Author(s): Shu Cheng; Oussama Choura; Camelia Chen; Moez Louati

Linked Author(s): Moez LOUATI, Shu CHENG

Keywords: Leak detection high-rise buildings transient pressure transfer learning CNNs

Abstract: This study explores machine learning-based leak detection for high-rise building water supply systems, addressing key factors like system scale, leak size, fluid-structure interaction, and noise sensitivity. Transfer learning with pre-trained CNNs (e. g., VGGNet, ResNet, EfficientNet) improves detection accuracy by adapting numerical simulation data to real-world experimental data from Tower 4 at the Hong Kong University of Science and Technology. Transient pressure signals from the numerical simulations are transformed into scalograms for model training, while fine-tuning ensures better alignment with experimental data. Preliminary results highlight challenges such as noise interference and the complexity of real-world systems, demonstrating the effectiveness of transfer learning. This work presents a scalable and adaptable framework for leak detection, contributing to efficient water management in high-rise buildings.

DOI:

Year: 2025