Abstract

The performance of the HVAC systems in buildings tends to decrease after few years of operation. Equipment and sensors degradation lead to remarkable wastes of energy and money, as well as to the increase of building occupants thermal discomfort. HVAC ongoing commissioning (OCx), the continuation of HVAC commissioning well into the occupancy and operation phase of a building life, has been recognized as a cost-effective strategy to reduce energy wastes, equipment degradation and thermal discomfort. Building Automation Systems (BAS) collect and store huge amount of data for the purpose of building systems control. Those data represent a golden mine of information that can be used for the OCx of the building HVAC systems.
This research work develops and validates new methods and models to be used for the OCx of HVAC systems using BAS measurements from commonly installed sensors. A Fault Detection and Identification (FD&I) method for chillers operation, and several virtual sensor models for variables of interest in Air Handling Units (AHUs) are presented.
A FD&I method based on Principal Components Analysis (PCA) has been developed and used to detect abnormal operation conditions in an existing chiller operation and identify the responsible variables. The proposed FD&I method has been trained using measurements from summer 2009, and then used to detect abnormal observations from the following seven summer seasons (2010-2016). When the detected abnormal observations were replaced with artificially generated fault-free data, the proposed FD&I method did not detect any abnormal value along those artificially faulty-free variables. In summer 2016 the building operators changed several HVAC system operation set points, the FD&I method was effective in detecting almost 100% of the observations and properly identifying those variables whose set point was changed.
For two different operation modes of an AHU several virtual outdoor air flow meters have been developed and the predictions have been compared against short-term measurements using uncertainty analysis and statistical indices. Three models have been investigated when the heat recovery coil was off. Results showed that the model with the simplest mathematical formulation was the most accurate, with the lowest value of uncertainty. When a heat recovery coil at the fresh air intake was on, two virtual flow meters have been developed to predict the outdoor air flow rate without the need of additional sensors. Both the models predicted the outdoor air ratio with good statistical indices: the Mean Absolute Error (MAE) was 0.015 for model a and 0.016 for model b.
Three methods for the virtual measurement and/or calibration of air temperature and relative humidity have been developed for different AHU operation modes. These methods are different in terms of modelling strategy, information needed and technical knowledge required for implementation. For instance, results from the correction of the faulty measurements of the outdoor air temperature along a 24 hours period using Method A showed a high virtual calibration capability: MAE = 0.2°C and the Coefficient of Variation, CV-RMSE = 1.7%.
A new definition of virtual sensor is proposed at the end of this research work. From a review of publications on virtual sensors for building application, the two most recurrent reason for the implementation of virtual sensor models (costs and practical issues) have been highlighted and integrated into the proposed new definition.