ME 3901 Engineering Experimentation

Mechanical Engineering Department

Worcester Polytechnic Institute

Error Analysis


·         Types of Error

·         Least Square Error Analysis

·         Uncertainty Analysis for Density

·         Hoop Stress Error Analysis

 


 

ME3901 is composed of several experiments

Need to learn:

How to measure

What tools/meters to use

The consequences of measurement errors and uncertainty

 

Data can occur as:

Single measured data – Some uncertainties (meter dependent) may not be discovered by repetition

Ex:  Measure fluid T with a thermometer many times.  These repeated measurements address the uncertainty in your ability to read the thermometer.  But no information is available on the meter itself.

 

Multi-sampled data – Measure T with several meters/devices

Thermometers

Thermopiles

Thermocouples T/Cs

Thermistors

 

Three types of error that can cause uncertainty in experimental measurements.

1.)           Gross Blunders:  Construction, Application, Staffing flaws

2.)           Fixed Errors:  Systematic errors, Bias errors – repeated readings to be in error by approximately the same amount, but for some unknown reason(s). Frequently, the systematic error is quantified as the (Average of the measured readings - True Value). (Unfortunately, the True Value if frequently unknown).

3.)           Random Errors:  Fluctuations – electronic, personnel, random phenomena frequently statistically distributed; at times difficult to separate from fixed errors. Frequently, the random error is quantified as the (Individual measured reading - Average of all measured readings).

 

If a measurement has small systematic errors it is Accurate.

 

If a measurement has small random errors it is Precise.

 

Frequently a systematic error can be mitigated and/or estimated, i.e. calibrate to remove systematic error

 

How to report or ‘analyze’ errors?

Common Sense

    (variable – who’s common sense!)

Largest error

Worse case error (Sum of all errors)

 

It is equally inaccurate to over-estimate as it is to under-estimate uncertainty.

 

A Least Squares Error Analysis

-        Most robust

-        Most widely accepted

 

Consider

P = 100 kPa

P = 100 kPa +/- 1 kPa

P = 100 kPa +/- 1 kPa (20 to 1)

(This latter statement requires total lab experience to put odds on uncertainty)

 

Most Results (R) are derived from multiple measurements, each of which has uncertainty

 

R = R(x1, x2, x3, … xN)

where xi are Independent variables

 

 

Want the uncertainty of the result wr

 In terms of the uncertainty of the independent measurements, wi

 

That is,

 

   

 

A Least Squares Uncertainty Estimate

 

Consider the problem – measure the density, r, of a fluid.

 

 

   

 

 

 

Are these measurements really independent?

Balance scale  f(ruler)

But l, d, and h ARE dependent

 

 

 

 

 

 

 

What is the uncertainty in V? 

Let the ruler be ‘shorter’ than actual

then the readings of l, d, and h will all be too large by approximately the same amount

 

i.e. a Systematic Error exists

 

 

 

 

 

 

 

                                        ( Much Larger Term!)

 

Independent Errors:

        Sum ( the squares of the error contributions)

 

Dependent Errors:

Square (the sum of the errors) ß  Much Larger

 

 

Dependent/Independent Error Conclusions

Dependent Errors cause greater total errors compared to Independent Errors.

 


Thus, if one had 3 measuring devices (different brand rulers) with similar uncertainties it would be significantly better to measure the l, d, and h with independent measuring devices!

 

This conclusion is contrary to most peoples’ intuitive experience.

 

Hoop Stress Error Analysis

Consider a measurement of hoop stress from a thin-walled pressure vessel.

 

 

 

 

 

 

Note:

As with all least square analyses, a large uncertainty will dominate multiple small uncertainties. 

 

If a single uncertainty is 5-10 times other uncertainties, then stating the resultant uncertainty as the largest value is reasonable.

 


Go to the: [next lecture | chapter top | previous lecture | cover ]