Author : Filippo Ranalli
Release : 2021
Genre :
Kind :
Book Rating : /5 ( reviews)

GET EBOOK

An Artificial Intelligence Framework for Multi-disciplinary Design Optimization of Steel Buildings - read free eBook in online reader or directly download on the web page. Select files or add your book in reader. Download and read online ebook An Artificial Intelligence Framework for Multi-disciplinary Design Optimization of Steel Buildings write by Filippo Ranalli. This book was released on 2021. An Artificial Intelligence Framework for Multi-disciplinary Design Optimization of Steel Buildings available in PDF, EPUB and Kindle. This dissertation describes the development of a new AI-driven multi-disciplinary structural design optimization (MDSDO) software framework to automate the design of the structural sub-systems of typical steel buildings, comprised of the composite floor system, the lateral system, and their respective connections, adopting a minimum-cost objective function, and meeting the relevant strength, drift, vibration, constructability, and ductility constraints per U.S. building code, design manuals, and industry standards. The cost objective function accounts for material, fabrication, and erection rates, and is evaluated through estimation models assembled with guidance provided by general contractors and steel fabricators in the U.S. who agreed to contribute up-to-date cost data, which is typically difficult to access for researchers. The core challenges of developing such a design tool revolve around the size of the optimization problem in terms of number of constraints, objective complexity, and design domain, as well as the need for software modularity to target specific structural sub-systems, while striving for the ease of adoption of this technology for new design in industry. Computational scalability to full-sized structures is also a key requirement to address, together with designing a way to incorporate accurate project-specific cost data in order to find minimum-cost solutions. In the proposed framework, the approach adopted to solve the optimization of steel buildings is to decompose the full design problem into three decoupled software modules in Python which are then solved individually with divide and conquer algorithmic solutions, coordinated through a multi-disciplinary architecture. The research approach focuses on the separate development and testing of each of the modules of the architecture on smaller benchmark sub-systems, progressively building software infrastructure to connect the stand-alone optimizations so as to run them collaboratively, with a focus on full-building design optimality, algorithmic run-time, and scalability. Both classic and more recent Artificial Intelligence methods play a key role across the optimization modules on the choice and design of the algorithms, as well as on the software engineering aspects for augmented computational efficiency. The optimization of each module is preceded by a series of automated pre-processing steps to parse and store the geometric, structural, and loading features of the Building Information Model (BIM) or Finite Element Analytical (FEA) model. Subsequently, the framework modules, one for each of the key structural sub-systems, may be run sequentially to automate the full design end-to-end, or as stand-alone logic to optimize select sub-systems in the structure. The first module performs the composite floor system design, in which each girder, filler beam, and column subjected to construction and ultimate gravity loads is sized to meet a series of constraints. Through dynamic programming, the algorithm exhaustively explores all feasible solutions in an engineer-specified discrete domain of slab and concrete properties, wide flange sections, degree of composite action, number of studs, camber, and shoring, and ultimately selects the most economical option. The second logic component automates the design optimization of the lateral frame system to withstand the seismic and wind load demands per building code regulations, using an innovative energy-based algorithm to envelope the critical load combinations and determine the cost-optimal sizes. Lastly, the third stand-alone software module of MDSDO consists of a connection optimization engine, whose functionality is to size and detail each individual gravity and lateral connection based on the load demands using non-linear solvers, yielding a design with better economy than the traditional schedule-based approach by which each connection is conservatively sized based solely on the geometry of its connecting elements. Validation of the optimization framework is illustrated by running each of the sub-system optimization modules of the multi-disciplinary architecture sequentially on an existing 4-story steel building in California subjected to gravity, wind, and seismic loading, with fine-tuned cost parameters and an analytical model assembled with guidance from the general contractor and structural engineer on the project. When compared to the original design, the results of the MDSDO show total steel cost savings of approximately 10%, with individual sub-system savings between 5 and 55%, amongst which the gravity connections show the largest potential for savings, and with a design run-time in the order of a few hours versus several weeks. The most cost-impacting structural feature differences between the original design and the optimization output are the reductions in shear studs, shear bolts, stiffener plates, shear plate dimensions, weld volume, and moment frame weight. A parametric study is furthermore conducted to understand to what extent varying slab and deck heights, composite moment frame girder action, and concrete type from the original design choices might affect the total cost. The most important theoretical improvement of the MDSDO framework over existing structural optimization methods in the literature is the capability to scale to full-size steel buildings, accounting for the full set of relevant strength, stiffness, vibration, ductility, and constructability prescriptions. These constraints are interpreted from U.S. codes and manuals, re-formulated with compact mathematical notation, and subsequently expressed in computer code. Scalability is achieved by reducing the design domain to realistic discrete values, and by designing new efficient algorithms for each of the gravity, lateral, and connection sub-systems. Moreover, the MDSDO relies on a cost objective function which is more complex and adaptable than the classic weight minimization approach, as it accounts for material, labor, and equipment rates sampled from industry data, which are used to estimate each of the detailing components of the structural sub-system designs. The modularity, interpretability, and ease of use of the MDSDO favors its applicability to new design of commercial, medical, and residential steel buildings with minimal additional effort from the engineer's part. The MDSDO is expected to impact industry by providing total installed cost reductions on steel frames ranging between 9% and 20%, the lower bound of which is deduced from the case study results of this dissertation, while the upper bound is estimated for projects whose budget is on the higher end of the spectrum. Moreover, the MDSDO is able to generate a feasible and cost-optimal design of a new building in the order of a few hours, thus allowing the engineer to save weeks of design time, resulting in a rather competitive edge, and when appropriate, the added ability to pursue the more advanced and time-consuming aspects of Performance-Based Engineering (PBE). The MDSDO also provides an additional layer of safety by automating the design checks, while also helping prevent issues with constructability by enforcing a series of constraints on relative member sizes. The author predicts that the adoption of the MDSDO in elastic design in industry may provide benefit to the owner, structural engineer, and general contractor, while potentially reducing the environmental impact of construction by identifying lighter solutions with reduced construction time.