Further work on beacon-based exception analysis for multimissions (BEAM), a method of real-time, automated diagnosis of a complex electromechanical systems, has greatly expanded its capability and suitability of application. This expanded formulation, which fully integrates physical models and symbolic analysis, is described architecturally in the figure.
In a typical application, BEAM takes the form of an embedded software suite executing onboard the system under study, though many off-board data analysis engines have been constructed as well. The BEAM software performs real-time fusion and analysis of all system observables. BEAM is intended to reduce the burden of diagnostic data collection and analysis currently performed by both human operators and computers. In the case of a spacecraft or aircraft, BEAM enables onboard identification and characterization of most anomalous conditions, thereby making telemetry of larger quantities of sensor information to ground stations unnecessary. Previously BEAM has been described in several prior NASA Tech Briefs articles: "Software for Autonomous Diagnosis of Complex Systems" (NPO-20803) Vol. 26, No. 3 (March 2002), page 33; "Beacon-Based Exception Analysis for Multimissions" (NPO-20827), Vol. 26, No. 9 (September 2002), page 32; and "Wavelet-Based Real-Time Diagnosis of Complex Systems" (NPO-20830), Vol. 27, No. 1 (January 2003), page 67.
The new formulation of BEAM expands upon previous advanced techniques for analysis of signal data, utilizing mathematical modeling of the system physics, and expert-system reasoning. These components are integrated seamlessly, making possible analysis of varied information about the monitored system, including time-correlated signal performance, state information, software execution, operator command execution, and convergence to state and physical models. BEAM software is highly adaptable and can be implemented at relatively low cost in terms of processor power and training, and does not require special sensors. Unlike some prior methods of automated diagnosis, BEAM affords traceability of its conclusions, which allows system experts to completely reconstruct its decision path for greater operator confidence or to aid analysis of novel conditions. Principal among BEAM's strengths is its excellent performance in detection and classification of such novelty, meaning faults of previously unknown — and untrainable — type.
In the BEAM architecture, discrete sensor information, state information, and commands are fed as input to the symbolic model, and quantitative sensor data is input to a simplified physical model of the system. These modules are designed to leverage existing system models, which can be high or low fidelity. The symbolic model aids signal-based analysis in terms of mode selection or other discrete outputs. The physical model improves sensitivity through separation of predictable and unpredictable signal components.
Time-varying quantities are analyzed in two groups: (1) signals with a high degree of correlation to others, or signals that are not isolated in a diagnostic sense, are passed to the coherence-analysis component of BEAM; (2) signals that may uniquely indicate a fault, as well as those already suspected to be faulty, are passed through feature-extraction components. This split allows BEAM to consider very complex faults in the system, including interference faults or miscommunication that escape univariate detection methods, while retaining robustness in poorly redundant systems or in the face of gross nonlinerity.
The components of BEAM described in the figure are summarized as follows:
- The model filter combines sensor data with predictions from a real-time physical model. The inclusion of physical models, where available, is the most efficient way to incorporate domain knowledge into signal-based data analysis.
- The symbolic data model interprets status variables and commands to provide an accurate, evolving picture of the system mode and requested actions.
- The coherence-based fault detector tracks the cobehavior of temporally varying quantities to expose changes in internal operating physics.
- The dynamical invariant anomaly detector tracks parameters of individual signals to sense subtle deviations and predict near-term behavior.
- The Informed Maintenance Grid (IMG) studies evolution of cross-channel behavior over the medium- and long-term operation of the system. It tracks consistent subthreshold deviations and exposes deterioration and loss of performance.
- The prognostic assessment yields a forward projection of individual signals, based upon their extracted parameters. It also provides a useful short-term assessment of impending faults and loss of functionality.
- The causal system model is a rule-based connectivity model designed to improve isolation of fault sources and identification of actor signals.
- The interpretation layer collates observations from all separate components and submits a single fault report in a format useable by recovery software, planners or other AI software, and/or human operators.
This work was done by Ryan Mackey, Mark James, Han Park, and Michail Zak of Caltech for NASA's Jet Propulsion Laboratory. For further information, access the Technical Support Package (TSP) free on-line at www.techbriefs.com/tsp under the Information Sciences category.
In accordance with Public Law 96-517, the contractor has elected to retain title to this invention. Inquiries concerning rights for its commercial use should be addressed to
Intellectual Property Group
Mail Stop 202-233
4800 Oak Grove Drive
Pasadena, CA 91109
Refer to NPO-21126, volume and number of this NASA Tech Briefs issue, and the page number.