Diagnostic algorithms that have been developed specifically for industrial processes are described. These algorithms are based on a representation of processes using the multilevel flow models (MFM) developed by M. Lind. Unlike most other model-based diagnosis systems, the MFM functional model explicitly represents teleological (or means-end) information. The algorithms use an explicit representation of goals and functional structure as a basis for diagnosis. The three types of diagnosis supported are measurement validation, alarm analysis, and fault diagnosis.

The author presents a concise review of MFM and describes how the algorithms use the MFM model for diagnosis. A set of simple, informative examples provide insight into how the models are constructed and how the structure of the model provides the information needed for the diagnostic algorithms. The paper explains the philosophy of the design well and describes the strengths and limitations of the representation and algorithms. A comparison of this type of model-based diagnosis with diagnosis based on qualitative physics is provided. A lengthy appendix compares this work with other work based on the MFM models, other functional models, qualitative behavior models, quantitative behavioral models, and hybrid models.

The breadth of application of this type of diagnosis system is illustrated by the Guardian system, which will represent critical functions of the human body and be used for diagnosis in emergency room situations. The highly abstract level of representation available with the MFM models and the efficient diagnostic algorithms based on the models appear to be a promising approach to the construction of practical diagnostic systems.