A hybrid clustering multi-source fault diagnosis method for chiller temperature sensors

Abstract

ABSTRACT Sensor faults have been observed to negatively impact the operation of the HVAC system. Among these faults is the complexity of multi-source sensor faults, which may result in fault confusion due to multiple fault points and different fault patterns. This paper proposes a fault diagnosis model applicable to single- and multi-source faults of HVAC system sensors. Based on the distribution patterns of chillers sensor data, the ensemble empirical mode decomposition soft threshold denoising Gaussian mixture model (EEMDSTD-GMM) is proposed. The study suggests a K-means-based pre-classification method for potentially confusing types of sensor faults. EEMDSTD-GMM-K-means has shown a better fault diagnosis capability under four single-source sensor faults and five multi-source sensor faults. Under the three examined fault levels, the results indicate a satisfactory performance with an average diagnosis rate of 98.7% for single-source faults and 96.5% for multi-source faults.