Variation is inevitable in any process, so it has to be dealt with effectively and economically. Reducing variation can be achieved in assembly lines strategically and dynamically. Implementing both the strategic and dynamic variation reduction techniques is expected to lead to further reduction in the number of failed final assemblies. The dissertation is divided into three major parts. In the first part, we propose to reduce variation for assemblies by developing efficient inspection plans based on (1) historical data for existing products, or simulated data for newly developed products; (2) Monte Carlo simulation; and (3) optimization search techniques. The cost function to be minimized is the total of inspection, rework, scrap and failure costs. The novelty of the proposed approach is three-fold. First, the use of CAD data to develop inspection plans for newly launched products is new, and has not been introduced in the literature before. Second, frequency of inspection is considered as the main decision variable, instead of considering whether or not to inspect a quality characteristic of a subassembly. Third, we use a realistic reaction plan (rework-scrap-keep) that mimics reality in the sense that not all out-of-tolerance items should be scrapped or reworked. At a certain stage, real-time inspection data for a batch of subassemblies could be available. In the second part of this dissertation, we propose utilizing this data in near real-time to dynamically reduce variation by assigning the inspected subassembly parts together. In proposing mathematical models, we found that they are hard to solve using traditional optimization techniques. Therefore, we propose using heuristics.Finally, we propose exploring opportunities to reduce the aforementioned cost function by integrating the inspection planning model with the Dynamic Throughput Maximization (DTM) model. This hybrid model adds one decision variable in the inspection planning; which is whether to implement DTM (assemble the inspected subassemblies selectively) or to assemble the inspected items arbitrarily. We expect this hybrid implementation to substantially reduce the failure cost when assembling the final assemblies for some cases. To demonstrate this, we solve a numerical example that supports our findings. / Ph. D.

Dans les systèmes manufacturiers de plus en plus complexes, les variations du processus de fabrication et de ses paramètres opératoires ainsi que leurs effets sur l’ensemble du système doivent être maîtrisés, mesurés et contrôlés. Cette thèse propose un cadre d’optimisation pour l’élaboration d’un plan d’inspection optimal qui permet une prise de décision opérationnelle afin d’assurer la satisfaction des objectifs stratégiques (réduction des coûts, amélioration de la qualité, augmentation de la productivité, …). La prise de décision se divise en trois questions : Quoi contrôler ? Comment contrôler ? Quand contrôler ? Le manque d'informations fiables sur les processus de production et plusieurs facteurs environnementaux est devenu un problème important qui impose la prise en compte de certaines incertitudes lors de la planification des inspections. Cette thèse propose plusieurs formulations du problème d’optimisation de la planification du processus d'inspection, dans lesquelles, les paramètres sont incertains et les machines de production sont sujettes aux défaillances. Ce problème est formulé par des modèles de programmation mathématique avec les objectifs : minimiser le coût total de fabrication, maximiser la satisfaction du client, et minimiser le temps de la production totale. En outre, les méthodes Taguchi et Monte Carlo sont appliquées pour faire face aux incertitudes. En raison de la complexité des modèles proposés, les algorithmes de méta-heuristiques sont utilisés pour trouver les solutions optimales. / Quality inspection in multistage production systems (MPSs) has become an issue and this is because the MPS presents various possibilities for inspection. The problem of finding the best inspection plan is an “inspection planning problem”. The main simultaneous decisions in an inspection planning problem in a MPS are: 1) which quality characteristics need to be inspected, 2) what type of inspection should be performed for the selected quality characteristics, 3) where these inspections should be performed, and 4) how the inspections should be performed. In addition, lack of information about production processes and several environmental factors has become an important issue that imposes a degree of uncertainty to the inspection planning problem. This research provides an optimization framework to plan an inspection process in a MPS, wherein, input parameters are uncertain and inspection tools and production machines are subject to breakdown. This problem is formulated through several mixed-integer mathematical programming models with the objectives of minimizing total manufacturing cost, maximizing customer satisfaction, and minimizing total production time. Furthermore, Taguchi and Monte Carlo methods are applied to cope with the uncertainties. Due to the complexity of the proposed models, meta-heuristic algorithms are employed to find optimal or near-optimal solutions. Finally, this research implements the findings and methods of the inspection planning problem in another application as hub location problem. General and detail concluding remarks are provided for both inspection and hub location problems.