ABSTRACT
Rocket nozzles are subject to various kinds of loadings that typically include the pressure due to hot exhaust gas, forces experienced as a result of high-speed flow of the propellant through the regeneration tubes, the thermal strains due to the high temperatures experienced, and bolt preloads. All these activities lead to several interesting phenomena that have to be analyzed for circumventing any problems that might occur during use. As experiments turn out to be an expensive option for testing the nozzles, engineers majorly rely on the well-established computational and numerical methods/simulations. The working environment/extreme conditions are virtually created, different materials are chosen, and the computer-aided designed models are imported for finally optimizing the rocket nozzles.
In the present chapter, we develop a self-sustained machine learning model for in-flight prediction of mechanical properties and the gaps generated in a rocket nozzle based on the material properties and environmental conditions comprising of bolt preload, gas pressure variation, forces from regeneration channels, and thermal condition inputted. After modeling the rocket nozzle comprising of two conical-shaped parts clamped together with a series of nuts and bolts, its finite element analysis (FEA) is carried out to obtain the deformation, stress, and gaps generated between the two parts of the nozzle. This procedure is repeated for different combinations of environmental conditions generated using design of experiments to produce a large set of data containing the corresponding outputs. Several machine learning models are trained using this data to create a model comprising of Random Forest and XGBoost that is able to predict the mechanical properties of the rocket nozzle in-flight with an accuracy of more than 97% within a second.
