Contributions to the Industrial Use of Design of Experiments

Abstract: A great variety of statistical methods are important for facilitating the continuous improvement of customer-oriented processes and products, and Design of Experiments (DoE) has proven to be one efficient approach. The considerable industrial interest together with substantial research have revealed further possibilities of development in DoE to increase the attractiveness of the method. This thesis concentrates on two aspects of DoE: analysis and dissemination.Experimental results are often analyzed with the normal probability plot. The plot is, however, considered troublesome due to its high proportion of informality. A new method, based on the Total Time on Test (TTT) technique for analyzing 2k-p experiments, is presented. The method consists of an informal graphical part and a formal part, where the former is a modified TTT plotand the latter utilizes the cumulative TTT statistic. The method facilitates automatic evaluations as the formal part needs no assumptions of the number of inert contrast. A simulation study indicates that the method presented is quite up to the standard of other presented methods. An extensive simulation study comparing eight of the most common methods, including the TTT based method, indicates that there are small differences in the power of the methods under effect sparsity although the differences increase as the number of active contrast increases.A method for identifying dispersion effects in inner and outer array experiments, a common design in robust design applications, is also presented. Simulation is applied to compare the power of the proposed method with earlier proposed methods under several variants of split plotstructures.Co-operation with Swedish organizations revealed problems in disseminating DoE throughout the organization. Two case studies are presented, where the earlier study focuses on important components in a dissemination strategy. The later study focuses on learning and the dissemination problem is studied in two different perspectives related to learning. The results indicate that for a lasting use of DoE both strategies, structures, easy-to-follow formulas and uncertainty are necessities.