Search

In the abrasive belt grinding process, actual material removal is an important parameter that affects its accuracy. At present, for obtaining the actual material removal, offline measurements are required to establish the mathematical prediction model. To improve the accuracy and efficiency of abrasive belt machine grinding, this paper proposes a novel method for monitoring material removal using multiple sensors and a two-dimensional (2D) convolutional neural network (2D-CNN) learning algorithm. In this method, features of multiple types (color, texture, and shape) are extracted from vision signals, and that of multiple domains (time, frequency, and time–frequency domain) are extracted from sound and tactile signals. These features are constructed into a 2D feature matrix as the input model, and the 2D-CNN prediction model is established between the multisensor features and the material removal rate of the abrasive belt grinding process. An experimental dataset is used to train and verify the established model. The results show that the proposed method can identify that sensor signals are sensitive to the material removal rate. After optimizing and tuning the model parameters, the coefficient of determination of the prediction results is as high as 94.5% and the root mean square error is 0.017. Therefore, the proposed method can be employed for the prediction of material removal rate for different belt specifications and different grinding parameters. Compared to traditional machine learning methods, this method can yield better training results without feature selection and optimization. [ABSTRACT FROM AUTHOR]