Adaptive Maintenance Knowledge Bases for Field Service

Abstract

A maintenance knowledge base (MKB) is an information system capable of offering solutions to diagnostic problems at a level comparable with that of experts in the field. The development of a maintenance knowledge base for a system is an iterative process; it is typically built using previous experience and expertise. In most cases, the development of diagnostic procedures in a MKB is a completely manual process; system engineers utilize existing sources of information, such as schematics, system design and reliability data, failure modes, effects and criticality analysis (FMECA) and maintenance logs (cases), in creating these procedures. A major problem with these diagnostic systems is that the knowledge bases (models and diagnostic decision trees or rules generated from them) are static, and that they are updated infrequently. Consequently, they are hard to adapt to newly-designed systems and/or to new maintenance logs collected from the field. In this paper, we will present an innovative approach that seamlessly combines model-based reasoning (MBR) and data-driven or case-based reasoning (CBR) for adaptive knowledge base creation, maintenance and update through multi-signal flow graph modeling. The adaptive MKB utilizes a diagnostic information model of the system (the prior) and combines it using a Bayesian framework with the historical and current reliability data/maintenance logs (data) that are part of a knowledge capture mechanism for the continuous refinement of the knowledge base (posterior). This adaptive MKB not only significantly reduces the upfront effort in creating the initial diagnostic model, but also has substantial potential in reducing the barrier of entry (and hence adoption) of model-based diagnostic methodology for system fault detection and diagnosis (FDD).