Infinitesimal perturbation analysis for second derivative estimation and design of manufacturing flow controllers

A simulation-based numerical technique for the design of near-optimal manufacturing flow controllers for unreliable flexible manufacturing systems uses quadratic approximations of the value functions that characterize the optimal policy and employs stochastic optimization to design the key coefficients of the quadratic approximations. First and second derivative estimates that drive the optimization algorithm are obtained from a single sample path of the system via infinitesimal perturbation analysis (IPA). Extensive computational experience is reported for one, two, and three-part-type production systems. The relative performance of first-order and second-order stochastic optimization algorithms is investigated. The computational efficiency of these algorithms is finally compared to conventional controller design algorithms based on state-space discretization and successive approximation.This research was supported by the National Science Foundation, Grant No. DDM-89-14277 and DDM-9215368.