Prediction of puncture key parameters in aramid stab-resistant fabrics based on surface damage images using artificial neural networks

Stab-resistant clothing significantly contribute to personal protection. In the field of stab resistance, traditional methods typically use the known impact conditions to evaluate the protection performance and damage of stab-resistant materials. However, these methods are unable to backtrack impact information from known damage, which makes it difficult to determine impactor characteristics. This study introduces a novel puncture damage prediction model capable of predicting the impact kinetic energy, peak puncture force, and number of penetration layers of aramid stab-resistant fabrics solely from surface damage images under various puncture conditions. First, the different puncture damages images and their corresponding parameters are obtained through dynamic stabbing tests and image acquisition system. Second, the segmentation network (named SAN_SE model) developed in this study overcomes the complexity of the surface texture of fiber-reinforced composites and achieves precise segmentation of damage regions. The training loss is stable at 1.5 × 10−4. Then a classification model is constructed to establish a relationship between the images and puncture parameters, followed by the application of transfer learning to derive a regression model from the classification model. The error of this regression model is below 6 %. Finally, a real-time puncture damage prediction system is constructed, applying this puncture damage prediction model to actual damage scenarios. The system achieves an accuracy of 88.57 % in predicting the number of penetration layers and could execute single images within 0.025s. The puncture damage prediction model proposed in this study is applicable to real-time monitoring systems in medical and military fields, such as injury assessment and counter-surveillance.