Back to results list
Show full item record
Please use this identifier to cite or link to this item:
|Title:||Fault detection and diagnosis methods for building HVAC systems||Authors:||Zhao, Yang||Degree:||Ph.D.||Issue Date:||2013||Abstract:||Faults in Heating, Ventilation and Air-conditioning (HVAC) systems would lead to uncomfortable indoor environment, poor indoor air quality, occupant complains and energy waste. Fault detection and diagnosis (FDD) tools are helpful to detect and isolate faults timely. Therefore, they are essential for reliable indoor environment control, saving maintenance efforts, and eliminating the associated energy waste. There is a growing interest in developing FDD tools for HVAC systems. Over the last decades, a considerable amount of FDD methods have been developed for chillers, air handling units (AHUs) and variable air volume (VAV) terminals. However, there is still a lack of reliable, affordable and scalable solutions. The main objective of this PhD project is to develop enhanced and reliable FDD methods for HVAC systems in buildings. Firstly, a comprehensive literature review is made. Then, the methodologies of four FDD methods are proposed for different applications. The first two methods are improvements of conventional FDD methods. The latter two methods are new ones. A simplified model-based FDD method with its customization tool is proposed. It is preferable when there are limited fault-free data to train models. The basic idea is to identify model parameters using limited training data, and then to generate benchmarks for fault detection using the calibrated models. It provides good applicability and convenience for actual applications. Based on this method, a simplified FDD strategy for centrifugal chillers is proposed. It adopts a simplified physical chiller model which can be calibrated using very limited operation or performance test data. Four schemes are developed to identify chiller model parameters based on available information and data from tests or from manufacturers. A new semi-physical sub-cooling model is adopted by the chiller model. Comparisons are made with four typical conventional FDD strategies using ASHRAE RP-1043 experimental data. The results show that this strategy has much higher detection and diagnosis ratios. An enhanced statistical FDD method is proposed to enhance incipient fault detection and diagnosis performance of the conventional gray-box model-based methods. It is suitable when measurements are sufficient. Support vector regression (SVR) algorithm is adopted to improve accuracies of reference performance index (PI) models. It is a non-linear regression approach which is based on structural risk minimization from statistical learning theory. Exponentially weighted moving average (EWMA) control charts are introduced to detect faults in a statistical way to improve the ratios of correctly detected points. The EWMA control charts reduce the Type II error ratios through taking into account the time series information using the weighting factor. This method is applied to a centrifugal chiller FDD strategy and a system-level FDD strategy respectively. Results show that the chiller FDD performance is improved significantly, especially at low severity levels. For example, in the case of condenser fouling, the proposed strategy achieved the ratios of correctly diagnosed points of 7.7%, 45.2%, 60.7% and 100.0% at four severity levels (SL-1 to SL-4) respectively at the confidence level of 99.73%. Using the conventional gray-box model-based method, this fault could not be correctly diagnosed at level SL-1, SL-2 and SL-3. The application on system-level fault detection is evaluated on a simulated commercial building at four severity levels and two uncertainty levels. Similar improvements are also observed.
A new pattern recognition-based FDD method is proposed using support vector data description (SVDD) algorithm. It is suitable when fault data are available. This method transforms the FDD problem as a typical one-class classification problem. The task of fault detection is to detect whether the process data are outliers of the fault-free class. The task of fault diagnosis is to find to which fault class do the process data belong. It overcomes shortcomings of available pattern recognition-based FDD methods in HVAC field. Evaluations are made using ASHRAE RP-1043 experimental data. It shows more powerful FDD capacity than other pattern recognition-based FDD methods, e.g., multi-class SVM-based FDD method and PCA-based fault detection method. A generic diagnostic Bayesian network (DBN)-based FDD method is proposed to simulate the actual diagnostic thinking of HVAC experts. It has better performance than other FDD methods when the diagnostic information is uncertain and incomplete. It benefits to allow merging different types of knowledge and information from diverse sources. The structure of the DBN is a graphical and qualitative illustration of the intrinsic causal relationships among causal factors, faults and fault symptoms. The parameters of the DBN represent the quantitative probabilistic relationships among them. It is effective in diagnosing faults based on uncertain, incomplete and conflicting information. DBNs are developed for chiller FDD and VAV terminal FDD respectively. A three-layer DBN is developed to detect and diagnose component faults in a 90-ton water-cooled centrifugal chiller as described in ASHRAE RP-1043. Only using measurements from building management system (BMS), the DBN has similar accuracy as rule-based chiller FDD methods when BMS measurements are complete. When BMS measurements are incomplete, the DBN still provides meaningful fault believes but rule-based chiller FDD methods fail to work. The diagnosis ratios are increased when evidences of nodes in additional information layer are used. Refrigerant overcharge and non-considerable gas can be correctly diagnosed with the help of evidences of nodes in additional information layer. A DBN is developed to detect and diagnose faults of the pressure independent VAV terminals in an office building located in Hong Kong. It is evaluated through conducting the ten typical VAV terminal faults on a dynamic simulation platform of an office building. All faults are correctly diagnosed with high confidences.
|Subjects:||Fault location (Engineering)
Hong Kong Polytechnic University -- Dissertations
|Pages:||xxix, 226 leaves : ill. ; 30 cm.|
|Appears in Collections:||Thesis|
View full-text via https://theses.lib.polyu.edu.hk/handle/200/7356
Citations as of May 28, 2023
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.