Using Computational Fluid Dynamics and Machine Learning to Predict Sled Profile During High Speed Water Braking at Holloman High Speed Test Track
Accurate prediction of a rocket sled test profile and water braking phenomena has potential to result in radical changes in designs of specific sleds and provide greater confidence of braking mechanism and recovery of critical infrastructures. Understanding the water’s behavior with the sled is critical to predicting how the water could damage the sled and affect recoverability of the sled and determine success of the missions. Traditionally, sled design for the test missions had been guided by empirical/hand calculations to estimate the forces on various components. The calculations involved various approximations to arrive at the force balance law and predict the acceleration/deceleration profile. In partnership with the Holloman High Speed Test Track (HHSTT), we performed preliminary simulations to develop a predictive model for the HHSTT sled tests for various velocity regimes. The CFD results from various geometry configurations for the sled and modeling parameters will be presented. The main goals of the CFD investigations are to improve the accuracy of the predicted profile that often depends on the complexity of the design and operating conditions. Having an atmospheric turbulence deconvolution model would be of great utility for the Air Force in its mission to maintain space situational awareness. The purpose or goal of this study is to increase US Air Force capability in space situational awareness, through the construction of a deconvolution model that, once trained, can remove turbulent effects and provide clearer images in a small amount of time (expedient), with relatively good accuracy (effective), and with a relatively low computational resource requirement (efficient). To develop a deconvolution model a Conditional Generative Adversarial Network (cGAN) will be used. The cGAN will make use of two Neural Nets (NN) that will be pitted against each other, one will be an image generator and the other an image discriminator. The Generator will be given turbulent images and will generate turbulent images while converging towards a 'pristine (non-turbulent)' image as it attempts to 'trick' the Discriminator. The discriminator will be given pristine, blurred, and generated (fake) images, its goal will be to distinguish whether images are generated or from the pristine dataset. By combining CFD with machine learning, cGANs specifically, we can train a net to produce CFD results, eliminating time and resources required to run CFD models.
Computational physics|Artificial intelligence|Fluid mechanics
Terrazas, Jose Armando, "Using Computational Fluid Dynamics and Machine Learning to Predict Sled Profile During High Speed Water Braking at Holloman High Speed Test Track" (2020). ETD Collection for University of Texas, El Paso. AAI28090665.