Abstract: Accurately prediction of the material removal profile(MRP) in the robotic grinding process plays an extremely important role in improving the precision of machining precision, as well as realizing the closed-loop control. However, accompanied by the high elasticity and wear resistance, it is extremely challenging to predict the MRP of the rubber materials used in large and complex components such as the sacrificial layer of the aircraft. An MRP prediction model of the rubber in robotic grinding of the large components is developed based on mega trend diffusion and extreme gradient boosting(MTD-XGBoost). Firstly, the grinding mechanism of the rubber is analysed and the independent factors of the MRP are determined. Then, a virtual sample generation method based on triangle membership function is proposed to solve the problems, such as the sample scarcity and poor model accuracy. Furthermore, clustering denoising is carried out on the sample data based on the k-means clustering algorithm, and the MRP prediction model is established to map the relationship between the MRP and the process parameters according to the Extreme Gradient Boosting and virtual sample generation method. Finally, a comparative experiment is carried out, the results of which show that the proposed virtual sample generation method can reduce the mean absolute error by 30.3% at most. Compared with Support Vector Machine, Bayesian regression and other methods, the proposed prediction model shows a better generalization performance under small datasets, and ultimately realized the MRP prediction of the rubber with the average relative error less than 10%.

Key words: robotic grinding, material removal profile, virtual sample, XGBoost