Model-based generic approaches for automated fault detection, diagnosis, evaluation (FDDE) and for accurate control of field-operated centrifugal chillers

Performance monitoring of large engineering systems with the objective of optimal control and fault detection, diagnosis and evaluation (FDDE) has drawn the attention of several researchers over the last 3–4 decades. However, there are several fundamental and applied issues that need to be researched into, evaluated, developed and tested before it can be used with confidence by the professional HVAC&R community. This research will specifically address: (i) the issue of automated FDDE for large vapor compression chillers by developing a generic (i.e., one which can be used for any vapor compression centrifugal chiller) model-based (as against model-free) approach that provides diagnostic insight into which primary chiller component is not performing as designed and has developed a fault, and (ii) provide accurate chilled water temperature control for vapor compression chillers by developing a transient physical model-based feed-forward control configuration (as against the traditional feed-back control configuration applied in centrifugal chillers).; A primary and distinctive facet of this approach is that this FDDE methodology and new control strategy will be based as much as possible on using existing sensors, transducers, and other hardware rather than requiring that these be installed specifically for this purpose. Each and every primary component of the chiller can be characterized by at least one performance parameter, the magnitude of which is indicative of the health of that component. Mathematical models for these characteristic parameters are presented, and how well these models can be trained with field-monitored data is illustrated. Large chillers are usually controlled by inlet guide vanes, which vary the refrigerant temperature in the evaporator. For such a control to respond quickly while maintaining reasonable accuracy, the guide vanes must be directly commanded by a function associated with the input signal. A transient physical model has been developed analytically, calibrated and validated against one-minute chiller transient data. Test results show that this new control strategy improves the control accuracy by 28% under the operating conditions selected.