process control and dynamic process planning

International Journal of Machine Tools & Manufacture 40 (2000) 239–257

Process control and dynamic process planning

R.J. Seethaler, I. Yellowley*

Department of Mechanical Engineering, University of British Columbia, Vancouver, V6T 1W5, Canada

Received 20 January 1999; accepted 17 May 1999

Abstract

Real time machine tool control and the planning activities which precede manufacture are usually inter-faced through a low level language which allows little more than position, feed, and speed information tobe passed between the two systems. The higher level systems used to describe geometry and tool pathsalso lack an adequate capability to describe manufacturing processes. The authors discuss the provision ofa much richer interface between the planning and control activities which both facilitates the identificationand scheduling of suitable monitoring tasks and allows the updating of process plan data from real timemeasurements. The result of such integration is an improvement in the efficiency of real time optimisation,and perhaps most importantly the possibility of quasi real time process planning. A system that is able toperform both initial process planning and plan refinement based upon low level feedback must alsoencompass the path generation activity, such a system is referred to by the authors as a dynamic processplanning system. The paper describes the fundamentals of the process models, identification algorithms,control strategies, and low level process plan generation used within such an integrated system. 1999Elsevier Science Ltd. All rights reserved.

Keywords:Process planning; Process control; Open architecture; CNC

1. Introduction

Process planning refers to the translation of design information to a suitable production plan;the activity consists of several phases which are hierarchical in nature, starting with relativelyhigh level decisions, and proceeding to detailed planning. The typical requirements of the phasesare shown in Table 1.

Generative process planning systems attempt to generate process plans from a CAD databaseby relying solely on models of the manufacturing process (i.e. without human interaction). This

* Corresponding author. Tel.: 001-604-822-2781; fax: 001-604-822-2403.E-mail address:[email protected] (I. Yellowley)

0890-6955/00/$ - see front matter 1999 Elsevier Science Ltd. All rights reserved.PII: S0890-6955(99)00054-1

240 R.J. Seethaler, I. Yellowley / International Journal of Machine Tools & Manufacture 40 (2000) 239–257

Table 1Process planning

High level planning phase (A) Selection of basic processes and sequence(B) Selection of specific machines and order(C) Design of holding devices, approximate ordering of main operations at eachmachine(D) Subdivision of operations, detailed operation order selection of tool types

Low level planning phase (E) Optimization of cutting conditions and toolsNC programming phase (F) Detailed evaluation of tool paths

(G) Cost estimating

approach has met with several difficulties. The major problem is the encountering of infeasibilityduring the planning process, as it proceeds from its highest to the lowest decision levels; suchan event will require backtracking. The very large number of feasible solutions makes findingthe best solution unlikely, and finding even good solutions rather difficult. Progress has beenmade in developing efficient algorithms for the low levels of process planning, both for singlepass [1–3], for multiple pass [4–6], and for the subdivision/scheduling of volume and operationsrespectively [7–9]. One should realise however that most of the methods suggested for processoptimisation rely on a knowledge of part geometry and process parameters which is unlikely tobe available in the initial planning stage. A truly optimal process plan leaves little room for errorin estimates of process parameters or indeed part geometry. Such errors will inevitably lead tothe failure of tools, and damage to machines and work pieces. Computer generated process plansthen usually use conservative estimates of process parameters that will lead to the safe operationof the machine tool, but generally far from optimal performance.

A partial solution to this problem is the provision of real time process control. Such a systemneeds significant intelligence if it is to be suitable for general use. The system should be able toidentify current geometry, process parameters and tool condition, it also needs the capacity toselect feeding velocity (and hopefully cutting speed although this is more difficult). The systemat best will be capable of achieving close to economic cutting conditions on each feature withminimal risk of constraint violation (breakage, torque, power, surface finish, etc.). The systemhowever cannot modify the higher level process plan (the sequencing of machining operationsand the selection of tool paths). (It also should be realised that there will be some extreme transi-ents which will defeat the best of real time control/identification algorithms without some formof warning.)

The authors believe that a real time process control system should also have the capacity tofeed back identified process parameters to the process planning system. The identified parameterscan then be used to optimize the higher level process plans for the remainder of the part, orsubsequent parts of the same batch.

Fig. 1 shows the requirements of such an integrated system. The real time process controlsystem running on the machine tool must be supplied with expected process parameters (thisinformation concerns workpiece, cutter, type of operation, etc., and is normally available withinthe planning system but rarely at the machine tool level). The extension of the interface betweenthe planning system and the control system should also include the capability of sending identifiedprocess parameters back to the planning system as described earlier. Most importantly the planning

241R.J. Seethaler, I. Yellowley / International Journal of Machine Tools & Manufacture 40 (2000) 239–257

Fig. 1. Dynamic process planning.

system should allow recovery from tool breakage or other catastrophic events by providing aframework for the development of new feasible strategies using remaining resources.

The paper which follows describes the elements required to allow the realisation of a real timeprocess control system. A rudimentary demonstration system is described and the application ofthe identification monitoring and control elements demonstrated using rough milling operationsas an example.

2. Process identification

A practical real time control system for milling operations needs to be able to avoid the majorconstraints (tool breakage and maximum torque), it also needs to be able to identify tool condition(both breakage and excessive runout). Ideally of course one would wish to track tool wear andsurface finish, these are presently difficult to achieve in a practical machine tool environment. Toachieve the requirements outlined, it is usually easiest to identify the current cutting conditionsfirst and then to use a comparison of actual and expected values of forces to identify tool condition,both processes are achieved in the frequency domain. Since the authors make considerable useof force relationships in the development of the real time system, a brief derivation of the requiredmaterial follows.

2.1. Force model

The force model shown in Fig. 2 is used to describe the forces in both end and face milling.The tangential and radial forces are assumed to be proportional to the undeformed chip thickness:

Ft=Kast sinf

for f1#f#f2

Ft=0

for 0,f#f1 andf2,f#2p

Fr=r1Ft

(1)


Fig. 2. Milling geometry.

whereFt=tangential force component,Fr=radial force component,Fx=feed force,Fy=force normalto feed force,N=number of flutes,K=specific cutting pressure,a=axial depth,r1=force ratio,st=feed per tooth,f=cutter rotation angle,f1=insert entry angle,f2=insert exit angle,fm=meanimmersion angle, andfs=swept angle of cut.

These forces are related to the feed and normal forces through the following transformation:

Fx=Ft cosf+Fr sinf

Fy=Ft sinf−Fr cosf(2)

The periodic nature of the cutting forces in milling lends itself well to frequency domain analysis.The convention for representing in plane cutting forces for milling operations in this paper isas follows:

Tr5RKaFar0

21O`

k51

ark cos(kf)1brk sin(kf)G (3)

It can be shown that when the major force components are assumed to be proportional to theinstantaneous area of the undeformed chip, the mean cutting forces are given by the followingexpressions [10]:

F̄x=KaNst

8p{(cos(2f1)−cos(2f2))+r1((2fs)+sin(2f1)−sin(2f2))}

F̄y=KaNst

8p{( −r1(cos(2f1)−cos(2f2)+(2fs)+sin(2f1)−sin(2f2))}

(4)

2.2. Identification of instantaneous cut geometry

In face milling with a given diameter of cutter, the cutting geometry is uniquely defined bythe insert entry anglef1, the insert exit anglef2, and the axial depth (a), (see Fig. 2). It is difficultto identify the two angles directly from cutting forces. However, the swept angle of cutfs (the


difference between these two angles) and the mean immersion anglefm (the average between thetwo angles) can be identified fairly readily from Fourier series coefficients of an orthogonal pairof in plane forces.

fs=f2−f1

fm=12(f2+f1)

(5)

A previous method for identification of the swept angle of cut has been presented which uses theratio of fundamental cutting force components over mean cutting force components. The algorithmhas the advantage that is it is largely independent of wear and cutting direction, and it is alsofairly easy to obtain [10].

The mean immersion angle is found from the ratio of the mean forces in thex and iny directionsas follows:

F̄y

F̄x

5

S r1(cos(2fm+fs)−cos(2fm−fs))

+sin(2fm−fs)−sin(2fm+fs)+2fsD

S(sin(2fm−fs)−sin(2fm+fs)+2fs)

−cos(2fm−fs)+cos(2fm−fs)D (6)

This relationship can be solved analytically for the mean immersion angle, the resulting quadraticrelationship is however quite complex. A very good approximation to the solution is found byinspection as follows:

fm<f∗m5tan−1SF̄y

F̄xD1tan−1(r1) (7)

The axial depth is obtained from the magnitude of the quasi mean resultant force and knowledgeof the immersion angles. If edge forces are neglected, then the quasi mean resultant force can beexpressed in terms of the fundamental of the torque series:

ÎF̄x2+F̄y

2=NKast

2Î1+r1

2Îa12+b1

2

=NKast

4pÎ1+r1

2Îfs(sin 2f1−sin 2f2)+sin2 f2+fs2

(8)

This relationship is now inverted to allow one to find the axial depth:

a54p

NKst! F̄x

2+F̄y2

(1+r12)(fs(sin 2f1−sin 2f2)+sin2f2+fs

2)(9)

2.3. Cutting parameter identification

When the cutting geometry is known exactly, calibration cuts can be used to identify the cuttingparameters. The mean forces inx andy give a good estimate of the cutting pressure and the force


ratio. Solving Eq. (4) for the cutting pressureK, and the force ratior1 yields the followingexpressions:

K58p

NKst

a1F̄x−b1F̄y

a12+b1

2 (10)

r15b1F̄x+a1F̄y

a1F̄x−b1F̄y

(11)

Tool condition monitoring is important from both an operating safety standpoint and a partquality perspective. Badly worn tools, or tools with excessive runout can lead to unacceptablesurface finish and poor part tolerances. If the tool fails catastrophically, then the machine andoperator safety are compromised. Tool wear rate plays a role in the machining cost and ideallyshould be selected in a manner which ensures the minimisation of process cost. Even thoughalgorithms for tracking tool wear from force measurements have been suggested [11–13], andprototype sensors have been fabricated, no commercially viable technique is yet available for therobust measurement of this parameter. The identification of tool breakage and runout, however,has been studied with more success [14]. In this work, an identification algorithm for relativeradial runout of individual teeth on a milling cutter in the general case is applied. The provisionof a generalized approach for face milling has great practical importance because the characteriz-ation of a simple eccentricity will rarely be sufficient. The examination of the specific case ofinserted tooth cutters, for instance, means that the geometry of individual pockets and inserts aremore important than spindle runout. It is also possible to utilize this approach to identify breakageas a change in relative runouts as cutting progresses. Such a method is then able to cope withrelatively large initial runouts and to identify the changes caused by a subsequent edge chip orbreakage. Fig. 3 illustrates the relationship between the absolute radial size of teethRt and therelative runout between teethArt. The relative runoutArt is the difference between the radial sizeof the current toothRt, and the radial size of the previous toothRt21 The total sum of the individualchip thickness added as a result of runout in a single revolution will be zero in the absence of

Fig. 3. Radial runout for individual teeth.


dynamics. The actual algorithm to infer relative radial runout from in-plane force measurementshas been described previously by the authors [14].

3. Real time process control

The inherent variability in tools, workpieces and indeed machine tools means that tool life,wear rates and even detailed force information cannot be predicted reliably during the initialprocess planning stage. In order to minimize cost, cutting speed and feed rate must be optimizedin real time, this implies that multiple constraints must be dealt with. An appropriate controlarchitecture should also allow the modification of the tool path itself in order to cut excessmaterial, or to ensure the maintenance of tolerances in the presence of wear and machine inaccur-acies.

There are two well publicised approaches to the real time optimization stage of metal cutting.The first approach is called adaptive control constraint (ACC); ACC attempts to optimize theprocess conditions by maximizing the metal removal rate. This is usually achieved by maintaininga single constraint such as force or torque at its maximum value. Since this approach will dramati-cally change the control system parameters with changing process parameters such as depth ofcut, system stability has been of great interest to many researchers. The solution to this problemis usually a parameter adaptive controller, where controller parameters such as gain are varied inorder to ensure system stability over a wide range of cutting conditions ([15,16]). This approachis however only promising if the optimum conditions are really governed by a single constraint.

The second approach to real time process optimization is called adaptive control optimization(ACO); it attempts to optimize a performance index based on an economic model. This approachhas not been pursued particularly vigorously, since a realistic economic index is difficult to define,and the required sensory information is often unavailable. Such a system requires one to achievean optimal cost while satisfying all the constraints, (tool breakage, torque, power and perhapssurface finish), (see Fig. 4).

Fig. 4. Adaptive, multiple constraint, optimization.


3.1. Control of cutter feedrate

The system demonstrated later in this paper selects feed on the basis of both a tool edgebreakage constraint and a tool shank breakage constraint. The tool edge breakage constraintensures that the equivalent chip thickness on each tooth stays below a set maximum value, andthe tool shank breakage constraint requires that the in plane resultant force also be constrained.

For tool edge breakage control, the equivalent chip thickness is determined from identifiedvalues of the axial depth of cuta, the maximum instantaneous undeformed chip thickness foundover one spindle revolution,h0max

, and the length of the active cutting edgele:

hemax5

h0maxa

le(12)

The maximum chip thickness is found from the feed per tooths, and the tool entry and exitangles (see Fig. 5):

if f2,p2→h0max

5st sin(f2) (13)

if f1.p2→h0max

5st sin(f1) (14)

if f1,p2

,f2→h0max5st (15)

The length of the active cutting edge is dependent on the tool geometry. (Particularly in the caseof faceted or radiused cutters.)

The system demonstrated in the final portion of this paper then applies two constraints to therealtime control of cutter feedrate. The command input to the system is formulated for each

Fig. 5. Maximum chip thickness for varying immersions.


constraint and the binding constraint used. For each constraint a simple ratio of identified para-meter and maximum allowable is utilised as follows.

FROnewedge5FROcurrent

heallowable

heidentified

(18)

and

FROnewshank5FROcurrent

Frallowable

Fridentified

(19)

The lower of these two feedrates is then implemented by the process control system once perspindle revolution:

if FROnewedge#FROnew

shank→FROnew=FROnewedge

if FROnewedge.FROnew

shank→FROnew=FROnewshank

(20)

3.2. Speed control

The optimisation of cost or production rate in milling requires that the feed is kept at amaximum value that is governed by tool breakage and machine torque constraints. The tool break-age condition relies on identified cutting geometry and needs to consider the intermittent natureof the cutting process in milling. The spindle speed is used to regulate the wear rate of the tool,which should be kept at an optimum value. Unfortunately there are no robust identification tech-niques available to determine wear rate; thus, spindle speed needs to be selected using approximatemodels of the process. It can be shown however, that the influence of slightly suboptimum cuttingspeeds will not affect the cost of the cutting operation significantly. The effect of feed and speedon the machining cost is demonstrated in Fig. 6, where cost is plotted against variations in spindle

Fig. 6. Machining cost functions for speed variations.


speed for three different feed rates. Clearly, the cost is affected little by±10% variations in spindlespeed from the optimal speed (this is for typical tool life exponents). Thus, a priori selection ofspeed is adequate in most circumstances. (It should be noted too that information from the operatormay be used to improve initial estimates.)

3.3. Miscellaneous control functions

A process control system for real machine tools needs to be able to handle a variety of situationswith different requirements, the previous discussion has only dealt with identification, optimisationand adaptive control in cutting. Other situations involve for instance rapid moves in air whereone is checking for collisions and cuts where one knows the cut dimensions with certainty andmay be required to send calibration values back to the planning system. Typical monitoring andcontrol tasks for varying cutting conditions are shown in Fig. 7.

4. Dynamic process optimization

A dynamic process planning system (as defined in this paper) allows for even more flexibilitythan the selection of appropriate real time control strategies. It needs to be able to collect identifiedprocess parameters and update high level process plans in real time. Such a system must copeefficiently with a number of machining conditions that traditionally either led to suboptimal manu-facturing conditions or required operator interference. The following paragraphs describe someof the functions that a dynamic process planning system should perform for different sets ofsensory feedback.

Tool wear and variations in the geometry of the work piece are the primary unknowns duringthe traditional planning process. Given good identification techniques for these variables, dynamicreplanning of higher level process plans is the first step in a dynamic process planning system.Tool wear sensors, as mentioned earlier, are unavailable; thus, operator input needs to be relied

Fig. 7. Monitoring and control functions for a real time process control system.


Table 2Machine data

Machine data base

Maximum torque Maximum powerMachine capital cost Tool change time

on for generating tool life data that can then be used to update the process plan. Variationsin geometry can be identified by the process control system that then uses this data to controlfeed rate.

Besides compensating for the deviations in wear rate and work piece geometry, a dynamicprocess planning system is also able to recover from emergency situations such as tool breakage.Tool breakage requires that the feed is stopped immediately after the condition is detected andthe tool safely retracted before a tool change is commanded. Often, identical tools to that brokenwill be unavailable, and the planning system will be required to replan the remaining part withthe tools that are still available. If more of the same parts need to be machined, then the entireprocess plan for the remaining batch need to be updated.

There are other phenomena which may lead to the reexamination and replanning of the highlevel plan, typical examples are the occurrence of chatter, and the production of an inadequatesurface finish. One would hope that the real time process control system is able to allow theavoidance of the worst effects of these phenomena, however high level replanning to avoid theconstraints must be incorporated.

Clearly there is a wide range of applications where an integrated planning and control systemcan be beneficial, especially for small batch sizes, or work pieces with variable geometries. Anysuch system is required to share data between the planning and the control system. The followingsection will demonstrate the data that is required for this fusion.

5. Shared planning and control data bases

Process planning and process control need to share a common data base. This database containsinformation on the manufacturing equipment and the process conditions. Some of this data istime-invariant, but most of it will change with varying manufacturing operations. The equipmentdata base contains information on the machine tool, the available tools, and the work material.Clearly the information on the machine tool is time invariant. The material data will change withevery new part, and the tool database needs to be updated continuously, since tools will wearand break during cutting operations. A listing of important parameters in the equipment databaseis shown in Tables 2–4.

Table 3Material data

Material data base

Material family Material hardness


Table 4Tool data

Tool data base

Cutter radius Number of flutes Flute angle Shank length Maximum axial depthEffective rake angle Nose radius Maximum equivalent Maximum force for Maximum feed for

chip thickness shank breakage control roughing/(semi)finishing

Fig. 8. Transient cutting conditions.

The process data base contains information on the geometrical and technological process para-meters. The geometrical process parameters need to reflect the width and depth of cut, as wellas the certainty of these parameters, since this certainty will indicate to the control system whetherthe upcoming operation can be viewed as a normal cutting operation or a calibration operation.

The technological process parameters include a description of the cutting process which tellsthe control system whether the cutter is expected to be in air, in a transient cut, or a normal cut.In both transient and normal cuts material is being removed from the work piece, but a transientcut is defined as a cut where axial depth or radial width are expected to change during the cut.This situation is encountered at all tool entry and exit conditions. A diagram of these cuttingprocesses is shown in Fig. 8. The technological process parameters also define whether a roughingor a finishing operation is taking place and what the required tool wear rate is for controlling thespindle speed. It is also possible to include maximum allowable force limits for part accuracyconstraints in this part of the process data base. A list of important parameters for the processdata base is shown in Table 5.

Table 5Process data

Process parameters

Axial depth Radial width Cut: up/down Calibration: On/OffMovement: Operation: Prescribed wear rate for Maximum force for partair/transient/normal Roughing/(Semi)Finishing speed control accuracy


Table 6G-Code extensions for geometric process data

Process parameters G-code Example

Up milling UP UPDown milling DN DNPredicted radial width RP RP 1.0Exact radial width RE RE 0.75Predicted axial depth AP AP 0.2Exact axial depth AE AE 0.3

6. Integration of process planning and control

There are a variety of approaches that can be taken to interface the CNC control and monitoringsystem. Given the need to standardize the system over a range of controllers, it seems sensibleto express the needed information in the form of an extension to normal G-code. This approachallows adaptation of existing G-code programs, by simply adding appropriate process controlcommands. Naturally, a new standard for these G-code extensions needs to be developed. Onesuch attempt is shown in Table 6 where G-code extensions for passing geometric process datato the control system are proposed [17].

For the work shown here however a different approach was followed. The UBC open architec-ture controller already allows extreme openness to new hardware and software. The basic machin-ing language is easily extended to include the requirements of dynamic process planning.

The control architecture is based on either the ISA, STD or STD32 backplane bus; the latterbus allows multiple masters to share the tasks of process planning and control. The backplane busis used primarily for distribution of motion information, the front plane bus is used for interslavecommunication and process integration as follows. Firstly, it allows for synchronization of positionloop closing between the slave axes, secondly it allows the examination of the state of a seriesof processes and the modification of feedrate based upon the result. This arrangement allows anyslave board to modulate the feed rate of the overall system in real time without the knowledgeof the control master processor. The basic system has been described in several previous publi-cations (e.g. [18]) and patents (e.g. [19]); a diagram of the system is shown in Fig. 9. There arethree logical units within a process optimization system; these need to be able to transfer para-

Fig. 9. Open control architecture.


meters to each other and perform tasks synchronously. First, the CAPP system generates processplans with parameter estimates of the process. Second, the traditional CNC control system pos-itions the tool according to the process plan that it receives from the CAPP system. Third, theprocess control system updates cutting conditions through the CNC system according to controltasks that it receives from the CAPP system and sensory inputs that are gathered during themanufacturing process.

Naturally the control tasks in the process control system need to be synchronized with theposition control of the CNC system. A distributed monitoring architecture is proposed here thatfulfills these requirements. A diagram of this architecture where two monitoring systems, a CNCmachine, and a CAPP system are interconnected is shown in Fig. 10. The dash-dotted linesbetween the process planning system and the CNC machine represent the traditional informationflow of position, speed, and feed information. The dashed lines show the flow of monitoring tasksthat are sent by a task scheduler in the CAPP system and received by schedulers on the monitoringboards. It is also possible that monitoring tasks have requests for the process planning system.These are routed through the same schedulers, but in the opposite direction. The dotted linesrepresent the connections between sensors on the CNC machine and tire monitoring boards. Thesolid lines represent a front plane bus which has two functions. The “sync” lines in this busallow for synchronization of position loop closing, spindle rotation, or other process relevantsynchronization events. The so-called state lines are used to modify machine feed rate. Thedynamic process planning system envisioned here has been developed to a stage that providesreliable sharing of process data between the planning system and the control system. The approachalso allows reliable real time process identification and process control. The system is able toidentify the radial width and the axial depth in face and endmilling operations which then allowscontrol of the machine tool feedrate. Due to a lack of reliable wear sensors, the spindle speed ofthe machine tool must be selected according to tool life estimates. The system is also able to

Fig. 10. Monitoring system.


identify cutting parameters, which can be used to calibrate the geometry identification algorithms.Finally a runout identification algorithm has been implemented that can be used to infer toolcondition for the dynamic process planning system.

7. Experimental investigation

The process control system has been tested using a variety of cutting conditions with steel andtitanium work pieces. The machine tool used was a Holke light duty knee mill which was retrofit-ted with the open control architecture in 1993. The controller on this particular machine uses an8 bit STD backplane, one of the monitoring CPU’s is an Intel 8OC196 slave with communicationto the master through biPort ram, the second, an Intel 80486, is mounted in a separate card cage.The interface to the second cage is achieved using a serial connection and the front plane arrayfor synchronisation and feedrate modification. The forces are measured using an inhouse designedlow cost/low bandwidth dynamometer, the approximate bandwidth of the dynamometer with a15 kg workpiece is 150 Hz in all three orthogonal directions. Force measurements are made usingthe 8OC196 based monitoring board and are triggered by a signal from a proximity probe whichis mounted against a 30 tooth spline on the machine spindle. In this section results obtained duringthe cutting of a 200 BHN steel will be discussed. The purpose of this section is to demonstratethe effectiveness of the basic methods for parameter identification, process control and runoutidentification.

7.1. Calibration cuts

Calibration cuts were performed during a slotting and a one quarter immersion operation. Dur-ing the steady state portions of the cuts, the monitoring system identifies values for cutting press-ure and force ratio. This identification task assumes accurate values of radial width and axialdepth that are supplied by the CAPP system. Fig. 11 shows a plot of the identified values. The

Fig. 11. Identified cutting parameters.


cutting pressure is what one would expect for such a material (3000 N/mm2), the rather highvalue of force ratio (r=0.8) indicates a fair amount of wear on the inserts.

7.2. Process control

Fig. 13 shows the action of the process control system in a typical situation of a quarter immer-sion face milling operation where the feedrate is regulated by the tool breakage constraint. Thisconstraint requires that the equivalent chip thickness is kept below 0.09 mm and that the resultantforce is kept under 450 N. In the cutting operation chosen here, the chip thickness constraintis dominant.

The equivalent chip thickness is determined from the immersion angle that is successfullyidentified at approximately 60°, and the axial depth which is found to be close to the expected0.6 mm.

During the entry and exit transients, the force magnitudes are below the noise cutoff levels;thus, conservative values of feed are selected by the process control system during these periods.During the steady state portion of the cut, the chip thickness control keeps the feed rate overrideclose to the expected 100% mark.

7.3. Monitoring

Tool condition includes tool wear and tool runout. Since no reliable wear sensors are available,radial runout is the only parameter that has been identified.

The identification algorithm is able to determine relative radial runout of individual insertsfrom Fourier series representations of in plane cutting forces where relative runout is defined asthe difference between two adjacent tooth sizes. In the example shown in Fig. 12, the runout ofa four tooth cutter was identified in three quarter immersion and quarter immersion, (the quarterimmersion occurs after approximately 700 revolutions. The three quarter after 400). The identifiedrunout values correspond well, both with each other, and with dial gauge measurements (seeTable 7 and Fig. 13).

In a dynamic process planning system this identification algorithm allows both the monitoringof cutter runout and the detection of edge breakage of the cutter in the presence of considerableinitial runout. A similar algorithm for identifying axial runout of individual inserts has also beenproposed that promises to provide the equivalent information for face milling [20].

8. Conclusions

It has been shown that the integration of process planning and control activities can yieldsavings on two levels. At the lower level, a real time process control system can be used toimprove the selection of feed and speed, at the higher the feedback of identified process parameterscan be used to improve process plans dynamically.

The integration of planning and control poses challenges over a wide area of research interest.Procedures and algorithms for process planning that are able to arrive at good solutions givenonly approximate process data need to refined. Robust process identification algorithms, the back-


Fig. 12. Chip thickness control for a quarter immersion cutting operation.

Table 7Radial runouts measured with a dial gauge

Relative runout (mm)

Insert 1 0.018Insert 2 20.015Insert 3 20.005Insert 4 0.002

bone of any process planning system, need to be available and the sensors that these identificationalgorithms rely on need to be low cost. Finally, a suitable planning and control architecture needsto be in place to ensure that the control system is able to adapt to changing process requirements.

This work has attempted to demonstrate a rudimentary implementation of such a system. Basicplanning, identification and control functions have been integrated by extending the UBC controlarchitecture. The work has demonstrated that current sensor technology allows for robust processidentification even when based on rather low cost/bandwidth sensors. The identification procedures


Fig. 13. Typical action of the control system.

have been shown to be effective for both cutting condition and tool condition purposes. A multiconstraint feed control algorithm has also been shown to be practical.

References

[1] P.G. Petropuolos, Optimal selection of rate variables by geometric programming, Int. J. Prod. Res. 11 (4) (1973)305–314.

[2] K.V. Ramaswamy, B.K. Lambert, Cost optimization for a single pass turning operation including inventory andpenalty costs, Int. J. Prod. Res. 12 (3) (1974) 331–344.

[3] T.C. Chang, R.A. Wysk, R.P. Davies, B. Choi, Milling parameter optimization through a discrete variable trans-formation, Int. J. of Prod. Res. 20 (4) (1982) 507–516.

[4] J. Crookall, N. Venkatramani, Computer optimization of multi pass turning, Trans. ASME: Int. J. of Prod. Res.9 (2) (1971) 247–259.

[5] B.K. Lambert, A. Walvekar, Optimization of multi-pass machinig operations, Int. J. Prod. Res. 16 (4) (1978)259–265.

[6] I. Yellowley, E.A. Gunn, The optimal selection of cut in multi-pass machining operations, Int. J. Prod. Res. 27(9) (1989) 1573–1588.

[7] T. Kishimano, K. Saito, The optimum sequence of operations in the multistage manufacturing system, in: Proceed-ings of the 16th Machine Tool Design Research Conference. 1976, pp. 57–61.

[8] I. Yellowley, A. Kusiak, Observations on the use of computers in the process planning of machined components,Trans. CSME 9 (2) (1985) 70–75.

[9] A. Kusiak, Optimal selection of machinable volumes, Trans. IIE 22 (1990) 151–160.[10] I. Yellowley, A note on the significance of quasi-mean resultant force and the modelling of instantaneous torque

and forces in peripheral milling operations, Transactions of the ASME: Journal of Engineering for Industry 110(1988) 300–303.

[11] S.C. Lin, R.J. Yan, Force based model for tool wear monitoring in face milling, Int. J. Mach. Tools Manufact.35 (1995) 1201–1211.


[12] M.A. Elbastawi, T.A. Papazafirou, R.X. Du, In-process monitoring of tool wear in milling using cutting forcesignature, Int. J. Mach. Tools and Manufacture 31 (1) (1991) 55–73.

[13] Y. Altintas, I. Yellowley, In-process detection of tool failure in milling using cutting force models, Trans. ASME,J. Eng. Ind. 111 (1989) 149–157.

[14] R.J. Seethaler, I. Yellowley, The identification of radial runout in milling operations, Accepted for Publication,Trans, ASME, J. Man. Sci. Tech., 1999.

[15] M.A. Elbastawi, R. Sagherian, Parameter adaptive control in peripheral milling, Int. J. Mach. Tools and Manufac-ture 27 (3) (1987) 399–414.

[16] O. Masory, Y. Koren, Stability analysis of a constant force adaptive control system, Trans. ASME: Jour. of Eng.for Ind. 107 (1985) 295–300.

[17] R.J. Seethaler, L. Yellowley, Dynamic process planning within an open CNC environment, in: Proceedings ofthe 32nd International MTDR Conference. 1997, pp. 489–494.

[18] I. Yellowley, R. Ardekani, L. Yang, R.J. Seethaler, The development of robust extensible control architecturesfor machine tools, in: SPIE, Photonics East Conference, Boston, USA 1996.

[19] I. Yellowley, P.R. Pottier, Slave Control System, United States Patent No. 5,222,017, 1992.[20] R.J. Seethaler, Integrated planning monitoring and control of milling operations, Ph.D. Thesis, University of British

Columbia, 1997.

process control and dynamic process planning

Documents