Novel pattern recognition-enhanced sensor fault detection and diagnosis for chiller plant

Research output: Journal Publications and Reviews (RGC: 21, 22, 62)21_Publication in refereed journalpeer-review

5 Scopus Citations
View graph of relations


Original languageEnglish
Article number110443
Journal / PublicationEnergy and Buildings
Online published8 Sep 2020
Publication statusPublished - 1 Dec 2020


In building application, sensor faults in chilled water system would cause extra electricity consumption or thermal comfort problems. In the computerized building management system, it is important to automatically detect, diagnose and correct different categories of sensor faults. To realise this aim, a novel 2-stage pattern recognition-enhanced sensor fault detection and diagnosis (PRe-SFDD) was formulated. At the first-stage pattern recognition, various featuring patterns were generated through sensor reading datasets from both fault-free and different faulty test cases. At the second-stage pattern recognition, one-day featuring patterns were used to diagnose the sensor faults of positive bias, negative bias, precision degradation and general drift; while 3-day featuring patterns would allow further recognise the drift fault to be positive or negative. Hence, different categories of sensor faults could be automatically detected, diagnosed and corrected through the proposed pattern recognition strategy. For a representative chiller plant, it was found that the successful diagnosis ratio of the 2-stage PRe-SFDD were 97.9%, 100%, 96.4%, 95.4% and 98.1% for positive bias, negative bias, precision degradation, positive drift and negative drift, respectively. In addition, characteristic curves of clustering score values were constructed for correction of the extent of sensor faults.

Research Area(s)

  • Chiller plant, Clustering, Data mining, Fault detection and diagnosis, Pattern recognition, Sensor