Bayesian Optimal Experimental Design for Intelligent Data Collection in Material Flow Analysis

Abstract

Material flow analyses (MFAs) are powerful tools for identifying and analyzing energy and material efficiency (resource efficiency) opportunities across a supply chain. MFAs are typically represented as directed graphs with key parameters including the mass of the material flows entering the system and the allocation of materials flowing through one node (typically representing a process or location) to other nodes in the system. Parametric uncertainty can hamper the credibility and usability of MFA results. Uncertainty may be reduced by collecting more data; however, an intelligent data acquisition strategy is needed given the limited resources available for completing a given MFA. In this article, we apply Bayesian optimal experimental design (BOED) derived from the Kullback-Leibler divergence to target the collection of high value data, which is then fed into a Bayesian framework to effectively reduce the MFA parametric uncertainty. The methodology is demonstrated using a case study on the 2012 U.S. steel sector. Bayesian inference is then used to validate the BOED results with data collected from the United States Geological Survey and the World Steel Association. This article’s methods allow efficient data collection to rapidly create MFAs with reduced and quantified parametric uncertainty, aiding decision makers in their efforts to pursue resource efficiency.