Composing model-based diagnostic systems

We present a novel approach to developing model-based diagnostic systems based on composing a solution from a prespecified and predeveloped, and hence reusable, set of software modules identified from a functional analysis of the complete model-based diagnostic process. This approach leads to a generic architecture, developed within the ARTIST project, that encompasses the main model-based diagnostic techniques currently being employed in applications. The architecture consists of predictor, candidate proposer, and diagnostic strategist modules that represent the primitive functionalities of application systems. Instantiations of each of these functionalities are provided for the main strategies, i.e., dependency recording and iterative search based techniques. From these modules three complete diagnostic systems have been developed and applied to two full-scale industrial applications and a laboratory-scale process-rig. Such an approach allows the reuse of existing modules, customized by the relevant domain knowledge, resulting in a much reduced development time and making the power and generality of model-based applications to diagnosis technically and economically viable for a wide range of industrial applications.