Contributions to the use of designed experiments in continuous processes a study of blast furnace experiments

Sammanfattning: Design of Experiments (DoE) contains techniques, such as factorial designs, that help experimenters maximize the information output from conducted experiments and minimize the amount of experimental work required to reach statistically significant results. The use of DoE in industrial processes is frequently and thoroughly described in literature. However, continuous processes in industry, frequently found in, for example, the mining and steel industries, highlight special issues that are typically not addressed in the DoE literature. The purpose of this research is to contribute to an increased knowledge of the use of DoE in continuous processes and aims to investigate if factorial designs and other existing techniques in the DoE field are effective tools also in continuous processes. Two studies have been performed. The focus of the first study, a case study of an industrial blast furnace operation, is to explore the potential of using factorial designs in a continuous process and to develop an effective analysis procedure for the experiments in a continuous process. The first study includes, for example, interviews, experiments, and large elements of action research. The focus of the second study is to explore how a-priori process knowledge can be used to increase the analysis sensitivity for unreplicated experiments. The second study includes a metastudy of experiments in literature as well as an experiment. The results show that it is possible to use factorial designs in a continuous process even though it is not straightforward and special considerations by the experimenter will be required. For example, the dynamic nature of continuous processes affects the minimum time required for each run in an experiment since a transient time period is needed between each run to allow the experimental treatments to reach full effect in the process. Therefore, the use of split-plot designs is recommended since it can be hard to completely randomize the experimental run order. It is also found that process control, during the conduction of the experiment, may be unavoidable in continuous processes. Thus, developing a process control strategy during the planning phase is found to be an important experimental success factor. Furthermore, the results indicate that the multitude of cross-correlated response variables typical for continuous processes can be problematic during the planning phase of the experiment. The many and cross-correlated response variables are also reasons to why multivariate statistical techniques, such as principal component analysis, can make an important contribution during the analysis. Moreover, a-priori process knowledge is confirmed to have a positive effect on analysis sensitivity for unreplicated experiments. Since experimental effects in continuous processes can be expected to be small compared to noise, a-priori process knowledge can also make a valuable contribution during analysis of experiments in continuous processes. Furthermore, activities like coordination of people, information and communication as well as logistics planning are found as important parts of the experimental effort in continuous processes.