Direct current arc furnace and control method

Abstract

The application discloses a direct current electric arc furnace and a control method thereof, and belongs to the technical field of the direct current electric arc furnace.The direct current electric arc furnace comprises a main frame, a furnace shell main body and an electrode lifting mechanism.The furnace shell main body is used for bearing molten materials, and an outer wall of the furnace shell main body is welded with a cooling water jacket for forming a circulating water channel, an acoustic wave guide pipe for conducting sound waves in the furnace and a furnace wall vibration base for mounting an acceleration sensor.The electrode lifting mechanism is slidably connected with the main frame through a guide structure, and the electrode lifting mechanism comprises an electrode holder for holding an electrode and a hydraulic lifting cylinder connected with the electrode holder.An electrode vibration base is arranged on the electrode holder.The application enables a controller to obtain more comprehensive information of the state in the furnace, and overcomes the limitation of traditional single electric parameter or temperature difference measurement.

Description

Technical Field

[0001] This invention relates to the field of DC submerged arc furnaces, specifically to a DC submerged arc furnace and its control method. Background Technology

[0002] In traditional electric arc furnace smelting processes, the lack of comprehensive real-time monitoring of the furnace interior makes it difficult for operators to accurately assess physical and chemical changes within the furnace. This leads to multiple problems, such as:

[0003] Incomplete and untimely information: Traditional methods mainly rely on indirect measurements, such as cooling water temperature difference and electrical parameters, to estimate the molten pool temperature. These methods cannot obtain the spatial distribution of the temperature field inside the molten pool, nor can they directly observe the location of critical events such as arc behavior or material collapse. This incompleteness of information limits the assessment of furnace conditions to a macroscopic level, making precise control difficult.

[0004] Difficulties in event identification and tracing: Existing vibration monitoring is usually single-point or localized, making it difficult to determine the specific location of mechanical events such as material collapse in the furnace. When an abnormality occurs in the furnace, operators can only determine whether the event exists, but cannot perform spatial positioning, which delays problem diagnosis and handling.

[0005] Lag and passivity in control: Traditional control strategies are mostly passive response-based, that is, they make significant adjustments only after adverse working conditions, such as large-scale material collapse, have already occurred. This may lead to drastic fluctuations in the smelting process and affect the stability of production.

[0006] These problems mainly stem from the limitations of data acquisition methods and data processing technologies. Traditional submerged arc furnaces have dispersed and single-function sensors, making it difficult to generate multi-dimensional, unified timestamp data streams. Furthermore, the lack of processing logic to fuse multi-source data and perform causal analysis makes it impossible to effectively distinguish and decouple electrical and mechanical events, and also prevents the prediction of future furnace conditions, resulting in a lack of foresight in control decisions.

[0007] The information disclosed in the background section above is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0008] The purpose of this invention is to provide a DC submerged arc furnace and its control method to solve the problems mentioned in the background art.

[0009] The technical solution of the present invention includes: a main frame, including a furnace shell body and an electrode lifting mechanism; the furnace shell body is used to carry molten material, and its outer wall is welded with a cooling water jacket forming a circulating water channel, an acoustic waveguide for transmitting sound waves inside the furnace, and a furnace wall vibration base for installing an acceleration sensor;

[0010] The electrode lifting mechanism is slidably connected to the main frame through a guide structure. The electrode lifting mechanism includes an electrode holder for holding the electrode and a hydraulic lifting cylinder connected to the electrode holder. The electrode holder is provided with an electrode vibration base.

[0011] The sensing system includes an acoustic sensor installed on the acoustic waveguide, an acceleration sensor installed on the furnace wall vibration base and the electrode vibration base, a high-frequency current sensor connected in series with the power supply busbar, and a temperature difference measuring unit for measuring the temperature difference between the inlet and outlet of the cooling water jacket.

[0012] The controller is electrically connected to the sensing system and is used to receive the data collected by the system.

[0013] Preferably, the furnace wall vibration base is evenly distributed along the circumferential and height directions on the outer wall of the furnace shell body to form a spatial monitoring array.

[0014] Preferably, a viewing window is provided on the side wall of the furnace shell body, and an infrared thermal imager is provided facing the viewing window.

[0015] Preferably, the lower end of the hydraulic lifting cylinder is fixed to a platform surrounding the furnace shell body via a trunnion, and the upper end of its piston rod is connected to the electrode holder via a ball joint.

[0016] Preferably, a flue gas duct is connected to the top of the furnace shell body, and the sensing system further includes a waste gas analyzer with a sampling port connected to the flue gas duct.

[0017] A control method for a DC submerged arc furnace includes the following steps:

[0018] The controller synchronously collects data from the sensing system to construct a multimodal sensing data stream, wherein the arc current harmonic data from the high-frequency current sensor is defined as electrical characteristic data, and the signals from the acoustic sensor and the acceleration sensor are defined as acoustic vibration characteristic data.

[0019] Based on the synchronous fluctuations of the multimodal sensing data stream, the controller compares the pacemaker timing of the electrical characteristic data and the acoustic vibration characteristic data to determine whether the event source is an electrical or mechanical event. Based on the determination result, the controller decouples the state of the multimodal sensing data stream and infers the internal state variables of the smelting process.

[0020] Based on the internal state variables, the controller uses a digital twin to predict the risk of adverse operating conditions occurring within a future time window, and simulates and selects preventive control strategies for the risks in order to generate optimal control commands.

[0021] The controller outputs the optimal control command to the electrode lifting mechanism (200) and the power supply system to adjust the electrode position and power supply parameters.

[0022] Preferably, when the source of the event is determined to be a mechanical event, the controller further compares the time difference of the signal arriving at the acceleration sensor at different physical locations to deduce the origin region of the mechanical event.

[0023] Preferably, the step of inferring the internal state variables of the smelting process includes: for identified electrical property events, analyzing the electrical characteristic data and acoustic signals of a specific frequency band to infer the concealment degree of the electric arc; for identified mechanical property events, combining their source area, vibration intensity, and subsequent changes in exhaust gas composition to infer the scale and material composition of the collapse.

[0024] Preferably, the optimal control command is a smooth and continuous operation sequence used to drive the electrode lifting mechanism to lift at a non-linear rate and simultaneously perform pulse-type power regulation on the power supply system.

[0025] Preferably, the method further includes the step of:

[0026] The predicted risks from the digital twin are compared with subsequent measured data to obtain the prediction deviation;

[0027] Based on the prediction deviation, the prediction deviation is set as a penalty signal, and the model parameters in the digital twin are adjusted in reverse to achieve self-correction of the model.

[0028] This invention provides an improved DC submerged arc furnace and its control method, which, compared with the prior art, have the following improvements and advantages:

[0029] 1. The sensing system of the present invention includes an acoustic sensor, an acceleration sensor, a high-frequency current sensor, a temperature difference measurement unit, an infrared thermal imager, and an exhaust gas analyzer. These sensors collect a variety of data. For example, the infrared thermal imager can provide a two-dimensional temperature field image of the molten pool surface, while the exhaust gas analyzer can monitor the flue gas composition in real time. This enables the controller to obtain more comprehensive furnace status information, which surpasses the limitations of traditional single electrical parameter or temperature difference measurement.

[0030] 2. The furnace wall vibration base of the present invention is distributed in a spatial array on the outer wall of the furnace shell. When mechanical events such as material collapse occur in the furnace, the generated vibration waves will reach different acceleration sensors with a small time difference. By analyzing these time differences, the controller can reversely calculate the three-dimensional source area of ​​the event in the furnace. This design elevates vibration monitoring from a simple judgment of whether an event exists to a spatial source tracing level, providing key spatial dimension information for accurate diagnosis of furnace working conditions.

[0031] 3. The controller of this invention can synchronously acquire multimodal data streams and determine whether the initial disturbance originates from an electrical or mechanical event by comparing the pacemaker timing of electrical characteristic data, such as arc current harmonics and acoustic vibration characteristic data signal waveforms. This decoupling capability is an advancement of existing technology because it avoids confusing all fluctuations into a single cause, making subsequent state inference more targeted.

[0032] 4. The controller of this invention utilizes a high-fidelity digital twin model, which can predict the risk of adverse operating conditions occurring within a future time window based on the currently inferred internal state. For example, when the model predicts that a large-scale material collapse may occur, the controller can generate a smooth and continuous operating sequence before the event occurs by simulating and selecting preventive control strategies. This perturbation-style active intervention aims to guide the material collapse in a gentle manner, thereby avoiding the occurrence of catastrophic material collapse and ensuring the stability of the smelting process. Attached Figure Description

[0033] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0034] Figure 1 This is a schematic diagram of the overall structure of the device;

[0035] Figure 2 This is a schematic diagram of the connection structure of the main body of the furnace shell;

[0036] Figure 3 This is a schematic diagram of the cooling water jacket structure;

[0037] Figure 4 This is a schematic diagram of the method structure of the present invention;

[0038] In the diagram: 100, furnace shell body; 110, cooling water jacket; 120, viewing window; 130, acoustic waveguide; 140, furnace wall vibration base; 150, flue gas duct; 200, electrode lifting mechanism; 210, hydraulic lifting cylinder; 220, electrode holder; 230, electrode vibration base; 300, sensing system; 310, high-frequency current sensor; 320, acoustic sensor; 330, acceleration sensor; 340, infrared thermal imager; 350, temperature difference measurement unit; 360, exhaust gas analyzer. Detailed Implementation

[0039] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0040] Example 1

[0041] Please see Figure 1-3 The present invention provides a DC submerged arc furnace, comprising:

[0042] The main frame includes a furnace shell body 100 and an electrode lifting mechanism 200; the furnace shell body 100 is used to carry molten materials, and its outer wall is welded with a cooling water jacket 110 forming a circulating water channel, an acoustic waveguide 130 for transmitting sound waves inside the furnace, and a furnace wall vibration base 140 for installing an acceleration sensor 330.

[0043] The electrode lifting mechanism 200 is slidably connected to the main frame through a guide structure. The electrode lifting mechanism 200 includes an electrode holder 220 for holding the electrode and a hydraulic lifting cylinder 210 connected to the electrode holder 220. The electrode holder 220 is provided with an electrode vibration base 230.

[0044] The sensing system 300 includes an acoustic sensor 320 mounted on the acoustic waveguide 130, an acceleration sensor 330 mounted on the furnace wall vibration base 140 and the electrode vibration base 230, a high-frequency current sensor 310 connected in series with the power supply busbar, and a temperature difference measuring unit 350 for measuring the temperature difference between the inlet and outlet of the cooling water jacket 110.

[0045] The controller, electrically connected to the sensing system 300, is used to receive the data collected by the system.

[0046] In this embodiment, a DC submerged arc furnace has a main frame that provides basic mechanical support and execution functions for the entire device. The furnace shell 100 not only serves as a container for the smelting reaction but also functions as a physical mounting base for various sensors. The electrode lifting mechanism 200 is responsible for performing the key action of power regulation. The sensing system 300 collects multi-dimensional physical signals during the smelting process. The controller, which can be a Siemens SIMATIC S7-1500 series PLC, is used to uniformly receive and process data from the sensing system 300. This structural design, by integrating the sensing function with the furnace structure, provides the necessary and high-quality data input foundation for subsequent intelligent control methods, aiming to overcome the shortcomings of traditional submerged arc furnaces that lack sufficient real-time internal status information for precise control.

[0047] The furnace wall vibration base 140 is evenly distributed along the circumferential and height directions on the outer wall of the furnace shell body 100 to form a spatial monitoring array.

[0048] In this embodiment, the furnace wall vibration base 140 is arranged in a spatial array on the outer wall of the furnace shell 100, which is a functional design. In the prior art, vibration monitoring is usually carried out at a single point or locally, making it difficult to accurately determine the location of mechanical events such as material collapse in the furnace. However, in this embodiment, the furnace wall vibration base 140 is uniformly welded in the circumference and height of the middle and lower part of the furnace body. For example, three or more points are selected on planes at different heights and distributed at 120 degrees or 90 degrees to form a three-dimensional monitoring network. The purpose is that when events such as material collapse occur in the furnace, the generated vibration waves will propagate to the acceleration sensors 330 on each base along different paths. By analyzing the small time and amplitude differences of the vibration waves reaching different sensors, the controller can reversely calculate the three-dimensional spatial source point of the event in the furnace. This design elevates vibration monitoring from a simple determination of the presence or absence of an event to the level of event tracing, providing spatial dimension information for accurate diagnosis of furnace operating conditions.

[0049] The calculation logic is as follows: The controller first accurately records the absolute timestamps of the vibration wave signal arriving at at least three non-collinear accelerometers 330. The time difference between the signals received by any two sensors defines a hyperboloid in space, and the source of the event must lie on this hyperboloid. When three sensors are used, two independent time differences can be formed, thus defining two different hyperboloids. Theoretically, the intersection of these two hyperboloids is the possible location of the event source. When a fourth or more sensors are introduced to form a spatial array, more hyperboloids will be generated. The intersection points or the densest intersection areas of these hyperboloids in space are determined by the controller as the three-dimensional source region of this mechanical event. This process does not depend on the exact time of the event, but only on the time difference, thus exhibiting high robustness.

[0050] The furnace shell body 100 is also provided with a viewing window 120 on its side wall, and an infrared thermal imager 340 is provided facing the viewing window 120.

[0051] In this embodiment, the viewing window 120 provides a channel for non-contact optical observation of the furnace interior. Specifically, the viewing window 120 is a flange structure inlaid with high-temperature resistant sapphire glass and equipped with a high-pressure nitrogen purging pipeline to maintain mirror cleanliness. This allows an external infrared thermal imager 340, such as the FLIRA700 series industrial thermal imager, to penetrate the furnace wall and directly observe the temperature distribution on the surface of the molten pool without obstruction. In traditional smelting processes, the molten pool temperature is mainly estimated indirectly based on the cooling water temperature difference and electrical parameters, which cannot reflect the uniformity of its spatial distribution. The two-dimensional temperature field image acquired by the infrared thermal imager 340 can intuitively reveal whether there are overheated or undercooled areas inside the molten pool. This information can be used to correct and verify the average temperature calculated based on the overall thermal balance, providing intuitive visual data support for the controller to more accurately assess the energy absorption efficiency of the molten pool and determine the smelting state.

[0052] The calculation logic for this correction and verification is as follows: First, the controller calculates the theoretical average temperature of the molten pool based on the law of conservation of energy by subtracting the furnace wall heat loss calculated from the cooling water temperature difference, the heat carried away by the flue gas, and other known losses from the total input electrical power. Simultaneously, the controller processes the image acquired by the infrared thermal imager to calculate the average radiation temperature of the entire molten pool surface. The controller then compares this measured average temperature with the aforementioned theoretical average temperature. If a persistent and significant deviation exists between the two, it indicates that some assumptions of the model, such as the heat absorption efficiency of the furnace charge or unmodeled heat losses, do not match the actual situation. Based on this deviation, the controller corrects the thermal balance model in the digital twin, thereby making the model's assessment of the overall energy state of the molten pool more accurate.

[0053] The lower end of the cylinder body of the hydraulic lifting cylinder 210 is fixed to a platform surrounding the furnace shell body 100 by a trunnion, and the upper end of its piston rod is connected to the electrode holder 220 by a ball joint.

[0054] The hydraulic lifting cylinder 210 in this embodiment is designed to improve the mechanical reliability of the system. As the power source for the electrode lifting mechanism 200, the hydraulic lifting cylinder 210 can be a standard heavy-duty hydraulic cylinder. The lower end of the cylinder body is connected to the platform via a trunnion, and the upper end of the piston rod is connected to the electrode holder 220 via a ball joint. This connection method decouples complex mechanical stresses. The trunnion allows the hydraulic cylinder body to swing within a small range to accommodate unavoidable alignment errors during installation and structural displacements caused by thermal expansion and contraction during equipment operation. Simultaneously, the ball joint at the piston rod end allows the electrode holder 220 to generate slight angular deviations during movement without transmitting bending moment to the piston rod. This flexible connection design avoids problems such as hydraulic cylinder jamming, abnormal seal wear, or piston rod bending that can occur with rigid connections, ensuring long-term stability and precision of the electrode lifting action.

[0055] The top of the furnace shell body 100 is connected to a flue gas duct 150, and the sensing system 300 also includes a waste gas analyzer 360 with a sampling port connected to the flue gas duct 150.

[0056] In this embodiment, the exhaust gas analyzer 360 adds a chemical composition analysis dimension to the sensing system 300; the flue gas duct 150 serves as a channel connecting the furnace top and subsequent processing equipment, venting the exhaust gas generated during the smelting process; the exhaust gas analyzer 360 can specifically be a Thermo Fisher PrimaPRO online mass spectrometer, whose sampling port is directly connected to the flue gas duct 150 to achieve continuous real-time monitoring of exhaust gas components; its purpose is to invert the reaction process inside the furnace through chemical signals; for example, when a large-scale raw material collapse occurs inside the furnace, the carbonaceous reducing agent will react violently with the high-temperature melt or furnace gas, causing the concentration of carbon monoxide or carbon dioxide in the flue gas to show a characteristic peak in a short period of time; this sudden change in chemical composition captured by the exhaust gas analyzer 360 provides the controller with a verification basis independent of acoustic and vibration signals for judging the occurrence, scale and material properties of the collapse event, improving the accuracy of state inference.

[0057] Example 2

[0058] Please see Figure 4 A control method for a DC submerged arc furnace includes the following steps:

[0059] The controller synchronously collects data from the sensing system 300 to construct a multimodal sensing data stream. Among them, the arc current harmonic data from the high-frequency current sensor 310 is defined as electrical characteristic data, and the signals from the acoustic sensor 320 and the acceleration sensor 330 are defined as acoustic vibration characteristic data.

[0060] Based on the synchronous fluctuations of the multimodal sensor data stream, the controller compares the pacemaker timing of electrical characteristic data and acoustic vibration characteristic data to determine whether the event source is an electrical or mechanical event. Based on the determination result, the controller decouples the state of the multimodal sensor data stream and infers the internal state variables of the smelting process.

[0061] The controller uses a digital twin based on internal state variables to predict the risk of adverse operating conditions occurring within a future time window, and simulates and selects preventive control strategies to generate optimal control commands.

[0062] The controller outputs the optimal control command to the electrode lifting mechanism 200 and the power supply system to adjust the electrode position and power supply parameters.

[0063] In this embodiment, the control method is executed by a controller. The controller synchronously collects data from all sensors in the sensing system 300, including a high-frequency current sensor 310 (e.g., a PEMCWT series Rogowski coil), an acoustic sensor 320 (e.g., a Kistler 8152C series), an accelerometer 330 (e.g., a PCB600 series), an infrared thermal imager 340, a temperature difference measurement unit 350, a PT100 platinum resistance thermometer, and a waste gas analyzer 360. This data forms a multimodal data stream with a unified timestamp. The controller utilizes the physical differences in signal propagation speeds to perform causal analysis of the event source. Since electrical signal propagation is nearly instantaneous, while sound waves and vibration waves have millisecond-level delays in the furnace medium, when the controller... When the controller detects a sudden and violent fluctuation in the data stream, it compares the starting trigger points of the electrical characteristic data and acoustic vibration characteristic data signal waveforms, i.e., the pacemaker timing. This timing is determined by real-time edge detection or threshold triggering of the raw signals from the high-frequency current sensor 310, acoustic sensor 320, and acceleration sensor 330. The order of the pacemakers determines whether the initial disturbance originates from electrical events such as arc instability or mechanical events such as material collapse. Based on this determination, the controller decouples the highly coupled signals and infers the deep state variables inside the furnace. The controller uses a high-fidelity digital twin model to predict the probability of adverse operating conditions that may occur in the next few minutes based on the inferred current internal state.

[0064] This model aims to create a high-fidelity virtual model that is synchronized in real time with the physical submerged arc furnace to simulate the dynamic changes in the smelting process and evaluate the effectiveness of different control strategies. Logically, the model contains multiple sub-models. For example, the heat balance model receives inputs such as electrical power and cooling water temperature difference to evaluate the overall energy state of the molten pool; the material collapse dynamics model receives internal state variables as inputs to predict changes in material structure and future collapse risks.

[0065] For predicted high-risk events, the controller simulates various perturbation control operations in a digital twin environment and selects the preventive strategy that can effectively reduce risks from them, generating the optimal control command. The controller sends this command sequence to the hydraulic system and power supply system of the electrode lifting mechanism 200 to perform precise preventive regulation.

[0066] When the source of the event is determined to be a mechanical event, the controller further compares the time difference of the signal arriving at the acceleration sensor 330 at different physical locations to deduce the origin region of the mechanical event.

[0067] In this embodiment, after determining that the source of the event is mechanical, the control method initiates spatial tracing calculation. The controller calls the data collected by the array of accelerometers 330 arranged on the furnace wall vibration base 140. The vibration waves generated by these sensors reach each sensor at different distances, resulting in a time difference that can be accurately measured at the nanosecond or millisecond level. The controller uses a built-in positioning algorithm to calculate the approximate three-dimensional location of the mechanical event inside the furnace based on the order and time difference of the signals received by each sensor, combined with the known spatial coordinates of the sensors and the propagation speed model of the vibration waves in the furnace shell structure.

[0068] The input to this calculation process is the absolute timestamps of vibration wave signals from at least three non-collinear accelerometers 330, as well as the known spatial coordinates of these sensors; the controller calculates the time difference between the signals received by any two sensors, which defines a hyperboloid in space on which the source of the event must be located; by solving for the intersection points or the densest intersection region of multiple hyperboloids, the three-dimensional source region of the event can be determined.

[0069] This step expands the monitoring of mechanical events from a one-dimensional time dimension to a three-dimensional spatial dimension, providing crucial spatial information for subsequent assessment of the local impact of material collapse on the furnace wall or for determining the shape of the material surface inside the furnace.

[0070] The steps for inferring the internal state variables of the smelting process include: for identified electrical events, analyzing electrical characteristic data and acoustic signals in specific frequency bands to infer the concealment of the electric arc; for identified mechanical events, combining their source area, vibration intensity, and subsequent changes in exhaust gas composition to infer the scale and composition of the material collapse.

[0071] The control method in this embodiment employs multi-information fusion logic when inferring internal state variables; after an event is decoupled and identified as an electrical event, the controller performs in-depth analysis of arc current harmonic data and acoustic signals.

[0072] By matching these signal characteristics with a pre-stored pattern library, the degree to which the electric arc is covered by the furnace charge can be inferred, i.e., the degree of concealment of the electric arc. When an event is identified as a mechanical event, the controller will integrate multiple information sources for judgment: using the accelerometer 330 array to determine its source area, then inferring its impact intensity based on the amplitude and energy of the vibration signal, and combining this with the subsequent changes in exhaust gas composition detected by the exhaust gas analyzer 360, to comprehensively assess the scale of the collapse and the approximate composition of the materials it contains;

[0073] For example, the acoustic spectrum and current harmonic characteristics of a stable, brightly burning arc and a concealed arc covered by material differ significantly. By matching the current signal characteristics with patterns in the historical database, the controller can infer the degree to which the arc is covered by the furnace charge, i.e., the degree of concealment of the arc. When an event is identified as a mechanical event, the controller integrates multiple information sources for judgment: first, it uses the accelerometer 330 array to determine its source area; then, it infers its impact intensity based on the amplitude and energy of the vibration signal; and simultaneously, it combines the subsequent changes in exhaust gas components detected by the exhaust gas analyzer 360, such as the amplitude and duration of the CO peak, to comprehensively assess the scale of the collapse and the approximate composition of the materials it contains, such as whether it is carbonaceous materials like coke, flux, or ore. This inference method transcends the limitations of a single sensor and achieves a comprehensive characterization of the multidimensional attributes of the same event.

[0074] The comprehensive evaluation logic for the scale of material collapse is as follows: The controller first uses the integral of the signal from the accelerometer 330, i.e., the vibration energy, as the initial baseline value for the scale of material collapse. This baseline value is then weighted and corrected using information from other sensors. For example, if the exhaust gas analyzer 360 detects a significant and persistent peak in CO concentration after the vibration event, the controller assumes that the collapse contains a large amount of newly entered carbonaceous material in the reaction zone, thus increasing the weight of the collapse scale evaluation. Conversely, if the gas composition does not change significantly, the weight will be appropriately reduced. Simultaneously, if the infrared image shows a significant and widespread drop in the surface temperature of the collapsed area after the event, it also corroborates that this is a large-scale cold material collapse, and the controller will further increase the confidence level of the scale evaluation. Through this fusion and cross-validation of multi-source information, a more reliable comprehensive index of the collapse scale is ultimately obtained than that from a single signal source.

[0075] The optimal control command is a smooth and continuous operation sequence, used to drive the electrode lifting mechanism 200 to lift at a non-linear rate, and simultaneously perform pulse power regulation on the power supply system.

[0076] In this embodiment, the optimal control command output by the control method is not a simple step signal, but a set of dynamically planned, time-coupled operation sequences.

[0077] Its optimality is reflected in its ability to simultaneously meet multiple objectives such as production stability, energy consumption optimization, and safety risk minimization. The instruction is that the controller, in the digital twin model, evaluates the impact of each strategy on future risks after simulating and selecting multiple preventive control strategies, and selects the strategy sequence with the highest comprehensive score as the output.

[0078] The aim is to achieve proactive, perturbation-based intervention in the smelting process, mitigating potential risks in a gentle manner. For example, when the digital twin predicts that a large-scale material collapse may occur in a certain area of ​​the furnace within the next one or two minutes due to material crusting or bridging, the controller may not generate instructions to significantly lift the electrode after the collapse, but rather to execute a composite instruction sequence before the collapse, lifting the electrode by 10 mm at a non-linear rate that starts slow and gradually increases over the next 60 seconds, while simultaneously momentarily reducing the power by 5% at the 20th and 40th seconds of the lifting process before restoring it. This smooth and composite operation aims to proactively disrupt the forming material bridge structure through small-amplitude mechanical and electrothermal disturbances, guiding it to collapse in a small-scale, harmless manner, thereby avoiding catastrophic large-scale material collapse and ensuring the stability of the smelting process.

[0079] The method also includes the following steps:

[0080] The predicted risks from the digital twin are compared with subsequent measured data to obtain the prediction bias;

[0081] Based on the prediction bias, the prediction bias is set as a penalty signal to adjust the model parameters in the digital twin in reverse, thereby achieving self-correction of the model.

[0082] The control method in this embodiment includes a model self-correction and evolution mechanism. In each control cycle, the controller continuously compares the events predicted by its digital twin, such as a prediction of a material collapse on the northeast side in 90 seconds, with subsequent actual sensor data. If the actual situation matches the prediction—for example, if the accelerometer 330 does indeed capture the corresponding vibration signal near the predicted time and area—the model is considered accurate. If it does not match or the deviation is large—for example, if the actual collapse is much larger than predicted—the controller quantifies this prediction deviation as a penalty signal. This penalty signal is input into an online reinforcement learning algorithm. This algorithm, through a gradient descent-based optimization process, uses the penalty signal as input to inversely adjust the core physical model parameters within the digital twin, thereby achieving model self-correction and evolution. I modified the algorithm to make its internal model increasingly closer to the real physical process, thereby achieving adaptation to changes in furnace conditions and maintaining the accuracy of its long-term predictions. The goal of this algorithm is to minimize future penalty signals. To achieve this goal, the algorithm will fine-tune the core physical model parameters inside the digital twin, such as adjusting key coefficients in the model related to the sound wave propagation speed of semi-molten materials or the melting endothermic rate of specific raw materials, rather than just the weights of data fusion. In this way, the digital twin can continuously learn from its own mistakes, making its internal model increasingly closer to the real physical process, thereby achieving adaptation to changes in furnace conditions, such as fluctuations in raw material composition and the increase in furnace age, and maintaining the accuracy of its long-term predictions.

[0083] The computational logic for model parameter self-correction can be summarized as an optimization process based on gradient descent, the core idea of ​​which is embodied in the following expression:

[0084] ;

[0085] in:

[0086] This represents the revised parameter values ​​of the new model, such as the melting rate of the material.

[0087] This represents the old value of the parameter before the correction;

[0088] The Greek letter eta represents the learning rate, which is a preset hyperparameter used to control the step size of each correction to ensure the stability of the convergence process.

[0089] This represents the prediction bias, also known as the aforementioned penalty signal, which is the difference between the model's predicted value and the subsequent measured value.

[0090] Represents prediction bias For old parameters The gradient, or sensitivity, indicates the direction of parameter adjustment. It is calculated by observing the impact of micro-perturbation parameters on the prediction deviation within the digital twin. The significance of this gradient is that it tells the controller which direction to adjust the parameters to most effectively reduce future prediction deviations. The entire process is automatically completed online by the controller, realizing the autonomous evolution of the model.

[0091] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A DC submerged arc furnace, characterized in that, include: The main frame includes a furnace shell body (100) and an electrode lifting mechanism (200); the furnace shell body (100) is used to carry molten material, and its outer wall is welded with a cooling water jacket (110) forming a circulating water channel, an acoustic waveguide (130) for transmitting sound waves inside the furnace, and a furnace wall vibration base (140) for installing an acceleration sensor (330). The electrode lifting mechanism (200) is slidably connected to the main frame through a guide structure. The electrode lifting mechanism (200) includes an electrode holder (220) for holding the electrode and a hydraulic lifting cylinder (210) connected to the electrode holder (220). The electrode holder (220) is provided with an electrode vibration base (230). The sensing system (300) includes an acoustic sensor (320) mounted on the acoustic waveguide (130), an acceleration sensor (330) mounted on the furnace wall vibration base (140) and the electrode vibration base (230), a high-frequency current sensor (310) connected in series with the power supply busbar, and a temperature difference measuring unit (350) for measuring the temperature difference between the inlet and outlet of the cooling water jacket (110). The controller is electrically connected to the sensing system (300) and is used to receive the data collected by the system.

2. The DC submerged arc furnace according to claim 1, characterized in that, The furnace wall vibration base (140) is evenly distributed along the circumferential and height directions on the outer wall of the furnace shell body (100) to form a spatial monitoring array.

3. The DC-DC submerged arc furnace according to claim 1, characterized in that, The furnace shell body (100) is also provided with a viewing window (120) on its side wall, and an infrared thermal imager (340) is provided facing the viewing window (120).

4. A DC submerged arc furnace according to claim 1, characterized in that, The lower end of the cylinder body of the hydraulic lifting cylinder (210) is fixed to a platform surrounding the furnace shell body (100) by a trunnion, and the upper end of its piston rod is connected to the electrode holder (220) by a ball joint.

5. A DC submerged arc furnace according to claim 1, characterized in that, The top of the furnace shell body (100) is connected to a flue gas duct (150), and the sensing system (300) also includes a waste gas analyzer (360) with a sampling port connected to the flue gas duct (150).

6. A control method for a DC submerged arc furnace, applied to a DC submerged arc furnace as described in any one of claims 1 to 5, characterized in that, Includes the following steps: The controller synchronously collects data from the sensing system (300) to construct a multimodal sensing data stream, wherein the arc current harmonic data from the high-frequency current sensor (310) is defined as electrical characteristic data, and the signals from the acoustic sensor (320) and the acceleration sensor (330) are defined as acoustic vibration characteristic data. Based on the synchronous fluctuations of the multimodal sensing data stream, the controller compares the pacemaker timing of the electrical characteristic data and the acoustic vibration characteristic data to determine whether the event source is an electrical or mechanical event. Based on the determination result, the controller decouples the state of the multimodal sensing data stream and infers the internal state variables of the smelting process. Based on the internal state variables, the controller uses a digital twin to predict the risk of adverse operating conditions occurring within a future time window, and simulates and selects preventive control strategies for the risks in order to generate optimal control commands. The controller outputs the optimal control command to the electrode lifting mechanism (200) and the power supply system to adjust the electrode position and power supply parameters.

7. The control method for a DC submerged arc furnace according to claim 6, characterized in that, When the source of the event is determined to be a mechanical event, the controller further compares the time difference of the signal arriving at the acceleration sensor (330) at different physical locations to deduce the origin region of the mechanical event.

8. The control method for a DC submerged arc furnace according to claim 6, characterized in that, The steps for inferring the internal state variables of the smelting process include: for identified electrical events, analyzing the electrical characteristic data and acoustic signals in a specific frequency band to infer the concealment degree of the electric arc; for identified mechanical events, combining their source area, vibration intensity, and subsequent changes in exhaust gas composition to infer the scale and material composition of the collapse.

9. The control method for a DC submerged arc furnace according to claim 6, characterized in that, The optimal control command is a smooth and continuous operation sequence used to drive the electrode lifting mechanism (200) to rise and fall at a non-linear rate, and simultaneously perform pulse power regulation on the power supply system.

10. The control method for a DC submerged arc furnace according to claim 6, characterized in that, The method also includes the following steps: The predicted risks from the digital twin are compared with subsequent measured data to obtain the prediction deviation; Based on the prediction deviation, the prediction deviation is set as a penalty signal, and the model parameters in the digital twin are adjusted in reverse to achieve self-correction of the model.

Images