Adaptive design and analysis procedures for expensive industrial experiments with inadequate prior knowledge

Statistically designed experimental procedures provide an insight knowledge of the system and have several advantages over other techniques. Although they require use of various statistical tools and techniques, they are getting more and more popular these days.;Current design and analysis procedures are applied in certain kinds of industries (e.g., automotive). However, there is a need to overcome the problems faced in the design and analysis of expensive industrial experiments. Such experiments require great attention and care as they are costly and may go incomplete. Current practices are not adequate to solve these problems. The aim of this research work is to develop techniques that can be effectively utilized to properly design and conduct the experiments for large scale industrial experiments.;This research work proposes a competent technique that can be used for conducting the experiments in an efficient way. It allows the experimenter to conduct experiments at more desirable points first, by quickly shifting to the optimum region. It thus minimizes, in the early stages, the less desirable design points to a minimum, collecting data at the points the response values at which are closer to the objective.;Costly experiments may not allow the experimenter to complete all runs resulting in incomplete factorial designs. Procedures to deal with such situation are also included in this research. Two methods for estimation of missing data devised as a part of this research are compared to two of the existing ones. Data analysis procedures using all of these methods are presented. Use of examples has supported the two new methods by yielding more accurate analysis results.;Due to the complex nature of large scale industrial experiments, it is sometimes feasible to conduct the experiments serially in a nonrandom fashion. Experiments performed with such randomization restriction need to be analyzed in an appropriate manner. Procedures for analyzing serially obtained observations are also included in this research.;These solutions will help the experimenters in careful planning of the experiments, carrying them out in a determined manner and analyzing them properly. The solutions provided in this dissertation can be effectively incorporated into an organizations' design and development process to overcome the problems addressed above.