Transactions of the Society of Heating,Air-conditioning and Sanitary Engineers of Japan
Online ISSN : 2424-0486
Print ISSN : 0385-275X
ISSN-L : 0385-275X
Current issue
Displaying 1-1 of 1 articles from this issue
Scientific Paper
  • Mengtian HUANG, Shohei MIYATA, Keiichiro TANIGUCHI, Yasunori AKASHI
    2025 Volume 50 Issue 341 Pages 9-16
    Published: August 05, 2025
    Released on J-STAGE: July 20, 2025
    JOURNAL FREE ACCESS

    HVAC (Heating, Ventilation, and Air Conditioning) systems are responsible for approximately 30–50% of a building’s operational energy consumption, emphasizing the need for improved energy efficiency. AFDD (Automated Fault Detection and Diagnosis) helps identify faults that impact system performance, such as sensor errors and equipment malfunctions. Although data-driven, machine learning-based methods are more accurate and reduce reliance on expert knowledge, they typically involve high initial investment. The diversity of HVAC system structures and insufficient information management often result in poor scalability and increased implementation costs. A plug-and-play workflow, which enables automatic adaptation to different HVAC systems without extensive customization, is therefore essential for AFDD application development. To address these challenges, our research proposes a unique workflow for generic data-driven AFDD application development. This workflow is based on Brick Schema, an open-source metadata schemas designed to standardize and structure HVAC system information. In the workflow, component-level simulation models of each piece of equipment are generated using equipment information provided by a Brick model of the subject system. These models can reproduce different fault patterns under various fault conditions. The simulated fault data is used to train neural networks (NNs) for automated fault detection and diagnosis. Model generation and parameter optimization are performed autonomously via SPARQL queries that access contextual information from the Brick model. The diagnostic performance and generality of the proposed method are evaluated through a case study involving two different HVAC systems: a primary circuit of a factory consisting of two chillers and two cooling towers, and a secondary circuit within an experimental residence including an AHU (Air Handling Unit). The validation highlights both the advantages of using metadata models for data-driven development and identifies areas for future research.

    Download PDF (1375K)
feedback
Top