An Adaptable Self-Monitoring Framework for Opaque Machines

Diagnostic systems for complex machines are highly specialized and cannot be applied in other domains without significant effort. Our goal is to improve the robustness of diagnostics with an adaptable monitoring framework for identifying and explaining anomalous behavior that can be easily modified for different domains or systems. We define a vocabulary for reasonable data---to precisely identify contradictions between expected and anomalous behavior and a language---to express rules, policies, and constraints/preferences of the user. We combine this framework with explanation mechanisms to describe the core reasons and support for a reasonableness judgment made by running the reasoner over the reasonable data, rules and the state of the related components.