469060 Estimating the Life Cycle Impact of Chemicals from Molecular Descriptors and Thermodynamic Properties Via Mixed-Integer Linear Programming
Tuesday, November 15, 2016: 8:30 AM
Union Square 13 (Hilton San Francisco Union Square)
Raul Calvo-Serrano, CENTRE FOR PROCESS SYSTEMS ENGINEERING, Imperial College, London, United Kingdom, María González Miquel, Centre for Process Integration, University of Manchester, Manchester, United Kingdom, Stavros Papadokonstantakis, Energy and Environment, Chalmers University of Technology, Gothenburg, Sweden and Gonzalo Guillén-Gosálbez, Centre for Process Systems Engineering, Imperial College of Science, Technology and Medicine, London, United Kingdom
The chemical industry is at present striving to decrease its environmental footprint during the transition towards a more sustainable manufacturing sector. In this context, there is a clear need to develop systematic methods to assess and optimize the environmental impact of chemicals, identify critical hotspots across their supply chains and define the best strategies to implement so as to enhance their sustainability level. In the recent past, it has become clear that the environmental impact should be assessed considering all the stages in the life cycle of the products/services being assessed. As a result, Life Cycle Assessment (LCA) has become (arguably) the prevalent approach to quantify the environmental burdens of products, finding applications in a wide variety of areas, including the evaluation of the environmental footprint of chemicals from cradle to grave.One of the main limitations of LCA is that it is highly data intensive, that is, it requires large amounts of data from several echelons of the product supply chain, which are typically owned by different companies that might be reluctant to share this environmental information or even lack the corresponding measurements. This is critical in the chemical industry, where chemicals are produced by complex interrelated chemical networks that exchange energy, mass and water and consume a wide range of intermediate products and feedstocks.
One possible way to simplify the LCA calculations is to resort to Streamlined LCA (SLCA) approaches, which attempt to reduce the upstream and downstream information required in a standard LCA by using proxy data, qualitative models and/or regression equations. These methods are typically tailored to a particular sector, as otherwise the underlying simplifications and assumptions would not hold, thereby leading to poor estimates and large errors.
SLCA methods have been developed for a wide range of production processes, including the building sector (where the size of the bill of materials often makes unpractical the use of standard approaches), urban water treatments and hydroelectric power plants, among others. In the context of chemical processes, SLCA has been applied in oil refining as well as to general chemical processes. In this work we introduce a novel approach for evaluating the life cycle impact of chemicals that provides better estimates and reduces the complexity of previous models, thereby opening new avenues for their use within the framework of computer aided molecular design tools. In essence, building on the seminal paper by Wernet et Al. (2008)1 that predicted the life cycle impact of chemicals from their molecular structure using neural networks, we present here an alternative method based on rigorous mixed-integer optimization techniques that uses key thermodynamic properties in addition to molecular descriptors in order to produce more accurate estimates of impact. Our methodology was applied to estimate 9 environmental impact categories of 88 chemicals using 17 structural descriptors and 15 thermodynamic properties. Numerical results show that the life cycle impacts can be predicted with relative errors ranging from 18 to 45% depending on the specific indicator. Furthermore, the inclusion of thermodynamic properties in addition to molecular descriptors improves greatly the quality of the estimate, with relative errors dropping from 22 to 18% when moving from a molecular-based to a combined-based estimation.
1 G. Wernet, S. Hellweg, U. Fischer, S. Papadokonstantakis and K. Hungerbühler, Environ. Sci. Technol., 2008, 42, 6717–6722.
Extended Abstract: File Not Uploaded