Buffer Allocation in Unreliable Production Lines Using Infinitesimal Perturbation Analysis and Genetic Algorithm

This paper presents an approach based on Genetic Algorithm (GA) and Infinitesimal Perturbation Analysis (IPA) technique to maximize the production rate in unreliable production lines. Unlike Traditional optimization techniques based on simulation which require a large number of simulations runs to find the optimal solutions, the proposed approach uses a long and unique simulation. Indeed, through this single simulation, IPA which forms the heart of the GA-IPA, gives a reliable estimate of the gradient of the production rate, and where the input solutions are provided by GA. This gradient is then integrated into a stochastic optimization algorithm that runs simultaneously with the simulation to select the optimal buffer allocation. Computational experiments on various production lines are presented and discussed. The average Production Rate (PR) is calculated with 5 runs for the largest problem and up to 50 runs for the smallest problem, showing on one hand that the developed GA-IPA algorithm clearly outperforms the seven benchmark algorithms taken from the literature, and proving on the other hand the rapid convergence of our algorithm.