Continuous feed electric arc furnace system

Abstract

A method and system are provided for determining the real-time feed rate (unit mass / unit time) of scrap metal material being fed into an electric arc furnace (EAF), the method and system identifying a real-time velocity of the scrap metal material and a volume of the scrap metal material. The method and system also classify the scrap metal material using a machine learning model based on a digital image of the scrap metal material and assign a density to the scrap metal material. The feed rate is determined based on the velocity and volume of the scrap metal material and the assigned density.

Description

[Technical field]

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS This application claims priority under 35 USC § 119 from U.S. patent application Ser. No. 63 / 318,518, filed March 10, 2022, the disclosure of which is expressly incorporated herein by reference in its entirety.

[0002] Inventive embodiments of the present disclosure relate generally to methods and systems for determining (e.g., in real time) the feed rate (unit mass / unit time) of scrap metal material being input to an EAF by a conveyor (or other transport device), as well as related methods and systems for operating the EAF system based in part on the determined feed rate of the scrap metal material. [Background technology]

[0003] The traditional method of steel production is based on the blast furnace (BOF) route, where coke is used to reduce and melt the iron ore in a blast furnace, followed by smelting in a BOF converter. The coke is burned to generate energy and the iron ore is reduced by reacting with carbon and carbon monoxide. In contrast, in the last few years, the electric arc furnace (EAF) steelmaking process has been gaining increasing interest worldwide over BOF technology.

[0004] The benefits of the EAF process are numerous, but one of the leading arguments for EAF, especially next to lower emissions, is flexibility. The process is much more flexible than the BOF route, which cannot be turned on and off at will depending on current market conditions. The trend towards more advanced, more flexible and cleaner processes calls for technologies that can be rapidly implemented and adapted to each steelmaker's individual needs.

[0005] Additionally, a unique steelmaking process, Consteel® technology, has been developed that continuously preheats and feeds metallic inputs (scrap metal, pig iron, hot iron briquettes, etc.) to an EAF while simultaneously controlling gaseous emissions. In particular, Consteel® technology uses a conveyor to feed raw materials (e.g., scrap metal) to the EAF. The input material is loaded onto the input conveyor from a scrap yard or rail car and is preheated by an off-gas process as it is continuously fed to the EAF, where it is melted by immersion in molten steel. In this regard, the EAF operates at constant flat-bed conditions, which is a key advantage over traditional batch processes.

[0006] Although the Consteel® system has advantages regarding the environment and energy consumption, it also has some drawbacks that prevent it from becoming more widely accepted. The main challenge, apart from the installation cost, is the estimation of the tapping weight. The Consteel® system typically relies on load cells located in the EAF base to estimate the EAF weight. These sensors are difficult to maintain and give different readings every time the EAF is tilted.

[0007] Thus, there remains a need in the art of continuous feed EAFs for a non-contact system that determines the scrap metal material entering the EAF (e.g., in real time) without requiring the use of one or more of load cells, crane feed data regarding the deposition of the scrap metal material on a conveyor, and / or operator input. Summary of the Invention [Problem to be solved by the invention]

[0008] The present invention has been made to solve the problems in the conventional techniques described above. [Means for solving the problem]

[0009] Some embodiments according to the present invention provide a method for determining (e.g., in real time) a scrap feed rate of an electric arc furnace (EAF) having a conveyor that delivers scrap metal material to the EAF. The method includes the steps of: (i) acquiring a first plurality of red-green-blue (RGB) digital images of a first portion of the scrap metal material being transported by the conveyor; (ii) determining, in a computer processor, a first velocity of the first portion of the scrap metal material by comparing a first frame and a second frame of the first plurality of RGB digital images of the first portion of the scrap metal material being transported by the conveyor; (iii) generating, in a computer processor, a first stereo camera depth image of the first portion of the scrap metal material, the computer processor correlating one or more pixels of the first depth image with 3D coordinates of the first portion of the scrap metal material; and (iv) generating, in a computer processor, a first depth image of the first portion of the scrap metal material by comparing a first frame and a second frame of the first plurality of RGB digital images of the first portion of the scrap metal material by comparing a first frame and a second frame of the first frame of the ... The method may include (a) determining, on the computer processor, a first volume of the first portion of the scrap metal material based at least in part on the 3D coordinates of the first portion of the scrap metal material; (b) determining, on the computer processor, a first classification of the first portion of the scrap metal material from among a group of known scrap metal material based at least in part on the first RGB image using a machine learning model and assigning a first density to the first portion of the scrap metal material; and (c) determining a scrap feed rate associated with the first portion of the scrap metal material being transported on the conveyor based on (a) the first volume of the first portion of the scrap metal material, (b) the assigned first density, and (c) the first velocity of the first portion of the scrap metal material.

[0010] In another aspect, some embodiments according to the present invention provide a method of operating an EAF. The method may include determining, in real time, a real-time scrap feed rate of scrap metal material being fed to the EAF by a conveyor, as described and disclosed herein. The method may also include comparing, in a computer processor, the real-time scrap feed rate to an operating condition of the EAF. Additionally, the method may include modifying (e.g., in real time) a conveyor speed of a conveyor feeding scrap metal material to the EAF, modifying (e.g., in real time) a current flow rate to electrodes in the EAF, or both.

[0011] In another aspect, some embodiments according to the present invention provide a system for determining a scrap feed rate of an EAF having a conveyor that introduces scrap metal material into the EAF. The system may include a sensor fusion module, which includes two monochrome cameras configured to provide stereoscopic vision, an RGB camera, an infrared spectrum projector, and a first computer processor in operative communication with the two monochrome cameras, the RGB camera, and the infrared spectrum projector. The first computer processor may be configured, for example, to: (a) determine a speed of the scrap metal material in real time based on a plurality of digital images acquired by the RGB cameras; (b) generate a stereoscopic depth image in real time and correlate one or more pixels of the depth image with 3D coordinates of the scrap metal material in real time; and (c) determine a volume of the scrap metal material in real time based at least in part on the 3D coordinates of the scrap metal material. The system may also include a scrap classification module in operative communication with the sensor fusion module, which includes a machine learning model. The machine learning model may be configured, for example, to: (a) identify, in real time, one or more distinct classes or types of scrap metal material from a group of known scrap metal material (e.g., known classes or types of scrap metal material) based on at least a portion of the first RGB image received in real time from the sensor fusion unit; (b) assign a density to the scrap metal material in real time; and (c) in the second computer processor, determine, in real time, a scrap feed rate for the scrap metal material being transported on the conveyor based on the volume of the scrap metal material, the assigned first density, and the speed of the scrap metal material.

[0012] In yet another aspect, some embodiments according to the invention provide an EAF system including an EAF including a furnace body defining an interior portion and a set of electrodes extending into the interior portion. The system may also include a conveyor, the conveyor including a first portion distal to the EAF for receiving scrap metal material (e.g., a new supply for the EAF, which may be loaded onto the conveyor from a rail car or a crane), a second portion proximal to the EAF for inputting the scrap metal material into the interior portion of the EAF, and a third portion located between the first portion and the second portion, the conveyor transporting the scrap metal material from the first portion through the third portion and the second portion to the interior portion of the EAF. The system may further include a sensor fusion module mounted above a section of the third portion of the conveyor. The sensor fusion module may include two monochrome cameras configured to provide stereoscopic vision, an RGB camera, an infrared spectrum projector, and a first computer processor in operative communication with the two monochrome cameras, the RGB camera, and the infrared spectrum projector. According to some embodiments of the present invention, the first computer processor may be configured to: (a) determine a velocity of the scrap metal material in real time based on a plurality of digital images acquired by the RGB camera; (b) generate a stereoscopic depth image in real time and correlate one or more pixels of the depth image with 3D coordinates of the scrap metal material in real time; and (c) determine a volume of the scrap metal material in real time based at least in part on the 3D coordinates of the scrap metal material. According to some embodiments of the present invention, the system may include a scrap classification module in operative communication with the sensor fusion module.The scrap classification module may include a machine learning model configured to: (a) identify, in real time, one or more distinct classes of scrap metal material from among a group of known scrap metal material (e.g., known classes or types of scrap metal material) based on at least a portion of the first RGB image received in real time from the sensor fusion unit; (b) assign a density to the scrap metal material in real time; and (c) in the second computer processor, determine, in real time, a scrap feed rate of the scrap metal material being transported on the conveyor based on the volume of the scrap metal material, the assigned first density, and the speed of the scrap metal material.

[0013] The present invention will now be described more fully hereinafter with reference to the accompanying drawings, in which some, but not all, embodiments of the invention are shown. Indeed, the invention may be embodied in many different forms and should not be construed as limited to the embodiments set forth herein; rather, these embodiments are provided so that this disclosure will satisfy applicable legal requirements. Like reference characters refer to like elements throughout. [Brief description of the drawings]

[0014] [Figure 1] 1 illustrates a continuous feed EAF system according to some embodiments of the present invention, which includes an EAF with a conveyor that is capable of continuously or semi-continuously feeding scrap metal material to the EAF for conversion into molten steel. [Diagram 2] 1 shows a schematic of a continuous feed EAF system according to some embodiments of the present invention, which includes three separate 3D cameras 110 mounted above separate locations along the length of a conveyor. [Diagram 3] 1 is an image of a sensor fusion module mounted above a conveyor according to some embodiments of the present invention. [Figure 4] 1 illustrates an example of a 3D camera according to some embodiments of the present invention. [Diagram 5] 1 shows an overview of a velocity measurement algorithm according to some embodiments of the present invention. [Figure 6] 1 shows an image of scrap metal material with velocity vectors superimposed on the digital image. [Figure 7] 1 shows a stereoscopically generated depth image and the corresponding RGB digital image. [Figure 8] 1 illustrates a 3D composite image formed by merging a depth image and an RGB image in a computer processor according to some embodiments of the present invention. [Figure 9] 1 illustrates a schematic of a system for operating an EAF system according to some embodiments of the present invention. [Figure 10] 13 illustrates a correlation between scrap speed and conveyor motor speed set point calculated using a system according to some embodiments of the present invention. [Figure 11] 1 illustrates a one-month scrap mix ratio for a system according to some embodiments of the present invention. [Figure 12] 13 shows how chunks of various scrap types accumulated while the EAF was connected to a system according to some embodiments of the present invention and the conveyor was running. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0015] In the following, embodiments of the present invention will be described in more detail with reference to the accompanying drawings, in which some, but not all, embodiments of the present invention are shown. Indeed, the present invention may be embodied in many different forms and should not be construed as being limited to the embodiments set forth herein, but rather, these embodiments are provided so that this disclosure will satisfy applicable legal requirements. As used in this specification and the appended claims, the singular forms "a," "an," and "the" include plural referents unless the context clearly dictates otherwise. As used herein, the term "and / or" includes any and all combinations of one or more of the associated listed items.

[0016] Embodiments of the present invention generally relate to methods and systems for measuring and / or determining (e.g., in real time) the feed rate (unit mass / unit time) of scrap metal material inputted by a conveyor (or other transport device) to an EAF. The feed rate of scrap metal material determined according to some of the methods and systems described and disclosed herein may be integrated into existing programmable logic controllers associated with the operation of the EAF to provide accurate estimates of tap weight (e.g., weight of molten steel exiting the EAF) and optimized energy usage (e.g., current consumed by electrodes of the EAF). According to some embodiments of the present invention, the methods and systems are capable of determining the real-time speed of scrap metal material being transported by a conveyor, where the speed of the scrap metal material is not necessarily the same as the conveyor speed, because components or pieces of scrap metal material may tumble, shift, gap, get stuck at certain points along the length of the conveyor, or move independently of the conveyor speed. In this regard, determining the actual scrap metal speed provides a significant advantage in determining the scrap metal feed rate in real time compared to simply assuming it is the same as the conveyor speed. According to some embodiments of the present invention, the method and system can also determine the volume of the scrap metal in real time, for example, by using stereoscopic vision to generate a depth image (e.g., three-dimensional information can be extracted by comparing relative pixel positions between two digital images acquired from two vantage points and examining the relative positions of objects or pixels in the two digital images) and calculating the volume of the scrap metal in a computer processor based on the depth image. For example, the method and system can also include obtaining an RGB digital image of the scrap metal and processing the RGB digital image using a machine learning model configured to identify and / or classify the nature of the individual scrap metal materials, operating the model which may have been previously trained and validated by machine learning operations.That is, the scrap metal material unloaded onto the conveyor and delivered to the EAF may include different types or classes of scrap metal material (e.g., different chemical compositions, different geometric shapes, etc.) For example, Table 1 provides a non-exhaustive list of some common classes of scrap metal material along with their respective chemical compositions, bulk densities, and morphologies.

[0017] [Table 1]

[0018] The machine learning model may, for example, identify a particular class or classes of scrap metal material in real time based on at least a portion of the RGB digital images. For example, the machine learning model may process a continuous or nearly continuous stream of continuous or nearly continuous RGB digital images (e.g., frames) in real time to identify when a first portion of scrap metal material passes that includes a first type or class of scrap metal material and is followed by a second portion of scrap metal material that includes a second type or class of scrap metal material. Based on the identification of the type or class of scrap metal material, a scrap density may be assigned to each component or piece of scrap metal material that is unloaded onto the conveyor and delivered to the EAF. According to some embodiments of the present invention, the assigned scrap densities may be utilized in real time along with respective velocity values ​​of the scrap metal material and respective volume values ​​of the scrap metal material to determine a real time feed rate (unit mass / unit time) of the scrap metal material in a computer processor.

[0019] According to some embodiments of the present invention, the methods and systems described and disclosed herein may provide one or more alarms or warnings in real time. For example, the machine learning model may be further configured to identify undesirable and / or hazardous materials (e.g., undesirable materials in an operating EAF) as they pass through (e.g., containers that contain or may contain oil or undesirable liquids (e.g., one or more oils, etc.) based on the RGB digital images. In this regard, the identification of the undesirable and / or hazardous materials may be provided to a computer processor that is configured to trigger an alarm or warning to an operator that the undesirable and / or hazardous materials are present in the scrap metal material, thereby providing the operator with an opportunity to remove the undesirable and / or hazardous materials before they are undesirably introduced into the EAF. Additionally or alternatively, the method and system may include a computer processor configured to trigger an alarm or warning when a piece of scrap metal material and / or a pile of scrap metal material is identified as having a volume exceeding a size threshold and / or an orientation extending outside a predetermined work zone, where such identification may be based on the volume identification results and / or the depth image. Additionally or alternatively, the method and system may include a computer processor configured to trigger an alarm or warning when a piece of scrap metal material is stuck at a particular location on the conveyor (e.g., stuck in a tunnel portion of the conveyor). For example, a portion of the conveyor adjacent and / or proximate to the EAF entrance for the scrap metal material may be covered with a tunnel, and waste heat (e.g., waste gas and / or waste steam) from the EAF is passed through the tunnel to preheat the scrap metal material before it is introduced into the EAF. For example, a stuck piece of scrap metal material can be identified by a speed identification of the scrap metal material.

[0020] FIG. 1 illustrates a continuous-feed EAF system 1 that includes an EAF 10 with a conveyor 20 that can continuously or semi-continuously feed scrap metal material 47 to the EAF for conversion to molten steel. The conveyor can include a distal end 24 to the EAF 10 and a proximal end 28 to the EAF 10. At least a portion of the proximal end 28 can be covered by a tunnel 30 through which waste heat (e.g., waste gases and / or waste steam) from the EAF 10 can be passed to preheat the scrap metal material before it is fed into the EAF. The EAF system 1 can include an exhaust gas treatment operation 35 operatively connected to the tunnel portion of the proximal end 28 of the conveyor 20. The distal end 24 of the conveyor 20 can be or include an input section where scrap metal material can be deposited (e.g., with a magnetic crane element 42) from a storage location 40 or a rail car 44. The EAF system shown in FIG. 1 also includes a sensor fusion module 100 mounted above a portion of a conveyor that is continuously or semi-continuously transporting scrap metal material 47 toward the EAF 10. The sensor fusion module 100 may include a three-dimensional (3D) camera or depth camera and, optionally, an associated computer processor. The 3D camera or depth camera may include, for example, two stereoscopic monochrome cameras, an RGB (red-green-blue) camera, and an infrared spectrum projector. The 3D camera or depth camera may also include an inertial measurement unit (IMU) including a gyroscope and an accelerometer, which may help counteract unwanted vibrations of the camera during use. One example of a 3D camera or depth camera is the Intel® RealSense™ Depth Camera D435. The 3D camera or depth camera is configured to capture digital images of the scrap metal material 47 moving, for example, down the conveyor 20 toward the EAF 10.

[0021] While Figure 1 generally illustrates a single location above the conveyor where a 3D camera may be mounted, some embodiments of the invention may include multiple 3D cameras mounted above and along a portion of the conveyor. For example, Figure 2 illustrates an exemplary embodiment in which scrap metal material of a first type or class 47a and (another) second type or class 47b are transported on a conveyor 20 toward an EAF 10, in which the sensor fusion unit includes three separate 3D cameras 110 mounted above separate locations along the length of the conveyor 20. Which of the three 3D cameras to use may be selected with a switch 110, whereby a digital image from the selected 3D camera is sent to a computer processor 120, which measures the velocity and volume of the scrap metal material beneath the selected camera 110. 2, the computer processor 120 may be operatively connected (e.g., directly wired or wirelessly connected) to another programmable logic controller 200, which may be associated with monitoring and / or controlling the operation of the EAF (e.g., positioning the electrodes, controlling current to the electrodes, etc.). As described throughout, methods and systems according to some embodiments of the present invention are capable of identifying and classifying a first type or class 47a and a second type or class 47b of scrap metal material using a machine learning model based at least in part on an RGB digital image captured by the 3D camera 110 in use. Operation of the machine learning model may be implemented on the computer processor 120 or on the programmable logic controller 200.

[0022] FIG. 3 illustrates a sensor fusion unit 100 mounted above the conveyor 20. As shown in FIG. 3, one or more controllable illumination sources 101 are mounted proximate to the sensor fusion module 100 and configured to provide constant illumination in real time. A constant illumination during operation can be useful in classifying scrap metal material because the machine learning model can include one or more convolutional layers that consider and / or weigh color scales at one or more locations of the scrap metal material. In this regard, a consistent or constant illumination reduces the possibility of errors due to the image quality of the RGB digital images used to identify and / or classify the scrap metal material. FIG. 4 illustrates an example of a 3D camera 110, which can be mounted within the sensor fusion module 100. The 3D camera can include two stereoscopic monochrome cameras 112 and one RGB camera 113 housed in a common casing.

[0023] In one aspect, some embodiments of the invention include a method for determining (e.g., in real time) a scrap feed rate of an EAF having a conveyor that introduces scrap metal material into the EAF, the method including the steps of: (i) acquiring a first plurality of RGB digital images of a first portion of the scrap metal material being transported by the conveyor; (ii) determining, in a computer processor, a first velocity of the first portion of the scrap metal material by comparing a first frame and a second frame of the first plurality of RGB digital images of the first portion of the scrap metal material being transported by the conveyor; and (iii) generating, in a computer processor, a first stereo camera depth image of the first portion of the scrap metal material, the computer processor correlating one or more pixels of the first depth image with 3D coordinates (e.g., x, y, z coordinates) of the first portion of the scrap metal material; (v) in the computer processor, identifying a first volume of the first portion of the scrap metal material based at least in part on the 3D coordinates of the first portion of the scrap metal material; (v) in the computer processor, identifying a first classification of the first portion of the scrap metal material from among a group of known scrap metal material based at least in part on the first RGB image using a machine learning model and assigning a first density to the first portion of the scrap metal material; and (vi) determining a scrap feed rate associated with the first portion of the scrap metal material being transported on the conveyor based on (a) the first volume of the first portion of the scrap metal material, (b) the assigned first density, and (c) the first speed of the first portion of the scrap metal material.

[0024] According to some embodiments of the invention, the first frame and the second frame may be successive frames of the plurality of RGB digital images. In this regard, the relative movement of one or more selected points (e.g., pixels or adjacent groups of pixels) of the first portion of the scrap metal material from the first frame to the second frame is compared in a computer processor to identify a first velocity of the first portion of the scrap metal material. The computer processor may be configured to perform, for example, various algorithms to identify the velocity of the first portion of the scrap metal material. According to some embodiments of the invention, the computer processor may be configured to perform a Horn-Schunck method based on the first frame and the second frame to identify the first velocity of the first portion of the scrap metal material. In this regard, the Horn-Schunck method has been found to be particularly suitable and computationally efficient, since the algorithm assumes that the flow is smooth throughout the image. This assumption in the algorithm minimizes flow distortions and favors solutions that exhibit greater smoothness. Furthermore, the pixel intensities of the RGB digital image of the first portion of the scrap metal material are constant between successive frames. Therefore, according to some embodiments of the present invention, it may be advantageous to use a controllable illumination source mounted proximate to the sensor fusion module and configured to provide constant illumination in real time. For example, Figure 5 illustrates an overview of a velocity measurement algorithm (i.e., the horn-shank method) according to some embodiments of the present invention, where a first velocity of the first portion of the scrap metal material is defined by a two-dimensional (2D) vector field generated in a computer processor, as shown in Figure 6 (i.e., an image of the scrap metal material with velocity vectors superimposed on the digital image).

[0025] As described above, according to some embodiments of the present invention, a computer processor may be configured to identify and / or trigger an alarm when a first component of a first portion of scrap metal material has fallen or rolled onto a conveyor based on the 2D vector field. For example, the computer processor may be configured to trigger an alarm or warning when a piece of scrap metal material gets stuck at a particular point on the conveyor (e.g., gets stuck in a tunnel portion of the conveyor).

[0026] According to some embodiments of the present invention, a method for determining (e.g., in real time) a scrap feed rate of an EAF having a conveyor for introducing scrap metal material into the EAF may further include merging the first depth image and the first RGB image in a computer processor to generate a first 3D composite image of the first portion of the scrap metal material and displaying the first 3D composite image on a user display. For example, FIG. 7 shows a stereoscopically generated depth image (e.g., red areas are closer to the camera and blue areas are farther from the camera) at the top of the figure, and a corresponding RGB digital image at the bottom of the figure. These images may be merged by a computer processor to form a 3D composite image as shown in FIG. 8. Displaying the 3D composite image on a user display may provide a visually easy to understand view of the scrap metal material in real time. Additionally or alternatively, the method may include performing a volumetric calibration operation, the volumetric calibration operation including mapping a plurality of pixels associated with a plurality of known locations of a calibration tool (e.g., an inanimate object of known dimensions placed on the conveyor while the conveyor is not moving) to a 3D coordinate system (e.g., x, y, z) and generating in the computer processor a volumetric computational model based on the mapped pixels, the volumetric computational model being configured in the computer processor to determine a volume of the scrap metal material based at least in part on the 3D coordinates of the imaged scrap metal material.

[0027] According to some embodiments of the invention, the first RGB image may be sent to a machine learning model to cause the machine learning model to converge on a solution, for example, in a computer processor, identifying and / or classifying the imaged scrap metal (e.g., a first portion of the scrap metal) by the machine learning model from among a group of known scrap metals based on at least a portion of the first RGB image. Once the machine learning model has converged on a solution for identifying the imaged scrap metal (e.g., NO2 Heavy), the computer processor may assign (for example) a first density to the first portion of the scrap material (e.g., NO2 Heavy has a density of 0.75 kg / m 3 ), a scrap feed rate associated with a first portion of the scrap metal material being transported on the conveyor may be determined based on (a) a first volume of the first portion of the scrap metal material, (b) the assigned first density, and (c) a first velocity of the first portion of the scrap metal material.

[0028] According to some embodiments of the present invention, the machine learning models disclosed and described herein may include one or more types of machine learning, such as, for example, neural networks (e.g., convolutional neural networks, recurrent neural networks, deep learning neural networks, natural language processing, etc.), transformers, support vector machines (SVMs), and clusters (e.g., clustered machine learning). In this regard, for example, use of the term "machine learning model" throughout this disclosure may encompass (i) a single type of machine learning model according to some embodiments of the present invention, and (ii) different types of machine learning models according to other embodiments of the present invention. By way of example only, a first type of machine learning model (e.g., a neural network) may be used to identify and classify scrap metal material from among one or more types of classifications, while a second type of machine learning model may be used to identify and / or classify undesirable or potentially undesirable materials (e.g., oil or other liquids that are undesirable in the EAF, sealed containers that may contain oil or other liquids that are undesirable in the EAF, etc.) that are mixed with the scrap metal material intended for input into the EAF. The first type of machine learning model may include, for example, a deep learning solution having multiple convolutional layers and / or multiple pooling layers and at least one fully connected layer configured for the identification and / or classification of one or more types or classes of scrap metal material. In this regard, for example, at least one of the convolutional layers may capture the highest scoring scrap type selected as the classification result for the input scrap image. According to some embodiments of the present invention, the first convolutional layer may consider and / or weigh the identified straight edges, arc edges, colors, and / or color gradients across the surface. According to some embodiments of the present invention, different layers may have different node layouts, where each node may have a respective weight. The respective weights may be calculated during the training phase, and the model results may be customized according to different custom wrap types utilized in the EAF.

[0029] According to some embodiments, the scrap classification model was developed using machine learning techniques. For example, an initial model for training may be selected, which may be done, for example, by backpropagation, which serves to feed the classified data into the model and then measures the performance of the model. The error rate is measured using a loss function. In this regard, for example, backpropagation serves to use gradient descent to measure the rate of change of the loss function with respect to the weighting of each connection, and a gradient descent step is used to reduce the error rate of each connection as close to zero as possible to ensure that the network converges to a solution. According to some embodiments of the present invention, training of an initial algorithm or initial model configured for general image identification may be performed by selecting one or more of the following: a percentage of images to be used for training and validation (e.g., 70% for training and 30% for validation), applying various preprocessing to the input images, setting training and validation thresholds, defining the number of iterations per epoch, selecting a gradient thresholding method, selecting a learning rate, and validating a confusion matrix. Once the trained model meets all user specifications, the model may be exported to a computer processor for use in image identification and / or image classification according to some embodiments of the present invention. For example, the algorithm or model may be exported as an ONNX (Open Neural Network Exchange) model. ONNX (Open Neural Network Exchange) is an open format for expressing artificial intelligence models that is widely supported and found in many frameworks, tools, and hardware. According to some embodiments, the algorithm or model may be retrained if the number of non-solutions reaches an unacceptable number during operation of some methods described and disclosed herein.

[0030] As discussed above, and as generally shown in FIG. 2, the scrap metal material being transported on the conveyor may include multiple different types or classes of scrap metal material that are provided separately or intermingled with one another.In this regard, some embodiments according to the present invention include a method for detecting scrap metal material from a conveyor, the method comprising: (i) acquiring a plurality of second RGB digital images of a second portion of the scrap metal material being transported on a conveyor, the second portion of the scrap metal material being different (e.g., different type or class) from the first portion of the scrap metal material, the first portion of the scrap metal material being closer to the EAF than the second portion of the scrap metal material; (ii) determining, in a computer processor, a second velocity of the second portion of the scrap metal material by comparing a first frame and a second frame of the second plurality of RGB digital images of the second portion of the scrap metal material being transported on the conveyor; (iii) generating, in a computer processor, a second stereo camera depth image of the second portion of the scrap metal material, the computer processor correlating one or more pixels of the second depth image with 3D coordinates of the second portion of the scrap metal material; and (iv) computing a second depth image of the second portion of the scrap metal material from a computer processor; The method may include (a) identifying, in the computer processor, a second volume of the second portion of the scrap metal material based at least in part on the 3D coordinates of the second portion of the scrap metal material; (b) identifying, in the computer processor, a second classification of the second portion of the scrap metal material from among a group of known scrap metal material based at least in part on the second RGB image using a machine learning model and assigning a second density to the second portion of the scrap metal material, wherein the second classification is different from the first classification; and (c) determining a scrap feed rate associated with the second portion of the scrap metal material being transported on the conveyor based on (a) the second volume of the second portion of the scrap metal material, (b) the assigned second density, and (c) the second velocity of the second portion of the scrap metal material, wherein the scrap feed rate associated with the second portion of the scrap metal material is different from the scrap feed rate associated with the second portion of the scrap metal material.In this regard, the method may include determining in real time a scrap feed rate for scrap metal material being transported by the conveyor (e.g., the scrap feed rate may vary in real time depending on the type or class of scrap metal material and / or their respective volumes).

[0031] In another aspect, some embodiments according to the present invention provide a method of operating an EAF. The method may include determining, in real time, a real-time scrap feed rate of scrap metal material being fed into the EAF by a conveyor, as described and disclosed herein. The method may also include comparing the real-time scrap feed rate to operational conditions of the EAF (e.g., current to the electrodes, etc.) in a computer processor (e.g., a programmable logic controller configured to control operation of furnace conditions, current to electrodes, tapping times, etc.). Additionally, the method may include modifying (e.g., in real time) a conveyor speed of a conveyor feeding scrap metal material into the EAF, modifying (e.g., in real time) a current flow rate to electrodes in the EAF, or both. Modifying the conveyor speed of a conveyor feeding scrap metal material into the EAF, modifying a current flow rate to electrodes in the EAF, or both may be performed by a programmable logic controller configured to control operation of furnace conditions. According to some embodiments of the invention, the conveyor speed may be decreased if the real-time scrap feed rate increases and / or the current flow rate may be increased if the real-time scrap feed rate increases. Additionally or alternatively, the conveyor speed may be increased if the real-time scrap feed rate decreases and / or the current flow rate may be decreased if the real-time scrap feed rate decreases.

[0032] According to some embodiments of the invention, the computer processor may be configured to identify and / or trigger a second alarm when a piece of scrap metal material and / or a pile of scrap metal material is identified as having a volume exceeding a size threshold and / or an orientation extending outside a predetermined work zone. In this regard, the size threshold may be based on a volume or size that may undesirably strike an electrode in the EAF during operation. The second alarm may, for example, enable an operator to at least partially lift the electrode upward to increase space to allow a larger piece of scrap metal to enter the EAF without damaging the electrode. Additionally or alternatively, the machine learning model in the computer processor is configured to trigger a third alarm and / or stop the conveyor upon identifying undesirable material mixed in with the scrap metal material.

[0033] In another aspect, some embodiments according to the present invention provide a system for determining a scrap feed rate of an EAF having a conveyor that introduces scrap metal material into the EAF. The system may include a sensor fusion module, which includes two monochrome cameras configured to provide stereoscopic vision, an RGB camera, an infrared spectrum projector, and a first computer processor in operative communication with the two monochrome cameras, the RGB camera, and the infrared spectrum projector. The first computer processor may be configured, for example, to: (a) determine a speed of the scrap metal material in real time based on a plurality of digital images acquired by the RGB cameras; (b) generate a stereoscopic depth image in real time and correlate one or more pixels of the depth image with 3D coordinates of the scrap metal material in real time; and (c) determine a volume of the scrap metal material in real time based at least in part on the 3D coordinates of the scrap metal material. The system may also include a scrap classification module in operative communication with the sensor fusion module, which includes a machine learning model. The machine learning model may be configured, for example, to: (a) identify, in real time, one or more distinct classes or types of scrap metal material from among a group of known scrap metal material (e.g., known classes or types of scrap metal material) based at least in part on the first RGB image received in real time from the sensor fusion unit; (b) assign a density to the scrap metal material in real time; and (c) determine, in the second computer processor, a scrap feed rate of the scrap metal material being conveyed in real time based on the volume of the scrap metal material, the assigned first density, and the speed of the scrap metal material. According to some embodiments of the invention, the system includes one or more controllable illumination sources mounted proximate to the sensor fusion module and configured to provide a constant illuminance in real time, as described above.

[0034] In yet another aspect, some embodiments according to the invention provide an EAF system including an EAF including a furnace body defining an interior portion and a set of electrodes extending into the interior portion. The system may also include a conveyor, the conveyor including a first portion distal to the EAF for receiving scrap metal material (e.g., a new supply for the EAF, which may be loaded onto the conveyor from a rail car or a crane), a second portion proximal to the EAF for inputting the scrap metal material into the interior portion of the EAF, and a third portion located between the first portion and the second portion, the conveyor transporting the scrap metal material from the first portion through the third portion and the second portion to the interior portion of the EAF. The system may further include a sensor fusion module mounted above a section of the third portion of the conveyor. The sensor fusion module may include two monochrome cameras configured to provide stereoscopic vision, an RGB camera, an infrared spectrum projector, and a first computer processor in operative communication with the two monochrome cameras, the RGB camera, and the infrared spectrum projector. According to some embodiments of the present invention, the first computer processor may be configured to: (a) determine a velocity of the scrap metal material in real time based on a plurality of digital images acquired by the RGB camera; (b) generate a stereoscopic depth image in real time and correlate one or more pixels of the depth image with 3D coordinates of the scrap metal material in real time; and (c) determine a volume of the scrap metal material in real time based at least in part on the 3D coordinates of the scrap metal material. According to some embodiments of the present invention, the system may include a scrap classification module in operative communication with the sensor fusion module.The scrap classification module may include a machine learning model configured to: (a) identify, in real time, one or more distinct classes of scrap metal material from among a group of known scrap metal material (e.g., known classes or types of scrap metal material) based at least in part on the first RGB image received in real time from the sensor fusion unit; (b) assign a density to the scrap metal material in real time; and (c) determine, in the second computer processor, a scrap feed rate of the scrap metal material being conveyed in real time based on the volume of the scrap metal material, the assigned first density, and the speed of the scrap metal material. According to some embodiments of the invention, the system includes one or more controllable illumination sources mounted proximate to the sensor fusion module and configured to provide a constant illuminance in real time, as described above.

[0035] 9 illustrates a schematic of a system for operating an EAF system according to some embodiments of the present invention, including a sensor fusion unit 100 in which a computer program 120 is in operative communication with a scrap metal volume measurement module 180 and a scrap metal velocity measurement module 190. The sensor fusion unit 100 transmits real-time velocity values ​​of the imaged scrap metal material and real-time velocity values ​​of the imaged scrap metal material to a programmable logic controller 200 (or other computer processor) associated with controlling the operation of the EAF. The sensor fusion unit 100 also transmits an RGB digital image 107 of the scrap metal material to a scrap classification module 210. The scrap classification module 210 uses machine learning models to identify the type or class of the scrap metal material and assign a density to the scrap metal material. The assigned density, the determined velocity, and the determined volume of the scrap metal material may be input to a mass flow estimation module, which outputs a real-time scrap metal feed rate. The programmable logic controller may also receive various feedback data 300 from various functions of the EAF system. The programmable logic controller may adjust the real-time scrap metal feed rate to the EAF by comparing the real-time scrap metal feed rate to the feedback data 300 and, in response to the comparison, defining or generating a new set point for the conveyor speed.

[0036] Working Example The present disclosure is further illustrated in the following examples, which should not be construed as limiting in any way, i.e., the specific features described in the following examples are illustrative only and not limiting.

[0037] A system according to some embodiments of the present invention was implemented in a steel mill with a Consteel EAF. In this particular steel mill, the feeding to the conveyor was done in layers. Therefore, two sensor fusion modules were installed above the conveyor. One sensor fusion module was mounted at the crushing inlet and the other sensor fusion module was installed just before the reheat furnace. FIG. 10 shows the correlation between the scrap speed and the conveyor motor speed set point calculated using the system according to some embodiments of the present invention. The scrap speed is shown to have a correlation of 0.87 with the conveyor speed, which reflects the actual inertia of the large scrap pieces vibrating in the conveyor pacing. The steel mill had another scrap estimation system based on crane position tracking and magnetic load cells. However, that system exhibited significant delays, which made the incoming scrap data inaccurate and unusable in most cases. During the commissioning of the system according to some embodiments of the present invention, the scrap tracking was switched from the crane-based system to the system according to some embodiments of the present invention, as the latter was found to be significantly more reliable. With the availability of two sensor fusion modules, the material tracking system was implemented to inform the steel mill of the exact time when a particular scrap (e.g. scrap metal material) was dropped into the EAF. Scrap classification was performed using a database of more than 10,000 images acquired by the first and second sensor fusion modules, and all images were pre-processed with various filters. More than 20 deep learning models were evaluated to select the best performing model with an accuracy of more than 95%. During training, only 70% of the dataset was used, while the remaining 30% was used only for validation. In this regard, the scrap types in the first dataset were manually labeled by process engineers, and the machine learning model automatically labeled the images thereafter. The initial dataset included images of crushed, turning, heavy metal 1, heavy metal 2, mill return, bussering, bundles, plates, and pig iron.Whenever the classification model score fell below 50, the image was archived in a separate location for later model retraining. The retraining process was performed only once. Once the scrap was classified, a density was selected for that scrap type. Knowing both the volume and velocity of the scrap with density being fed into the EAF in real time allowed for a high degree of accuracy in estimating the actual EAF feed rate. FIG. 11 shows the scrap mix of the system over a one-month period, estimated only by the automated scrap classification system according to some embodiments of the present invention, and showing 99% accuracy with respect to the actual plant scrap mix. In this regard, only 1% of the images were classified as unknown, while the rest were the correct mix of the steel mill for that period.

[0038] 12 shows how the mass of various scrap types accumulated while the EAF was connected to a system according to some embodiments of the present invention and the conveyor was running, and the total heat weight was calculated as 40 tons. The calculated total weight of 40 tons matched the tapping weight of that heat.

[0039] These and other modifications and variations to the embodiments of the present invention may be made by those skilled in the art without departing from the spirit and scope of the present invention, which is more particularly set forth in the appended claims. Moreover, it will be understood that aspects of the various embodiments may be interchanged in whole or in part. Moreover, those skilled in the art will appreciate that the above description is merely illustrative and is not intended to limit the invention, as further set forth in such appended claims. Thus, the spirit and scope of the appended claims should not be limited to the exemplary description of the forms contained herein.

Claims

1. 1. A method for determining a scrap feed rate for an electric arc furnace (EAF) having a conveyor for feeding scrap metal material into the EAF, the method comprising: (i) acquiring a first plurality of red-green-blue (RGB) digital images of a first portion of the scrap metal material being transported on the conveyor; (ii) determining, in a computer processor, a first velocity of the first portion of the scrap metal material by comparing first and second frames of the plurality of RGB digital images of the first portion of the scrap metal material being transported on the conveyor; (iii) generating, in the computer processor, a first stereoscopic depth image of the first portion of the scrap metal material, the generating step correlating, in the computer processor, one or more pixels of the first depth image with 3D coordinates of the first portion of the scrap metal material; (iv) in the computer processor, identifying a first volume of the first portion of the scrap metal material based at least in part on the 3D coordinates of the first portion of the scrap metal material; (v) identifying, in the computer processor, a first classification of the first portion of scrap metal material from among a group of known scrap metal material based at least in part on the first RGB image with a machine learning model and assigning a first density to the first portion of scrap material; (vi) determining the scrap feed rate associated with the first portion of the scrap metal material being transported on the conveyor based on (a) the first volume of the first portion of the scrap metal material, (b) the assigned first density, and (c) the first velocity of the first portion of the scrap metal material; The method includes:

2. 2. The method of claim 1, wherein the first frame and the second frame are successive frames of the plurality of RGB digital images, and relative movement of one or more selected points of the first portion of the scrap metal material from the first frame to the second frame is compared in the computer processor to determine the first velocity of the first portion of the scrap metal material.

3. 3. The method of claim 1, wherein the computer processor is configured to perform a Horn-Shank technique based on the first frame and the second frame to determine the first velocity of the first portion of the scrap metal material.

4. The method of any one of claims 1 to 3, wherein the first velocity of the first portion of the scrap metal material is defined by a two-dimensional (2D) vector field generated in the computer processor.

5. 5. The method of claim 4, wherein the computer processor is configured to identify and / or trigger an alarm when a first component of the first portion of the scrap metal material has fallen or rolled on the conveyor based on the 2D vector field.

6. 6. The method of claim 1, further comprising the step of performing a volumetric calibration operation, the volumetric calibration operation including mapping a plurality of pixels associated with a plurality of known locations of a calibration tool placed on the conveyor to a 3D coordinate system, and generating in the computer processor a volumetric calculation model based on the mapped pixels, the volumetric calculation model being configured in the computer processor to determine a volume of the first portion of the scrap metal material based at least in part on the 3D coordinates of the first portion of the scrap metal material.

7. 7. The method of claim 1, further comprising: merging, in the computer processor, the first depth image and the first RGB image to generate a first 3D composite image of the first portion of the scrap metal material; and displaying the first 3D composite image on a user display.

8. The method according to any one of claims 1 to 7, wherein the machine learning model includes multiple convolutional layers and / or multiple pooling layers and at least one fully connected layer, and at least one of the convolutional layers captures the scrap type with the highest score selected as a classification result for the input scrap image.

9. (i) acquiring a plurality of second RGB digital images of a second portion of the scrap metal material being transported on the conveyor, the second portion of the scrap metal material being different from the first portion of the scrap metal material, the first portion of the scrap metal material being located closer to the EAF than the second portion of the scrap metal material; (ii) determining, in the computer processor, a second velocity of the second portion of the scrap metal material by comparing a first frame and a second frame of the second plurality of RGB digital images of the second portion of the scrap metal material being transported on the conveyor; (iii) generating, in the computer processor, a second stereoscopic depth image of the second portion of the scrap metal material, the generating step correlating, in the computer processor, one or more pixels of the second depth image with 3D coordinates of the second portion of the scrap metal material; (iv) determining, in the computer processor, a second volume of the second portion of the scrap metal material based at least in part on the 3D coordinates of the second portion of the scrap metal material; (v) identifying, in the computer processor, a second classification of the second portion of scrap metal material from among the group of known scrap metal material based at least in part on the second RGB image with the machine learning model and assigning a second density to the second portion of scrap material, the second classification being different from the first classification; (vi) determining the scrap feed rate associated with the second portion of the scrap metal material being transported on the conveyor based on (a) the second volume of the second portion of the scrap metal material, (b) the assigned second density, and (c) the second velocity of the second portion of the scrap metal material, wherein the scrap feed rate associated with the second portion of the scrap metal material differs from the scrap feed rate associated with the second portion of the scrap metal material; The method of any one of claims 1 to 8, further comprising:

10. A method as claimed in any preceding claim, wherein the scrap feed rate of the scrap metal material being transported by the conveyor is provided in real time.

11. The method according to any one of claims 1 to 10, wherein the machine learning model executes an algorithm developed by machine learning.

12. 1. A method of operating an electric arc furnace (EAF), comprising: (i) determining in real time a real time scrap feed rate of scrap metal material being conveyed to the EAF by a conveyor according to any one of claims 1 to 11; (ii) comparing, in a computer processor, the real-time scrap feed rate with an operating condition of the EAF; (iii) modifying a conveyor speed of the conveyor feeding the scrap metal material into the EAF, modifying a current flow rate to an electrode in the EAF, or both; The method includes:

13. 13. The method of claim 12, wherein the conveyor speed is decreased if the real-time scrap feed rate increases and / or the current flow rate is increased if the real-time scrap feed rate increases.

14. 13. The method of claim 12, wherein the conveyor speed is increased if the real-time scrap feed rate decreases and / or the current flow rate is decreased if the real-time scrap feed rate decreases.

15. 15. The method of claims 11-14, wherein the computer processor is configured to identify and / or trigger a second alarm when the piece of scrap metal and / or the pile of scrap metal is identified as having a volume that exceeds a size threshold and / or an orientation that extends outside a predefined work zone.

16. The method of any one of claims 11 to 15, wherein the machine learning model in the computer processor is configured to trigger a third alarm and / or stop the conveyor upon identifying undesirable substances mixed into the scrap metal material.

17. 1. A system for determining a scrap feed rate for an electric arc furnace (EAF) having a conveyor for feeding scrap metal material into the EAF, the system comprising: (i) a sensor fusion module including two monochrome cameras configured to provide stereoscopic vision, an RGB camera, an infrared spectrum projector, and a first computer processor in operative communication with the two monochrome cameras, the RGB camera, and the infrared spectrum projector, the first computer processor configured to: (a) determine a velocity of the scrap metal material in real time based on a plurality of digital images captured by the RGB cameras; (b) generate a stereoscopic depth image in real time and correlate one or more pixels of the depth image with 3D coordinates of the scrap metal material in real time; and (c) determine a volume of the scrap metal material in real time based at least in part on the 3D coordinates of the scrap metal material. (ii) a scrap classification module in operative communication with the sensor fusion module and including a machine learning model, the machine learning model configured to: (a) identify, in real time, one or more distinct classes of the scrap metal material from among a group of known scrap metal materials based at least in part on a first RGB image received in real time from the sensor fusion unit; (b) assign a density to the scrap metal material in real time; and (c) in a second computer processor, determine, in real time, the scrap feed rate of the scrap metal material being transported on the conveyor based on the volume of the scrap metal material, the assigned first density, and the velocity of the scrap metal material; A system including:

18. 20. The system of claim 17, further comprising one or more controllable illumination sources mounted proximate to the sensor fusion module and configured to provide a constant illumination intensity in real time.

19. 1. An electric arc furnace (EAF) system comprising: (i) an EAF including a furnace body defining an interior portion and a set of electrodes extending into the interior portion; (ii) a conveyor including a first portion distal to the EAF for receiving scrap metal material, a second portion proximal to the EAF for depositing the scrap metal material into the interior portion of the EAF, and a third portion located between the first portion and the second portion, the conveyor transporting the scrap metal material from the first portion through the third portion and the second portion to the interior portion of the EAF; (iii) a sensor fusion module mounted above a section of the third portion of the conveyor, the sensor fusion module including two monochrome cameras configured to provide stereo vision, an RGB camera, an infrared spectrum projector, and a first computer processor in operative communication with the two monochrome cameras, the RGB camera, and the infrared spectrum projector, the first computer processor configured to: (a) determine a speed of the scrap metal material in real time based on a plurality of digital images acquired by the RGB cameras; (b) generate a stereo depth image in real time and correlate one or more pixels of the depth image with 3D coordinates of the scrap metal material in real time; and (c) determine a volume of the scrap metal material in real time based at least in part on the 3D coordinates of the scrap metal material. (ii) a scrap classification module in operative communication with the sensor fusion module and including a machine learning model, the machine learning model configured to: (a) identify, in real time, one or more distinct classes of the scrap metal material from among a group of known scrap metal materials based at least in part on a first RGB image received in real time from the sensor fusion unit; (b) assign a density to the scrap metal material in real time; and (c) in a second computer processor, determine, in real time, the scrap feed rate of the scrap metal material being transported on the conveyor based on the volume of the scrap metal material, the assigned first density, and the velocity of the scrap metal material; A system including:

20. 20. The system of claim 19, further comprising one or more controllable illumination sources mounted proximate to the sensor fusion module and configured to provide a constant illumination intensity in real time to the section of the third portion of the conveyor.

Images