Experiments and capability analysis in process industry

Abstract: The existence of variation has been a major problem in industry since the industrial revolution. Hence, many organizations try to find strategies to master and reduce the variation. Statistical analysis, such as process capability analysis and Design of Experiments (DoE), often plays an important role in such a strategy. Process capability analysis can determine how the process performs relative to its requirements or specifications, where an important part is the use of process capability indices. DoE includes powerful methods, such as factorial designs, which helps experimenters to maximize the information output from conducted experiments and minimize the experimental work required to reach statistically significant results. Continuous processes, frequently found in the process industry, highlight special issues that are typically not addressed in the DoE literature, for example, autocorrelation and dynamics. The overall purpose of this research is to contribute to an increased knowledge of analyzing DoE and capability in process industry, which is achieved through simulations and case studies of real industrial processes. This research focus on developing analysis procedures adapted for experiments and comparing decision methods for capability analysis in process industry. The results of this research are presented in three appended papers. Paper A shows how the use of a two-level factorial experiment can be used to identifying factors that affect the depth and variation of the oscillation mark that arises from the steel casting process. Four factors were studied; stroke length of the mold, oscillation frequency, motion pattern of the mold (sinus factor), and casting speed. The ANOVA analysis turned out to be problematic because of a non- orthogonal experimental design due to loss of experimental runs. Nevertheless, no earlier studies where found that shows how the sinus factor is changed in combination with the oscillation frequency so that the interaction effect could be studied. Paper B develops a method to analyze factorial experiments, affected by process interruptions and loss of experimental runs, by using time series analysis. Paper C compares four different methods for capability analysis, when data are autocorrelated, through simulations and case study of a real industrial process. In summary, it is hard to recommend one single method that works well in all situations. However, two methods appeared to be better than the others. Keywords: Process industry, Continuous processes, Autocorrelation, Design of Experiments, Process capability, Time series analysis.