The main objective of this thesis was to use a modeling approach to simulate the concentration of grain contamination in an existing corn processing system. The majority of the plant input is GMO corn, which is processed into multiple end products. In addition to GMO corn, each year, the plant processes six 11-day runs of non-GMO corn. During these non-GMO runs, the overall GMO contamination of the products produced is required to be less than 0.9% per the 3rd party labeling organization Non-GMO Project standards. Each run takes in approximately 1,400 lots (1000 bu/lot), with a sub-sample from each lot being tested at entry into the plant for contamination. Lots are accepted or rejected based upon a contamination acceptance threshold set by the plant management. This thesis models the current system to assess its performance. After establishing the model for the existing system, this thesis explores the impact of operational changes that might reduce costs and increase confidence in the ability to meet 3rd party labeling requirements for non-GMO products. This thesis partially fulfills the Master of Science degree requirement in Agricultural and Biosystems Engineering.

The first chapter provides context and establishes the scope and reasoning behind the work. It includes a literature review summarizing the history of GMO production and regulation in the United States and incorporates history on legislation concerning the creation, commercialization, and consumption of GMO products, as well as market trends on desirable foods. It discusses the challenges in separating GMO and non-GMO supply chains, the testing methods for detecting GMO contamination, and concludes with the blending methods currently used to reduce costs in grain facilities.

The second chapter introduces a modeling approach for assessing the current system using a discrete time simulation. The program uses the time-series entry-point contamination levels provided by a cooperating entity grain processor to calculate the contamination in the storage system. The variability and periodicity of these data were explored, and the data were found to fit a beta distribution. Because of an assumption of perfect blending, the average contamination thereby computed within the storage system is also the contamination in the outflow from the storage system, which is subsequently sent to processing. We then used the model to examine how the acceptance threshold level impacts outflow contamination levels. This allowed us to explore the feasibility of accepting lower quality corn to be blended with higher quality corn, all while obtaining the required contamination percentages going into the processing system.

Chapter three examines a critical operational question, namely whether segregating the incoming lots by contamination into bin sub groups would improve the confidence in outflow contamination levels. This chapter proposes a method to determine how many bin sub groups should be used, the percentage of each bin sub group to put into the final flow to the mill, and a decision tree analysis on what to do when a particular bin sub group runs out. To increase the range of contamination levels beyond those provided in the real data, we generated beta-distributed artificial data sets to run through the model. This allowed an examination of how various numbers of bin subgroups, and operational rules, impact outflow contamination levels.

The thesis concludes with a summary of the findings and a discussion of potential future avenues of exploration.