Comparing Preparation and Calibration Techniques for Ultrasonic Bolt Tension Measurement

Transcription

1 Comparing Preparation and Calibration Techniques for Ultrasonic Bolt Tension Measurement Conducted by: September 12, 2007 No part of this publication may be reproduced in any form, in any electronic retrieval system or otherwise, without the prior written permission of. Copyright 2007

2 Ultrasonic Bolt Prep and Cal September Introduction and Objectives 1.1 Introduction The use of ultrasonics in calculating bolt tension through measurement of strain has increased in recent years due largely to improvements in application of the technology. Advances in user interface including the use of Windows-based systems has both improved ease of use and enabled traceability of settings and setup that were once established by knob twisting and then were lost or were unverifiable. One main source of repeatability errors, the differences in acoustic coupling each time the transducer is applied to the fastener, has been eliminated in some systems by integrating or attaching the transducer s pulse generation to the fastener so only electrical connection is made on successive measurements. Some liquid-coupled systems are also addressing this issue through echo triggering designed to negate this effect. Another potential source of error that the test system OEM can only partially impact through signal processing and software and features is bolt preparation and calibration. This is usually the test practitioner s responsibility, and as such invokes discussion and debate as to best practices. This report will attempt to provide some comparative data to aid practitioners in evaluating their alternatives. 1.2 Test Objective While all ultrasonic measurement systems have the capability of measurement using data base material constants, such as acoustic velocity and stress factor, for many applications greater accuracy is desired. Because within a given material type these material constants are not truly constant, the fasteners being tested are often calibrated by elongating them while simultaneously measuring the increasing time delay of the ultrasonic pulse and the load required to produce that delay through elongation. The procedure for performing that calibration and for preparing the bolt ends is not universally defined, either by independent specification or by general acceptance that there is an obviously superior approach. The purpose of this report is to provide quantitative guidance in addressing the most common questions test practitioners and their customers have regarding preparation and calibration. These questions are: 1. Is it better to calibrate the bolt by directly tensioning it, or by elongating it by rotation relative to a mating nut member? 2. How sensitive is calibration accuracy to the parallelism of the prepared bolt ends? 3. How good a surface finish is required of the prepared bolt ends? 4. Related to questions of required parallelism and finish, what are the best methods of preparing the ends of the fastener flat and parallel to one another? 5. Is better accuracy achieved by measuring tension based on the individual calibration of each bolt or based on an average calibration of a subset of bolts? 6. Does the bolt s grip length have an influence on the best calibration method to choose? Page 2

3 September 2007 Ultrasonic Bolt Prep and Cal 1.3 Test Plan To provide the answers to these test objectives the following test plan was developed. The explanation for the headings in Table 1 will be provided in the sections describing each test. Objective Test ID Cal Method Table 1 Test Plan Grip Finish QTY Cal Basis Effect of 1 Tensile med fine 10 Self finish/angle Calibration 2 Tensile med extra fine 10 Self linearity 3 Tensile med ground 10 Self Calibration linearity Tension Accuracy 4 Tensile short ground 5 Self 4a Tensile short ground 5 Ave 5 Tensile long ground 5 Self 5a Tensile long ground 5 Ave 6 Torque short ground 5 Self 6a Torque short ground 5 Ave 7 Torque long ground 5 Self 7a Torque long ground 5 Ave 8 Tensile short extra fine 5 Self 8a Tensile short extra fine 5 Ave 9 Tensile long extra fine 5 Self 9a Tensile long extra fine 5 Ave Grip Lengths (mm) - Short: 23.4, Med: 31.4, Long: Test Equipment 2.1 Test Fasteners The fastener used for all tests was an M10 x 1.5 property class 10.9 hex head cap screw 50 mm long and threaded 26mm with a phosphate/oil finish. They were procured from a national industrial supplier along with mating hex nuts of the same property class and finish. 2.2 Ultrasonic Measurement The basis of measurement is the Micro Controls MC900 transient recorder with ultrasonics option (Figure 1). This four channel system operates in the MS Windows environment and permits simultaneous recording of other sensor inputs in addition to the ultrasonic signal. The transducer used is a 3mm x 3mm x 0.1 mm thick 7.5 Mhz piezo ceramic sensor bonded to the prepared bolt. A separate magnetic pickup and cable transfers the sensor signal to the transient recorder. These elements are shown in Figure 2. Page 3

4 Ultrasonic Bolt Prep and Cal September 2007 Figure 1 MC900 transient recorder with ultrasonics Figure 2 Piezo senor bonded to bolt and signal pickup (inset) 2.3 Load Measurement Two sources of load measurement were required for the various test phases. A 80 kn (16,000 lb) Sintech universal test machine (i.e. tensile tester or UTM) was used in tensile bolt calibration, while an RS Technologies 100kN (22,500 lb) tension load cell was the source of load measurement for torque-based bolt calibration and was the standard for tests of calibration accuracy comparisons. They are pictured in Figure 3 and 4, respectively. Page 4

5 September 2007 Ultrasonic Bolt Prep and Cal Figure 3 Sintech universal test machine Figure 4 RS Technologies tension load cell 2.4 Torque Input To provide consistent tightening for torque-based calibration, 401 N-m (295 ft-lb) Stanley DC nut runner was fixtured to a 542 N-m (500 ft-lb) Midwest Specialties torque which was balanced to float the nut runner in position. The nut runner is shown in position in Figure 5. An external torque transducer was fitted to the nut runner so that torque tension traces could be captured for reference if desired. Page 5

6 Ultrasonic Bolt Prep and Cal September 2007 Figure 5 Fixtured DC Nutrunner with external torque-angle transducer 3.0 Test 1 - Finish and Parallelism of Bolt Preparation Techniques 3.1 Objective Two interrelated test objectives listed in Section 1.2 were the desire to provide guidance as to the effect of the finish and parallelism of the bolt ends on calibration and measurement accuracy, and what values different preparation methods might be expected to produce. Several preparation methods are used by practitioners; surface grinding, milling, turning and disk sanding. To contain the size of this test only surface grinding and disk sanding were tested as they would be considered the two extremes of potential accuracy. In this section the results of measuring the surface finish and parallelism of bolts prepared by these methods will be summarized. In Section 4.0 the results of calibrating bolts by these methods will be compared. 3.2 Test Method Finish and parallelism measurements were taken for three preparation processes. A summary of these processes is shown in Table 2. Test ID 1 Disk Sand - Fine Table 2 Bolt Preparation Processes Process Fixturing Comments In V-Block w/ 12 disk (see Fig. 6a) 60 grit pass followed by 120 grit Page 6

7 September 2007 Ultrasonic Bolt Prep and Cal 2 Disk Sand Extra Fine 3 Surface Grind In V-Block w/ 12 disk (see Fig. 6) Bolts threaded into fixture block to end of threads pass 60 grit pass followed by 180 grit pass See Fig. 6b for an appearance comparison Figure 6a Disk sanding setup Page 7

8 Ultrasonic Bolt Prep and Cal September 2007 Figure 6b Comparison of disk sanded (left) and surface ground (right) bolts 3.3 Test Results After the bolts were prepared the surface finish of each bolt was measured with a Mitutoyo SJ- 201P profiliometer. Three readings were taken on each bolt. A summary of the average of those three readings is shown in Table 3. Table 3 Comparison of surface finish by preparation method Surface Finish, Ra (μm) Ave Min Max Std Dev Disk - Fine Disk - Extra Fine Surface Ground Surface Finish, Ra (μin) Disk - Fine Disk - Extra Fine Surface Ground After surface finish readings were taken each bolt was placed shank down on a surface plate and two pairs of measurements were taken at the perimeter of the head with a 0.001mm ( in) resolution Mitutoyo gage head and amplifier (Fig.7). Page 8

9 September 2007 Ultrasonic Bolt Prep and Cal Figure 7 Parallelism measurement setup. Height measurements were taken with a 0.001mm ( in) resolution gage head relative to a surface plate on which the shank was placed. The angle of the head relative to the shank as defined by the readings at the 0 and 180 position were calculated as was the angle relative to the 90 and 270 line. The average of those angles for the ten bolts in each group are summarized in Table 4. Table 4 Angularity between bolt head and shank, degrees Ave Disk - Fine Disk - Extra Fine Surface Ground Viewed independently, the data in Tables 3 and 4 is of limited value. In Section 4.0 and Section 5.0 the impact of these differences will be better illustrated as the linearity of calibration and the error of ultrasonic tension readings will be measured and compared. 4.0 Test 2 Calibration Linearity and Repeatability of Bolt Preparation Techniques 4.1 Objective In this test the impact different bolt preparation processes described in Section 3.0 have on bolt calibration will be examined. At the same time the two fundamentally different means of tensioning the bolt to allow the correlation between elongation and load will be examined. As there is disagreement as to whether torque or tensile calibration is more accurate, it follows that Page 9

10 Ultrasonic Bolt Prep and Cal September 2007 there are proponents of both methods and it is not entirely clear which is the more commonly used method. The torque backers point to the fact that since torsion will be applied to the bolt in the actual joint when measurements will be taken, it only follows that it be calibrated in the same manner. While the tension crowd generally agrees that s true in theory, they will counter that the quantitative benefit is unclear and unless the actual nut member and the bearing surface/finish are used in the calibration, the simulation benefit is reduced. Also, as the underhead and thread surfaces are disturbed, the calibrated bolts are generally not used for testing, thus increasing the customer s cost. 4.2 Test Method The two means of creating bolt elongation for calibration are by directly tensioning the bolt, generally in a UTM, or by applying torque to the bolt or mating nut in a load cell or UTM. A common means of fixturing the bolt for tensile calibration is with the use of a nut and washer plate of the correct size and thread pitch. An example is shown in Figure 8. Figure 8 An example of fixturing for tensile bolt calibration. Torque-based calibration can also be performed in a load frame as shown in Figure 8, although the fixturing must be altered to provide access for the tool applying the torque and for resisting rotation. Another means of torque calibration, and the one that will used in this test, is in a load cell as shown in Figure 9. A set of plates is required for each fastener size which commonly include a pocket to restrain square test washers that provide a fresh non-rotating bearing surface for each rundown, and another pocket to prevent anti-rotation of the fixed nut or bolt head. Page 10

11 September 2007 Ultrasonic Bolt Prep and Cal Figure 9 Tension load cell used for torque calibration. A replaceable test washer can be seen under the bolt head. Independent of what means of calibration is used, the grip length of the test setup must simulate the actual joint as this has a direct impact on the relationship between load and elongation. In this test the piezo sensor was always applied to the tail of the bolt. While this generally not a preferred tightening location when there is an option, in this case it was done in an attempt to maximize torsional stress applied to the bolt, as this is one of the reasons sited for differences between tensile and torque-based calibration. 4.3 Test Results Before summarizing the calibration results it is helpful to examine a typical calibration output. Figure 10 is a graph of load vs. time delay for ten bolts undergoing calibration. Time delay is the increase in time for the pulse emitted from the piezo sensor to return after reflecting off the opposite end of the bolt, compared to the time required when the bolt is unloaded. The green lines are the results of calibrating the individual bolts comprising the group used to generate the calibration file. Although it is difficult to see, the red line is a 1 st order best-fit average of the ten plots created through linear regression. The C1 and C0 coefficients represent the slope and intercept of that line. The Correlation Coefficient (Corr Coeff) and the Standard Error of the Estimate (Std Err Est) are measures of how well that average line fits the individual data. As comparative Standard Errors calculated from our tests will be plotted, the source of the value should be defined: Std Err Est = ( Y Y ) N Where: Y = actual value Y = predicted value N = number of data points 2 Page 11

12 Ultrasonic Bolt Prep and Cal September 2007 Figure 10 Average calibration plot showing the best-fit line (red) and the individual calibration plots (green) from which it was calculated. The result of calibration can also be represented as an error plot. Figure 11 shows an example of this, where the error of each bolt s individual calibration relative to the best-fit line is plotted. The coefficients found at the bottom of the plot are the same ones found in Figure 10. Data from this plot will be used to generate the comparative data compiled in the remainder of this section. Page 12

13 September 2007 Ultrasonic Bolt Prep and Cal Figure 11 Error plot of individual bolt calibration. Before listing the comparative results it may be helpful to review, Table 1, the test plan introduced in Section 1. Objective Test ID Cal Method Table 1 Test Plan Grip Finish QTY Cal Basis Effect of 1 Tensile med fine 10 Self finish/angle Calibration 2 Tensile med extra fine 10 Self Linearity 3 Tensile med ground 10 Self Calibration Linearity Tension Accuracy 4 Tensile short ground 5 Self 4a Tensile short ground 5 Ave 5 Tensile long ground 5 Self 5a Tensile long ground 5 Ave 6 Torque short ground 5 Self 6a Torque short ground 5 Ave 7 Torque long ground 5 Self Page 13

14 Ultrasonic Bolt Prep and Cal September a Torque long ground 5 Ave 8 Tensile short extra fine 5 Self 8a Tensile short extra fine 5 Ave 9 Tensile long extra fine 5 Self 9a Tensile long extra fine 5 Ave Grip Lengths (mm) - Short: 23.4, Med: 31.4, Long: 39.4 The results that follow summarize the linearity and repeatability of the individual calibration plots by comparing them to the best fit line as shown in Figure 11. These results are provided for Test ID s 1 thru 9. The test groups with a suffix, such as 4a do not have an independent calibration, but instead use the average calibration file for the primary ID, such as 4. For example, Group 4 consists of 5 bolts that were calibrated. From that group of calibrations an average calibration file was created (the red line in Figure 10). That average calibration file will used for tension readings on 5 additional bolts that were prepared the same way but were never calibrated (Group 4A). In the next section each bolt will be placed in a load cell and tensioned to three different levels with both the load cell and ultrasonic tension values recorded. One of the comparisons that will be made is the difference in error between the bolts that use their own (self) calibration basis and those that use the average calibration file of other bolts from the same batch as the basis for calibration. Table 5 and Figures 12 and 13 compare the calibration errors of each test group. Test ID Table 5 Comparison of Calibration Errors by Group Cal Meth Prep Meth Grip Std Err Est Max Error% + Max Error% - 1 Tensile Fine Med Tensile Extra Fine Med Tensile Ground Med Tensile Ground Short Tensile Ground Long Tensile Extra Fine Short Tensile Extra Fine Long Torque Ground Short Torque Ground Long Page 14

15 September 2007 Ultrasonic Bolt Prep and Cal Comparison of Std Errors of the Estimate Std Err Est Test ID Figure 12 Standard errors of the estimate for the nine test groups Range of Maximum Error Maximum Error, % Max Error + Max Error Test ID Figure 13 The maximum individual error readings of each test group Finally, it is common across a range of tensile readings or calibrations to exercise the setup, by tensioning it prior to taking the actual reading or calibration. That was done on each of the tensile calibrations. As opposed to standard practice, results of the preliminary tensioning were recorded and a calibration file created so that the calibration of the first run could be compared to the second to get insight on the effectiveness of the practice. The results, shown in Figure 14, show that this preliminary tensioning does appear to be of value, providing an average 12% Page 15

16 Ultrasonic Bolt Prep and Cal September 2007 decrease in Standard Error of the Estimate. The tensile calibration files used in all test comparisons were second runs. Effect of Pre-Tensioning on Standard Error Run 1 Run 2 Std Err Est Tensile Cal Group Figure 14 Standard error of the estimate comparison of the 1 st and 2 nd runs of tensile calibrations. 5.0 Test 3 Accuracy of Ultrasonic Tension Readings by Test Group 5.1 Objective While Test 1 and Test 2 compared important elements that influence the accuracy of ultrasonic bolt tension readings they are in reality invisible to the final product which is the bolt tension that is recorded during a test, such as determining the relationship between torque and tension in the actual joint, or monitoring bolt tension over time or during a dynamic test. In this section we will compare the test groups by the accuracy in which ultrasonic bolt tension readings match an independent load cell reading. 5.2 Test Method The bolts in Groups 4 through 9 of Table 1, including the a suffix groups, will be the subject of this test. If the accuracy of a measurement is to be determined it must be compared against a standard. For this test determining what that standard should be was not obvious. Because the tooling available for the UTM was not configured to allow torque-based elongation, the tensile and torque calibrated bolts were calibrated to two different load standards. The options would be to tension the bolts in the same setups in which they were calibrated, or to compare all groups using the same setup. As the load frame used to provide the tensile calibration was not an effective setup for torque application it was decided that all readings would be taken on the tension load cell used for creating the torque-based calibrations (Fig. 4), and this would be the standard against which all ultrasonic bolt tension readings would be compared. Page 16

17 September 2007 Ultrasonic Bolt Prep and Cal Each of the bolts in test Groups 4 through 9a was inserted into the load cell. The bolt was turned manually until the desired tension was reached on the load cell. the tension readings from both the load cell and the ultrasonic channel were then recorded. The procedure was then repeated in tensioning the bolt to the next level. Three tension readings were taken for each bolt; at 30 kn, 35kN and 40 kn. One of the reasons these values were chosen is that the 40 kn reading is beyond the maximum tension at which the bolts were calibrated, providing the opportunity to compare the accuracy of that reading the those within the calibration range. All bolts were calibrated to 36 kn, 75% of the bolt s proof load. 5.3 Test Results Figure 14a thru 14d compare the differences between the ultrasonic readings and the load cell expressed as percent error. The number in the legend corresponds to the group ID number. Short Grip - Self Cal Error 7% 6% 5% 4% 3% 2% 1% 0% 30 kn 35 kn 40 kn Tension Torque (#6) Tensile (#4) Disc (#8) Error 7% 6% 5% 4% 3% 2% 1% 0% Short Grip - Ave Cal 30 kn 35 kn 40 kn Tension Torque (#6a) Tensile (#4a) Disc (#8a) Page 17

18 Ultrasonic Bolt Prep and Cal September % Long Grip - Self Cal Error 2.0% 1.5% 1.0% 0.5% Torque (#7) Tensile (#5) Disc (#9) 0.0% 30 kn 35 kn 40 kn Tension Long Grip - Ave Cal 2.5% Error 2.0% 1.5% 1.0% 0.5% Torque (#7a) Tensile (#5a) Disc (#9) 0.0% 30 kn 35 kn 40 kn Tension Figure 14 a-d Error of the ultrasonic readings relative to the load cell readings Page 18

19 September 2007 Ultrasonic Bolt Prep and Cal Figure 15 summarizes the comparative error of all groups by averaging the errors of all three tension readings. 7% Average Error Across All Loads Error 6% 5% 4% 3% 2% 1% Self Cal Ave Cal 0% Short Torque Short Tensile Short Disc Long Torque Long Tensile Long Disc Configuration Figure 15 Average error of all three tension readings by group Figure 16a thru 16d are an attempt to display the error data in a manner that visually conveys the linearity and range of the difference in readings between the ultrasonic and load cell channels. 1.0 Short Grip - Self Cal 0.5 Delta Tension, kn Torque, max Torque, min Tensile, max Tensile, min Disc, max -3.0 Disc, min Tension, kn Page 19

20 Ultrasonic Bolt Prep and Cal September Short Grip - Ave Cal Delta Tension, kn Torque, max Torque, min Tensile, max Tensile, min Disc, max Disc, min Tension, kn Long Grip - Self Cal 2.5 Delta Tension, kn Torque, max Torque, min Tensile, max Tensile, min Disc, max Disc, min Tension, kn Long Grip - Ave Cal 2.5 Delta Tension, kn Torque, max Torque, min Tensile, max Tensile, min Disc, max Disc, min Tension, kn Figure 16 a-d Measurement differentials by test configuration Page 20

21 September 2007 Ultrasonic Bolt Prep and Cal Probably the result that stands out most prominently is the high error of the short grip length relative to the long grip, particularly for the tensile calibrated groups. This is an area where it is possible that the test method influenced the results. As grip length decreases it is increasingly more important that the grip established during calibration matches the actual grip in the joint. To illustrate this Figure 17 shows the effect of adding 2 mm to the grip length before taking comparative readings in the load cell. The ultrasonic readings went from about 3% low to about 5% high, quite an impact for a relatively small change. A likely reason for this is the thread length of this bolt relative to the grip length. Because the short grip length only exposes a few threads, the 2 mm increase in grip increases the number of exposed threads by a large percentage. Since the cross-sectional area of the threaded portion is smaller than the unthreaded shank a disproportional level of elongation occurs in the threaded portion. Error 6% 5% 4% 3% 2% 1% 0% -1% -2% -3% -4% Short Grip - Ave Cal Grip + 2mm Std Grip 30 kn 35 kn 40 kn Tension Figure 17 The effect of variation in grip length between calibration and test for test group 4a. However the behavior of tensile vs. torque-based tensioning relative to grip length should be examined in a more fundamental way. When calibration is performed through tensile elongation the distance between the bolt head and the nut member increases with applied tension, while in torque-based calibration that distance remains essentially the same but the mating threads are advanced, thus shortening the bolt. Therefore, holding all else equal, the bolt calibrated by being elongated with torque will calibrate as being a bit stiffer that being calibrated with tension. This effect is somewhat normalized relative to grip length because although the amount which the tensile grip length will change increases with increasing grip length as the bolt becomes less stiff, the effect of that change is decreased because it is a smaller proportion of the total grip. 6.0 Comments on Test Results 1. As this test was based on only a single bolt style and sample sizes that are smaller than ideal, the applicability of results should be viewed narrowly and as guidance only. Page 21

22 Ultrasonic Bolt Prep and Cal September Considered independent of parallelism and based on experience in addition to this test, surface finish readings greater than approximately 2.0 µm (79 µin) R a are not desirable, primarily because the amplitude of the signal is diminished. The gain required to maintain a given amplitude with the Fine Sanded group was approximating 20% greater than the Extra Fine sanded group ( 2.63 µm vs µm R a ). With bolts that don t require much gain to achieve good amplitude this may not be a problem but high-gain applications increase signal noise. Finishes finer that approximatly1.0 µm (39 µin) R a don t appear to add any additional value and may decrease sensor bond strength. Note that these comments may not translate precisely to liquid-coupled ultrasonic systems as the sensor-to-bolt interface is fundamentally different. 3. The affect of parallelism is a continuous spectrum rather than the more bounded sweet spot of finish. All incremental improvements in parallelism and flatness provide incremental improvements in accuracy, albeit getting smaller as control gets greater. As such it is hard to state a cut-off value. The angular error of the extra fine sanded bolts were about 50% greater than the ground bolts but the standard errors of the three groups prepped with this method were essentially equal to the five groups that were ground. While this could be a case of small sample sizes or narrow application, additional experience indicates that the disk sanding method can be used with good results, and is particularly beneficial when bolt quantities are small. This method is more operator-dependant than other methods and is at least as sensitive to setup quality as others so it should be used with caution and with a means to check the result. Flatness/parallelism generated from milling and turning tends to fall in the middle, although surface finish should be monitored because in can be worse than any of the methods tested here. 4. The fact that in 33 of 36 instances the average calibration files (applied to bolts that were never calibrated) resulted in smaller errors that bolts that used their own individual calibration file may seem surprising. However, whether that should be surprising depends on the source of calibration variation from bolt-to-bolt. If the variation were mainly due to differences in the inherent acoustic properties within each bolt, self calibration should be superior. However, simply taking the same bolt and calibrating it multiple times in the same setup (beyond the initial exercise run) will result in small variations on each calibration. This indicates that the manner in which the ultrasonic pulse travels through the bolt varies a small amount each time it is tensioned. Add to this the fact that the way each successive bolt sits the tooling is not identical, and the conclusion is that more often than not the tooling and test signal variation is larger than the inherent variation in material properties so that the average calibration file better accounts for this variation. 5. The outcome of the Pullers vs. Twisters debate is not clear-cut. The torque-based calibrations showed a fairly significant reduction in Standard Error, while the tensile tension readings were slightly more accurate. To complicate things, use of the load cell gave the torque calibrated groups a bit of a home-field advantage in tension readings as both the calibration and readings were performed using the same standard. There was no evidence in this test that the adding torsional stress to the bolt during calibration had any benefit for torque calibration. However, the ability to accurately duplicate actual grip length during calibration clearly has Page 22

23 September 2007 Ultrasonic Bolt Prep and Cal Page 23