SOW - Algorithm Development 2020 IHMC_SWarner.docx

07 July 2020

STATEMENT OF WORK

BACKGROUND

To combat the operational safety threat associated with unexplained physiological episodes (UPEs) the U.S. Navy and Air Force have developed sensors to integrate into tactical aviation to detect life-support system malfunctions and changes in operator functional state. These sensor packages include measures of FiO2, SpO2, and CO2 state, heart rate, breath rate, cabin pressurization, breathing gas pressure, and acceleration forces. Multisensory arrays will provide a wealth of information to diagnose and mitigate UPEs. However, in dynamic-aircraft environments with active multisensory data streams, measurement density and high frequency human/machine activity variability are likely to be issues and create high rates of signal processing, large data acquisition bandwidths and other complexities that prevent capturing the total contributions to pilot performance from aircraft, pilot and/or the environment. Additionally, the recording and processing technologies for the multisensory data and human-machine performance typically requires developmental support from different technical experts in electrical engineering, computer science and statistics and significant time and resources. Tracking such complex human and synthetic systems also entails many different inputs and multiple vendors who supply their own data formats and integrating their data streams and types is challenging, expensive and requires long development cycles.

In response to such challenges, IHMC will develop a prototype hardware and software system for automated processing of Unexplained Physiological Event (UPE) relevant multivariate data derived from multisensory arrays using machine visual analytics. The rapidly deployable machine vision system incorporates machine learning and cognitive neural networks aspects and will allow the capture of multivariable image data, labeling, neural net training, AI processing and variable(s) prediction. The minimally invasive multi-sensor data fusion and learning system will be evaluated and optimized for classification and ranking of vital signs. The system will be augmented with expert rules and decision tree logic and evaluated as a discovery tool for multi-factors contributing to UPEs for future UPE diagnosis and mitigation. The hardware and software system performance will put a minimal impact on human and machine operation and be capable of operating under harsh environmental conditions.

SCOPE

In an effort to diagnose and mitigate UPEs, a prototype hardware and software system for automated processing of multivariate data from multiple sensor packages and incorporation of machine visual analytics will be developed.

System requirements include:

· Measureable data converted/dispersed from multisensory arrays.

· Machine vision system (machine learning and cognitive neural networks).

· Capture of multivariable data, labeling, and neural net training.

· Artificial Intelligence processing and variable prediction.

· Classification and ranking of vital signs.

· Discovery tool for multi-factors contributing to UPEs.

· Machine Learning Processing supplemented with expert rules and decision tree logic.

· Minimal impact on human and machine operation.

· Capable of operating under harsh environmental conditions.

DELIVERABLES

(1) Executable algorithm prototype – including relevant hardware and software

(2) Materials and services required for preventive maintenance and operation

(3) Training and certificates

SECURITY

Unclassified IHMC maintains facility clearance: Secret

PERIOD OF PERFORMANCE

One Year from Date of Award

TECHNICAL POINT OF CONTACT (TPOC)

David Fries (laboratory), IHMC, 850.202.4468 Joe Pierce (shipping), IHMC, 850.202.4462 Joe Campbell (security), IHMC, 850.275.0350