Computational Approaches for Life Cycle Inventory Database Development
This research aims to advance the current practice of developing Life Cycle Inventory (LCI) databases into a faster, less expensive process that still generates reliable LCI data. The research will (1) create a framework for modeling and analyzing LCI networks, (2) develop computational models for estimating missing LCI data, and (3) apply these models to evaluate LCI data quality and predict LCI data for emerging technologies. The education plan will (1) engage a diverse group of LCA practitioners during the course of the project, (2) deliver open source software add-ons for LCA practitioners to easily use the computational models developed in the proposed research, (3) develop an education theory grounded curriculum module incorporating research outcomes for broader dissemination, and (4) train undergraduate and graduate students with diverse background in STEM fields by engaging them in the research program and other education activities.
This research will develop computational approaches for estimating missing data in Life Cycle Inventory (LCI) databases based solely on limited known data, without relying on time-consuming, expensive empirical data collection. The approach transfers the latest knowledge from network science to LCI database development. An LCI database represents the interdependence of unit processes and environmental interventions. The ensemble of such interdependence characterizes the structure of the underlying technology network (or LCI network). If sufficient enough, observed LCI data, although limited, can be used to extract structural features of the underlying LCI network. Such structural features, in turn, can be used to predict the structure of the unknown area of the LCI network, which is equivalent to estimating the unknown data in the LCI database. This research will first create a framework for modeling and analyzing LCI networks. This framework will then be used to develop and validate a variety of link prediction models to estimate missing data for LCI databases. Finally the validated link prediction models will be used to evaluate LCI data quality and predict LCI data for emerging technologies for testbed databases selected in consultation with stakeholders.