Chapter 2. RAMS+C optimization and Genetic Algorithms
74
the others introduced as constraints. Multi-objective optimization proves a series of best tradeoff solutions called the Pareto-optimal set. Optimization of RAMS+C for safety systems is a complex non-linear integer, combinatorial and stochastic problem. Many times it is difficult to define the dependability functions in explicit analytical form. It is here when the advantages that GAs convey for efficiently treating illbehaved problems can be exploited. GAs mimic the natural evolution process based on stochastic search techniques. They are able to handle high dimensional, non-linear and discrete problems with discontinuous functions. They produce a pool of several optimal solutions which provides great flexibility for decision-making. Two of the most efficient multi-objectives GAs are the Fonseca & Fleming MOGA and the NSGA-II. The advantages of these two GAs is exploited in this thesis for solution of SIS optimization problems. They are used for implementing the optimization cases of chapter 3 to 6, where a detailed description of the working principles of these algorithms is provided. RAMS+C multi-objective optimization with GAs has a recent history, where several aspects of it has been approached in the search for better designs and test and maintenance polices. An overview of these approaches has been provided in this chapter, and some issues not fully explored have been detected. The optimization cases previously addressed by other researchers have not comprised the compliance with the requirements with the international standards IEC 61508, including the level of modelling detail necessary for real-life SIS. Detected niches and opportunities for research of system optimization with GAs that will be explored in this work include modelling of Common Cause Failure and diagnostic coverage, integration of diverse redundancy for improvement of system performance, time-dependent modelling for more precise modelling of unavailability in optimization of test intervals and strategies and integration of MooN voting architectures optimization of system design and test.