Abstract
Faults in Heating, Ventilation, and Air Conditioning (HVAC) systems of buildings result in significant energy waste in building operation. With fast-growing sensing data availability and advancement in computing, computational modeling has demonstrated strong capability to detect and diagnose HVAC system faults, hence, ensuring efficient building operation. This paper comprehensively reviews the state-of-the-art computing-based fault detection and diagnosis (FDD) for HVAC systems. Overall, the reviewed computing-based FDD methods are classified as two major approaches: knowledge-based and data-driven approaches. We then identify multiple important topics, including data availability, training data size, data quality, approach generality, capability, interpretability, and required modeling efforts, along with corresponding metrics to summarize the most updated FDD development. Generally, the knowledge-based approaches are further divided as physics-based modeling, Diagnostic Bayesian Network, and performance indicator-based methods while data-driven approaches include supervised learning, unsupervised learning, and regression and statistics-based methods. State-of-the-art FDD development, remaining challenges, and future research directions are further discussed to push forward FDD in practice. Availability of fault data, capability of existing methods to deal with complex fault situations (such as simultaneous faults), modeling interpretability for data-driven methods, and required engineering efforts for physics-based methods are identified as remaining challenges in FDD development. Improving modeling fidelity and reducing modeling efforts are essential for applying physics-based methods in real buildings. Meanwhile, addressing fault data availability, increasing algorithm adaptability, and handling multiple faults are essential to further enhance the applicability of data-driven FDD approaches.
| Original language | English |
|---|---|
| Article number | 112395 |
| Journal | Renewable and Sustainable Energy Reviews |
| Volume | 161 |
| DOIs | |
| State | Published - Jun 2022 |
Funding
Finally, researchers have combined PCA with clustering approaches and other methods for FDD as well [165]. Typically, PCA was either used before data clustering for data transformation, hence, easier pattern recognition by clustering analysis, or after data clustering such that the performance of PCA models could be improved by targeting certain groups of data. Using PCA as the data pre-processing technique, Yan et al. [166] have utilized PCA to pre-process the data and further identified data clusters for fault identification of AHUs. Also, Li et al. [167] have implemented PCA to project the original sensing data and SVDD (Support Vector Data Description) for FDD. On the other hand, performing data clustering before applying PCA, Du et al. [15] have employed subtractive clustering for identifying system operation conditions in the first place and then PCA to detect sensor faults accordingly. Similarly, Li and Wen [168] have matched the patterns of building operation firstly and built PCA models for fault detection. Li and Hu [169] used Density-Based Spatial Clustering (DBSCAN) to recognize the building operation status and established sub-PCA models for FDD.In addition to using supervised classification, unsupervised learning and regression, and statistics-based approaches alone, researchers have combined these approaches to detect and diagnose system faults [195]. The coupling of regression-based approaches with either supervised classification or unsupervised learning is the most commonly used combination in FDD. Typically, regression models are used to provide operation baseline or support performance indicators calculation. Then, either supervised classification or unsupervised clustering is further applied for fault identification and isolation. For the coupling of regression and supervised classification, Yan et al. [68] have utilized CART (classification and regression tree) based on residual features derived from the comparison between regression model prediction and actual data. Mulumba et al. [50] and Yan et al. [196] have derived autoregressive time series models with exogenous variables for AHU and chiller modeling, respectively, and further combined them with SVM for fault detection and diagnosis. Sun et al. [29] have proposed to model air source heat pump system (ASHP) with the convolution-sequence model and diagnose faults with the convolutional neural network model. Similarly, for the application of deep learning, Eom et al. [30] have utilized the convolutional neural network to detect and diagnose refrigerant charge fault of ASHP systems. Zhou et al. [197] trained fuzzy models and the Neural Network to diagnose chiller faults based on performance indicators calculated from regression models. On the other hand, for the coupling of regression-based modeling and unsupervised learning, Van Every et al. [198] have leveraged the Gaussian process with probabilistic prediction and support vector novelty detector for HVAC FDD. Capable of dealing with unknown faults, Du et al. [199,200] have associated the Neural Network with subtractive clustering to detect abnormalities in AHUs. To deal with sensor faults, PCA was employed to recover sensing data first. The performance indicator residuals (difference between actual and referenced performance indicators) were then inferred to isolate system faults [201]. Except for the involvement of regression-based modeling, unsupervised learning could be coupled with supervised classification for FDD as well. Fan et al. [65] have combined the Neural Network with clustering techniques for FDD of AHUs. Similarly, Du et al. [25] have also decomposed the sensing data with Wavelet analysis first and further employed Neural Network for fault diagnosis. Fan et al. [202] have implemented PCA and SMOTE (synthetic minority oversampling technique) to balance the training dataset in an SVM-based FDD strategy. Han et al. [60] have proposed to detect and diagnose faults of vapor-compression refrigeration systems by coupling PCA with SVM. Li et al. [62] have utilized density-based clustering for data pre-processing, classification and regression training for fault identification, and linear regression for fault isolation. Finally, the unsupervised learning has also been combined with the control chart for FDD of VRF [37] and VAV terminals [203].
Keywords
- Artificial intelligence
- Computing algorithm
- Data-driven
- Fault detection and diagnosis
- Heating
- Machine learning
- Physics-based modeling
- Ventilation and air conditioning systems