A COMPARATIVE STUDY OF PHYSICS-BASED MODEL AND NEURAL NETWORK FOR THE 2SLG LAUNCHER
Abstract
Accurate prediction of muzzle velocity is critical for hypervelocity impact tests using two-stage light-gas (2SLG) launchers. This paper evaluates two complementary modeling approaches: a physics-based Python code applying lumped-parameter interior ballistic analysis, and a data-driven neural network (NN) trained on synthetic datasets generated by the physics-based model. The physics-based model captures isentropic reservoir blowdown, adiabatic pump-tube compression, and coupled piston-projectile dynamics. The NN model, featuring two hidden layers, achieves strong predictive performance with RMSE = 30.1 m/s and R² = 0.943. Both models identify the reservoir pressure, the light gas pressure, the diaphragm burst pressure, and the projectile mass as the top four controlling parameters in tests. The physics-based model enables local sensitivity analysis and efficient training-data generation, while the NN model facilitates near-instantaneous global parametric exploration. Together, they provide practical guidelines for 2SLG test parameter settings and experiment planning.
DOI
10.12783/ballistics26/37620
10.12783/ballistics26/37620
Refbacks
- There are currently no refbacks.