目的 研究激光熔覆关键工艺参数（激光功率、扫描速度、送粉速率）与单道熔覆层宏观形貌（宽度、高度、熔池深度）之间的数量关系，以实现对WC-Co50复合熔覆层形貌的预测，从而为牙轮钻头的修复提供参考。方法 设计不同的实验参数，利用4 kW光纤激光器在牙轮钻头钢15MnNi4Mo表面熔覆单道WC-Co50复合涂层。采用工业显微镜观察单道熔覆层的横截面宏观形貌，并测量其三维尺寸。在上述形貌参数的基础上，分别运用多元线性回归分析和人工神经网络方法，建立关键工艺参数与熔覆层宏观形貌之间的关系模型，并将实验结果与模型预测结果进行对比。结果 总体来讲，神经网络对熔覆层形貌的预测结果更为精确，平均相对误差为5.3187%；多元线性回归分析预测的平均相对误差为6.0028%。分析表明，对熔覆层宽度的预测结果最精确，两种方法的平均相对误差仅为1.2999%；对高度及熔池深度的预测结果稍差，平均相对误差分别为8.0586%和7.6237%。结论 两种预测方法都具有较高的精度，但神经网络法函数关系不明确，运算过程复杂，需要通过进一步的算法优化来提高预测精度。

The work aims to study the quantitative relationship among the key technological parameters (laser power, scanning speed, powder feeding rate) and the geometrical characteristics (width, height, bath depth) of single-track cladding layer, so as to predict the geometrical size of WC-Co50 composite coating and provide a reference for the repair of cone bit. Different experiments were designed and single-track WC-Co50 composite coatings were deposited on the surface of cone bit 15MnNi4Mo steel by 4 kW fiber laser. Meanwhile, the cross-sectional macrostructure of the single-track coating was observed and its three-dimensional size was measured by industrial microscopy. Based on the above data, multiple linear regression analysis and artificial neural network were used to establish the mathematical models between the key process parameters and the geometrical characteristics of the cladding layer. Then, the experimental results were compared with those predicted by the models. The neural network model was more accurate in predicting the cladding layer characteristics, with an average relative error of 5.3187%, while the mean relative error of multiple linear regression analysis model was 6.0028%. The analysis indicated that the prediction result of the coating width was the most accurate, and the average relative error of the two methods was only 1.2999%. Meanwhile, the prediction results of the height and molten pool depth were slightly worse, with average relative errors of 8.0586% and 7.6237% respectively. The both prediction methods have high accuracy, but the function of the neural network is vague, and the calculation process is complex. The both prediction methods have high accuracy. However, the function relationship of neural network is not clear and the operation process is complex, so further algorithm optimization is needed to improve the prediction accuracy.