No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Heat-loss cycle prediction in steelmaking plants through artificial neural network
Abstract
A critical factor in steelworks concerns setting the steel release temperature from the ladle furnace. The challenge resides in estimating in advance the reduction the steel temperature will undergo during its non-processing time until the subsequent casting process. A poor estimation results in productivity and yield losses in casting and unnecessary energy consumption in the ladle. Given process complexity, a pure mathematical description is not available. This work develops a predictive neural model for the reduction in steel temperature between the ladle and the caster considering the main sources of heat losses. The case study refers to a steelmaking plant in Brazil. After model identification and validation, and a sensitivity analysis study, thirty troublesome steel runs that resulted in unplanned shutdowns during casting were investigated. The neural approach provided a correlation between factory-collected values and model estimates of 0.895, with a satisfactory Mean Absolute Error (MAE) of 3.03 °C , against 0.308 and 4.97 °C, respectively, given by the experimental plant model used by the process team, and ‒0.087 and 8.53 °C, respectively, obtained with a linear regression analysis used for comparison purposes. More reliable estimation of the reduction in steel temperature leads to more efficient and economic operations.
... Assembled learning is used (the use of regression and classification mixed models) as predictor of temperature of the molten steel, the classification is made by k-nearest neighbors in [35]. The case of [36] presents a neural network forecaster of temperature for energy saving based in the monitoring of temperature (heat-loss) in the process to avoid the use of extra energy in the cycles of heating and cooling to reduce the energy waste in the process particularly in period between noncasting and casting times. Their proposal presents a mean absolute error in the order of 8.53 °C produced by a linear regression and 4.97 °C produced by the ANN forecaster. ...
... The author declares no competing interests. < 0.55% [44] 0.4% [33] ± 5 °C [34] ± 5% [36] 8.53 °C with ANN [36] 4.97 °C with linear regression [37] 6 °C with ANN [37] 14 °C with physical model [38] 10 °C [41] Tested Springer Nature journal content, brought to you courtesy of Springer Nature Customer Service Center GmbH ("Springer Nature"). Springer Nature supports a reasonable amount of sharing of research papers by authors, subscribers and authorised users ("Users"), for smallscale personal, non-commercial use provided that all copyright, trade and service marks and other proprietary notices are maintained. ...
... The author declares no competing interests. < 0.55% [44] 0.4% [33] ± 5 °C [34] ± 5% [36] 8.53 °C with ANN [36] 4.97 °C with linear regression [37] 6 °C with ANN [37] 14 °C with physical model [38] 10 °C [41] Tested Springer Nature journal content, brought to you courtesy of Springer Nature Customer Service Center GmbH ("Springer Nature"). Springer Nature supports a reasonable amount of sharing of research papers by authors, subscribers and authorised users ("Users"), for smallscale personal, non-commercial use provided that all copyright, trade and service marks and other proprietary notices are maintained. ...
In this paper, an interval type-2 radial basis function neural network system is presented that produces an accurate forecast of the temperature of liquid steel in a secondary metallurgical monitoring and control process. The main goal of this proposal is to get the precise temperatures and times to aggregate the additives to the liquid bath to produce high-strength, low-alloy steel and not waste materials due to the oxidation caused by high temperatures. Also, the proposal reduces the replacement times of measurement instruments that are damaged by the high temperatures. The novelty of this proposal is the lack of this class of neural networks in steelmaking processes and the fact that this is an emergent technology that has only been around for less than 10 years. The produced results show an error in order of 0.12% that is a figure well below 0.30% which is the typo error produced by the measurement devices such as thermocouples of the k-type that are damaged at 1600° and produce erroneous measurements in temperatures superior to 1200 °C and less than the classic interval type-2 fuzzy logic systems.
... The definition of abnormal values can be decided by the operators based on experience or metallurgical principles. Carlsson et al. [40] provided the following example: if the furnace has a maximum capacity of 100 t, then the measured values above 100 t are considered to be abnormal. The study of Duarte et al. [41] regarding the prediction of steel temper- ...
... Several methods are used to calculate the variable importance. Manojlovic et al. [25] and Carlsson et al. [40] used Shapley additive explanations (SHAP). Fig. 6 illustrates the impact of various variables on the electricity consumption in electric arc furnace (EAF) steelmaking by SHAP [25]. ...
... The variable with the highest SHAP value is the quantity of produced steel (billets, t) due to its inclusion in the equation of electricity production. However, the variable of tap-to-tap time ranks first in the variable importance in Carlsson's research [40]. Considering the endpoint P content prediction in BOF steelmaking, the mean impact value is used to detect the variable importance among the 13 input variables in a previous work [43]. ...
With the development of automation and informatization in the steelmaking industry, the human brain gradually fails to cope with an increasing amount of data generated during the steelmaking process. Machine learning technology provides a new method other than production experience and metallurgical principles in dealing with large amounts of data. The application of machine learning in the steelmaking process has become a research hotspot in recent years. This paper provides an overview of the applications of machine learning in the steelmaking process modeling involving hot metal pretreatment, primary steelmaking, secondary refining, and some other aspects. The three most frequently used machine learning algorithms in steelmaking process modeling are the artificial neural network, support vector machine, and case-based reasoning, demonstrating proportions of 56%, 14%, and 10%, respectively. Collected data in the steelmaking plants are frequently faulty. Thus, data processing, especially data cleaning, is crucially important to the performance of machine learning models. The detection of variable importance can be used to optimize the process parameters and guide production. Machine learning is used in hot metal pretreatment modeling mainly for endpoint S content prediction. The predictions of the endpoints of element compositions and the process parameters are widely investigated in primary steelmaking. Machine learning is used in secondary refining modeling mainly for ladle furnaces, Ruhrstahl–Heraeus, vacuum degassing, argon oxygen decarburization, and vacuum oxygen decarburization processes. Further development of machine learning in the steelmaking process modeling can be realized through additional efforts in the construction of the data platform, the industrial transformation of the research achievements to the practical steelmaking process, and the improvement of the universality of the machine learning models.
... The use of empirical models and data-driven approaches is widely employed in industrial applications [9][10][11], and this scenario is no different in the steel industry [12][13][14][15]. This is primarily due to two factors. ...
... A value close to 1 indicates a strong positive correlation, and a value close to −1 indicates also a strong, negative correlation, though. A value close to zero indicates a weak or non-existent correlation represented by cases in which the calculated p-values are greater than 5% (test significance) [12,24]. ...
Dephosphorization is a reaction of important role in steelmaking process, and the correct adequacy of endpoint phosphorus content would improve the quality and productivity of steel in basic oxygen furnace (BOF) processing. Aiming to meet the required steel specifications and reduce process time, two different empirical strategies were established for predicting the endpoint phosphorus content in BOF steelmaking process: linear regression and neural network. Eight variables that affect the endpoint phosphorus content (selected as output) were determined as the input variables of the models. The performances of predictions were evaluated simultaneously with the sensitivity analysis of the model to variations in the values of its input variables. Sensitivity analysis is essential as it reveals the impact of input variables on results, although it is often neglected due to its complexity and the need for multiple simulations. Integrating sensitivity analysis with prediction techniques allows for identifying key variables and making decisions. Both empirical models are suitable and reliable for decision making in the process and can be used as tools for predicting the endpoint phosphorus content, where the neural network has higher accuracy. The sensitivity analysis showed that the two variables that most affect the response of the empirical models were the percentage of oxygen volume of oxygen blown until the sub-lance in relation to the estimated total volume, and the phosphorus concentration in the sub-lance. Received: 3 May 2024 | Revised: 15 July 2024 | Accepted: 20 August 2024 Conflicts of Interest The authors declare that they have no conflicts of interest to this work. Data Availability Statement Data available on request from the corresponding author upon reasonable request. Author Contribution Statement Diego Henrique de Souza Chaves: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Data curation, Writing - original draft, Writing - review & editing, Visualization; Iara Campolina Dias Duarte: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Data curation, Writing - original draft, Writing - review & editing, Visualization; Esly Ferreira da Costa Junior: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Resources, Data curation, Writing - review & editing, Visualization, Supervision, Project administration, Funding acquisition; Andréa Oliveira Souza da Costa: Conceptualization, Investigation, Resources, Data curation, Writing - review & editing, Visualization, Supervision, Project administration, Funding acquisition.
... Although the standard backpropagation (BP) algorithm has some problems, such as a low training rate, it can be optimized to solve these problems with good prediction accuracy and generalizability. Therefore, the use of machine learning models, such as artificial neural networks, has also attracted the attention of scholars in various fields [11,12] and these are widely used in the processes of predicting and controlling various nonlinear problems [13][14][15]. M.H. Sabzalian et al. proposed a lung cancer diagnosis system based on an improved bidirectional recurrent neural network and metaheuristic technique. ...
With the continuous optimization of the steel production process and the increasing emergence of smelting methods, it has become difficult to monitor and control the production process using the traditional steel management model. The regulation of steel smelting processes by means of machine learning has become a hot research topic in recent years. In this study, through the data mining and correlation analysis of the main equipment and processes involved in steel transfer, a network algorithm was optimized to solve the problems of standard back propagation (BP) networks, and a steel temperature forecasting model based on improved back propagation (BP) neural networks was established for basic oxygen furnace (BOF) steelmaking, ladle furnace (LF) refining, and Ruhrstahl–Heraeus (RH) refining. The main factors influencing steel temperature were selected through theoretical analysis and heat balance principles; the production data were analyzed; and the neural network was trained and tested using large amounts of field data to predict the end-point steel temperature of basic oxygen furnace (BOF) steelmaking, ladle furnace (LF) refining, and Ruhrstahl–Heraeus (RH) refining. The prediction model was applied to predict the degree of influence of different operating parameters on steel temperature. A comparison of the prediction results with the production data shows that the prediction system has good prediction accuracy, with a hit rate of over 90% for steel temperature deviations within 20 °C. Compared with the traditional steel temperature management model, the prediction system in this paper has higher management efficiency and a faster response time and is more practical and generalizable in the thermal management of steel.
Ensuring the production of superior castings devoid of foundry flaws while minimizing costs is a perpetual concern for foundries. Enhancing and refining manufacturing procedures are essential to attaining this objective. Computer simulation of foundry operations presents a contemporary substitute for costly and labour-intensive investigations in actual foundries, providing a reliable representation of casting. An in-depth investigation of the casting simulation outcomes enables the anticipation of several hazards that may lead to faults in castings, thereby diminishing their quality and, importantly, increasing production costs. This research analyses a computer simulation of ductile iron valve body casting at a Fine Cast Foundry. The valve body exhibited significant shrinkage defects, leading to a casting rejection rate of approximately 26%. A simulation of the casting process was conducted in accordance with actual shop-floor conditions. The initially designed gating and feeder system was found inadequate in mitigating the shrinkage issues. A revised configuration of the feeder system was proposed based on simulation insights, which, when implemented, led to a reduction in shrinkage-related defects from 26% to 1%. To ensure dimensional accuracy and realistic representation, 3D models were developed using NX 10.0 parametric software, and ProCAST was employed for casting simulation. Additionally, the Taguchi Design of Experiments (DOE) method was applied to optimize key green sand mould parameters. This statistical approach led to a significant reduction in sand mould-related defects, bringing the rejection rate down from 9% to 1.25%.
To address the national dual carbon strategy and to decrease the consumption of iron and steel materials by enterprises, accurate prediction of the converter tapping weight can help to control the amount of ferroalloy addition in tapping. In the actual production process, this relies on manual experience to estimate the converter tapping weight, but the accuracy of the estimated results is not very good. At present, there are many applications of neural networks, but few people make network predictions for the converter tapping weight. In this study, the key factors influencing the converter tapping weight are analyzed by the Pearson correlation method. The kernel principal component analysis–genetic algorithm–backpropagation (KPCA–GA–BP) neural network model has been established to forecast the converter tapping weight. The hidden layer nodes, learning rate, and training times of the model were optimized through a trial-and-error process. The training results show that the hit ratios for the converter tapping weight prediction within error ranges of ± 1 t, ± 2 t, and ± 3 t are 74.0%, 91.0%, and 97.5%, respectively, which significantly surpass the conventional manual predictions. Compared with the traditional artificial prediction, the accuracy of ± 2 t and ± 3 t has been increased by 56.0% and 37.5%, respectively. At the same time, the established KPCA–GA–BP model has been added to the intelligent system of steelmaking alloy addition, and effectively applied to the production, making the steelmaking process more intelligent.
To achieve the goal of reducing energy consumption in the steel industry, accurately predicting the temperature of converter steelmaking is critical for controlling the steelmaking process. Due to the complexity of the converter steelmaking process, precisely predicting the end-point temperature remains a significant challenge. To achieve precise control of the end-point temperature of converter steelmaking, a method combining key feature amplification (KFA) with grey wolf optimizer (GWO) improved affinity propagation (AP) clustering algorithm and gradient-boosting decision tree (GBDT) was proposed to establish a steel temperature prediction model. Firstly, the primary factors influencing the temperature of converter steelmaking were identified based on metallurgical mechanisms. Secondly, the maximal information coefficient (MIC) was utilized to eliminate factors with minimal impact on converter steelmaking temperature, thus determining the input variables for the model. Then, the metallurgical mechanisms and MIC calculation results were integrated with AP clustering to amplify the key feature weights of the model. For the parameter optimization problem in AP clustering, the GWO was employed to find the optimal operating condition classification. Finally, GBDT models were constructed for each operating condition dataset, culminating in the establishment of the KFA-GWO-AP-GBDT end-point converter steelmaking temperature prediction model. The results demonstrate that, compared to K-means, AP, and KFA-AP clustering, KFA-GWO-AP clustering exhibits the best performance and can effectively classify operating conditions. Compared to unclustered and other clustered models, the KFA-GWO-AP-GBDT model achieved prediction accuracy rates of 52.18, 83.36, and 96.58 pct within ± 5 °C, ± 10 °C, and ± 15 °C, respectively, showcasing the best overall performance. This model holds significant practical importance for achieving precise control of converter steelmaking temperature, reducing production costs, and enhancing converter efficiency.
To meet the goals of the national "Dual Carbon" strategy and reduce energy consumption in the steel industry, accurate prediction of steel composition is crucial for precise control over alloy addition in steelmaking. Several models have been created to predict the composition of the converter endpoint with a high level of accuracy. However, the different shortcomings of each have prevented large-scale application in real production environments. CBR prediction model has limited scope to solve the problem. CNN model has complex data processing and no memory. RELM model has randomly given input layer weights and hidden layer deviations. In this study, correlation analysis was used to analyze the factors influencing the carbon content at the endpoint of converter steelmaking. A feasible model was established and applied to predict the carbon content at the endpoint of converter using t-distributed stochastic neighbor embedding (t-SNE), particle swarm optimization (PSO), and backpropagation (BP) neural network. The learning rate, training times, and hidden layer nodes number of the prediction model were optimized. The prediction hit ratios for the carbon content in the error ranges of ± 0.003%, ± 0.01%, and ± 0.02% are 61%, 86%, and 98%, respectively. Meanwhile, apply the established model to actual production, the carbon content of the product can be stably controlled between the lower and median limits, the control effect is significantly better than traditional methods. The results demonstrate that the t-SNE-PSO-BP model performs better than the known models. The accurate prediction of the carbon content at the endpoint of converter can greatly contribute to realizing a “narrow composition control” of the molten steel. Realize accurate prediction of carbon content at the endpoint of converter smelting, and has been effectively applied to industrial production.
Under the traditional method of predicting the endpoint carbon content of the converter, the hit rate of the middle and lower limits of the carbon content in the product is 48%. The t-SNE-PSO-BP model predicts the carbon content at the endpoint of the converter model, and the product carbon content can be controlled stably between 0.21–0.23%. According to the study results and actual application effects, use the t-SNE-PSO-BP model to predict the carbon content at the endpoint of the converter is appropriate, and is conducive to the “narrow composition control” of the steel composition in the converter steelmaking process.
Tapping weight is an important parameter in converter blowing process, wherein precisely predicting the quantity of steel required in an alloy baking converter can effectively guide the requirement of alloy ingredients. In practical production, the main approach is empirical estimation, despite its low accuracy. Employing a general neural network model for prediction requires to gather the converter blowing parameters and endpoint temperature measurement sampling parameters as the model inputs. However, this data cannot be obtained until the blowing process reaches its endpoint, rendering it impractical for alloy batching that requires advance preparation for baking. In this study, a principal component analysis–whale optimization algorithm–backpropagation algorithm (PCA–WOA–BP) neural network tapping prediction model is developed using raw material parameters available before alloy baking as input. This model is integrated into the intelligent alloy reduction model of a factory. The model achieves a regression coefficient R² of 0.823, with 98.53% of furnaces having a prediction error of less than 2 t. The tapping weight of 20 consecutive heats in actual production is predicted; the error range is less than 80% within 1 t, and the converter tapping weight is predicted accurately.
Accurate prediction of temperature and carbon content of liquid steel plays an important role in steelmaking process. In order to enhance the accuracy of predicting the basic oxygen furnace (BOF) end-point temperature and carbon content of liquid steel, a hybrid model based on principal component analysis (PCA) − genetic algorithm (GA) − backpropagation (BP) neural network is proposed. PCA is used to reduce the dimensionality of the input variables and eliminate the collinearity among the variables, then the obtained principal components are seen as new input variables of the BP neural network. GA is employed to optimize the initialized weights and thresholds of the BP neural network. Data from a 250t BOF of H steel plant in China is used to test and validate the model. The results show that the prediction accuracy of the single output models is higher than that of the dual output models. The PCA-GA-BP neural network model with single output shows higher prediction performance than others. The root mean square error of temperature between predicted and actual values is 7.89, and that of carbon content is 0.0030. Therefore, the model can provide a good reference for BOF end-point control.
Highly accurate prediction of converter molten steel end temperature is important foundation of realizing intelligent smelting in converter procedure. Single model prediction of converter end temperature has problems such as weak generalization ability and difficulty in increasing accuracy. To tackle these problems, this review proposes a modeling method based on Bayesian formula that dynamically integrate multiple models with different features. First, various data patterns and features are explored using different modeling methods and respective prediction model for converter molten steel end temperature is constructed. Then, the value range of the objective is discretized and the confidence levels of the prediction results from different models are computed based on Bayesian formula. Last, the prediction values of different models are weighted according to their confidence level and a comprehensive prediction result is obtained. Testing and comparison are carried out on the integration model and the three single models (support vector regression, random forest, and BP neural network) using actual production data. Results show that the integration model based on Bayesian formula can further increase the prediction accuracy effectively on top of single model prediction accuracy.
Availability of large amounts of data from connected processes and breakthrough innovations in computing tools create exciting opportunities for applications of Artificial Intelligence (AI) in Steel Industry. Data-driven predictive analytics and Machine Learning (ML) are already playing a significant role in shaping the Intelligent Plant of the future. Combining long-standing process expertise with a modern digital infrastructure, Danieli Automation has included AI as a strategic asset in DIGI&MET smart factory model. As illustrated through the paper, business benefits guaranteed from this model are already clear from the operational feedback of several real-life experiences from steelmaking to rolling and finishing lines.
To facilitate industrial vessel lining design for various material properties and lining configurations, a method, being composed of the back‐propagation artificial neural network (BP‐ANN) with multiple orthogonal arrays, is developed, and a steel ladle from secondary steel metallurgy is chosen for a case study. Ten geometrical and material property variations of this steel ladle lining are selected as inputs for the BP‐ANN model. A total of 160 lining configurations nearly evenly distributed within the ten variations space are designed for finite element (FE) simulations in terms of five orthogonal arrays. Leave‐One‐Out cross validation within various combinations of orthogonal arrays determines 7 nodes in the hidden layer, a minimum ratio of 16 between dataset size and number of input nodes, and a Bayesian regularization training algorithm as the optimal definitions for the BP‐ANN model. The thermal and thermomechanical responses of two optimal lining concepts from a previous study using the Taguchi method are predicted with acceptable accuracy.
Data mining and analytics have played an important role in knowledge discovery and decision making/supports in the process industry over the past several decades. As a computa-tional engine to data mining and analytics, machine learning serves as basic tools for information extraction, data pattern recognition and predictions. From the perspective of machine learning, this paper provides a review on existing data mining and analytics applications in the process industry over the past several decades. The state-of-the-art of data mining and analytics are reviewed through eight unsupervised learning and ten supervised learning algorithms, as well as the application status of semi-supervised learning algorithms. Several perspectives are highlighted and discussed for future researches on data mining and analytics in the process industry.
In galvanising line of cold rolling mill, mechanical properties, i.e. yield strength (YS) and ultimate tensile strength (UTS), are achieved by controlling the key process parameters within specified limits. In this paper, a feed-forward back-propagation artificial neural network (ANN) is proposed to predict the mechanical properties of a coil from its chemical composition, thickness, width and key galvanising process parameters. Principal component analysis is used to avoid redundancy and collinearity effects in input variables for the ANN. The model predicted the YS and UTS with an accuracy of ±10 megapascal (MPa) for 90% of the data. The model was implemented in the continuous galvanising line of Tata Steel, India. An online quality monitoring system was developed to monitor the predicted mechanical properties and process parameters of a galvanised coil. This system helps quality team in decision making.
Big data analytics is the journey to turn data into insights for more informed business and operational decisions. As the chemical engineering community is collecting more data (volume) from different sources (variety), this journey becomes more challenging in terms of using the right data and the right tools (analytics) to make the right decisions in real time (velocity). This article highlights recent big data advancements in five industries, including chemicals, energy, semiconductors, pharmaceuticals, and food, and then discusses technical, platform, and culture challenges. To reach the next milestone in multiplying successes to the enterprise level, government, academia, and industry need to collaboratively focus on workforce development and innovation. Expected final online publication date for the Annual Review of Chemical and Biomolecular Engineering Volume 8 is June 7, 2017. Please see http://www.annualreviews.org/page/journal/pubdates for revised estimates.
The main factors that determine heat losses in a steelmaking ladle are reviewed. Controlling the temperature of the liquid steel from the primary refining until casting is essential to achieve product quality as well as energy savings. Among other objectives this study sought to assess the relative weight of heat losses to the refractory lining and the thermal losses through the slag layer. Temperature, slag thickness and properties, refractory temperature and physical properties and convection characteristics are the defining factors. Numerical integration was carried out with a computer code specifically developed for the purpose .The results show that heat loss through the refractory linings at side wall is relatively high compared to that through the bottom and slag layer, Heat loss through the double slag layer depends on temperature, slag thickness, heat transfer coefficient and time.
Aiming at the characteristics of the practical steelmaking process, a hybrid model based on ladle heat status and artificial neural network has been proposed to predict molten steel temperature. The hybrid model could overcome the difficulty of accurate prediction using a single mathematical model, and solve the problem of lacking the consideration of the influence of ladle heat status on the steel temperature in an intelligent model. By using the hybrid model method, forward and backward prediction models for molten steel temperature in steelmaking process are established and are used in a steelmaking plant. The forward model, starting from the end-point of BOF, predicts the temperature in argon-blowing station, starting temperature in LF, end temperature in LF and tundish temperature forwards, with the production process evolving. The backward model, starting from the required tundish temperature, calculates target end temperature in LF, target starting temperature in LF, target temperature in argon-blowing station and target BOF end-point temperature backwards. Actual application results show that the models have better prediction accuracy and are satisfying for the process of practical production.
A multi layered feed forward neural network model is being developed for the prediction of end blow oxygen in the LD converter using a two step process. In the first step intermediate stopping temperature is being predicted and using this as an input the end blow oxygen is predicted. In both the cases two hidden layers had given the best results compared to the single layer neural network. Intermediate and end blow temperatures played a vital role in end blow oxygen and intermediate stopping temperature predictions. The model acts a guide for the operator and thereby enhances the yield of the converter steel making process.
The heat transfer in a steelmaking ladle was studied. The evaluation of heat transfer of the steel was performed by measuring steel temperature in points including all refining steel process. In the ladle, the temperatures in the refractories and the shell were also measured. To evaluate the thermal profile between the hot and cold faces of the ladle in the slag line position, an experiment which shows the importance of thermal contact resistance was carried out. Higher heat losses in the tapping and the vacuum were verified. The temperature measurements of the ladle indicate distinct thermal profiles in each stage of steel refining. Moreover, as each stage of the process depends on the previous one, the complexity of the ladle thermal control is incremental. So a complete model of heat losses in the ladle is complex.
A statistical model for predicting the liquid steel temperature in the ladle and in the tundish is devel-oped. Given a large data set in a steelmaking process, the proposed model predicts the temperature in a seconds with a good accuracy. The data are divided into four phases at the mediation of five temperature measurements: before tapping from the converter (CV), after throwing ferroalloys into the ladle, before and after the Ruhrstahl-Heraeus (RH) processing, and after casting into the tundish in the continuous cast-ing (CC) machine. Based on the general state space modeling, the bootstrap filter predicts the tempera-ture phase by phase. The particle approximation technique enables to compute general-shaped probability distributions. The proposed model gives a prediction not as a point but as a probability distribution, or a predictive distribution. It evaluates both uncertainty of the prediction and ununiformity of the temperature. It is applicable to sensitivity analysis, process scheduling and temperature control. KEY WORDS: statistical modeling and simulation; liquid steel temperature control; general state space model; bootstrap filter; steelmaking.
A mathematical model for heat transfer during solidification in continuous casting of automobile steel, was established on researching under the influence of the solidifying process of bloom quality of CCM in the EAF steelmaking shop, at Shijiazhuang Iron and Steel Co. Ltd. Several steel grades were chosen to research, such as, 40Cr and 42CrMo. According to the results of the high temperature mechanical property tests of blooms, the respective temperature curves for controlling the solidification of different steels were acquired, and a simulating software was developed. The model was verified using two methods, which were bloom pinshooting and surface strand temperature measuring experiments. The model provided references for research on the solidifying process and optimization of a secondary cooling system for automobile steel. Moreover, it was already applied to real production. The calculated temperature distribution and solidification trend of blooms had offered a reliable theory for optimizing the solidifying process of blooms, increasing withdrawal speed, and improving bloom quality. Meanwhile, a new secondary cooling system was designed to optimize a secondary cooling water distribution, including choice and arrangements of nozzles, calculation of cooling water quantity, and so on.
In recent years, machine learning (ML) has become a key enabling technology for the sciences and industry. Especially through improvements in methodology, the availability of large databases and increased computational power, today’s ML algorithms are able to achieve excellent performance (at times even exceeding the human level) on an increasing number of complex tasks. Deep learning models are at the forefront of this development. However, due to their nested non-linear structure, these powerful models have been generally considered “black boxes”, not providing any information about what exactly makes them arrive at their predictions. Since in many applications, e.g., in the medical domain, such lack of transparency may be not acceptable, the development of methods for visualizing, explaining and interpreting deep learning models has recently attracted increasing attention. This introductory paper presents recent developments and applications in this field and makes a plea for a wider use of explainable learning algorithms in practice.
This paper introduces the development of empirical predictive models and detection methods that are incorporated into a water detection framework for an industrial steelmaking electric arc furnace (EAF). The predictive models investigated in this work are designed based on different techniques such as statistical fingerprinting, artificial neural network (ANN), and multiway projection to latent structures (MPLS). Robustness issues related to each method are discussed and performance comparisons have been done for the presented techniques. Furthermore, model fusion theory has been applied to improve the prediction accuracy of the developed models’ defined output- the value of off-gas water vapor- which is known as one the most vital variables to guarantee a safe and reliable operation. Finally based on the proposed predictive models, a water leak detection methodology is introduced and implemented on an industrial AC EAF and a comprehensive discussion has been done to evaluate the performance of the developed algorithm. To this aim, two fault detection methods have been applied. Fault detection method #1 has been designed using statistical fingerprinting technique, while the other one has been developed based on machine learning-based models and also fusion of the models’ outputs.
Mechanical property prediction is considerably demanded by steel manufacturers because it helps to avoid quality problems and increase productivity. However, prediction of yield strength, tensile
strength and elongation of steel after annealing is a hard task because the relation between mechanical properties and process parameters like cold rolling reduction rate, annealing time, annealing temperature and alloying elements of steel is highly nonlinear. Moreover, analytical
models mostly depend on experimental constants which are hard to obtain for industrial processes due to the wide range of the process parameters. Using neural network models is more practical, but it may lead to unphysical results. To increase accuracy and avoid unphysical results, analytical annealing models are redeveloped and integrated into the neural network models. Moreover, more than 50,000 tensile test data belonging to the production of past 3 years are used during model construction. Additionally, models are developed for different steel categories based on historical data, and these are aluminum killed, bake hardening, dual phase, high strength low alloy, interstitial free, and rephosphorized steels. According to the results, models have less than 1.7% error for tensile strength, 3.3% error for yield strength and 4.7% error for elongation.
A prediction model based on the principal component analysis (PCA) and back propagation (BP) neural network is proposed for BOF end-point phosphorus content, based on the characters of BOF metallurgical process and production data. PCA is used to reduce dimensionality of the factors influencing end-point phosphorus content, and eliminate the correlations among the factors, and then the obtained principal components are used as BP neural network input vectors. The combined PCA-BP neural network model is trained and tested by history data, and is further compared with multiple linear regression (MLR) model and BP neural network model. The results of the comparison show that the PCA-BP neural network model has the highest prediction accuracy and PCA improved the generalization capability. Finally, online prediction system of BOF end-point phosphorus content based on PCA and BP neural network is developed and applied in actual productive process. Field application results indicate that the hit rate of end-point phosphorus content is 96.67%, 93.33% and 86.67% respectively when prediction errors are within ±0.007%, ±0.005% and ±0.004%. The combined PCA-BP neural network model has achieved the effective prediction for end-point phosphorus content, and provided a good reference for end-point control and judgment of quick direct tapping of BOF.
The paper presents the novel method of fast and contactless estimation of iron oxide (FeO) concentration in steel slag during discharge process in steelworks. The infrared imaging and artificial neural network are the key tools used in the research. The imaging system consists of three cameras that work in the different wavelength ranges. The novel idea based on steel and slag radiation parameters extracted from the sequences of images is described. Radiation data that are the most correlated with FeO content in steelmaking slag are selected and use as input variables for ensemble of Artificial Neural Networks (ANN). Different parameters and configuration of the ANN were tested to define the most effective ones. The final result of FeO content estimation is presented and validated with the values obtained from the chemical analysis.
With the availability of large databases and recent improvements in deep learning methodology, the performance of AI systems is reaching or even exceeding the human level on an increasing number of complex tasks. Impressive examples of this development can be found in domains such as image classification, sentiment analysis, speech understanding or strategic game playing. However, because of their nested non-linear structure, these highly successful machine learning and artificial intelligence models are usually applied in a black box manner, i.e., no information is provided about what exactly makes them arrive at their predictions. Since this lack of transparency can be a major drawback, e.g., in medical applications, the development of methods for visualizing, explaining and interpreting deep learning models has recently attracted increasing attention. This paper summarizes recent developments in this field and makes a plea for more interpretability in artificial intelligence. Furthermore, it presents two approaches to explaining predictions of deep learning models, one method which computes the sensitivity of the prediction with respect to changes in the input and one approach which meaningfully decomposes the decision in terms of the input variables. These methods are evaluated on three classification tasks.
Because of demand for lower emissions and better crashworthiness, the use of hot stamped 22MnB5 boron steel has greatly increased in manufacturing of automobile components. However, for many applications it is required that only certain regions in hot stamped parts are fully hardened whereas other regions need be more ductile. The innovative process of tailored hot stamping does this by controlling the localized microstructures through tailored cooling rates by dividing the tooling into heated and cooled zones. A barrier to optimal application of this technique is the lack of reliable phase distribution prediction model for the process.
We present a novel Artificial Neural Network (ANN) based phase distribution prediction model for tailored hot stamping. The model was developed and validated using data generated from extensive thermo-mechanical physical simulation experiments and instrumented nanoindentation based phase quantification method. Advanced statistical techniques were used for preventing overfitting, for making the optimal use of available experimental data and for quantification of prediction uncertainty. The final predictions made by the ANN model during its independent validation have shown good agreement with the experimentally generated data and have a RMS prediction error of just 7.7%, which is a significant improvement over the existing models.
In the steelmaking continuous-casting (SCC) process, scheduling problem is a key issue for the iron and steel production. To improve the productivity and reduce material consumption, optimal models and approaches are the most useful tools for production scheduling problems. In this paper, we firstly develop a mixed integer nonlinear mathematical model for the SCC scheduling problem. Due to its combinatorial nature and complex practical constraints, it is extremely difficult to cope with this problem. In order to obtain a near-optimal schedule in a reasonable computational time, Lagrangian relaxation approach is developed to solve this SCC scheduling problem by relaxing some complicated constraints. Owing to the existence of the nonseparability coming from the product of two binary variables, it is still hard to deal with this relaxed problem. By making use of their characteristics, the subproblems of the relaxed problem can be converted into different difference of convex (DC) programming problems, which can be solved effectively by using the concave-convex procedure. Under some reasonable assumptions, the convergence of the concave-convex procedure can be established. Furthermore, we introduce an improved conditional surrogate subgradient algorithm to solve the Lagrangian dual (LD) problem and analyze its convergence under some appropriate assumptions. In addition, we present a simple heuristic algorithm to construct a feasible schedule by adjusting the solutions of the relaxed problem. Lastly, some numerical results are reported to illustrate the efficiency and effectiveness of the proposed method.
A mathematical model suitable as a tool for improving temperature control in the steel plant is presented. With this tool, a number of variables such as holding time, material choice and refractory layer thicknesses can be studied with regard to their influence on the steel temperature evolution during casting. In addition, the model can be used as a decision support and forms a basis for automation of temperature control.
The task of the “level-2᾿ process control system is the pre-calculation and setup of the mill’s actuators prior to processing. Today’s rolling mill process models are fairly mature. Typically, the adaptation of these mathematical models to process variations has been done using a fragmented, look-up table based approach. This approach has drawbacks due to the required size for the look-up tables for large product ranges and the lack of interpolation capability. The application of neural networks within a new control strategy has reduced these problems significantly. The application strategy for neural networks within hybrid systems, their role in various rolling mill applications (rolling force, stock temperature, width) and the benefits achieved by implementations of this technology are described.
Due to the lack of simple and effective data filtering method for multi-variable and numerous samples in BOF endpoint forecasting model, a method of outlier identification and judgment was introduced and applied to data screens for improving BOF endpoint forecasting model. The outside values as potential outliers are calculated using the method of five-number summary which is a robust estimation of the population parameter, and then the potential outliers are judged with the clustering method. By comparing the exceptional data from clustering analysis with the outside values from the five-number summary, the intersection of these two groups is regarded as the final outliers to be deleted; in addition, the exceptional data but not outside values are regarded as final exceptional data to be further analyzed; and the outside values but not exceptional data are regarded as final outliers to be deleted too. Finally, to verify the data selection, an improved BP-based neural network model is used to predict the end-point carbon content and temperature. By using this data pretreatment method, the absolute values of the mean and maximum training residuals of endpoint carbon and temperature decreased by 26.7%, 41% and 17.3%, 34.5% respectively; and those of the prediction decreased by 10%, 44.9% and 9.4%, 22.9% respectively. It is shown that the proposed method improves effectively the neural network model for BOF endpoint forecasting.
To control the molten steel temperature in a Ladle Furnace accurately, it is necessary to build a precise (i.e. accurate and good generalized) temperature prediction model. To solve this problem, ensemble modeling methods have been applied to predict the temperature. Now, in the production process, large-scale data with more helpful information are sampled, which provides possibilities to improve the precision of the temperature prediction. Although most of the existing ensemble temperature models have strong learning ability, they are not suitable for the large-scale data. In this paper, to solve the large-scale issue, the Tree-Structure Ensemble General Regression Neural Networks (TSE-GRNNs) method is proposed. Firstly, small-scale sample subsets are constructed based on the regression tree algorithm. Secondly, GRNN sub-models are built on sample subsets, which can be designed very quickly and cannot converge to poor solutions according to local minima of the error criterion. Then, the TSE-GRNNs method is applied to establish a temperature model. Experiments show that the TSE-GRNNs temperature model is more precise than the other existing temperature models, and meets the requirements of the RMSE and the maximum error of the molten steel temperature prediction in Ladle Furnace.
Steel strip surface defects recognition is very important to steel strip production and quality control, which needs further improvement. In this paper, an end-to-end surface defects recognition system is proposed for steel strip surface inspection. This system is based on the symmetric surround saliency map for surface defects detection and deep convolutional neural networks (CNNs) which directly use the defect image as input and defect category as output for seven classes of steel strip defects classification. The CNNs are trained purely on raw defect images and learned defect features from the training of network, which avoiding the separation between feature extraction and image classification, so that forms an end-to-end defects recognition pipeline. To further illustrate the superiority of the defect recognition methods with CNNs, an authoritative and standard steel strip surface defect dataset - NEU is also used to evaluate the defect recognition effect using CNNs. Experimental results demonstrate that the proposed methods perform well in steel strip surface defect detection of different types and achieve a high recognition rate for defect images. In addition, a series of data augmentation methods are discussed to analyze its effect on avoiding over-fitting for defects recognition.
The steel-making and continuous casting process (SCCP) is the bottleneck in iron and steel production. SCCP often involves various uncertainties such as emergency customer orders, inaccurate estimate of processing time or unpredictable machine breakdown. All these dynamic disturbance factors disturb the rhythm of the regular production and results in the reduction of productivity. How to make a robust and adaptable dynamic reactive scheduling in a computationally efficient manner and be able to assess the quality of the schedule to improve the steel production during SCCP are the key factors for iron and steel manufacturing productivity. This problem is difficult due to various processing routs selection, complicated dynamic disturbances, and combinational explosion of the optimization search space. This paper presents a novel Lagrangian relaxation neural network (LRNN) for the scheduling of SCCP by combing recurrent neural network optimization ideas for constraint handling. An improved multiagent framework (IMAF) with an emphasis on robustness, adaptability, and optimization speed is introduced, and then a combinatorial auction mechanism based on improved surrogate subgradient algorithm with an emphasis on computational efficiency is applied in the mulitagent system architecture. The neuron-based stochastic dynamic programming (NSDP) method is adapted to obtain the subproblems. The approach provides both a theoretical basis and some experimental justifications for a dynamic scheduling combining real-time and predictive decision making. It resolves various disruptions as flexibly as dispatching rules while providing more stability. The approach has been tested by using practical data from Shanghai Baoshan Steel Plant of China. Numerical results demonstrate solution quality, computational efficiency, and values of new features in the IMAF architecture.
Since the Brazilian inclusion in the global market, search for productivity and product quality improvement became essential for the companies to survive. However, due to energy costs rise, national steel industries are investing in electrical power generation in partnership with energy supply companies aiming at overall cost reduction. Therefore, actions that search for energy consumption reduction and productivity increase became priority for their research and development projects. The ladle furnace of V&M is one of the largest energy consuming units in the steel plant, consuming up to 2,400 MWh on average a month. Due to process complexity, system optimization became difficult to be implemented using conventional parametric approaches. However, applications of computational intelligence have been used as important alternative approaches to process modeling. Due to the little knowledge about the ladle furnace dynamics and the high variability of specific energy consumption, the use of neural networks was applied as a non parametric approach. This paper demonstrates the use of neural networks in complex industrial problems by applying it to the steel temperature prediction of the ladle furnace process. This paper shows that the neural network used yielded high generalization capability by obtaining smaller mean error on the test data than the expected error specified by the steel temperature measurement instrument. In addition, this paper shows that the use of this neural thermal model resulted in productivity increase, operational and energy cost reduction.
Controlling temperature of molten steel is crucial for product quality in continuous casting. In this paper, sensitivity analysis is carried out on a statistical model for predicting temperature in tundish, and important and influential operations on temperature are identified.
End-point prediction is one of the most difficult problems in basic oxygen furnace (BOF) steelmaking process. To address this problem, some researchers have proposed some methods based on flame image processing and pattern classification. Because of the dynamically changing flame and real-time needs during the blowing process, there are still some issues that need to be solved. We propose a novel method based on accurate and fast multi flame features extraction and general regression neural network (GRNN). Firstly, flame images were acquired, and then the background of each image was removed via color similarity determination algorithm; secondly, color, texture, and boundary features were extracted; the fast and robust boundary and texture features were extracted by using the proposed methods, and these features were tested for their validity to the end-point prediction via comparing them with some other similar methods; finally, the prediction model was built using multi-features and GRNN. The experimental results demonstrated that it is accurate and fast to use the proposed method to the BOF end-point predict.
The hybrid method composed of clustering and predicting stages is proposed to predict the endpoint phosphorus content of molten steel in BOF (Basic Oxygen Furnace). At the clustering stage, the weighted K-means is performed to generate some clusters with homogeneous data. The weights of factors influencing the target are calculated using EWM (Entropy Weight Method). At the predicting stage, one GMDH (Group Method of Data Handling) polynomial neural network is built for each cluster. And the predictive results from all the GMDH polynomial neural networks are integrated into a whole to be the result for the hybrid method. The hybrid method, GMDH polynomial neural network and BP neural network are employed for a comparison. The results show that the proposed hybrid method is effective in predicting the endpoint phosphorus content of molten steel in BOF. Furthermore, the hybrid method outperforms BP neural network and GMDH polynomial neural network.
A temperature prediction model has been developed for controlling the casting superheat temperature. For ease of implementation, the model is intentionally made simpler having a combination of a one dimensional heat transfer model and a simple regression model. The model is based on the fact that the BOF temperature of the liquid steel along with the bath cooling behaviours controls the aimed casting superheat temperature. Starting with the steel liquidus temperature and calculating the required steel temperature backwards throughout the process line gives the targeted BOF tap temperature. The on-line picking up of actual data helps the model to predict for the next stage more accurately in forward direction. Based on the predicted steel temperature, plant operators can take any necessary corrective action like additional ladle heating and extra/reduced argon stirring to ensure the aim final ladle/tundish temperatures at the casting are achieved.
In this paper we describe an optimization procedure for planning the production of steel ingots in a steelmaking-continuous casting plant. The strict requirements of the production process defeated most of the earlier approaches to steelmaking-continuous casting production scheduling, mainly due to the lack of information in the optimization models. Our formulation of the problem is based on the alternative graph, which is a generalization of the disjunctive graph of Roy and Sussman. The alternative graph formulation allow us to describe in detail all the constraints that are relevant for the scheduling problem. We then solve the problem by using a beam search procedure, and compare our results with a lower bound of the optimal solutions and with the actual performance obtained in the plant. Computational experience shows the effectiveness of this approach.
The efficient and reliable control of a reheating furnace is a challenging problem, due to: (a) the many different types of billets to be processed, (b) the strong intercorrelation among process variables, (c) the large dimension of the input and output space, (d) the strong interaction among process variables, (e) a large time delay, and (f) highly nonlinear behavior. Thus, conventional reheating furnace operation has been heavily dependent upon look-up tables which list the optimal set points. This paper describes a modified modular neural network for the supervisory control of a reheating furnace. Based on the divide-and-conquer concept, a modular network is capable of dividing a complex task into subtasks, and modeling each subtask with an expert network. To model such activities, a gating network is used for the classification and allocation of the input data to the corresponding expert network. To overcome the correlation effects among process variables and the problem of dimensionality, principal component analysis (PCA) has been employed to remove the correlation and reduce the problem dimension. From PCA analysis, it was possible to decide on the optimal dimension for the problem, to describe the dynamic behavior of the furnace. The proposed neural network has been trained and tested using operational data from the reheating furnace and has been implemented on the wire rod mill process of POSCOTM.
The operation and control of blast furnaces poses a great challenge because of the difficult measurement and control problems associated with the unit. The measurement of hot metal composition with respect to silica and sulfur are critical to the economic operation of blast furnaces. The measurement of the compositions require spectrographic techniques which can be performed only off line. An alternate technique for measuring these variables is a Soft Sensor based on neural networks. In the present work a neural network based model has been developed and trained relating the output variables with a set of thirty three process variables. The output variables include the quantity of the hot metal and slag as well as their composition with respect to all the important constituents. These process variables can be measured on-line and hence the soft sensor can be used on-line to predict the output parameters. The soft sensor has been able to predict the variables with an error less than 3%. A supervisory control system based on the neural network estimator and an expert system has been found to substantially improve the hot metal quality with respect to silicon and sulfur.
In continuous casting, the cooling-solidification process must be based on the adaptation of heat transfer, which is directly connected to casting conditions such as casting speed, casting temperature, and cooling parameters. Most control schemes are based on the static relation between casting speed and water flow rate in each cooling zone; this constitutes an open loop that does not consider the dynamic surface temperature, which is an important parameter for the final slab quality. In steelmaking, the casting-speed changes affect the global heat transfer. An optimal operation requires an adjustment of the process control variables, i.e., global heat transfer. A learning neural network (NN) allows the identification and the control of a nonlinear heat transfer model in the continuous casting process. A heat transfer model was developed using the dynamic heat balance. A comparison between the experimental open loop results and those of the model simulation is considered. Following adaptation, the model is used for controlling the slab surface temperature in closed loop, using NN technology and PID controllers. The NN identification and control strategy gives a stable temperature closed loop control comparatively to the conventional PID.
Controle da temperatura do aço l ıquido em uma aciaria el etrica (Temperature control of liquid steel in an electric steelwork)
- N F Ferreira
Ferreira, N. F. (2000). Controle da temperatura do aço
l ıquido em uma aciaria el etrica (Temperature control of
liquid steel in an electric steelwork). Tese de Doutorado.
An artificial neural network model for the comprehensive study of the solidification defects during the continuous casting of steel
- I Ghosh
- N Chakraborty
Ghosh, I., & Chakraborty, N. (2018). An artificial neural
network model for the comprehensive study of the
solidification defects during the continuous casting of
steel. Computer Communication & Collaboration,
6(1-2), 1-14.
Model for prediction of the reduction of steel temperature between the ladle furnace and the continuous casting in the steelmaking process
- I C D Duarte
Duarte, I. C. D. (2009). Model for prediction of the reduction of steel temperature between the ladle furnace and
the continuous casting in the steelmaking process.
Department of Chemical Engineering, UFMG (in
Portuguese).
Steel thermal control in the steel plant of V&M do Brasil. IX ABM Seminar in Process Automation
- T Fujii
- G Lenna
- P Sampaio
- L Muradas
Fujii, T., Lenna, G., Sampaio, P., & Muradas, L. (2005).
Steel thermal control in the steel plant of V&M do
Brasil. IX ABM Seminar in Process Automation (pp.
462-470) (in Portuguese).
Computational approaches to materials design: Theoretical and practical aspects
- S Datta
- J P Davim
Datta, S., & Davim, J. P. (2016). Computational
approaches to materials design: Theoretical and practical
aspects. Engineering Science Reference.