Outline and Equation Sheet for M E 345

Author: John M. Cimbala, Penn State University

Latest revision, 27 March 2009

Introduction
• Primary dimensions – mass, length, time, temperature, current, amount of light, and amount of matter.
• Significant digits – the rules for multiplication and division, and for addition and subtraction.
• Rounding off – round up if the least significant digit is odd, and truncate if the least significant digit is even.

Dimensional Analysis
• Law of dimensional homogeneity – Every additive term in an equation must have the same dimensions.
• The method of repeating variables – there are 6 steps:
1. List the parameters and count them, n.
2. List the primary dimensions of each parameter.
3. Set the reduction, j, as the number of primary dimensions in the problem. Then k = n − j . [Reduce j by
one if necessary.]
4. Choose j repeating variables.
5. Construct the k Πs, and manipulate as necessary.
6. Write out the final dimensionless functional relationship Π1 = function ( Π 2 , Π 3 ,...Π k ) and check your
algebra.
• Dimensional analysis is often extremely useful in setting up and designing experiments.

Errors and Calibration

• Systematic errors (bias errors) – consistent, repeatable errors.
• Random errors (precision errors) – scatter in data, a lack of repeatability, unrepeatable, inconsistent errors.
• Accuracy – accuracy error is the measured value minus the true value.
• Precision – precision error is the reading minus the average of readings.
• Other errors – zero, linearity, sensitivity, resolution, hysteresis, instrument reapeatability, drift.
• Calibration – static (time not relevant) vs. dynamic (time is relevant) calibration.
1 n x − xtrue
• Mean bias error – defined as MBE = ∑ i , and usually expressed as a percentage.
n i =1 xtrue

Basic Statistics
n n

∑d ∑( xi − x)
2 2
n i
1
• Definitions – sample mean x = ∑ xi , sample standard deviation S = i =1
= i =1
, sample
n i =1 n −1 n −1
variance (S2), sample median (half lower, half higher), sample mode (most probable value – one that occurs
most frequently), population (all values) vs. sample (a selected portion of the total population).
• Excel – learn how to use Excel’s built-in statistics functions.
2
1 n ⎛ xi − xtrue ⎞
• Root mean square error – defined as RMSE = ∑⎜
n i =1 ⎝ xtrue ⎠
⎟ , and usually expressed as a percentage.

Histograms and Probability Density Functions

• Histogram plots – frequency, bins or classes, bin width or class width, Sturgis and Rice rules.
• Normalized histograms – how to normalize the vertical scale and the horizontal scale.
• PDF – how to create a probability density function from a histogram.
• Expected value – same as population mean.
• Standard deviation – same as population standard deviation.
x−μ
• Normalized PDF – transform from x to z using z = and f ( z ) = σ f ( x ) .
σ
• The Gaussian or normal PDF – how to predict probabilities for systems with purely random errors using the
⎛ z ⎞
exp ( −ξ 2 )d ξ . See table of A(z).
1 2 ξ =η
error function, A ( z ) = erf ⎜ ⎟ where erf (η ) = ∫
2 ⎝ 2⎠ π ξ =0
 • Other PDFs – lognormal, chi-squared distribution, student’s t, degrees of freedom, confidence level and
significance level.
S
• How to estimate the population mean μ from a sample using the student’s t PDF: μ = x ± tα / 2 with
n
confidence level 1 – α; use table of critical values for the student’s t PDF.
• How to estimate the population standard deviation σ from a sample using the χ2 PDF:
S2 S2
( n − 1) 2 ≤ σ 2 ≤ ( n − 1) 2 with confidence level 1 – α; use table of critical values for the χ2 PDF.
χ α /2 χ 1−α / 2

• The central limit theorem (CLT) – standard error of the mean = standard deviation of sample means =
S σ
S x = overall . As n gets large, we write the CLT in terms of population standard deviations: σ x = .
n n

Correlation and Regression

i=n

∑( x i − x )( yi − y )
• Linear correlation coefficient – rxy = i =1
. By definition, rxy must always lie between
i=n i =n

∑( x − x) ∑( y − y)
2 2
i i
i =1 i =1

−1 and 1, i.e., −1 ≤ rxy ≤ 1 . To determine if there is a trend in the data, use the table of critical values of rxy.
• Regression analysis – least-squares fits (linear, polynomial, and multiple variables), standard error of
estimate. Equations are provided in the lecture notes, but we typically use Excel to do the analysis.
i=n i =n i=n

∑y i
2
− b∑ yi − a ∑ xi yi
Standard error of estimate – also called standard error: S y , x = i =1 i =1 i =1
.
n−2

Outlier Points
• Outliers in a set of data points – use the modified Thompson tau technique: compare maximum absolute
value of the deviation with Thompson τ times sample standard deviation (τ S) to determine outliers:
If δ > τ S, reject the outlier.
i=n

∑( y − Yi )
2
i
• Outliers in a set of data pairs – for polynomial fit of order m, S y , x = , df = n − (m + 1) . i =1

df
Examine the standardized residual ei / Sy,x to determine outliers. There are two criteria for a data pair to be
called an outlier:
o ei / S y , x > 2
o The standardized residual is inconsistent with its neighbors (as judged by a plot of ei / Sy,x vs. x).
Experimental Uncertainty Analysis
• Maximum uncertainty – assumes all errors have the same sign and add up (unlikely). Normally, we do not
use the maximum uncertainty. Typically, we use instead the RSS uncertainty, as described below:
• RSS uncertainty or expected uncertainty – assumes some errors are positive and some are negative (more
2 2
i= N
⎛ ∂R ⎞ wR wR ,RSS i= N
⎛ wxi ⎞
likely). Use wR ,RSS = ∑ ⎜ wxi
i =1 ⎝
⎟ or, if R = C ⋅ x1 ⋅ x2 ... ⋅ xN ,
∂xi ⎠
a1 a2 aN

R
=
R
= ∑ ⎜ ai
i =1 ⎝
⎟ .
xi ⎠
i=K
• Combining elemental uncertainties using RSS uncertainty analysis – use u x = ∑u i =1
i
2
,where K is the

number of elemental uncertainties, and ui are the individual elemental uncertainties.

Experimental Design
• Choosing a test matrix – one parameter at a time (full factorial) vs. Taguchi’s experimental design arrays
(fractional factorial) test matrices. N = L (L = levels, P = parameters) for a full factorial experiment.
P

• Level averages – E.g., average over all the runs where parameter a is at level 1.
• Optimum Taguchi design array – Meets two criteria:
o Each level of each parameter appears the same number of times in the array.
o Repetitions of parameter-level combinations are minimized as much as possible.
• Taguchi orthogonal arrays – know how to spot a poorly designed experimental design array.
• Response surface methodology (RSM) – goal is to efficiently hunt for the optimum values of parameters a, b,
c, etc. such that response y is maximized or minimized.
⎛ a − amid value ⎞
• Coded variables – transform the physical variables into coded variables, x1 = 2 ⎜ ⎟⎟ ,
⎜ a
⎝ range ⎠
a + amax
amid value = min , arange = amax − amin (with similar equations for the other variables). This is necessary to
2
properly use the RSM technique.
⎛ ∂y ∂y ∂y ⎞
• Direction of steepest ascent – vector ⎜ , , ⎟ determined by regression analysis on coded variables.
⎝ ∂x1 ∂x2 ∂x3 ⎠
∂y ∂y ∂y ∂y
For known Δx1, Δx2 = Δx1 and Δx3 = Δx1 .
∂x2 ∂x1 ∂x3 ∂x1
arange
• Marching – Transform back to physical variables, Δa = Δx1 , then march in the direction of steepest
2
ascent until y is no longer improving.
Hypothesis Testing
• Null hypothesis – a theory that is being considered or tested. Typically, the null hypothesis represents
“nothing is happening.”
• There are two parts to the null hypothesis:
o The critical value (extreme value), μo.
o The “sides” or “tails” of the null hypothesis – select the least likely scenario (μ = μ0, μ < μ0, or μ > μ0).
• Alternative hypothesis – also called the research hypothesis – the complement of the null hypothesis.
x − μ0
• t-statistic – calculate for the extreme value (critical value) of the null hypothesis, t = .
S/ n
• p-value – calculate based on the t-statistic; p-value is the probability of wrongly rejecting the null hypothesis,
if it is in fact true. In Excel, use TDIST(t, df, tails) (p-value is the area under the tail(s) of the t-PDF). See
also the tables of p-values for a given value of df for the t-distribution, one tail. Note: The values in the table
are for one-tail hypothesis tests; multiply by 2 for two tails.
• A small p-value means that the null hypothesis is unlikely to be true – we reject the null hypothesis if the p-
value is less than some level, typically 0.05 (5%) to standard engineering confidence level.
• Paired samples hypothesis testing when n is the same for the two samples – use δ = xB – xA as the variable,
set the null hypothesis as μ 0 = μδ = 0 (no difference between the means of xB and xA), and set the value of
δ − μ0
the critical t-statistic to t = .
Sδ / n
• Two sample hypothesis testing when n is not the same for the two samples – set the null hypothesis as
⎡ ⎛ S A2 S B 2 ⎞
2
⎤
⎢ ⎜ + ⎟ ⎥
x A − xB ⎢ ⎝ nA nB ⎠ ⎥
μ A = μ B , set the critical t-statistic to t = , and use df = NINT ⎢
2 2 2 2 ⎥
⎢ 1 ⎛ S A ⎞ + 1 ⎛ SB ⎞ ⎥
2 2
SA S
+ B
nA nB ⎢⎣ n A − 1 ⎜⎝ n A ⎟⎠ nB − 1 ⎜⎝ nB ⎟⎠ ⎥⎦
(Welch’s equation) as a kind of weighted value of the degrees of freedom.
Digital Data Acquisition
• Binary to decimal and vice-versa – how to convert: sum columns or successive division, respectively.
(Vmax − Vmin ) quantization error = ± 1 resolution = ± 1 (Vmax − Vmin )
• A/D systems – resolution = , .
2N 2 2 2N
• Discrete sampling – Δt = 1/ f s , ω = 2π f , watch out for clipping and aliasing.
2
• Ailasing – If f s > 2 f , then there is no aliasing (Nyquist criterion), if f < f s < 2 f , then f a = f s − f .
3
2 ⎛ f ⎞ f
If f s < f , then f a = ⎜ a ⎟ f folding where f folding = s , and need to use the folding diagram.
3 ⎜f ⎟ 2
⎝ folding ⎠
• Signal reconstruction – the Cardinal series: reconstruct a digital signal back into an analog signal.
∞ ∞
• Fourier series – f ( t ) = c0 + ∑ an sin ( nω0 t ) + ∑ bn cos ( nω0 t ) where fundamental frequency is f 0 = 1/ T and
n =1 n =1
T T T
1 2 2
ω0 = 2π f 0 . Coefficients are c0 = ∫ f ( t ) dt , an = ∫ f ( t ) sin ( nω0 t ) dt , and bn = ∫ f ( t ) cos ( nω0 t ) dt .
T0 T0 T0
• Harmonic amplitude plots – plot of amplitude of each coefficient a1 , a2 , etc. versus harmonic number n.
fs N
• Fourier transforms, DFTs and FFTs – discrete version of Fourier series; f max = f folding = = Δf where
2 2
N = total number of discrete data points taken, T = total sampling time, Δt = T / N = time between data
points, f s = 1/ Δt = N / T = sampling frequency, and Δf = 1/ T = frequency resolution.
• Leakage – appears when the discrete data acquisition does not stop at exactly the same phase as it started.
• Excel – how to create a frequency spectrum in Excel.
• Windowing – reduces but does not totally eliminate leakage.
• Analyzing the frequency content of a signal – sample at higher and higher frequencies until no aliasing.
• Anti-aliasing filter – use a low-pass filter to remove high frequencies to avoid aliasing.
Basic Electronics Review
⎛ 1Ω ⎞ ⎛ 1F ⎞ ⎛ 1 H ⋅ A/s ⎞ ⎛1 A ⋅s ⎞
• Unity conversion factors – ⎜ ⎟ =1 , ⎜ ⎟ =1, ⎜ ⎟ = 1 , and ⎜ ⎟ =1 .
⎝ 1 V/A ⎠ ⎝ 1 C/V ⎠ ⎝ 1 V ⎠ ⎝ 1C ⎠
dV dI
• Ohm’s law and basic equations – ΔV = IR , I = C and V = L at any instant in time.
dt dt
R2
• Voltage divider – Vout = Vin where Vout is measured across resistor R2.
R1 + R2
• Resistors in series – Rtotal = R1 + R2 + ... + RN and I = I1 = I 2 = ... = I N .
1
• Resistors in parallel – Rtotal = and I total = I1 + I 2 + ... + I N .
1 1 1
+ + ... +
R1 R2 RN
1
• Capacitors in series – Ctotal = and I = I1 = I 2 = ... = I N .
1 1 1
+ + ... +
C1 C2 CN
• Capacitors in parallel – Ctotal = C1 + C2 + ... + C N and I total = I1 + I 2 + ... + I N .
• Inductors in series – Ltotal = L1 + L2 + ... + LN and I = I1 = I 2 = ... = I N .
1
• Inductors in parallel – Ltotal = and I total = I1 + I 2 + ... + I N .
1 1 1
+ + ... +
L1 L2 LN
Vpeak 1
• Impedance – Z = : resistor Z = R , capacitor Z = , inductor Z = iω L .
I peak iωC
Filters
ωcutoff 1 Vout 1 ⎛ f ⎞
• 1st-order low-pass filter – f cutoff = = , G= = , φ = − arctan ⎜ ⎟.
2π 2π RC Vin ⎛ f ⎞
2
⎝ f cutoff ⎠
1+ ⎜ ⎟
⎝ f cutoff ⎠
ωcutoff 1 Vout 1 ⎛ f ⎞
• 1st-order high-pass filter – f cutoff = = , G= = , φ = arctan ⎜ cutoff ⎟ .
2π 2π RC Vin ⎛ f ⎞
2
⎝ f ⎠
1 + ⎜ cutoff ⎟
⎝ f ⎠
Vout 1
• Higher-order filters – for a Butterworth low-pass filter of order n, G = = .
Vin ⎛ f ⎞
2n

1+ ⎜ ⎟
⎝ f cutoff ⎠

Operational Amplifiers (op-amps)

• Open-loop gain – Vo = g (V p − Vn ) .
• Closed-loop configuration (with feedback loop) – V p ≈ Vn .
R2
• Example circuits – buffer: Vout = Vin , inverting amplifier: Vout = − Vin = GVin , inverter: Vout = −Vin ,
R1
⎛ R ⎞
inverting summer: Vout = − (V1 + V2 ) , noninverting amplifier: Vout = ⎜ 1 + 2 ⎟ Vin = GVin .
⎝ R1 ⎠
• Other circuits – active filters, clipping circuits (high and low voltage clipping), etc.
g
• Common-mode rejection ratio (CMRR) – CMRR = 20 log10 , and GCM is the common-mode gain.
GCM
• Gain-bandwidth product (GBP) – op-amp acts like low-pass filter at high frequencies:
o Noninverting op-amp amplifier: GBP = GBPnoninverting = f c Gtheory, noninverting = constant [GBP supplied by
manufacturer’s specs].
− R2
o Inverting op-amp amplifier: GBPinverting = GBPnoninverting , GBPinverting = f c Gtheory, inverting = constant .
R1 + R2
• Loading – input and output loading can cause voltage change due to internal resistances. Ri = input resistance
or input impedance.
Stress, Strain, and Strain Gages
F δL
• Axial stress and axial strain – σ a = , εa = , σ a = Eε a .
A L
ε δ w δt δD
• Transverse strain & Poisson’s ratio – ν = Poisson's ratio = − t , ε t = = ; εt = for round rods.
εa w t D

•
1
Stress-strain relationship on a surface; principal axes – ε x =
E
( σ x −νσ y ) , ε y = (σ y −νσ x ) .
1
E
ρL δ R δR/R
• Wire resistance – R = , = S ε a , where S is the strain gage factor, defined as S = .
A R εa
R3 R1 − R4 R2
• Wheatstone bridge – Vo = Vs , bridge is balanced when R3 R1 = R4 R2 . Approximate
( R2 + R3 )( R1 + R4 )
Vo R2,initial R3,initial ⎛ δ R1 δ R2 δ R3 δ R4 ⎞
relationship when resistances change: ≈ 2 ⎜
− + − ⎟.
Vs ( R2,initial + R3,initial ) ⎝ R1,initial R2,initial R3,initial R4,initial ⎟⎠
⎜
 4 Vo 1 n
• Quarter, half, and full bridge – ε a ≈ or Vo ≈ ε a SVs , where
n Vs S 4
o n = 1 for a quarter bridge
o n = 2 for a half bridge
o n = 4 for a full bridge
4 (Vo − Vo,reference ) 1 n
• Unbalanced bridge – ε a ≈ or Vo ≈ Vo,reference + ε a SVs .
n Vs S 4

Dynamic System Response

dny d n −1 y d2y dy
• General governing equation (ODE) – an n
+ an −1 n −1
+ ... + a 2 2
+ a1 + a0 y = bx .
dt dt dt dt
dy y − yi −t
y − yf −t

• First-order system – τ + y = Kx , K = b / a0 , τ = a1 / a0 , = 1 − e τ , Γ (t ) = = eτ .
dt y f − yi yi − y f
1 d 2 y 2ς dy a a1
• Second-order system – + + y = Kx , K = b / a0 , ωn = 0 , ς = .
ωn dt
2 2
ωn dt a2 2 a0 a2
⎡ 1 ⎤
o Underdamped, ζ < 1,
y − yi
y f − yi
= 1 − e −ςωnt ⎢
⎢⎣ 1 − ς 2 (
sin ωn t 1 − ς 2 + φ )⎥⎥⎦ , φ = sin (
−1
)
1 − ς 2 . Undamped

ωn 1 a0
natural frequency and ringing frequency: f n = = , and ωd = ωn 1 − ς or f d = f n 1 − ς .
2 2

2π 2π a2
y − yi
o Critically damped, ζ = 1, = 1 − e −ωnt (1 + ωn t ) , and there is no phase shift.
y f − yi
⎡ ⎤
o Overdamped, ζ > 1,
y − yi
y f − yi ⎢⎣
(
= 1 − e −ςωnt ⎢cosh ωn t ς 2 − 1 + )ς
ς −1
2
sinh ωn t ς 2 − 1( )⎥⎥⎦ .
Temperature Measurement
• Mechanical devices – liquid-in-glass thermometers, bimetallic strips, pressure thermometers.
• Thermojunctive devices – thermocouples, sensing junctions, reference junctions, type K, T, etc.
• Thermocouple laws – notation: V1− 2 = V1− R − V2 − R .
o Law of intermediate metals: A third (intermediate) metal wire can be inserted in series with one of the
wires without changing the voltage reading (provided that the two new junctions are at the same
temperature).
o Law of intermediate temperatures: If identical thermocouples measure the temperature difference
between T1 and T2, and the temperature difference between T2 and T3, then the sum of the corresponding
voltages V1–2 + V2–3 must equal the voltage V1–3 generated by an identical thermocouple measuring the
temperature difference between T1 and T3, i.e., V1− 3 = V1− 2 + V2 − 3 for any 3 temperatures T1 , T2 , and T3 .
o Statement of the law of additive voltages: For a given set of 3 thermocouple wires, A, B, and C, all
measuring the same temperature difference T1 − T2, the voltage measured by wires A and C must equal
the sum of the voltage measured by wires A and B and the voltage measured by wires B and C, i.e.,
V1− 2,A&C = V1− 2,A&B + V1− 2,B&C .
• Thermopile – Several thermocouples in series.
• Thermoresistive devices – RTDs (metal, R increases with increasing T), thermistors (semiconductor, R
decreases with increasing T).
1/ 4
W ⎛ε ⎞
Radiative devices – e.g., infrared pyrometer: E = εσ T , σ = 5.669 × 10
−8
•
4
2 4 ,
TH = ⎜ assumed ⎟ Tind .
mK ⎝ ε actual ⎠
• Tables – how to use thermocouple tables and RTD tables, and interpolation. Reference temperature is 0oC for
all temperature tables.
Measurement of Mechanical Quantities
• Position – mechanical devices; interferometer; potentiometer; linear variable displacement transducer
aΔt
(LVDT); ultrasonic transducer: pulse-echo x = , through transmission x = aΔt ; capacitance sensor:
2
C2
C = K ε 0 A / d , ε 0 = 8.854 × 10 −12 ; laser displacement meter.
N ⋅ m2
⎛ rotation ⎞ ⎛ radian ⎞⎛ rotation ⎞⎛ 60 s ⎞
• Angular velocity and rpm – ⎜ N rpm ⎟ = ⎜ω ⎟⎜ ⎟⎜ ⎟ ; contacting tachometer,
⎝ min ⎠ ⎝ s ⎠⎝ 2π radian ⎠⎝ min ⎠
P
noncontacting tachometer: N rpm = . For stroboscopic tachometers, watch out for aliasing.
nteeth
2π
• Torque and power – Wshaft = ωT = N rpmT ; dynamometers: prony brake, cradled DC motor, eddy current.
60

Pressure Measurement
• Types of pressure – absolute, gage: Pgage = Pabs − Patm , vacuum: Pvac = Patm − Pabs , Pvac = − Pgage .

• Liquid manometers – Pbelow = Pabove + ρ g Δz ; McCleod gage used to measure very low pressures.
• Mechanical pressure gages – Bourdon tube, deadweight tester.
• Electronic pressure transducers – diaphragm type, with displacement measured by:
o Strain gage – a strain gage is mounted on the diaphragm itself, sensing strain in the diaphragm.
o Capacitance – the diaphragm is mounted close to a fixed parallel plate, capacitance is measured.
o LVDT – diaphragm is attached to the core of a linear variable displacement transducer.
o Optical – various optical techniques used to measure the degree of diaphragm deformation.
• Piezoelectric pressure transducers – compression causes a voltage that can be calibrated with pressure.
Velocity Measurement
• Linear velocity transducer (LVT) – Similar in principle to LVDT, but measure velocity, not displacement.
2V cos θ
• Doppler radar velocimeter – Doppler frequency shift Δf D = .
λ
• Displacement sensors – V (t ) = dx(t ) / dt , noise is amplified by differentiation.
Acceleration sensors – V (t ) = V0 + ∫ a (t ) dt , noise is attenuated by integration.
t
•
t0

• Velocity of fluids – Lagrangian methods follow fluid particle moving with the flow: V (t ) = dx(t ) / dt ;
Eulerian methods measure velocity field with a probe sitting in the flow.
fλ
• Laser Doppler velocimeter (LDV) – particles in fringe pattern of focal volume, V = fs = .
2sin(α / 2)
• Hot-wire and hot-film anemometer – wire heated to constant temperature; King’s law E 2 = a + bV n .
2 ( P1 − P2 )
• Pitot-static probe – Pitot formula V = .
ρ
• Other – rotating velocimeter (cup anemometer, turbine or vane anemometer), electromagnetic velocimeter.
Volume Flow Rate Measurement
• Mass flow rate vs. volume flow rate – m = ρV . Notation: V = volume, V = velocity, V = volume flow rate.
• End-line measurement – V = V / Δt vs. in-line measurement – various types:
2 ( P1 − P2 )
• Obstruction – orifice, flow nozzle, and Venturi: β = d / D , V = Ao Cd , Cd = discharge coeff.
ρ (1 − β 4 )
• Positive displacement – measure how many “trapped” or “captured” volumes pass through per unit time.
• Turbine and paddlewheel – spin a turbine or paddlewheel connected to a shaft, measure rpm and calibrate.
• Roatmeters or floatmeters or variable-area flowmeters – float hovers when forces balance.
• Miscellaneous flowmeters – ultrasonic (transit time and Doppler shift), electromagnetic, vortex, others.