Sensor-fault detection, diagnosis and estimation for centrifugal chiller systems using principal-component analysis method

An online strategy is developed to detect, diagnose and validate sensor faults in centrifugal chillers. Considering thermophysical characteristics of the water-cooled centrifugal chillers, a dozen sensors of great concern in the chiller-system monitoring and controls were assigned into two models based on principal-component analysis. Each of the two models can group a set of correlated variables and capture the systematic trends of the chillers. The Q-statistic and Q-contribution plot were used to detect and diagnose the sensor faults, respectively. In addition, an approach based on the minimization of squared prediction error of reconstructed vector of variables was used to reconstruct the identified faulty-sensors, i.e., estimate their bias magnitudes. The sensor-fault detection, diagnosis and estimation strategy was validated using an existing building chiller plant while various sensor faults were introduced.