Technical capacity for development of sequentially training neural networks in industrial architecture design

This paper aims to analyze the possibilities of developing sequentially learning neural networks in the context of industrial architecture design. The study revealed a lack of empirical data on the application of neural network methods to automate the creation of architectural models of industrial objects in the modern information environment. Moreover, available studies are limited to a general description of the problem, offering no comprehensive solutions. The applied methods are based on training neural networks using images; however, they are adapted for modeling autonomous behavior in architectural design. The proposed standards stipulate that algorithmically controlled neural networks serve as a tool to facilitate subsequent design, rather than being a source of knowledge. Theoretical and experimental studies have been conducted to develop new engineering design techniques to create a model of autonomous behavior in architecture. The results demonstrate that the integration of engineering design with neural network methods contributes to the standardization of design processes, increasing functional efficiency and quality of realized objects. The developed methods can be used at early design stages to switch from conventional visual approaches to the use of algorithmically controlled models, which offer opportunities for engineering design automation.