Although other classifications exist, we can think of most existing solutions to the problems of performing diagnostics and prognostics as belonging to one of two types: data-driven (also called model-free) techniques (including signal processing algorithms and knowledge-based methodologies) and model-based techniques.
Data-driven techniques rely on comparative assessments of the status of a system under testing with other known occurrences. For as long as the behavior of the system under testing is deemed to remain similar to that of a previously known, healthy configuration, the former is deemed to be healthy. When the measured behavior deviates from this reference, a fault is detected, and a comparison with the conditions previously observed in analogous faulty systems can take place. Under the appropriate conditions, this new comparison is bound to isolate and identify the fault effectively and very efficiently. Thus, the ability of data-driven techniques to perform the task of diagnosis is given by training of a classification algorithm.
The model-based approach is generally more robust in the sense that it can deal more easily with new or unforeseen situations, since this technique can incorporate and replicate, per its mathematical models, a wider range of behaviors, even if previously unobserved in actual systems. If the state of a system deviates from expected operational ranges, model-based techniques can continue to work by updating physical parameters that describe the new situation. Because of this ability, model-based techniques can also dispense with the use of the extensive training and historical information that its data-driven counterpart requires.
A comparison between the applicability of the data-driven and the model-driven approaches is shown graphically in the image below.
Inman, D. J., C. R. Farrar, et al., Eds. (2005). Damage Prognosis : For Aerospace, Civil and Mechanical Systems, John Wiley and Sons.