基于跨层级协同强化的 CFRP 孔边挤压强度预测

To address the challenge of insufficient experimental data on the bearing strength at the hole edge of carbon fiber-reinforced polymer (CFRP) composites, which hindered the full understanding of the influence mechanisms of various features and led to difficulties in accurately predicting extrusion strength, a CFRP extrusion strength prediction method based on cross-hierarchical synergistic reinforcement (CHSR) was proposed. Firstly, due to the limited experimental data volume relative to the high dimensionality of input features, it is challenging to comprehensively extract knowledge embedded in all features simultaneously. To address this, a modular strategy is employed to partition features into low-level features and high-level features based on their complexity. Secondly, a low-scale model, namely the low-level feature mapping model, was constructed. The low-level feature knowledge was extracted through neural network pre-training, and the knowledge was collaboratively transferred to the fixed domain of the high-scale model. Subsequently, on the basis of locking the underlying knowledge in the fixed domain of the high-scale model, reinforcement learning was applied to dynamically optimize the weights and thresholds of high-level feature representations, enabling adaptive refinement of their contributions to the prediction task. Then, cross-hierarchical gradient collaborative propagation was implemented to harmonize feature interactions between low-level and high-level domains, thereby achieving accurate prediction of hole-edge compressive strength under complex feature coupling. Finally, taking prediction of X850 material bearing strength at the edge of the bolted connection hole as a typical case, variable layup sample prediction was carried out. The results show that the mean error of the two paving layers is 0.10%, 0.48%, the standard deviation error is 2.64%, 3.50%, and the allowable value error is 0.29%, 0.13%. Compared with the traditional machine learning algorithm, CHSR can achieve high precision prediction of X850 bearing strength at the hole edge.