You are on page 1of 10
13 “Reversal” Techniques for Coordinate Measuring Machine Calibration Robert J. Hocken CONTENTS 13.1 Classic Reversals 361 13.2. Self-Calibration on Coordinate Measuring Machines 367 133 Summary 370 In this chapter, the term reversal techniques is used to cover a wide range of self. calibration techniques that are used or could be used on coordinate measuring machines (CMMs). These are a subset of a wider set of self-calibration techniques described in the literature. A good review of the subject, wihich was used extensively in this chapter, can be found in the study by Evans, Hocken, and Ester (1996) and the references therein. 13.1 CLASSIC REVERSALS ‘These reversals requite a repositioning of the instrument or the artifact between at least two measurements. The simplest forms the level reversal shown in Figure 13.1. ‘Suppose that the level is imperfect due to a piece of dirt attached to one foot (a com- ‘mon problem in masonry). When the level is in one position, it measures the angle of the surface, s, plus the angle of the level, When rotated 180° and put atthe same place on the surface, it measuses the angle ofthe surface minus the angle ofthe level Spliting the difference gives the correct angle of the surface. fone watches a good mason laying a brick wall, level reversal is seen constantly as itis hard to keep small pieces of mortar from adhering to the surface of the level. ‘When initially leveling a measurement machine with either a bubble level or an elec- tronic level, level reversal is recommended.” 7 Similar procedures ave wed in surveying where the iescope ofa teodolia le treed 180” about a ‘horizontal ati and then 180" about a vertial ais andrenighted osu target I! he angles are averaged, ‘eight of standard errors quareness and cllimation eer can be semeved. The procedize i om monly called “ac elt, fee right 361 362 Coordinate Measuring Machines and Systems FIGURE 13.1. Level reversal Indicator outputs Sf Sealghedge 3100) Table oH ” l f= M00 +500) FIGURE 13.2 Straighiness reversal Another classic reversal useful for CMM calibration is called straightedge reversal." This procedure is used to measure the straightness of motion of a mov- ing cartiage and is illustrated in Figure 13.2. Fust, a calibrated indicator is used to measure the changing distance between the carriage and a straightedge aligned with the carriage. Next, the straightedge is rotated 180° about the measurement line on its face and is remeasured in the “reverse” position. According to Evans eta, if one calls the machine slide straightness M(x) and the straightedge straightness S(x), then the indicator readings in the two positions are given by the equations in Figure 13.2. ‘M(») and S(a) can then be calculated by taking half the sum and differences of I, and. 1, After the calculation of M(x) and S(a), a straight-line term should be removed, as itis the result of the lack of perfect alignment. Traditionally, the straight line was "The oldest reference found on avaghtedge reversal i rom 1909 (Cobleigh 1908). Here the reversal tecbrigue was tied to make astraghtedge by continually reversing and resoving material ui ines ‘uaced alg the straightedge in the reversed and forward dzectionsaze the same Reversal” Techniques for Coordinate Measuring Machine Calibration 363 removed graphically by zeroing the end points (or by graphically estimating a central line), but a Teast squares line is now preferred. Note that the straightedge must remain stable throughout the process, the rever sal must be such that the measurement line is at the same distance from the table, that is no offset, and there should be no movement of the straightedge along the axis direction, x, in the figure (movement of the straightedge in that direction is commonly called “shear”, Errors caused by position changes will depend on the out-of-strightness of the straightedge (for sheat) and the angular error of the mov- ing table (for offset). The reversal procedure works best if both errors mentioned are slowly varying functions of the position. For machines that have continuous motion rotary axes, another classic reversal is quite useful. I is universally called the Donaldson reversal alter ils originator (Donaldson 1972). Ibis precisely analogous to the straightedge reversal but measures radial motion (ASME 2010) of a rotating axis and allows the use of an imperfect ‘master ball just as the straightedge reversal uses an imperfect straightedge. This procedure is shown schematically in Figure 13,3, Fist, the ball out-ol-roundness is ‘measured in one position, and then the ball is rotated 180° with respect to the rotary (able and is remeasured with the indicator on the opposite side of center. Let X(@) be the radial motion of the rotary axis and R(@) be the out-of-roundness of the mas- {cr ball, Then, the equivalent equations to the straightedge case are represented in Figure 13.3, with similar sum and difference solutions for R(@) and X(0) Eccentricity should be remaved from the solutions. Further, if the rotation angle is not exactly 180°, errors due to master ball deviations from roundness can be seen, as is the case with shear in a straightedge, Results from this method can be quite accurate (Marsh 2009) Another reversal that is useful for rotary axes is due to Estler (Estler 1986; Salsbury 2003h). It is called the Ester Face Motion Reversal and is shown dia- ‘grammatically in Figure 13.4, This method allows the separation of tilt and axial motions from oul-of-flatness of the axis face, In the first setup, two indicators Indlestor outputs ve ho 1 _ 1@)= RO) - FIGURE 18,3 Donaldson reversal, 364 Coordinate Measuring Machines and Systems FIGURE 13.4 Esler reversal, are used, one on the rotation axis and the other a distance r off the axis, The spindle is rotated 360° and the indicator readings and angles are recorded. In the second setup, the partis rotated 180° and the second indicator is repositioned to follow it, Again, a rotation and a measurement are performed. The first indica tor is not really needed for the second measurement, but if left on the center, it improves knowledge of the axial motion, The data are analyzed by first averag- ing the results of the on-axis indicator, call it A(®) following Evans, Hocken, and Estler (1996), Then, one can calculate the tlt error motion of the spindle 7(@) and the flatness of the part P(@), where hah =A@) aan A) = I= PO)+T0)+ 1 = POO) +7T (0) +A) a f= PO)=1T0)+ AW) po 4th 2 033) T®)= Reversals are also commonly used for squareness measurements. The simplest of these is the so-called square reversal shown diagrammatically in Figure 135. It shows a dial-gage reading against an imperfect square, which has one face aligned with the y axis of a machine, The dial gage is traversed in the x direction, and a trace is recorded. Then the square is rotated 180° (Bipped) and the operation repeated, Talf the difference between the angles of the two traces is the outof-squateness of the machine axis. A similar case is used to measure the parallelism between a linear axis and a rotary axis when the square is just a straightedge aligned on a rotary axis (pindle or positioning), as shown in Figure 13.6. Here the trace is made, and then the rotary axis is rotated 180°, and the probe is rebracketed across centers and is used to trace the straightedge again, Half the difference in angles gives the parallelism Many other examples are in the BS.S7 standard (ASME 1998), Reversal” Techniques for Coordinate Measuring Machine Calibration 365 MQ) 1) FIGURE 18.5 Squareness ceversal rueet FIGURE 13.6 Parallelism reversal ‘Staightness interferometers were intoduced by Hewlett-Packard in the 1970s and are now produced by several companies. Typically, they use an angled refer- ence mirror and a Wollaston prism, as shown in Figure 137. Lateral movement of the two optical elements with respect to each other (y direction in the figure) will ccause changes in the optical path lengths of the two beams and therefore a change in the optical count If the mizrors of the straightess reflector ate not perfectly flat, the deviations from flatness yield incorzect results as the optics move laterally. This error may he removed by another reversal technique introduced by Bryan and Carter (1989, ), ‘The data for an axis are taken first with the straightness mitror in one position, 366, Coordinate Measuring Machines and Systems Laser beam t \Wollston prism Straighines Flector FIGURE 13.7 Schematic of straightness interferometer, FIGURE 13,8 Mirror fatness effects ‘Then, its rotated 180° around the axis, as shown in Figure 13.8 and the straightness is measured at the same points. The average of the two results removes the fat ness error, Interested readers are referred tothe study by Evans, Hocken, and Estler (4996) for the detailed equations ané alignments. For example. the same positions 6on the mirror must be used) Note that all these classic reversal techniques are used to measure parametric ‘machine errors and thus are useful for manufacturers as well as machine users Reversal techniques may also be used for determining the properties ofthe probe tip on measuring machines, Sometimes, the whole tip is mapped for use in ultra accu- racy CMMs (see Chapter 16, Typical Applications and King, Mel, and Thalmann 2007), whereas in others, itis only desired to measure the probe diameter along one line, as shown in Figure 139. This technique was developed by Physikalisch- Technische Bundesanstalt (PTB; Evans, Hocken, and Ester 1996). Note that from the aree measurements shown inthe figure, the diameters of the probe tips can be calculated with simple algebra For CMMs, however, there exists a very different class of “reversal” like tech- nigues that are usually referred to as “self-calibration.” They are described in Section 13.2 Reversal” Techniques for Coordinate Measuring Machine Calibration 367 a FIGURE 13.9 PTB probe diameter measurement 13.2 SELF-CALIBRATION ON COORDINATE MEASURING MACHINES (On machines with multiple degrees of freedom, uncalibrated artifacts may be used {or self-calibration, The concept is to measure the artifact in a number of carefully chosen positions and from the measurement results deduce the machine errors and the artifact errors (Sartori and Zhang 1995; Belforte etal. 1987; Balsamo etal. 1996). ‘This should not be confused with measuring machine errors with a calibrated arti= fact, These measurements ean only be used to find geometric errors as the results fare independent of the chosen length scale (metric), which must be defined by the ‘machine user, Ideally, such measurements could yield complete geometric informs- ‘don about the machine, but practical artifacts have limited numbers of measurement points, and the time of measurement is also constrained. Therefore, errors with a higher spatial frequency than the point density can be missed or erroneously inter- preted due to aliasing. Most modern CMMs, however, have slowly varying (in space) ‘geometric and scale errors and are thus ideally suited for this type of self-calibration There are two very different methods. In the first method, the general procedure is as follows: An artifact, usually a agri plate or 2 ball plate is measured in at least (Wo positions in two dimensions (2D) and more in three dimensions (3D), Ball bars, space frames, and even plane surfaces have also been used (Clément, Bourdet, and Weill 1981; Jouy and Clément 1986), Next, in any case, a model of machine errors is developed. This model may be very simple (Hocken and Borchardt 1979) containing only a few dominant terms or guite complex requiring a set of basis functions and perhaps some assumptions about the form of the machine errors (Raugh 1985, 1997), After multiple measurements, a Teast squares fit for the artifact coordinates and the machine error parameters is performed on the measurement data, Some experiments have fit the data to get repre sentations of the normal parametric errors (see also Chapler 12, Error Compensation of Coordinate Measuring Machines; Fu 2000), others go directly to an X, ¥ error ‘map. Often, itis not mentioned that the machine would need a diferent error map {or different probe offsets, The earliest form of this type of calibration was found by Reeve (1974), He used a very simple model for the machine errors because he was concerned with the calibration of reference ball plates rather than the machine: his ‘machine model contained only out-of-squareness, In his model, the plate was rotated tomany different angles about the z axis of a measuring machine, and the X, Y center 368 Coordinate Measuring Machines and Systems coordinates were measured. In his study, he gives an example in which the plate is measused at four angles but that is not a limitation of the analysis. Many more angles can be used. The procedure “randomizes” the effects of the normal para metric errors and gives an excellent value for the out-of-squareness (randomization is also used for the measurement of a silicon sphere, as described in Chapter 16, ‘Typical Applications). The fit is highly nonlinear and required long run times on a central computer when it was originally written. Following this a linearized version was developed, which required only a 90° rolation, although the rotation had to be within a small angle of 90° (HHocken and Borchardt 1979) This method used a machine model that contained out-of-squareness and a “metric erton,” which was the difference between the scales of the x and y axes. In normal use, one axis was assumed to have an accurate scale, Besides these two errors, this method calculated the difference between the actual rotation and 90° as well as any positional offset caused by the mechanical rotation, How this method ‘works can be seen from an oversimplified example (Evans, Hocken, and Estler 1996), In that study, a ball bar, as shown in Figure 13.10, was rotated exactly 90° with no linear motion of the rotation point, Call the measured position of the second ball (the ball not atthe origin) X,, ¥, in the first position and X,, ¥, in the second position. ct is the outof-squareness and ys the incremental scale error in the x axis, then Xd eyk+a¥ y-¥ X= (+ y)¥ -ax Y=-x ws and the solutions for the machine squareness and scale difference are Xs = YY = Xi 4 Yoh Ye +¥? XY + XY, “YY (3s) ‘A further advance in this procedure was developed by Raugh (1985) for the cali- bration of the stages on direct writing e-beam machines at Hewlett-Packard. In his, ‘model, the errors for each axis were parameterized, and the least squares fit was, ‘made for the coefficients of the ft parameters as well as the squareness and scale differences, Again a [airly large program was required. He continued to try and develop a complete theory (Raugh 1997) but has yet to complete this work. He did discover, however, that the procedures of Reeve and Hocken (Raugh 1997) would miss certain rotationally symmetric errors unless a finite translation was added to the procedure ‘The above procedures were in 2D and the techniques were extended to 3D by ‘measuring a number of planes. See the study by Kruth, VanHerck, and Jonge (1994) and references therein Reversal” Techniques for Coordinate Measuring Machine Calibration 368 FIGURE 13.10 Idealized bal bar rotation ‘This method of multiple measurements of the same artifact and least squares ft {ing was also used on rotary tables (Kunzmann etal. 1989). The abstract of his patent states the following: ‘The four components of the deviations of a rotary table from an ideal axis of rola tion ean be obsained by a single measuring method wherein atest body (4) having a plurality of well-defined measurement points is placed on the rotary table (1) and the Positions of said measuring points are then determined by means of a coordinate reasuring instrument @), for each of various angular postions of the rotary table, there being thus a set of ascertained measurement-point coordinates for each of the angular positions ofthe rotary table. Prom the ses af measurement-point coordinates, the travel deviations fsub.a, Esub and f subs of the axis of rotation, and the angle position deviation psubsw ae then determined by calculation In American nomenclature (ASME 2010), the deviations are called axial, radial, and tilt motions ff, fespectively) as well as angular displacement accuracy (p,). Nearly independent of these developments for measuring machines (even the Raugh model that was checked by using the e-beam machine as a scanning elec- {ton microscope), efforts were ongoing in the microelectronics world to improve the accuracy of mask making by self-calibration (Ye 1996; Ye et al. 1997). In this stage calibration method introduced by Stanford University, three views are chosen (Figure 13.11a through c): (1) the initial view: (2) the rotation view, where the plate is rotated counterclockwise 90° about the center of the sample arrays; and (3) the translation view, where the plat is translated along +x by one sample site interval a Because ofthe existence of the alignment errors, the equations for each view couple the stage errors with the alignment errors. Consequently, solving forall these unknown. errors involves a large computation (Ye et al. 1997). The beauty of Stanford's method 370 Coordinate Measuring Machines and Systems The Dh Ade [To Le xy lol Pas ix Ps, sa Os Inka views view 9 Rotation view view 1 ranation view: iw 2 c ® © FIGURE 18.11 ‘The three views for Stanford's stage calibration method. ol Ie Ley | FIGURE 13.12 A reversal method that requires rotating the plate about the y axis, is its algorithen, in which the alignment errors are decoupled by a clever mathematical treatment. Furthermore, by applying the discrete Fourier transform (DFT) algorithin, the discrete coefficients ofthe stage error map are further decoupled orthogonally, and therefore the calibration result is less sensitive to random measurement errors ‘Another similar technique uses plates that can be accessed from both sides. It is in some sense similar to the straightedge reversal in that a rotation out of the plane ‘of measurement is required, The three Views required are shown in Figure 13.12 ‘This method yielded values for the parametric errors, which were compared to those obtained by traditional calibration with good agreement (Fa 2000). An extended algorithm that provided a 3D solution to self-calibration for testing CMMs ‘was presented in 2006 (Dang, Yoo, and Kim 2006). Using a 3D grid artifact of steel balls, four separate sels of measurements were taken in different positions ofthe arti= fact. Then, the errors related tothe artifact were identified using algebraic manipula tion, hence a complete 4D error map of the machine systematics was constructed, ‘The results were checked by experiment, Accuracy was limited by the measurement repeatability of the CMM tested. 13.3. SUMMARY In this chapter, many of the reversal {echniques common in machine calibration were described. Several of the methods described, such as square reversal, are already part of American national standards for machine tools (B5,57) and are in ‘wide use. As pointed out in a review by Evans, Hocken, and Estler (1996), such tests are also common in optical metrology, The interested reader is referred to that study {for many non-CMM applications.

You might also like