Multi-objective Optimisation for Sustainable Steel Structures Employing Artificial Intelligence

Aim
The power of the approach proposed in this research project can be exercised performing a series of studies focusing on long-span structural systems such as roofs for archaeological sites, airport terminals, concert halls, and train/bus stations. Such structural forms pose a special modelling challenge: they often have large open spaces with unusual shapes and few interior columns, so they rely on systems of triangular space trusses and frames working together to support the load of the building. The use of computer simulation early in the design process - when the share of the building in determined - can have a major impact on embodied energy (Life Cycle Analyses - LCA studies) as well. Careful choice of the geometry and layout of the structure can reduce internal forces and decrease the amount of energy-intensive structural materials required for support.

Accurate service-life prediction of particular long-span members is vital for taking appropriate measures in a time- and cost-effective manner. However, the conventional prediction design models rely on simplified assumptions for typically used members (standard sizes) often leading to inaccurate estimations. Although data driven approaches mainly used today to enhance the performance prediction, they still depend on empirical formulas with many limitations. This project will engage with Artificial Intelligence (AI) methods recently developed for structural engineering applications, as is proving to be an efficient alternative approach to classic modelling techniques, and attempt to reduce the percentage of uncertainty of the results as well as saving significant human time and effort spent in experiments.

Methodology
This study will focus on finding an efficient scheme for the optimisation of both shape and member stiffness distributions in order to create long-span steel members (beams) with higher buckling strength than the one created by just using empirical approaches or even only topology optimisation techniques (recently adopted by practising engineers). Geometric and material characteristics will be optimised for a target to maximise linear buckling load under static as well as dynamic (vibration) actions. Buckling optimisation will be studied for first time at this scale using advanced algorithms of Altair's Hyperworks software tools. Together with machine (supervised) learning Neural Network algorithms (via regression analyses), the limitations of classical prediction models will be demonstrated. Parametric nonlinear finite element (FE) analyses will be performed using ANSYS software to feed the machine learning algorithm with validated data. In addition, both the initial energy required for making structural materials and components as well as the future operational energy will be quantified and compared for the design of energy-efficient structures.

Impact
The knowledge generated by this project can push solutions in interesting and unexpected ways and lead to new building designs and regulations (including low- and high- storey lightweight structures) via the design of long-span lightweight and stiff (support-less) structural members that are high-performance, innovative and architecturally expressive.