Evolutionary algorithms for the optimization of reliability in parallel/series redundant systems

The allocation of redundancy and availability in parallel/series system is known to be an intractable problem for which no deterministic algorithm is expected to succeed. The difficulties come from the constraints put on the problem and the mixed variable types. This paper presents two problem-specific evolutionary algorithms for solving the cost minimization and the availability maximization versions of this problem. For the first case, the proposed algorithm makes use of evolutionary operators that map feasible points into other feasible points. Specialized operators for boundary search allows an optimal utilization of available resources, since the search is restricted to solutions corresponding to a full utilization of these resources. In the second case, boundary search operators can not be explicitly stated. The proposed alternative is a repair algorithm that maps infeasible solutions to feasible ones as close as possible to the boundary of feasibility, if not exactly on it