Scope and MotivationsEpistemic uncertainties due to lack of knowledge can be found in many real-life design problems. The uncertainties cannot be modeled using the conventional statistical tools, e.g. Gaussian distribution model. Instead, some tools like Evidence Theory or Belief Function Theory can be used to model the uncertainties. Plausibility and belief, which describe the upper and lower bounds of the possible results, are used to evaluate the uncertainty impacts.
In some more complex problems, the epistemic uncertainties takes a more complex form, e.g. the uncertainties with Basic Probability Assignment (BPA) structure of the estimated values and variances. In some other design problems, both epistemic and aleatory uncertainties can be found as well. In this case, interaction between the epistemic and aleatory uncertainties should be considered in the design optimization.
The design optimization under uncertainties can be formulated as multi-objective optimization problem, with the objectives to minimize system function values and the objective to maximize corresponding beliefs. In all mentioned cases, the step-like optimal Pareto front should be obtained.
Another problem the designers face frequently in real-life engineering optimization is the fidelity management in the design optimization. A common way to tackle the design problems with multi-fidelity models is to start the optimization with the low fidelity model, and then use the expensive high fidelity model to refine the design solutions. The strategy works well in some problems but have a risk to trap into the pseudo optimal solutions if the values the low level model predicts is not consistent well with the high fidelity ones in that region. Therefore, the model fidelity management strategy for determining when the high fidelity model should be used is required. The strategy can be kriging, space mapping, and trust region method, etc. However, most of the strategies are designed for the single objective design cases, e.g to maximize the lift-drag ratio of the airfoil under specified Mach number. For a more complex design problem with multi-objective optimization under uncertainties, such strategies should be improved before they can be used in the design optimization.
In recent years, methods of optimization under epistemic uncertainties and multi-fidelity models are developed, but in a separate way. However, in many cases, both the epistemic uncertainties and the multi-fidelity models are involved. The session aims to develop efficient and combinatory strategies for these problems, and incorporate efficiently the evidence computation and model fidelity management in the design optimization. As the evidence computation can cost huge amount of computational resources if numbers of epistemic uncertainties are involved, high efficiency approximation techniques are required, and bench mark functions for the design optimization should be developed as well.
Session TopicsThe design problem under epistemic uncertainties and multi-fidelity models can be found in many practical problems, particularly in aerospace engineering design problems. The issues involved include MOO algorithm, uncertainty modeling, model fidelity management, and the strategies to integrate them to implement the optimization efficiently. The session seeks to promote the discussion and presentation of novel works related with (but not limited to) the following issues:
- Application of robust design optimization in engineering problems
- Uncertainty modeling
- Parameter reduction technique
- Model fidelity management
- Surrogate of expensive model
- Multi-objective optimization for the problems with step-like optimal front
- Approximation techniques for evidence computation
Important DatesPaper Submission: 15 January 2016
Decision Notification: 15 April 2016
Note that the submission should go to the special session in the submission system at IEEE-WCCI/CEC2016.
All accepted papers in the special sessions will be included in the published conference proceedings.
Session OrganisersDr. Liqiang Hou
State Key Laboratory of Astronautic Dynamics
Xi'an Satellite Control Center, Xi'an, China