System of Systems Architecture Evaluation Using Evolutionary Computation

Evolutionary computation and evolutionary algorithms represent a developing science and technology that can be effectively applied to the generation and evaluation of system of systems architectures. A general technique used by systems engineering professionals is a binary matrix representation of a system or system of systems. The specific meaning and semantics of the binary relationship depends of the type of representation used. Typical representations are, “N squared”, design structure matrix, dependency structure matrix, and implication matrix. A key feature of these typical representations is their direct relationship to the structure required in an evolutionary computational approach. Evolutionary algorithms can be applied to the evaluation and optimization of these matrix structures. A new evolutionary algorithm has been developed that applies specifically to the generation and evaluation of systems and system of systems. This new evolutionary algorithm incorporates a fuzzy inference system in the calculation of the best fit evaluation. The current industrial and social environment is populated with a vast array of existing and developing systems. Any new system must take this context into account. Evolutionary computation is applied to assist the system architect and engineer in the evaluation of these complex configurations and interface sets. The new evolutionary computing techniques are applied to system of systems architecting tasks using a well defined set of measures of effectiveness (MOE). The systems architecting task is divided into three general areas organized around the roles and responsibilities associated with the system architect, the system customer and the system engineer. The system architect is responsible for the complete system operation and MOE balance, focused on life-cycle cost and risk. The customer is responsible for the mission profile and mission functions. Operational effectiveness and operational suitability areas are the responsibility of the systems engineers. Affordability, risk, operational effectiveness and operational suitability are the four MOE used to evaluate the candidate system of systems architectures. Introduction: Systems engineering classically uses measures of effectiveness to evaluate total system performances (Goode and Machol, 1956). The system architecture performance measures in this approach are defined and used in the context of classical systems engineering measures of effectiveness. System effectiveness is defined as a quantifiable measure of the degree to which the candidate system under evaluation is expected to meet and/or perform the stated mission need and objectives using the provided mission profile. It is important to select and match the measure of effectiveness with the given customer problem. The system customer provides the mission context and the mission function architecture which represents the decomposition of the mission function that needs to be performed. During the architecture generation and evaluation process, each candidate system architecture is evaluated to determine how well the system architecture will perform the required mission function as well as to consider other system architecture factors, such as suitability, cost and risk. The primary measure of effectiveness that can be used is system effectiveness versus life cycle cost (Dahlberg, 2004). This classical system MOE is composed of four sub-components: operational effectiveness, operational suitability, life cycle cost and risk. These four sub-components are selected to provide logical and semantic consistency in the development and use of this specific MOE. The operational effectiveness sub-component is focused on how well the candidate system under evaluation meets the complete mission function. The operational suitability sub-component considers the specific physical and design features associated with a specific physical system solution. The life cycle cost component is focused on the system cost evaluation while the risk component considers all sources of risk associated with a specific candidate system. The classical system MOE approach has been modified to accommodate the unique attributes of fuzzy numbers, fuzzy inference systems and evolutionary algorithms. System Evaluation Using Fuzzy Measures of Effectiveness Every evolutionary algorithm must incorporate a fitness function that evaluates the current solution population and selects the best solutions for use in the production of the next generation of solutions. A fuzzy inference system is used as the fitness function for the evolutionary algorithm discussed in this paper. A diagram of the evolutionary computation process is shown in Figure 1, which outlines the basic steps of evolutionary computation. Architectures Chromosome Crossover Fuzzy Assessment Selection Mutation Figure 1. Evolutionary Computation Process The fuzzy measures of effectiveness are used in the fuzzy assessment as well as the selection steps of the evolutionary computation process. All four subcomponents of the system measure of effectiveness have been modified to integrate with the application of fuzzy logic. The primary modifications are associated with replacing “percentage measures” with “degree of membership” measures as well as the inclusion of a weighting mechanism to vary the weights and impacts of each subcomponent on the final system measure of effectiveness. The context for the application of this set of modified measures is shown in Figure 2, ‘General Architecture Development Context.’