神經網絡模型訓練對於特定領域的人工智慧物聯網 (AIoT) 應用是必不可少的。通常顯示卡在模型訓練期間平均可能消耗數百瓦，而搭載 GPU 的嵌入式設備在出於相同的目的可能僅消耗幾瓦，但需要更長的訓練時間。在本論文中，使用了 NVIDIA RTX 2080 Ti 顯示卡和 NVIDIA Jetson Nano 嵌入式設備進行模型訓練的實證研究。將測量到的能量消耗和訓練時間，用以比較兩個平台之間的差異。結果表明，令人驚訝的是雖然使用 Jetson Nano 的訓練時間 比使用獨立顯示卡的訓練時間慢 30 倍，但 Jetson Nano 的總能耗實際上只有一半。結果表明，當考量能源消耗的重要性大於時間性的時候，可以選擇在搭載 GPU 的嵌入式設備上進行模型 訓練以達到節省能源的效果，反之則使用配有獨立顯示卡的電腦是更佳的選擇。在這些 AI 模型訓練中，像 Nvidia Jetson Nano 這樣的配備 GPU 的嵌入式設備可能在耗能方面具有更好的性能。
此外，此論文也探討了關於 AIoT 用於預測性維護的案例研究，以說明配有 GPU 的嵌入式系統在模型訓練中的優勢。在實作預測性維護的案例研究中，也使用了 NASA 提供的渦輪引擎退化模擬資料集。而案例研究結果指出在時間性上的延遲是可以被接受的情況下，配備 GPU 的嵌入式裝置是可以有效的節省能源。

Neural network model training is indispensable for domain-specific Artificial Intelligent Internet-of-Things (AIoT) applications. Typically, a GPU graphics card may take several hundreds watts during model training, while an embedded GPU device may take only couple watts for the same purpose at the cost of a longer training time. This thesis presents an empirical study on the model training using NVIDIA Geforce RTX 2080 Ti graphics card and NVIDIA Jetson Nano embedded device. The two platforms were compared in terms of energy consumption and training response time. The empirical result shows that, surprisingly, while the training time using the Jetson Nano may be 30 times longer than that using the RTX 2080 Ti graphics card, the total energy consumed by Jetson Nano may be 50% less. This suggests that, in AIoT applications where the response time of model training is less critical, one may instead choose to use GPU embedded devices to save energy. The GPU-equipped embedded systems, like NVIDIA Jetson Nano, may have better performance of energy consumption in these AI model training. Predictive maintenance, which uses the machine learning to predict the maintenance timing, is a good example of domain-specific AIoT applications, because the training response time is less critical. This thesis includes a case study of predictive maintenance by using the turbofan engine degradation simulation data set from NASA. The case study shows that with the use of an embedded GPU device, it can save more energy while meeting the timeliness requirement of the application.

[1] Industrial Internet Consortium, “Industrial Internet Reference Architecture,” Jun 2019. [Online]. Available: https://www.iiconsortium.org/IIRA.htm
[2] “Nvidia jetson series.” [Online]. Available: https://www.nvidia.com/zh-tw/
autonomous-machines/embedded-systems/
[3] N. Otterness, M. Yang, S. Rust, E. Park, J. H. Anderson, F. D. Smith, A. Berg, and
S. Wang, “An evaluation of the nvidia tx1 for supporting real-time computer-vision workloads,” in 2017 IEEE Real-Time and Embedded Technology and Applications Symposium (RTAS), 2017, pp. 353–364.
[4] S. Bianco, R. Cadene, L. Celona, and P. Napoletano, “Benchmark analysis of representative deep neural network architectures,” IEEE Access, vol. 6, pp. 64 270–64 277, 2018.
[5] J. Jo, S. Jeong, and P. Kang, “Benchmarking gpu-accelerated edge devices,” in
2020 IEEE International Conference on Big Data and Smart Computing (BigComp).
IEEE, 2020, pp. 117–120.
[6] A. A. S¨uzen, B. Duman, and B. S¸en, “Benchmark analysis of jetson tx2, jetson
nano and raspberry pi using deep-cnn,” in 2020 International Congress on HumanComputer Interaction, Optimization and Robotic Applications (HORA). IEEE, 2020, pp. 1–5.
[7] T. Amert, N. Otterness, M. Yang, J. H. Anderson, and F. D. Smith, “Gpu scheduling on the nvidia tx2: Hidden details revealed,” in 2017 IEEE Real-Time Systems Symposium (RTSS). IEEE, 2017, pp. 104–115.
[8] S. Mittal, “A survey on optimized implementation of deep learning models on the nvidia jetson platform,” Journal of Systems Architecture, vol. 97, pp. 428–442, 2019.
[9] A. A. S¨uzen, B. Duman, and B. S¸en, “Benchmark analysis of jetson tx2, jetson
nano and raspberry pi using deep-cnn,” in 2020 International Congress on HumanComputer Interaction, Optimization and Robotic Applications (HORA), 2020, pp. 1–5.
[10] A. Basulto-Lantsova, J. A. Padilla-Medina, F. J. Perez-Pinal, and A. I. BarrancoGutierrez, “Performance comparative of opencv template matching method on jetson tx2 and jetson nano developer kits,” in 2020 10th Annual Computing and Communication Workshop and Conference (CCWC), 2020, pp. 0
[11] R. K. Mobley, An introduction to predictive maintenance. Elsevier, 2002.
[12] M. Paolanti, L. Romeo, A. Felicetti, A. Mancini, E. Frontoni, and J. Loncarski,
“Machine learning approach for predictive maintenance in industry 4.0,” in 2018
14th IEEE/ASME International Conference on Mechatronic and Embedded Systems
and Applications (MESA). IEEE, 2018, pp. 1–6.
[13] A. Kanawaday and A. Sane, “Machine learning for predictive maintenance of industrial machines using iot sensor data,” in 2017 8th IEEE International Conference on Software Engineering and Service Science (ICSESS). IEEE, 2017, pp. 87–90.
[14] A. Saxena, K. Goebel, D. Simon, and N. Eklund, “Damage propagation modeling for aircraft engine run-to-failure simulation,” in 2008 international conference on
prognostics and health management. IEEE, 2008, pp. 1–9.
[15] K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in Proceedings of the IEEE conference on computer vision and pattern recognition,
2016, pp. 770–778.
[16] S. Hochreiter and J. Schmidhuber, “Long short-term memory,” Neural computation, vol. 9, no. 8, pp. 1735–1780, 1997.