An efficient sensor and thermal coupling fault diagnosis methodology for building energy systems

The simultaneous occurrence of sensor fault and thermal fault results in misdiagnosis of the current single-type fault diagnosis methods, and significantly reduces the diagnosis accuracy of the two types of faults. Hence, this study proposes a methodology that can diagnose sensor and thermal coupling faults for building energy systems. Four sub-models are developed to achieve coupling fault diagnosis, i.e., sensor fault detection model, sensor fault diagnosis model, sensor fault reconstruction model and thermal fault diagnosis model. A voting mechanism is proposed to efficiently separate sensor faults and thermal faults which eliminates the impact of the two types of faults on each other's diagnosis. It develops candidate features that are linearly correlated with thermal fault diagnostic features and uses Euclidean Distance between these features to detect and locate sensor faults. Four sensor faults coupled with seven thermal faults are investigated on an experimental chiller. Results show that sensor faults resulted in 13.7%–53.4% decline in the accuracy of thermal fault diagnosis. After the sensor bias faults were reconstructed, the thermal fault diagnosis accuracy increased 38.1%–51.2%. And it reverted to over 90% after the sensor stuck faults were restored. This work can provide guidance for sensor and thermal coupling fault diagnosis in buildings.