Improving life cycle-based exploration methods by coupling sensitivity analysis and metamodels
Exploration methods combine parametric energy assessments and data visualization to support building designers at early design stages. When exploration methods come to Life-Cycle Assessment (LCA) and the Global Warming Potential (GWP) assessment, a larger number of input parameters induces a very high computation load. Previous researches suggested using Sensitivity Analysis (SA) to decrease the space exploration thanks to their sampling techniques, and input sensitivities. However, this theoretical framework has almost never been applied to building LCA so far and underline two major issues. Upon SA techniques, which one is most suitable for LCA input specificities? How is it possible to extend the exploration process outside the limits of SA samples? This article addressed these questions thanks to an extensive state-of-the-art, the description of a new method combining Sobol SA and Artificial Neural Network (ANN), and a case study. The Sobol method delivered satisfying results with the computation of quantitative indices. Then, an Artificial Neural Network trained on the data generated by the SA was used to predict the GWP of new design alternatives in a small amount of time, and with a coefficient of determination higher than 0.9. Finally, the proposed method adapted exploration methods to the LCA complexity.
2019-01-01
44
70
84
REVIEWED