%0 Journal Article %1 WANG20191 %A Wang, Jinjiang %A Yan, Jianxing %A Li, Chen %A Gao, Robert X. %A Zhao, Rui %D 2019 %J Computers in Industry %K %P 1 - 14 %R https://doi.org/10.1016/j.compind.2019.06.001 %T Deep heterogeneous GRU model for predictive analytics in smart manufacturing: Application to tool wear prediction %U http://www.sciencedirect.com/science/article/pii/S0166361519300260 %V 111 %X Smart manufacturing arises the growing demand for predictive analytics to forecast the deterioration and reliability of equipment. Many machine learning algorithms, especially deep learning, have been investigated for the above tasks. However, long-term prediction is still considered as a challenging issue. To address this problem, this paper presents a hybrid prediction scheme accomplished by a newly developed deep heterogeneous GRU model, along with local feature extraction. Specifically, to capture the temporal pattern hidden in the sequential input, a local feature extraction method is designed by integrating expertise knowledge into the deep learning model for enhanced feature learning. Next, an intermediate layer is designed in the deep heterogeneous GRU model structure to capture the inherent relation for long-term prediction. The proposed model is optimized by systematic feature engineering and optimal hyperparameter searching. Finally, experimental studies on tool wear test are performed to validate the superiority of the presented model over conventional approaches.

@article{WANG20191, abstract = {Smart manufacturing arises the growing demand for predictive analytics to forecast the deterioration and reliability of equipment. Many machine learning algorithms, especially deep learning, have been investigated for the above tasks. However, long-term prediction is still considered as a challenging issue. To address this problem, this paper presents a hybrid prediction scheme accomplished by a newly developed deep heterogeneous GRU model, along with local feature extraction. Specifically, to capture the temporal pattern hidden in the sequential input, a local feature extraction method is designed by integrating expertise knowledge into the deep learning model for enhanced feature learning. Next, an intermediate layer is designed in the deep heterogeneous GRU model structure to capture the inherent relation for long-term prediction. The proposed model is optimized by systematic feature engineering and optimal hyperparameter searching. Finally, experimental studies on tool wear test are performed to validate the superiority of the presented model over conventional approaches.}, added-at = {2020-12-14T10:02:01.000+0100}, author = {Wang, Jinjiang and Yan, Jianxing and Li, Chen and Gao, Robert X. and Zhao, Rui}, biburl = {https://www.bibsonomy.org/bibtex/2210e45c3d28fb017372e1ac33f50e8b1/ninawue}, doi = {https://doi.org/10.1016/j.compind.2019.06.001}, interhash = {2ae02db63fe75be3e3b0793c457ad443}, intrahash = {210e45c3d28fb017372e1ac33f50e8b1}, issn = {0166-3615}, journal = {Computers in Industry}, keywords = {}, pages = {1 - 14}, timestamp = {2020-12-14T10:02:01.000+0100}, title = {Deep heterogeneous GRU model for predictive analytics in smart manufacturing: Application to tool wear prediction}, url = {http://www.sciencedirect.com/science/article/pii/S0166361519300260}, volume = 111, year = 2019 }