Simulation-based benchmarking of VWART for fault detection in heat interface units

In recent years, buildings have become increasingly digitised and monitored. As a result, in many cases, it is found that the energy performance of buildings falls short of expectations. Several studies indicate that this performance gap can be related to the presences of faults in the building’s energy system, which can be responsible for 4-30% of the energy consumption [1]. Therefore, interest in fault detection and diagnosis (FDD) techniques for building energy systems has increased.

However, building energy systems are inherently diverse, encompassing a wide range of components, configurations, and control strategies. This diversity makes it challenging to develop generic, robust, and reliable FDD algorithms. These FDD algorithms rely on historical and/or case-specific operational data for their development. Consequently, they may not fully capture the nuances of the different system configurations [1].

In this research, we address the advantages of using a simulation-based approach for the development of reliable, robust, and generic FDD algorithms. More specifically, the approach is applied to a type of collective residential heating system called the combined heat distribution circuit (CHDC), also known as the flatstation concept. By simulating various configurations of the CHDC and common fault scenarios identified in previous research [2], operation conditions are emulated, and a fault detection strategy involving key performance indicators, such as the Volume-Weighted Average Return Temperature (VWART), and thresholds are developed.

The most impactful system parameters on the performance indicators, such as the temperature regime, type of components, controls, and type of occupant, are explored. Based on this knowledge, statistical analysis, clustering techniques, and logistic regression are used to develop the FD strategy and define proper thresholds. Subsequently, the simulation-based monitoring approach is assessed by applying the methodology for one the faults encountered in previous research [2] using data from residential heat meters. Finally, we reflect on the applicability of the simulation-based approach and future improvements.