A software system designed to provide a purely signal-based diagnosis of virtually any time-varying system has been developed. This software is part of an overall concept called “Beacon-based Exception Analysis for Multimissions,” or BEAM. This concept provides for real-time autonomous diagnostics and prognostics of virtually any complex system (e.g., intelligent spacecraft or advanced aircraft) by use of software executed on an embedded computer.

BEAM provides for the onboard identification and isolation of anomalous conditions, making it unnecessary to telemeter large quantities of raw data for analysis on the ground. BEAM thereby reduces operator or pilot workload by isolating all anomalies that could affect safety, navigation, or performance. BEAM was conceived as an incrementable autonomy technology, capable of operating with no human intervention but also supplying condensed information to aid human operating decisions. Thus, the system is useful at both extremes, viz, total autonomy and complete operator control, and at every level in between.

This particular component of BEAM, called the System Invariant Estimator (SIE), uses strictly signal-processing methods and is therefore highly reusable. Many approaches to the problem of self-diagnosis exist, such as model-based and neural network solu- tions. However, in many cases these approaches require significant investment in modeling and training, perform poorly beyond the model or training envelopes, and perform poorly in the presence of data uncertainty. Because of BEAM’s underlying architectural differences from these approaches, a priori modeling is useful but not necessary, the training process is relatively simple and is incrementable, and the system is capable of correctly isolating anomalies well outside the training envelope.

The SIE provides a standard method for real-time fusion and analysis of all time-varying system observables, including sensor data as well as derived quantities and certain quantifiable software indications. These data sources can be from similar sensors or from radically varying types. The intermediate information products of the SIE retain considerable physical meaning, which allows complete traceability of the diagnostic state and reconstruction of the BEAM conclusions. The SIE provides detection capability, in both space (signal localization) and time, for both sudden and gradual changes in any system. The approach is readily scalable to systems of higher complexity and is resistant to the usual problem of combinatoric explosion as system size increases.

The SIE functions by considering the cobehavior of time-varying quantities, in particular their dynamics, as sensed from an operating system. Suitable sensors are widely used and typically include so-called performance sensors, i.e., temperatures, pressures, and the like. Certain repeatable relationships between physical quantities, and, hence, sensor values, exist in the system as dictated by the physics of its operation. These relationships are repeatable and relatively insensitive to changes in the environment or normal fluctuations.

The SIE constructs a single quantitative object to capture this cobehavior across the entire system or subsystem. This object reflects both known relationships, such as voltage/current relationships that are easily modeled, and unknown relationships, such as thermal transmission through system structure that is not well understood. The object is studied with respect to stability, for the purpose of detecting instantaneous changes or mode switches, and its longterm convergence, which reveals the presence of incipient faults, degradations, or minor and local shifts away from desired performance. This allows us to consider the entire space of faults, including those for which model information or even data is not available.

The SIE is a computationally efficient calculation grounded entirely in signal theory. It is trained using raw sensor data during known nominal operation, i.e., “supervised” training. This data would ideally cover all nominal modes. However, should this be impossible to obtain, or if only approximate data is available, the SIE can be executed in a learning mode. This is possible because the SIE is capable of capturing “novel” data as part of its anomaly detection, and should this data prove to be acceptable upon review, it can be incrementally added to the SIE training. The only needed human effort is providence of nominal data; all other training and detection is completely autonomous.

The SIE can accept time-correlated input data from relatively large sources. Its specific outputs are the presence of off-nominal behavior or degradation, the signals implicated in the fault, and a predictive assessment of loss of functionality. It also identifies specific pair-wise resonances contributing to the fault (critical for assessing control-loop-induced failures), renormalization functions to quantify intramodal stability and degradation, and capture of off-nominal flight data to allow adaptation to new modes or specific hardware.

Unlike similar neural network approaches, each information product of BEAM has an implicit physical meaning. Where uncertainties about the diagnosis exist, each step towards constructing the ISE (information space estimate) and its analysis can be performed individually. This provides operators the maximum utility in managing system information. This property also allows BEAM to optimally summarize fault information for downlink.

The present software has been applied to a broad variety of systems and has demonstrated enormous potential in improving system operability while reducing operator workload. Systems previously studied have spanned from 6 to 1,600 observables, and at various data rates from 0.016 Hz to 10 kHz.

This work was done by Sandeep Gulati and Ryan Mackey of Caltech for NASA’s Jet Propulsion Laboratory. For further information, access the Technical Support Package (TSP) free on-line at www.nasatech.com/tsp  under the Electronic Components and Systems 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 JPL Mail Stop 202-233 4800 Oak Grove Drive Pasadena, CA 91109 (818) 354-2240 Refer to NPO-20803, volume and number of this NASA Tech Briefs issue, and the page number.

This Brief includes a Technical Support Package (TSP).
Reusable Software for Autonomous Diagnosis of Complex Systems

(reference NPO-20803) is currently available for download from the TSP library.

Don't have an account? Sign up here.