On the Application of Designed Experimentation for Customer Focused Product Development

Sammanfattning: Today, the use of statistical methods and tools for solving problems in industry is growing steadily. The increased application of statistics in industry is attributable to the many theoretical advancements made by researchers over the years, but also to the increased awareness in industry that statistical methods and tools constitute an important resource of great strategic value. The foundation of this thesis lies in the ideas presented by Walter A. Shewhart in the early 1930s and the influence they have on today's activities in industry. The thesis uses an integrated framework for linking customer satisfaction to product development activities to establish the relationship between Shewhart's statistical perspective and customer focused product development. In particular, the important role of designed experimentation in this framework is highlighted. In detail the following topics are considered.An important challenge facing product developers is to link customer satisfaction to development activities carried out within a company. Success in the marketplace is very much dependent on the ability to translate satisfaction into product properties and to do it as early as possible in the product development process. The use of Quality Function Deployment combined with Customer Satisfaction Modeling is presented as means for linking internal product development and improvement activities to increased customer satisfaction.Separating active factors from inert ones represents an important problem in the analysis of designed experiments. An extensive simulation study comparing eight procedures complementing the normal probability plot for this separation is presented. The study shows that only small differences appear between different procedures, regardless of complexity, when the sparsity assumption is not violated. As the number of active factors increases, so do the differences in the performance of the studied procedures. Among the best performing procedures is one that rests on a Bayesian foundation and utilizes generic a priori knowledge for the selection problem. To further improve its performance a modification is proposed that will allow introduction of more elaborate and foremost less generic domain knowledge. The performance of the modified procedure is evaluated in a simulation study and it is found that considerable improvements can be made unless the introduced domain knowledge strongly contradicts the reality. A procedure for construction of new contrasts supplementing the original experimental design is introduced that will allow more efficient use of the available data. The new contrasts can be defined in various ways and thereby be given different interpretations in order to serve special purposes according to the agenda of the analyst. Apart from an enhanced interpretation of the experimental results, these contrasts contribute to a more elaborate representation of the experimental error when using normal probability plotting.In an effort to encourage use and to illustrate the possible benefits from using Conjoint Analysis, a step-by-step introductory workflow is presented. One of the apparent problems when designing a conjoint analysis study is the conflict between including many product attributes and not overloading the respondents. Non-geometric Plackett-Burman designs represents a class of orthogonal designs that provide an opportunity to resolve this conflict. The use of non-geometric Plackett-Burman designs for conjoint analysis is advocated and a procedure that takes advantage of the special properties of the non-geometric Plackett-Burman designs is proposed and demonstrated.