CAL4AHU - Calibration of air handling unit models for fault detection and diagnosis
Over the last two centuries, the world’s population has increased significantly from 1 billion in 1800 to 7.4 billion in 2018. The world’s energy use (150 TWh in 2018) tracks the global population trend. By 2060, both population and energy use are expected to reach 10.2 billion and 280 TWh, respectively.
Buildings consume 40% of the global energy representing more than 35% of CO2 emissions. Studies show that most of the current buildings will still be in use 30 years from now. The academic and industry communities focusing on global energy issues agree that the development of new knowledge to increase the efficiency of the current building energy technologies is essential. However, the introduction of new and sustainable energy technologies requires significant investment in both time and resources to move from current energy technologies to new ones, particularly in existing buildings.
Energy efficiency measures (EEM) in existing buildings provides short-and-medium-term solutions. New knowledge must still be developed and implemented in current building energy systems aimed at reducing the energy use by introducing minimal changes in their physical infrastructure (lowering costs). In other words, existing building energy systems operation must be optimised.
In most buildings, the heating, ventilation and air conditioning (HVAC) systems consume 50% of building energy. Apart from being considerably expensive, HVAC system suffer from systematic operational inefficiencies and faults that can be corrected via low-cost EEM thus making HVAC systems an ideal target for the EEM implementation. These measures could include continuous commissioning, control system optimisation, building insulation improvement, occupancy load redistribution, occupant behaviour optimization and fault detection and diagnosis (FDD).
There is a new trend of developing and using computational models of HVAC systems to support the implementation of EEM’s including model-based FDD and control optimisation. In addition, such computational models and simulation provide excellent tools to evaluate the feasibility of EEM from both technical and economical perspectives as they provide a technology framework to carry out multiple test scenarios at low cost and high accuracy.
One of the main challenges when developing computational models to support EEM’s for HVAC systems is in the calibration process required to guarantee the necessary accuracy of the computational models. Air handling units (AHU) comprise a significant component of HVAC systems and current modelling and calibration approaches for AHU models are not accurate enough and/or demand excessive computational resources particularly for fault detection and diagnosis applications.
In this thesis, a systematic methodology for the automated calibration of AHU computational models for use in FDD is investigated that requires a minimum dataset and produces a trade-off between (model) accuracy and (calibration) complexity. The most relevant calibration methodologies for AHU models are reviewed, and one, based on physical models and machine learning techniques, is proposed. This methodology is implemented along with a first principle-based modelling library to produce a tool that facilitates the automation of the workflow from AHU system modelling and simulation to the model calibration.
The proposed methodology is demonstrated in a real AHU located at roof level on the fifth floor in Cork School of Music, in Ireland as part of the IERC-funded EMWiNS project. The main contribution of this thesis to the EMWiNS project was to develop and provide accurate calibrated computational models of the psychrometric processes occurring in the AHU specifically focusing on temperature changes. These computational models enabled model-based FDD which was compared with a standard AHU Performance Assessment Rules (APAR) implementation. In addition, a tool (CAL4AHU) was systematically developed which allows the automated calibration of the computational models that represent these psychrometric processes.
The computational models and the automated calibration approach are assessed using a recognised validation methodology based on statistical tests. The case study included two model-based FDD methods using qualitative and quantitative approaches.
Experiments and the necessary data collection designed to test the CAL4AHU are described for the AHU under study. The proposed calibration methodology and the associated computational models performed excellently, detecting and isolating a set of faults common in the operation of industrial grade AHUs.