Above Burden Temperature Data Probes Interpretation to Prevent Malfunction of Blast Furnaces - Part 1: Intelligent Information Preprocessing (original) (raw)
Related papers
DATA MINING APPLIED TO IDENTIFYING FACTORS AFFECTING BLAST FURNACE STAVE HEAT LOADS
A long blast furnace (BF) campaign life is achieved through careful selection and control of various inputs to the process (raw materials physical and chemical quality characteristics) and adequate operator interventions (burden distribution changes and casting practices). One of the key controlling factors for long BF life is to minimise and reduce the variability of stave heat loads (SHL). Since Port Kembla No. 5 Blast Furnace (BF5) started operating in June 1991, there have been several excursions in SHL levels and the underlying drivers behind these variations, as well as the drivers necessary to achieve aim total heat load (1500 GJ/day) or below, are not well understood. In order to examine this problem more comprehensively, the SHL was first segmented into its 16 constituent heat loads (4 regions by 4 drums). An Ishikawa or cause-effect diagram was constructed to identify most of the contributing factors and define a set of day-averaged campaign data for analysis. A machine learning procedure, the so-called data mining technique, was adopted to provide an objective means of identifying valid and useful structure or patterns in this large, complex, multi-factorial and highly coupled data set. The technique involves visualisation, transformation and multi-dimensional scaling of data, as well as decision tree analysis and rule set generation. Through a number of data mining trials on the historical data supplied from the operation of BF5, a number of key variables together with several structural, or contextual relationships between them, were uncovered. These, in turn demonstrated a significantly high accuracy in predicting nine defined bands of SHL ranging from low to very high.
ISIJ International, 2010
The adaptation of blast furnaces to the new technologies has increased the operation information so that the sensor information can be known at every moment. However this often results in the supply of excessive data volume to the plant operators. This paper describes an industrial application for self-organized maps (SOM) in order to help them make decisions regarding blast furnace control by means of pattern recognition and the matching of temperature profiles supplied by the thermocouples placed on the above burden. The classification of patterns via easy color coding indicates to the operator what the blast furnace operational situation is, thus making the necessary corrections easier. KEY WORDS: ironmaking; blast furnace; neural networks; self-organized maps (SOM); forecasting. Fig. 1. Data statistics of the top gas temperature profiles in a blast furnace.
Expert system of diagnostics blast furnace process
IOP Conference Series: Materials Science and Engineering, 2020
The expert system of diagnostics blast furnace process is presented. It is based on a logical-mathematical model for assessing the progress of blast furnace smelting. The model provides an opportunity to evaluate the normal operation mode of blast furnace and further deviations from this mode such as overdeveloped gas flows (peripheral and central), violation of thermal melting conditions (hot and cold course of melt), violation of smooth descent of burned materials in the furnace (tight furnace operation, higher and lower suspension of burden). The functional capabilities of developed software are represented.
Metallurgical Research & Technology
The estimation of thermal level in blast furnace is of utmost importance, because the processes occurring inside the blast furnace are complex in nature and any drift in thermal level could lead to abnormal furnace state. The present review is made to understand the methods for estimating thermal level in blast furnace, and the drift in estimation of the thermal level. The thermal level estimation is divided into 3 categories, viz. mathematical models, statistical models and decision support systems. The mathematical models are based on the first principle of thermodynamics and give an estimate of the thermal level in blast furnace. On the other hand, the statistical models are mainly the data-based approach that uses the historical data to predict the instability in blast furnace. Lastly, the decision support systems are the prescriptive models that give the recommendations for making the necessary corrections in the process parameters to avoid occurrence of abnormality in blast fu...
Fuzzy Classifier Design for Development Tendency of Hot Metal Silicon Content in Blast Furnace
IEEE Transactions on Industrial Informatics, 2018
Since the hot metal silicon content simultaneously reflects the product quality and the thermal state of the blast furnace, accurately predicting the development tendency of hot metal silicon content has the immensely guiding role for blast furnace operators. This paper focuses on fuzzy classifier design for the development tendency of hot metal silicon content based on blast furnace operation data. The cross characteristic of binary classification problem was found via embedding highdimensional blast furnace data into a two-dimensional space. Then, presented a nonparallel hyperplanes based fuzzy classifier, which conquered the cross classification still holding the interpretability advantage as fuzzy classifier. The proposed method was tested on No.2 blast furnace of Liuzhou Steel in China, that demonstrated the excellent performance compared with some other classifier algorithms.
Data Mining Using Unguided Symbolic Regression on a Blast Furnace Dataset
Lecture Notes in Computer Science, 2011
In this paper a data mining approach for variable selection and knowledge extraction from datasets is presented. The approach is based on unguided symbolic regression (every variable present in the dataset is treated as the target variable in multiple regression runs) and a novel variable relevance metric for genetic programming. The relevance of each input variable is calculated and a model approximating the target variable is created. The genetic programming configurations with different target variables are executed multiple times to reduce stochastic effects and the aggregated results are displayed as a variable interaction network. This interaction network highlights important system components and implicit relations between the variables. The whole approach is tested on a blast furnace dataset, because of the complexity of the blast furnace and the many interrelations between the variables. Finally the achieved results are discussed with respect to existing knowledge about the blast furnace process.
Application of symbolic regression on blast furnace and temper mill datasets
2012
In this paper a data mining approach for variable selection and knowledge extraction from datasets is presented. The approach is based on unguided symbolic regression (every variable present in the dataset is treated as the target variable in multiple regression runs) and a novel variable relevance metric for genetic programming. The relevance of each input variable is calculated and a model approximating the target variable is created. The genetic programming configurations with different target variables are executed multiple times to reduce stochastic effects and the aggregated results are displayed as a variable interaction network. This interaction network highlights important system components and implicit relations between the variables. The whole approach is tested on a blast furnace dataset, because of the complexity of the blast furnace and the many interrelations between the variables. Finally the achieved results are discussed with respect to existing knowledge about the blast furnace process.
Ann based prediction of blast furnace parameters
2007
The paper presents a method to predict blast furnace parameters based on ar tificial neural network (ANN). The pr ediction is important as the parameters cause the degradation of the pr oduction process. The productivity as well as quality can be improved by knowing these parameters in advance. In this context, the ir on making process in the modern blast furnace is briefly illustrated. Characterisation of the input and the output parameters as well as the design of a feed for ward neural network (FFNN) is outlined. The implementation issues ar e discussed to predict the parameters like hot metal temperatur e (HMT) and percentage of impurity of silicon content in molten ir on. The simulation and plant trial r esults are compar ed to show the effectiveness of the approach.
Data Mining Applications in Steel Industry
Encyclopedia of Data Warehousing and Mining, Second Edition
The industrial plants, beyond subsisting, pursue to be leaders in increasingly competitive and dynamic markets. In this environment, quality management and technological innovation is less a choice than a must. Quality principles, such as those comprised in ISO 9000 standards, recommend companies to make their decisions with a based on facts approach; a policy much easily followed thanks to the all-pervasive introduction of computers and databases. With a view to improving the quality of their products, factory owners are becoming more and more interested in exploiting the benefits gained from better understanding their productive processes. Modern industries routinely measure the key variables that describe their productive processes while these are in operation, storing this raw information in databases for later analysis. Unfortunately, the distillation of useful information might prove problematic as the amount of stored data increases. Eventually, the use of specific tools capa...