目的 通过训练不同实验参数条件下的确定性剪切增稠抛光的实验数据，建立基于BP神经网络的确定性剪切增稠抛光材料去除率模型，为实现抛光点材料的确定去除控制提供基础。方法 以BK7平面玻璃为抛光对象展开确定性剪切增稠抛光正交实验，根据正交实验分析结果，比较抛光头转速、抛光头与工件之间的间隙以及抛光液浓度三个因素，对抛光点材料去除率影响的权重，确定BP神经网络的输入参量。根据经验公式初步确定网络隐含层节点个数，并综合比较不同隐含层节点数目下的模型性能来确定整体网络结构，使用训练集实验数据训练网络模型，建立抛光点的材料去除率模型。结果 模型预测结果与实验结果对比表明，所建立的峰值去除率BP神经网络预测模型输出结果与实验结果之间的相对误差在6.8%以内，验证了所建立材料去除率模型的准确性。结论 传统理论模型难以精确描述确定性剪切增稠抛光的工艺参数与抛光区域材料峰值去除率之间复杂的非线性映射关系，而BP神经网络的自学习自适应能力能够克服这种问题，为确定性剪切增稠抛光去除率模型的建立提供新的思路。

The work aims to establish the removal rate model of deterministic shear thickening polishing (DSTP) based on BP neutral network by training the data of DSTP under different conditions, to provide basis for the deterministic removal control of material at polishing point. DSTP orthogonal experiment was conducted on BK7 flat glass, and the results were analyzed to compare the weight of each factor’s (polishing head speed, polishing gap, concentration of polishing slurry) impact on material removal rate of polishing point, to determine the input parameters of BP neural network. The number of hidden layer nodes was determined by experimental equation initially, and then the whole structure of network model was constructed by comparing the performance of the model under different hidden layer nodes. The final material removal rate model was established based on the experimental data designed network trained by the model. The comparison between the model prediction results and the experimental results showed that the relative error was less than 6.8% between the output of the established BP neural network prediction model and the measured results, proving the accuracy of the prediction model. The traditional theoretical model is difficult to accurately describe the complicated nonlinear mapping relationship between the process parameters of DSTP and the peak removal rate. The self-learning and adaptive ability of BP neural network can overcome this problem and provide a novel strategy for building the removal rate model of the deterministic shear thickening polishing process.