17.12.2012 Views

Engineering Instrumentation and Measurement Experiments

Engineering Instrumentation and Measurement Experiments

Engineering Instrumentation and Measurement Experiments

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

MAE 300 - <strong>Engineering</strong><br />

<strong>Instrumentation</strong> <strong>and</strong> <strong>Measurement</strong><br />

Calibration of an Electronic Load Cell<br />

<strong>and</strong> a Bourdon Tube Pressure Gage<br />

Experiment No. 2<br />

Kai Gemba, 003678517<br />

California State University<br />

Department of Mechanical <strong>and</strong> Aerospace <strong>Engineering</strong><br />

MAE 300 - <strong>Engineering</strong> <strong>Instrumentation</strong> <strong>and</strong> <strong>Measurement</strong><br />

Instructor: Rahai, Hamid R.<br />

October 14, 2007


Abstract<br />

Calibration procedures were performed by students on an electronic load-cell scale <strong>and</strong> mechanical<br />

Bourdon tube-type pressure-gage scale. Both apparatuses were incrementally loaded to<br />

full scale <strong>and</strong> data was recorded for both the load <strong>and</strong> unload direction. The data was plotted,<br />

statistically evaluated <strong>and</strong> the method of least squares was applied to determine a best-fit analytical<br />

expression for the calibration function. The Bourdon Gage output was closely correlated<br />

to a linear calibration function throughout the test range, R 2 vales ranging from 0.996 to 0.9993.<br />

Hysteresis in the unload direction was less than the output gage precision. The strain gage output<br />

was less linear but a fist order polynomial was used, too. R 2 values ranging from 0.981 to 0.978.


Experiment No. 2<br />

Contents<br />

1 Objective of experiment 1<br />

2 Background <strong>and</strong> Theory 1<br />

2.1 Intercept Calculations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2<br />

2.2 St<strong>and</strong>ard error of the linear relationship . . . . . . . . . . . . . . . . . . . . . . . . . 2<br />

2.3 Coefficient of Determination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3<br />

3 Experimental procedure 3<br />

4 Experimental data 4<br />

5 Calculations <strong>and</strong> Results 5<br />

5.1 Bourdon Gage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5<br />

5.2 Strain Gage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7<br />

5.3 Sample Calculation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9<br />

6 Discussion of results 10<br />

6.1 Bourdon Gage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10<br />

6.2 Strain Gage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10<br />

7 Conclusions <strong>and</strong> recommendations 10<br />

Bibliography 11<br />

i


Experiment No. 2<br />

List of Figures<br />

1 Bourdon Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5<br />

2 Bourdon Calibration Residuals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6<br />

3 Strain Gage Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7<br />

4 Strain Gage Calibration Residuals . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8<br />

List of Tables<br />

1 Stain Gage Calibration Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4<br />

2 Bourdon Gage Calibration Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4<br />

3 Curve Fit Calculation for Bourdon Gage . . . . . . . . . . . . . . . . . . . . . . . . . 9<br />

ii


Experiment No. 2 Calibration of a Electronic Load Cell <strong>and</strong> a Bourdon Tube-Type Pressure Gage<br />

1 Objective of experiment<br />

The stated objective of the experiment is to familiarize students with the calibration procedure <strong>and</strong><br />

method of least squares<br />

2 Background <strong>and</strong> Theory<br />

The following descriptions of the equipment <strong>and</strong> the analytic methods for evaluating the data were<br />

obtained from the web resources noted in the bibliography <strong>and</strong> are consistent with the presentation<br />

in the course textbook.<br />

In a Bourdon tube gauge, a C shaped, hollow spring tube is closed <strong>and</strong> sealed at one end. The<br />

opposite end is securely sealed <strong>and</strong> bonded to the socket. As the gauge pressure increases the tube<br />

will tend to uncoil, while a reduced gauge pressure will cause the tube to coil more tightly. This<br />

motion is transferred through a linkage to a gear train connected to an indicating needle. The needle<br />

is presented in front of a card face inscribed with the pressure indications associated with particular<br />

needle deflections.<br />

Two points of the wheatstone bridge are connected to an exciter voltage (from the battery or AC<br />

adapter) <strong>and</strong> an output analog voltage feed to a A/D Converter. The output voltage being fed in<br />

the A/D varies in proportion to the load applied to the platform of the scale. This occurs since the<br />

weighing platform of a scale is connected to the end of the load cell via a post. The applied force<br />

is transferred from the platform, through the post <strong>and</strong> onto the aluminum beam. When the beam<br />

bends the strain gauges bend resulting in their resistance value to change <strong>and</strong> the voltage changes<br />

in proportion to the load applied to the platform.<br />

Least squares or ordinary least squares (OLS) is a mathematical optimization technique which, when<br />

given a series of measured data, attempts to find a function which closely approximates the data<br />

(best fit). It attempts to minimize the sum of the squares of the ordinate differences (called residuals)<br />

between points generated by the function <strong>and</strong> corresponding points in the data. Specifically, it is<br />

called least mean squares (LMS) when the number of measured data is 1 <strong>and</strong> the gradient descent<br />

method is used to minimize the squared residual. LMS is known to minimize the expectation of<br />

the squared residual, with the smallest operations (per iteration). But it requires a large number of<br />

iterations to converge.<br />

An implicit requirement for the least squares method to work is that errors in each measurement be<br />

r<strong>and</strong>omly distributed. The Gauss-Markov theorem proves that least square estimators are unbiased<br />

<strong>and</strong> that the sample data do not have to comply with, for instance, a normal distribution. It is also<br />

important that the collected data be well chosen, so as to allow visibility into the variables to be<br />

solved for (for giving more weight to particular data, refer to weighted least squares).<br />

1


Experiment No. 2 Calibration of a Electronic Load Cell <strong>and</strong> a Bourdon Tube-Type Pressure Gage<br />

2.1 Intercept Calculations<br />

We sum the observations, the squares of the Ys <strong>and</strong> Xs <strong>and</strong> the products X × Y to obtain the<br />

following quantities.<br />

<strong>and</strong> Sy similarly.<br />

<strong>and</strong> S + Y Y similarly<br />

SX = x1 + x2 + ... + xn<br />

SXX = x 2 1 + x 2 2 + ... + x 2 n<br />

SXY = x1y1 + x2y2 + ... + xnyn<br />

We use the summary statistics above to estimate the slope β.<br />

β ≈ mSXY − SXSY<br />

nSXX − SXSX<br />

We use the estimate of β to estimate the intercept, α:<br />

α ≈ SY − βSX<br />

(2)<br />

n<br />

A consequence of this estimate is that the regression line will always pass through the center of the<br />

data.<br />

2.2 St<strong>and</strong>ard error of the linear relationship<br />

The following is excerpted from Applications, Basics, <strong>and</strong> Computing of Exploratory Data Analysis,<br />

Velleman <strong>and</strong> Hoaglin, Wadsworth, Inc. 1981.<br />

A fundamental step in most data analysis <strong>and</strong> in all exploratory analysis is the computation <strong>and</strong><br />

examination of residuals. [...] Most analysts propose a simple structure or model to begin describing<br />

the patterns in the data. Such models differ widely in structure <strong>and</strong> purpose, but all attempt to<br />

fit the data closely. We therefore refer to any such description of the data as a fit. The residuals<br />

are, then, the differences at each point between the observed data value <strong>and</strong> the fitted value, or the<br />

actual value <strong>and</strong> the predicted value: Residual = Actual value − Predicted value.<br />

The median-median line provides one way to find a simple fit, <strong>and</strong> its residuals, r, are found for<br />

each data value, (xi, yi) as<br />

ri = yi − (axi + b)<br />

A pessimist might view residuals as the failure of a fit to describe the data accurately. He might even<br />

speak of them as errors, although a perfect fit, which leaves all residuals equal to zero, would arouse<br />

suspicion. An optimist sees in residuals details of the data’s behavior previously hidden beneath<br />

the dominant patterns of the fit. Both points of view are correct. The best fits leave small residual,<br />

<strong>and</strong> systematically large residuals may indicate a poorly chosen model. Nevertheless, even a good<br />

fit may do nothing more than describe the obvious.<br />

Any method of fitting models must determine how much each point can be allowed to influence<br />

the fit. Many statistical procedures try to keep the fit close to every data point. If the data include<br />

an outlier, these procedures may permit it to have an undue influence on the fit. As always in<br />

exploratory data analysis, we try to prevent outliers from distorting the analysis. Using medians in<br />

fitting lines to data provide resistance to outliers, <strong>and</strong> thus the line fitting technique [...] is called<br />

the resistant line [2].<br />

2<br />

(1)


Experiment No. 2 Calibration of a Electronic Load Cell <strong>and</strong> a Bourdon Tube-Type Pressure Gage<br />

What we will do is to look at a scatter plot of our residuals. This scatter plot may be able to tell us<br />

what we might try if we do not have a good linear fit. In addition, the residuals measure a signed<br />

distance of the predicted value from the actual value. We get the st<strong>and</strong>ard error by<br />

�<br />

�<br />

� n�<br />

�<br />

� r<br />

� i=1<br />

std.err =<br />

2 i<br />

(3)<br />

n − 2<br />

This a measure of type of average distance from the data to the predicted values.<br />

2.3 Coefficient of Determination<br />

Excel reports the coefficient of determination R 2 . which is the proportion of variability in a data<br />

set that is accounted for by a statistical model. In this definition, the term variability st<strong>and</strong>s for<br />

variance or, equivalently, sum of squares. There are several common <strong>and</strong> equivalent expressions for<br />

R2. The version most common in statistics texts is based on an analysis of variance decomposition<br />

as follows:<br />

In the above definition,<br />

R 2 = SSR<br />

SST<br />

= 1 − SSE<br />

SST<br />

SST = �<br />

(yi − y) 2<br />

i=1<br />

SSR = �<br />

( ˆyi − y) 2<br />

i=1<br />

SSR = �<br />

( ˆyi − ˆyi) 2<br />

i=1<br />

That is, SST is the total sum of squares, SSR is the regression sum of squares, <strong>and</strong> SSE is the sum<br />

of squared errors. In some texts, the abbreviations SSE <strong>and</strong> SSR have the opposite meaning: SSE<br />

st<strong>and</strong>s for the explained sum of squares (which is another name for the regression sum of squares)<br />

<strong>and</strong> SSR st<strong>and</strong>s for the residual sum of squares (another name for the sum of squared errors).<br />

3 Experimental procedure<br />

The Bourdon gage was accompanied by a set of weights labeled in nominal increments of 100 psi<br />

<strong>and</strong> 500 psi. The weights were incrementally loaded on the Bourdon platform <strong>and</strong> the pressure<br />

valve associated with the piston supporting the platform was opened until the piston raised to<br />

mid-stroke. The platform was rotated to resolve any static friction <strong>and</strong> the output pressure values<br />

were recorded. This procedure was repeated until the limit of the scale was reached <strong>and</strong> then the<br />

procedure was reversed until the scale was unloaded. The pressure gage increments were 100 psi.<br />

Output values were interpolated, nominal precision of the instrument should be considered +/-<br />

50 psi. The strain gage load cell supported an aluminum platform approximately 8 inches square.<br />

The tare was adjusted to zero on the digital display system before loading the platform. A series<br />

of weights in nominal increments of 1 pounds were added to the platform. At maximum load, the<br />

procedure was reversed until the scale was unloaded. The output was recorded to a precision of<br />

0.001 mv for each load state however the nominal precision of the output was +/- .005 mv.<br />

3<br />

(4)


Experiment No. 2 Calibration of a Electronic Load Cell <strong>and</strong> a Bourdon Tube-Type Pressure Gage<br />

4 Experimental data<br />

Table 1 <strong>and</strong> table 2 display the experimental data.<br />

Table 1: Stain Gage Calibration Data<br />

Input [lbs] Load [mv] Unload [mv]<br />

0 0.000 0.000<br />

1.5 0.075 0.100<br />

2.5 0.150 0.170<br />

3.5 0.210 0.230<br />

4.5 0.270 0.290<br />

5.5 0.330 0.350<br />

6.5 0.390 0.400<br />

7.5 0.450 0.440<br />

8.5 0.490 0.470<br />

9.5 0.520 0.510<br />

11 0.570 0.560<br />

13 0.640 0.640<br />

Table 2: Bourdon Gage Calibration Data<br />

Input [psi] Load [psi] Unload [psi]<br />

0 0 0<br />

100 125 120<br />

300 555 550<br />

500 990 1000<br />

700 1400 1420<br />

1200 2400 2450<br />

1700 3410 3425<br />

4


Experiment No. 2 Calibration of a Electronic Load Cell <strong>and</strong> a Bourdon Tube-Type Pressure Gage<br />

5 Calculations <strong>and</strong> Results<br />

5.1 Bourdon Gage<br />

The Bourdon Calibration Figure 1 shows the raw data <strong>and</strong> a least-squares linear curve fit to the<br />

data. The Bourdon Calibration first order polynomial shows a first order least squares fit to the<br />

data. The Bourdon Calibration Residuals illustrates the error of the data to the associated analytical<br />

expression. The following analytical calibration were calculated from the data: Loading direction f(x)<br />

= 2.0299x - 36.3082 +/- 50psi Calculated st<strong>and</strong>ard error of data 27 psi with a st<strong>and</strong>ard deviation<br />

of errors 25 psi. Unloading direction, f(x) = 2.0509x - 37.7478 +/- 50psi. Calculated st<strong>and</strong>ard error<br />

of data was found to be 36 psi <strong>and</strong> st<strong>and</strong>ard deviation of errors 35 psi.<br />

Output (psi)<br />

4000<br />

3500<br />

3000<br />

2500<br />

2000<br />

1500<br />

1000<br />

500<br />

0<br />

-500<br />

Bourdon Calibration<br />

0 500 1000 1500 2000<br />

Input (psi)<br />

Figure 1: Bourdon Calibration<br />

5<br />

Load Linear Fit<br />

y = 2.0299x - 36.3802<br />

R² = 0.9996<br />

Unload Linear Fit<br />

y = 2.0509x - 37.7478<br />

R² = 0.9993<br />

Unload (psi)<br />

Load (psi)


Experiment No. 2 Calibration of a Electronic Load Cell <strong>and</strong> a Bourdon Tube-Type Pressure Gage<br />

Output (psi)<br />

50<br />

40<br />

30<br />

20<br />

10<br />

0<br />

-10<br />

-20<br />

-30<br />

-40<br />

-50<br />

-60<br />

Bourdon Calibration Residuals<br />

0 500 1000 1500 2000<br />

Input (psi)<br />

Figure 2: Bourdon Calibration Residuals<br />

6<br />

Unload<br />

(psi)<br />

Load (psi)


Experiment No. 2 Calibration of a Electronic Load Cell <strong>and</strong> a Bourdon Tube-Type Pressure Gage<br />

5.2 Strain Gage<br />

The Strain Gage Calibration Figure 3 shows the raw data. A cursory examination of this graph<br />

reveals a linear character. The Strain Gage Polynomial Curve Fit shows a fist order polynomial<br />

least squares fit to the data to the data points of the set. The Strain Gage Calibration Residuals<br />

illustrates the error of the data to the associated analytical expression. The following analytical<br />

calibration were calculated from the data. Calculated st<strong>and</strong>ard error of loading data; 0.029 mv, its<br />

st<strong>and</strong>ard deviation 0.028 mv<br />

f(x) = 0.0510 × x + 0.028 ± .029<br />

Calculated st<strong>and</strong>ard error of unloading data: 0.030 mv, its st<strong>and</strong>ard deviation .0284 mv.<br />

Output (mv)<br />

0.800<br />

0.700<br />

0.600<br />

0.500<br />

0.400<br />

0.300<br />

0.200<br />

0.100<br />

0.000<br />

f(x) = 0.048 × x + 0.049 ± 0.030<br />

Strain Gage Curve Fit<br />

0 5 10 15<br />

Input (lbs)<br />

Figure 3: Strain Gage Calibration<br />

7<br />

Load Curve Fit<br />

y = 0.051x + 0.028<br />

R² = 0.981<br />

Unload Curve Fit<br />

y = 0.048x + 0.049<br />

R² = 0.978<br />

Load (mv)<br />

Unload (mv)<br />

Linear (Load (mv))<br />

Linear (Unload (mv))


Experiment No. 2 Calibration of a Electronic Load Cell <strong>and</strong> a Bourdon Tube-Type Pressure Gage<br />

Output (mv)<br />

0.050<br />

0.040<br />

0.030<br />

0.020<br />

0.010<br />

0.000<br />

-0.010<br />

-0.020<br />

-0.030<br />

-0.040<br />

-0.050<br />

-0.060<br />

Strain Gage Calibration Residuals<br />

0 2 4 6 8 10 12 14<br />

Input (lbs)<br />

Figure 4: Strain Gage Calibration Residuals<br />

8<br />

Unload (mv)<br />

Load (mv)


Experiment No. 2 Calibration of a Electronic Load Cell <strong>and</strong> a Bourdon Tube-Type Pressure Gage<br />

5.3 Sample Calculation<br />

The sample calculations are implicitly expressed in the calculated tabular data above <strong>and</strong> background<br />

theory discussion. The calculations for Bourdon Gage linear coefficient are show below<br />

Table 3: Curve Fit Calculation for Bourdon Gage<br />

Xi Yi Error load Error 2 Xi × Y i X 2 i<br />

0 0 36 1324 0 0<br />

100 125 -42 1731 12500 10000<br />

300 555 -18 309 166500 90000<br />

500 990 11 131 495000 250000<br />

700 1400 15 239 980000 490000<br />

1200 2400 1 0 2880000 1440000<br />

1700<br />

� �<br />

3410<br />

�<br />

-4<br />

�<br />

20<br />

�<br />

5797000<br />

�<br />

2890000<br />

4500 8880 0 3753.7 10331000 5170000<br />

The intercept can be calculated as follows, using summation values from table 3<br />

a0 = XiXi × Yi − Yi × X 2 i<br />

XiXi − 7X 2 i<br />

= 4500 × 10331000 − 8880 × 5170000<br />

4500 2 − 7 × 5170000<br />

The slope can be calculated as follows, using summation values from table 3<br />

a1 = XiYi − 7Xi × Yi<br />

XiXi − 7X 2 i<br />

= 4500 × 8880 − 7 × 10331000<br />

4500 2 − 7 × 5170000<br />

9<br />

= −36.3802<br />

= 2.0299


Experiment No. 2 Calibration of a Electronic Load Cell <strong>and</strong> a Bourdon Tube-Type Pressure Gage<br />

6 Discussion of results<br />

6.1 Bourdon Gage<br />

The Bourdon data was reported to a precision of +/- 10 psi, however the output gage increment was<br />

100 psi so the nominal precision should be reported at +/- 50 psi. The Bourdon gage calibration<br />

showed high correlation with a linear curve fit to the data throughout the range. The hysteresis in<br />

the unload cycle was in all cases less then the nominal precision of the output gage which means<br />

that one calibration curve for both loading directions would be sufficient for most applications.<br />

Comparatively, a second order polynomial curve fit to the data produced a comparable correlation<br />

coefficient. The plot of the errors associated with the calibration curves show increasing correlation<br />

with higher loading as might be expected from losses associated with friction <strong>and</strong> other mechanical<br />

factors. If the Bourdon gage was going to be used exclusively in the lowers section of the range,<br />

errors are a lot smaller. The calculated st<strong>and</strong>ard error from the linear curve fit was less than the<br />

nominal precision of the gage which reaffirms the suitability of a simple first order polynomial.<br />

6.2 Strain Gage<br />

The strain gage data was reported to a precision of .001 mv, however the fluctuation in the last<br />

digit dictated a nominal precision of +/- .005 mv. The strain gage calibration data was less linear<br />

throughout the range, but not irregular. Visual inspection of the raw data in Figure 3 shows that a<br />

nonlinear fit would have better correlation with even the data points. The st<strong>and</strong>ard error associated<br />

with the linear curve fit was about .027 mv. The lower number of data points in the curve-fit range<br />

was a factor in the relatively high st<strong>and</strong>ard error. The strain gage scale also showed high hysteresis<br />

or bias error in the unload direction. With a device like this, it is important to a specific procedure<br />

to obtain accurate <strong>and</strong> repeatable results. A plot of the errors of the data to the analytical curve<br />

show decreasing correlation with increasing load, which is self-evident in the raw data plot.<br />

7 Conclusions <strong>and</strong> recommendations<br />

A mechanical <strong>and</strong> an electronic scale were evaluated by students to familiarize them with calibration<br />

procedures <strong>and</strong> the analytical methods of least squares curve fitting. The scales were incrementally<br />

loaded with nominal weights <strong>and</strong> the gage outputs were recorded. The procedure was repeated<br />

in the unload direction. Statistical evaluation comparing the data to the analytical results with<br />

consideration for the nominal gage precision was used to determine the most suitable reporting<br />

precision. The Bourdon mechanical scale calibration data showed good correlation with a linear<br />

curve. The electronic strain gage calibration data was more nonlinear but a first order polynomial<br />

was used for a good fit over the entire set of the range. The st<strong>and</strong>ard error was calculated for<br />

both data sets. For both experiments, the st<strong>and</strong>ard error exceeded the output precision (the load<br />

precision being close). Since Bourdon errors were smaller the Strain Gage errors, the first experiment<br />

would be more reliable. This exercise was effective for illustrating the procedures <strong>and</strong> limitations of<br />

calibration methodology as might be applied to a quality control process. The different apparatuses<br />

also contrasted the different character of calibration data from linear throughout the range for the<br />

Bourdon gage, to a more non-linear with for the strain gage. The concepts of bias error <strong>and</strong> precision<br />

error were illustrated by hysteresis between the load <strong>and</strong> unload calibration direction.<br />

10


Experiment No. 2 Bibliography<br />

References<br />

[1] Rahai, H.R et al, 2007, ”MAE 300 <strong>Instrumentation</strong> <strong>and</strong> <strong>Measurement</strong>”, CSULB MAE Department.<br />

[2] Velleman, Paul F., 1984, ”Applications, Basics, <strong>and</strong> Computing of Exploratory Data Analysis ”,<br />

Wadsworth Pub Co.<br />

11

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!