Machine learning-driven multi-objective optimization of straight-blade paddle mixers for efficient mixing in melt-cast explosives systems.

Abstract

In the production of melt-cast explosives, efficient mixing in straight-blade paddle mixers is critical to product performance. Taking the classical trinitrotoluene-cyclotrimethylenetrinitramine (TNT-RDX) system as an example, this paper presents a machine learning-based multi-objective optimization algorithm that addresses the conflicting requirements of maximizing mean flow velocity and minimizing maximum pressure difference during mixing. By training random forest regression models on historical experimental and simulation data, the algorithm accurately predicts mixing outcomes from structural and operational parameters (e.g., blade number, width, length, angle, and rotation speed). These surrogate models are coupled with the nondominated sorting genetic algorithm II (NSGA-II) multi-objective evolutionary optimizer to identify Pareto-optimal solutions. Combined with the experimental verification of the simulated materials, the design method of the straight paddle system comparable to other high-efficiency stirring paddles was obtained. In addition, the formation mechanism of non-Newtonian fluid in the system of TNT-RDX is investigated by quantum chemical method, and the optimization results of the straight paddle for the non-Newtonian fluid system are numerically computed, which proves the versatility of the straight paddle design method.