Spatial layout design in architecture through deep neural network algorithms
الموضوعات : Space Ontology International Journal
Mahsa Hamouni
1
(
Department of Architecture, Faculty of Art & Architecture, South Tehran Branch, Islamic Azad University, Tehran, Iran
)
حسین سلطانزاده
2
(
گروه معماری، دانشکده معماری و شهرسازی، واحد تهران مرکزی، دانشگاه آزاد اسلامی، تهران، ایران.
)
Seyed Hadi Ghoddusifar
3
(
استادیار معماری دانشگاه آزاد اسلامی تهران جنوب
)
muharram Mansoorizadeh
4
(
Department of Computer Science, Faculty of Engineering, Bu-Ali Sina University, Hamedan, Iran
)
الکلمات المفتاحية: Spatial layout design, Generative design, Deep learning, GAN,
ملخص المقالة :
The architectural layout design problem is a well-known algorithmic problem in computer-aided architectural design. It is the assignment of discrete space elements to their corresponding locations while attempting to satisfy geometrical and topological goals in their layout. This activity is exploratory and iterative in nature. This level of complexity has encouraged researchers to explore this problem by taking advantage of computational strength for predicting space layouts. Numerous methods have been proposed to solve the space layout design problem. Each one can be viewed as a rule-based strategy that makes an effort to simulate space layout design using some high-level rules. However, in addition to the quantitative criteria that may be tested and assessed in a logical process, numerous non-quantitative elements exist. These qualitative criteria are frequently based on a variety of factors and are challenging to describe; thus, hard-coding them would not be possible or effective. It would be best if the program could learn these rules from existing examples. Some potential solutions can be found in the rapidly expanding field of machine learning. The generative adversarial network (GAN), is a machine learning algorithm that has shown remarkable outcomes in the development of 2D designs. In this paper, GAN was applied to generate space layouts with given boundaries. An special training dataset is generated to train the model. The training dataset consists of 660 existing apartment layouts in Hamadan. After completing the process of training model, the space layouts of 12 new apartments were tested. Also, the performance of the model was evaluated based on both quantitative and qualitative methods. The results show that the proposed model has been successful in generating space layout plan from a predefined boundary. We propose this model to facilitate the architectural layout design process especially in projects with complex topological constraints.
