Fault detection and diagnosis for buildings and HVAC systems using combined neural networks and subtractive clustering analysis

Various faults occurred in the buildings and heating, ventilation and air conditioning (HVAC) systems usually lead to more energy consumption and worse thermal comfort inevitably. The soft faults such as the sensor biases are difficult to discover in the real buildings. A robust diagnostics tool is presented to improve the energy efficiency and thermal comfort of buildings through removing various faults. The combined neural networks, integrating the basic neural network and auxiliary neural network, are developed to detect the abnormities in the air handling unit that is the widely used in commercial buildings. As a data mining technology, clustering analysis is used to classify the various faulty conditions adaptively in the buildings in this paper. Through subtractive clustering analysis, the different faults can be separated into different space zones in the data space. Besides the known faults in the library, the new unknown faults can be recognized and complemented into the faults library adaptively. The fixed biases, drifting biases and complete failure of the sensors and chilled water valve faults commonly occurred in the buildings are validated in this paper.