A Multi-Objective Stochastic Approach to Combinatorial Technology Space Exploration


Abstract

Historically, aerospace development programs have frequently been marked by performance shortfalls, cost growth, and schedule slippage. New technologies included in systems are considered to be one of the major sources of this programmatic risk. Decisions regarding the choice of technologies to include in a design are therefore crucial for a successful development program. This problem of technology selection is a challenging exercise in multi-objective decision making. The complexity of this selection problem is compounded by the geometric growth of the combinatorial space with the number of technologies being considered and the uncertainties inherent in the knowledge of the technological attributes. These problems are not typically addressed in the selection methods employed in common practice. Consequently, a method is desired to aid the selection of technologies for complex systems design with consideration of the combinatorial complexity, multi-dimensionality, and the presence of uncertainties. Several categories of techniques are explored to address the shortcomings of current approaches and to realize the goal of an efficient and effective combinatorial technology space exploration method. For the multi-objective decision making, a posteriori preference articulation is implemented. To realize this, a stochastic algorithm for Pareto optimization is formulated based on the concepts of SPEA2. Techniques to address the uncertain nature of technology impact on the system are also examined. Monte Carlo simulations using the surrogate models are used for uncertainty quantification. The concepts of graph theory are used for modeling and analyzing compatibility constraints among technologies and assessing their impact on the technology combinatorial space. The overall decision making approach is enabled by the application of an uncertainty quantification technique under the framework of an efficient probabilistic Pareto optimization algorithm. As a result, multiple Pareto hyper-surfaces are obtained in a multi-dimensional objective space. Each hypersurface represents a specified probability level, which in turn enables probabilistic comparison of various options. Other more traditional technology selection and scanning techniques such as the greedy algorithm, one-on one-off technique and designs of experiments are also explored. An advisor to recommend the best selection technique from amongst these options based on the complexity and scope of the problem is also an important contribution of this research. Various techniques used for creating the exploration and decision making methodology are experimented on a benchmark knapsack problem. These techniques are used in a synergistic manner to formulate the Pareto Optimization and Selection of Technologies (POST) methodology. POST is implemented on an example technology exploration and selection problem for a 300 passenger commercial aircraft. This is a large problem with 29 technologies, 11 objectives and 4 constraints. Initially, the technologies and their system impacts are defined along with their uncertainties. The computational complexity is evaluated and the problem dimensionality reduced using a dominance structure preserving approach. Probabilistic Pareto optimization is implemented with the reduced dimensionality and three Pareto layers each corresponding to a predefined probability level are created. These Pareto layers are exported to a visualization and analysis environment enabled by JMPr. The technology combinations on these Pareto layers are explored using various visualization tools and one combination is selected. The main outcome of this research is a method based on a consistent analytical foundation to create a dynamic tradeoff environment in which decision makers can interactively explore and select technology combinations.