目的 建立能够准确预测不同喷丸工艺参数下工件表层显微硬度的数学模型。方法 以42CrMo钢作为研究对象，采用正交实验法设计喷丸实验方案和逐点测量法测量0~320 μm层深内的显微硬度，以实验数据为样本，基于BP神经网络建立42CrMo钢受喷后表层显微硬度的预测模型，并利用遗传算法(GA)对BP神经网络结构进行优化，建立基于GA-BP神经网络的42CrMo钢受喷后表层显微硬度模型。结果 将实验数据集用于模型的训练，BP神经网络模型和GA-BP神经网络模型训练的相关系数R均为0.97左右，两种模型的训练效果均较好。对比20组测试集的模型预测值和实验值发现，BP神经网络模型预测值与实验值之间的相对误差的最大值和平均值分别为3.5%和1.1%，相比之下，经遗传算法优化的BP神经网络(GA-BP)模型预测值与实验值的相对误差的最大值和平均值仅为2.9%和0.7%。GA-BP神经网络模型具有更高的预测精度和稳定性。结论 经GA遗传算法优化的BP神经网络(GA-BP)更适合用于建立受喷工件表层显微硬度的预测模型，可为其在工程上的应用提供一定的参考。

The work aims to establish a mathematical model that can accurately predict the surface microhardness of peened parts under different shot peening parameters. Taking 42CrMo steel as the research object, the shot peening experiment plan was designed by orthogonal experiment method and the point-by-point measurement method was used to measure the microhardness in the depth of 0~320 μm. The BP neural network was used to establish the surface microhardness prediction model of 42CrMo steel after shot peening. Meanwhile, genetic algorithm (GA) was used to optimize the structure of BP neural network, and the surface microhardness prediction model of 42CrMo steel after shot peening based on GA-BP neural network was established. Velocity, diameter, coverage and depth from surface were set as the input parameters, and the surface microhardness was set as the output parameter in both two models. The experimental data was divided into two parts, where the training set was used for the training of the two models, the correlation coefficient R of BP neural network model and GA-BP neural network model was about 0.97, and the training effect of the two models was good. By comparing the predicted value of two models and the experimental value of 20 groups of test set, it was found that the maximum and average relative errors between the predicted value of the BP neural network model and the experimental value were 3.5% and 1.1%, respectively. The maximum and average relative errors between the predicted value of the GA-BP neural network model and the experimental value were only 2.9% and 0.7%, respectively. The GA-BP neural network model had higher prediction accuracy and stability. The BP neural network optimized by genetic algorithm (GA-BP) is more suitable for establishing the prediction model of the surface microhardness of peened parts, which can provide some guidance for the industrial application.