|Univerza v Ljubljani , Fakulteta za elektrotehniko|
|Research projects (co)funded by the Slovenian Research Agency .|
|Member of University of Ljubljana||UL Faculty of Electrical Engineering|
|Project||Development of decision support methods based on smart sensors for steel recycling process in electric arc furnace|
|Period||1.1.2016 - 31.12.2018|
|Range on year||
|Research Organisation||link on SICRIS|
|Abstract||Recycling of the steel in electric arc furnaces (EAFs) represents approximately one third of the global, one half of the European and complete Slovenian production of the steel. Increased demands regarding the quality of the steel, economical, ecologic and technological aspects together with fierce market competition dictate optimized and software-supported manufacturing processes. Due to the nature of the EAF steel-recycling process (high temperatures and electric currents), performance of the crucial process measurements is difficult if not practically impossible. Consequently, monitoring and control of the melting process is performed using the operator's experience and is based on indirect measurements (e.g. power-on time, consumed energy, arc stability etc.) and not on the actual conditions in the EAF (e.g. stage of melting, bath composition, bath temperature), which leads to suboptimal operation, i.e. lower energy and raw material efficiency, increased off gas and CO2 emissions, decreased quality of the steel; and consequently higher operational costs. Furthermore, operational efficiency is influenced also by variable composition of the input materials (steel scrap, non-metallic additives). The issue can be resolved using a combination of advanced process-modelling methods, smart sensors, optimization techniques and decision support methods. The afore mentioned methods involve available process measurements, optimization methods and decision support algorithms, while their integration into a complete software solution forms a so called embedded system for optimal operation of the EAF. The system uses process measurements as inputs, in order to provide a better insight into the current EAF conditions and to suggest the most appropriate action to the user, leading to more efficient operation of the EAF. Using smart sensors based on mathematical models, which are designed in compliance with the physical laws and using available measurements as inputs, crucial process values, which are not measured, can be estimated in parallel to the EAF process with high accuracy. Since the EAF processes are complex, nonlinear and time variant, the development of optimization methods represents a challenging task, which requires the implementation of the most efficient methods for this purpose, i.e. evolutionary and genetic algorithms, fuzzy inference systems and particle swarms. In the frame of this project, some of the already developed mathematical EAF models (electrical, chemical, heat- and mass-transfer) will be used and adopted for the needs of smart sensing and real-time optimization. Due to the complexity of the steel-bath behavior, heat transfers from liquid to solid steel and chemical reactions, some of the existent models will be re-validated, modified and re-parameterized using a more complex computational fluid dynamics (CFD) approach and a material-properties database. The final result of the project will be a simulation environment, comprising of EAF process models, smart sensors, complex optimization methods and decision support algorithms. The developed software will be tested and validated using simulation studies of the current (measured) and optimal (operation according to suggested actions) operation of the EAF, which will demonstrate the differences between both EAF controls and their effect on economic balance. A combination of a real process, decision support system based on computer simulation and process models cannot be found in the EAF operation up to date. The proposed project thus represents an original approach to optimization of the steel-recycling process (higher steel yield, lower energy, raw material and additive consumption, shorter production times, higher steel quality etc.). The introduction of the enhanced EAF control, based on online optimization and decision-support tools, indirectly leads to improved economic, ecological and technological aspects of the mills, with such system installed.|
|Researchers||link on SICRIS|
|The phases of the project and their realization|| The work plan of the proposed project is
divided into five main workpackages:
adjustment of the existent mathematical models (electrical, hydraulical, thermal, chemical and mass-transfer) for their integration into smart sensors, process optimization and decision support system,
development, upgrade and parameterization of the EAF models using computational fluid dynamics (CFD),
upgrade of the designed models into smart sensors and development of the optimization framework to find the optimal scenarios and melting programs,
development of the decision support system and integration of the process models, smart sensors and optimization algorithms and their implementation into one software solution,
performance of comparative simulation studies between current (non-optimal) and optimized EAF operation.
Below, a detailed presentation of the problem, methods, knowledge and deliverables for each workpackage is given.
Adjustment of the existent mathematical models (electrical, hydraulical, thermal, chemical and mass-transfer) for their integration into smart sensors, process optimization and decision support system.
1.1 Description of the 1. workpackage:
Mathematical models, which were developed in a basic domestic project for Slovenian research agency – ARRS (J2-2310 - Monitoring and Control of Steel Melt Quality in Electric Arc Furnace) describe the relations between input data, physical phenomena and output signals during the EAF steel-recycling process. The developed models are based on dynamic formulations and are thus, in contrast to static formulations, capable of estimating the process values in each time moment of the melting-process simulation using available input signals and prior values of the states. Using fundamental physical laws as a modelling basis (Kirchhoff laws, laws of thermochemistry and thermodynamics, conservation of energy and mass etc.), a complex model of the EAF processes has been developed, which describes the conditions in the EAF during the steel melting with sufficient accuracy. The numerical integration method used to solve the differential equations was selected as a fixed step Euler method. All developed models were validated using available operational measurements, such as electrical values, initial chemical composition of the scrap, end-point chemical composition of the steel, end-point chemical composition of the slag, tap-to-tap time, energy consumption, steel, additive, oxygen and carbon consumption, steel yield and an accurate course of the melting program (transformer and reactor taps; times and amounts of charging; duration and rate of oxygen lancing; duration and rate of carbon injection; times and amounts of slag-forming additives – CaO, MgO, Al2O3; times and amounts of metallic additives – FeMn, FeSi, SiMn Al etc.). Comparative simulation studies between measured and simulated data have shown that the developed model is accurate and appropriate for further use. Since the goal of the project (J2-2310) was to design the models for simulation purposes, their inclusion into an EAF simulator and for offline process optimization, the models were not built for the needs of online process-optimization, smart sensors or decision support system.
Due to the before mentioned limitations, the existent models need to be modified in order to be suitable for their inclusion into the proposed technologies (smart sensors, online process optimization and decision support system). Since optimization of the process is a highly time-consuming task, solving speed of the models is also one of the issues which needs to be enhanced. The solution of the later lies in a replacement of the integration method for a faster variable-step method, which ensures faster solving of the models as the Euler method, while preserving the accuracy of the calculations. Substitution of the integration method is not simple, as besides the appropriate transformation of the program code, proper solving of the discontinuities appearing in the process (instantaneous addition of the steel, additives, oxygen, carbon etc.) needs to be handled. Discontinuities represent one of the biggest issues when solving the continuous processes, as the discontinuity of the state leads to infinite derivative. Such problems do not appear in the current implementation of the solving method, as the procedure is programmed directly in the software code and is not influenced by the occurrence of discontinuities. This however does not apply to the dedicated methods for numerical solving of differential equations – solvers. Modifications of the current functionality of the models also include a development of the additional modules for online input data feed and output data storage. The later can be implemented by either data files or data bases. Such modifications are necessary whether the developed models are to be used for the needs of smart sensors and online process optimization. Optimization procedure also requires all process data to be archived (input/output data, states), which enables it to check different melting scenarios in each simulation step, i.e. simulation can be performed from the selected time on with different input data. In this manner, optimal melting scenarios can be found quicker, since it is not required to run the simulation for the complete heat over and over again.
1.2 Description of methods and knowledge in the 1. workpackage:
To successfully implement the 1. workpackage of the project, methods and knowledge of numerical integration methods, simulation of the continuous processes with discontinuities and data bases will be required. Existent models will be modified in order to solve them using the integrated solving method. For this reason, the models need to be reformulated to express the process states. In this manner, the model is reformed to two main parts; the first, represented by all calculations involving differential equations; and the second, represented by other algebraic equations and relations between differential equations, inputs and outputs. The selected integration method will be based on numerical differentiation formulas (NDFs), which are particularly used to solve the stiff systems. When implementing the integration method, special attention must be paid to the occurrence of state discontinuities, which can lead to numerical issues when solving the differential equation due to the theoretically infinite derivative at discontinuity. This can either lead to inaccurate calculation and computational errors, longer solving times or even failure of the method. The issue can be solved in several manners; one of the simpler approaches is to reform the discontinuous event (instantaneous change in infinitely short time) to continuous event (continuous change in very short time), for which the integration method will perform poorer, but will avoid numerical issues. Besides the modification of the integration method, the model needs to be supplemented by additional modules for online reading/writing of the input/output data and archiving of all process values (inputs, outputs and states). Data files or data bases can be used to implement such module, depending on the read/write access speed. Whether data base should be selected, MySQL is one of the options. One the other hand, whether data files should be selected, MS Excel is one of the options.
1.3 Deliverables of the 1. workpackage:
Deliverables of the 1. workpackage are the following:
replacement of the existent integration method (fixed step Euler) for a variable order method based on numerical differentiation formulas (NDFs), in order to speed up the model solving,
handling of the discontinuities when using the NDF solver, in order to properly solve such issues,
upgrade of the existent models with additional modules of online input data feed and output data storage, in order to use them in smart sensors and optimization procedures,
upgrade of the existent models with a functionality of archiving of all process values (inputs, outputs, states), in order to speed up the search for optimal melting scenarios.
Development, upgrade and parameterization of the EAF model using computational fluid dynamics (CFD)
2.1 Description of the 2. workpackage:
The computational modelling of the arc furnace includes a complex interaction of electromagnetic phenomena, heat and mass transfer phenomena and chemical reactions in the molten steel. Many of the interrelated physical, chemical and dynamic aspects are still not very well understood. The process involves multiphase, high temperature reduction reactions, energy conversion and distribution from electric power through arcs and conduction. Submerged arc furnaces have the interesting property that the electrical side of the furnace can be measured and controlled very well, but that one cannot characterize the particulate feed or define the distributed properties of the process very accurately. To better understand and control these furnaces one has to combine electrical and metallurgical aspects in such a way that both aspects are taken care of in the most optimal way.
With the progress of thermodynamic databases and the availability of faster and more powerful computing system, coupling fluid flow simulations and thermodynamic databases is becoming more frequent. Commercial computational thermodynamic software packages provide a special module that can be adopted by computational fluid dynamics (CFD) or numerical simulation codes to calculate thermodynamic equilibration in conjunction with numerical calculations. Respectively, complex chemical reactions and phase equilibria occurring throughout the steelmaking process can be calculated over wide ranges of temperature, oxygen potential and pressure.
2.2 Description of methods and knowledge in the 2. workpackage:
In the framework of the present project, a qualitative CFD model framework, which allows for the investigation of the changing electrical data on operating conditions, will be developed. All the aspects of the furnace will be taken into account, except the complexity of the geometry that is furnace dependent. It is expected that the achieved state of the knowledge of the project will in the next stage allow for modelling of the specific furnace and will be supported by the industry. The model will respectively be based on axisymmetric geometry. A ferrophosphorous layer, a slag layer, a solid burden layer and open space between the ceiling and solid feed will be taken into account. The typical wall thicknesses and composition will be used.
The model of the interior of the furnace will be based on a combination of the porous material, semi-molten material and completely molten material. The mathematical description of the problem will be based on volume averaging concept. Mass, energy, momentum, and species transfer equations will be considered together with low Reynolds number turbulent dissipation rate and turbulent energy generation rate equations. Variable viscosity and chemical reactions will be taken into account. The semi-molten mushy zone will be modelled as a Darcy-Brinkman porous media with Kozeny-Karman permeability relation, where the morphology of the porous media will modelled by a variable value. The incompressible turbulent flow of the molten steel and slag will be described by the Low-Reynolds-Number (LRN) k-epsilon turbulence model, closed by the Abe-Kondoh-Nagano closure coefficients and damping functions. The respective equations lead to solution of the transport equation. The numerical method will be established on explicit time-stepping, collocation with multiquadrics radial basis functions on non-uniform five-nodded influence domains, and adaptive upwinding technique. The velocity pressure coupling of the compressible flow will be resolved by the explicit Chorin’s fractional step method. The radiation will be taken into account through P-1 model that leads to solving of Poisson equation. The reaction kinetics will be based on coupling of the energy and species equations with the commercial steel thermodynamics data base. The advantages of the use of the meshless method are its simplicity and efficiency, since no polygonisation is involved, easy adaptation of the nodal points in areas with high gradients, almost the same formulation in two and three dimensions, high accuracy and low numerical diffusion.
Further on, the developed numerical model of the EAF will be used in simulation studies of different melting scenarios, which will verify its accuracy and usability. Finally, the obtained CFD model will be used to re-parameterize and re-validate the existent mathematical models.
2.3 Deliverables of the 2. workpackage:
Deliverables of the 2. workpackage are the following:
definition and numerical implementation of the CFD model, in order to design a basis for accurate development of the meshless EAF melting model,
verification of the model and sensitivity studies, in order to ensure proper operation and usability of the model for further studies,
performance of different simulation studies of the melting scenarios,
implementation of the developed CFD model to parameterize and validate existent chemical, mass- and heat-transfer models.
Upgrade of the designed models into smart sensors and development of the optimization framework to find the optimal scenarios and melting programs.
3.1 Description of the 3. workpackage:
Supplemented, parameterized and re-validated models from the previous workpackages of the project will be integrated into the frame of smart sensors and included in the optimization framework to obtain optimal melting scenarios and melting programs. Smart sensor is a common name for a software solution, which uses available process measurements and integrated algorithms to estimate or reconstruct the missing process data. Usually, the smart sensors are used to either estimate the process values, where its measurements cannot be performed or to detect the failures of the physical sensors. In the later case, smart sensor is used for comparison of the measured and estimated process values and is able to detect failures of the physical sensors. In the case of an EAF, the smart sensors will be used to estimate the process values, which are due to the nature of the process (high temperatures and electric currents) not measured, but represent one of the crucial information for optimal process control. Those are temperature of the steel and slag, chemical composition of the steel and slag and stage of melting. Using the developed mathematical models and integrating them together with the process measurements into smart sensors, satisfactory estimation of the process values can be made, leading to better insight and control of the EAF.
Besides the estimation of the process values, smart sensors represent the basis of the optimization-framework system and decision support system. Complex, nonlinear and time variant EAF processes represent the biggest issue in the development of the optimization methods. Due to the nonlinearity and time variability of the processes, predicting the behavior of the system in advance is challenging. For this reason, optimization algorithms require implementation of the advanced and more efficient methods, which are more robust when solving such problems; however, they are more time consuming. The issue can be resolved using a combination of offline process optimization and optimal melting scenario database. As the time complexity of the method is of second importance at offline optimization, more complex approaches can be used to search for the optimal melting scenarios, which usually yield better results. Furthermore, obtained optimal scenarios can be saved into a data base and later be used to search for the most appropriate melting program, including optimal energy feed, oxygen lancing, carbon injection, slag-foaming elements and other metallic additives. Comparing the current conditions in the EAF to the saved data, a similar EAF state can be obtained for which the optimal melting program is already known. Further on, online optimization is used to optimize a shorter time interval in advance, in order to eliminate the minor differences between the current state and the archived melting scenario. In this manner, real-time optimization of the EAF process is possible. However, it is not necessary that a particular simulation will always follow the same optimal scenario. In literature, a combination of online monitoring and process optimization is referred to as performance supervision.
To develop the described optimization environment, excellent knowledge of the steel-recycling process is needed, as the search for optimal melting scenarios requires to know the limitations of the process, actuators, physical phenomena and physical characteristics of the present materials.
3.2 Description of methods and knowledge in the 3. workpackage:
To successfully implement the 3. workpackage of the project, methods and knowledge of system theory, modelling and optimization will be implemented. All previously modified models will be integrated into the frame of smart sensors and used to develop an optimization environment. As already mentioned, the EAF consists of complex, nonlinear and time-variant processes. For this reason, optimization algorithm will integrate the most efficient, advanced methods, such as evolutionary and genetic algorithms and particle swarms. The advantage of these methods is their insensitivity to occurrence of the local optimization minimums. When using conventional optimization (simplex or iterative methods) on highly nonlinear problems with multiple (local) minimums, one of these local minima is often suggested as a final result. For this reason, heuristic-based methods will be used, which usually do not exhibit this kind of issues. When developing an optimization environment, special attention must be paid to limitations of the process, which represent one of the key factors of the solution usability. Limitations of actuators (transformer power, oxygen, carbon and additive rates, offgas flows etc.), physical phenomena (melting, solidification, chemical reactions etc.) and physical properties of the materials (compound stability, heats of fusion, transformations of the physical states, thresholds of chemical reactions etc.) define the margin between possible and impossible EAF conditions, which need to be considered, whether the obtained optimization results are realistic. The results of the optimization, i.e. optimal melting scenarios and melting programs will be saved in the data base, which will account for faster search of the best possible solution of the given EAF operation. Afterwards, online optimization with simpler, conventional methods, will eliminate potential differences between the current and the saved scenario in real time.
3.3 Deliverables of the 3. workpackage:
Deliverables of the 3. workpackage are the following:
upgrade of the modified mathematical models for the needs of smart sensors for online estimation of the process values,
development of the optimization environment, including offline process optimization, in order to obtain optimal melting scenarios and melting programs,
development of the optimal melting scenario data base, in order to increase the performance of the overall optimization,
implementation of online optimization in order to improve the final results and eliminate potential differences between the current and the saved scenario, in order to ensure optimal control of the EAF in real time.
Development of the decision support system and integration of the process models, smart sensors and optimization algorithms and their implementation into one software solution.
4.1 Description of the 4. workpackage:
The work performed in the first three workpackages, i.e. modification of the process models for the needs of smart-sensor development in 1st workpackage, parameterization and validation of the models using the CFD approach in 2nd workpackage and the integration of all previous into the frame of optimization environment in the 3rd workpackage, represents a basis for the highest software solution level, i.e. development of the decision support system. As has already been mentioned, the efficiency of the steel-recycling process is influenced by various factors. Lack of process measurements, operator’s experience and input material variability lead to substantial fluctuations in several production-efficiency indicators, which indicates underutilization of the production potential. Using systemic solutions based on modelling, optimization and decision support, the system can be upgraded and optimized, minimizing the impact of the known disturbances and leading to higher efficiency of the process. Today, most of the EAFs are operated according to the operator’s experience, melting program and indirect measurements of the actual EAF conditions (e.g. power-on time, energy consumption, arc stability etc.). The melting programs are based on several assumptions, such as: initial scrap composition is equal to all heats, the amount of molten steel is dependent on energy input, amount of hot heel is always equal etc., which usually leads to significant deviations from the (theoretically) optimal control. Due to the nature of the EAF process, amended initial conditions can lead to completely different melting progress (e.g. chemical reactions can occur at different times with different intensity), which the predefined melting programs do not consider. Whether the initial conditions in the EAF are close to the conditions assumed by the melting programs, the EAF operation can be highly efficient; however, practical experience shows different results. It would be reasonable to perform the melting program according to the actual conditions in the EAF, indicated by the data such as melting stage, bath composition and temperature; however, online measurement of these values (except the temperature) is up to now not possible.
The above issue can be resolved by implementing an advanced decision support system, available process measurements, estimated process values and optimization algorithms. Basic research of the proposed high-tech solution thus represents a technological leap from non-optimal to optimal EAF control. Investing further research, the proposed solution could be used in existent or new EAF assemblies with relatively low initial investment. The literature review reveals that EAF-process support by means of the software solutions exists; however, it is used only for monitoring of the melting processes. A combination of the real process and a decision support system based on computer simulation and process models cannot be found up to now. Implementation of the proposed system not only increases the consistency of the process, but also other production-efficiency indicators (higher steel yield, lower energy and material consumption, shorter tap-to-tap times, higher steel quality, lower environmental impact etc.) and consequently economical, ecologic and technologic aspects of the mill.
The proposed optimization environment, together with decision support algorithms, represents the biggest added value in the EAF control, as the decisions of the most appropriate actions in the process operation are made by a computer system and no longer by the operator. In this manner, operator’s experience are no longer important when making the crucial decisions, as those are made by a supporting system ensuring consistency and lower fluctuations in the key production indicators.
4.2 Description of methods and knowledge in the 4. workpackage:
To successfully implement the 4. workpackage of the project, methods and knowledge of decision support algorithms and simulation will be used. The systems developed in previous workpackages represent the basis for the decision support system (DSS). The DSS can be described as a computer-aided information system that supports operation decision-making activities. It is usually used when accepting crucial decisions at the processes which are unpredictable and fast changing. Each DSS consists of three key elements, i.e. 1) data base, 2) process models and 3) user interface. To develop the DSS two different data bases will be used. The first is the optimal melting programs data base, built in the 3. workpackage using offline optimization. The second is the knowledge data base, used by the DSS in a combination with current process-value estimations, input data, first database and online optimization, to determine the most appropriate action in each moment. To develop the DSS, decision-making methods, decision trees, fuzzy inference systems and regression analysis will be used. Proper combination of the previous and their integration assures efficient and useful DSS. Moreover, to present the crucial information, a user interface has to be developed, representing all the necessary process data, including a possible (optimal) EAF control in a clear and understandable way. To design the user interface, programming skills will be used. In the end, the developed system will be tested on simulation studies and measured operational data of the EAF.
4.3 Deliverables of the 4. workpackage:
Deliverables of the 4. workpackage are the following:
preparation of the knowledge data base, used in a combination with the optimal melting scenarios,
development of the decision support system and integrating all previously developed process models, smart sensors and optimization algorithms, in order to allow monitoring, optimization and optimal EAF control,
development of the user interface, which represents the crucial process data and suggested the most appropriate actions during the melting process to the user in a clear and understandable manner, in order to achieve optimal EAF operation.
Performance of comparative simulation studies between current (non-optimal) and optimized EAF operation.
5.1 Description of the 5. workpackage:
Successful realization of all previous workpackages leads to improved operation of the EAF on multiple levels (increased repeatability, higher steel yield, lower energy and material consumption, shorter production times etc.). To validate the effectiveness of the performed work, comparative studies between prior and current EAF operation will be performed. To determine the previous, non-optimal EAF operation, measured EAF data shall be used, which will clearly represent the efficiency of the EAF operation prior to improvements. The measurements include energy and material consumption, tap-to-tap times, product quality and steel yield. All these data indicates the efficiency of the productions, which this study tries to improve. To determine the EAF efficiency after the improvements, simulation studies will be performed with all developed systems (smart sensors, optimization and decision support system) and validated according to the same criteria as before. In this manner, clear difference between both operations (i.e. non-optimal and optimal) will be presented together with all crucial efficiency indicators.
5.2 Description of methods and knowledge in the 5. workpackage:
To successfully implement the 5. workpackage of the project, methods and knowledge of statistical data analysis will be used, in order to validate the efficiency of the performed work. In this manner, differences in both EAF operations will be presented, together with possibilities on further development of the presented approach.
5.3 Deliverables of the 5. workpackage:
Deliverables of the 5. workpackage are the following:
estimation of the current EAF-operation efficiency by means of energy and material consumption, steel quality, tap-to-tap times and steel yield,
estimation of the EAF-operation efficiency after improvements by means of energy and material consumption, steel quality, tap-to-tap times and steel yield,
comparative study of both EAF operations regarding all crucial process-efficiency indicators, including economic balances.
|Citations for bibliographic records||link on SICRIS|