# Fault Detection and Identification of Blast Furnace Ironmaking Process Using the Gated Recurrent Unit Network

^{1}

^{2}

^{*}

## Abstract

**:**

## 1. Introduction

## 2. Methodologies

#### 2.1. GRU Neural Network

#### 2.1.1. The Structure of LSTM Cell

#### 2.1.2. The Structure of GRU Cell

#### 2.2. Support Vector Data Description

## 3. Fault Detection and Identification Strategy

#### 3.1. Fault Detection

#### 3.2. Fault Identification

- Obtain historical NOC data;
- Remove extreme values and normalize the training data to have a zero mean and unit variance.
- Set initial parameters of GRU model and train the model;
- If the GRU model is valid, the GRU residuals will be fed into the SVDD model, and the threshold ${R}^{2}$ of ${D}^{2}$ statistic is obtained.

- Collect online samples;
- Normalize the online samples;
- Use the GRU model trained in the offline process to make prediction and get the residuals;
- Calculate the ${D}^{2}$ statistic using SVDD;
- Determine whether to alarm by comparing the ${D}^{2}$ statistic and the threshold ${R}^{2}$. If ${D}^{2}$ is greater than ${R}^{2}$, the process is faulty, otherwise it is normal.
- If the process is faulty, isolate and identify which variables are most severely affected.

## 4. Application Studies

#### 4.1. Case 1: Hanging Fault

#### 4.1.1. Residual Generation Using the GRU Network

#### 4.1.2. Fault Detection and Identification

#### 4.2. Case 2: Abnormal Molten Iron Temperature

## 5. Conclusions

## Author Contributions

## Funding

## Conflicts of Interest

## References

- Amano, S.; Takarabe, T.; Nakamori, T. Expert system for blast furnace operation at Kimitsu works. ISIJ Int.
**1990**, 30, 105–110. [Google Scholar] [CrossRef] [Green Version] - Liao, S. Expert system methodologies and applications–a decade review from 1995 to 2004. Exp. Syst. Appl.
**2005**, 28, 93–103. [Google Scholar] [CrossRef] - Tian, H.; Wang, A. A Novel Fault Diagnosis System for Blast Furnace Based on Support Vector Machine Ensemble. ISIJ Int.
**2010**, 50, 738–742. [Google Scholar] [CrossRef] [Green Version] - Liu, L.; Wang, A.; Sha, M. Multi-class classification methods of cost-conscious LS-SVM for fault diagnosis of blast furnace. Ind. J. Iron Steel Res. Int.
**2011**, 18, 17–23. [Google Scholar] [CrossRef] - An, R.; Yang, C.; Zhou, Z.; Wang, L. Comparison of Different Optimization methods with support vecto machine for blast furnace multi-fault classification. IFAC-PapersOnLine
**2015**, 48, 1204–1209. [Google Scholar] - Vanhatalo, E. Multivariate process monitoring of an experimental blast furnace. Qual. Reliab. Eng. Int.
**2010**, 26, 495–508. [Google Scholar] [CrossRef] - Zhang, T.; Ye, H.; Wang, W. Fault diagnosis for blast furnace ironmaking process based on two-stage principal component analysis. ISIJ Int.
**2014**, 54, 2334–2341. [Google Scholar] [CrossRef] [Green Version] - Shang, J.; Chen, M.; Zhang, H. Increment-based recursive transformed component statistical analysis for monitoring blast furnace iron-making processes: An index-switching scheme. Control Eng. Pract.
**2018**, 77, 190–200. [Google Scholar] [CrossRef] - Pan, Y.; Yang, C.; An, R. Robust principal component pursuit for fault detection in a blast furnace process. Ind. Eng. Chem. Res.
**2017**, 57, 283–291. [Google Scholar] [CrossRef] - Zhou, B.; Ye, H.; Zhang, H. Process monitoring of iron-making process in a blast furnace with PCA-based methods. Control Eng. Pract.
**2016**, 47, 1–14. [Google Scholar] [CrossRef] - Cai, J.; Zeng, J.; Luo, S. A state space model for monitoring of the dynamic blast furnace system. ISIJ Int.
**2012**, 52, 2194–2199. [Google Scholar] [CrossRef] [Green Version] - Vanhatalo, E.; Kulahci, M. Impact of autocorrelation on principal components and their use in statistical process control. Qual. Reliab. Eng. Int.
**2015**, 32, 1483–1500. [Google Scholar] [CrossRef] [Green Version] - Dong, Y.; Qin, S.J. A novel dynamic pca algorithm for dynamic data modeling and process monitoring. J. Process Control.
**2018**, 67, 1–11. [Google Scholar] [CrossRef] - Chiang, L.; Braatz, R.; Russell, E.L. Fault Detection and Diagnosis in Industrial Systems; Springer Science & Business Media: Berlin, Germany, 2002; Volume 44, pp. 197–198. [Google Scholar]
- Qin, S. Survey on data-driven industrial process monitoring and diagnosis. Annu. Rev. Control
**2012**, 36, 220–234. [Google Scholar] [CrossRef] - Wang, J.; Yan, J.; Li, C. Deep heterogeneous GRU model for predictive analytics in smart manufacturing: Application to tool wear prediction. Comput. Ind.
**2019**, 111, 1–14. [Google Scholar] [CrossRef] - Hochreiter, S.; Schmidhuber, J. Long Short-Term Memory. Neural Comput.
**1997**, 9, 1735–1780. [Google Scholar] [CrossRef] - Funahashi, K.; Nakamura, Y. Approximation of dynamical systems by continuous time recurrent neural networks. Neural Netw.
**1993**, 6, 801–806. [Google Scholar] [CrossRef] - Kim, P.; Lee, D.; Lee, S. Discriminative context learning with gated recurrent unit for group activity recognition. Pattern Recognit.
**2018**, 76, 149–161. [Google Scholar] [CrossRef] - Li, G.; Hu, Y.; Chen, H. An improved fault detection method for incipient centrifugal chiller faults using the PCA-R-SVDD algorithm. Comput. Sci.
**2014**, 116, 104–113. [Google Scholar] [CrossRef] - Pearson, R.; Neuvo, Y.; Astola, J.; Gabbouj, M. Generalized Hampel filters. EURASIP J. Adv. Signal Process.
**2016**, 87. [Google Scholar] [CrossRef] - Chang, Z.; Zhang, Y.; Chen, W. Electricity price prediction based on hybrid model of adam optimized LSTM neural network and wavelet transform. Energy
**2019**, 187. [Google Scholar] [CrossRef]

Sample Availability: Samples of the compounds are available from the authors. |

**Figure 4.**Flowchart of the GRU-support vector data description (SVDD)-based fault detection and identification methodology.

**Figure 5.**Prediction results for ${u}_{1}$ using models with different parameter settings (the red line corresponding to predictions, the blue line corresponding to true values). (

**a**) Prediction results for ${u}_{1}$ with ${d}_{h}=64$ and ${d}_{h}^{{}^{\prime}}=100$; (

**b**) Prediction results for ${u}_{1}$ with ${d}_{h}=64$ and ${d}_{h}^{{}^{\prime}}=200$; (

**c**) Prediction results for ${u}_{1}$ with ${d}_{h}=32$ and ${d}_{h}^{{}^{\prime}}=100$; (

**d**) Prediction results for ${u}_{1}$ with ${d}_{h}=32$ and ${d}_{h}^{{}^{\prime}}=100$.

**Figure 6.**The prediction results of the GRU model in Case 1 (the red lines represent prediction and the blue lines represent actual values).

**Figure 7.**Monitoring results for the hanging fault. (

**a**) Monitoring results of GRU-SVDD, (

**b**) Monitoring results of LSTM-SVDD, (

**c**) Monitoring results of principal component analysis (PCA)-SVDD.

**Figure 10.**Prediction results using GRU and monitoring results using SVDD for Case 2. (

**a**) Prediction results using GRU, (

**b**) Monitoring results using SVDD.

**Figure 11.**The fault contribution rate for each variable in Case 2. (

**a**) The Contribution of different variables in observations 2250 to 2450, (

**b**) The Contribution of different variables in observations 2450 to 3000.

No. | Variable | |
---|---|---|

${u}_{1}$ | quantity of blast | |

${u}_{2}$ | temperature of blast | |

${u}_{3}$ | pressure of blast | |

${u}_{4}$ | the quantity of oxygen blasted | |

${u}_{5}$ | CO concentration in top gas | |

${u}_{6}$ | ${\mathrm{CO}}_{2}$ concentration in top gas | |

${u}_{7}$ | ${\mathrm{H}}_{2}$ concentration in top gas |

Methods | Detection Rate | |
---|---|---|

${D}_{GRU\phantom{\rule{3.33333pt}{0ex}}-\phantom{\rule{3.33333pt}{0ex}}SVDD}^{2}$ | $93.52\%$ | |

${D}_{LSTM-SVDD}^{2}$ | $92.27\%$ | |

${D}_{PCA\phantom{\rule{3.33333pt}{0ex}}-\phantom{\rule{3.33333pt}{0ex}}SVDD}^{2}$ | $72.82\%$ |

No. | Variable | |
---|---|---|

${u}_{1}$ | quantity of blast | |

${u}_{2}$ | temperature of blast | |

${u}_{3}$ | pressure of blast | |

${u}_{4}$ | quantity of oxygen blasted | |

${u}_{5}$ | temperature of cold blast | |

${u}_{6}$ | top pressure | |

${u}_{7}$ | CO concentration in top gas | |

${u}_{8}$ | ${\mathrm{CO}}_{2}$ concentration in top gas | |

${u}_{9}$ | ${\mathrm{H}}_{2}$ concentration in top gas | |

${u}_{10}$ | pressure of cold blast |

© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## Share and Cite

**MDPI and ACS Style**

Ouyang, H.; Zeng, J.; Li, Y.; Luo, S.
Fault Detection and Identification of Blast Furnace Ironmaking Process Using the Gated Recurrent Unit Network. *Processes* **2020**, *8*, 391.
https://doi.org/10.3390/pr8040391

**AMA Style**

Ouyang H, Zeng J, Li Y, Luo S.
Fault Detection and Identification of Blast Furnace Ironmaking Process Using the Gated Recurrent Unit Network. *Processes*. 2020; 8(4):391.
https://doi.org/10.3390/pr8040391

**Chicago/Turabian Style**

Ouyang, Hang, Jiusun Zeng, Yifan Li, and Shihua Luo.
2020. "Fault Detection and Identification of Blast Furnace Ironmaking Process Using the Gated Recurrent Unit Network" *Processes* 8, no. 4: 391.
https://doi.org/10.3390/pr8040391