臺灣博碩士論文加值系統

根據台灣電力公司2022年的資料統計，燃氣及燃煤發電約占總發電量的50% ~ 60%，這些發電的能量來自於鍋爐設備燃燒燃料及氣體，理想燃燒狀況的空氣量應當是計量空氣，在實際的應用中，為了避免不完全燃燒的情形發生，通常會加入適當的過量空氣，加入的空氣量由燃燒設備、燃料而定，過量空氣不足會發生燃燒不完全，且會造成空氣污染，空氣過量過多則會造成熱損失，因此需要配合鍋爐和燃料，通入適量的過量空氣，為了達到這一需求，本論文嘗試發展一套智慧影像辨識設備，以化學螢光不同波段的強度比值獲得燃燒訊息的原理，訓練不同的神經網絡(包含VGG-16、VGG-19、ResNet-50、LSTM以及一個由本文搭建的卷積神經網絡)，以提供一個建議的神經網絡用於控制過量空氣的輸入量，控制的方式是使用一種名為Modbus的國際標準的通訊協議與AI程式交換數據，使用RS485作為硬體接口，因為此種接口具有傳輸距離較長的優勢且受到業界廣泛使用，為了模擬控制鍋爐中的火焰，本論文搭建一個直徑40毫米的蜂窩陶瓷燃燒器，並附有混合艙與逆止閥，使燃料氣體與過量空氣能夠充分混合並確保實驗安全，為了降低成本使本設備更容易拓展並成為現有電廠的外掛設備，在考量影像品質與成本之後，使用市面上常見的商業相機與擷取卡作為電腦從相機端取得影像的工具，在論文的最後，使用G/R的訊號強度，比較有無使用LSTM控制的結果。

According to the statistics of Taiwan Electric Power Corporation in 2022, gas and coal-fired power generation accounts for about 50% to 60% of the total power generation. The energy for these power generation comes from the combustion of fuel and gas by boiler equipment. The air volume for ideal combustion conditions should be measured. Air, in practical applications, in order to avoid incomplete combustion, appropriate excess air is usually added. The amount of air added depends on the combustion equipment and fuel. Insufficient excess air will cause incomplete combustion and cause air. Pollution, too much air will cause heat loss. Therefore, it is necessary to cooperate with boilers and fuels to pass in an appropriate amount of excess air. In order to meet this demand, this paper attempts to develop a set of intelligent image recognition equipment. The intensity of chemical fluorescence in different bands The principle of ratio to obtain burning information, training different neural networks (including VGG-16, VGG-19, ResNet-50, LSTM and a convolutional neural network constructed by this paper) to provide a proposed neural network for controlling excess The air input is controlled by using an international standard communication protocol called Modbus to exchange data with the AI program, using RS485 as the hardware interface, because this interface has the advantage of long transmission distance and is widely used in the industry. In order to simulate the control of the flame in the boiler, this paper builds a honeycomb ceramic burner with a diameter of 40 mm, with a mixing chamber and a check valve, so that the fuel gas and excess air can be fully mixed and ensure the safety of the experiment. The equipment is easier to expand and become the external equipment of the existing power plant. After considering the image quality and cost, the common commercial cameras and capture cards on the market are used as the tools for the computer to obtain images from the camera. At the end of the paper, G/R is used. The signal strength of, compares the results with and without LSTM control.

摘要 ............................................................................................................ i
Abstract...................................................................................................... ii
誌謝 .......................................................................................................... iii
目錄 .......................................................................................................... iv
表目錄 ...................................................................................................... vi
圖目錄 ..................................................................................................... vii
符號說明.................................................................................................... x
一、緒論.................................................................................................... 1
1.1 燃燒穩定與燃燒不穩定..................................................................... 1
1.1.1回火(Flashback)........................................................................ 2
1.1.2燃燒流體動力學不穩定性 (Hydrodynamic Instability)............ 5
1.2 檢測燃燒不穩定的工具..................................................................... 6
1.3 數位影像前處理............................................................................... 12
1.4 多孔燃燒器的結構與功能............................................................... 13
1.5 預測燃燒不穩定的工具－人工神經網絡........................................ 15
1.5.1感知器 ................................................................................... 16
1.5.2梯度下降法(gradient descent)............................................... 18
1.5.3啟動函數(Activation Functions) ............................................. 20
1.5.4MLP 架構................................................................................ 23
1.5.5反向傳播法............................................................................ 24
1.5.6過度擬合(Overfitting) ............................................................ 26
1.5.7優化網絡架構........................................................................ 27
1.5.8長短期記憶模型(LSTM)......................................................... 27
1.6 研究重點 .......................................................................................... 30
v
二、實驗設備.......................................................................................... 31
2.1 燃燒實驗台 ...................................................................................... 31
2.2 燃燒控制設備 .................................................................................. 32
2.3 圖像擷取設備 .................................................................................. 33
2.4 冷卻設備與保護裝置....................................................................... 33
三、實驗方法.......................................................................................... 35
3.1 回火測溫實驗 .................................................................................. 35
3.2 影像處理以及把燃燒不穩定的過程數值化(檢測回火).................. 37
3.3 ROI 強度波動診斷(檢測燃燒不穩定) .............................................. 38
3.4 火焰形心偏移診斷(檢測火焰流體動力學不穩定(hdi)................... 40
3.5 深度學習 .......................................................................................... 41
3.6 燃燒器控制實驗............................................................................... 43
四、結果與討論...................................................................................... 44
4.1 燃燒不穩定的觀察與蒐集訓練資料 ............................................... 44
4.1.1回火測溫實驗與影像處理(蒐集回火訓練資料)................... 44
4.1.2ROI 強度波動診斷(蒐集回火訓練資料)................................ 51
4.1.3焰形心偏移診斷(蒐集火焰流體動力學不穩定訓練資料) ... 52
4.2 深度學習(DL) .................................................................................. 53
4.3 燃燒器控制實驗............................................................................... 61
4.4 結論 .............................................................................................. 61
五、 未來工作......................................................................................... 62
參考文獻.................................................................................................. 64
Extended Abstract ......................................................................................... 67

[1]A. Kalantari and V. McDonell, "Boundary layer flashback of non-swirling premixed flames: Mechanisms, fundamental research, and recent advances," Progress in Energy and Combustion Science, vol. 61, pp. 249-292, 2017/07/01/ 2017.
[2]A. Lipatnikov, J. J. P. i. e. Chomiak, and c. science, "Molecular transport effects on turbulent flame propagation and structure," vol. 31, no. 1, pp. 1-73, 2005.
[3]R. Borghi, "On the structure and morphology of turbulent premixed flames," in Recent advances in the Aerospace Sciences: Springer, 1985, pp. 117-138.
[4]O. Kwon, G. J. C. Faeth, and Flame, "Flame/stretch interactions of premixed hydrogen-fueled flames: measurements and predictions," vol. 124, no. 4, pp. 590-610, 2001.
[5]G. Koroll, R. Kumar, E. J. C. Bowles, and Flame, "Burning velocities of hydrogen-air mixtures," vol. 94, no. 3, pp. 330-340, 1993.
[6]D. Korschelt et al., "Mechanism of instabilities in turbulent combustion leading to flashback," 1981.
[7]M. Konle, F. Kiesewetter, and T. J. E. i. F. Sattelmayer, "Simultaneous high repetition rate PIV–LIF-measurements of CIVB driven flashback," vol. 44, no. 4, pp. 529-538, 2008.
[8]B. Lewis and G. J. T. J. o. C. P. von Elbe, "Stability and structure of burner flames," vol. 11, no. 2, pp. 75-97, 1943.
[9]R. K. Cheng, "Controlling emissions, intensity, and stability of turbulent premixed combustion," in Proc. of Symposium on Smart Control of Turbulence, The University of Tokyo, 1999, pp. 46-57.
[10]K. R. Anderson, J. Hertzberg, S. J. C. S. Mahalingam, and Technology, "Classification of absolute and convective instabilities in premixed bluff body stabilized flames," vol. 112, no. 1, pp. 257-269, 1996.
[11]S. Sivasegaram, J. J. C. Whitelaw, and flame, "Oscillations in confined disk-stabilized flames," vol. 68, no. 2, pp. 121-129, 1987.
[12]M. Heitor, A. Taylor, J. J. C. Whitelaw, and Flame, "Influence of confinement on combustion instabilities of premixed flames stabilized on axisymmetric baffles," vol. 57, no. 1, pp. 109-121, 1984.
[13]H. Büchner, C. Hirsch, W. J. C. s. Leuckel, and technology, "Experimental investigations on the dynamics of pulsated premixed axial jet flames," vol. 94, no. 1-6, pp. 219-228, 1993.
[14]Y.-C. Chao, Y.-L. Chang, C.-Y. Wu, and T.-S. J. P. o. t. C. I. Cheng, "An experimental investigation of the blowout process of a jet flame," vol. 28, no. 1, pp. 335-342, 2000.
[15]A. G. Gaydon, H. G. Wolfhard, and S. Penner, "Flames, their structure, radiation and temperature," 1960.
[16]Y.-c. Chien, "Development of a Chemiluminescence Diagnostic System for Combustion Measurements," Development of a Chemiluminescence Diagnostic System for Combustion Measurements, 2007.
[17]Y. Liu, J. Tan, H. Wang, and L. Lv, "Characterization of heat release rate by OH* and CH* chemiluminescence," Acta Astronautica, vol. 154, pp. 44-51, 2019/01/01/ 2019.
[18]刘瑶, 谭建国, 王浩, and 高政旺, "甲烷-空气低旋流火焰中自由基的辐射特性," 推進技術, vol. 40, no. 9, pp. 2022-2029, 2019.
[19]Y.-c. Chien, "火焰自然螢光量測系統之研發," 成功大學航空太空工程學系學位論文, pp. 1-50, 2007.
[20]J. Kojima, Y. Ikeda, and T. J. P. o. t. C. i. Nakajima, "Spatially resolved measurement of OH*, CH*, and C2* chemiluminescence in the reaction zone of laminar methane/air premixed flames," vol. 28, no. 2, pp. 1757-1764, 2000.
[21]H. Gao, Z. Qu, X. Feng, and W. Tao, "Combustion of methane/air mixtures in a two-layer porous burner: A comparison of alumina foams, beads, and honeycombs," Experimental Thermal and Fluid Science, vol. 52, pp. 215-220, 2014/01/01/ 2014.
[22]P.-F. Hsu, W. D. EVANS, J. R. J. C. S. HOWELL, and Technology, "Experimental and numerical study of premixed combustion within nonhomogeneous porous ceramics," vol. 90, no. 1-4, pp. 149-172, 1993.
[23]V. Khanna, R. Goel, J. J. C. s. Ellzey, and technology, "Measurements of emissions and radiation for methane combustion within a porous medium burner," vol. 99, no. 1-3, pp. 133-142, 1994.
[24]A. J. Barra, J. L. J. C. Ellzey, and flame, "Heat recirculation and heat transfer in porous burners," vol. 137, no. 1-2, pp. 230-241, 2004.
[25]A. J. Barra, G. Diepvens, J. L. Ellzey, M. R. J. C. Henneke, and Flame, "Numerical study of the effects of material properties on flame stabilization in a porous burner," vol. 134, no. 4, pp. 369-379, 2003.
[26]W. M. Mathis, J. L. J. C. S. Ellzey, and Technology, "Flame stabilization, operating range, and emissions for a methane/air porous burner," vol. 175, no. 5, pp. 825-839, 2003.
[27]M. T. Smucker, J. L. J. C. s. ELLZEY, and Technology, "Computational and experimental study of a two-section porous burner," vol. 176, no. 8, pp. 1171-1189, 2004.
[28]F. Rosenblatt, "The perceptron: a probabilistic model for information storage and organization in the brain," Psychological review, vol. 65, no. 6, p. 386, 1958.
[29]D. E. Rumelhart, G. E. Hinton, and R. J. J. n. Williams, "Learning representations by back-propagating errors," vol. 323, no. 6088, pp. 533-536, 1986.
[30]Y. LeCun, Y. Bengio, and G. J. n. Hinton, "Deep learning," vol. 521, no. 7553, pp. 436-444, 2015.
[31]Y. LeCun et al., "Backpropagation applied to handwritten zip code recognition," vol. 1, no. 4, pp. 541-551, 1989.
[32]B. A. Richards et al., "A deep learning framework for neuroscience," vol. 22, no. 11, pp. 1761-1770, 2019.
[33]T. P. Lillicrap, A. Santoro, L. Marris, C. J. Akerman, and G. J. N. R. N. Hinton, "Backpropagation and the brain," vol. 21, no. 6, pp. 335-346, 2020.
[34]G. Cybenko, "Continuous valued networks with two hidden layers are sufficient: Technical Report," ed: Department of Computer Science, Tufts University, 1988.
[35]E. R. Davies, "CHAPTER 25 - Biologically Inspired Recognition Schemes," in Machine Vision (Third Edition), E. R. Davies, Ed. Burlington: Morgan Kaufmann, 2005, pp. 725-755.
[36]D. E. Rumelhart, G. E. Hinton, and R. J. Williams, "Learning representations by back-propagating errors," nature, vol. 323, no. 6088, pp. 533-536, 1986.
[37]S. Sra, S. Nowozin, and S. J. Wright, Optimization for machine learning. Mit Press, 2012.
[38]J. L. Elman, "Finding structure in time," Cognitive science, vol. 14, no. 2, pp. 179-211, 1990.
[39]Y. Bengio, P. Frasconi, and P. Simard, "The problem of learning long-term dependencies in recurrent networks," in IEEE international conference on neural networks, 1993, pp. 1183-1188: IEEE.
[40]S. Hochreiter and J. Schmidhuber, "Long short-term memory," Neural computation, vol. 9, no. 8, pp. 1735-1780, 1997.
[41]J. Chung, C. Gulcehre, K. Cho, and Y. Bengio, "Empirical evaluation of gated recurrent neural networks on sequence modeling," preprint, 2014.
[42]D. Bahdanau, K. Cho, and Y. Bengio, "Neural machine translation by jointly learning to align and translate," arXiv preprint arXiv:. 2014.
[43]A. Graves, A.-r. Mohamed, and G. Hinton, "Speech recognition with deep recurrent neural networks," in 2013 IEEE international conference on acoustics, speech and signal processing, 2013, pp. 6645-6649: Ieee.
[44]L. Baraldi, C. Grana, and R. Cucchiara, "Hierarchical boundary-aware neural encoder for video captioning," in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017, pp. 1657-1666.
[45]J. Li, S. Chou, G. Huang, W. Yang, Z. J. E. T. Li, and F. Science, "Study on premixed combustion in cylindrical micro combustors: transient flame behavior and wall heat flux," vol. 33, no. 4, pp. 764-773, 2009.
[46]J. Pan, J. A. Libera, N. H. Paulson, and M. Stan, "Flame stability analysis of flame spray pyrolysis by artificial intelligence," The International Journal of Advanced Manufacturing Technology, vol. 114, no. 7, pp. 2215-2228, 2021/06/01 2021.
[47]G. E. Hinton, N. Srivastava, A. Krizhevsky, I. Sutskever, and R. R. J. a. p. a. Salakhutdinov, "Improving neural networks by preventing co-adaptation of feature detectors," 2012.