Ontology-driven generation of Bayesian diagnostic models for assembly systems

A major challenge limiting the practical adoption of Bayesian networks for diagnosis in manufacturing systems is the difficulty of constructing the models from expert knowledge. A key possibility for tackling this limitation is believed to be through utilising the available sources of design information that is readily available as part of the engineering design process. Some of the most notable sources of such design information include formal domain models such as product-process-equipment design ontologies which are becoming a widely accepted mean for formally capturing and communicating design information. This makes these ontologies a valuable knowledge source for automatic and semi-automatic generation of Bayesian networks, instead of the entirely expert-driven traditional approach. However, design ontologies lack on the fault-related information side as they are primarily aimed at capturing the intended behaviour of the designed system. To bridge this gap, we propose integrating failure mode and effect analysis (FMEA) information into design ontologies and using the resulting integral models for the generation of Bayesian diagnostic networks. We also propose a method for the generation process and demonstrate the validity of the approach with an industrial case study.