Improving the accuracy of roundness measurement

Abstract The application of error separation techniques to roundness measuring instruments is investigated experimentally, using digital computers. The accuracy of such techniques is substantiated by comparative measurements with two independent multi-orientation systems. Results obtained with these and other systems confirm that the errors of conventional roundness measuring instruments can be reduced by a factor of ten or more and meaningful polar graphs obtained at radial magnifications of up to one million. These advances in accuracy can be achieved in relatively poor working environments and with little extra skill on the part of the instrument operator.

1 Introduction 

Advances in the design and manufacture of precision bearings are making increasing demands on the accuracy with which departures from roundness need to be measured. In the case of certain bearing components, notably those used in precision gyroscopes, accuracies are now being demanded which are of the same order as, or even less than, the rotational accuracy of a roundness measuring instrument's reference axis of rotation. A typical figure for the radial error? of a top class instrument spindle is 0.025 pm (1 pin). It is doubtful whether this  figure can be improved significantly without considerable development and a large increase in manufacturing cost.A most attractive and cost effective solution to improving instrument accuracy is to identify and remove the systematicerrors of rotation of the instrument spindle. Whitehouse (1 976)has shown theoretically how multi-orientation and multiprobe techniques may be used to remove both systematic and random spindle errors. In this paper, practical systems based on multiorientation techniques are examined with the aim of extending the current limits of roundness measurement.

2 Limitations of conventional instruments

Most high accuracy roundness measuring instruments are limited in radial magnification to x 10 000 or x 20 000. This upper limit is chosen on the basis that at these magnifications, errors in the profile graph due to spindle error are barely  perceptible and do not have significant effect on the profile. Much higher magnifications can be used but the resultant profile graph represents a combination of both spindle and component errors and needs very careful interpretation in order to be meaningful. To illustrate this point, it is interesting to examine the effect of measuring a precision component, in this case an optically polished glass hemisphere as used for testing roundness instruments, with a radial magnification of x 106. Operation at such a high magnification is greatly facilitated by operating the instrument on-line to a digital computer, using techniques similar to those described by Chetwynd and Kinsey (1973). In this particular case, a Talyrond 73 instrument was used on-line to a Hewlett Packard 2116C computer, with sampling of the transducer signal initiated by a radial grating mounted on the instrument spindle. The component was centred initially under manual controi such that its magnified profile could be contained within the limits of the polar chart at a magnification of x 20 000. Eccentricity terms were then calculated and subtracted from the transducer signal using the limacon approximation (Whitehouse 1973) as criterion for best fit. Remaining data were magnified by the computer and then displayed on the polar chart at a magnification of x 106. This process is essentially the same as physically centring the component and remeasuring it. However, the latter operation is exceedingly difficult to perform at magnifications in excess of 5 x 10*since the manual centring controls do not possess the required sensitivity. Using the computer, centred polar graphs can be obtained at very high magnification with an absolute minimum of setup time. Figure 1 shows some results obtained with this system and illustrates the variation of both graph shape and roundness

parameter values with relative orientation between component  and spindle. All graphs are shown with the chart centre coincident with the least squares centre of the profile and the well known peak-to-valley (P+V), mean-line-average (MLA) and root-mean-square (RMS) parameters are listed. The variation of these parameters with orientation, plotted in graphical form, indicates a cyclical variation with a period of approximately 180" which suggests that the spindle and component errors are predominantly of second harmonic order. This is confirmed by the graphs of the individual spindle and component errors, shown for comparison purposes together withcorresponding parameter values. These errors were extracted from the combined error profiles using the reversal technique described subsequently. 

Examination of the values listed in figure 1 soon reveals that any attempt to compensate for spindle error by subtracting a singleparameter value from the combined error value can give, at best, only a marginal improvement in the estimate of component error and, at worst, a completely misleading estimate. To increase the accuracy of a roundness measuring instrument, it is therefore not sufficient to merely increase its gain without in some way accurately compensating for its spindle error.

3 Multi-orientation techniques

To separate the spindle and component errors requires more information than can be obtained from a single roundness profile. Sufficientadditional information can be obtained either by multiprobe techniques, in which more than one transducer is used simultaneously or by multi-orientation methods, in which several measurements are made with differing relative orientations between component and spindle. A detailed analysis of the possibilities of these techniques is given by Whitehouse (1976); discussion here is confined to a brief outline of the most suitable methods for high accuracy roundness measurement. Multiprobe techniques are particularly suited to in situ measurement but are less attractive at very high accuracies since very careful electrical and mechanical matching of thtransducers is necessary. A further difficulty is that of ensuring coincidence of the plane of measurement for all the transducers concerned. After some consideration, it was decided to restrict the investigation to multi-orientation methods in which only one transducer is used. All multi-orientation methods are dependent on both the spindle and component errors exhibiting good short term repeatability during the measurement cycle. Under controlled environmental conditions, this requirement does not present too much difficulty for the component but it does require that the spindle errors exhibit only a very small amplitude of random fluctuations.

In practice, some random noise, whether specific to the spindle or not, will be introduced into the measurement system. However, providing this is not excessive, it may be effectively removed by averaging the profile over several revolutions and cancelling out random errors by integration.

The effect of averaging can be seen from figure 1 in which the combined error graphs are obtained from the operation of a single stage 0-20 umt filter over one revolution whereas the individual error graphs are produced after averaging over 40 revolutions.

Two multi-orientation methods, differing in both technique and analysis, were selected for investigation and comparison.

The first method, which we shall term the 'multistep technique' (Reason 1966, Spragg and Whitehouse 1968) entails taking a whole series of roundness profiles in each of which the component is stepped through equal angles relative to the spindle.

It is possible, in principle, to separate the errors from just two positions in terms of the differences between samples (this is illustrated for straightness measurement by Thwaite 1973).However, considerably better performance in terms of noise rejection and better behaved types of calculations is obtained by using a series of steps completing a circle of orientations. Analysis of the traces is performed by digital computer using the Fourier coefficients of the profiles.

The second method, termed the ‘reversal technique’ requires two traces to be taken, in the second of which the orientation of both the component and transducer is reversed relative to the spindle. This method, which is better known in straightness measurement, has been applied to roundness measurement by Donaldson and Bowen (see Bryan et al 1967 and Donaldson 1972). In these two traces the relative position of stylus and component is unchanged, while the effect of the spindle error on the stylus at any position is equal and opposite. The component error can be extracted simply as the mean of the two traces while the difference gives the spindle error.

Both multistep and reversal techniques have their respective advantages and disadvantages. The multistep method offers some advantages in setup and lends itself to automated measurement cycles. However, it always suffers to some extent from harmonic distortions and is limited for accuracy to errors involving only low numbers of undulations per revolution. It is also less suited to rotating worktable roundness measuring instruments than it is to rotating stylus instruments. The reversal technique does not suffer from harmonic suppression and the calculations involved are simple. It can be applied to either rotating workpiece or rotating stylus instruments although modification of the instrument may be necessary and automation of the measurement cycle is not so straightforward. Both methods are capable of effecting significant improvement in the accuracy of roundness measurement and are sufficiently different in both theory and method of analysis to justify their use as comparative techniques.

4 Applying the techniques

Repeatability of spindle error is a necessary condition for the successful application of error separation techniques. In general, only very high precision spindles are found to be highly repeatable and it is necessary to use such spindles in order to achieve significant advances in accuracy. However, even given a high quality instrument, it is still difficult to extract improved performance without the use of digital computers. The advantages gained from the computer, whether working on-line (with a direct link to the instrument) or off-line (using, for instance, punched paper tape), are considerable. In addition to its speed of calculation, the computer’s ability to handle large volumes of data enables techniques such as time averaging of the profiles to be usefully exploited.

An off-line error separation system based on the multistep technique has been in use for a number of years at the British Calibration Service Laboratory of Rank Taylor Hobson, Leicester. A photograph of the arrangement is shown in figure 2. The instrument used is a Talyrond 73, a rotating stylus instrument which utilizes a hydrodynamic spindle with a maximum radial error of 0-025 /im deviation from the least squares circle (LSC). The component - in this case a master glass hemisphere as used for spindle checking - is mounted on an indexing table controlled by a modified Geneva mechanism. The accuracy of rotation of the indexing table is unimportant provided it ensures that the profile signal is kept within the effective range of the transducer. A simple encoder is mounted on the spindle to initiate sampling of the profile at 50 equi¬ angular points per revolution. Roundness data are then recorded at each of ten orientations of the component, each orientation being stepped from the previous one by 36°. An Improving the accuracy of roundness measurement  eleventh step gives an additional set of data which should be identical to the first and can be used to identify system drift. At each step, sampling is repeated over four revolutions giving some degree of time averaging as protection against spurious disturbances. The whole system is driven by a hardware controller so that the complete sequence of measurements and steps is made automatically. A total of 2200 data points are logged in a measurement cycle lasting approximately 11 min.The data tape is then presented to a digital computer which averages the data for the four revolutions at each orientation and then uses the set of orientations to calculate the spindle and component errors. The error analysis is printed numeri¬ cally and can be hand plotted if desired. About 2 min are required to process the results on a typical minicomputer and it is quite feasible to use this level of computer power on-line in a dedicated system.The BCS multistep system has been used for checking Talyrond spindles which are held to 0-025 jam LSC and master hemispheres which are held to 0-05 /am LSC and has given excellent service over five years of operation. An illustration of the stability of the errors of a standard spindle is given in figure 3f. This polar chart consists of measurements made over t Reproduced by kind permission of the Director of the British Calibration Service, a period of eight months with five different components having roundness errors from 0-005 fim to 0-07 p.m LSC. Apart from routine oiling the spindle was in constant use for BCS work and was not serviced in any way. It can be seen that repeat¬ ability of the spindle error is of a high order with a maximum radial spread of 0 005 jam. In fact the spread of results is only a little greater than the repeatability of the system (see next section).

Although the BCS system has given excellent repeatability of results, its accuracy has not been verified directly by indepen¬ dent means. To remedy this situation, it was decided to apply the reversal technique to another Talyrond 73 instrument and to compare true profile graphs of a component measured by both systems. A Talyrond 73 was modified both to work on¬ line to a minicomputer at high gain and to allow the reversal technique to be readily applied. The general requirements for interfacing precision surface metrology instruments have been described adequately elsewhere (see, for example, Kinsey and Chetwynd 1973) and are sufficiently straightforward not to be repeated here. A simple encoder, similar to that used in the BCS system, was attached to the spindle allowing accurate repeatability of sampling position at a sampling rate of 512 radial ordinates per revolution. The transducer mount of the instrument was modified to enable the whole transducer to be re-oriented at 180° to its normal position with minimal effect on the balance of the spindle. Re-orientation of the transducer was effected without reversing its direction of measurement and hence changes in the plane of measurement were mini¬mized. Reversal of the component through 180° was accom¬plished using alignment by eye against a simple angular scale.

Working on-line in this system allows much larger total amounts of data to be handled, together with data error checking in ‘real time’. This results in the possibility of taking many more samples during each revolution and also enablelarge numbers of revolutions to be handled and averaged, if necessary. It is also possible to output onto polar graphs the separated and centred errors of spindle and component and to artificially magnify the ‘genuine’ or combined error instru-mental profile (see Chetwynd and Kinsey 1973)。

It will be appreciated that the surroundings of the instrument will tend to influence both the random noise level and long term drift. In the case of this particular system, physical constraints of space necessitated siting the instrument in a relatively bad environment to allow easy on-line access to the computer. To restrict noise and to allow reasonably small numbers of revolutions to give adequate averaging, a single stage analogue filter having a cut-off of about 20 UPR was used. This response is not unduly restrictive for the type of test pieces used since it is primarily the lower harmonics that are of interest.

Averaging was normally performed over 20 revolutions for each orientation in any one run. This gave the same number of revolutions of data as that used with the BCS system. Having  acquired the data, the algorithm was presented with two profiles, each being the mean of the appropriate set of original traces and being mathematically centred using the least squares limacon criterion. Each profile was similarly centred before averaging, a feature which is probably not essential provided the worst case centring error is less than, say, 0-001 of the component radius as will almost certainly be the case for work of the highest precision.

5 Comparative results
A master glass hemisphere was first measured on the on-line Talyrond 73 using the reversal technique and then transferred to another Talyrond 73 equipped to measure using the BCS multistep method. In both cases the plane of measurement was  assessed by eye at 6 mm above the top of the hemisphere mount with the stylus arm inclined at 20° to the vertical. A computer-plotted polar graph of the hemisphere, obtained by using the reversal technique, is shown in figure 4(a) with an original radial magnification of 106. Results for the multistep technique, obtained from the computer printout, are also plotted on this graph. The deviation of 50 equi-angularly spaced points on each profile from the mean values obtained by the two tech¬ niques is shown in figure 4(6). A standard deviation from the mean value of only 0-001 gm (10 Ä) is indicated for each profile. The agreement obtained between the two sets of results is remarkable, particularly since the instruments, operators and environments were different for each case. Of the small variations that do exist, some, in part, could be due to the difference in frequency response between the two techniques and to the fact that the tw'o planes of measurement were probably not precisely coincident.

For both cases of measurement a period of approximately 20 min w'as allowed for component stabilization after handling of the component, A maximum radial difference between repeat measurements of 0-0025 gm was observed using the multistep technique. With the reversal technique a slightly greater maximum difference of 0-004 gm was obtained but this could be attributed to the poorer working environment in which the on-line Talyrond was used. The excellent agreement obtained between the two techniques confirms that, given a suitable spindle, both are capable of effecting significant

improvement in roundness measurement accuracy.

The results described so far have been obtained with a rotating stylus instrument of high quality and proven repeat¬ ability. Tests were also carried out using a lower accuracy  rotating worktable instrument - the Talyrond 100. This instrument is intended primarily for workshop use and employs a simple dry bearing spindle with a radial error of 0-1 gm LSC. The normal maximum radial magnification of this instrument is x 5000. By sampling the transducer signal in time with a start pulse derived from the rotating worktable (which is driven by a synchronous motor), a simple on-line system was established. Measurement of the same hemisphere as was used for the comparison measurements was then accomplished using the reversal technique with averaging over 20 revolutions for each orientation. The resultant polar graph, computer-plotted with a radial magnification of 106,is shown in figure 5。Comparison with figure 4(a) reveals a maximum radial difference of 0-006 gm (0-25 gin) between the two profiles.This represents an improvement in spindle accuracy of about 16 : 1.With dry bearing and roller bearing spindles the relative proportion of random errors to systematic errors is generally greater than with hydrodynamic or hydrostatic bearings. To achieve a given accuracy of measurement it is therefore necesssary to employ considerably more time averaging than with higher accuracy spindles with less random error. For example,20 revolutions in each orientation are necessary to achieve0-006 gm accuracy with the Talyrond 100 spindle; one revolu¬ tion in each orientation is sufficient with the Talyrond 73

spindle.

6 Spindle errors

In addition to extracting the true component profile in the plane of measurement, multi-orientation techniques are capable of extracting the spindle error associated with that particular plane, component radius and direction of measure¬ ment. They thus provide a pow-erful means of examining the performance of an instrument spindle with a precision that has not been attainable previously. In this section the nature of spindle errors is examined briefly, with particular reference to results obtained with Talyrond 73 hydrodynamic spindles.

Rotation of a spindle is generally accompanied by three undesirable degrees of motion, namely radial, axial and angular motion. Radial motion refers to movement of the spindle axis in a direction perpendicular to itself; axial motion  refers to movement along the spindle axis; and angular or coning error refers to tilt of the spindle axis. All three effects can be related to geometric errors in the datum bearings of the spindle and are present to some degree in all types of spindle. In the case of a Talyrond 73 spindle, which consists of a lower hemispherical bearing and an upper journal bearing (see Reason 1966), vertical movement in the lower bearing will lead to axial error, equal radial movement at each bearing will lead to radial error and opposing radial movements at the bearings will lead to angular error. The main effect of radial and angular errors is to introduce radial changes at the point of measurement which appear directly as part of the combined component and spindle errors. A further effect of angular and axial errors is to change the track of the stylus about the component by a small amount. This effect can generally be ignored on precision cylindrical components but can be significant when the transducer and surface to be measured are angled to the vertical (as, for example, when measuring a cone or master glass hemisphere). If the transducer is tilted by an angle ®, then the measured spindle error may be represented as where er and ea are the ‘pure’ radial and axial errors of the spindle; ef and ef are the appropriate error components(corresponding to the particular plane and radius of measure-ment) contributed by angular motion of the spindle. In the following discussion, unless otherwise indicated, the terms radial error and axial error refer to the combined errors(ier+ er') and (ea + ea')respectively, designated by Er and £a.Equation (1) thus becomes Both Er and £a can be extracted using multi-orientation techniques. This exercise has been performed with a Talyrond 73 spindle known to exhibit relatively pronounced errors so that changes of spindle error with <t> can be clearly identified. The results are shown in figure 6. Radial and axial errors were determined by using respectively a precision right cylinder and a right-angled stylus attachment onto an optical flat. In both cases the errors were separated using the multistep technique. In figure 7 the two profiles are shown combined for ® = 20°(applying equation (2)) and the result is compared to the measured spindle error obtained from applying the multistep technique to a master hemisphere with the stylus inclined at 20 =. The close agreement between the constructed and measured cases (maximum radial difference 0-0025 ,um(0-1 pin)) confirms the adequacy of equation(2).It is interesting to note that the spindle error obtained from testing with the master glass hemisphere (the method normally used for determining spindle radial error) is 0-017 jum (0-65 pin) LSC whereas that obtained when measuring the cylindrical specimen is 0-025 pm (1-0 pin) LSC. For this particular spindle there appeared to be some correlation between its radial and axial errors resulting in the spindle error being less at intermediate values of O with a minimum error at approximately 45°. Whether this correlation is typical of the design requires further investigation although, in general, errors with this type of spindle, particularly the axial error, are less pronounced than those shown.

Angular or coning error is usually assessed by measuring a master hemisphere at various distances along the spindle axis and expressing the change in spindle error in terms of so many pm nr or /ain in-1. Far more is revealed about the coning error by examining the spindle error profiles at various distances from and along the axis ; from these results a composite picture can be built up of the way in which the spindle errors change with component size and disposition relative to the spindle. The advent of error subtraction techniques has made such a task relatively straightforward and figure 8 shows results obtained using the multistep method with precision cylindrical ring gauges as components. Rings of 50, 100 and 150 mm diameter were measured with stylus arm lengths of 64, 127 and 254 mm. Component errors were then subtracted from the combined error profile to give the radial spindle error profile corresponding to the particular ring gauge size and its position. It can be seen that, for a particular plane of measurement, the radial error profile is reasonably constant. The profile shape is fairly consistent throughout the series of measurements but its magnitude increases approximately as the length of the stylus arm. This tends to indicate that the radial error term eT' due to angular motion predominates over the ‘pure’ radial error term er. The maximum departure from the least squares circle is indicated below each profile. Examination of  these values reveals a maximum coning error of approxi¬ mately 0-25 /xm nr1 which is equivalent to an angular tilt of about 0-05".

7 Environmental effects 

Mention has already been made of the influence of environmental conditions on instrument accuracy. When working to the exceptional precision afforded by multi-orientation techniques, great care must be taken to minimize these effects, chief of which are those of vibration and temperature. High frequency vibrations - due, for example, to resonance in parts of the instrument excited by floor vibration or by the spindle drive - are invariably noticed in roundness instruments which are used at magnifications in excess of 20 000. At particular rates of sampling these high frequency signals, when digitized into the computer, can produce aliased signals which are well within the frequency bandwidth of interest. For example, on one occasion a vibration with a frequency of about 250 Hz (2500 UPR) was found to produce an aliased signal with a frequency of around 15 Hz (150 UPR). Since the sampled profiles are typically only capable of resolving to about 25 Hz (250 UPR), a practical solution to this problem is to introduce a

low pass filter (as mentioned in §4) into the system before digitizing.

Temperature variations in the form of draughts and inter¬ mittent sources of radiant heat produce temperature gradients within the component and the instrument structure. In addition, mechanical drift can occur in spindle and table movements for some considerable time after centring adjustments have been made. The most common manifestation of such small scale instability is that a roundness graph taken from the instrument at high gain will show a spiralling pattern or failure to close. This indicates relative drift between the instru¬ ment stylus and the component over the duration (usually 10 s) of one spindle rotation. When working with the stylus at an angle to the spindle axis (as, for instance, on a hemisphere)axial drift of the spindle due to variations in oil film thickness will also give rise to a radial signal variation and so show as a spiral trace. Measurements carried out with the on-line Talyrond, situated in a relatively poor environment, indicated that the spiralling error does reduce as the instrument settles from start-up but that it does not decay uniformly to zero. By programming the computer to monitor the spiral error over a large number of spindle revolutions it was found that, after the initial settling period, the failure to close varied in a manner which was, to a good degree of confidence, random about zero mean with a standard deviation of about 0-002 /xm (0-08 /fin).It will take a large number of averaged traces to effectively reduce this spiral effect to the level of the errors being investi¬ gated and it was therefore decided to detect the difference between the beginning and end of the circle of data as a test for local drift. This method was applied to the reversal technique by using the magnitude of the failure to close as a criterion for rejecting drift-affected traces. Instead of averaging over a fixed number of consecutive traces, the averaging is performed only over that number of traces for which the failure to close is less than some specified value, all other traces being ignored. At the end of a measurement sequence the computer can print out the number of revolutions required to produce the specified number of acceptable traces, thus giving an indication of the stability of the conditions in which the instrument is being operated. This system also has the merit that it releases the operator from close attention since it will automatically wait until the instrument has settled down before commencing measurements!

The effect of introducing this simple test was dramatic. Using a criterion of acceptability of 0-0025 /xm (0-10 /xin) for the failure to close, error graphs were found to repeat over several hours with a maximum radial difference reduced from 0-004 /xm (0-16 /xin) without the test to 0-001 /xm (0-04 /xin) atthe 95 % confidence level. Over several weeks, maximum radial difference values of 0-0025 /xm (0-10 /xin) were obtained forcomponents which had been removed from the instrument after measurement and then remeasured later with a completely new setup. These results are all the more remarkable in that they were obtained in an environment far removed from that of a standards room.

 8 Practical high accuracy systems

The results described so far have demonstrated the effectiveness of multi-orientation methods when applied with the aid of a digital computer. Both systems described can be used  successfully with little more skill than that required for operation of a conventional roundness measuring instrument. However, the total time of 20 min or so required for setup and measurement, although relatively short for measurements to this precision, is still restrictive for some inspection require ments. This is particularly true in the case of many precision components at the present state of manufacturing accuracy of, say 0-02 to 0-1 /xm (1 to 4 /xin). At this level the full potential of multi-orientation methods is not required but uncorrected spindle errors are large enough to cast doubts on the magnitude and shape of the measured component errors. Arguing on the basis of the rapidly escalating cost of improving the spindle mechanically, even for a factor of improvement of only 2 or 3, alternative computer based systems are suggested here which may go a long way towards meeting inspection requirements in this problem area. In the two systems to be described, virtually normal instrument operation is retained, but accuracies of the order of 0-005 /xm (0-2 /xin) can be achieved in much the same time as measurement to 0-05 /xm (2 /xin) using the instrument alone.

8.1 Spindle error subtraction
This is a very simple concept based on the good stability of spindle error showm by the Talyrond 73 spindle (see §4).Basically it consists of measuring the spindle error using a multi-orientation technique and storing this shape in the on line computer memory. Then for all normal measurements it is only necessary to take one reading of the component from which is subtracted, point for point, the stored spindle error. In practice one is likely to use the average of several repeat traces to reduce the effect of random error and also to use spiral detection, depending on the accuracy required. Both of these, however, are under control of the computer and the operator merely sets up normally, instructs the computer, and then waits for the results to appear in the form of a numerical or graphical output.An important consideration in applying this technique is that of the various forms of spindle error, as described in §6. The use of a tilted stylus arm leads to a combination of radial and axial errors and so the calibration should be performed with the stylus angled as it will be for the measurements.Alternatively, individual calibrations for radial and axial errors could be made with the computer constructing from these a suitable combination for the stated stylus angle at any particu¬ lar measurement. Although more difficult to measure, the coning error could, in principle, be similarly treated. In practice, the shortest stylus arm length is almost invariably used for high accuracy measurement. For the few cases where longer stylus arms are necessary (e.g. for deep bore measurement) spindle error values corresponding to this case could also be stored. Note, as always, that for optimum accuracy the calibration conditions should be as close as is practical to those of the measurement.

A system working in this way was tested with the on-line Talyrond. Radial and axial errors were determined using the reversal technique and stored on magnetic disc. High accuracy measurements could then be made at a single setting of the component on the instrument. Comparison of results with those obtained from application of the reversal method indicated a typical maximum radial difference of 0’006 ,um(0-25 pin) for measurement over one revolution and 0-0025 pm(0-10 pin) over 20 revolutions. These figures were obtained on components with radii within 10 mm of that used for spindle error calibration and were not found to change materially over the period of several weeks in which measurements were made. With a highly repeatable spindle, recalibration of the system should only be required at infrequent intervals or when measurements of the highest possible accuracy are required.

A polar graph obtained with this system, with averaging over 20 revolutions, is shown in figure 9. This graph can be compared with that in figure 4(a) for the same component, obtained directly by multi-orientation methods. The time required for setup and measurement in this case, however, is approximately one quarter of that required for reversal.

8.2 Component error subtraction

In the system described above, the instrument must be capable of allowing stylus reversal or use with an indexing table even though these features are not used except in calibration. The need for instrument modification or extra equipment could be removed by using a system based on calibration with a circular standard of known error. Calibration could then be accomplished by measuring this standard and reading into the computer a tape record of its errors. The spindle errors can be found by subtracting this record from the measured profile and can then be stored and used as in §8.1. All comments on accuracy stated in §8.1 apply also to this system. One practical limitation of this technique is that it is restricted to detection of radial errors only. In practice this is probably not a serious limitation since the measurement of roundness on most precision components (such as cylinders or spheres) is confined to radial errors.

The stability of the component and the definition of its plane of calibration are of fundamental importance with this tech¬ nique. A suitable standard could take the form of a sphere of fused quartz, carefully mounted to avoid distortion, with a base permitting kinematic location on the instrument table. The plane of calibration could then be defined as the equatorial plane of the sphere. This can be determined with reasonable accuracy by turning point detection when moving the standard relative to the stylus in the direction of the spindle axis. As with spindle error subtraction, the frequency of necessary calibra¬ tion will depend on the accuracy which needs to be guaranteed. It w'ould also be advisable to have the reference standard recalibrated from time to time by the standards laboratory which could then use a multi-orientation system to issue a new error tape.

For both systems described above, the accuracy of measurement on component sizes far removed from that used for calibration will depend on the angular error of the spindle. For the spindle examined in §6, calibration on a 50 mm diameter  component with a standard 64 mm stylus arm would result in a  maximum radial error due to coning of 0-005 /xm (0-2 /jin) at a  diameter of 100 mm and 0-011 pm (0-45 /xin) at a diameter of 150 mm. Corresponding figures with a 127 mm stylus arm are 0-010 pm (0-4 pin) and 0-013 pm (0-5 pin) and, for a 254 mm arm, 0-015 /im (0-6 /xin) and 0-020 /xm (0-8 /xin) respectively. These figures are considerably less than the corresponding errors of the spindle (see figure 8).
The computing power necessary for these and previous systems is well within the capabilities of a small minicomputer and an on-line system based on a Talyrond 73 is shown in figure 10. The computer (in this case a Data General Nova 2/4 with 16k core store) is mounted under the instrument desk with operation controlled via a teletypewriter. In addition to the instrument’s polar recorder, profiles can also be examined on a display unit mounted above the transducer amplifier. Error separation could be used in conjunction with automatic compensation of eccentricity, computer magnification of the profile and additional features such as harmonic analysis and the plotting of best fit and envelope-based reference lines to give a system of high versatility, speed and accuracy. The extra cost of the computer and associated equipment is currently of the same order as that of the instrument alone. Present trends indicate that this extra cost relative to the instrument is likely to decrease in the future.

9 Suggestions for further work

The results obtained so far have taken roundness measurement well into the region known as nanotechnology (based on the nanometre which is equivalent to 10 Ä, about 10 atomic diameters). In this area the physics of interaction between the stylus and the surface to be measured provides a fundamental limit to the ultimate accuracy that can be achieved. Effects such as elastic deformation and local welding will depend to a large extent on such factors as the physical and surface properties of the component and stylus, the measuring force and shape of the stylus and the relative velocity between surface and stylus. In the case of the tungsten carbide/glass combination used for most of the measurements described in this paper, noise levels were found to vary markedly with the stylus force and degree of cleanliness of the glass surface. This was a contributory factor in restricting the upper limit of frequency response that could be obtained using the reversal technique. Some attention has already been paid to the stylus/surface interface by Whitehouse (1974) but more work needs to be done to assess ways in which these effects can be quantified and minimized.

Environmental effects are another important limitation on the accuracies that can be achieved. For work of the highest accuracy it is desirable to provide the best possible working environment in terms of constant temperature, steady air conditions and freedom from vibration. In practice such ideal working conditions are not often available. However, as shown in §7, simple drift checking procedures can be built into the system to reduce these effects, at the expense of increasing the time of measurement. Further research into more sophisticated methods of reducing environmental effects could reduce the time required for measurement to a given accuracy.

In addition to the above effects, both the multistep and reversal techniques have their respective theoretical and practical limitations. The main disadvantage of the multistep method is its harmonic suppression. That of the reversal method is the need to stop the spindle and reverse the stylus. Both these features can lead to residual errors which, although unimportant for many purposes, act as limitations to the ultimate accuracy of measurement that can be achieved. One method of overcoming these disadvantages, although possibly at the expense of introducing other problems, is to mount an additional identical transducer at 180° to the standard instru¬ ment transducer. By electrically switching between the two transducers after reversing the component through 180°, the reversal technique can be carried out without stopping the spindle and without any change in loading conditions on the  spindle. Reversal of the component could be effected using a motorized indexing table, as in the BCS multistep technique, to give a fully automatic measurement cycle. A further reduction in measurement time could be achieved by multiplexing between the two transducers when sampling, thus obtaining twice as much data per revolution. The two transducer outputs could also be combined to obtain diametral, as well as radial, variations in the shape of the part, a feature of interest to some bearing manufacturers. Such a system, although considerably more expensive and complex than any previously described and with the drawback of requiring careful matching of the trans¬ ducers and their plane of measurement, could give very high accuracy of measurement within a relatively short measure¬ ment cycle.

10 Conclusions

The feasibility of using error separation techniques to improve the accuracy of roundness measurement has been demonstrated experimentally. The most effective way of implementing these methods is with the aid of a digital computer, preferably working on-line. Several practical computer based systems have been evaluated and reduction in instrument errors by a factor of ten or more has been achieved. Results from these systems have led to greater insight into the behaviour of preci¬ sion instrument spindles. Although representing a significant advance in the accuracy of roundness measurement, the techniques used here are, in principle, capable of still further refinement.

Acknowledgments

The authors acknowledge the valuable discussions held with colleagues and wish to thank the Directors of Rank Taylor Hobson for permission to publish.

转载于:https://www.cnblogs.com/chaining/p/9454507.html