Predict
To forecast what is going to happen in the future, often on the basis of present and past indications.
Real Time Trajectory Reshaping
Barron Associates’ unique and innovative approach to real time trajectory reshaping is denoted as the Optimum-Path-To-Go (OPTG) algorithm. Polynomial networks describing the “best” remaining path to the end of the mission segment are interrogated on-line at regular intervals to obtain the optimal trajectory commands, given the current state and “health” of the vehicle. OPTG developments to date have shown very promising results, as Barron Associates continues to refine and improve the approach.
Machinery Diagnostics and Prognostics
A novel approach for performing gear and bearing diagnostics (i.e., fault detection and isolation) and prognostics (i.e., forecasting remaining useful life (RUL)) has been developed by Barron Associates. The approach is based on a bispectral statistical change detection (SCD) algorithm. The approach detects and isolates faults to the level of specific machinery components (e.g., gears and bearings) and forecasts RUL. Important attributes of the SCD approach include: (1) it does not require a priori training data; (2) it is operating-regime independent (i.e., works across different operating conditions, torque levels, flight regime, etc. without knowledge of same); and (3) it can be implemented using relatively inexpensive computer hardware, since only low-frequency vibrations need to be processed.
Algorithmic results are available for naturally-occurring (i.e., non-seeded faults) in a CH-47D helicopter gearbox and for a laboratory gear and pinion setup, both of which were overloaded to accelerate fault initiation. For the CH-47D data, RUL was predicted 8 minutes prior to a gearbox failure; the initial gear tooth microcrack initiated only 23 minutes prior to complete gearbox failure. For a gear and pinion system, gear failure was accurately predicted 50 hours prior to end of RUL.
Medical Outcome Assessment
Barron Associates has developed quantitative methods for field assessment of injury severity based on very restricted sets of vital sign and related data. Such assessments would be used to augment decision-making by emergency personnel. In application to civilian trauma databases, Barron Associates' polynomial neural network (PNN)-based mortality prediction algorithms out-performed all conventional pre-hospital triage algorithms for which comparison was possible. Additionally, our analysis of various hemorrhagic shock databases produced innovative modeling methodologies that exploited hemorrhage dynamics. This resulted in synthesis of models capable of utilizing continuous data steams to effectively capture complex physiological trajectories.
The Barron Associates algorithms are targeted for use by the military in the Life Support for Trauma and Transport (LSTAT) system, which is a smart stretcher/portable ICU recently approved by the Food and Drug Administration, that turns essentially any transport vehicle into an ambulance.
Fault Detection and Isolation
This work concerns the development of an algorithmic approach for detecting and isolating sensor, actuator, and plant faults in complex dynamical systems, such as aircraft flight control systems and turbine engines. The method is based on the use of extended constrained Kalman filters, which are able to detect and isolate such faults by exploiting analytic redundancy that exists among various subsets of available actuator input and sensor output data. A statistical change detection technique based on a modification of the generalized likelihood ratio is used to detect faults in real time. The approach has been demonstrated successfully in simulation of both a nonlinear jet engine and a full six degree-of-freedom (6-DOF) nonlinear business jet application.