(519e) A Hybrid Clustering Approach for Fault Detection in HVAC Systems

Heating, ventilation, and air conditioning (HVAC) systems, used in modern buildings to maintain air quality and thermal comfort, are responsible for a significant amount of energy consumption (approximately 40% of the total global energy consumption). The occurrence of different faults in these systems can lead to higher energy use in addition to reduced comfort levels. This has motivated research on designing/developing efficient fault detection and isolation (FDI) techniques for HVAC systems [1]. In one direction, first-principles-based residuals are developed to capture the difference between observed and expected behavior to detect possible faults [2]. In addition, data-driven models, such as artificial neural networks (ANN), have attracted considerable attention for FDI designs due to existing challenges in developing and maintaining first-principles models [3].

Hybrid approaches, developed by integrating data-driven methods with first-principles knowledge, have been gaining significant attention recently. These approaches have shown superior performance compared to purely data-driven and first-principles modeling techniques [4], so they can bring potential value to FDI designs for HVAC systems. In an earlier contribution, a hybrid-based fault detection strategy was proposed by integrating the first-principles residuals with measured data to build a principal component analysis (PCA)-based fault detection algorithm [5]. This method is developed based on prior knowledge of normal operating conditions (data samples corresponding to normal conditions are available). In situations where no labels (related to normal and fault operating conditions) are available, it becomes useful to leverage an appropriate unsupervised clustering technique to find underlying patterns in the data. This can facilitate the development of an efficient FDI framework.

Motivated by the above considerations, a hybrid-based clustering approach is proposed for HVAC fault detection. First-principles-based residuals are first integrated with the data (measurements in different locations of the air handling unit (AHU)). PCA and autoencoder (AE) are then utilized to extract latent features. These features are used to perform clustering analysis using different methods such as K-means, density-based spatial clustering of applications with noise (DBSCAN), and ordering points to identify the clustering structure (OPTICS). This method is compared to a purely data-driven approach, where only measured data are used for feature extraction and development of the clustering methods. The superior performance of the proposed hybrid clustering method is demonstrated using simulation tests for different fault scenarios.

References

[1] Mirnaghi, M.S. and Haghighat, F., 2020. Fault detection and diagnosis of large-scale HVAC systems in buildings using data-driven methods: A comprehensive review. Energy and Buildings, 229, p.110492.

[2] Seem, J.E. and House, J.M., 2009. Integrated control and fault detection of air-handling units. HVAC&R Research, 15(1), pp.25-55.

[3] Shahnazari, H., Mhaskar, P., House, J.M. and Salsbury, T.I., 2019. Modeling and fault diagnosis design for HVAC systems using recurrent neural networks. Computers & Chemical Engineering, 126, pp.189-203.

[4] Ghosh, D., Hermonat, E., Mhaskar, P., Snowling, S. and Goel, R., 2019. Hybrid modeling approach integrating first-principles models with subspace identification. Industrial & Engineering Chemistry Research, 58(30), pp.13533-13543.

[5] Hassanpour, H., Mhaskar, P., House, J.M. and Salsbury, T.I., 2020. A hybrid modeling approach integrating first-principles knowledge with statistical methods for fault detection in HVAC systems. Computers & Chemical Engineering, 142, p.107022.