Professional Documents
Culture Documents
Lecture-13
So far.
So far.
Risk assessment
Paradigm shift
Accident Investigation as problem solving
Exxon Valdez
Common contributors to accidents
Problem solving process, 5 steps
Heuristic Approach
Process Chart
So far.
So far.
Petri Nets
Decision Trees, Binary Decision Trees
FTA
ETA
Solution implementation
Approval
Planning
So far.
Implementation
Planning
MS Project , Critical Path
Approval
Resources
MOC
Onion Model
Follow Up
So far.
Error analysis
Experimental data analysis
Error types
Examine the
need for the
experiment
Define
objectives for
the experiment
Choose
responses to
measure
Identify the
important
variables
Design the
experiment
Basic Building
Blocks for
Experimental
Design
IMPORTANT: Error Analysis
Perform
experiment
Analyze
results
Act upon
results
Report !
ERROR ANALYSIS
PRESENTATION
OF DATA
REPORT
Precision
Uncertainty
Statistical Analysis
Error Distribution
Gaussian distribution
t-distribution
Chi-square distribution
Population
theory
Sample
theory
Some Definitions
Metrology
The science (and art) of measurement
Accuracy
The difference between what you measure and
the true or standard value
Precision
Difference the difference or range in repeated
measurement of the same thing. Also the
number of digits you reasonably report.
Some Definitions
Resolution
Smallest increment of change you can determine;
sometimes called least count.
Resolution is is usually less than or equal to
Precision.
Sensitivity
Change in output per unit change in input.
Range
The interval of the smallest and largest inputs
(measurands) the system can handle without
distortion
Many times range and sensitivity are at odds.
Types of Errors
Bias error (systematic error) are those that occur
the same way each time a measurement is
made. Bias errors do not show a distribution,
therefore no statistical technique may be used.
Calibration errors
Consistent human errors
Defective equipment
Loading errors (effective of sensor on quantity
measured)
Total error
Bias error
Precision
error
Truth
Measured value
Error Analysis
Types of error that should be avoided in
experimental design: Type I and II error
TYPE I: Declare that something has an
effect on the outcome, when really it does
not (False Alarms)
TYPE II: When we fail to discover a real
effect on the outcome (Missed Data)
Schematically
x1 x2 x3 xn
INPUTS
controllable factors
PROCESS to be
Understood
OUTPUT, y
uncontrollable factors
z 1 z2 z3 z m
Schematically
x1 x2 x3 xn
controllable factors
PROCESS to be
Understood
OUTPUT, y
uncontrollable factors
z 1 z2 z3 z m
Schematically
x1 x2 x3 xn
controllable factors
PROCESS to be
Understood
OUTPUT, y
uncontrollable factors
z1 z2 z3 zm how to minimize these?
Designing an Experiment
A well designed experiment can result in
an experimentally determined model
(Empirical Model)
Good data (and reporting) stands the test
of time.
Galileos drawings of the moons of Jupiter
Mayan observations of the sun
Data Collection
Good data collection is essential
Why it is worth investing in a lab manual,
and recording all observations.
Data Collection
Why it is valuable to include raw data in a
journal report, since it allows others the
opportunity to evaluate the measurements
may come to different conclusions
may notice a problem that you didnt
Data Collection
But before we rush off to collect data we
need to adopt a strategy to follow in order
to collect it
Want a
Strategy of Experimentation
recognition/statement of problem
choosing response variable
choosing factors, levels, ranges
choose experimental design
do experiment
statistical analysis of data
conclusions/recommendations
INPUTS
controllable factors
PROCESS to be
Understood
OUTPUT, y
uncontrollable factors
z 1 z2 z3 z m
Inputs/Outputs
What are my inputs?
How do we measure the output (y)?
How do we assess improvement in output
(statistics play a big part here)?
Sample Problem
Goal: Maximize production of C
Reaction:
A+B
catalyst
Sample Problem
Factors:
amount of catalyst
T of reaction
stirring/no stirring (degree of agitation)
operator
supplier of A and B
A Method to Solve
try a set of variables measure y
use intuition to change a variable to see
change in measured response
C (g)
Catalyst
Temperature
Agitation
Golf Example
Factors identified
type of driver (regular or oversized)
type of ball
walking or riding
drinking water or beer
play AM or PM
play cool or hot
golf shoe spikes (metal or soft)
windy or calm
Interaction Effect
Set of Scores
Determining Effects
Beverage
Ball
Driver
Improved Approach
Factorial Design
Some Terms
Run
Variable/Factor
Randomization
Replication
Blocking
Run
A measurement from one set of
experimental parameters
Variable/Factor
An experimental value that can be
changed.
Often want a range for the variable:
minimum and maximum temperature
length of experiment (time)
amount of additive
etc.
Randomization
allocation of experimental material and
order of runs is random
ensure that each experimental run is an
independent observation
in other words no carry over effect
between experiments
useful for averaging out effects of
extraneous factors (those zs)
Replication
independent repeat of a run
allows for estimation of experimental error
difference between a repeated
measurement
Blocking
design technique which can improve
precision
used to reduce/eliminate nuisance factors
(something which may influence result but
is not of interest; e.g., two batches of
materials supplied for testing)
Zr/Hf crucibles
Blocking
This is a area requiring more depth
Sample Mean
Sample Variance
n-1
Regression Model
Linear Model
y = 0 + 1 x +
Quadratic Model
y = 0 + 1 x + 2 x 2 +
Factorial Design
Terms
main effects (x1, x2, x3, )
interaction effects (x1x2, x1x3, x2x3, )
coded scale (-1 to +1)
Terms
regression model representation for a twofactor factorial design
y = 0 + 1x1 + 2x2 + 12x1x2 +
response surface
Relative Efficiency