Application of neural networks trained with multizone models for fast detection of contaminant source position in buildings

Vladimir Vukovic, Jelena Srebric

    Research output: Contribution to journalConference article

    19 Citations (Scopus)

    Abstract

    This study presents an overview of the development of a novel approach for real-time detection of contaminant source locations in buildings. The approach uses the ability of parallel computational architectures, neural networks, to perform non-linear mapping of indoor contaminant concentration patterns to source locations. Such an approach is inverse compared to the traditionally used techniques for predicting contaminant dispersion from the known source position. The existing realtime prediction methods for contaminant dispersion include (I) utilization of powerful supercomputers and computational fluid dynamics, (2) application of statistical techniques toprecalculated databases of possible contaminant release scenarios, (3) genetic algorithm classification in combination with multizone or outdoor dispersion models, and (4) measuring techniques, such as computed tomography. This study presents advancements from our initial investigation to include more complex applications and neural network generalization. The initial investigation successfully detected a contaminant source within nine indoor zones based on the contaminant concentrations computed by multizone models. This success initiated a generalization of the neural network procedure into the Locator of Contaminant Sources algorithm. Furthermore, the applicability of the developed tool was extended to include a sensor optimization algorithm. Prospective practical applications of the developed algorithms include detection of chemical, biological, or radiological contaminant sources during incidental events, as well as determination of a minimum number and allocation of sensors required for such detection.

    Original languageEnglish
    Pages (from-to)154-162
    Number of pages9
    JournalASHRAE Transactions
    Volume113
    Issue number2
    Publication statusPublished - 1 Dec 2007
    Event2007 ASHRAE Annual Meeting - Long Beach, CA, United States
    Duration: 23 Jun 200727 Jun 2007

    Fingerprint Dive into the research topics of 'Application of neural networks trained with multizone models for fast detection of contaminant source position in buildings'. Together they form a unique fingerprint.

  • Cite this