Additional Validation of Physics-Informed Transfer Learning for Out-of-Sample Vapor Pressure Prediction: Organic Nitrate Esters

Author(s):

Michael J. McQuaid, Joshua L. Lansford, and Christopher P. Stone

Abstract:

Toward creating Hertz–Langmuir–Knudsen (HLK) equations for calculating volatilization rates for energetic materials (EMs) formulated with nitrocellulose (NC), we evaluated a directed-message passing neural network (D-MPNN) model’s predictions for the saturated vapor pressures (P_sat) of organic nitrate esters. Comparing D-MPNN-predicted and measurement-based values for 32 different alkyl nitrates at temperatures (T) from 250 to 1000 K, we observed reasonable agreement in most cases. Moreover, in cases in which discrepancies were significant, they could be rationalized. Based on those findings, we concluded that the model’s P_sat(T) predictions warranted use in formulating the HLK equations we sought. Volatilization-rate predictions produced by those equations were compared with values produced by a well-regarded pyrolysis law for double-base propellants. The results indicated that the HLK equations have the potential to produce better predictions for the rate at which (pyrolyzed) NC will volatilize at temperatures less than 400 K. Thus, they may find application in models for simulating the response of NC-containing EMs to thermal loads relevant to ignition and cook-off scenarios.