Finding Unknown-Unknowns

Using Big Data to create turnkey, adaptive goal-driven systems

0 Ratings

Imagine you own a system vital to your mission success, but for which subject matter experts are not readily available. However, the system has historical logs of numerous parameters per minute describing its mostly normal state of operations over time or a database into which you could save these parameters. You need to know when this system starts to behave abnormally or outputs system status parameters that have unknown precursor signatures that could lead to a system failure.


The Air Force needs to automatically integrate numerous complex sources of Space Situation Awareness (SSA) data to provide a consistent, collaborative, and timely representation of the space situation. Any Air Force system such as a satellite, aircraft, space catalog, engine, or even communications, outputs thousands of State of Health (SOH) measurements every minute. This information describes the state of the system. Twenty analysts study the values of these measurements, each individual specializing in a subset. A small percentage of the measurements have "stop light limits" that require an analyst to manually report the system's status to the commander. Such time-consuming labor can be avoided with the Blue Force Status (BFS) system. BFS automatically provides unknown abnormal data patterning and abnormal measurement correlation detection, characterization, and event tracking.


System analysts need to be flagged on the most abnormal system behaviors, find similar behaviors and their responses used in the past, and be able to characterize new behaviors and responses. Without BFS only the known "needles in the haystacks" are found, not the "abnormally bent hay." BFS replaces outdated means of fault detection that miss the precursors and events that kill the system. The BFS especially addresses factory-set limits that quickly go out of date and are not maintained.


DF&NN; developed a computational intelligence system that learns normal system behavior offline and then in real-time scores how abnormal the system behavior in each time window is. DF&NN; teamed with AFRL, Space and Missile System Center, and Lockheed Martin Corporation to operationalize the system.

"BFS uses relationship learning to provide automated real-time abnormal behavior detection, characterization, and tracking in complex "big data" sets from large electro-mechanical devices or networks" — Christopher Tschan


The BFS system is a computer program that takes in a lot of data points and learns normal patterns in it over time. The system uses the representation of these patterns to measure how abnormal real-time patterns are over moving over a particular time period. It automatically detects abnormal behavior that is “within stoplight limits.” The system is data-driven so it does not require expert time other than to define the top-level system goals such as the length of the time of interest and reporting sensitivity. The system detects in real-time and is linearly scalable up to an arbitrarily large set of system measurements. Based upon user goals, BFS decides when to retrain to learn new normal system behavior, what data to retrain on, what data to test on, and when to promote the new BFS system. A unique characteristic of the DF&NN; BFS is that not only are the components that learn normal behavior automatically adapted to the data, but the architecture of the BFS components themselves are adapted based on the data, without access to the BFS expert, based upon Occam’s Razor. To achieve these bio-inspired solutions we are not concerned with learning the world (e.g., not passing the Turing test), but with achieving a system that fulfills a specific dynamic purpose and maintains itself for each application. As a result sufficient performance is achieved without any system domain expert or algorithm expertise.


Air Force system subject matter experts can do what they do best without pouring over big data sets. They are given prioritized flags on when and where to look, what was the cause, when this abnormal behavior has happened before, how it was characterized, and what the recommended response was.

DF&NN; has received a patent on this BFS technology for commercialization. This BFS development provided us with a "big data" abnormality detection, characterization, and event tracking set of broadly applicable to

The technology and patent increase US competitiveness indirectly through intelligent "big data" processing results. Advantages of this increased automation include more efficient aerospace system operations; the ability to respond more quickly to unanticipated situations than is possible today; increased safety; reduction in operations cost; and eventual reduction in personnel needed to support operations.

The technology has provided the first step in "deep learning" sparse neural networks capabilities. "I am 100% more efficient with the BFS system." — Armand Chenard, 22 SOPS Signals Analyst at 22 SOPS government employee

Data Fusion & Neural Networks (DF&NN;), LLC

1643 Hemlock Wy, Broomfield CO 80020

DF&NN; provides Data Fusion & Resource Management (DF&RM;) computational intelligence systems tailored to meet DoD needs. DF&RM; systems are developed using the DF&NN; Dual Node Network (DNN) technical architecture that decomposes the problem into 5 'dual' DF&RM; levels.

Christopher Bowman Christopher Bowman

Christopher Bowman


Duane Desieno Duane Desieno

Duane Desieno

Intelligent Systems Programmer Analyst


Threat Detection, Validation, and Mitigation Tool for Counterspace and Space Situation Awareness (SSA) Operations





For more exciting Air Force launch stories, visit



Space On Earth


One Stop Space Shop