隨著深度學習應用於製造業越加廣泛，只要擁有大量製程資料及量測資料，就能夠透過模型獲得兩者之間的關聯性，穩定產線的良率。然而，工具機產業時常需要因應製程的改變或者因量測樣本困難，無法時常即時地取得大量的量測資料來回饋深度學習模型。因此，在這樣的背景下發展了遷移學習，透過相似資料之間的知識遷移，讓模型更新不需大量的樣本，透過模型之間的權重凍結與層的微調，減少訓練完成需要的時間及資源。
運用遷移學習於有限樣本環境之深度神經網路逐漸更新機制，為基於產線樣本收集的模式，發展而生的一套遷移學習方法。以卷積類神經網路VGG 19為例，讓產線每個生產批次回傳的成對樣本，透過機制設計的資料之間的相似度 ( TLI ) 以及樣本累積數量( RS ) 指標選擇適宜的遷移策略，適時地調整凍結層以及結構，讓模型於各個生產階段也能擁有堪用的監測水準，直至完成更新。學術上我們將樣本相似度以及樣本數量進行實際量化，產生指標對應四個具備物理意義的策略，豐富遷移學習的機制，讓卷積的凍結方針更有依據。產線價值上，我們減少深度學習模型重新建模所需要的樣本，讓其能夠更快地實行產線監測，使深度學習應用於業界更具實用性及意義。

As deep learning has becoming much more widely adopted in the manufacturing industries, once a large amount of process data and metrology data are collected, the correlation between the two can be obtained through the model to stabilize the yield of the production lines. However, due to the reasons that the machine tool industry should adjust often in response to changes in manufacturing processes or because of the difficulty of measuring samples, collecting a large amount of measurement data in real time to feedback the deep learning models is not an easy task. Thus, transfer learning (TL) has been developed to solve such problems. Knowledge transfer among similar data eliminates the constraint of needing a large number of samples for model update. Through the freezing of weights between models and the fine-tuning of layers, the time and resources required to complete the training can be reduced.
Transfer-Learning based Gradual Improvement Scheme (TLGIS) applies TL to the environment of limited samples, and it is a set of transfer learning methods developed based on the sample collection model of the production lines. Take the Convolutional Neural Network VGG 19 as an example, let the paired samples returned from each production batch of the production line be selected appropriately through the similarity (TLI) and cumulative number of samples (RS) between the data designed by the mechanism. According to the migration strategy, the frozen layer and structure are adjusted in a timely manner, so that the model can also have a usable monitoring level in each production stage until the update is completed.
Academically, the sample similarity and the number of samples are quantified to generate indicators corresponding to four strategies with physical meaning, enrich the transfer learning mechanism, and establish the base for the freezing policy of convolution. In terms of values of the production lines, the samples required for remodeling the deep learning model are reduced to speed up the production line monitoring, which makes deep learning applications more practical and meaningful in the industry.

[1] H. Tieng, Y. Li, K. Tseng, H. Yang and F.-T. Cheng, “An Automated Dynamic-Balancing-Inspection Scheme for Wheel Machining,” IEEE Robotics and Automation Letters, vol. 5, no. 2, pp. 2224-2231, April 2020, doi: 10.1109/LRA.2020.2970953.
[2] H. Tieng, T. Tsai, C. Chen, H. Yang, J. Huang, F. Cheng, “Automatic Virtual Metrology and Deformation Fusion Scheme for Engine-Case Manufacturing,” IEEE Robotics and Automation Letters, vol. 3, no. 2, pp. 934-941, Apr. 2018, doi: 10.1109/LRA.2018.2792690.
[3] F.-T. Cheng, H. Tieng, H.-C. Yang, M.-H. Hung, Y.-C. Lin, C.-F. Wei, and, Z.-Y. Shieh, “Industry 4.1 for wheel machining automation,” IEEE Robotics and Automation Letters, vol.1, no.1, pp. 332-339, Jan. 2016, doi: 10.1109/LRA.2016.2517208.
[4] 陳宏勳。2019。應用AVM 於大型臥式搪铣床。碩士論文。台南：成功大學 製造資訊與系統研究所
[5] Y. LeCun, Y. Bengio, and G. Hinton, “Deep learning,” nature, vol. 521, pp. 436-444, May 2015.
[6] Simonyan K, Zisserman A. “Very Deep Convolutional Networks for Large-Scale Image Recognition,” arXiv preprint arXiv:1409.1556. 2014 Sep 4.
[7] F. Yang, W. Zhang, L. Tao, and J. Ma, “Transfer learning strategies for deep learning-based PHM algorithms,” Applied Sciences, vol. 10, no 7, pp. 2361, Mar. 2020, doi: doi.org/10.3390/app10072361
[8] Chi-Cheng Jou, and Ta-Wen Hung, ‘‘Applying Time-Series Similarity Measurement to Fault Detection and Classification” Institutional Responsory of NCTU, Sep. 2016
[9] W. Wang, G. Lyu, Y. Shi and X. Liang, "Time Series Clustering Based on Dynamic Time Warping," 2018 IEEE 9th International Conference on Software Engineering and Service Science (ICSESS), 2018, pp. 487-490, doi: 10.1109/ICSESS.2018.8663857
[10] H. Tieng, C.-F. Chen, F.-T. Cheng, and H.-C. Yang, “Automatic virtual metrology and target value adjustment for mass customization,” IEEE Robotics and Automation Letters, vol.2, no. 2, pp. 546-553, Apr. 2017, doi: 10.1109/LRA.2016.2645507
[11] D. Neupane, and J. Seok, “Bearing fault detection and diagnosis using case western reserve university dataset with deep learning approaches: a review,” IEEE Access, vol.8, pp. 93155-93178, Apr. 2020.
[12] C. Tastimur, M. Karakose and E. Akin, “Vibration Signal Processing Based Bearing Defect Diagnosis with Transfer Learning,” 2019 1st International Informatics and Software Engineering Conference (UBMYK), 2019, pp. 1-5, doi: 10.1109/UBMYK48245.2019.8965451.
[13] J. Cui, R. Zhou and Z. Wang, “Time Frequency Feature Analysis of Rolling Bearing Fault Based on Deep Transfer Learning,” 2021 4th IEEE International Conference on Industrial Cyber-Physical Systems (ICPS), 2021, pp. 679-685, doi: 10.1109/ICPS49255.2021.9468216.
[14] T. Lu, F. Yu, B. Han and J. Wang, “A Generic Intelligent Bearing Fault Diagnosis System Using Convolutional Neural Networks With Transfer Learning,” IEEE Access, vol. 8, pp. 164807-164814, 2020, doi: 10.1109/ACCESS.2020.3022840.
[15] Fei Shen, Chao Chen, R. Yan and R. X. Gao, “Bearing fault diagnosis based on SVD feature extraction and transfer learning classification,” 2015 Prognostics and System Health Management Conference (PHM), 2015, pp. 1-6, doi: 10.1109/PHM.2015.7380088.
[16] X. Wang, C. Shen, M. Xia, D. Wang, J. Zhu, and Z. Zhu, “Multi-scale deep intra-class transfer learning for bearing fault diagnosis,” Reliability Engineering & System Safety, vol. 202, pp.107050, Oct. 2020, doi: doi.org/10.1016/j.ress.2020.107050
[17] P. Cao, S. Zhang, and J. Tang, “Preprocessing-free gear fault diagnosis using small datasets with deep convolutional neural network-based transfer learning,” IEEE Access, vol.6, pp.26241-26253, May 2018, doi: 10.1109/ACCESS.2018.2837621
[18] Z. Chen, K. Gryllias, and W. Li, “Intelligent fault diagnosis for rotary machinery using transferable convolutional neural network,” IEEE Transactions on Industrial Informatics, vol.16, no.1, pp. 339-349, May 2019, doi: 10.1109/TII.2019.2917233
[19] W. Zhang, G. Peng, and C. Li, ‘‘Bearings fault diagnosis based on convolutional neural networks with 2-D representation of vibration signals as input,’’ MATEC Web Conf, vol. 95, Oct. 2017, Art. no. 13001.
[20] Y. Bengio, “Learning deep architectures for AI”, Foundations and trends in Machine Learning, vol.2, no.1, pp.1-127, Jan. 2009, doi: doi.org/10.1561/2200000006