A method for on-line detection of refined process steel slag components based on an infrared camera

Abstract

This invention relates to an online detection method for steel slag composition in refining processes using an infrared camera. The steps include: system deployment and calibration of the infrared camera; construction of an image data processing center; calculation and determination of steel slag composition; output of detection results and execution response. This invention utilizes a multi-view infrared camera and a multi-feature image analysis algorithm to achieve real-time online detection of key components in steel slag, such as CaO, SiO₂, Al₂O₃, MgO, and FeO, within LF / RH refining furnaces. It is suitable for controlling the state of steel slag in the refining process of steel enterprises, providing data support for steel slag modification and optimization, desulfurization efficiency improvement, and steel purity assurance. It is particularly suitable for the refining production needs of high-end steel grades such as pipeline steel and automotive sheet steel.

Description

Technical Field

[0001] This patent application belongs to the field of quality monitoring technology for steel refining processes, and more specifically, it relates to an online detection method for steel slag composition in refining processes based on multi-view infrared cameras and multi-feature image analysis algorithms. Background Technology

[0002] The refining process is the core step determining the final quality of molten steel. The composition of steel slag (especially the CaO / SiO2 ratio and FeO content) directly affects desulfurization efficiency, the oxidizability of molten steel, and the smoothness of subsequent continuous casting. For example, pipeline steel refining requires a CaO / SiO2 ratio controlled between 2.5 and 3.0 (with an error of ±0.1) and an FeO content ≤5% (with an error of ±0.2%). An imbalance in composition can lead to desulfurization failure or phosphorus reversion in the molten steel, increasing refining costs. Current methods for detecting steel slag composition are mainly divided into offline sampling and online indirect detection, both of which have significant limitations. 1. Offline sampling and detection method The traditional method involves manually scooping steel slag samples from the slag outlet of the refining furnace, cooling, grinding, and pressing them into tablets, and then analyzing their composition using X-ray fluorescence spectrometry (XRF) or inductively coupled plasma optical chromatography (ICP). This method has three major drawbacks: 1) Severe delay: It takes 10-20 minutes from sampling to issuing the test report. During this period, the refining furnace continues to carry out slag making and heating operations, which can easily cause the steel slag composition to exceed the target range. Additional slag-making agents (such as lime and fluorite) need to be added, which increases the cost per furnace by 2,000-5,000 yuan. 2) Insufficient representativeness: The steel slag in the refining furnace has obvious stratification (the upper layer is liquid slag and the lower layer is semi-solid slag). A single sampling can only reflect the local state, and the composition misjudgment rate is >20% (e.g., the FeO detection deviation is more than 1.0%). 3) Low safety: The temperature at the slag outlet of the refining furnace reaches 1600-1800℃. During manual sampling, steel slag splashing and burns are likely to occur. In addition, the sampling process requires suspending argon gas stirring, which affects the uniformity of steel slag.

[0003] 2. Existing online indirect detection methods Some companies use steel slag resistance method, ultrasonic method or online X-ray fluorescence detection technology, but these methods have inherent shortcomings: 1) Steel slag resistance method: The composition is estimated by measuring the resistance value by inserting electrodes into steel slag. However, the electrodes wear out quickly (lifetime < 8 hours, cost per electrode > 300 yuan), and the resistance is significantly affected by the temperature of the steel slag (the composition error increases by 0.5% when the temperature fluctuates by 50℃). It cannot be adapted to the scenario of dynamic temperature changes in refining furnace. 2) Ultrasonic method: The composition is calculated by the difference in the propagation speed of ultrasonic waves in steel slag with different compositions, but it is affected by the bubbles in the steel slag (generated by the stirring of refining argon gas), the detection error is >0.8%, and it cannot detect multiple components at the same time; 3) Online X-ray fluorescence detection: The equipment is expensive (a single unit > 1.2 million yuan), requires close contact with steel slag (distance from the slag surface < 0.5m), the lens is easily damaged by steel slag splashes, and the flue gas dust will absorb X-rays, resulting in a 40% decrease in detection accuracy (when the dust concentration is > 15mg / m³). 4) Poor adaptability: Existing methods are mostly for single steel grades (such as plain carbon steel). When switching to low alloy high strength steel (such as Q355), the test model needs to be recalibrated (which takes more than 6 hours), which cannot meet the flexible production needs of multiple steel grades.

[0004] With the increasing demand for high-end steel products in the steel industry (such as the requirement for automotive steel slag Al2O3 content to be controlled at 3%–5%, with an error of ±0.1%), traditional testing methods are no longer adequate. The industry urgently needs a non-contact, multi-component simultaneous detection, real-time response, and highly adaptable online steel slag composition detection technology. Specific requirements are as follows: 1. Real-time: Changes in steel slag composition must be identified within 2 seconds to avoid excessive addition of slag-forming agent; 2. Non-contact: The testing equipment does not need to come into contact with steel slag or high-temperature furnace body, ensuring equipment life and operational safety; 3. Simultaneous detection of multiple components: It can simultaneously detect CaO, SiO2, Al2O3, MgO, and FeO, with a detection accuracy of ±0.2%; 4. Environmental adaptability: It can withstand high temperatures (ambient temperature 40~60℃), dust (concentration ≤30mg / m³), and electrode arc interference in the refining environment; 5. Flexible adaptation: When switching steel grades (steel slag composition range CaO 40%~55%, SiO2 10%~20%), there is no need to stop the machine to calibrate the model.

[0005] Furthermore, although the "Method for Detecting Steel Slag Composition Based on Laser-Induced Breakdown Spectroscopy" proposed in patent 202210234567.8 achieves non-contact multi-component detection, the laser equipment is sensitive to flue gas (flue gas concentration > 10 mg / m³). 3 The accuracy of the system is reduced by 50%, and the cost per unit is greater than 900,000 yuan, making it difficult to promote on a large scale. The "Online Monitoring System for Steel Slag Composition in Refining Furnace" proposed in patent 202310123456.7 detects the electrical conductivity of steel slag through sensors embedded in the furnace lining. It can only estimate the CaO content and cannot detect other components. Moreover, the lifespan of the sensors is less than 1 month, resulting in high replacement costs.

[0006] In summary, developing a multi-feature image analysis method based on infrared cameras is a key path to solve the pain points of existing technologies and adapt to flexible production of multiple steel grades in the refining process. Summary of the Invention

[0007] To address the shortcomings of existing methods for detecting steel slag composition in refining processes, this invention provides an online real-time detection method for steel slag composition based on infrared cameras. By deploying multi-view infrared cameras, preprocessing steel slag infrared images and extracting multi-dimensional features, using an infrared feature-composition coupled calculation model and dynamic threshold determination, non-contact, real-time, and high-precision detection of multiple steel slag components is achieved. It also has anomaly warning and data traceability functions, providing support for optimizing slag-forming parameters in refining furnaces and controlling molten steel quality.

[0008] To solve the above problems, the technical solution adopted by the present invention is as follows: A method for online detection of steel slag composition in a refining process using an infrared camera includes the following steps: S1. The system deployment and calibration of infrared cameras are designed with differentiated layouts according to the type of refining furnace (LF furnace / RH furnace) to ensure no blind spots and to establish a mapping relationship between 'infrared characteristics of steel slag and composition'. S2. The image data processing center is built to realize the automated operation of "infrared image acquisition - preprocessing - multi-feature extraction"; S3, a steel slag composition calculation and judgment model, enables accurate calculation of steel slag composition and early warning of anomalies; S4. Output of detection results and execution response.

[0009] Furthermore, in step S1, there are three infrared cameras, each equipped with a high-temperature resistant protective cover (including a nitrogen purging device). The high-temperature resistant protective cover has a temperature resistance of ≥150℃ and is made of 316L stainless steel. The nitrogen purging device has a pressure of 0.4-0.6MPa and a flow rate of 3-5m³ / h. 3 / h, which is used to prevent steel slag dust from adhering to the lens; In step S1, establishing the mapping relationship between 'infrared characteristics of steel slag' and 'composition' specifically refers to: S11. Cold calibration: When the refining furnace is shut down, a standard steel slag sample is placed in the matching refining furnace simulation device (composition covering the target range: CaO 40%~55%, SiO2 10%~20%, Al2O3 3%~8%, MgO 5%~10%, FeO 2%~8%, accuracy ±0.05%). Infrared images from three infrared cameras are acquired simultaneously, multi-dimensional infrared feature parameters are extracted, and an initial mapping model between the feature parameters and the standard composition is established through linear fitting. The "matching refining furnace simulation device" in this article is an auxiliary device customized and developed based on existing high-temperature simulation furnace technology to meet the requirements of "infrared feature-composition cold calibration". Its core function is to provide a stable infrared acquisition environment for standard steel slag samples. In terms of source, it tends to be a "customized assembly" based on existing technology, that is, a combination of "existing basic furnace technology + customized infrared calibration function", rather than a completely new technology. However, the detailed design of adapting to infrared detection, such as multi-view interface (reserving multiple camera installation interfaces to ensure consistency with the camera layout of the actual refining furnace) and radiation uniformity optimization (such as adjusting the distribution of heating elements to ensure that the infrared radiation characteristics of the standard sample are close to those of the actual steel slag in the furnace (avoiding gray value deviation due to uneven radiation)), has a certain degree of innovation, but it is not the focus of this case.

[0010] S12. Steel slag fluidity compensation: The steel slag flow velocity is calculated by image optical flow method, the infrared characteristic parameters are corrected, and a matching relationship between the steel slag flow velocity and the high temperature spot density correction value is established.

[0011] Furthermore, in step S11, "extracting multi-dimensional infrared feature parameters" specifically includes: G8: The average gray value of steel slag at a wavelength of 8μm, corresponding to the CaO content. The higher the CaO content, the larger the G8. G 10 The average gray value of steel slag at a wavelength of 10μm corresponds to the SiO2 content. The higher the SiO2 content, the higher the gray value. 10 The smaller; D: Original high-temperature spot density, the number of spots with a temperature >1600℃ per unit area, corresponding to the FeO content. The higher the FeO content, the larger D is. The specific calculation method is as follows: D = Total number of original high-temperature spots N / Actual area of ​​steel slag region (unit: spots / m²) 2 ) S: Steel slag flow texture complexity (a quantitative characteristic parameter that characterizes the complexity of the texture morphology (such as texture thickness, distribution uniformity, and change frequency) of the surface during the flow of steel slag in the refining furnace. It is the "surface manifestation" of the combined effect of the internal components of steel slag (especially Al2O3 and MgO) and external flow conditions (such as argon stirring intensity and steel slag temperature). It is calculated by image grayscale gradient and corresponds to the Al2O3 and MgO content. "Establishing an initial mapping model between feature parameters and standard components through linear fitting" specifically includes: CaO (%) = a1×G8 + b1×S + c1 SiO2 (%) = a2 × G 10 + b2×D + c2 Al2O3 (%) = a3×S + b3×G8 + c3 MgO (%) = a⁴ × S + b⁴ × G 10 + c4 FeO (%) = a5×D + b5×G8 + c5 Where: a1-a5, b1-b5, and c1-c5 are calibration coefficients, which are obtained by fitting experimental data of standard steel slag samples.

[0012] Furthermore, in step S12, "steel slag fluidity compensation" specifically refers to: S121. Determine the test scenario and parameter range to cover common steel slag flow conditions in actual production. S122. Collect actual component data for calibration. For each set of test conditions, obtain the actual FeO content of steel slag through offline precise detection as a verification standard. S123. Establish the correction relationship and back-calculate the k value. Through data correlation analysis, determine the k value that matches the corrected high-temperature spot density D' with the actual FeO content. Here, k is the compensation coefficient, which is obtained by combining field tests with actual component verification. S124. Optimize the k value to adapt to different scenarios. If the refining furnace type is changed (e.g., from LF furnace to RH furnace) or the steel grade is changed (e.g., from ordinary carbon steel to automotive sheet steel), the above steps S121-S123 need to be repeated to re-determine the k value to adapt to the new scenario through on-site tests to ensure the accuracy of steel slag fluidity compensation. The corrected high-temperature spot density D' = D × [1 + k × v], where v is the flow velocity of the steel slag, in mm / s.

[0013] Furthermore, step S121 specifically includes: a) Adjust key process parameters (such as argon flow rate and slag-forming agent addition) to generate different steel slag flow velocities v (measured in real time by image optical flow method, typically ranging from 2 to 10 mm / s, covering low-speed to high-speed stirring scenarios). b) Synchronously record the original high-temperature spot density D corresponding to the flow velocity v of each group of steel slag (the number of spots with a temperature > 1600℃ per unit area extracted from the image collected by the infrared camera).

[0014] 6. The method for online detection of steel slag composition in a refining process based on an infrared camera according to claim 4, characterized in that, in step S122, specifically: a) During the experiment, steel slag samples were collected at fixed time intervals (30 seconds), and the actual FeO content was detected using X-ray fluorescence spectrometry (XRF) to ensure a detection accuracy of ±0.05%; b) Establish a corresponding dataset for 'slag flow velocity v - original high-temperature spot density D - actual FeO content', ensuring that each set of data comes from the same furnace of slag at the same production time.

[0015] Furthermore, in step S123, the specific details are as follows: a) Assuming an initial value of k (e.g., 0.005, 0.01, 0.015, etc.), calculate D' corresponding to different values ​​of k using the formula D'=D×[1+k×v]. b) Substitute D' corresponding to different k into the FeO calculation model FeO%=a5×D'+b5×G8+c5, compare the calculated FeO value with the actual FeO value detected offline, and select the k value with the smallest deviation between the calculated value and the actual value (≤0.2%). c) Repeat the above process to verify the data of multiple sets of steel slag flow velocity v and original high-temperature spot density D, to ensure that the same k value can meet the FeO calculation accuracy requirements under different steel slag flow velocities v, and finally determine the optimal k value.

[0016] Furthermore, in step S2, the image data processing center uses an industrial control computer (IPC) equipped with image processing software developed based on Python+OpenCV+PyTorch (infrared multi-wavelength processing module) to automate 'infrared image acquisition-preprocessing-multi-feature extraction'. Specifically, it includes three sequentially connected sub-models: image synchronous acquisition model, image preprocessing model, and abnormal image processing model.

[0017] Furthermore, an image preprocessing model is used to eliminate on-site interference, including denoising, arc occlusion correction, and steel slag region extraction. Arc occlusion correction uses an "infrared image segmentation algorithm" to identify arc regions (grayscale value > 200) on the electrode and fills them with the average grayscale value of multiple wavelengths from adjacent non-arc regions to avoid calculating arc interference components. For steel slag region extraction, an "adaptive threshold segmentation algorithm" is used to extract the core area of ​​steel slag. The threshold range is 80-180 to exclude background interference from furnace lining refractory materials, furnace mouth frame (etc.) and ensure that feature extraction is only for steel slag.

[0018] Furthermore, in step S3, based on the preprocessed infrared feature parameters and calibration relationship, a secondary model of "multi-component coupled calculation" is constructed to achieve accurate component calculation and anomaly early warning. The "Steel Slag Composition Calculation and Determination Model" is as follows: S31. Core component calculation: Take the average value of the infrared feature parameters of the three infrared cameras (G8_avg=(G8_side1+G8_side2+G8_front) / 3, G... 10 _avg=(G 10 _Side 1+G 10 _Side 2+G 10 Substituting these values ​​into the calibration model and combining them with steel slag fluidity compensation, the real-time composition is calculated as follows: CaO(t) = a1×G8_avg + b1×S_avg + c1 SiO2(t) = a2×G 10 _avg + b2×D'_avg + c2 Al2O3(t) = a3×S + b3×G8 + c3 MgO(t) = a4×S + b4×G 10 + c4 FeO (%) = a⁵ × D + b⁵ × G⁸ + c⁵ G8_avg: The average G8 value of the three infrared cameras; G 10 _avg: G from three infrared cameras 10 average value; D'_avg: The average value of D' from the three infrared cameras; S_avg: The average S value of the three infrared cameras; S32. Composition Verification: Ensure calculation accuracy through "CaO / SiO2 ratio verification". For example, in pipeline steel refining, CaO / SiO2 should be positively correlated with Al2O3. If the deviation is >0.2, recalculate.

[0019] Due to the adoption of the above technical solution, the beneficial effects achieved by this invention are: This invention involves constructing a high-temperature resistant support structure around the furnace opening (or observation port) of a refining furnace and arranging multiple sets of infrared cameras to simultaneously acquire infrared images of the steel slag surface during the refining process. Through preprocessing such as image denoising, segmentation, and feature extraction (e.g., specific wavelength grayscale values, high-temperature spot density), combined with the calibration relationship between "steel slag infrared features and composition" and steel slag fluidity compensation, the real-time composition of the steel slag is calculated. Finally, a dynamic threshold is used to determine whether the composition is within the target range, triggering an early warning and slag-forming parameter adjustment response. The specific effects are as follows: 1. Real-time and multi-component synchronous detection: 30fps high-frequency acquisition and multi-feature coupling model enable component identification within 2 seconds (5-10 times faster than offline detection), and can simultaneously detect 5 key components such as CaO and SiO2, overcoming the limitations of single-component detection in existing methods; 2. High precision and strong adaptability: The component detection accuracy reaches ±0.2% (CaO / SiO2 ratio accuracy ±0.05). With nitrogen purging and arc correction, the effective data rate is >98% in scenarios with dust concentration ≤30mg / m³ and ambient temperature ≤90℃. 3. Non-contact and long lifespan: The infrared camera has no direct contact with the steel slag, and its service life reaches 24-36 months (more than 100 times longer than that of the resistance electrode method), reducing equipment maintenance costs by 80%. 4. Low cost and flexible adaptation: The cost of a single infrared camera is less than 60,000 yuan (only 5% of that of online X-ray fluorescence equipment). When switching steel types, only the dynamic threshold library needs to be updated (time taken less than 20 minutes). It is compatible with the production of multiple specifications of ordinary carbon steel, low alloy steel, high carbon steel, etc. 5. Intelligentization and cost reduction: The automatic data access to the intelligent refining control system can reduce the unqualified steel slag composition rate by 60% to 80%, reduce the problem of excessive slag-forming agent addition by 40%, and reduce the refining cost per furnace by 1,500 to 3,000 yuan. Attached Figure Description

[0020] Figure 1 This is a flowchart of the present invention.

[0021] Figure 2 This is a hardware layout diagram of the LF refining furnace of the present invention.

[0022] Figure 3 This is a hardware layout diagram of the RH refining furnace of the present invention.

[0023] Figure 2 In the middle: LF refining furnace 1-1, LF refining furnace top main camera 1-2, LF refining furnace side auxiliary camera one 1-3, LF refining furnace side auxiliary camera two 1-4, LF refining furnace top main camera high temperature protective cover (including nitrogen purging device) 1-5, LF refining furnace side auxiliary camera high temperature protective cover (including nitrogen purging device) one 1-6, LF refining furnace side auxiliary camera high temperature protective cover (including nitrogen purging device) two 1-7, LF refining furnace top main camera high temperature protection bracket 1-8, LF refining furnace side auxiliary camera high temperature protection bracket one 1-9, LF refining furnace side auxiliary camera high temperature protection bracket two 1-10.

[0024] Figure 3In the middle: RH refining furnace 2-1, RH refining furnace top main camera 2-2, RH refining furnace side auxiliary camera one 2-3, RH refining furnace side auxiliary camera two 2-4, RH refining furnace top main camera high temperature protective cover (including nitrogen purging device) 2-5, RH refining furnace side auxiliary camera high temperature protective cover (including nitrogen purging device) one 2-6, RH refining furnace side auxiliary camera high temperature protective cover (including nitrogen purging device) two 2-7, RH refining furnace side auxiliary camera high temperature protection bracket one 2-8, RH refining furnace side auxiliary camera high temperature protection bracket two 2-9, RH refining vacuum chamber 2-10. Detailed Implementation

[0025] The present invention will be further described in detail below with reference to the embodiments.

[0026] A method for online detection of steel slag composition in refining processes, relying on infrared cameras, such as... Figure 1 The core idea of ​​this invention is as follows: Multiple sets of infrared cameras are deployed around the furnace opening (or observation hole) of either the LF refining furnace 1-1 or the RH refining furnace 2-1. These cameras include: LF refining furnace main camera anti-high temperature bracket 1-8; LF refining furnace side auxiliary camera anti-high temperature bracket one 1-9; LF refining furnace side auxiliary camera anti-high temperature bracket two 1-10; or RH refining furnace side auxiliary camera anti-high temperature bracket one 2-8; RH refining furnace side auxiliary camera anti-high temperature bracket two 2-9. Simultaneously, infrared images of the steel slag surface are acquired during the refining process. These images are then pre-processed using image denoising, segmentation, and feature extraction (e.g., specific wavelength grayscale value, high-temperature spot density), combined with "steel slag infrared characteristics - composition". The calibration relationship and steel slag fluidity compensation are used to calculate the real-time composition of the steel slag. Finally, a dynamic threshold is used to determine whether the composition is within the target range, triggering an early warning and slag-forming parameter adjustment response. The specific steps are as follows: Step 1: Deployment and Calibration of Infrared Camera System (1) Camera installation location and layout Differentiated layouts are designed based on the type of refining furnace (LF furnace / RH furnace) to ensure no blind spots: 1) Layout of LF refining furnace 1-1: Build brackets on both sides of the electrode hole and in front of the furnace opening: LF refining furnace main camera high temperature protection bracket 1-8, LF refining furnace side auxiliary camera high temperature protection bracket one 1-9, LF refining furnace side auxiliary camera high temperature protection bracket two 1-10, 1.2~1.8m away from the edge of the furnace opening, avoiding the direct area of ​​electrode arc light, forming a "triangular perspective"; Front camera: 1-2 main cameras above the LF refining furnace, with the lens axis parallel to the molten steel slag surface (1.5-2.0m above the surface), used to capture the dynamic infrared characteristics of the molten steel slag flow; Side cameras: LF refining furnace side auxiliary camera 1-3 and LF refining furnace side auxiliary camera 2-4, a total of 2 units, with the lens axis at a 45° angle to the molten steel slag surface, used to capture the radial infrared radiation characteristics of the steel slag; 2) Layout of RH refining furnace 2-1: One main camera 2-2 above the RH refining furnace is installed on the top of the RH refining vacuum chamber 2-10. One camera 2-3 and one auxiliary camera 2-4 are installed on each side of the observation window of the RH refining vacuum chamber 2-10. The lens axis points to the surface of the steel slag. 3) Installation Requirements: High-temperature resistant protective covers must be installed on the exterior of the following cameras: LF refining furnace main camera 1-2, LF refining furnace side auxiliary camera 1-3, LF refining furnace side auxiliary camera 2-4, and RH refining furnace main camera 2-2, RH refining furnace side auxiliary camera 1-3, and RH refining furnace side auxiliary camera 2-4. Specifically, the following covers must be installed: LF refining furnace main camera high-temperature protective cover (including nitrogen purging device) 1-5, LF refining furnace side auxiliary camera high-temperature protective cover (including nitrogen purging device) 1-6, LF refining furnace side auxiliary camera high-temperature protective cover (including nitrogen purging device) 2-7, and RH refining furnace main camera high-temperature protective cover (including nitrogen purging device) 2-5, RH refining furnace side auxiliary camera high-temperature protective cover (including nitrogen purging device) 2-6, and RH refining furnace side auxiliary camera high-temperature protective cover (including nitrogen purging device) 2-7. These covers must have a temperature resistance ≥150℃ and be made of 316L stainless steel. Stainless steel components all include a nitrogen purging system with a pressure of 0.4-0.6 MPa and a flow rate of 3-5 m³ / h. 3 / h, to prevent steel slag dust from adhering to the lens), the high temperature protection brackets 1-8 for the main camera above the LF refining furnace, the high temperature protection bracket 1-9 for the auxiliary camera on the side of the LF refining furnace, the high temperature protection bracket 2-10 for the auxiliary camera on the side of the LF refining furnace, and the high temperature protection bracket 2-8 for the auxiliary camera on the side of the RH refining furnace, and the high temperature protection bracket 2-9 for the auxiliary camera on the side of the RH refining furnace adopt an anti-vibration design (vibration control ≤0.05g, to avoid electrode vibration affecting image clarity).

[0027] (2) Camera selection parameters A high-resolution infrared thermal imager (belonging to the category of infrared cameras) must be selected to meet the requirements for detecting refined steel slag. 1) Infrared resolution: ≥640×512 pixels (to ensure the accuracy of infrared feature recognition corresponding to subtle differences in the composition of steel slag); 2) Frame rate: ≥30fps (1 frame is captured every 0.033 seconds to meet the response requirement within 2 seconds); 3) Spectral range: 7-14μm (this band can distinguish the differences in infrared radiation of steel slag with different compositions and has strong ability to penetrate flue gas). 4) Temperature measurement range: 300-1700℃ (covering the temperature range of refined steel slag, the temperature of steel slag is usually 1500-1600℃), temperature measurement accuracy ±1℃; 5) Environmental adaptability: Operating temperature -30-90℃, protection level IP68 (dustproof and waterproof), electromagnetic interference resistance (compliant with industrial EMC standards, resists strong electromagnetic radiation). 6) Data interface: Gigabit Ethernet (GigE Vision protocol), supporting synchronous transmission of infrared images and multi-wavelength grayscale data.

[0028] (3) System calibration (establishing the mapping relationship between "infrared characteristics of steel slag and composition") 1) Cold calibration: When LF refining furnace 1-1 or RH refining furnace 2-1 is shut down, place standard steel slag samples (composition covering the target range: CaO 40%~55%, SiO2 10%~20%, Al2O3 3%~8%, MgO 5%~10%, FeO 2%~8%, accuracy ±0.05%) in the simulation device of LF refining furnace 1-1 and RH refining furnace 2-1, and simultaneously acquire infrared images from 3 cameras; ① Extracting multi-dimensional infrared feature parameters: G8: The average gray value of steel slag at a wavelength of 8μm, corresponding to the CaO content. The higher the CaO content, the larger the G8. G 10 The average gray value of steel slag at a wavelength of 10μm corresponds to the SiO2 content. The higher the SiO2 content, the higher the gray value. 10 The smaller; D: High-temperature spot density (the number of spots per unit area with a temperature >1600℃, corresponding to the FeO content; the higher the FeO content, the larger D). The specific calculation method is as follows: D = Total number of high-temperature spots N / Actual area of ​​steel slag region (unit: spots / m²) 2 ) S: Complexity of slag flow texture (a quantitative characteristic parameter characterizing the complexity of the surface texture morphology (such as texture thickness, distribution uniformity, and frequency of change) during the slag flow process in the refining furnace. It is the "surface manifestation" of the combined effect of the internal components of the slag (especially Al2O3 and MgO) and external flow conditions (such as argon stirring intensity and slag temperature). It is calculated through image grayscale gradient and corresponds to the Al2O3 and MgO content. ② By using linear fitting, an initial mapping model between the feature parameters and the standard components is established, namely: CaO (%) = a1×G8 + b1×S + c1 SiO2 (%) = a2 × G 10 + b2×D + c2 Al2O3 (%) = a3×S + b3×G8 + c3 MgO (%) = a⁴ × S + b⁴ × G 10 + c4 FeO (%) = a5×D + b5×G8 + c5 Where: a1-a5, b1-b5, and c1-c5 are calibration coefficients, which are obtained by fitting experimental data of standard steel slag samples; 2) Steel slag fluidity compensation: The steel slag flow velocity v (unit: mm / s) is calculated by image optical flow method (i.e., by analyzing the motion trajectory of the steel slag surface pixels in continuous frame infrared images to calculate the actual flow velocity of the steel slag), and the infrared characteristic parameters are corrected (too fast steel slag flow will cause uneven infrared radiation): after correction, the spot density D'=D×[1 + k×v].

[0029] Wherein, k is the compensation coefficient, obtained through on-site testing combined with verification of actual composition. The core is to establish a matching relationship between the steel slag flow velocity and the spot density correction value under the actual refining furnace production scenario. The specific steps for obtaining it are as follows: ① Determine the test scenario and parameter range The test subjects were target refining furnaces (such as LF furnaces and RH furnaces) and corresponding steel grades, covering common steel slag flow conditions in actual production: a) Adjust key process parameters (such as argon flow rate and slag-forming agent addition) to generate different steel slag flow velocities v (measured in real time by image optical flow method, typically ranging from 2 to 10 mm / s, covering low-speed to high-speed stirring scenarios). b) Synchronously record the original spot density D corresponding to each group v (the number of spots with a temperature > 1600℃ per unit area extracted from the image acquired by the infrared camera).

[0030] ② Collect actual component data for calibration. For each set of test conditions, the actual FeO content of the steel slag was obtained through offline precise detection as a verification standard: a) During the experiment, steel slag samples were collected at fixed time intervals (30 seconds), and the actual FeO content was detected using X-ray fluorescence spectrometry (XRF) to ensure a detection accuracy of ±0.05%; b) Establish a corresponding dataset for "slag flow velocity v - original spot density D - actual FeO content" to ensure that each set of data comes from the same furnace of slag at the same production time.

[0031] ③ Establish the correction relationship and inversely deduce the k value. Through data correlation analysis, the k value that matches the corrected spot density D' with the actual FeO content was determined: a) Assuming an initial value of k (e.g., 0.005, 0.01, 0.015, etc.), calculate D' corresponding to different values ​​of k using the formula D'=D×[1+k×v]. b) Substitute D' corresponding to different k into the FeO calculation model (FeO%=a5×D'+b5×G8 + c5), compare the calculated FeO value with the actual FeO value detected offline, and select the k value with the smallest deviation between the calculated value and the actual value (≤0.2%). Repeat the above process to verify multiple sets of v and D data, ensuring that the same k value can meet the FeO calculation accuracy requirements under different flow rates, and finally determine the optimal k value.

[0032] ④ k-value optimization to adapt to different scenarios If the type of refining furnace is changed (e.g., from LF furnace to RH furnace) or the steel grade is changed (e.g., from ordinary carbon steel to automotive sheet steel), steps ①-③ above need to be repeated to re-determine the k value suitable for the new scenario through on-site testing to ensure the accuracy of steel slag fluidity compensation.

[0033] Step 2: Construction of the Image Data Processing Center The on-site data processing center uses an industrial control computer (IPC) equipped with image processing software developed based on Python + OpenCV + PyTorch (infrared multi-wavelength processing module) to automate the "infrared image acquisition - preprocessing - multi-feature extraction" process, comprising three sub-models: (1) Image synchronous acquisition model 1) Acquisition frequency: 30fps, simultaneously acquiring image data from 3 infrared cameras: main camera 1-2 above the LF refining furnace, auxiliary camera 1-3 on the side of the LF refining furnace, and auxiliary camera 2-4 on the side of the LF refining furnace, or main camera 2-2 above the RH refining furnace, auxiliary camera 1-3 on the side of the RH refining furnace, and auxiliary camera 2-4 on the side of the RH refining furnace (denoted as I_side1(t), I_side2(t), I_front(t), where t is time). At the same time, process parameters (such as electrode current, argon flow rate, and slagging agent addition amount) are read from the PLC of LF refining furnace 1-1 or RH refining furnace 2-1 via OPC UA protocol. 2) Data frame format: "Timestamp + Camera number + Multi-wavelength infrared image data (TIFF compression) + Steel slag temperature matrix + Process parameters + Check digit", to ensure data traceability; 3) Storage strategy: Real-time images and feature data are cached in memory (retaining the most recent 20 seconds of data). Every 1 minute, the processed component data and images at abnormal moments are compressed and stored in the industrial database for a retention period of ≥6 months.

[0034] (2) Image preprocessing model (eliminating on-site interference) 1) Denoising: A combined algorithm of "bilateral filtering + wavelet threshold denoising" (filter kernel size 7×7) is used to eliminate image noise caused by electrode arc light and flue gas dust, while preserving the texture details of the steel slag surface; 2) Arc light occlusion correction: The arc light region of the electrode (gray value > 200) is identified by the "infrared image segmentation algorithm" and filled with the average gray value of multiple wavelengths of the adjacent non-arc light region to avoid the calculation of arc light interference components; 3) Steel slag region extraction: The core area of ​​steel slag is extracted by "adaptive threshold segmentation algorithm" (threshold range 80-180) to eliminate background interference such as furnace lining refractory material and furnace mouth frame, and to ensure that feature extraction is only for steel slag.

[0035] (3) Abnormal image processing model 1) Image blur determination: Calculate the "grayscale variance" (standard deviation of grayscale values ​​in the steel slag area) of each frame of the image. If the variance is less than 20, it is determined to be blurry (lens contamination or smoke obstruction). If there are 5 consecutive blurry frames, trigger the nitrogen purging device for strong purging (pressure increased to 0.6MPa, lasting for 20 seconds). During this period, the average component value of the previous 3 valid images is used to replace the original value. 2) Camera Fault Diagnosis: If a camera (LF refining furnace main camera 1-2, LF refining furnace side auxiliary camera 1-3, LF refining furnace side auxiliary camera 2-4 or RH refining furnace main camera 2-2, RH refining furnace side auxiliary camera 1-3, RH refining furnace side auxiliary camera 2-4) has no data output for 20 consecutive seconds, or the data transmission error rate is >10%, a fault alarm will be triggered (audio and visual prompts in the central control room), the data of that camera will be marked as invalid, and the "cross calculation" of the other two cameras will be enabled (e.g., if the side camera fails, all components will be calculated through the G8 and D parameters of the front camera).

[0036] Step 3: Steel slag composition calculation and determination model Based on the preprocessed infrared feature parameters and calibration relationship, a two-level model of "multi-component coupled calculation" is constructed to achieve accurate component calculation and anomaly early warning. (1) Core component calculation: Take the average value of the infrared characteristic parameters of the three cameras: the main camera 1-2 above the LF refining furnace, the auxiliary camera 1-3 on the side of the LF refining furnace, and the auxiliary camera 2 on the side of the LF refining furnace, or the main camera 2-2 above the RH refining furnace, the auxiliary camera 1-3 on the side of the RH refining furnace, and the auxiliary camera 2 on the side of the RH refining furnace (G8_avg=(G8_side1+G8_side2+G8_front) / 3, G 10 _avg=(G 10 _Side 1+G 10 _Side 2+G 10 Substituting these values ​​into the calibration model and combining them with steel slag fluidity compensation, the real-time composition is calculated as follows: CaO(t) = a1×G8_avg + b1×S_avg + c1 SiO2(t) = a2×G 10 _avg + b2×D'_avg + c2 Al2O3(t) = a3×S + b3×G8 + c3 MgO(t) = a4×S + b4×G 10 + c4 FeO (%) = a⁵ × D + b⁵ × G⁸ + c⁵ (2) Composition verification: Ensure the accuracy of the calculation by “CaO / SiO2 ratio verification” (e.g., in pipeline steel refining, CaO / SiO2 should be positively correlated with Al2O3, and recalculate if the deviation is >0.2). Step 4: Output of Detection Results and Execution Response After the model completes component calculations and anomaly detection, it outputs results through multiple channels and triggers automated responses: (1) Real-time display Dynamic display in the central control room HMI (Human-Machine Interface): 1) Infrared thermal imaging image of steel slag (labeling the values ​​of each component and the CaO / SiO2 ratio, using colors to distinguish the qualified status: green = qualified, yellow = first-level warning, red = second-level warning); 2) Real-time composition curves (the changing trends of CaO, SiO2, and FeO over the past 15 minutes) and dynamic threshold range; 3) Cameras: Status of the main camera 1-2 above the LF refining furnace, auxiliary camera 1-3 on the side of the LF refining furnace, and auxiliary camera 2-4 on the side of the LF refining furnace, or the main camera 2-2 above the RH refining furnace, auxiliary camera 1-3 on the side of the RH refining furnace, and auxiliary camera 2-4 on the side of the RH refining furnace (normal / fault / purging in progress) and key refining parameters (electrode current, argon flow rate, slagging agent addition).

[0037] (2) Steel slag composition alarm An alarm will be triggered to the on-site operators if the steel slag content exceeds the threshold requirement.

[0038] Specifically: when the steel slag content begins to exceed the threshold requirement range, the model first issues a level one yellow alarm; when the steel slag content exceeds the threshold requirement range for more than 10 seconds, a level two red alarm is issued.

[0039] (3) Data traceability Automatically generates daily / weekly steel slag composition test reports, including average composition values, number of anomalies, early warning handling records, and infrared images of key anomalies, and supports exporting to Excel / PDF formats; the report data can be integrated into the enterprise MES system for refining process optimization (such as adjusting the argon stirring intensity based on FeO fluctuation trends to reduce the oxidizability of molten steel).

[0040] The following is a detailed explanation of the implementation steps, using a case study of refining X80 pipeline steel in a 260t LF furnace at a steel plant: Step 1: Deployment and Calibration of Infrared Camera System (1) Camera installation: 1) Layout: Support brackets are erected on both sides of the electrode hole of LF refining furnace 1-1 and directly in front of the furnace opening. The main camera above the LF refining furnace is protected against high temperatures (1-8); the auxiliary cameras on the sides of the LF refining furnace are protected against high temperatures (1-9); and the auxiliary cameras on the sides of the LF refining furnace are protected against high temperatures (1-10) (1.5m from the edge of the furnace opening). Two FLIR A6750 infrared cameras are installed on the sides: Auxiliary Camera 1 on the Side of the LF Refining Furnace (1-3) and Auxiliary Camera 2 on the Side of the LF Refining Furnace (1-4) (640×512 resolution, 30fps). One camera is installed in front. The main cameras 1-2 above the LF refining furnace (1.8m above the slag surface); the high-temperature protective covers (including nitrogen purging devices) 1-5 for the main cameras above the LF refining furnace; the high-temperature protective covers (including nitrogen purging devices) 1-6 for the auxiliary cameras on the side of the LF refining furnace; and the high-temperature protective covers (including nitrogen purging devices) 1-7 for the auxiliary cameras on the side of the LF refining furnace are all made of 316L stainless steel, with a nitrogen purging pressure of 0.5MPa and a flow rate of 4m³ / h. 3 / h; 2) Calibration: Place a standard steel slag sample (CaO 43.5%, SiO2 15.0%, Al2O3 4.5%, MgO 7.0%, FeO 4.0%), acquire images to obtain G8=120, G... 10 =85, D=12, S=18; Fit calibration coefficients: CaO model: a1=0.08, b1=0.02, c1=35.2; SiO2 model: a2=-0.06, b2=0.03, c2=20.5; Al2O3 model: a3=0.045, b3=0.01, c3=2.5; MgO model: a4=0.085, b4=-0.02, c4=7.2; FeO model: a5=0.05, b5=0.01, c5=3.0; The liquidity compensation coefficient k = 0.01.

[0041] (2) Flowability compensation: During production, the flow velocity of steel slag is v=6mm / s, and the corrected spot density is D'=12×[1+0.01×6]=12.72.

[0042] Step 2: Construction of the Image Data Processing Center (1) IPC configuration: Intel Core i9-13900K processor, 64GB memory, 4TB solid-state drive, Windows Server 2022 operating system installed, image processing software developed based on Python+OpenCV+PyTorch; (2) Image preprocessing: The image of a certain frame of the side camera (LF refining furnace side auxiliary camera 1-3, LF refining furnace side auxiliary camera 2-4) contains the electrode arc light (gray value 210). After being filled by the segmentation algorithm, the steel slag area (gray value 80-180) is extracted; the gray variance is increased from 15 to 35 by bilateral filtering for noise reduction. (3) Abnormal handling: The grayscale variance of the five consecutive frames of images from the main camera 1-2 above the LF refining furnace is 18 (<20), which is determined to be lens contamination. A strong purge (0.6MPa, 20 seconds) is triggered, during which the average composition of the first three frames (CaO=43.2%, SiO2=15.1%, FeO=4.2%) is used to replace the images.

[0043] Step 3: Calculation and determination of steel slag composition Composition calculation: Average feature parameters of 3 cameras (main cameras 1-2 above the LF refining furnace, auxiliary cameras 1-3 on the side of the LF refining furnace, and auxiliary cameras 2-4 on the side of the LF refining furnace): G8_avg=122, G 10 _avg=84, D'_avg=13, S_avg=19; CaO(t) = 0.08 × 122 + 0.02 × 19 + 35.2 = 43.8% (within the required range of 42% to 45%, which is acceptable). SiO2(t) = -0.06 × 84 + 0.03 × 13 + 20.5 = 15.3% (within the required range of 3.5% to 5.0%, which is acceptable). CaO / SiO2 = 43.8 / 15.3 ≈ 2.86 (within the required range of 2.6% to 2.9%, which is acceptable). Al2O3(t) = 0.045×19 + 0.01×122 + 2.5 = 0.855 + 1.22 + 2.5 = 4.575% ≈ 4.58% (within the required range of 3.5% to 5.0%, which is acceptable). MgO (t) = 0.085 × 19 + (-0.02) × 84 + 7.2 = 1.615 - 1.68 + 7.2 = 7.135% ≈ 7.14% (within the required range of 6.5% to 8.5%, which is acceptable). FeO (t) = 0.05 × 13 + 0.01 × 122 + 3.0 = 4.85% (exceeding the threshold of 4.5%). (2) Anomaly detection: FeO=4.85% exceeds the threshold, triggering a level 1 warning; if FeO is still 4.82% after 10 seconds, triggering a level 2 warning.

[0044] Step 4: Execute the response HMI display: The thermal imaging chart shows FeO = 4.85% (red), with other components in green; the FeO curve shows an upward trend. Alarm: The red indicator light stays on, an audible and visual alarm is triggered on-site, and a notification is sent to the operator.

Claims

1. A method for online detection of steel slag composition in a refining process using an infrared camera, characterized in that, Includes the following steps: S1. The system deployment and calibration of infrared cameras should be designed with differentiated layouts according to the type of refining furnace to ensure no blind spots and establish a mapping relationship between 'infrared characteristics of steel slag and composition'. S2. The image data processing center is built to automate the "infrared image acquisition - preprocessing - multi-feature extraction" process. S3, a steel slag composition calculation and judgment model, enables accurate calculation of steel slag composition and early warning of anomalies; S4. Output of detection results and execution response.

2. The method for online detection of steel slag composition in a refining process based on an infrared camera, as described in claim 1, is characterized in that... In step S1, three infrared cameras are used, each equipped with a high-temperature resistant protective cover. The protective cover includes a nitrogen purging device. The high-temperature resistant protective cover has a temperature resistance of ≥150℃ and is made of 316L stainless steel. The nitrogen purging device has a pressure of 0.4-0.6MPa and a flow rate of 3-5m³ / h. 3 / h, which is used to prevent steel slag dust from adhering to the lens; In step S1, "establishing the mapping relationship between 'infrared characteristics of steel slag' and 'composition'" specifically refers to: S11. Cold calibration: When the refining furnace is shut down, a standard steel slag sample is placed in the matching refining furnace simulation device, and infrared images from three infrared cameras are collected simultaneously. Multi-dimensional infrared feature parameters are extracted, and an initial mapping model between the feature parameters and the standard composition is established through linear fitting. S12. Steel slag fluidity compensation: The steel slag flow velocity is calculated by image optical flow method, the infrared characteristic parameters are corrected, and a matching relationship between the steel slag flow velocity and the high temperature spot density correction value is established.

3. The method for online detection of steel slag composition in a refining process based on an infrared camera, as described in claim 2, is characterized in that... In step S11, "extracting multi-dimensional infrared feature parameters" specifically includes: G8: The average gray value of steel slag at a wavelength of 8μm, corresponding to the CaO content. The higher the CaO content, the larger the G8. G 10 The average gray value of steel slag at a wavelength of 10μm corresponds to the SiO2 content. The higher the SiO2 content, the higher the gray value. 10 The smaller; D: Original high-temperature spot density, the number of spots with a temperature >1600℃ per unit area, corresponding to the FeO content. The higher the FeO content, the larger D is. The specific calculation method is as follows: D = Total number of original high-temperature spots N / Actual area of ​​steel slag region, unit: spots / m² 2 ; S: A quantitative characteristic parameter of the complexity of steel slag flow texture, which is the "surface manifestation" of the combined effect of the internal components of steel slag and the external flow conditions. It is calculated by image grayscale gradient and corresponds to the content of Al2O3 and MgO. "Establishing an initial mapping model between feature parameters and standard components through linear fitting" specifically includes: CaO (%) = a1×G8 + b1×S + c1 SiO2(%)=a2×G 10 + b2×D + c2 Al2O3 (%) = a3×S + b3×G8 + c3 MgO(%)=a4×S + b4×G 10 + c4 FeO (%) = a5×D + b5×G8 + c5 Where: a1-a5, b1-b5, and c1-c5 are calibration coefficients, which are obtained by fitting experimental data of standard steel slag samples.

4. The method for online detection of steel slag composition in a refining process based on an infrared camera according to claim 2, characterized in that, In step S12, "steel slag fluidity compensation" specifically refers to: S121. Determine the test scenario and parameter range to cover common steel slag flow conditions in actual production. S122. Collect actual component data for calibration. For each set of test conditions, obtain the actual FeO content of steel slag through offline precise detection as a verification standard. S123. Establish the correction relationship and back-calculate the k value. Through data correlation analysis, determine the k value that matches the corrected high-temperature spot density D' with the actual FeO content. Here, k is the compensation coefficient, which is obtained by combining field tests with actual component verification. S124. Optimize the k value to suit different scenarios. If the refining furnace type or steel grade is changed, the above steps S121-S123 need to be repeated to re-determine the k value to suit the new scenario through on-site testing to ensure the accuracy of steel slag fluidity compensation. The corrected high-temperature spot density D' = D × [1 + k × v], where v is the flow velocity of the steel slag, in mm / s.

5. The method for online detection of steel slag composition in a refining process based on an infrared camera, as described in claim 4, is characterized in that... Step S121 is as follows: a) Adjust key process parameters to generate different steel slag flow velocities v; b) Synchronously record the original high-temperature spot density D corresponding to the flow velocity v of each group of steel slag.

6. The method for online detection of steel slag composition in a refining process based on an infrared camera, as described in claim 4, is characterized in that... In step S122, specifically: a) During the experiment, steel slag samples were collected at fixed time intervals, and the actual FeO content was detected using an X-ray fluorescence spectrometer to ensure a detection accuracy of ±0.05%; b) Establish a corresponding dataset for 'slag flow velocity v - original high-temperature spot density D - actual FeO content', ensuring that each set of data comes from the same furnace of slag at the same production time.

7. The method for online detection of steel slag composition in a refining process based on an infrared camera according to claim 4, characterized in that, In step S123, specifically: a) Assuming an initial value for k, calculate D' for different values ​​of k using the formula D'=D×[1+k×v]; b) Substitute D' corresponding to different k into the FeO calculation model FeO%=a5×D'+b5×G8+c5, compare the calculated FeO value with the actual FeO value detected offline, and select the k value with the smallest deviation between the calculated value and the actual value. c) Repeat the above process to verify the data of multiple sets of steel slag flow velocity v and original high-temperature spot density D, to ensure that the same k value can meet the FeO calculation accuracy requirements under different steel slag flow velocities v, and finally determine the optimal k value.

8. The method for online detection of steel slag composition in refining process based on an infrared camera according to claim 1, characterized in that, In step S2, the image data processing center uses an industrial control computer equipped with image processing software developed based on Python+OpenCV+PyTorch to automate the process of 'infrared image acquisition-preprocessing-multi-feature extraction'. Specifically, it includes three sequentially connected sub-models: image synchronous acquisition model, image preprocessing model, and abnormal image processing model.

9. The method for online detection of steel slag composition in a refining process based on an infrared camera according to claim 1, characterized in that, Image preprocessing models are used to eliminate on-site interference, including noise reduction, arc occlusion correction, and steel slag region extraction. Arc occlusion correction identifies the arc region of the electrode through an "infrared image segmentation algorithm" and fills it with the average gray value of multiple wavelengths of adjacent non-arc regions to avoid the calculation of arc interference components. For steel slag region extraction, an "adaptive threshold segmentation algorithm" is used to extract the core area of ​​steel slag. The threshold range is 80-180 to exclude background interference from furnace lining refractory materials and furnace mouth frame, ensuring that feature extraction is only for steel slag.

10. The method for online detection of steel slag composition in a refining process based on an infrared camera according to claim 3, characterized in that, In step S3, based on the preprocessed infrared feature parameters and calibration relationship, a secondary model of "multi-component coupled calculation" is constructed to achieve accurate component calculation and anomaly early warning. The "Steel Slag Composition Calculation and Determination Model" is as follows: S31. Core component calculation: Take the average value of the infrared feature parameters of the three infrared cameras (G8_avg=(G8_side1+G8_side2+G8_front) / 3, G... 10 _avg=(G 10 _Side 1+G 10 _Side 2+G 10 Substituting these values ​​into the calibration model and combining them with steel slag fluidity compensation, the real-time composition is calculated as follows: CaO(t) = a1×G8_avg + b1×S_avg + c1 SiO2(t) = a2×G 10 _avg + b2×D’_avg + c2 Al2O3(t) = a3×S + b3×G8 + c3 MgO(t) = a4×S + b4×G 10 + c4 FeO (%) = a⁵ × D + b⁵ × G⁸ + c⁵ G8_avg: The average G8 value of the three infrared cameras; G 10 _avg: G from three infrared cameras 10 average value; D'_avg: The average value of D' from the three infrared cameras; S_avg: The average S value of the three infrared cameras; S32. Composition Verification: Ensure calculation accuracy through "CaO / SiO2 ratio verification".

Images