Please use this identifier to cite or link to this item: https://scholarbank.nus.edu.sg/handle/10635/211996
DC FieldValue
dc.titleFAULT DETECTION AND DIAGNOSIS FOR CHILLERS USING RECURRENT NEURAL NETWORKS
dc.contributor.authorSIM SER LYN
dc.date.accessioned2021-12-28T02:37:15Z
dc.date.available2021-12-28T02:37:15Z
dc.date.issued2021-12-13
dc.identifier.citationSIM SER LYN (2021-12-13). FAULT DETECTION AND DIAGNOSIS FOR CHILLERS USING RECURRENT NEURAL NETWORKS. ScholarBank@NUS Repository.
dc.identifier.urihttps://scholarbank.nus.edu.sg/handle/10635/211996
dc.description.abstractHeating, Ventilation and Air-Conditioning (HVAC) chiller systems require constant maintenance, where faulty operations can result in untimely costly replacements and an overall drop in HVAC system efficiency of the building. A big part of building energy consumption is attributed to chillers where early fault diagnosis can prevent further deterioration and reduced energy usage. The aim of this dissertation study is to develop a reliable and accurate fault detection and diagnosis tool for centrifugal chillers. This project analyses the various researched FDD methods that are currently utilised, before proposing the selected generic artificial intelligent FDD strategy of a sequential Recurrent Neural Network (RNN) with long short-term memory (LSTM) layers. As compared to other commonly used neural network structures such as ANN and CNN, RNN is a more suitable candidate for chillers due to its ability to identify recurrent relationships and capture sequential information presented by soft faults. In particular, the LSTM architecture retains fault memory over long periods of time, due to the presence of a special ‘memory block’ (Hochreiter & Schmidhuber, 1997). The chiller data used in training the RNN model is derived from ASHRAE RP-1043, in which 7 typical single faults were identified at 4 severity levels. Our RNN model was capable of detecting and diagnosing chiller faults effectively, with a 92.2% accuracy rate, alongside a micro-average Area Under Receiver Operating Characteristic (AUROC) score of 0.955.
dc.subjectRecurrent Neural Network
dc.subjectFault Detection
dc.subjectFault Detection and Diagnosis
dc.subjectNeural Network
dc.subjectHeating Ventilation and Air-Conditioning System
dc.subjectBuilding Maintenance
dc.typeDissertation
dc.contributor.departmentDEPT OF THE BUILT ENVIRONMENT
dc.contributor.supervisorYAN KE
dc.description.degreeBachelor's
dc.description.degreeconferredBachelor of Science (Project and Facilities Management)
Appears in Collections:Bachelor's Theses

Show simple item record
Files in This Item:
File Description SizeFormatAccess SettingsVersion 
Sim Ser Lyn DBE.pdf5.39 MBAdobe PDF

RESTRICTED

NoneLog In

Page view(s)

81
checked on Nov 17, 2022

Download(s)

9
checked on Nov 17, 2022

Google ScholarTM

Check


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.