Machine learning-based multi-objective optimization of ceramic composite armor plates via the NSGA-II genetic algorithm

To address increasing battlefield demands for enhanced mobility and protection capabilities, ceramic composite armor has gained prominence as a leading candidate material system due to its exceptional lightweight properties and superior ballistic performance. Owing to the fact that the long duration and limited efficiency of experimental tests preclude the generation of comprehensive datasets needed for robust multi-objective optimization, integrating numerical simulation with machine learning models has become an established methodology. A finite element model validated through experimental calibration was integrated with Latin hypercube sampling to generate critical ballistic performance data for steel/SiC/UHMWPE multi-layered armor plates. These computational datasets were utilized to train a multi-layer perceptron (MLP) neural network, establishing predictive correlations between layer thickness variations and two conflicting objectives: specific energy absorption (SEA) and residual kinetic energy. A comparative assessment of predictive performance metrics among three machine learning models (MLP, SVM, and RF) was conducted. The optimized MLP demonstrated superior predictive accuracy and lower training loss relative to SVM and RF counterparts. Subsequent implementation of the NSGA-II algorithm on the MLP model generated a comprehensive Pareto frontier that quantifies inherent design trade-offs, while a knee point identification approach based on the minimum distance selection method facilitated optimal solution selection. The optimized ceramic composite armor plate achieved 24.3 % SEA improvement coupled with 14.5 % residual kinetic energy reduction compared to baseline design. This integrated machine learning and evolutionary optimization approach delivers actionable guidelines for advancing lightweight armor systems.