Multimodal Reasoning for Federated Hybrid Diagnosis of Modular Electromechanical Systems

Introduction Marketplace pressures towards "mass customization" are pushing industrial equipment and consumer products to be manufactured in greater numbers of configurations and models. One of the more significant product variety coping strategies to emerge from industry is the increasing emphasis on modularity and reuse of components, assemblies and systems over multiple product generations and families. The mechanical engineering design research community has responded to these pressures with increased attention to product design-for-modularity methodologies and research [Ulrich] [Newcomb et al.]. Electromechanical systems above a certain level of complexity are typically difficult and cosily to model for diagnosis. For most products, the diagnostic modeling task is generally performed by the system integrator (or OEM original equipment manufacturer) after the design is completed. A single reasoning approach is usually adopted for ease of model construction, testing and maintenance. To the extent that embedded diagnostics are not required, Supply Chain vendors do not share in the system-level diagnostic modeling task, even though their component module may constitute a significant failure risk [Lee]. This current approach, however, is poorly suited to the needs of manufacturers who increasingly "compesC’ new products by reusing pre-existing modules in strategic locations. "Unimodal" centralized diagnostic modeling and reasoning technologies do not scale well, and are illsuited to keeping pace with the need to field larger varieties of products (and product diagnostic systems) lower cost, and with less development, test and maintenance effort. This industry-sponsored research explores the interplay between electromechanical design-for-modularity theory and practice and the construction and maintenance of modular multimodal diagnostic systems. We are specifically interested in: