Title: DESIGN OF EXPERIMENT
1DESIGN OF EXPERIMENT
INTRODUCTION
TO
TOOLS AND TECHNIQUES
TRAINING PROGRAMS DEVELOPED FOR DURACELL
CHINA BY SCENTIA INTERNATIONAL INC.
2DESIGN OF EXPERIMENT
(3) Day Session
SECTION ONE Introduction to Design of
Experiment Quality
- Purposes and Practices of DOE as linked to the
use of quality tools
- Define how DOE is currently applied in
manufacturing operations
- Identify key resources necessary to conduct
successful DOE Programs
- Learn basic advantages and limitations of DOE
process
- Identify specific goals for use of DOE in
manufacturing processes
SECTION TWO Key Elements for Successful
Application of DOE
- Introduction to Experimentation
- Selection of criteria for design of DOE projects
- Learn how to choose sampling data collection
alternatives
- Learn useful statistical tests for variables
factors that effect DOE
Conduct a step by
step, in-class
experiment
Taguchi
loss and analysis to DOE applications
SECTION THREE Select and design an in-house
DOE project
- Determine baseline criteria for your DOE project
- Define the experimental levels for your
in-house DOE project
- Generate present a timeline and resource list
for your DOE project
- Define specific goals for your selected DOE
project
(2) Day Session
SECTION FOUR Review in-house DOE and advanced
topics
- Team summary review of experiment design,
application, and results
- Analysis of common roadblocks experienced in
DOE design and application
- Learn modeling methods for DOE and using DOE
in problem solving
SECTION FIVE Learn how to improve
production operations
- Learn to direct product and process design
using DOE
- Learn how to improve management of production
operation product design
- Show how DOE can reduce variance in your
manufacturing operations system
- Evaluate accuracy of the DOE results in
seeking solutions to process problems
3Purposes and practices of Design of Experiments
- A Method to
- describe
- predict
- and control
- variables in a process in order to understand and
improve the process or product. - In order to improve a process or system one must
first understand it. If it is not understood,
then any change made to it is tampering. The
effect then is almost always detrimental to the
quality of the process and product.
4Purposes of DOE
- To gain knowledge about the effects of variables
(factors) in a process - Identify primary and secondary variables
- Determine optimum levels of variables
- Determine interactions of variables
5Purposes of DOE
- Identify key factors to design (or re-design) a
process or product for better quality/efficiency. - Identify critical design parameters (1.2)
- Provides a base to monitor measurements and tests
- Can help in Failure Modes and Effects Analysis
(FMEA)
6Practices of DOE
- Uses statistical principles to obtain accurate
results - Yields more information than one-at-a-time
experiments - Statistically valid design gives statistically
valid results - Uses confidence testing to determine significance
of effects
7Practices of DOE
- Leads to good engineering practice such as
- Highly structured and thoughtful design and
implementation encourages good design practice. - Provides a base to mathematically model a process
using response surface methodology. (session 2) - Promotes systematic and logical testing that
validates and verifies intuition and experience. - Encourages good decisions through documentation
and data analysis.
8Practices of DOE
- DOE builds on basic quality tools
- Depends on an in-control process
- Control charts
- Histograms
- Flow Charts
- Identification of factors
- Cause and Effect diagrams
- Pareto charts and Check sheets
- Monitoring of changes made by DOE
- Control Charts
- Scatter plots
9Purposes and practices of Design of Experiments
- Summary
- DOE provides an mechanism for active change in
the improvement process. - A very effective high level tool that can be used
in detail operations, sequential processes or
overall systems.
10DOE in Manufacturing Injection Molding Example
11Tensile test Pareto Chart
- Effects of Molding Variables on Tensile Strength
12Interactions plot
- What are the interactions that would effect the
process? Example of fiber length and fiber
concentration
13Interactions Plot
- What are the interactions that would effect the
process? Example of temperature and mold
thickness.
14Interactions plot - 3-way example
- What are the interactions that would effect the
process? Example of all three factors.
15DOE in Manufacturing
- Composites injection molding example
- Better able to manage the process for cost,
time and quality - Better able to predict the effects of adjustments
of the process - Can better meet customer requirements
16DOE in manufacturing
- The process in now easier to manage in the system
for - Production amounts
- Production times
- Inventory/material required
- Relationship to other operations
17Identify resources necessary for successful DOE
- Remember DOE is a prevention cost activity
- More difficult to measure savings
- Training and planning are critical and on-going
18Identify DOE resource needs
- Management support necessary
- Time and money for training
- Need management perspective to see impact on the
whole system - Downtime, if necessary of process
- Material/product costs (25-30 on experiment 1)
- Knowledge and expertise
- Statisticians (if any) can be of help
- Experts involved in, or related to, the process
being tested
19Identify DOE resource needs
- A method to help determine time and material
resource needs - What product/process will be targeted and why?
- How much time/material will be necessary to
conduct experiment 1? (then multiply by four) - What will be the effect on production?
- When will be the best time to conduct the
experiment - Production schedule personnel
- External conditions that may affect results
- How much time to analyze and document the
experiment
20Advantages and limitations of DOE
- Advantages
- Statistical foundation yields a lot of
information for relatively low cost - Provides main, secondary and interaction effects
of factors being tested - Basic designs can be conducted and evaluated
without significant statistical knowledge or
expertise - Much more information than obtained from
one-at-a-time experimentation - Pro-active tool for directing improvements
21Advantages and limitations of DOE
- Limitations
- Not valid if process is not in-control or
statistically stable - Requires more analytical skill than basic quality
tools - Sometimes is used on processes when simpler tools
would suffice - Care must be take not to confuse results, for
example - Maintain homogeneity in material
- External conditions should be as stable as
possible - Requires good planning and documentation
22Identifying specific goals for DOE
- Ask Why am I using this tool on this problem?
- This will help identify the desired response
variable (the goal) for the experiment and ensure
DOE should be used. - Ask How will improving the desired response
variable improve the process? - Validates the need for conducting the experiment
23Identifying specific goals for DOE
- Ask How will improving the targeted process
affect the system? - Protects against sub-optimization of a process at
the expense of the system - Ask Can I maintain constant external
conditions? - Helps to prevent introduction of special causes
that would corrupt the analysis - If the external conditions cannot be maintained
constant then they can be planned for
24Identifying specific goals for DOE
- Ask What are the main factors to be considered
in this experiment? - Using a Cause and effect diagram will help answer
this question - All who are part of the process (management to
line people) should be involved - Ask Do I have baseline information for a
benchmark? - If no baseline has been obtained this should be
the first goal of data collection
25End Section One
26Main Tab Page for Section Two
27Introduction to the experimental method
28Introduction to the experimental method
The Factorial Pattern Row A B C D 1 2 -
__________k1 3 - 4 - - ________
__k2 5 - 6 - - 7 - - 8 - - -
__________k3 9 - 10 - - 11 - -
12 - - - 13 - - 14 - - - 15 - - - 16
- - - -__________k4
29Introduction to the experimental method
- How the factors and interactions are tested
(refer to the factorial pattern on the previous
page) - Factor A against desired response in rows 1 2
(also rows 3 4) - Rows 1 8 together show effect of C on effect of
B and effect of A (3-way) - Rows 1-4 show the change of A at high B and at
low B (2-way) - Software does all this automatically
30Introduction to the experimental method
- Summary
- Full factorials cover full design space
- Full factorials are easy to lay out - repeating
pattern in standard order - Multiple factors are tested and interactions
31Design of the DOE
- Cause and effect diagram (right side)
- Review section 1 item 5
- Determine how to measure the response variable
32Design of the DOE
- Screening designs (more later)
- Cause and effect diagram (left side)
- Historical data as baseline
33Design of the DOE
- Blocking
- Planning for known variation (noise) that may be
inherent in the experimental sample. For example - two replications from same lot
- 1/2 of design in lot A and 1/2 in lot B
- Randomization
- To avoid time trends and other variation not
known - Never run in sequence order
- Avoids the effects of hidden variable
- Validates statistical conclusions
- Use tables or numbers from hat
34Design of the DOE
- Replication
- Measure experimental variability
- Improve estimate of effects
- Determine if change in factor levels is special
(induced) or common cause variation
- A replication of a run is an independent and
random application of the run including set up. - A repeat is application of a run without a new
setup.
35Data Collection
- Strive for homogenous material
- Same lots etc.
- Reduce effects of time through randomization or
new setup (example chemical bath degradation) - Use blocking and randomization to minimize
external effects
36Data Collection
- Record keeping
- Emphasize integrity and accuracy of documentation
- Resisting estimating (let the data perform its
function) - Note special conditions in the data collection
sheet - Always include as much information as possible -
name, time, date, location, operator, shift, etc.
37Results will be no better than the quality of
data you obtain and record. Complete accuracy
and integrity in the data is critical
Data Collection
38Statistical tests for DOE variables and factors
- Calculating effects
- A effect is the difference in the averages
- Main effects
- E(A)?A - ?A-
- Interaction effects
- E(AB)1/2(?A - ?A-)B - (?A - ?A-)B-
39Statistical tests for DOE variables and factors
- Standard deviation of experiment runs
- Se?(?Si2/k)
- Standard deviation of effects
- SeffSe ?(4/N)
- where N in the total number of trials
- T-statistic
- using t-tables (appendix) and degrees of freedom
where - degrees of freedom
- d.f.( of observations per run-1) X ( of runs)
- See example with in class experiment
40Statistical tests for DOE variables and factors
- Charting the results
- Pareto charts of results with significance limits
41Statistical tests for DOE variables and factors
- Interaction plots
- parallel lines mean no interaction
- normal lines mean strong interaction
42Statistical tests for DOE variables and factors
- Cube Plots
- Shows the effects at planes and corners of the
factorial
43In class experiment
- Determine the effects of
- Heat treatment
- Vendor
- Size
- on the durability of paper clips.
- Will use 3 factors at 2 levels as defined by the
experiment sheet
44In class experiment
- Groups of four people with the following tasks
assigned - Test engineer _________________
- Runs the test (bends the clips)
- Analyst ______________________
- Leads the team in calculating results
- Recorder_____________________
- Ensures all data is properly recorded
- Lead engineer_________________
- Ensures runs and treatments are performed in
proper order and form
45In class experiment
46In class Experiment
47In class experimentTime order plot of data
48In class experimentSummary Table for
experimental results
49In class experiment
50In class experiment
- Calculate effects
- Calculate t-ratios
- Construct Pareto Chart of effects
- Construct Cube plot
- Construct interaction charts
- See appendix for chart blanks
51End Section Two
52Main Tab Page for Section Three
53Review of Taguchi loss and DOE applications
- Deviation from target is loss
- Loss can be calculate for point or sample
- Loss at a point is L(X) k(x-m)2
- Loss of the sample set is Lks2 (in per unit)
- kA/ª2 (A is the cost of failure at the spec
limit, ª2squared distance from target to failure
value) - xmeasured value
- mtarget value
- sstandard deviation of the sample
54Review of Taguchi loss and DOE applications
- DOE builds a model to identify both product and
process target values - Robust design (product or process) will include
DOE experiments. - Testing existing target values on products or
process using DOE and historical data
55Group design of a DOE
- Flight time of a paper helicopter
- Customers have been complaining about the short
flight times of PHL (Paper Helicopter Ltd.)
helicopters. Management want your group to
redesign the helicopters to increase flight times.
56Group design of a DOEHelicopter standard design
57Group design of a DOE
- Factors that may affect flight time
- Approved possible modifications
- Factor Standard Allowed change
- Paper type Copier Heavy colored
- Paper clip none yes
- Taped body no yes (3 inches)
- Taped wing joint no yes
- Body width 75mm 100mm
- Wing length 75mm 100mm
- Each team must select only three factors
58Group design of a DOE
- Responsibilities and duties
- Lead engineer________________
- Final decision on prototype
- Test Engineer________________
- Final say on conducting tests
- Assembly Engineer____________
- Final say on building issues
- Recorder____________________
- Leads data collection and analysis
59Group design of a DOE
60Group design of a DOE Time order plot of data
61Group design of a DOE Summary Table for
experimental results
62Group design of a DOE
63Your own DOE project
- Determine baseline for the project
- Is the process in-control?
- If not then special causes must be eliminated
before proceeding - Do you have historical data available?
- If not then determine data collection method and
collect data - If the process is in-control then proceed. If not
see step 1.
64Determine factors and levels for your DOE
- Use a cause and effect diagram with others in
involved to determine factors to be tested - Use only three factors
65Determine factors and levels for your DOE
- To determine levels
- To be done with others involved with, expert in,
or affected by the process. - What is the current baseline level?
- What levels would those involved suggest be the
edges of the operating range (or just ones all
want to test) - Document the levels and the reason for the
selected level
66Generate Time line
- Using a Gantt chart or project timeline determine
the following task milestone dates - Finalize project design
- Determine necessary resources
- Conduct experimental runs
- Analyze data
- Present to group and instructor (July 26th)
67Generate DOE resource list
- A method
- What product/process will be targeted and why?
- How much time/material will be necessary to
conduct experiment 1? (then multiply by four) - What will be the effect on production?
- When will be the best time to conduct the
experiment - Production schedule personnel
- External conditions that may affect results
- How much time to analyze and document the
experiment
68Identifying goals for your DOE
- Why am I using this tool on this problem?
- How will improving the desired response variable
improve the process? - How will improving the targeted process affect
the system?
69Identifying goals for your DOE
- Can I maintain constant external conditions?
- What are the main factors to be considered in
this experiment? - Should have been completed already
- Do I have baseline information for a
benchmark? - Should have been completed already
70Good luck and see you in July
- If you have questions feel free to email
Professor Hawks at - hawksv_at_byu.edu
71End Section Three