A milling cutter state recognition method based on multi-source heterogeneous data fusion

In multi-varieties and small-batch aerospace enterprises, due to the flexible and mixed production, multi-process cross-parallel, complex process, and production environment, the tool wear data set is small, multi-source heterogeneous, and under-sampled. As a result, the tool state recognition is difficult and low precision. To handle these issues, a method of machine tool status recognition based on multi-source heterogeneous data fusion is proposed. First, the compressed sensing method is used to compress multi-source data to balance samples and improve sample sparsity, and random Gaussian noise is added to the compressed data to enhance the robustness of the network. Then, a stack sparse auto-encoder network based on improved D-S evidence theory and the Dropout method is formulated. The unsupervised learning combined with supervised learning and the Dropout method is designed to train the network to cope with the problem of overfitting of a deep learning network and the recognition accuracy of small sample multi-source heterogeneous data. Finally, to assess the effectiveness of the proposed method, comparison experiments are carried out between the proposed method, the artificial feature extraction method, and stacked sparse autoencoder (SSAE) method. The identification results show that the proposed method has better recognition accuracy and generalization performance, which can accurately reflect the production line state of multi-varieties and small batch aerospace enterprises.