Inversion of Thermophysical Properties of Melt-cast Explosives via Artificial Neural Networks

The conventional methods for determining the thermophysical properties predominantly rely on the experimental measurements, which are often time-consuming, labor-intensive, and inadequate for obtaining the continuous temperature-dependent functions. To address these limitations, this paper proposes an inversion method based on a two-layer Backpropagation (BP) neural network optimized with Bayesian regularization. This method is used to accurately acquire key thermophysical parameters (thermal conductivity and specific heat capacity) in the solidification process of melt-cast explosives. A mapping relationship is constructed using the cooling time differences from multiple radial temperature measurement points as inputs and the corresponding thermophysical properties at specific temperatures as outputs. A substantial set of training samples are generated via finite element simulation, and the network is subsequently trained and validated with experimental data. The results demonstrate that the neural network model achieves a coefficient of determination (R²) of 0.994 for the inversion of both thermal conductivity and specific heat capacity. The average error between the inverted results and the measured data is less than 5%, indicating high predictive accuracy and strong engineering applicability. This method provides a new approach for acquiring the thermophysical parameters of melt-cast explosives.