PRE-MACHINING CNC CONTOUR VALIDATION

Tony Schmitz, John Ziegert

University of Florida, Machine Tool Research Center
Gainesville, FL 32611

Abstract
The current procedure followed to manufacture a new part by CNC machining is to write the part program,

machine a test part and measure the test part for conformance to the required dimensions and tolerances. If the test

part dimensions are not correct, the part program is modified and the process repeated until a successful part is

machined. In many applications, such as the aerospace industry, where material cost and machining time are high,

this iterative process becomes economically unacceptable.

Research has been conducted to test the feasibility of using the Laser Ball Bar (LBB), a spatial coordinate

measuring device, to measure dynamic continuous-path contours of CNC part programs to micrometer accuracy

prior to machining. In this way, a virtual test part can be measured and compared to the design drawings to

validate the CNC part program. This reduces or eliminates the costly and time-consuming steps involved in the

machining of physical test parts.

This paper outlines the testing method and results acquired using one LBB to measure dynamic part paths

employing sequential trilateration. A circular contour was measured using an encoder trigger for data capture.

The radial error motions of the spindle used to generate the circular contour were also measured using a

capacitance probe nest to verify the LBB results. Comparable error waveforms between the LBB and cap probe

measurements verified the possibility of using the LBB to measure dynamic continuous-path contours. Future work

using three LBBs simultaneously is also outlined.

Keywords: Laser Ball Bar; CNC machining; dynamic measurements

Introduction
One of the most important uses of CNC machine tools is the cutting of complex

continuous part paths, or contours. In continuous-path numerical control systems, there is

contact between the cutting tool and workpiece throughout the part path while up to five axes are

in motion. Therefore, the final workpiece dimensions are directly related to the positional
 2
relationship between the tool and workpiece. The ability to monitor this relationship and

predict the final part dimensions is important for today’s manufacturing engineer.

For a new part or production run to be manufactured with the use of CNC machine tools,

the current procedure is to write the CNC part program using the engineering drawings, execute

the part program to machine a test part or prototype, and then inspect the test part, normally with

the aid of a coordinate measuring machine, to check for conformance to design tolerances. This

feedback of the actual part dimensions (with an adequate degree of precision) is currently the

only way to certify the performance of the CNC program. If the test part does not meet the

specified tolerances, as is often the case for a first trial, the CNC program is modified, another

test part is machined and the process repeated.

This iterative process may be acceptable, although not efficient, in situations where the

material is inexpensive and machining time is short. However, in many cases, such as the

aerospace industry, the material is costly (expensive forgings to be machined) and machining

time is dramatically increased. In such instances, this iterative process of producing

dimensionally correct parts becomes economically unacceptable.

Both time and money could be saved if there were a way to directly measure the machine

tool’s dynamic contouring accuracy over an arbitrary 3-D path without the necessity of

machining an expensive test part. At the same time, the efficiency of the CNC machining

process could be greatly increased. The use of the Laser Ball Bar (LBB), a spatial coordinate

measuring device, to take dynamic measurements of a path could reduce or replace the need to

machine and inspect a test part. In effect, the spatial coordinates of the dynamic tool path

measured by the LBB could function as a virtual test part1.

This particular approach must be differentiated from the conventional approach to

 3

machine tool metrology. At this time, the normal procedure is to measure the machine tool error

motions including geometric, thermal and perhaps process errors. These errors may then be used

as pre-machining compensation in the machine tool controller in an effort to give theoretically

perfect motions. The machine’s accuracy and, therefore, the part dimensions are based largely

on the success of this time-consuming process. Although this research is certainly not a

replacement for this body of work, it is also not the same. The purpose of this work is to try and

predict the final part dimensions for the execution of a specific CNC part program on a given

machine tool prior to cutting the part, not to measure the geometric or servo errors of the

machine tool.

Background
One of the main categories in the modern evaluation of CNC machining centers is the

assessment of the contouring capabilities of the machine tool. The evaluation of this contouring

accuracy can be divided into two main classifications: post-process and in-process testing.

Post-process testing includes those tests which are performed after machining has been

completed. The most popular post-process inspection tool is the coordinate measuring machine

(CMM). Other post-process methods used to evaluate a CNC machine tool’s contouring

accuracy involve the use of either master parts or well-defined contours. One example of this

technique is the use of standard part paths, such as the part program corresponding to the

National Aerospace Standard test part 979 (NAS979), to machine a master part. This part can

then be measured to evaluate flatness, squareness, parallelism, roundness, etc. A similar, but

somewhat more efficient method has been termed master part tracing. This procedure simulates

machining by replacing the tool with a gage and the workpiece with a master part of known
 4

accuracy. The machine is then programmed to follow the ideal path given by the master part.

Deviations from the ideal path registered by the tool gage represent the contouring error2,3.

The other main type of contour measurement is in-process testing. This category can

then be subdivided into in-process and in-cycle gaging. In-process gaging refers to testing

carried out during the actual machining process, while in-cycle gaging defines measurements

taken after the part is finished, but before it is removed from the machine.

In-process gaging is generally achieved by adding analog transducers to the machine tool

to directly assess the test part’s size during the cutting operation. An example of in-process

measurement entitled Workpiece-Referred Form Accuracy Control, or WORFAC, has been

suggested by Uda et al.4 It consists of a highly sensitive optical surface sensor and a microtool

servo to adjust for changes in the distance between the tool holder and workpiece. Although

accuracy improvements were shown with the use of this feedback system, it is still limited to

simple geometries (cylindrical turning operations) and requires complex tooling and set-up.

In-cycle gaging can be accomplished by replacing one or more of the tools in a CNC

machining center’s turret with measuring probes. These probes can then be indexed into the tool

position to probe key features of the test part while it remains in the machine and, unfortunately,

is still in a thermally unstable state.

An adaptive error correction method has been proposed by Mou et al. which combines

both in-cycle and post-process testing with a geometric-thermal error model5,6. This technique

proposes the use of in-cycle and post-process gaging to interactively determine the changes in

the geometric-thermal model over time.

Although these processes provide worthwhile means of evaluating a CNC machine tool’s
 5

contouring accuracy, they all suffer from the necessity of machining and inspecting a test part

for conformance to required tolerances. The purpose of this research was to investigate the

possibility of using the Laser Ball Bar as a measurement tool to determine this CNC path

accuracy without the requirement of producing a costly test part. Commercial products which

are moving toward this goal include the Heidenhain 2-D grid plate and Renishaw Ball Bar.

These tools, however, are path limited. The Ball Bar allows only circular or hemispherical paths

and records just radial deviation. The grid plate can measure only planar part paths. For 3 or 5-

axis part paths, the Laser Ball Bar system is required to dynamically measure these contours.

Laser Ball Bar

The LBB is a precision linear displacement measuring device developed by Ziegert and

Mize at the University of Florida7. It consists of a two-stage telescoping tube with a precision

sphere mounted at each end. A heterodyne displacement measuring interferometer is aligned

inside the telescoping tube and measures the relative displacement between the two spheres. See

Figure 1. The LBB has been shown to be accurate to sub-micrometer levels during static

measurements of spatial coordinates8.

Once initialized, the LBB uses trilateration to measure the spatial coordinates of points

along a CNC part path. The six sides of the tetrahedron formed by three base sockets (attached

to the machine table) and a tool socket (mounted in the spindle) are measured and, by geometry,

the coordinates of the tool position can be calculated.

In sequential trilateration, the same part path is traversed three times, measuring the

lengths of one of the base-to-tool socket legs at a finite number of points during each repetition.

See Figure 2. Note that the tool socket must be in exactly the same position (for a given point)
 6

for each of the three measurements. If the tool socket were in a slightly different position at

point 5 for the leg 3 measurement than for legs 1 and 2, for example, the spatial coordinates

calculated by trilateration would be incorrect. In order to achieve accurate coordinate

measurements, the sampling trigger must be spatially repeatable. When quasi-static

measurements are performed, i.e. the tool stops at each measurement position, the machine

repeatability governs the accuracy at the measured coordinate. For most machine tools, the short

term repeatability is substantially better than the absolute positioning accuracy and this process

yields satisfactory results. However, for dynamic path measurements, the LBB must be

triggered at specific points along the path to collect data.

In simultaneous trilateration, three LBBs ride on a single sphere at the tool point to

completely define the tetrahedron with one execution of the CNC program. For this method,

since all three leg lengths are captured simultaneously, the spatial repeatability of the sampling

trigger is no longer a concern. Figure 3 illustrates this method. The tool socket shows the three

individual LBBs, each using a 3-point contact magnetic socket, attached to a single sphere. In

order to minimize interference between the three magnetic sockets as they move over the sphere

surface, and to insure an adequate range of motion, the socket diameters (and consequently the

neodymium magnet diameters) must be relatively small. This requires minimization of the LBB

weight.

The actual tool point positioning errors collected by the LBB can then be used in

conjunction with a suitable kinematic model to evaluate the overall parametric accuracy of a

machine tool. These off-line error measurements can provide pre-calibration compensation

when stored in the machine tool controller and used to correct the commanded position
 7

according to its location in the work volume.

The parametric error map for a 3-axis milling machine was constructed by Kulkarni using

both the LBB technique and the standard methods described in the ASME B5.54 Standard,

“Methods for Performance Evaluation of Computer Numerically Controlled Machining

Centers"9. The agreement between the two results verified the LBB method10. A second study

was completed by Srinivasa which measured the positioning errors on a 2-axis turning center

over a thermal duty cycle and correlated these errors with the temperature gradients within the

machine tool using a neural network. A PC-based error compensation system was then used to

compensate for the errors predicted by the neural network11.

The above methods give a rapid, efficient way to measure the quasi-static errors of a

machine tool and thus evaluate its static positioning accuracy. However, these methods do not

characterize the dynamic positioning accuracy of the machine tool.

Dynamic Path Measurements

The purpose of this research was to appraise the dynamic measurement performance of

the LBB to evaluate its potential for use in dynamic path measurements. Since only one LBB

was available at the time this study was conducted, it was decided to test the feasibility of using

sequential trilateration to perform these measurements. In other words, it was desired to

measure the spatial coordinates of the tool point at predefined intervals along the contour during

the execution of the CNC part program while the machine was in motion using one LBB. These

dynamic path measurements using a single LBB require that the three leg lengths from the base

sockets to the tool socket are acquired as the tool passes through the same points along its path

on three consecutive runs. This necessitates a spatially repeatable triggering method to initiate
 8

data capture.

Two separate techniques were implemented and tested to provide the repeatable

measurement trigger. The first sampling procedure used a time-based scheme. In this method, a

sample was taken at the start of the motion and then consecutive samples were taken at well-

defined time intervals (every 5 ms) throughout the part path. This method was shown to be

unacceptable since the acceleration profile of the controller varied slightly from run to run on the

two-axis turning center used in this study. This caused the tool to be in a slightly different

position (up to 15 µm at 1000 mm/min) along the path during each leg measurement.

The second triggering scheme involved the use of the feedback signals from the machine

axis encoders. In this method, the axis encoders were continuously sampled and the length of

the LBB recorded at discrete distance intervals along the part path. This sampling method was

chosen because the positioning repeatability (and therefore, the encoder output) is the smallest

source of error in the contouring performance of a machine tool.

Experimental Results
Initial dynamic path measurements using the X and Z-axis encoder sampling algorithm

were completed for both right-angle and semi-circular contours. A look-up table of encoder

positions for each path was constructed. During measurements, the X and Z encoder positions

were continuously sampled and a data point was captured (with a repeatable 200 µs delay) when

the encoder positions matched the values in the look-up table. Experimental results showed

general path degradation with higher velocities. This path degradation was especially evident

for the right-angle contour. Figures 4 and 5 show measurement results for path velocities of 250

and 1000 mm/min, respectively. (The missing point in the data is due to an error in the look-up
 9

table.) At velocities of 1 m/min and higher, the corner was increasingly rounded due to the

inherent steady-state positional error, or velocity lag, in the positional servo. The velocity lag,

which is proportional to the feed rate, causes the commanded motion to be executed with a slight

delay. In an X-Z right-angle motion, for example, this gives the X motion time to decelerate and

stop before the Z motion starts. This effectively eliminates overshoot, but can also round the

corner at higher velocities.

Since the actual part path for the right angle contour changed as the feed rate was

increased, it was impossible to distinguish between path errors caused by the controller (due to

the velocity lag) and any measurement errors introduced by the LBB. Therefore, in order to

provide a contour which would be independent of velocity, the spatial coordinates of a circular

contour provided by the spindle rotation with a socket mounted off-center were measured by

sequential trilateration. The circular path, 20.8 mm in diameter, traced by the socket as the

spindle rotated was divided into 256 parts. An LBB data point was taken every 4 spindle

encoder counts (1024 counts/revolution for the spindle encoder gave 256 data points). The path

velocity (spindle speed) was then varied for consecutive tests to investigate the dynamic

measurement performance of the LBB. The radial error motions for spindle speeds of 10 and 60

rpm (tangential velocities of 653 and 3921 mm/min) are shown in Figures 6 and 7.

From these results, it can be seen that the radial error fits approximately within a ±1 µm

band with some noise. The seemingly smoother waveform at the slower spindle speed is due to

aliasing at the lower sampling rate (note that the sampling rate is directly proportional to the

spindle speed). The cause of the repeatable seemingly asynchronous waveform with some

superimposed high-frequency noise was not immediately apparent. Possible sources such as
 10

spindle motion errors, machine noise, precision sphere/3-point socket alignment errors or

bearing friction in the LBB have been suggested.

In order to validate the LBB results, the spindle error motions were measured at the same

feed rates (but higher sampling frequencies) using two mutually-perpendicular capacitance

probes and a precision artifact12 and compared to the radial error motions attained by the LBB

tests. A grade 5, 1" diameter test ball was attached magnetically to a 3-point contact socket.

This socket was attached to a wobble plate which was clamped in the spindle chuck. The radial

error motion for a 60 rpm test using this method is shown in Figure 8.

A comparison of this result with the data obtained using the LBB for the same speed

shows that both error motions fit approximately within a ±1 µm band and both exhibit seemingly

asynchronous motion, although it is difficult to define the period of the waveform with only one

revolution of data. Although this error range is at the limit of the LBB accuracy, the waveforms

were similar and there were no additional errors (beyond the ±1 µm band) introduced by the

LBB measurements. These results suggest that the LBB is a suitable dynamic measuring device.

In order to explore the possible sources of the 120 Hz high-frequency noise found in both

tests, measurements were taken using the cap probe nest to sample the sphere movements with

the machine on, but no spindle motion. Data was collected at a rate of 20 kHz for a period of 2

seconds. The result of these measurements showed a random noise level of ±0.5 µm at the same

120 Hz frequency seen in the LBB and capacitance probe tests. It should be noted that the high

frequency noise content is more evident in the cap probe test simply because the sampling

frequency was much higher and the possibility of aliasing is reduced. The results of this test are

shown in Figure 9.
 11

Conclusions
Experimental results show that the LBB is a suitable device for dynamic path

measurements since the magnitude and waveform of the radial error motions measured with the

LBB corresponded to those recorded using the cap probe nest for the 2-D circular part path.

Although the magnitude of these error motions is at the limit of the LBB accuracy, the fact that

no additional errors were introduced by the LBB and the similarity in the measurements (taking

sampling rates and the possibility of aliasing into account) suggest that the LBB is suitable for

dynamic measurements. Dynamic measurements of three-dimensional CNC tool paths, or

virtual test parts, could also be completed and used to validate the conformance of the CNC part

program to the required engineering specifications.

The ultimate goal of this research was not to predict the machine tool geometric or servo

errors, but rather to predict the final dimensions of a part for a specific CNC program on a given

machine. This paper evaluates the LBB as a potential tool to perform these measurements. Most

of the difficulties encountered in this work were caused by the necessity of sampling the LBB

length at exactly the same points in space for the three tool-to-base socket measurements during

three consecutive runs. Despite this difficulty, the experimental results indicate the LBB holds

promise as a tool for dynamic measurement of arbitrary 3-D tool paths.

Future work includes modifying the LBB design to allow simultaneous trilateration using

three LBBs. This would permit the simultaneous capture of all three leg lengths at a finite

number of points during a single execution of the part program and eliminate the sampling and

timing issues. In this configuration, a time-based scheme would then be suitable since all three

leg lengths are recorded simultaneously and small changes in the controller acceleration profile

from one run to the next would no longer be a concern.

 12

Additionally, a cutting force model could be added to the measurement algorithm to

refine the predicted part dimensions as a function of both the measured path coordinates and the

applicable cutting forces. The cutting force model could act as a filter to post-process the

measured data and predict the final part dimensions. The virtual test part dimensions could then

be validated by comparison to actual machined test parts. Once the implementation of

simultaneous trilateration using 3 LBBs is completed, the next step will be to compare measured

path coordinates with the machine tool coordinates (collected simultaneously) in an effort to

correct any errors in the NC program.

Figure 1: Laser Ball Bar

 13

Figure 2: Sequential Trilateration Figure 3: Simultaneous Trilateration

X-Z Right-Angle Path X-Z Right Angle Path

250 mm/min 1000 mm/min

688 688

686 686

684 684

682 682
Z (mm)
Z (mm)

680 680

678 678

676 676

674 674
81 82 83 84 85 86 87 88 81 82 83 84 85 86 87 88
X (mm) X (mm)

Figure 4: 250 mm/min Test Figure 5: 1000 mm/min Test

 14

Radial Error vs. Angular Position Radial Error vs. Angular Position
10 rpm 60 rpm
0.002
0.002

0.001
0.001

Error (mm)
Error (mm)

0.000
0.000

-0.001 -0.001

-0.002 -0.002
0 90 180 270 360 0 90 180 270 360
Angular Pos. (deg) Angular Pos. (deg)

Figure 6: 10 rpm LBB Test Figure 7: 60 rpm LBB Test

Figure 8: Capacitance Probe Spindle Error Motions

Figure 9: Machine Noise Measurement

 15

References
1. Schmitz, Tony. “Machining Virtual Test Parts,” M.S. Thesis, University of Florida,
Gainesville, FL, 1996.

2. Tlusty, J. and Koenisberger, F. “Specifications and Tests of Metal Cutting Machine Tools
Vol.1,” Proceedings of the Conference, University of Manchester Institute of Science and
Technology, Feb. 19 and 20, 1970, 45-49.

3. Bryan, J.B. and Pearson, J.W. Machine Tool Metrology. Technical Paper IQ68-753, ASTME,
1968.

4. Uda, Y., Kohno, T. and Yazawa, T. “In-Process Measurement and Workpiece Referred Form
Accuracy Control System (WORFAC): Application to Cylindrical Turning Using an Ordinary
Lathe,” Precision Engineering, 18(11), January 1996, 50-55.

5. Mou, J., Donmez, M.A. and Cetinkunt, S. “An Adaptive Error Correction Method Using
Feature-Based Analysis Techniques for Machine Performance Improvement, Part 1: Theory
Derivation,” Transactions of the ASME, Vol. 117, November 1995, 584-590.

6. Mou, J., Donmez, M.A. and Cetinkunt, S. “An Adaptive Error Correction Method Using
Feature-Based Analysis Techniques for Machine Performance Improvement, Part 2:
Experimental Verification,” Transactions of the ASME, Vol. 117, November 1995, 591-600.

7. Ziegert, J.C. and Mize, C.D. “Laser Ball Bar: A Precision Instrument for Machine Tool
Metrology,” Precision Engineering, 16(4), October 1994, 259-267.

8. Mize, C.D., Ziegert, J.C., Pardue, R. and Zucker, N. “Spatial Measurement Accuracy Tests of
the Laser Ball Bar,” Final Report for CRADA No. Y-1293-02244 between Martin Marietta
Energy Systems and Tetra Precision, Inc., August 1994.

9. ASME B5.54. “Methods for Performance Evaluation of Computer Numerically Controlled

Machining Center,” New York, NY, 1992.

10. Kulkarni, Rajeev. “Design and Evaluation of a Technique to Find the Parametric Errors of a
Numerically Controlled Machine Tool Using a Laser Ball Bar,” M.S. Thesis, University of
Florida, Gainesville, FL, 1996.

11. Srinivasa, Narayan. “Modeling and Prediction of Thermally Induced Errors in Machine
Tools Using a Laser Ball Bar and a Neural Network,” Ph.D. Dissertation, University of Florida,
Gainesville, FL, 1994.

12. Tlusty, J. “ System and Methods of Testing Machine Tools,” Microtecnic, Vol. XIII, 1959,
166-167.