PCD Milling Cutter Remaining Useful Life Prediction for Titanium and Aluminum Mirror Milling by Using S2S-LSTM Deep Learning Technology
Shang-Liang Chena, Kuei-Ming Leeb, Yen-Hsiang Huangc, Yu-Ting Lud, Yu-Fu Line and Ho-Chuan Huangf
aInstitute of Manufacturing Information and Systems, National Cheng Kung University, Tainan, 70101, R.O.C. bInstitute of Manufacturing Information and Systems, National Cheng Kung University, Tainan, 70101, R.O.C. cInstitute of Manufacturing Information and Systems, National Cheng Kung University, Tainan, 70101, R.O.C. dMicro Mechanical Manufacturing R&D Department, Metal Industries Research & Development Centre, Kaohsiung, 811, R.O.C. eMicro Mechanical Manufacturing R&D Department, Metal Industries Research & Development Centre, Kaohsiung, 811, R.O.C. fDepartment of Intelligent Commerce, National Kaohsiung University of Applied Sciences, Kaohsiung, 824, R.O.C.
|
Abstract:
Both the titanium alloy and aluminum alloy cutting by using Polycrystalline Diamond (PCD) milling cutter for obtaining mirror milling surface results are important processing technologies in the industry. To improve the production efficiency or enhance the cutting performance of this cutting technology, the Remaining Useful Life (RUL) prediction of PCD milling cutter becomes one of the major issues nowadays. The Sequence to Sequence Long Short-term Memory (S2S-LSTM) is used in this research as the prediction model to carry out PCD milling cutter’s RUL prediction, and two times of PCD milling cutting experiments for titanium and aluminum alloy are designed and carried out. In the experiments, the data of the vibration signal, sound signal, and the surface roughnesses of the workpieces are measured and used as the datasets. The prediction model yielded F1-scores of 98.1% and 95.8% by using the validation datasets of the two experiments. The proposed model is also compared with other AI (Artificial Intelligent) models, such as RNN (Recurrent Neural Network), GRU (Gated Recurrent Unit), and LSTM under the same batch size, epoch, learning rate, and other hyper-parameters.
|