Engineering and Statistical Analysis of Solid and Liquid Missile Systems

Abstract

Engineering and statistical analysis is applied to gain insights and information regarding liquid and solid missile systems. This dissertation outlines five main key topic areas, physics modeling, data generation, model generation, model robustness/sensitivity, and model explainability. The Auburn University Liquid Rocket Code (AULRC) is used to generate waypoint parameters such as time of flight. Models are then generated to predict time of flight and the performance of said model is assessed using various metrics. Data generated from the AULRC is used for quantification (regression) purposes throughout this dissertation. Similarly, the Auburn University Solid Rocket Code (AUSRC) is used to generate waypoint parameters such as max thrust, however, the AUSRC is used to build classes of rockets for classification models. Both programs use a Latin Hypercube (LHC) algorithm to randomly generate data between minimum and maximum values. Regression models developed will be classical linear regression including ridge and lasso and will include higher order interaction terms and more complex models developed include neural networks (NNETs). Classification models developed will be classical linear/quadratic discriminant analysis and NNETs. All the models were built using either Scikit-Learn or TensorFlow with Keras generated using Python. The classification models assume that all the data is available while in real-life scenarios this is not always true. Therefore, parameters are chosen to be “missing” and are replaced using imputation methods. The sensitivity and robustness of the models can be assessed by evaluating the classification metrics on the imputed data. To explain models and their predictions, the feature contributions can be assessed to see which parameters are most influential and how the model makes a prediction. For the classical models, coefficients can be used to assess which parameters had the most influence, but for NNETs, the SHapley Additive exPlanation (SHAP) values can be used to determine which parameters were most important in the model and assess how model makes predictions.