TY - JOUR
T1 - Fault diagnosis of heating systems using multivariate feature extraction based machine learning classifiers
AU - Gharsellaoui, Sondes
AU - Mansouri, Majdi
AU - Trabelsi, Mohamed
AU - Refaat, Shady S.
AU - Messaoud, Hassani
N1 - Funding Information:
This publication was made possible by NPRP grant [NPRP10-0101-170082] from the Qatar National Research Fund (a member of Qatar Foundation ) and the co-funding by IBERDROLA QSTP LLC. The statements made herein are solely the responsibility of the authors.
Funding Information:
The main contribution of this paper is to merge the advantages of machine learning (ML), statistical feature extraction and selection (FES) using principal component analysis (PCA) and multiscale representation. The developed FDD technique, so-called multiscale PCA (MSPCA)-based ML is developed so that the FES problem is addressed using MSPCA and the classification problem is performed using ML classifiers. These ML classifiers incorporate decision tree [11], support vector machines (SVM) [12], K-Nearest Neighbors (KNN) [13] and Naive Bayes (NB) [14]. The wavelet transformation is an efficient separation of the deterministic characteristics of random noise, so it represents a powerful tool for transforming time domain signals into time-frequency domain. The proposed MSPCA-based ML approaches are developed to increase the reliability and safety of the heating system, to detect, and isolate faults in HVAC systems.Other ML techniques were developed for prediction and classification purposes. These techniques include linear regression [21], Least Absolute Shrinkage and Selection Operator (LASSO) [22], Elastic net [23], decision tree regression Bagging [11], decision tree regression boosting [24], artificial neural networks (ANN) [25], support vector regression (SVR) [12], extreme learning machines (ELM) [26], relevance vector machines (RVM) [27], Kernel ridge regression (KRR) [28], regularized linear regression (RLR) [29] and Gaussian process regression (GPR) [30]. It has been shown in Refs. [31–33] that GPR outperformed other parametric and nonparametric machine learning approaches such as neural networks (NNs) and kernel regression. For most anomaly detection applications, abnormal data are generally insufficient [34]. This problem aggravates as dimensionality rise. As noted in Ref. [35], destructive experiments are usually required to collect positive data in many industrial cases, which can lead to high costs. Furthermore, although supervised algorithms generally have great accuracy in detecting anomalies which have occurred before, their ability to generalize in situations that have never occurred before (“unhappened” anomalies) is poor [36,37]. When there is a lack of sufficiently labeled data, often the case, in reality, the detection of anomalies frequently uses unsupervised approaches. In unsupervised fault detection approaches, normal operating conditions are modelled beforehand and faults are detected as deviations from normal behavior. Various unsupervised learning algorithms have been adopted for this aim, such as the k nearest neighbors, the self-organizing Map (SOM) and other approaches based on clustering [38]. Up to date, many fault detection and diagnosis (FDD) techniques have been developed for the reliability analysis and maintenance decision on HVAC. Authors in Ref. [39] proposed a model-based multi-layer FDD methodology in HVAC systems. By analyzing the residues faults are detected and SVM classifiers are applied for diagnosis. A separate SVM is trained for each of the 4 hypotheses (3 different faulty modes and the healthy mode); Features used incorporate water temperatures, supply air and control signals. Since this approach assumes the availability of fault residues generated by replay, it is difficult to know to what extent it would be successful in a real building, where the replay is not possible. Katipamula et al. detect and diagnose using decision tree the outdoor air economizer faults in Ref. [40]. The decision tree is generated manually based on “rules derived from engineering models of proper and improper air-handler performance”. The decision variables are the air temperatures, building energy, air humidity, damper position, and fan schedules. Several approaches have been proposed in this field. One approach presented in Ref. [41] uses statistical machine learning techniques for FDD. An approach proposed in Refs. [42,43] for FDD of HVAC systems using kalman filter, especially for valve actuator failures. An artificial intelligence approach reported in Ref. [44] for the FDD of an air-handling unit using dynamic fuzzy neural network. In Ref. [45], researchers propose that they can achieve 20%–30% of energy saving by recommissioning malfunctioning HVAC systems. The developed technique is used to detect and analyze faults and anomalies in building systems monitoring.In Ref. [54], the authors considered a data-driven technique for FDI of chillers in HVAC systems. They employed multiway partial least squares (MPLS), multiway dynamic principal component analysis (MPCA), and support vector machines (SVMs) in order to diagnose the faults of interest in the chiller. The simulation of a chiller under diverse fault conditions is performed using a standard chiller simulator from the American Society of Heating, Refrigerating, and Air-Conditioning Engineers (ASHRAE) cite american1989ashrae.This publication was made possible by NPRP grant [NPRP10-0101-170082] from the Qatar National Research Fund (a member of Qatar Foundation) and the co-funding by IBERDROLA QSTP LLC. The statements made herein are solely the responsibility of the authors.
Publisher Copyright:
© 2020 Elsevier Ltd
PY - 2020/7
Y1 - 2020/7
N2 - An essential tool for an effective monitoring of heating systems is the accurate Fault Detection and Diagnosis (FDD). Moreover, the implementation of high-efficiency FDD techniques leads to the reduction of building's energy consumption and the improvement of the comfort level. Thus, this paper proposes an enhanced FDD approach based on multiscale representation, principal component analysis (PCA), and machine learning (ML) classifiers. The main goal of this paper is to merge the benefits of the aforementioned techniques to improve the efficiency of the detection and diagnosis of failures occurring in heating systems. First, a multiscale decomposition (MSD) is used to extract the dynamics of the systems at different scales. The multiscale representation is very suitable for the monitoring of heating systems, which are generally characterized by different dynamics in time and frequency frameworks. Then, the multiscaled data-set are introduced into the PCA to extract more efficient features. Finally, the ML classifiers are applied to the extracted and selected features to address the problem of fault diagnosis. The effectiveness and higher classification accuracy of the developed approach is demonstrated using two examples: synthetic data and simulated data extracted from heating systems.
AB - An essential tool for an effective monitoring of heating systems is the accurate Fault Detection and Diagnosis (FDD). Moreover, the implementation of high-efficiency FDD techniques leads to the reduction of building's energy consumption and the improvement of the comfort level. Thus, this paper proposes an enhanced FDD approach based on multiscale representation, principal component analysis (PCA), and machine learning (ML) classifiers. The main goal of this paper is to merge the benefits of the aforementioned techniques to improve the efficiency of the detection and diagnosis of failures occurring in heating systems. First, a multiscale decomposition (MSD) is used to extract the dynamics of the systems at different scales. The multiscale representation is very suitable for the monitoring of heating systems, which are generally characterized by different dynamics in time and frequency frameworks. Then, the multiscaled data-set are introduced into the PCA to extract more efficient features. Finally, the ML classifiers are applied to the extracted and selected features to address the problem of fault diagnosis. The effectiveness and higher classification accuracy of the developed approach is demonstrated using two examples: synthetic data and simulated data extracted from heating systems.
KW - Fault classification
KW - Fault diagnosis
KW - Feature extraction
KW - Heating systems
KW - Machine learning (ML)
KW - Principal component analysis (PCA)
UR - http://www.scopus.com/inward/record.url?scp=85080064874&partnerID=8YFLogxK
U2 - 10.1016/j.jobe.2020.101221
DO - 10.1016/j.jobe.2020.101221
M3 - Article
AN - SCOPUS:85080064874
SN - 2352-7102
VL - 30
JO - Journal of Building Engineering (JOBE)
JF - Journal of Building Engineering (JOBE)
M1 - 101221
ER -