CNN-GRU network-based force prediction approach for variable working condition milling clamping points of deformable parts

Improper clamping is one of the major causes part deformation. Improving fixture arrangement through force analysis points an effective means to suppress or improve machining However, existing research focuses on monitoring and off-line optimization point force, which has a certain lag deformation control, it difficult predict due time-varying coupling effect multiple factors such as process parameters, cutting in process. Inspired by excellent performance convolutional neural networks gated recurrent feature extraction learning temporal association laws, this paper proposes CNN-GRU-based method for predicting state under variable working conditions. Firstly, prediction model during milling with conditions established. Secondly, network designed extract features dynamic coupled Then, units constructed learn correlation law between forces achieve machining. Finally, was verified piston skirt. The results show that CNN-GRU can effectively force. In addition, higher computational efficiency accuracy compared CNN-LSTM, CNN-RNN CNN-BP.