Transcript Sundar Krishnamurty - Decision Based Design Open Workshop

On the Role of Decision Analysis
in Engineering Design
Sundar Krishnamurty
Department of Mechanical and Industrial Engineering
University of Massachusetts-Amherst
Email: [email protected]
2000 DBD Workshop
DBD Workshop, September 2000
Sundar Krishnamurty, Umass-Amherst
Should all design activities be
constructed under a decision theory
framework?
DBD Workshop, September 2000
Sundar Krishnamurty, Umass-Amherst
42%
25
%
17
%
0
%
A
B
50%
C
33
%
A
D
17
%
E
17
%
B
0
%
D
C
0
%
E
80%
20
%
A
17
%
A
B
33
%
B
0
%
C
0
%
D
0
%
E
17
%
17
%
17
%
C
D
E
57%
0
%
A
14
%
14
%
B
C
14
%
D
E
1. The fundamental element of engineering design is decision making. All design
activities, including those in concept generation, need to be constructed under the
decision theory framework. Therefore decision analysis is the most important
aspect of design.
2. Design involves both art and science. While the majority of activities are desired
to be logical, some activities are intuitive such as brainstorming for the generation
of design alternatives which heavily relies on creativity. Decision analysis can
support design activities that are logical, however, other theories and techniques
need to be developed to enhance intuitive activities.
3. There are benefits and limitations of decision analysis throughout the design
process. Based on the design situation, a designer need to decide how to use
decision analysis in engineering design appropriately.
4. There are activities in the design process that are decision activities but do not
require the same type of decision-making as that used for selecting a final design -which necessarily involves values, uncertainty, and risk assessment. Decision
analysis could become too sophisticated or inapplicable for those activities.
5. The quantitative decision analysis will leave out many human and artistic
qualities that engineering design possesses. The use of decision analysis in
engineering
DBD Workshop, September 2000
Sundar Krishnamurty, Umass-Amherst
Normative Decision Analysis in
Decision-Making (Dr. Howard)
Ingenuity
Environment
Uncertain
Complex
Dynamic
Competitivee
Finite
Confusion, worry
Choice
Normative Decision Analysis
Alternatives
Perception Information Probability - assignments
Structure
Value assessment
Preferences
Time Preference
Philosophy
Risk Preference
Think
DBD Workshop, September 2000
Praise, Blame
Logic
Insight
Decision
Act
Outcome
Joy , sorrow
Sundar Krishnamurty, Umass-Amherst
Excerpts from Ron Howard’s
1960s paper
• Decision Analysis does not eliminate judgment,
intuition, feelings, opinions, or anything like that.
– Rather, it provides a mathematical framework to quantify
them and express them in a form where logic can operate
on them, rather than buried in human’s mind where we
cannot get access to them.
DBD Workshop, September 2000
Sundar Krishnamurty, Umass-Amherst
Excerpts from Ron Howard’s
1960s paper
• Can you obtain better outcomes by decision analysis
that by following intuition?
– “It is an act of faith.” A logical procedure based on sound
principles you believe in is better than another approach.
• When is it worthwhile to do decision analysis?
– Do a “back-of-the-envelope” analysis: How much would it
be worth more to do a more refined analysis? This itself
can be treated as a decision and analyzed as such.
• Decisions vs. Outcomes
DBD Workshop, September 2000
Sundar Krishnamurty, Umass-Amherst
Excerpts from Dr. Raiffa’s
Lecture Notes (1960s)
• “To make the correct engineering decisions requires extensive knowledge
and experience in engineering. Mathematical ability alone will not
suffice..” Admiral Rickover
• Sprit of decision analysis is to divide and conquer
• Decision Analysis requires explicit articulation of a though process
• Everything is reduced to dollar signs and lives saved. What about the
quality of life?
• Decision Analysis conveniently features those aspects that are readily
amenable to analysis and ignore like a plague intangibles that really count.
What would you do otherwise?
DBD Workshop, September 2000
Sundar Krishnamurty, Umass-Amherst
Is Multiattribute Utility Analysis a
useful method in Engineering
Design? What is your approach to
engineering decision making that
involves multiple attributes and
multiple decision makers?
DBD Workshop, September 2000
Sundar Krishnamurty, Umass-Amherst
Decision-making in Engineering
Design
• Decision: A choice from among a set of alternatives,
an irrevocable allocation of resources
– Options
– Expectations
• Outcome under Uncertainty: Process, Model, Computational
– Preferences & utility
• Possibly
Continuous Alternatives and Limitless
design options
• Options May be Unknown a-priori
• Heavy reliance on Simulation and Computational
Models
DBD Workshop, September 2000
Sundar Krishnamurty, Umass-Amherst
Decision-based Design
• Generation of option/design space
• Exploration/Formulation of option space
• Computation of outcome expectation
resulting from the choice of a particular
option
• Establishment of a value system to rank the
expectations
DBD Workshop, September 2000
Sundar Krishnamurty, Umass-Amherst
Expected Utility
• Optimal solution rule (Preferred Option)
– obtaining the highest overall Expected Utility value
• Need to develop the preference structure and utility functions
– Consistent with vN-M Axioms
– “Mathematical techniques that attempt to construct the preference order by
directly eliciting the decision makers preferences..”
• Bottom Line:
– Correct Preference
Correct Utility
Correct Decision
– Utility Theory: Mathematically rigorous approach to expressing human
preferences
• MAU: Keeney & Raiffa, von Winterfeldt & Edwards, others
• MAU in Engineering Design: Thurston, Krishnamurty, others
DBD Workshop, September 2000
Sundar Krishnamurty, Umass-Amherst
What are the top research issues for
developing DBD approaches to
design?
DBD Workshop, September 2000
Sundar Krishnamurty, Umass-Amherst
• Quality will decide:
– Common sense, good judgment, experimental & knowledge
based, Probability & Statistics, utility theory, validation
(Bayesian), etc.
• Equally important:
– Attending DBD workshops, Gordon Conferences
– Organizing Special Editions in Journals & Sessions in
Conferences
– Serving on NSF Panels
Be supportive!!
DBD Workshop, September 2000
Sundar Krishnamurty, Umass-Amherst