隨著近年數位轉型，智慧物聯網、大數據等技術蓬勃發展，因此 伺服器使用量快速增加，帶動全世界資料中心大規模的快速增長。而 當伺服器機櫃負載配置不當、風扇故障或是其他問題，導致積熱以及 因壓力差產生熱回流，皆會影響伺服器機櫃散熱效率，使得資料中心 損耗電能之營運成本。因此，資料中心伺服器機櫃熱異常之監控受到 極大的重視。紅外線熱相機廣泛地應用於工業上非侵入性檢測以及熱 異常監控，目前已有應用的範例，例如: 電力系統之熱點偵測或是太陽能電力板上熱區域分割。基於此，本研究旨在建立熱點偵測追蹤系 統。首先，蒐集伺服器不同負載狀態下產生熱回流之熱影像，透過影 像辨識技術 YOLOv4 及 CLAHE 影像預處理，以所訓練神經網絡模型 對伺服器機櫃狀態進行二元分類定位，偵測熱回流之產生。並且透過 調整 SORT 演算法，將歐式距離閥值與方向變化結合卡爾曼濾波器， 進行即時多物件追蹤，在無距離感測器之情況下，仍可對伺服器精準定位。再者，以視覺迴授控制步進馬達與線性滑台，實現熱相機滑塊 定位控制。本研究為首篇結合影像辨識與物件追蹤技術，以純視覺偵 測機櫃伺服器發生熱迴流之研究。

Due to digital transformation in recent years, the rapid development of technologies such as the Artificial Intelligence of Things (AIoT), Big Data, and others, has led to significant growth in the scale of data centers worldwide, along with an increase in server usage. Improper server load configuration or fan failures can lead to heat accumulation and pressure differentials, affecting the cooling efficiency and operational costs of enterprises. As a result, the monitoring of thermal anomalies in data centers has become of the utmost importance. Infrared thermal cameras play a crucial role in industrial non-intrusive detection and thermal anomaly monitoring. They have been applied in detecting hotspots in power systems and heat zones segmentation on solar power panels. Based on this, the research aims to establish a hot spot detection and tracking system. Firstly, thermal images of servers under different load conditions and those with hot air recirculation will be collected. These images will be processed using state-of-the-art image recognition technique, YOLOv4, and CLAHE image preprocessing. A neural network model will be trained to perform binary classification and localization of server status in the rack servers to detect hot air recirculation. Addition ally, the SORT algorithm will be modified by combining Euclidean distance threshold with direction changes using the Kalman filter for real-time multi object tracking. This approach ensures accurate server positioning without the needs of distance sensors. Moreover, a linear slider system will be implemented using stepper motor and ball screw driver which enables visual feedback signal controls the positioning of the thermal camera slider. This research introduces a novel approach to integrate thermal image recognition and object tracking techniques for visual detection of heat recirculation in rack servers.

第一章 緒論 1
1.1 前言 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 研究動機及目標 . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 文獻回顧 . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 本文架構 . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
第二章 伺服器系統與軟硬體介紹 11
2.1 機櫃伺服器系統 . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 硬體設備 . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2.1 紅外線熱相機 . . . . . . . . . . . . . . . . . . . . 14
2.2.2 K 型熱電偶 . . . . . . . . . . . . . . . . . . . . . . 14
2.2.3 數據擷取器 . . . . . . . . . . . . . . . . . . . . . 15
2.2.4 Arduino UNO 板 . . . . . . . . . . . . . . . . . . . 15
2.2.5 線性滑台 . . . . . . . . . . . . . . . . . . . . . . . 16
2.2.6 微動開關 . . . . . . . . . . . . . . . . . . . . . . . 17
2.2.7 步進馬達與驅動板 . . . . . . . . . . . . . . . . . 17
2.3 軟體及套件 . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3.1 pySerial . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3.2 PyVISA . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3.3 OpenCV . . . . . . . . . . . . . . . . . . . . . . . 20
2.3.4 IRImagerDirect SDK . . . . . . . . . . . . . . . . . 20
2.3.5 Tkinter . . . . . . . . . . . . . . . . . . . . . . . . 20
第三章 系統設計與實踐 21
3.1 自動化資料擷取系統 . . . . . . . . . . . . . . . . . . . . 21
3.1.1 資料蒐集系統架構 . . . . . . . . . . . . . . . . . 21
3.1.2 實時圖形化介面 . . . . . . . . . . . . . . . . . . . 23
3.2 熱影像資料蒐集 . . . . . . . . . . . . . . . . . . . . . . . 25
3.2.1 熱回流資料集 . . . . . . . . . . . . . . . . . . . . 25
3.2.2 ROI 取得 . . . . . . . . . . . . . . . . . . . . . . . 27
3.2.3 影像資料增強 . . . . . . . . . . . . . . . . . . . . 29
3.3 伺服器狀態辨識與定位 . . . . . . . . . . . . . . . . . . . 30
3.3.1 CLAHE 影像預處理 . . . . . . . . . . . . . . . . . 31
3.3.2 YOLOv4-tiny . . . . . . . . . . . . . . . . . . . . . 34
3.4 機櫃實時多物件追蹤 . . . . . . . . . . . . . . . . . . . . 38
3.4.1 多物件追蹤演算法 . . . . . . . . . . . . . . . . . 39
3.4.2 追蹤問題描述 . . . . . . . . . . . . . . . . . . . . 44
3.4.3 Modified-SORT 演算法架構調整 . . . . . . . . . . 45
3.5 伺服器熱回流偵測追蹤系統實踐 . . . . . . . . . . . . . . 50
3.5.1 機構硬體配置 . . . . . . . . . . . . . . . . . . . . 50
3.5.2 系統運作流程 . . . . . . . . . . . . . . . . . . . . 52
第四章 實驗結果 53
4.1 紅外影像偵測熱回流響應之時延程度 . . . . . . . . . . . 53
4.1.1 熱相機與熱電偶溫度量測結果 . . . . . . . . . . . 53
4.1.2 熱相機辨識暫態熱回流時延性量測結果 . . . . . 56
4.2 影像辨識模型效能 . . . . . . . . . . . . . . . . . . . . . 59
4.2.1 影像預處理與訓練架構 . . . . . . . . . . . . . . . 59
4.2.2 效能評估指標 . . . . . . . . . . . . . . . . . . . . 63
4.2.3 YOLOv4-tiny 效能分析 . . . . . . . . . . . . . . . 66
4.3 多物件追蹤演算法效能 . . . . . . . . . . . . . . . . . . . 75
4.3.1 效能評估指標 . . . . . . . . . . . . . . . . . . . . 76
4.3.2 多物件追蹤演算法效能分析 . . . . . . . . . . . . 77
第五章 結論與未來工作 83
5.1 結論 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
5.2 未來工作 . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
參考文獻 85

[1] M. Law. (2022) Energy efficiency predictions for data centres in
2023. [Online]. Available: https://datacentremagazine.com/articles/
efficiency-to-loom-large-for-data-centre-industry-in-2023, accessed:
2022-12-30.

[2] T. Rooks. (2022) Data center energy consumption [Online].
Available: https://www.dw.com/en/data-centers-energy-consumption steady-despite-big-growth-because-of-increasing-efficiency/a 60444548,accessed: 2022-01-24.

[3] E. Masanet, A. Shehabi, N. Lei, S. Smith, and J. Koomey,
“Recalibrating global data center energy-use estimates,” Science, vol.
367, no. 6481, pp. 984–986, 2020.

[4] eASPNet. (2022) 2023 面對全球需求爆炸，【未來資料中心】發展
趨勢是什麼？. [Online]. Available: https://www.easpnet.com/blog/
green-data-center/, accessed: 2022-08-03.

[5] ASHRAE, ”Data Center Networking Equipment–Issues and Best
Practices”, ASHRAE Technical Committee (TC), 2016.
[6] F. B. Insights. (2020) Thermal imaging market. [Online]. Available:
https://www.fortunebusinessinsights.com/ thermal-imaging-market 102588/, accessed: 2020-09-01.

[7] S. Nada and K. Elfeky, “Experimental investigations of thermal
managements solutions in data centers buildings for different
arrangements of cold aisles containments,” Journal of building
engineering, vol. 5, pp. 41–49, 2016.

[8] J. Wan, X. Gui, S. Kasahara, Y. Zhang, and R. Zhang, “Air flow
measurement and management for improving cooling and energy
efficiency in raised-floor data centers: A survey,” IEEE Access, vol. 6,
pp. 48 867–48 901, 2018.

[9] E. K. Lee, H. Viswanathan, and D. Pompili, “Model-based thermal
anomaly detection in cloud datacenters using thermal imaging,” IEEE
Transactions on Cloud Computing, vol. 6, no. 2, pp. 330–343, 2018.

[10] A. N. N. Afifah, Indrabayu, A. Suyuti, and Syafaruddin, “Hotspot
detection in photovoltaic module using otsu thresholding method,”
2020 IEEE International Conference on Communication, Networks
and Satellite (Comnetsat), pp. 408–412, Batam, Indonesia, December
17-18, 2020.

[11] B. N. Wiysobunri, H. S. Erden, and B. U. Toreyin, “A deep
learning approach to fault detection and classification in data centers,”
Basarim2020 High Performance Computing Conference,” Ankara ,
Turkey, October 8 - 9, 2020.

[12] Y.-H. Peng, C.-M. Lee, K.-Y. Tung, and R. Chen, “Rack inlet
temperature prediction based on deep learning,” 2022 25th
International Conference on Mechatronics Technology (ICMT),
pp. 1–4, Kaohsiung, Taiwan, November 18 - 21, 2022.

[13] Ö. Özer, D. Kumlutaş, and U. A. Yücekaya, “A method for volumetric
visualization of temperature distribution: three-dimensional meshed
infrared thermography,” Experiments in Fluids, vol. 60, pp. 1–17,
2019.

[14] P. R. Muniz, S. P. Cani, and R. d. S. Magalhaes, “Influence of
field of view of thermal imagers and angle of view on temperature
measurements by infrared thermovision,” IEEE Sensors Journal,
vol. 14, no. 3, pp. 729–733, 2013.

[15] S. Muniyappan, A. Allirani, and S. Saraswathi, “A novel approach
for image enhancement by using contrast limited adaptive histogram
equalization method,” 2013 Fourth International Conference
on Computing, Communications and Networking Technologies
(ICCCNT), pp. 1–6, Tiruchengode, India, July, 4 - 6, 2013.

[16] A. Bochkovskiy, C.-Y. Wang, and H.-Y. M. Liao, “Yolov4:
Optimal speed and accuracy of object detection,” arXiv preprint
arXiv:2004.10934, 2020.

[17] W. Liu, D. Anguelov, D. Erhan, C. Szegedy, S. Reed, C.-Y. Fu,
and A. C. Berg, “Ssd: Single shot multibox detector,” Computer
Vision–ECCV 2016: 14th European Conference, Part I 14, pp. 21–37,
Amsterdam, The Netherlands, October, 11 - 14, 2016.

[18] P. Jiang, D. Ergu, F. Liu, Y. Cai, and B. Ma, “A review of yolo algorithm
developments,” Procedia Computer Science, vol. 199, pp. 1066–1073, 2022.

[19] J. Redmon, S. Divvala, R. Girshick, and A. Farhadi, “You only look
once: Unified, real-time object detection,” Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 779–788, Las Vegas, USA, June 27 - 30, 2016.

[20] A. Neubeck and L. Van Gool, “Efficient non-maximum suppression,”
18th international conference on pattern recognition (ICPR’06), vol. 3,
pp. 850–855, Hong Kong, China, August, 20 - 24, 2006.

[21] P. Bharati and A. Pramanik, “Deep learning techniques—r-cnn to mask
r-cnn: a survey,” Computational Intelligence in Pattern Recognition:
Proceedings of CIPR 2019, pp. 657–668, 2020.

[22] J. Redmon and A. Farhadi, “Yolov3: An incremental improvement,”
arXiv preprint arXiv:1804.02767, 2018.

[23] A. Bochkovskiy. (2020) Yolov4-tiny released: 40.21770 fps
tkdnn/tensorrt. [Online]. Available: https://github.com/AlexeyAB/
darknet/issues/6067, accessed: 2021-10-22.

[24] A. Bewley, Z. Ge, L. Ott, F. Ramos, and B. Upcroft, “Simple online
and realtime tracking,” 2016 IEEE international conference on image
processing (ICIP), pp. 3464–3468, Phoenix, Arizona, September 25 -
28, 2016.

[25] L. Leal-Taixé, A. Milan, I. Reid, S. Roth, and K. Schindler,
“Motchallenge 2015: Towards a benchmark for multi-target tracking,”
arXiv preprint arXiv:1504.01942, 2015.

[26] R. E. Kalman, “A new approach to linear filtering and prediction
problems,” 1960.

[27] H. W. Kuhn, “The hungarian method for the assignment problem,”
Naval research logistics quarterly, vol. 2, no. 1-2, pp. 83–97, 1955.

[28] N. Wojke, A. Bewley, and D. Paulus, “Simple online and realtime
tracking with a deep association metric,” 2017 IEEE international
conference on image processing (ICIP), pp. 3645–3649, Beijing,
China, September 17-20, 2017.

[29] W. Luo, J. Xing, A. Milan, X. Zhang, W. Liu, and T.-K. Kim, “Multiple
object tracking: A literature review,” Artificial intelligence, vol. 293,
p. 103448, 2021.

[30] C. Mansley, J. Connell, C. Isci, J. Lenchner, J. O. Kephart, S. McIntosh,
and M. Schappert, “Robotic mapping and monitoring of data centers,”2011 IEEE International Conference on Robotics and Automation, pp.5905–5910, Shanghai, China, May, 9 - 13, 2011.

[31] M. Everingham, L. Van Gool, C. K. Williams, J. Winn, and
A. Zisserman, “The pascal visual object classes (voc) challenge,”
International journal of computer vision, vol. 88, pp. 303–338, 2010.

[32] K. Bernardin and R. Stiefelhagen, “Evaluating multiple object tracking
performance: the clear mot metrics,” EURASIP Journal on Image and
Video Processing, vol. 2008, pp. 1–10, 2008.