vernier scales, etc.

186 J. Ramón Alique and R. Haber

• Post-process. Just as measurements must be taken before the process to establish

the initial conditions, the final condition of the elements being studied (mainly

the part and the cutting tool) must also be checked after machining is complete.

On-line monitoring

Machine status monitoring is seen as a maintenance task that is independent of the

actual machining process and prepares the machine for beginning or continuing

the machining process. In this case, the monitoring process will take account of

factors such as the condition of the bearings and gears, the emergence of vibration

in the moving parts (particularly under non-load conditions), the suitability of the

temperature of machine components and the consumption of the drives.

Table 5.1 Measuring instrument type classified by machining signals

Signal

type

Physical

quantity

(sensor type)

Location

Parameter to

process

Measurement

Power Motor shaft,

electric cabinet

Spindle power

Feed shaft power

Cracking and tool

wear

Cutting power/

driving power ratio

Torque

(static,

dynamic)

Table, tool Tool torque Tool wear

Tool fracture

No cutting chip

condition

Thread depth

Distance/

displacement

Force transmission

element

Indirect force meas-

urement

Crack identification

Pressure forces

Force

1–3 axis force

sensor

Area under forces

parallel to dis-

placement axis

Single- or multiple-

axis force

Process identifica-

tion

Acoustic

emission

(AE)

Linked to work

piece. Near signal

source

Proportional cutting-

force signals

High-frequency

energy signals of

cutting process and

tool fracture

Acoustic

emission

(AE) fluid

sensor

Near noisy area Proportional cutting-

force signals

High-frequency

strength waves of

work pieces in

movement or rotat-

ing components, or

materials with very

rough surfaces

Acoustic

emission

Rotating

acoustic

emission

(AE) sensor

Near signal source Proportional cutting-

force signals

High-frequency

energy signals of

cutting process and

tool fracture

5 Advanced Controls for New Machining Processes 187

Table 5.1 (continued)

Signal

type

Physical

quantity

(sensor type)

Location

Parameter to

process

Measurement

Accelero-

meters

Near signal

source: spindle or

work piece

Mechanical vibration

of machine structure in

machining process

Tool wear

Give increase

Collisions

Excess vibrations

in ball bearing

and spindles

Ultrasonic

sensor

Near signal source Ultrasonic-range

vibrations

Tool wear

Give increase

Collisions

Excess

vibrations in ball

bearing

and spindles

Vibration

Laser barrier Perpendicular to

measurement

surface

Tool fracture and wear Thermal

deformations

Work-piece

roughness

Chatter marks

Feed marks

Thermo-

couple

In contact with

work piece

Temperature effect on

wear and machining

process

Temperature

gradient

Tempera-

ture

Thermo-

graphic

camera

Near signal source Temperature effect on

wear and machining

process

Temperature gradi-

ent distribution in

cutting process

Although some in-depth research has already been done in this area, it is not

yet widely applied, and only very simple sensors are being used, generally based

on the application of constant-value upper and lower limits throughout the entire

machining process.

The aim of monitoring the machining process is to detect disturbances or oper-

ating modes that have a negative effect on the process and part quality. Process

monitoring can include all the phenomena that bear any correlation with the sur-

face quality of the part being machined (e.g., vibration), the factors that result in

machine process planning errors (e.g., precision in terms of dimensions) and un-

expected machining conditions (e.g., collisions).

The most appropriate measuring instruments to select depend on process and

computing-power constraints, as well as the variable that is to be monitored, and

these depend in turn on the parameters that are to be estimated. The literature

published in this field suggests the measuring instruments listed in Table 5.1,

depending on the goals and constraints.

188 J. Ramón Alique and R. Haber

5.5.1 Correct Monitoring Conditions

Attempts to develop the appropriate monitoring instrumentation reveal that both

the cutting process and the machine tool have limitations. The most significant are

as follows:

• Spindle speed.

− High-order frequencies caused by mechanical vibrations.

− Appearance of high-order harmonics.

• Bandwidth, sensitivity, impact limit and resonant sensor frequency.

− Bandwidth. A high bandwidth should be considered if the presence of high

main frequencies and high harmonics is suspected.

− Sensitivity. Critical when measuring signal values that are very small but

very significant.

− Impact limit. May not be so sensitive if the equipment does not involve ele-

ments that are subject to direct impact.

− Resonant frequency. This could be considerably affected, depending on the

equipment. Could become saturated unexpectedly early. The possibility of

harmonics appearing above the main frequency increases with the number

of moving parts that come into contact with one another at high relative

speeds.

• Sampling frequency: High main frequencies require high sampling frequencies,

for the acquisition of the largest amount of data possible for each tool cycle.

• Acquisition resolution: This determines the precision with which measurements

are taken. The greater the resolution, the greater the precision.

• Download speed: At high acquisition speeds, the card must be able to

download the data into the memory quickly. The internal memory must be

large enough to be able to store all the data from the sample.

• PC requirements: In real-time monitoring and control, any acquired data must

be processed as rapidly as possible. The PC processing capacity must be such

that it can store the information received on its hard disk before the memory

used for acquisition is entirely used up.

• Platform for communication with the PLC: This must allow the flow of data

between the PC and the PLC (Programmable logic controller) to happen in real

time. This means that the protocol must allow for high data transmission

speeds; communication with the PLC must be as direct as possible, and any in-

termediate delays must be avoided; and finally the operating system must be in

real time in order to guarantee the least possible loss of time in the sending and

receiving of data between remote applications and the monitored system. The

advantage of this type of operating system lies in the way it manages priorities.

In contrast to other systems like Windows™ and Linux, these systems assign

greater priority to the execution of the applications that require it.

5 Advanced Controls for New Machining Processes 189

5.5.2 Machining Characteristics and their Measurement

References [24,

25] have identified and classified the most important features that

must be monitored during machining. Each feature to be monitored can be identi-

fied with one or more measuring parameters. Table 5.2 shows these machining

features and their potential measurement factors, offering an idea of the flexibility

with which one or more features can be monitored.

The function performed by monitoring each feature is shown in Table 5.3 [25].

Each function shown in this table can be performed using off-line and on-line

strategies. Everything related with estimation forms part of the off-line (and, in

this case, pre-process) strategy, while everything related with maintenance forms

part of the on-line, in-process monitoring.

The general structure of a condition monitoring system (CMS) is shown in

Fig. 5.11 [1] and contains the following functional units: 1) sensory signals, 2)

signal processing, 3) extraction of sensory characteristic features, 4) decision mak-

ing and classification, and 5) corrective actions. The aim in practice consists of

selecting a multi-sensor system and ways of processing the resulting signals that

will allow for a decision-making and classification function with minimum levels

of error as regards possible process failures. As Table 5.3 shows, various meas-

urement parameters can be used to evaluate machining characteristics.

Table 5.2 Characterisation of machining and measurement factors

Measurement

parameters

Feature to monitor

Chatter

Cutting

forces

Chip

forming

Cutting

temperature

Tool

wear

Tool

failure

Surface

errors

Cutting depth × × × ×

Spindle speed × × ×

Acoustic emission × × × ×

Cutting force × × × × × ×

Feed vibrations × ×

Feed × × × ×

Tool wear ×

Infrared image ×

Chip contact length ×

Surface finish ×

Part dimension ×

Tool geometry ×

Tool vibrations × ×

Spindle power ×

Tool deflection ×

Cutting speed ×

Cutting temperature

190 J. Ramón Alique and R. Haber

Fig. 5.11 The general structure of a condition monitoring system

Table 5.3 Functions associated with the features monitored

Feature Function

Chatter Estimate chatter

Avoid chatter

Suppress chatter

Cutting forces Estimate cutting forces

Maintain specs

Chip formation Estimate formation

Maintain specs

Cutting temperature Estimate temperature

Maintain specs

Tool wear Estimate wear

Maintain specified wear rate

Compensate for wear

Tool failure

Estimate, avoid and detect failure

Surface errors Estimate surface finish and tolerances

Maintain specs

A detailed analysis of tool condition monitoring (TCM) and surface quality

prediction is given in Fig. 5.11, in view of the particular importance of these two

processes.

5.5.3 Two Case Studies

For many years now, a great deal of research has focused on the area of monitor-

ing tool conditions, given that tool failure is the cause of around 20% of machine

down time and tool wear has a clear impact on the quality of the parts produced,

5 Advanced Controls for New Machining Processes 191

their dimensional precision, their surface quality and their surface integrity [9].

Artificial vision techniques have been successful in the laboratory and have given

very precise results. Nevertheless, they have only worked in the laboratory and

always in a post-process context, i.e., when the tool is no longer in contact with

the part. The illumination has turned out to be a highly critical element. The ma-

jority of the developments made in the laboratory have been restricted to 2D

measurements. Few results have been produced in 3D, and these only for crater

wear.

The forces, both static and dynamic, that appear during machining are affected

by tool wear. The radial force component would seem to be the most sensitive to

wear. Sometimes the relationships between forces are utilised. The relationship

between feed force and cutting force would seem to be highly sensitive to wear on

the flank of the tool.

Likewise, during the cutting process, the part undergoes considerable plastic

deformation, which generates acoustic signals (acoustic emissions). These high-

frequency acoustic signals (50

KHz to 1

MHz) are highly sensitive to tool wear.

A strong correlation has been shown between the acoustic emission RMS value

and tool wear. The variance is one of the most sensitive parameters, with the high-

est range being shown at the end of a tool life. These laboratory results have not

yet been put into industrial practice either, due to the lack of sufficient knowledge

about the physical significance of acoustic emission signals and the great sensitiv-

ity of such signals to sensor location.

To date, the operation of almost all systems used to monitor tool condition has

been based on very simple strategies that involve detecting when tool condition

trespasses some specific pre-set limit. These strategies need to be recalibrated

constantly, given the wide variation in the conditions under which machining is

done. Adding greater “intelligence” to the monitoring process, along with the

capacity to adapt to the production process itself, while maintaining a simplicity of

operation for the end user is still the focus of research for many research teams.

From the most recent results reported, it can be concluded that the sensors most

commonly used to monitor tool condition are: 1) acoustic emission sensors,

2) vibration/acceleration sensors, 3) static and dynamic cutting force sensors, and

4) cutting torque sensors. The place where sensors are located on the machine is

an important factor to consider, since the system must, in all cases, be as non-

invasive as possible.

Once the signals received from the sensors have been properly processed, pat-

terns must be extracted, either in the time domain (time series models, e.g., AR,

ARMA) or in the frequency domain. Once patterns have been extracted, the deci-

sion-making process must take place. For this, two types of methods may be used:

statistical/stochastic methods or artificial intelligence (AI) methods. The most

heavily used statistical methods are statistical analysis, time series analysis and

discriminant analysis.

There is currently a wide variety of types of AI methods: 1) fuzzy logic, 2) arti-

ficial neural networks, 3) hybrid systems (e.g., neurofuzzy systems), and 4) prob-

abilistic networks (e.g., Bayesian networks). Neural networks are currently widely

192 J. Ramón Alique and R. Haber

used in research projects, in which both supervised networks (multi-layer percep-

tron, among others) and non-supervised networks are used. The latter are currently

very much in fashion, with networks such as ART (adaptive resonance theory),

KSOM (Kohonen maps) and RBF (radial basis function). All these neural struc-

tures are of the competitive-learning type.

Tool condition has a noticeable effect on surface roughness, surface integrity,

dimensional quality, machine vibration levels and machine performance in gen-

eral. It is therefore important to ensure that these systems not only account for the

fresh/worn dichotomy, but also cover intermediate wear parameters that allow for

the development of more effective, sophisticated tool change strategies.

Surface quality is measured in terms of what is called surface integrity, which

not only describes the topology of a surface but also takes account of its mechani-

cal and metallurgical properties that subsequently play a vital role in terms of

fatigue strength, resistance to corrosion and the part’s service life. Highly sophisti-

cated laboratory techniques are generally used, such as X-ray diffraction and met-

allographic inspection. These are always applied post-process and in all cases cost

both time and money.

Surface topology is represented by the so-called surface texture that measures

several quantities related with the deviations in a part as compared with the nomi-

nal surface. Deviations are classed from first to sixth order, in accordance with

international standards. First- and second-order deviations refer to form (i.e., flat-

ness, circularity, etc.) and to waviness, respectively. Surface roughness refers to

deviation from the nominal surface of the third up to sixth-orders. Ra is the most

widely used parameter to describe surface texture due to its direct influence on

friction, fatigue, electrical and thermal contact resistance, and appearance [20].

Nowadays roughness measurements are taken using modern roughness meters

(profilometers), though their application is restricted to the post-process applica-

tions, with the corresponding inherent problems such as inspection times, the need

for random testing (i.e., not every part produced is tested) and the production of

waste, since the fact that inspection occurs post-process means that waste produc-

tion cannot be avoided, with all that this entails in terms of both cost and the effect

on the environment. At the present time, the quality assurance process involves

two phases. The first uses Taguchi-type methods during the design stage to ensure

the quality of the product being designed. This is a pre-process action, making it

essential to incorporate a second intermediate stage to ensure quality during the

machining process.

Ensuring the surface quality of a part during the machining process is no easy

task. Vision systems have been used which involve measuring reflected light in-

tensity in order to estimate surface roughness. However, the practical results have

as yet not offered much hope, as the measurements obtained have not been very

precise or reliable. Neither sensors based on artificial vision nor sensors based on

acoustic emissions have actually worked on the factory floor. This has made it

essential to develop sensors based on predictive models.

Predictive models can basically be separated into two main categories: models

based on machining theory, which are analytical models that could be termed