Using Neural Networks for Estimating Cruise Missile Reliability

ACC believes its current methodology for predicting the reliability of its Air Launched Cruise Missile (ALCM) and Advanced Cruise Missile (ACM) stockpiles could be improved. They require a predictive model that delivers the best possible 24-month projection of cruise missile reliability using existing data sources, collection methods and software. It should be easily maintainable and developed to allow a layperson to enter updated data and receive an accurate reliability prediction. The focus of this thesis is to improve upon free flight reliability, although the techniques could also be applied to the captive carry portion of the missile reliability equation. The following steps were taken to ensure maximum accuracy in model results. 1. Add more detail to flight test reliability calculation. 2. Convert the ground test data into a usable form (reduce). 3. Engage in an exercise in feature selection. 4. Develop a Matlab model prototype. 5. Validate the model via problems with known solutions. 6. Apply an appropriate data fusion technique to the different network outputs (logistic regression, feed-forward and radial basis function). 7. Put the model into the form of a usable tool for the end-user. The end product is the ALCM/ACM Reliability Estimation System (AARES), a VBA-based model that meets all user criteria.