Partager

Approche de détection et diagnostic automatiques des défaillances basée sur les données qui utilise l’émission acoustique dans les systèmes CVC de bâtiment.

A data-driven AFDD Approach using acoustic emission in building HVAC systems.

Numéro : 3130

Auteurs : HUANG J., YANG Z., LI G., WU T., O'NEILL Z., WEN J., CANDAN K. S.

Résumé

Building automatic fault detection and diagnosis (AFDD) technologies have shown great potential for energy savings. Literature on building AFDD research mainly focuses on traditional data available from building automated systems (BAS) or one-time measurements. In this research, we investigate the capability of acoustic emission (AE), a non-traditional data source, to support AFDD in real building heating, ventilation and air-conditioning (HVAC) systems. Experiments were conducted to generate four different AE datasets under different operational scenarios for HVAC systems, where faults were manually injected. The first dataset consists of acoustic data collected from acoustic sensors placed at two different positions (inside/outside) of the same air-cooled chiller under abnormal and normal operations; the second dataset includes acoustic data collected from two identical air-conditioner (AC) outdoor condenser units under abnormal and normal operations; the third one contains acoustic data collected from multiple air diffusers in an experimental residential home under abnormal and normal operations; and the fourth dataset is acoustic data collected under various severity levels of fault conditions occurring in a condenser unit for different time periods. Short-time Fourier Transform (STFT) is used to transform the time series to time-frequency spectrogram, and two different approaches, standard machine learning (ML) and end-to-end deep learning (DL), are used as AFDD strategies to validate the efficacy of AE for the fault detection. For the ML approach, averaged frequency at each time is derived as features fed into random forest classifier; for the DL approach, spectrograms are directly fed into multilayer perceptron. 5-fold cross validation (CV) is repeated 10 times to reduce randomness and avoid overfitting. Experimental results show that AFDD using acoustic data by both the ML and the DL present satisfactory detection performances. For random forest classifier, the averaged fault detection rates are 0.93, 1.00, 1.00 and 0.88 for the four datasets respectively. For multilayer perceptron model, the averaged fault detection rates are 0.97, 1.00, 1.00 and 0.88 respectively. We conclude the use of AE has great potential to support AFDD in the building systems.

Documents disponibles

Format PDF

Pages : 10 p.

Disponible

Gratuit

Télécharger

Détails

  • Titre original : A data-driven AFDD Approach using acoustic emission in building HVAC systems.
  • Identifiant de la fiche : 30032971
  • Langues : Anglais
  • Source : 2024 Purdue Conferences. 8th International High Performance Buildings Conference at Purdue.
  • Date d'édition : 15/07/2024

Liens


Voir d'autres communications du même compte rendu (63)
Voir le compte rendu de la conférence

Indexation