A hybrid implicit/explicit automated reasoning approach for condition-based maintenance

A hybrid approach for implicit and explicit automated reasoning for condition-based maintenance (CBM) has been developed. The development of the new approach is motivated by describing the relevance of automated reasoning to CBM, and to the smart ship project, and by reviewing the challenges in designing an automated reasoning system. The concept of the new approach is then described and illustrated with an example. In an illustrative example, an intelligent oil analysis system, a set of explicit rules is built, for ascertaining whether the oil filter is clogged based on pressure, temperature, and flow measurements. Then the rules are used to train a layered feedforward neural network to capture the explicit knowledge. Finally, a much smaller rule set is derived from the neural network which represents the same explicit knowledge that was obtained from a domain expert. The neural network can be retrained on actual sensor data to refine the knowledge base and to incorporate implicit system knowledge. Rules can also be extracted from the resulting network. The notable features of the novel approach are its ability to encapsulate explicit and implicit knowledge and its flexibility in allowing updates of the knowledge base and extraction of a parsimonious and consistent set of rules from the network. The approach is truly hybrid as it combines fuzzy logic and rule-based systems, to capture the expert's explicit knowledge, and neural networks, to provide a parsimonious representation of the knowledge base and for their adaptability. The approach can benefit shipboard CBM efforts in automating; intelligent condition monitoring and by providing an efficient implementation mechanism for distributed diagnostics.