Neural Network Methods for Preventing Unstarts

We will use an innovative neural network algorithm to perdict and control National Aero-Space Plane(NASP) unstarts at the earliest possible moment. Unstarts present a design challenge to hypersonic aircraft. An unstart occurs when the pressure at the aft end of the engine flow path reaches a critical level. This overpressure causes a "choking" of the air flow. The flow tends to spill around the inlet and the pressures in the flowpath become excessively high. The thrust goes to zero. These high pressures can damage the engine and if the design is not sufficiently robust, an unstart could lead to loss of the vehicle. By predicting the unstart at the earliest possible moment, the prediction algorithm will give the control algorithm enough time to prevent the unstart. A few extra milliseconds additional lead time could improve control. We will compare traditional methods of pattern recognition against a neural network approach to determine if critical time could be gained with the neural approach. Technical Abstract F33657-93-C-2233