480945 Hypothesis Investigation without a Designed Experiment

Monday, March 27, 2017
Exhibit Hall 3 (Henry B. Gonzalez Convention Center)
Jenny Chaves, Analytical Technology Center, The Dow Chemical Company, Midland, MI, Mary Beth Seasholtz, The Dow Chemical Company, Midland, MI and Dana A. Livingston, Global Ag Process Research, The Dow Chemical Company, Pittsburg, CA

A designed experiment enables one to infer causal relationships between input variables and the variable of interest. When investigating a hypothesis related to a chemical process issue, a well-designed experiment can provide strong evidence to confirm or refute the claim. What about situations when a designed experiment is not feasible? This situation is common in a chemical manufacturing environment due to the high costs and potential safety concerns associated with recreating the process conditions that caused the original problem. One of Dow’s plants faced this challenge as they sought to investigate the claim that the source (supplier) of a raw material was responsible for the lower concentration of active material in the final product. The raw material of interest is highly toxic and has to be manufactured on-site whenever possible to minimize its transport. Occasionally it is obtained from an off-site source when it cannot be produced in sufficient quantities on-site. A combination of visual methods and multivariate modeling techniques were used to examine historical plant data over a period of 3 years. The frequency of the data was hourly (averaged), and approximately three dozen variables were used from multiple sources: online process instruments, online plant analyzers, and the quality assurance lab. The hypothesis was convincingly refuted in the absence of a designed experiment.

Extended Abstract: File Not Uploaded
See more of this Session: Big Data Analytics Poster
See more of this Group/Topical: Spring Meeting Poster Session and Networking Reception