Master's Thesis Johanna Wilkens


Fault Modelling for Components of Air Handling Units

component faults Copyright: EBC Overview of the selected component faults (middle) divided into fault categories (left) showing their cause (colours) and their impact (right)

Suboptimal performance of building energy systems increases energy consumption up to 18 % compared to a fault-free operating system. The increased energy consumption is caused by undetected faults, which also lead to comfort limitations. Automatic fault detection and diagnosis (FDD) could enable preventive maintenance making it possible to predict the best time for the maintenance and repair process depending on the actual condition of the energy system. Even though different FDD algorithms have been developed, a lack of test and training data complicates their implementation in building systems. The search for complete fault datasets in real measured data and the cost intensive and elaborate execution of experiments to create this data is time consuming and not all fault patterns can be quantified with these methods.

This paper introduces a fault model library developed for simulations with the AixLib, an open source Modelica library for creating these missing fault data sets for FDD algorithms. The faults are selected according to their frequency of occurrence and their impact on energy consumption. The models are dynamic, component-oriented and flexibly implementable in different building models. The results of this paper in comparison with the reference case of a fault-free operation and with publications of quantified faulty behaviour proof the realistic effect of the fault models on the energy system.

Since the fault models can be easily implemented in different models, the fault library enables the creation of realistic building specific data representing a faulty operational state. In addition, the simulations emphasize the importance of FDD methods since each of the simulated failures either increase energy consumption or reduced comfort conditions.