Anda di halaman 1dari 15

Chapter 1: Introduction to Modeling

A. The Scientific Method


The heart of sound engineering practice is
the scientific method:
systematic approach for problem solving
constant interplay between the concrete
and abstract
Concrete: Actual engineering processes
Abstract: Mathematical models
Example:
Uniformity of layers of silicon grown on the
wafers.

Concrete
Non-Uniform
Wafers

Abstract

@
@
@
@
R

Clearly Dene
Problem

Comment

What is Non-Uniform?
How Do We Measure?

Postulate
Important Factors
?

Formulate
Working Model
?

Conduct
Experiment


@
@
@
@
@
R

Deposition Time (t)


Deposition Temp. (T )
Argon Flow Rate (A)
y = f (t, T, A)
A First Approximation:
y = 0 + t t + T T + A A

Plan
Experiment

Develop Systematic Strategy


for Changing t, T , and A;
See Eect on Uniformity

Estimate
Model

Use Data to Quantify Eects

Determine
Important Factors

Some Factors Are Important;


Others Are Not

Revise
Working Model

Drop Unimportant Factors;


Possibly Add New Factors

Plan Conrmatory
Experiment
Conduct
Conrmatory
Experiment

Repeat as Necessary
2

B. Probabilistic Models
The scientic method requires:
data collection
data analysis
Problem: Data always exhibit variability.
Variability obscures our ability to make decisions.
The scientic method has always used abstract models to address concrete problems.
For example, consider Ohms Law
V = IR
where
V is the voltage,
I is the current,
R is the resistance.
Note: This is a deterministic model.
If we know I and R, we claim to know V
exactly.
3

Consider an EE lab with 20 students


All set up circuits
same current
same resistance
All measure the voltage.
How many dierent voltages do you think
will be observed?
A better approach uses a probabilistic model
V = IR +
where is a random error.
In this case, is best viewed as an experimental error.

C. Regression Models
Consider the thickness of the silicon layer
on a wafer for an integrated circuit.
Let yi be the thickness of the silicon layer
on the ith wafer.
A deterministic model is
y = 0 + tt + T T + AA
where
the s are constants,
t is the deposition time,
T is the deposition temp, and
A is the Argon ow rate.

A better model is
yi = 0 + tt + T T + AA + i.
where i is the random error associated with
the ith wafer.
We note that this model is an example of a
first-order Taylor Series approximation!
The scientic basis for many of our regression models is Taylor Series.
George Box: All models are wrong; some
are useful.
The simple linear model:
yi = 0 + 1xi + i
The multiple linear regression model:
yi = 0 + 1x1i + 2x2i + . . . + k xki + i
In general, k represents the number of regressors (independent variables).

Other linear models:


yi = 0 + 1xi + 11x2i + i
Note: we can let x1i = xi, x2i = x2i , and let
2 = 11.
We then can rewrite the model as
yi = 0 + 1x1i + 2x2i + i
A linear model means linear in the parameters (the s).
log(yi) = 0 + 1xi + 11x2i + i
1
yi = 0 + 1 + i
xi

D. Issues in Data Collection


The Scientic Method requires data.
Important issues in data collection:
purpose of the study
characteristic of interest
measurement of the characteristic of interest
presumed engineering model
parameters of interest
physical constraints
Example: Battery Plate Process

Problem: Plates with blisters causes premature failure of battery cells.


purpose: eliminate blisters on plates
characteristic of interest: number of plates
showing blisters in a test cell of 60 plates
initial model: y = 0 + 1x1 + 2x2 + 3x3 +
where
y: the number of blisters
x1: plate porosity
x2: concentration of an important electrolyte
x3: load applied to the plates during
charging
s: constants relating the eect of
the xs on the number of blisters
: random error
parameters of interest: the s in the
model
constraints: can test only four cells at a
time.
Note: Constraints require a modication
in the model!
9

E. Methods for Collecting Data


Three basic methods of collecting data are:
retrospective studies
observational studies
designed experiments
Retrospective studies use previously collected data.
Primary advantage:
cost since the data are already collected
Disadvantages:
1. may not have the information required
data collected are easy to do so
data were required for some problem
a long time ago
2. poor data quality
3. often have problems with missing data
4. typically document interesting phenomena.
10

Observational studies interact with the process only as much as is required to obtain
relevant data.
Usually we use observational studies to monitor processes.
Observational studies usually use a sampling plan to collect their data.
Some common sampling plans are:
simple random sampling
stratied random sampling
systematic sampling
Of these three, the most common is simple
random sampling.

11

Designed experiments intentionally disturb


the process and then observe the results.
Example
Consider the yield (concentration of desired
product) from a distillation column.
We call the yield our response.
Factors which may inuence the yield:
reboil temperature
column pressure
ow rate into the column
In a designed experiment, we manipulate
the factors, allow the system to reach equilibrium, and then we observe the response.

12

Suppose we wish to conduct an experiment


using a high and low level for each factor.
For example: Reboil temp. 120 150
pressure
2 3 atm.
ow rate
100 150 GPM
What would seem a reasonable strategy?
Use all possible combinations.
Reboil
Temp. Pres. Flow Rate
120
2
100
150
2
100
120
3
100
150
3
100
120
2
150
150
2
150
120
3
150
150
3
150

13

T
C

This is called a 23 factorial experiment.

14

Common reasons for conducting designed


experiments are:
to screen important factors
to predict the behavior of an important
characteristic
to optimize a characteristic of interest
to make products and processes robust
to known sources of variability.

15

Anda mungkin juga menyukai