Building drainage system risk assessment using convolutional neural networks

Abstract

Building drainage systems (BDS) have become essential infrastructure in modern buildings as urban life continues to expand. However, a malfunctioning BDS significantly increases the risk of harmful consequences as potential transportation routes for foul gases and pathogens to leak indoors. Constructing a risk assessment system for BDS is essential and will benefit the public health in residential area. This research focus on introducing an alternative approach on BDS risk assessment via Convolutional Neural Network (CNN). The rapid advancement in artificial intelligence enables risk assessment through computer-based image recognition. A total of 500 cases are utilized in deep learning, resulting in the development of a classification system for identifying high-risk BDS. Based on the given dataset, the classification system achieved a maximum accuracy of 76.00%. Three influential parameters will be examined to study the impact on the model’s performance: dataset size, positive and negative case enhancement, and manually pre-categorized datasets. The developed solution has the potential to be applied to other research areas in the architecture field, where design diagrams serve as the primary medium for conveying construction information. It is expected to enable more efficient large-scale risk assessments and handle a greater volume of evaluations compared to traditional human inspection.

Practical Application:This research aims to construct an alternative method to evaluate the risk value of a given building drainage system. By introducing Convolutional Neural Network to the realm of drainage system, an image-based inspection system is developed. Merely by inputting plumbing riser diagrams, the system is able to distinct high-risk system from stable system with an overall accuracy of 76.00%. It is expected that with the implementation of AI and the handiness of computers, the building drainage risk assessment system could handle a larger number of cases more efficiently, moreover, the equality and objectiveness are more aligned.