Robotic abrasive belt grinding can be widely used to improve the surface quality of complex workpieces. Due to the elastic characteristics in the grinding process, feasible processing parameters cannot be fully predicted by existing cutting models given a desired cutting depth. Thus, abrasive belt grinding in industrial production still relies mainly on operator experience and intelligence. Robust theoretical guidance is provided for input parameters setting of robotic belt grinding. A new method called Inverse Input Prediction Model (IIPM) is proposed. IIPM has two processes to calculate the optimal input parameters for robotic abrasive belt grinding. First, the forward process is to establish the correspondence between known processing parameters and the unknown grinding depth. Support Vectors Regression (ε-SVR) is an application of support vector machine (SVM) to regress process parameters. The forward process use ε-SVR method to construct a nonlinear regression of grinding parameters and grinding depth. Second, the inverse process of dynamic grid search (DGS) method is proposed to predict the optimal processing parameters according to the required grinding depth. The experimental results in the forward process demonstrate that the ε-SVR model has the lowest prediction error compared with BP neural network and polynomial regression. DGS and particle swarm optimization (PSO) search methods are used in the inverse process to predict grinding parameters at the required grinding depth. Experiment results of the inverse model prediction show that the predicted depth is very closely fit the actual grinding depth. The proposed method can be used to find the stable and reliable sequence of continuous path grinding parameters by given the desired grinding depth.

机器人砂带研磨可广泛用于改善复杂工件的表面质量。由于磨削过程中的弹性特性，在给定所需切削深度的情况下，现有切削模型无法完全预测可行的加工参数。因此，工业生产中的砂带磨削仍然主要依赖于操作员的经验和智慧。为机器人皮带磨削的输入参数设置提供了可靠的理论指导。提出了一种称为逆输入预测模型（IIPM）的新方法。IIPM有两个过程可以计算出机器人砂带磨削的最佳输入参数。首先，正向过程是建立已知加工参数和未知磨削深度之间的对应关系。支持向量回归（ε-SVR）是支持向量机（SVM）用于回归过程参数的应用程序。正向过程使用ε-SVR方法构造磨削参数和磨削深度的非线性回归。其次，提出了动态网格搜索（DGS）方法的逆过程，以根据所需的磨削深度预测最佳加工参数。正演过程中的实验结果表明，与BP神经网络和多项式回归相比，ε-SVR模型的预测误差最低。在逆过程中使用DGS和粒子群优化（PSO）搜索方法来预测所需磨削深度处的磨削参数。逆模型预测的实验结果表明，预测深度与实际磨削深度非常接近。