A Deep Learning Solution for Fragment Mass Estimation using Holistic Data

Three dimensional (3D) stereoscopic data from high-speed video (HSV) has become a ubiquitous diagnostic technique for characterizing fragmenting warheads in arena tests. Their utility has been evaluated by the Department of Defense with varying degrees of success, depending on the metric of interest. Estimating fragment trajectories and speeds offers the highest confidence from these systems. However, fragment mass remains a difficult parameter to estimate. Reliable fragment mass measurements are a vital component to the U.S. Navy’s vision of developing advanced arena test diagnostics and are needed to validate computational models. Therefore, the U.S. Navy requires an innovative solution to bridge this gap related to 3D stereoscopic HSV systems through the development of a general-purpose fragment mass estimation library. Protection Engineering Consultants (PEC) and Southwest Research Institute (SwRI) have teamed to addresses these challenges for the development of a mass estimation library for 3D stereoscopic data. A successful mass estimation solution needs to reach beyond the exclusive use of grainy HSV data and utilize a more holistic data approach. We propose to develop a Bayesian framework that contains deep learning models to estimate fragment mass and uncertainty by leveraging these holistic datasets. Training data development will be the backbone of this solution. The final output will be a best estimate mass plus an uncertainty measurement.