基于密度聚類和動態時間彎曲的結晶器黏結漏鋼預報方法的開發

段海洋; 王旭東; 姚曼

doi:10.13374/j.issn2095-9389.2019.04.02.004

基于密度聚類和動態時間彎曲的結晶器黏結漏鋼預報方法的開發

doi: 10.13374/j.issn2095-9389.2019.04.02.004

段海洋^{1, 2},
王旭東^{1, 2, ,},
姚曼^{1, 2}

1.
大連理工大學材料科學與工程學院，大連 116024
2.
遼寧省凝固控制與數字化制備技術重點實驗室，大連 116024

基金項目: 國家自然科學基金資助項目(51974056，51474047)；中央高校基本科研業務費資助項目

詳細信息

通訊作者:
E-mail：hler@dlut.edu.cn

中圖分類號: TG249.7
計量
- 文章訪問數: 1684
- HTML全文瀏覽量: 978
- PDF下載量: 48
- 被引次數: 0
出版歷程
- 收稿日期: 2019-04-02
- 刊出日期: 2020-03-01

Development of prediction method for mold sticking breakout based on density-based spatial clustering of applications with noise and dynamic time warping

DUAN Hai-yang^{1, 2},
WANG Xu-dong^{1, 2
, ,},
YAO Man^{1, 2}

1.
School of Materials Science and Engineering, Dalian University of Technology, Dalian 116024, China
2.
Key Laboratory of Solidification Control and Digital Preparation Technology (Liaoning Province), Dalian 116024, China

More Information

Corresponding author: E-mail: hler@dlut.edu.cn

摘要

摘要: 針對漏鋼時結晶器銅板溫度呈現出的“時間滯后”和“空間倒置”等典型特征，本文通過引入動態時間彎曲(DTW)和機器學習中的密度聚類(DBSCAN)方法，提取、匯集并區分結晶器溫度的典型變化模式，在此基礎上開發出一種新型的漏鋼預報方法。借助動態時間彎曲度量不同拉速、鋼種或工藝操作條件下結晶器熱電偶溫度的相似性，并運用密度聚類方法聚集和分離正常工況、黏結漏鋼狀況下的溫度樣本，在此基礎上檢測和預報結晶器漏鋼。結果證實，相較于傳統的邏輯判斷和人工神經元網絡預報結晶器漏鋼的方法，基于聚類的漏鋼預報方法無需人為設置閾值或參數，能夠依據漏鋼歷史樣本中溫度變化的共性規律，提取并融合熱電偶溫度在時間、空間上典型的變化特征，準確區分和預報結晶器漏鋼，具有較好的自適應性和魯棒性。
- 連鑄 /
- 結晶器 /
- 漏鋼預報 /
- 密度聚類 /
- 動態時間彎曲
Abstract: As the core component of continuous casting machines, complex behaviors of fluid flow, heat transfer, mass transfer, and solidification occurring inside the mold are the key factors affecting the slabs quality. Breakout is one of the most catastrophic accidents in continuous casting process, which brings severe impacts on personal security, smooth producing, slab quality, and caster equipment. In particular, with the development of the high-speed casting technology, quality defects and sticking breakouts caused by high-load emerge frequently and missing or false alarms for online prediction of breakout occasionally occur. Thus, accurate identification and prediction for the mold breakout is a top priority for online processing control. Considering the typical temperature characteristics of “time lag” and “space inversion” during a breakout, this paper introduced the concepts of dynamic time warping (DTW) and density-based spatial clustering of applications with noise (DBSCAN) in machine learning. On the basis of collecting and distinguishing the typical change modes of mold temperature, an integrated novel method for predicting breakout was developed. The proposed method applied DTW to measure the similarity of mold thermocouple temperature under different casting speeds, steel grades, and other operating conditions, while DBSCAN was used to cluster and separate the temperature samples between normal casting status and sticking breakout. On the basis of the above mentioned method, the results show that the mold sticking breakout can be effectively detected and predicted. Compared with the traditional method based on logical judgment and artificial neural network, the clustering-based breakout prediction method does not require manual setting of thresholds or parameters. According to the common rule of temperature variation in historical samples of breakout, the typical characteristics of temperature in time and space can be extracted and fused, and the breakout can be accurately distinguished and predicted, which shows good self-adaptability and robustness.
- continuous casting /
- mold /
- breakout prediction /
- density-based spatial clustering of applications with noise /
- dynamic time warping

HTML全文

圖 1 示意圖. （a）結晶器熱電偶分布；（b）黏結漏鋼熱電偶溫度變化

Figure 1. Schematic diagram: (a) thermocouple distribution of mold; (b) thermocouple temperature variation of breakout

下載: 全尺寸圖片幻燈片

圖 2 不同工況下的溫度變化. （a）正常；（b）漏鋼；（c）誤報

Figure 2. Temperature comparison of different situations: (a) normal; (b) breakout; (c) false alarm

下載: 全尺寸圖片幻燈片

圖 3 溫度及其特征提取. （a）正常工況溫度；（b）正常工況溫度預處理結果

Figure 3. Temperature and features extraction: (a) temperature of normal status; (b) processing results of normal status

下載: 全尺寸圖片幻燈片

圖 4 溫度及其特征提取. （a）誤報溫度；（b）誤報溫度預處理結果

Figure 4. Temperature and features extraction: (a) temperature of false alarm; (b) processing results of false alarm

下載: 全尺寸圖片幻燈片

圖 5 溫度及其特征提取. （a）漏鋼溫度；（b）漏鋼溫度預處理結果

Figure 5. Temperature and features extraction: (a) temperature of breakout; (b) processing results of breakout

下載: 全尺寸圖片幻燈片

圖 6 歐氏距離和動態時間彎曲映射對比

Figure 6. Mapping comparison of Euclidean and DTW

下載: 全尺寸圖片幻燈片

圖 7 參數鄰域半徑選擇示意圖

Figure 7. Diagram of parameter Eps selection

下載: 全尺寸圖片幻燈片

圖 8 訓練樣本密度聚類可視化結果

Figure 8. Training samples visual result after DBSCAN clustering

下載: 全尺寸圖片幻燈片

圖 9 漏鋼預報流程

Figure 9. Flowchart of breakout prediction

下載: 全尺寸圖片幻燈片

圖 10 不同工況下測試樣本的熱電偶溫度. （a）正常；（b）誤報

Figure 10. Thermocouple temperature of test samples under different working mode: (a) normal; (b) false alarm mode

下載: 全尺寸圖片幻燈片

圖 11 黏結漏鋼測試樣本熱電偶溫度. （a）漏鋼樣本實例1；（b）漏鋼樣本實例2

Figure 11. Thermocouple temperature of test samples at breakout mode: (a) sample 1; (b) sample 2

下載: 全尺寸圖片幻燈片

圖 12 測試樣本密度聚類可視化結果

Figure 12. Testing samples visual result after DBSCAN clustering

下載: 全尺寸圖片幻燈片

表 1 歐氏距離和動態時間彎曲距離計算結果對比

Table 1. Comparison of calculation results for Euclidean and DTW distance

Item	Sequence	x	y	z
Euclidean	x	0	11.46	15.12
	y	11.46	0	10.53
	z	15.12	10.53	0
Dynamic Time Warping	x	0	2.42	9.39
	y	2.42	0	9.07
	z	9.39	9.07	0

下載: 導出CSV

表 2 訓練樣本密度聚類結果

Table 2. DBSCAN clustering result of training samples

Working condition	DBSCAN clustering result （label）
Normal	?1	?1	?1	?1	?1	?1	?1	?1	?1	?1
Normal	?1	?1	?1	?1	?1	?1	?1	?1	?1	?1
False alarm	?1	?1	?1	?1	?1	?1	?1	?1	?1	?1
	?1	?1	?1	?1	?1	?1	?1	?1	?1	?1
	?1	?1	?1	?1	?1	?1	?1	?1	?1	?1
Breakout	0	0	0	0	0	0	0	0	0	0
	0	0	0	0	0	0	0	0	0	0
	0	0	0	0	0	0	0	0	0	0

下載: 導出CSV

表 3 測試樣本密度聚類結果

Table 3. DBSCAN clustering result of samples testing

Working condition	DBSCAN clustering result （label）
Normal	?1	?1	?1	?1	?1	?1	?1	?1	?1	?1
	?1	?1	?1	?1	?1	?1	?1	?1	?1	?1
	?1	?1	?1	?1	?1	?1	?1	?1	?1	?1
False alarm	?1	?1	?1	?1	?1	?1	?1	?1	?1	?1
False alarm	?1	?1	?1	?1	?1	?1	?1	?1	?1	?1
breakout	0	0	0	0	0	0	0	0	0	0
breakout	0	0	0	0	0	0	0	0	0	0

下載: 導出CSV

259luxu-164

參考文獻(18)

[1]	Zhang Y X, Wang W L, Zhang H H. Development of a mold cracking simulator: The study of breakout and crack formation in continuous casting mold. Metall Mater Trans B, 2016, 47(4): 2244 doi: 10.1007/s11663-016-0705-y
[2]	Qin X, Zhu C F, Zheng L W, et al. Molten steel breakout prediction based on thermal friction measurement. J Iron Steel Res Int, 2011, 18(4): 24 doi: 10.1016/S1006-706X(11)60045-9
[3]	Liu X X, Liu P Z, Zhou J Q. Numerical simulation and prediction for sticking type breakout behavior in slab continuous casting. J Univ Sci Technol Beijing, 1997, 19(2): 143 劉曉霞, 劉佩忠, 周筠清. 板坯連鑄粘結型漏鋼過程模擬及預報. 北京科技大學學報, 1997, 19(2):143
[4]	Hao P F, Xu X H, Pei Y Y, et al. The period on collecting breakout data sampling and the fashion of mold thermal-monitoring in continuous casting breakout. J Northeast Univ Nat Sci Ed, 1997, 18(4): 400 郝培鋒, 徐心和, 裴云毅, 等. 連鑄漏鋼預報系統數據采樣與熱電偶埋設方式. 東北大學學報: 自然科學版, 1997, 18(4):400
[5]	He F, Zhang L Y. Mold breakout prediction in slab continuous casting based on combined method of GA-BP neural network and logic rules. Int J Adv Manuf Technol, 2018, 95(9-12): 4081 doi: 10.1007/s00170-017-1517-1
[6]	Liu Y, Wang X D, Du F M, et al. Computer vision detection of mold breakout in slab continuous casting using an optimized neural network. Int J Adv Manuf Technol, 2017, 88(1-4): 557 doi: 10.1007/s00170-016-8792-0
[7]	Blazek K E, Saucedo I G. Characterization of the formation, propagation, and recovery of sticker/hanger type breakouts. ISIJ Int, 1990, 30(6): 435 doi: 10.2355/isijinternational.30.435
[8]	Lu M J, Lin K J, Kuo C H, et al. Sticker breakout theory and its prediction in slab continuous-casting // Proceedings of 67th Steelmaking Conference. Dallas, 1993: 343
[9]	Liu Y Z, Wang X D, Jia Q Z, et al. Investigation in the behavior of sticking breakout and its propagation in slab continuous casting mould. Steelmaking, 2009, 25(3): 45 劉永貞, 王旭東, 賈啟忠, 等. 結晶器內粘結漏鋼及其傳播行為的研究. 煉鋼, 2009, 25(3):45
[10]	He F, Wu P F, Xu Q Y, et al. Mould temperature change and propagation behavior of sticking-type breakout during continuous casting. J Iron Steel Res, 2016, 28(2): 27 何飛, 吳鵬飛, 徐其言, 等. 連鑄粘結漏鋼的結晶器溫度變化及其傳播行為. 鋼鐵研究學報, 2016, 28(2):27
[11]	Liu Y, Wang X D, Shi G Q, et al. Process factors of sticking breakout for wide and thick slab continuous casting. J Univ Sci Technol Beijing, 2014, 36(6): 757 劉宇, 王旭東, 施桂錢, 等. 寬厚板連鑄黏結漏鋼的工藝因素. 北京科技大學學報, 2014, 36(6):757
[12]	Ester M, Kriegel H P, Sander J, et al. A density-based algorithm for discovering clusters in large spatial databases with noise // KDD-96 Proceedings. Second International Conference on Knowledge Discovery and Data Mining. Portland, 1996: 226
[13]	Salvadora S, Chan P. Toward accurate dynamic time warping in linear time and space. Intell Data Anal, 2007, 11(5): 561 doi: 10.3233/IDA-2007-11508
[14]	Kate R J. Using dynamic time warping distances as features for improved time series classification. Data Min Knowl Discov, 2016, 30(2): 283 doi: 10.1007/s10618-015-0418-x
[15]	Anh D T, Thanh L H. An efficient implementation of k-means clustering for time series data with DTW distance. Int J Bus Intell Data Min, 2015, 10(3): 213 doi: 10.1504/IJBIDM.2015.071311
[16]	Keogh E, Ratanamahatana C A. Exact indexing of dynamic time warping. Knowl Inf Syst, 2005, 7(3): 358 doi: 10.1007/s10115-004-0154-9
[17]	Wan Y, Chen X L, Shi Y. Adaptive cost dynamic time warping distance in time series analysis for classification. J Comput Appl Math, 2017, 319: 514 doi: 10.1016/j.cam.2017.01.004
[18]	Wang W T, Wu Y L, Tang C Y, et al. Adaptive density-based spatial clustering of applications with noise (DBSCAN) according to data // Proceedings of 2015 International Conference on Machine Learning and Cybernetics (ICMLC). Guangzhou, 2015: 445