نوع مقاله : مقاله پژوهشی
نویسندگان
1 کارشناسی ارشد فناوری معماری، دانشکدۀ معماری و شهرسازی، دانشگاه شهید بهشتی
2 دانشیار دانشکدۀ معمارى و شهرسازى دانشگاه شهید بهشتى
3 استادیار دانشکدۀ معماری و شهرسازی دانشگاه شهید بهشتی
چکیده
کلیدواژه‌ها
عنوان مقاله [English]
نویسندگان [English]
Background and objectives: Estimating and determining the sizes of structural components in the early stages of design is always one of the important issues architectural designers are involved with. It is largely based on personal experience carried out by people without engineering background. This is highly erratic and inefficient, forcing designers to rely on assumptions. This, in turn can result in reworkings and increased costs. On the other hand, whilst a large variety of analyses and design methods are used to determine sizes, there is no decision-making method for early design stages. it is not possible, therefore, to consider options that require precision at this stage. The main goal of this research is to obtain a method and a tool to estimate the approximate f structural components’ sizes without the need to use time-consuming and complex formulas or the use of simulation software: a tool that can reduce the decision-making gap between the designer and the structural engineer and prevent any possible subsequent lawsuits.
Materials and methods: In general, structural decision-making and design involve many variables such as choosing the type of structural system, choosing the type of materials, appropriate structural profile, applicable regulations, and occasionally other considerations. Attemptswere made in this study to include some of these important considerations in a rather limited framework to achieve a decision-making tool that will bring the architect and the structural engineer closer to each other. To start, a set of simulated structures in different states was created, followed by creation of machine learning models with Artificial Neural Network and Support Vector Machine. Consequently, it became clear that the Artificial Neural Network is better than the Support Vector Machine when it comes to predictions for steel structures, with average R2 score of 0.93 and average MAPE of 0.05.In concrete buildings, the Support Vector Machine algorithm performed better with R2 score of 0.93 and average MAPE of 0.06.
Results and conclusion: For about 50 years, there have been various rough diagrams to guide the user, providing very simple and practical information on the initial estimation of sizes based on experience and some analysis. While it is important to know the theories behind these elements, it is also important to be able to quickly assess the consequences of choosing a particular span, column arrangement, or loading. With the aim of helping engineers, and especially architectural designers, this study introduced a tool, that can efficiently be used to take inputs such as land dimensions, span dimensions, number of stories, and type of materials used to produce an initial estimate that is close to what is achieved by structural analysis and design tools. The methodology used in this study can be expanded to include variables such as different materials used in the construction industry, examining complex architectural forms and structures with irregular arrangements of structural members, increasing the number of floors, and examining roofs with different structures.
کلیدواژه‌ها [English]
)