Bayesian Networks for Automated Equipment Diagnostic Support
Abstract
As combat vehicles and other legacy systems age and are required to perform additional capabilities on increasingly remote battlefields, the Marines responsible for them currently lack tools necessary to diagnose and fix these critical assets independent from higher echelon corrective maintenance service support. For a Light Armored Reconnaissance detachment conducting distributed maritime operations and tasked with providing organic precision fires, small unit leaders and maintainers are responsible for performing all levels of diagnostics with minimal direct support, a situation that threatens expeditionary advanced base operations when vehicles inevitably fail. At the operator level, current troubleshooting procedures are primitive and fail to capitalize on recent breakthroughs in computation and causal reasoning algorithms. An automated program driven by a causal Bayesian network allows the maintainer to input observed symptoms into a model that directs their attention to the most probable causes of failure. Expert knowledge, Bayesian learning techniques, and automated reasoning are applied to determine network structure, model parameters, and the degree to which various symptoms affect output. When linked to a user interface, the maintainer can now quickly and accurately diagnose a degraded system from a handheld device, hundreds of nautical miles from the nearest maintenance bay.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 01, 2021
- Accession Number
- AD1151106
Entities
People
- Zachary C. Polson
Organizations
- Naval Postgraduate School