A singular perturbation model of reliability in systems control (Q794615)

In many engineering systems the intensity of the input noise is small and the state trajectory is mainly due to the deterministic part of the system structure. When this is the case for a white input noise, the mean exit-time from the reliability region reduces to the solution of a singularly perturbed partial differential equation for which an asymptotic solution can be evaluated analytically in a fairly general case. The mean exit-time is then used to measure the reliability of some decentralized control policies for linear stochastic systems. The concepts of reliability region and of the appropriate degree of reliability are proposed. In system analysis they are used to measure the instability of the perturbed system in the given domain. The synthesis of a most reliable control is demonstrated to be feasible. For large, linear, weakly coupled systems, the degree of suboptimality of the decentralized control is interpreted as an appropriate degree of reliability of the system.