System-of-Systems Engineering - Liophant Simulation
System-of-Systems Engineering - Liophant Simulation System-of-Systems Engineering - Liophant Simulation
Surrogate ModelsAre surrogate models the silver bullet? No,because:• They contain statistical error that maybe significant• Their applicability is limited by thedomain range of the DoE• The adequacy of a given surrogatemodel type depends on the nature ofbehavior it is meant to capture• They carry implicit assumptions thatcan be overlookedWhat kinds of surrogate models are there? Different surrogate models will capturevarying levels of complex behavior with agiven degree of accuracy/error using acorresponding regression/training data setResponse Surface Equations• …altogether, surrogates can bemisused or abused for analysis• Selecting the “right” type of surrogate(validation) is as important asconstructing the surrogate “right”(verification)Gaussian Process Regression,Or KrigingArtificial Neural NetworksDeveloping Methods and Techniques for System of Systems EngineeringDr. Dimitri Mavris, dimitri.mavris@ae.gatech.edu42
Challenges to Surrogate Modeling forStochastic Processes• For a given set of inputs, the outputs are a distribution, nota single point– Repetitions of each DoE case are required to capture thedistribution– If the variability is high, a large number of repetitions arerequired• The distribution may be non-normal, or be modal in nature– More difficult to accurately model a non-normal distribution– In the case of modal behavior, must determine the drivers forthe modes• Multiple surrogates will be required for each response– The number of surrogates required will be dependant on theassumed distributionDeveloping Methods and Techniques for System of Systems EngineeringDr. Dimitri Mavris, dimitri.mavris@ae.gatech.edu43
- Page 1 and 2: Developing Methods and Techniques f
- Page 3 and 4: The Aerospace Systems Design Labora
- Page 5 and 6: Motivation- Emphasis on “Early-Ph
- Page 7 and 8: What is a system?“A set of functi
- Page 9 and 10: System of Systems• “A set or ar
- Page 11 and 12: What is SoSE?• System of Systems
- Page 13 and 14: The Evolution of New Ideas“All tr
- Page 15 and 16: What is needed for this New Paradig
- Page 17 and 18: An Architecture-based Approach to S
- Page 19 and 20: The System Alternative Space• The
- Page 21 and 22: The Architecture-based Technology E
- Page 23 and 24: Extension to Systems-of-Systems Ana
- Page 25 and 26: Fundamental Properties of a Complex
- Page 27 and 28: Enablers for Complex SoS• Design
- Page 29 and 30: Design of Experiments• Design of
- Page 31 and 32: Design of Computer Simulation• Co
- Page 33 and 34: Agent-Based Behavioral Modeling•
- Page 35 and 36: System Dynamics Modeling• System
- Page 37 and 38: Discrete Event Simulations• Discr
- Page 39 and 40: Enablers for Complex SoS• Design
- Page 41: Surrogate ModelsWhy do we use surro
- Page 45 and 46: Fitting Modal ResponsesDeveloping M
- Page 47 and 48: Enablers for Complex SoS• Design
- Page 49 and 50: Enablers for Complex SoS• Design
- Page 51 and 52: Pareto Optimal Solutions in Many Di
- Page 53 and 54: Interactive Electronic Design Revie
- Page 55 and 56: Collaboration and Integration in So
- Page 57 and 58: Concluding Remarks• System of Sys
- Page 59 and 60: Selecting a DesignDeveloping Method
- Page 61 and 62: A New Paradigm Shift: Simulation Ba
- Page 63 and 64: Which are simple? Which arecomplica
- Page 65 and 66: What is a Markov Chain• Markov Ch
- Page 67 and 68: Markov Chains• The transition pro
Surrogate ModelsAre surrogate models the silver bullet? No,because:• They contain statistical error that maybe significant• Their applicability is limited by thedomain range <strong>of</strong> the DoE• The adequacy <strong>of</strong> a given surrogatemodel type depends on the nature <strong>of</strong>behavior it is meant to capture• They carry implicit assumptions thatcan be overlookedWhat kinds <strong>of</strong> surrogate models are there? Different surrogate models will capturevarying levels <strong>of</strong> complex behavior with agiven degree <strong>of</strong> accuracy/error using acorresponding regression/training data setResponse Surface Equations• …altogether, surrogates can bemisused or abused for analysis• Selecting the “right” type <strong>of</strong> surrogate(validation) is as important asconstructing the surrogate “right”(verification)Gaussian Process Regression,Or KrigingArtificial Neural NetworksDeveloping Methods and Techniques for <strong>System</strong> <strong>of</strong> <strong>System</strong>s <strong>Engineering</strong>Dr. Dimitri Mavris, dimitri.mavris@ae.gatech.edu42