Dependency modelling pitfalls

Model based diagnostic systems are becoming quite popular partially because of the large number of successful applications. Many tools exist for the use of the model based approach, each with a diagnostic advisor. However, experience at modeling over 250 systems has shown that the process of modelling is not straight forward and requires a number of considerations not generally dealt with in the literature. Since the modelling process involves some aspects of art and some aspects of science, the ability to model incorrectly and obtain inaccurate answers is great. This is especially true in the area of automated model development where the concept of a test may be improperly or poorly defined. Some of the more subtle nuances of the modelling approach are discussed and illustrative examples to demonstrate these subtleties are presented.<>