No abstract

No abstract

No abstract

No abstract

Space Shuttle Main Engine fault detection systems typically rely on sensor data analysis via redundant rulebased expert systems along with visual observations for the real-time assessment of engine health. A novel alternative to the traditional health monitoring approach is predicated upon the acquisition and subsequent neural network processing of electromagnetic plume emissions. Spectrometric examination of an emission spectrum provides a means for the identification and quantification of metallic species indigenous to the main engine plume flow. Knowledge of the metallic species eroding could pinpoint the specific location of component degradation within the engine as well as identify serious component failures at an early stage. Such an approach is advantageous because it allows for the detection of numerous internal failures that would otherwise go unnoticed by traditional monitoring methods. This paper details a radial basis function neural network architecture that is capable of inferring metallic state from a given plume spectrum. Specifically, a comprehensive discussion of the methodologies necessary for the development and implementation of the neural network approach are provided. The resulting neural networks are validated with actual test-stand data from an actual Space Shuttle Main Engine at NASA's Stennis Space Center. Nomenclature y = desired function for neural approximation x = neural network input feature pattern g j = Gaussian kernel function µ j = kernel function center position σ j = kernel function spread constant w = optimal array of network weighting coefficients G = array of Gaussian kernel responses to input patterns I i = vector of spectral intensities, (Watts•cm-2 •str-1 •nm-1