表面滚压强化工艺对FV520B钢表面完整性的影响及预测模型建立

目的 基于多元回归法和BP神经网络建立预测模型，实现对滚压后试件表面完整性指标的精准控制，从而指导实际加工生产。方法 以FV520B钢为研究对象，以滚压工艺参数（压强、进给量、滚压速度）为影响因素，以材料表面完整性指标（表面粗糙度、表面硬度、塑性变形层深度）为评价指标，设计了正交试验。通过对正交试验数据进行方差分析和信噪比分析，探究了滚压工艺参数对FV520B钢表面完整性的影响。基于正交试验数据构建了多元回归预测模型和BP神经网络预测模型，并对2种模型的有效性和精准度进行了分析和比较。结果 进给量对表面粗糙度有显著影响，随着进给量的增大，表面粗糙度也显著增大。压强和进给量对塑性变形层深度均有显著影响，且塑性变形层深度随着压强的增大而增大，随着进给量的增大而减小。多元回归法建立的预测模型的拟合度较差，而BP神经网络预测模型的实验值和预测值的相对误差均在10%以下，预测效果较好。结论 相比于多元回归预测模型，BP神经网络预测模型具有误差小、泛化性能好等优点，能够实现对滚压后试件表面完整性指标的精准控制，为实际的加工生产提供一定的指导。

Effect of Surface Rolling Strengthening Process on Surface Integrity of FV520B Steel and Establishment of Prediction Model

The work aims to establish a prediction model based on multiple regression method and BP neural network to achieve accurate control of surface integrity indicators of samples after surface rolling strengthening treatment process, so as to guide the actual processing production. With FV520B steel as the research object, orthogonal tests were designed with rolling process parameters (pressure, feed rate, and rolling speed) as affecting factors and measured surface integrity parameters (surface roughness, surface microhardness and plastic deformation layer) as evaluation indicators. The effect of rolling parameters on the surface integrity of FV520B steel was investigated by ANOVA and signal-to-noise ratio analysis on the orthogonal test data. A multiple regression prediction model and a BP neural network prediction model were constructed based on the orthogonal test data, and the effectiveness and accuracy of the two models were analyzed and compared. The feed rate had an evident effect on the surface roughness, which increased significantly with the increase of the feed rate. Both pressure and feed rate played a key role in the depth of the plastic deformation layer, which increased with the increase of the pressure, and decreased with the increase of the feed rate. The prediction model established by the multiple regression method had a poor fit, while the BP neural network prediction model had a better prediction with the relative error of both the experimental and predicted values below 10%. Compared with the multiple regression prediction model, the BP neural network prediction model has the advantages of small error and good generalization performance, which can realize the accurate control of surface integrity indicators of the sample after surface rolling strengthening treatment process and provide some guidance for the actual processing production.