Los Alamos National Laboratory
Our food production infrastructure is a highly complex system of systems (SOS). Many of the foods we eat are complex, manufactured products with many constituent ingredients, any one of which could be a source of contamination. The supply-distribution links connecting these systems are themselves complex networked systems only partially characterized at best. Yet they may amplify or mitigate risk depending on their structure. This makes assessing the risk to our foods a daunting task. Traditional probabilistic risk assessment (PRA) relies on detailed system information typically limiting its application to estimating risk associated with specific facilities. Heuristic approaches based purely on expert judgment are more broadly applicable, but lack the mathematical rigor necessary to reliably evaluate large-scale, complex systems. This project aims to address these challenges by developing a risk assessment methodology based on abstract models of food manufacturing facilities and product supply chains combining systems dynamics and complex network models into a hybrid modeling system of systems. We will develop the methods essential features using a model system of snack cake manufacture depending on the production of milk, eggs, sugar, flour and cocoa. We intend to work closely with the food industry to obtain the data needed to characterize the model system. We will define vulnerability and consequence in a generic way appropriate to estimating risk in the absence of detailed system information. Examining morbidity and mortality as the primary risk, we will correlate risk in the system with the amount of contaminant relative to a reference dose and will propagate it through the systems dynamics models and distribution networks. Vulnerability will be defined as the likelihood that a given amount of contaminant, relative to reference dose, can be introduced at a particular stage. This will be modeled as a subjective probability distribution based on expert judgment and reasoning models, and propagated through the system using Bayesian and Monte Carlo techniques. Using the proposed approach, we will perform computational experiments based on two categories of agents: viable organisms and toxins. For a given agent, and a given ingredient as the contamination source, we will use optimization to find the worst case regime. We will then combine results using statistical surrogate models and design-optimization to find the region of greatest overall risk. If successful, the proposed methodology would provide a principled way of exploring risk in very complex systems where it is not practical to engineer highly detailed models.