Employee Portal
 

Engineering System Selection and Optimization under Uncertainty

ETC MADA/MADO

The objective of this research was to explore and develop Multi-Attribute Decision Analysis (MADA) and Multi-Objective Robust Optimization (MORO) models and methods for making one or more “best’ choices for engineering (including Naval) systems under uncertainty.

A MADA model was developed for a notional case study dealing with payload (rocket) selection for a Fire Scout helicopter. A Fire Scout helicopter is used on Navy ships to defend from attacks by small boats. For the case study, the decision variables considered were the type and quantity of payloads. The conflicting attributes were to minimize the cost and weight of the helicopter payload while maximizing the probability of kill for an unknown number of boats which may be of unknown size. In the MADA model, the three attributes were scaled and weighted to reflect their relative importance and combined into a single value function. Next, an Influence Diagram was devised, using thePrecisionTree® tool, to calculate the ‘payoff’ of each of the payload alternatives. The influence diagram was used to rapidly explore different choices and compute corresponding payoffs and also facilitate sensitivity analysis and more in-depth assessment of solutions. An alternative approach was also investigated and developed, using the utility analysis theory of MADA, which took into account risk taking behavior of the decision maker during the payload selection. A MORO technique for dealing with reducible and irreducible uncertainty was also investigated and developed.

The work involved offline sampling, meta-modeling of objective and constraint functions and finally a two-level MORO approach that accounted for reducible and irreducible uncertainty for parameters. This approach is able to generate optimized solutions that are comparable to those obtained without approximation and with much less computational effort. For such systems, sometimes, it is feasible to decrease the amount of uncertainty in parameters, of course at a cost, to achieve better performance