Pull based production systems : performance, modelling and analysis

Abstract: This doctoral thesis addresses performances, modelling and analyses of pull based production systems. The thesis consists of four main parts: a literature review, a comparison study of Kanban and CONWIP, an industrial practical case study, and the development of a new tool used for presentation and analysis of simulated performance results. Through a literature review existing performance measures, modelling techniques and analysing approaches were discovered and discussed. The literature review revealed a lack of performance measures regarding the variation in system performances. To compensate for this a new variation measure for analysing pull based production systems was found. Simulation was chosen as an analysing tool, because of the dynamic nature of practical production systems and the desire to analyse complex real systems not fitting other approaches.A comparison of Kanban and CONWIP, the two most basic pull based control methods, has been conducted through a simulation study. A five machine line system for production of a single product was used as a case. The result was seven rather general observations. A main finding is that Kanban and CONWIP systems need to be analysed due to variation measures and not only average measures, Kanban and CONWIP perform equally in terms of throughput due to average work in process (WIP), but unequally due to maximum storage capacity. An industrial practical case of a mass-produced single product have been studied and analysed with simulation. A new production line consisting of several chemical and mechanical processes was to be designed and installed. The design was tested for sensitivity to the experts' estimated parameters down time and required buffer capacities. The down times were categorized in five groups due to length, and three of these were analysed further. The system was sensitive to changes in all three groups but due to different performance measures and with different results of variations. The study also showed that the variations of the flow differed in the system from input to output. Based on performed simulations a new tool for presentation and analysis of performance data related to the system configurations has been invented. The result is a 3D Performance Surface method. Here, the performance's lead time, WIP and throughput are plotted into a three dimensional geometrical space. In this model the relation between different system settings and performances is analysed. Variation measures can also be included. The advantage with this geometrical tool is that the performances of not simulated configurations can be estimated.