A New Network Training Model for the Processing of Work Equipment Timing Data

Currently, most of the work tool wear state detection methods are based on machine learning models. This does not require complex data preprocessing of the raw data. By adaptively extracting high-dimensional features hidden in the tool processing signal and designing a reasonable network depth, ideal results can be obtained. However, traditional neural network training models often cannot effectively solve the problem of the correlation between timing signals generated in automated production lines, which has the problems of poor correlation and low training accuracy. This article proposes a joint collaborative network training model D-GRU based on Convolutional Neural Networks (CNN) and Deep Bidirectional Gated Recurrent Units (BiGRU), and innovatively introduces the idea of attention mechanism into the improved D-GRU network model. Then, the simulation results demonstrate that the network can effectively address the issue of timing signal back-and-forth correlation being ignored by a single convolutional neural network, thereby avoiding the gradients associated with recursive neural networks.