A prediction model for surface roughness classification of computer numerical control drills based on fusion of local time–frequency features and global time–frequency features of the Mobilenet_v3_small model is proposed. Correct rates of the training set, validation set, and test set are 85.2%, 84%, and 85.4%. Compared with industrial lightweight network models, this model improved the correctness rates on the training set, validation set, and test set by about 10%, 9%, and 13%, respectively.

In the research and development process of underwater motion control of the beaver-like robot, the hind leg and tail are the main actuators. In order to estimate the driving force of the robot autonomously during underwater movement, we propose this model, which can quickly and easily infer the force of the hind leg on the body. The correctness of this model is verified by experiments and simulations.