基于改進降噪自編碼器半監督學習模型的熱軋帶鋼水梁印識別算法

陳兆宇; 荊豐偉; 李杰; 郭強

doi:10.13374/j.issn2095-9389.2021.01.04.004

基于改進降噪自編碼器半監督學習模型的熱軋帶鋼水梁印識別算法

doi: 10.13374/j.issn2095-9389.2021.01.04.004

北京科技大學高效軋制與智能制造國家工程研究中心，北京 100083

基金項目: 國家自然科學基金資助項目（51674028）

詳細信息

通訊作者:
E-mail: guoqiang@nercar.ustb.edu.cn

中圖分類號: TP273.5
計量
- 文章訪問數: 508
- HTML全文瀏覽量: 199
- PDF下載量: 44
- 被引次數: 0
出版歷程
- 收稿日期: 2021-08-04
- 網絡出版日期: 2021-06-18
- 刊出日期: 2022-07-06

Recognition algorithm of hot-rolled strip steel water beam mark based on a semi-supervised learning model of an improved denoising autoencoder

National Engineering Research Center for Advanced Rolling and Intelligent Manufacturing, University of Science and Technology Beijing, Beijing 100083, China

More Information

Corresponding author: E-mail: guoqiang@nercar.ustb.edu.cn

摘要

摘要: 針對水梁印識別困難且工作量大問題，提出一種基于改進降噪自編碼器半監督學習模型的熱軋帶鋼水梁印識別算法。該算法在降噪自編碼器（Denoising auto-encoder, DAE）的基礎上對編碼層的每一層添加隨機噪聲，在隱藏層后添加分類層，并對數據添加偽標簽，在解碼的同時進行分類訓練，使得DAE具有半監督學習能力。通過提取熱軋帶鋼粗軋出口溫度數據中的溫差特征，用相應特征對模型進行訓練。實驗結果表明，算法能夠準確識別出帶鋼的水梁印，在模型精確度上，與主流分類識別模型對比，提出的模型在帶標簽樣本數量較小時，分類精度相比其他模型高5.0%～10.0%；在帶標簽樣本數量較大時，提出的模型分類精度達到93.8%，現場能夠根據模型的識別結果提高生產效率。
- 熱軋帶鋼 /
- 降噪自編碼器 /
- 半監督學習 /
- 水梁印 /
- 加熱爐
Abstract: The water beam mark is a common problem in slab heating, which causes quality defects on strip steel. In hot strip rolling, the heating quality of the slab considerably influences the rolling stability and quality of the finished strip. The water beam mark caused by the heating process and equipment is a common defect in the slab heating. A slab water beam imprint has a great influence on the control precision of the rolling force and thickness of the finished strip. Presently, recognizing the water beam mark is difficult and the workload in the industry is heavy. To solve these problems, this study proposed a recognition algorithm of a hot-rolled strip steel water beam mark based on a semisupervised learning model of an improved denoising autoencoder (DAE). Based on the DAE, random noise was added to each layer of the coding layer, a classification layer was added after a hidden layer, and fake labels were added to the training data. Decoding and classification training are conducted simultaneously. These methods result in the model becoming semisupervised. In this study, we extract the temperature difference of the strip temperature data at the outlet of the roughing mill and use it to train the model. Experimental results showed that the algorithm can accurately recognize the water beam mark of strip steel. The classification accuracy of the proposed model is 5.0%–10.0% higher than other mainstream models when the number of tag proportions is small. When the number of tag proportions is large, the accuracy of the proposed model reaches up to 93.8%. According to the result, the production efficiency can be improved using this model.
- hot-rolled strip /
- denoising auto-encoder /
- semi-supervised /
- water beam mark /
- furnace

HTML全文

圖 1 粗軋出口溫度實測數據

Figure 1. Measured data of the rough rolling delivery temperature

下載: 全尺寸圖片幻燈片

圖 2 水梁印識別算法流程圖

Figure 2. Flow chart of the water beam seal recognition algorithm

下載: 全尺寸圖片幻燈片

圖 3 自編碼器模型

Figure 3. AE model

下載: 全尺寸圖片幻燈片

圖 4 降噪自編碼器模型

Figure 4. DAE model

下載: 全尺寸圖片幻燈片

圖 5 改進的DAE結構示意圖

Figure 5. Improved DAE structure diagram

下載: 全尺寸圖片幻燈片

圖 6 噪聲對模型準確率影響. (a)輸入層只有均值變化；(b)輸入層只有方差變化；(c)編碼層只有均值變化；(d)編碼層只有方差變化

Figure 6. Influence of noise on model accuracy: (a) only the mean value of the input layer changes; (b) only the variance of the input layer changes; (c) only the mean value of the encode layer changes; (d) only the variance of the encode layer changes

下載: 全尺寸圖片幻燈片

圖 7 粗軋出口溫度數據. (a)帶鋼頭部溫度；(b)帶鋼全長溫度

Figure 7. Measured data of rough rolling delivery temperature: (a) temperature of the strip head; (b) total strip temperature

下載: 全尺寸圖片幻燈片

圖 8 精軋出口厚度. (a)帶鋼頭部厚度；(b)帶鋼全長厚度

Figure 8. Thickness of the strip delivery of the finishing mill: (a) thickness of the strip head; (b) total strip thickness

下載: 全尺寸圖片幻燈片

圖 9 步進式加熱爐內部結構示意圖. (a)加熱爐左視圖；(b)加熱爐主視圖

Figure 9. Internal structure diagram of the walking beam furnace: (a) left view of the heating furnace; (b) main view of the heating furnace

下載: 全尺寸圖片幻燈片

圖 10 水梁印位置分布識別結果

Figure 10. Recognition results of the location distribution of water beam marks

下載: 全尺寸圖片幻燈片

表 1 不同位置添加噪音后在不同標簽占比下的模型精度結果

Table 1. Model accuracy results after adding noise at different layers under different label proportions %

Experiment condition	Accuracy when the label proportion is 1%	Accuracy when the label proportion is 2%	Accuracy when the label proportion is 5%	Accuracy when the label proportion is 6%	Accuracy when the label proportion is 10%	Accuracy when the label proportion is 33%	Accuracy when the label proportion is 50%
Input layer: no noise Encode layer: no noise	73.25	78.80	79.20	80.40	86.00	89.20	91.20
Input layer: add noise Encode layer: no noise	73.50	80.00	82.00	82.20	88.60	90.20	92.00
Input layer: no noise Encode layer: add noise	67.50	79.80	80.40	81.20	87.20	90.60	91.60
Input layer: add noise Encode layer: add noise	75.75	81.20	83.80	83.20	89.40	91.60	93.80

下載: 導出CSV

表 2 不同噪聲的對比模型

Table 2. Compare models with different noise

Experiment number	The mean of the input layer noise	The variance of input layer noise	The mean of the encode layer noise	The variance of encode layer noise
a	Variable	1	0	1
b	0	Variable	0	1
c	0	1	Variable	1
d	0	1	0	Variable

下載: 導出CSV

表 3 不同標簽占比下不同神經網絡分類精度

Table 3. Classification accuracy of different neural networks under different label proportions %

Model	Accuracy when the label proportion is 1%	Accuracy when the label proportion is 2%	Accuracy when the label proportion is 5%	Accuracy when the label proportion is 6%	Accuracy when the label proportion is 10%	Accuracy when the label proportion is 33%	Accuracy when the label proportion is 50%
BP	64.20	71.60	77.80	80.80	89.40	91.40	92.00
DBP	69.80	69.00	75.60	81.20	84.20	91.20	92.60
DBN	65.40	77.60	77.20	81.20	89.40	91.40	91.60
LSTM			73.20	74.60	78.00	84.80	93.80
CNN-LSTM			78.80	80.20	87.40	89.80	94.40
Improved-DAE	75.75	81.20	83.60	83.20	89.70	91.60	93.80

下載: 導出CSV

表 4 改進降噪自編碼器部分識別結果

Table 4. Partial recognition results of improved DAE ℃

Coil number	1	2	3	4	5	6	7	8
1 #	18.55	12.24	14.13	11.01	12.41	13.39	11.19	16.12
2 #	29.16	12.76	19.78	13.70	15.56	20.98	11.76	45.16
3 #	17.09	7.17	16.24	12.55	10.61	21.48	11.78	25.73
4 #	27.37	9.55	18.01	19.16	13.82	22.96	13.28	34.90
5 #	23.72	10.40	18.38	13.08	13.81	22.76	14.91	35.66

下載: 導出CSV

www.77susu.com

參考文獻(27)

[1]	Lu Z Q, Hong Z, Ji L. Research on simulator of walking beam in billet heating furnace. Manuf Autom, 2017, 39(4): 135 doi: 10.3969/j.issn.1009-0134.2017.04.034 魯照權, 洪志, 季亮. 鋼坯加熱爐步進梁模擬器研究. 制造業自動化, 2017, 39(4):135 doi: 10.3969/j.issn.1009-0134.2017.04.034
[2]	Qi F S, Wang Z S, Li B K. Slab heating uniformity based on multi-field coupling heat transfer in reheating furnace. J Northeast Univ Nat Sci, 2019, 40(10): 1413 doi: 10.12068/j.issn.1005-3026.2019.10.009 齊鳳升, 王子松, 李寶寬. 基于加熱爐多場耦合傳熱的板坯加熱均勻性. 東北大學學報(自然科學版), 2019, 40(10):1413 doi: 10.12068/j.issn.1005-3026.2019.10.009
[3]	Xiao N, Hu W C, Jiang H, et al. Numerical simulation of water beam black imprint in walking beam reheating furnace. Ind Furn, 2013, 35(5): 40 doi: 10.3969/j.issn.1001-6988.2013.05.012 肖楠, 胡文超, 江華, 等. 步進梁式加熱爐水梁黑印數值模擬. 工業爐, 2013, 35(5):40 doi: 10.3969/j.issn.1001-6988.2013.05.012
[4]	Chen G F. Analysis and improvement for water beam mark during heavy plate rolling // Proceedings of China Iron & Steel Annual Meeting. Beijing, 2019: 8 陳國鋒. 厚板軋機關于水梁印問題的分析及優化 //第十二屆中國鋼鐵年會論文集. 北京, 2019: 8
[5]	Sun Z B. Simulation research on heating furnace water beam's black mark. Ind Heat, 2016, 45(5): 30 doi: 10.3969/j.issn.1002-1639.2016.05.009 孫志斌. 加熱爐水梁黑印模擬研究. 工業加熱, 2016, 45(5):30 doi: 10.3969/j.issn.1002-1639.2016.05.009
[6]	Wang Q. Quality analysis system of 1780 mm hot rolling mill. Metall Ind Autom, 2006, 30(3): 17 doi: 10.3969/j.issn.1000-7059.2006.03.004 王強. 1780熱軋質量分析系統. 冶金自動化, 2006, 30(3):17 doi: 10.3969/j.issn.1000-7059.2006.03.004
[7]	Zhu Y B, Zhang F, Zhang Y J, et al. Particle swarm optimized neural network for strip crown prediction model research. Metall Ind Autom, 2019, 43(2): 11 朱永波, 張飛, 張勇軍, 等. 基于粒子群優化的帶鋼凸度神經網絡預測模型研究. 冶金自動化, 2019, 43(2):11
[8]	Li F J, Liu H F, Zhang X. On-line identification of roll eccentricity based on PSO-RBF neural network. J Iron Steel Res, 2017, 29(11): 906 李灃驥, 劉鴻飛, 張興. 基于PSO-RBF神經網絡的軋輥偏心在線辨識. 鋼鐵研究學報, 2017, 29(11):906
[9]	Cai Z X. Application of Artificial Intelligence in metallurgy automation. Metall Ind Autom, 2015, 39(1): 1 doi: 10.3969/j.issn.1000-7059.2015.01.001 蔡自興. 人工智能在冶金自動化中的應用. 冶金自動化, 2015, 39(1):1 doi: 10.3969/j.issn.1000-7059.2015.01.001
[10]	Wei L X, Wei X Y, Sun H, et al. Prediction of aluminum hot rolling force based on deep network. Chin J Nonferrous Met, 2018, 28(10): 2070 魏立新, 魏新宇, 孫浩, 等. 基于深度網絡訓練的鋁熱軋軋制力預報. 中國有色金屬學報, 2018, 28(10):2070
[11]	Peng L G, Wang D G, Li J, et al. Data-driven adaptive setting algorithm for coiling temperature model parameter. Chin J Eng, 2020, 42(6): 778 彭良貴, 王登剛, 李杰, 等. 數據驅動的卷取溫度模型參數即時自適應設定算法. 工程科學學報, 2020, 42(6):778
[12]	Ma W, Li W G, Zhao Y T, et al. Prediction of hot-rolled roll force based on deep learning. J Iron Steel Res, 2019, 31(9): 805 馬威, 李維剛, 趙云濤, 等. 基于深度學習的熱連軋軋制力預測. 鋼鐵研究學報, 2019, 31(9):805
[13]	Lin W L, Ding X F, Shuang Y H. Prediction on rolling force of oblique rolling piercing based on BP neural network. Forg Stamp Technol, 2018, 43(10): 175 林偉路, 丁小鳳, 雙遠華. BP神經網絡對斜軋穿孔軋制力的預測. 鍛壓技術, 2018, 43(10):175
[14]	Liu J H, Wang G X, Liu Y K. Prediction of rolling force based on grey theory and neural network. Forg Stamp Technol, 2015, 40(10): 126 劉杰輝, 王桂霞, 劉永康. 基于灰色理論和神經網絡的軋制力預測. 鍛壓技術, 2015, 40(10):126
[15]	Zhang Y F, Lu Z Q. Remaining useful life prediction based on an integrated neural network. Chin J Eng, 2020, 42(10): 1372 張永峰, 陸志強. 基于集成神經網絡的剩余壽命預測. 工程科學學報, 2020, 42(10):1372
[16]	Wu K, Liu X Z, Zhang X X, et al. Feature extraction of hot strip rolling data based on PCA-DBN. Metall Ind Autom, 2020, 44(3): 21 武凱, 劉新忠, 張笑雄, 等. 基于PCA-DBN的熱連軋數據特征提取. 冶金自動化, 2020, 44(3):21
[17]	Rasmus A, Berglund M, Honkala M, et al. Semi-supervised learning with ladder networks. Adv Neural Info Process Syst, 2015, 28: 3546
[18]	Springenberg J T. Unsupervised and semi-supervised learning with categorical generative adversarial networks // International Conference on Learning Representations. San Juan, 2016
[19]	Hinton G E, Zemel R S. Autoencoders, minimum description length, and Helmholtz free energy. Adv Neural Inf Process Syst, 1994, 6: 3
[20]	Vincent P, Larochelle H, Bengio Y, et al. Extracting and composing robust features with denoising autoencoders // Proceedings of the 25th International Conference on Machine Learning. Helsinki, 2008: 1096
[21]	Pezeshki M, Fan L, Brakel P, et al. Deconstructing the ladder network architecture // International Conference on Machine Learning. New York, 2016: 2368
[22]	Hinton G E, Salakhutdinov R R. Reducing the dimensionality of data with neural networks. Science, 2006, 313(5786): 504 doi: 10.1126/science.1127647
[23]	Hochreiter S, Schmidhuber J. Long short-term memory. Neural Comput, 1997, 9(8): 1735 doi: 10.1162/neco.1997.9.8.1735
[24]	Liu Y, Gao Z Y, Zhou X M, et al. Industrial data-driven intelligent forecast for chatter of cold rolling of thin strip with LSTM recurrent neural network. J Mech Eng, 2020, 56(11): 121 doi: 10.3901/JME.2020.11.121 劉陽, 郜志英, 周曉敏, 等. 工業數據驅動下薄板冷軋顫振的LSTM智能預報. 機械工程學報, 2020, 56(11):121 doi: 10.3901/JME.2020.11.121
[25]	Zhang P, Yang T, Liu Y N, et al. Feature extraction and prediction of QAR data based on CNN-LSTM. Appl Res Comput, 2019, 36(10): 2958 張鵬, 楊濤, 劉亞楠, 等. 基于CNN-LSTM的QAR數據特征提取與預測. 計算機應用研究, 2019, 36(10):2958
[26]	Lu J X, Zhang Q P, Yang Z H, et al. Short-term load forecasting method based on CNN-LSTM hybrid neural network model. Autom Electr Power Syst, 2019, 43(8): 131 doi: 10.7500/AEPS20181012004 陸繼翔, 張琪培, 楊志宏, 等. 基于CNN-LSTM混合神經網絡模型的短期負荷預測方法. 電力系統自動化, 2019, 43(8):131 doi: 10.7500/AEPS20181012004
[27]	Jia C Y, Kang K X, Gao W, et al. Fault prediction of electro-hydraulic servo valve based on CNN+LSTM neural network. Chin Hydraul Pneum, 2020(12): 173 doi: 10.11832/j.issn.1000-4858.2020.12.027 賈春玉, 康凱旋, 高偉, 等. 基于CNN+LSTM神經網絡的電液伺服閥故障預測. 液壓與氣動, 2020(12):173 doi: 10.11832/j.issn.1000-4858.2020.12.027