Deep learning-based online image analysis of the water suspension granulation process for 1,3,5-trinitro-1,3,5-triazacyclohexane-based polymer-bonded explosives

Abstract

Polymer-bonded explosives (PBXs) are typically prepared via water suspension granulation (WSG). However, few studies have reported on online monitoring during WSG. In this study, the WSG process for 1,3,5-trinitro-1,3,5-triazacyclohexane (RDX)-based PBXs was monitored using online imaging and infrared detection technology. The images of the particles and droplets obtained during the WSG process were then identified by employing the Mask Region-based Convolutional Neural Network (Mask R-CNN). The characteristic peak of ethyl acetate identified in the online infrared spectra during the WSG process is a key indicator for binder concentration. The experimental results show that the changes in the particle size distribution and particle morphology of the explosive modeling powders are consistent with the actual results, online infrared detection can provide effective guidance for the addition of the binder solution. This new and effective approach can be used to study WSG and has considerable potential for process optimization and quality control.