Unsupervised Clustering of Solid Propellant Missile Systems Using Performance Parameters and Telemetry Data

This paper describes the application of statistical learning techniques to unsupervised clustering of solid rocket motor-powered missile system designs. Two clustering scenarios are presented: one shows how to cluster missile data based on missile performance parameters, and the other shows how to use the "max" parameters of the trajectory for clustering. The first scenario reveals that missiles can easily be clustered using unsupervised training on well-partitioned data such as specific impulse and chamber pressure. This scenario will only utilize the raw data extracted from the missile, rather than the class of missiles from which the missile design was generated. This scenario, because it utilizes unsupervised learning, will also determine the number of clusters to use by only looking at the data extracted. The second scenario utilizes the max parameters of the flyout, such as apogee and downrange distance, to cluster the missile data in an unsupervised manner. Various clustering methods using different inputs are shown. Both scenarios utilize Scikit Learn in Python to generate the clusters and compare performance using methods such as K-Means and Spectral Clustering. The main goal is to cluster missile data based solely on collected data, rather than known information. Using K-Means and Spectral Clustering allows for the generation of clusters that are akin to the actual classes used for data generation. The paper will describe the methods, the data generation, associated results, and future directions for the work.