Improving the PCA-based Chiller Sensor Fault Diagnosis with Relative Contribution

Built environment system focused on creating the suitable indoor air quality, while consuming a great deal of energy. Sensors played key role for monitoring the current real situation for HVAC&R equipment and system. More and more machine learning algorithms were implemented to maintain the optimal operation for the built environment systems and devices. Among them, Principe Component Analysis, as one of the most useful unsupervised machine learning algorithms, was widely employed to diagnose the abnormal sensor for chiller. However, the complete decomposition contributions could not correctly finding the faulty sensor due to its non-largest corresponding contribution. The relative contribution were employed to diagnose the abnormal one according with the rate between the contribution and its expect. Validated by the dataset of ASHRAE RP-1043, results showed that the relative contribution worked well even with the faulty server level of ±0.5°C for the cooling water out temperature sensor.