Utilizing neural networks to predict wing leading edge slats airloads

摘要

This paper describes a method to predict pressure on the leading edge slats of a business jet using Artificial Neural Networks. Neural Networks are used to avoid the problem with unbiased estiamtors that stems from multicollinearity in the data. Pressure prediction is achieved through mathematical modeling of flight test data collected during the aircraft certification process. Mathematical model parameter selection and pressure port groupings are achieved through a combination of statistical analysis and engineering understanding of the problem. Use is made of Kohonen's Self-Organizing Map as a means of detecting redundant and conflicting data from the flight tests, and in selection of the trainign and test sets for the predictor networks. The backpropagation paradigm is then used to carry out the prediction process. Prediction errors were statistically examined to detect and correct any bias in the modeling process.