Gestion conjointe de production et qualite appliquee aux lignes de production non fiables

This research is concerned with unreliable production lines. Two types of machines are considered here: a machine for which part of the production is part substandard in quality and a machine whose production is 100% in conformity. The thesis is organized according to three principal contributions.;In the first part of our research and for a given choice of the maximum allowable total storage parameter, the performance of constant work-in-process (CONWIP) disciplines in unreliable transfer lines subjected to a constant rate of demand for parts, is characterized via a tractable approximate mathematical model. For a (n -- 1) machines CONWIP loop, the model consists of n multi-state machine single buffer building blocks, separately solvable once a total of (n -- 1) 2 unknown constants shared by the building blocks are initialized. The multi-state machine is common to all building blocks, and its n discrete states approximate the joint operating state of the machines within the CONWIP loop; each of the first (n -- 1) blocks maps into a single internal buffer dynamics, while the nth building block characterizes total work-in-process (wip) dynamics. The blocks correspond to linear n component state equations with boundary conditions. The unknown (shared) constants in the block dynamics are initialized and calculated by means of successive iterations. The performance estimates of interest, mean total wip, and probability of parts availability at the end buffer in the loop are obtained from the model and validated against the results of Monte-Carlo simulations. In the second part of our research, we address the optimal production control problems for an unreliable manufacturing system that produces items that can be regarded as conforming or non conforming. A new stochastic hybrid state Markovian model with three discrete states, also called modes is introduced. The first two, operational sound and operational defective are not directly observable, while the third mode, failure, is observable. Production of defective parts is respectively initiated and stopped at the random entrance times to and departure times from the defective operational mode. The intricate piecewise-deterministic dynamics of the model are studied, and the associated Kolmogorov equations are developed under the suboptimal class of hedging policies. The behavior of the model is numerically investigated, optimized under hedging policies, and subsequently compared to that of a tractable extension of the two-mode Bielecki-Kumar single machine model, where both conforming and defective parts are simultaneously produced in the operational mode, while the ratio of produced non conforming to conforming parts remains fixed.;Finally, we consider a fluid model of an unreliable production line consisting of n machines and n fixed buffer sizes. These machines produce a single part type with two different quality levels: conforming and non-conforming parts. The ratio of non-conforming parts to conforming ones is assumed to be a constant, which may vary depending on the machine. The production line can contain inspection stations whose function is to reject the non-conforming parts from the system. It is assumed that the production line must meet a constant rate of demand for good parts. The objective is to develop an approximate modeling framework and an optimization algorithm for unreliable transfer line inter machine buffer sizing, so as to minimize, under a constant demand for parts rate, the average long term combined storage and shortage costs, while accounting for parts quality and specifying the optimal location of inspection stations. Decomposition / aggregation methods developed in (Sadr et Malhame (2004b)) and their dynamic programming based optimization algorithm are adapted to the current model. In addition, numerical results based on the approximate theory, and those obtained from Monte-Carlo simulation, are contrasted.