Transcript GoldSim 2006 User Conference

The Submodel Element
GoldSim 2006 User Conference
Slide 1
Vancouver, B.C.
The Concept of a Submodel Element


What does a GoldSim element actually do?
– Takes inputs
– Produces outputs
– It may have a random component to its behavior (e.g.
Stochastic elements).
What could you do if you were able to take an entire
GoldSim model and embed it as one element within a
parent model?
– The inner model could either:
• Do a deterministic simulation,
• Do a Monte Carlo simulation, or
• Do an optimization

– The inner model could be either dynamic or static
The Glacier release of GoldSim will provide this
capability.
GoldSim 2006 User Conference
Slide 2
Vancouver, B.C.
Uncertainty Analysis
 Imagine a model of a reliability system…
– The system undergoes random failures and
variable repair times, etc.
– You can use GoldSim to estimate its
reliability, availability, throughput, etc.
 But what if you are uncertain about some of
its key input parameters?
– Can you estimate its performance?
– Can you develop confidence bounds on your
estimate?
– Can you decide whether to go into production
vs doing more testing?
GoldSim 2006 User Conference
Slide 3
Vancouver, B.C.
Uncertainty Analysis (cont)
 You can address these issues by using a
‘double Monte Carlo’ analysis:
– An outer Monte Carlo loop samples the
uncertain system parameters.
– An inner Monte Carlo loop simulates the
system’s random performance given specific
values for the uncertain variables.
– The result: an understanding of the
uncertainty in the system’s performance, e.g.
“The probability that the mean warranty
return rate will exceed 5%”.
GoldSim 2006 User Conference
Slide 4
Vancouver, B.C.
Uncertainty Analysis Example
 Define two uncertain parameters: the Mean
life (3 yrs) and Slope Factor (2) for a Weibull
failure mode.
 Add a Submodel element that contains a
Reliability Function element, with the defined
failure mode.
 A warranty cost of $100 occurs for every
failure in the first two years.
 What is the 95% confidence limit for the
mean warranty cost?
– See Submodel1.gsm and Submodel1a.gsm
GoldSim 2006 User Conference
Slide 5
Vancouver, B.C.
Optimizing a Random
(or Uncertain) System

What is the optimum design for a system that is
subject to random effects?
– The optimum for one possible set of random effects
may be quite different from the optimum for another
set (e.g. if the earthquake occurs or not…).
– So you need to evaluate your objective function over
the full range of possible performance (i.e., to get a
PDF for the objective function).
– The true optimum design has to be based on
STATISTICS for the objective function, eg:
• “Minimize the expected total lifecycle cost while
allowing no more than a 2% likelihood of a
catastrophic failure”

An optimizing parent can use a Submodel element to
calculate the desired statistic(s) to optimize.
GoldSim 2006 User Conference
Slide 6
Vancouver, B.C.
Dynamic Optimization
 What if your model needs to make an optimal
decision at one or more points in time during
a simulation?
– Set up a static submodel that will find the
optimal solution and return it to the main
model.
 This could be used to simulate resource
allocation decisions or other ‘local’
optimization problems.
GoldSim 2006 User Conference
Slide 7
Vancouver, B.C.
Example: Optimize the Life-Cycle Performance of
a Manufactured Component (SubModel2.gsm)
This example minimizes the life-cycle costs for a manufactured
product, subject to a constraint that the likelihood of a
catastrophic failure is less than 2%.
The Submodel element does a stochastic simulation of the
products operational performance and costs.
3.14
16
MeanLife
GSM
TheSystem
X
X
AllCosts
3.14
16
SlopeFactor
GoldSim 2006 User Conference
Slide 8
X
X
Prod_Cost
Vancouver, B.C.