Journal
JOURNAL OF INTELLIGENT MANUFACTURING
Volume 34, Issue 2, Pages 885-902Publisher
SPRINGER
DOI: 10.1007/s10845-022-01954-9
Keywords
Heterogeneous asymmetric convolution kernel; DenseNet; Depth-gated recurrent unit; Feature extraction; Tool wear prediction; Dilated convolution kernel
Ask authors/readers for more resources
Accurately predicting the machining tool condition during the cutting process is crucial for improving tool life, production quality, productivity, labor and maintenance costs, and reducing workplace accidents. This study proposes an integrated prediction scheme based on deep learning algorithms to address the critical challenge of extracting and fusing feature and fusion information from multi-sensor signals related to tool wear. The results show that the proposed model outperforms other tool wear prediction models in terms of both accuracy and generalization.
An accurate prediction of the machining tool condition during the cutting process is crucial for enhancing the tool life, improving the production quality and productivity, optimizing the labor and maintenance costs, and reducing workplace accidents. Currently, tool condition monitoring is usually based on machine learning algorithms, especially deep learning algorithms, to establish the relationship between sensor signals and tool wear. However, deep mining of feature and fusion information of multi-sensor signals, which are strongly related to the tool wear, is a critical challenge. To address this issue, in this study, an integrated prediction scheme is proposed based on deep learning algorithms. The scheme first extracts the local features of a single sequence and a multi-dimensional sequence from DenseNet incorporating a heterogeneous asymmetric convolution kernel. To obtain more perceptual historical data, a dilation scheme is used to extract features from a single sequence, and one-dimensional dilated convolution kernels with different dilation rates are utilized to obtain the differential features. At the same time, asymmetric one-dimensional and two-dimensional convolution kernels are employed to extract the features of the multi-dimensional signal. Ultimately, all the features are fused. Then, the time-series features hidden in the sequence are extracted by establishing a depth-gated recurrent unit. Finally, the extracted in-depth features are fed to the deep fully connected layer to achieve the mapping between features and tool wear values through linear regression. The results indicate that the average errors of the proposed model are less than 8%, and this model outperforms the other tool wear prediction models in terms of both accuracy and generalization.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available