Reaching a comprehensive diagnosis is achieved by replicating multiple behaviors of a faulted system via a model. These simulated behaviors can be compared to particular instances of observed behaviors in an actual system in operation to determine the current fault or damage status of the latter, per the matching situation in the model.
Thus, a model utilized in model-based diagnostics must provide the ability to simulate an engineering system under varying types of faults and varying amounts of damage, because comprehensive fault diagnostics involve not just the aspect of fault detection, but fault isolation and fault identification as well. Effort has thus been invested in designing an architecture that can simulate the changing behavior of a system as a fault develops. A result of this effort, which can be generally applicable to many kinds of systems, is the scheme shown below.
This modeling approach makes it possible to try to diagnose a specific fault in the plant by searching for a correspondence between observed plant behaviors and one of several simulations that represent varying degrees of damage. Thus, this technique can be likened to a reverse engineering process of fault identification.