Metamodel-based design optimization of structural one-way slabs based on deep learning neural networks to reduce environmental impact

This article presents a methodology for the construction and use of metamodels with Deep Learning (DL) methods that are useful for making multi-criteria decisions in the design and optimization of one-way slabs. The main motivation behind this research has been to examine the possibilities of improving slab design by including this methodology in future tools, which is capable of calculating thousands of solutions in real time based on the designer's specifications. The process of creating these metamodels begins by developing a database of millions of combinations of slab designs. These combinations are calculated with a heuristic algorithm that provides the following results: rigidity, deflection, cost per square meter, CO2 emissions and embodied energy. Once a database including the entire universe of possible solutions has been created, a metamodel is developed that is capable of “condensing” the implicit knowledge contained in the database. This metamodel is included within a Decision Support System (DSS) that produces thousands of solutions for slabs that all comply with a range of specifications designated by the design plan. Furthermore, the methodology described herein proposes the use of Pareto-optimal solutions and graphic tools to help designers make multi-criteria decisions regarding the solutions that best fit their needs. A case study is presented to illustrate this proposal: optimizing slab design in two buildings according to technical, economic and sustainability criteria. The results indicate that the multi-criteria solutions obtained would entail a significant reduction in both emissions and embodied energy as compared to mono-criteria solutions, without significantly increasing costs.