Journal of Infrastructure Intelligence and Resilience 4 (2025) 100167 - Fire sprinkler layout drawing automation using pix2pixHD GAN

Oct 24, 2026

Fire sprinkler system design is widely required in modern buildings, but manual drawing preparation remains slow, labor-intensive, and error-prone. This study presents an intelligent framework that automates fire sprinkler layout drawing generation. A dataset of 120 sprinkler design drawings trained a pix2pixHD GAN, enabling automatic sprinkler placement for new random architectural floorplans with 99.5% protection coverage. The GAN produces 13% fewer sprinklers than professional engineers while improving design-drawing preparation efficiency by 76%, supporting code-compliant safety design with better cost-benefit outcomes.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages15

Reading Time38 min

AccessFree to download

Introduction

Background and challenges in manual fire sprinkler drawing

Complexity of code compliance and error sources

Journal of Infrastructure Intelligence and Resilience 4 (2025) 100167 Contents lists available at ScienceDirect Journal of Infrastructure Intelligence and Resilience journal [homepage:](homepage: www.journals.elsevier.com/journal-of-infrastructure-intelligence-and-resilience)[ www.journals.elsevier.com/journal-of-infrastructure-intelligence-and-resilience](homepage: www.journals.elsevier.com/journal-of-infrastructure-intelligence-and-resilience) | AI-powered automatic building floorplans | | design of fire sprinkler layout for random | | | | --- | --- | --- | --- | --- | | Yanfu Zenga , Xinyi Liu b, Yifei Dinga, Zhe Zheng c,d, Tianhang Zhang a, Xinyan Huang a,* , Xinzheng Luc,**
a Dept. of Building Environment and Energy Engineering, Hong Kong Polytechnic University, China b Arup International Consultants, Shenzhen, China
c Dept. of Civil Engineering, Tsinghua University, Beijing, China
d Dept. of Investment and Technology Innovation, 8 Wuliangye Yibin Co., Ltd., Yibin, China | | | | | | A R T I C L E I N F O | | | A B S T R A C T | | | Handling Editor: Ms. Yanwei Li | | | Fire sprinkler system is a commonly designed safety provision in modern buildings, yet the current manual drawing preparation process is burdened by time-consuming tasks, heavy workloads, and human errors. This study introduces an intelligent framework aimed at automating the drawing preparation process for fire sprinkler layout. A database of 120 sprinkler design drawings was compiled to train a pix2pixHD generative adversarial network (GAN). After training, the GAN model can generate sprinkler placement with a protection coverage of 99.5% for new and random architectural floorplans. Apart from ensuring code-compliant design, the total number of sprinklers designed by GAN is 13% lower than those arranged by professional engineers. By adopting this intelligent method, the time needed for design drawing preparation can be saved by 76%, and the costbenefit of the sprinkler design can be improved by using reasonable fewer sprinklers. | | | Keywords:
Smart design
Fire services system
Building fire
Deep learning
Generative adversarial network | | | | | 1. Introduction Building fire service/protection systems play essential roles in safeguarding occupants and properties in the built environment, so they are strictly regulated and require careful design and installation (Stollard, 2014). In the current design process, architects take the first step in proposing basic building drawings like building shape, floorplans and functions to deliver the fundamental design concepts and spatial arrangement of the building. After that, building services engineers or fire engineers are responsible for incorporating fire service systems as necessary to protect the building, including sprinklers, fire detectors, emergency lighting, smoke ventilation, and other elements regulated by fire codes. These engineering drawings depict the placement of fire protection equipment throughout the building, and also indicate the positions of fire emergency exits, stairwells, and other escape routes, as well as the configuration for fire compartmentation (Zeng et al., 2024a). Despite the development of the digital design platforms such as computer-aid design (CAD) and building information modelling (BIM)(Chen et al., 2021; Sch¨onfelder et al., 2024; Wang et al., 2015; Zhou et al., 2012), the preparation of the engineering drawings still heavily relies on the engineers on manual, such as the distribution of sprinkler heads and calculation of egress travel distance. This manual design process suffers from prolonged time consumption and low efficiency. Overruns of time and cost are commonly seen, and statistics show that although the construction industry represents 13% of global GDP, it has experienced only a 1% annual profit increase over the past 20 years (McKinsey, 2020), making it one of the slowest-growing and least profitable sectors worl

FAQs

What problem does the study address in fire sprinkler layout design?

Manual fire sprinkler layout drawing preparation is time-consuming, workload-heavy, and prone to human errors, followed by additional manual code-compliance checking effort.

How is the proposed method trained and what model is used?

A dataset of 120 sprinkler design drawings is compiled to train a pix2pixHD generative adversarial network (GAN).

What performance improvements does the GAN achieve compared with professional engineers?

For new random floorplans, it reaches 99.5% protection coverage and designs with 13% fewer sprinklers, while reducing time needed for drawing preparation by 76%.

Oct 24, 2026

Upload by Jordan Avery