A method and system for on-line real-time dynamic detection of carbon content of converter steel

Abstract

This application relates to the technical field of carbon detection, and more particularly to an online real-time dynamic detection method for carbon content in molten steel produced in converter steelmaking. The method includes acquiring multi-source sensor data, including acquiring surface temperature images of molten steel collected by an infrared thermal imager; acquiring multi-band spectral data, including controlling a visible band sensor to acquire radiation intensity in the 500-600 nm range and controlling a near-infrared band sensor to acquire reflectivity characteristics in the 850-1100 nm range; analyzing and fusing the data, analyzing the temperature images to generate a two-dimensional temperature field of the molten pool; analyzing the multi-band spectral data to extract spectral feature vectors related to carbon content; fusing the two-dimensional temperature field and spectral feature vectors, and generating a carbon content distribution map on the molten steel surface based on a pre-trained carbon content machine learning model; and outputting optimized detection results by processing multi-node data through a distributed computing framework to output a dynamic carbon content detection report. This application achieves breakthroughs overcoming the limitations of traditional static detection, providing real-time component monitoring, full molten pool coverage, and dynamic feedback.

Description

Technical Field

[0001] This application relates to the technical field of carbon detection, and in particular to an online real-time dynamic detection method and system for carbon content in molten steel produced in converter steelmaking. Background Technology

[0002] In the steelmaking process, precise control of the carbon content in molten steel directly determines the final steel quality and smelting efficiency. As the mainstream production process, converter steelmaking's dynamic distribution of carbon elements within the molten pool directly affects the decarburization reaction process and its endpoint control.

[0003] Currently, the industry generally relies on intermittent manual sampling and offline laboratory analysis. Operators use a sampling gun to extract molten steel samples from specific locations in the converter pool, and after cooling, send them to the laboratory for chemical composition analysis.

[0004] However, this scheme has significant drawbacks. For example, single-point sampling cannot reflect the spatial distribution of carbon content in the entire molten pool, especially in dynamic flow areas, and the sampling frequency is limited by manual operation. Secondly, in high-temperature and dusty environments, accompanied by violent agitation of molten steel, the sensor is difficult to operate stably for a long time, making the existing probe prone to failure due to thermal stress or signal drift. In addition, it takes 3-5 minutes from sampling to result feedback, which cannot support rapid and accurate carbon control at the end of converter blowing.

[0005] Based on the shortcomings of the above-mentioned solutions, the current related technologies are unable to meet the core requirements of the modern steel industry for real-time composition monitoring, full molten pool coverage, and dynamic feedback. In particular, the lack of effective online detection methods has become a key bottleneck restricting intelligent steelmaking, especially in the high-temperature environment of violently agitated molten steel. Summary of the Invention

[0006] To address at least one of the aforementioned technical problems, this application provides a method and system for online real-time dynamic detection of carbon content in molten steel produced in converter steelmaking.

[0007] In a first aspect, this application provides a method for online real-time dynamic detection of carbon content in molten steel produced in converter steelmaking, employing the following technical solution, including the following steps:

[0008] S1. Acquire multi-source sensor data, specifically including:

[0009] S101. Acquire the surface temperature image of molten steel collected by an infrared thermal imager, wherein the infrared thermal imager is located above the converter;

[0010] S102. Acquire multi-band spectral data, including controlling the visible band sensor to acquire radiation intensity in the range of 500-600nm and controlling the near-infrared band sensor to acquire reflectivity characteristics in the range of 850-1100nm.

[0011] S2. Data parsing and fusion, specifically including:

[0012] S201. Analyze the temperature image to generate a two-dimensional temperature field of the molten pool;

[0013] S202. Analyze the multi-band spectral data and extract the spectral feature vectors related to carbon content;

[0014] S203. By fusing the two-dimensional temperature field and spectral feature vector, a carbon content distribution map of the molten steel surface is generated based on a pre-trained carbon content machine learning model.

[0015] S3. Output optimized detection results, specifically including:

[0016] S301. Process multi-node data through a distributed computing framework and output a dynamic carbon content detection report.

[0017] By employing the aforementioned technical solution, infrared thermal imaging technology is integrated with multi-band spectral analysis technology. A pre-trained machine learning model is used to process and correlate these two different types of physical information, ultimately generating a carbon content distribution map of the molten steel surface. This method completely changes the traditional, lagging mode that relies on manual sampling and offline analysis, achieving non-contact, online, and full-surface carbon content detection. Through multi-source information fusion, the shortcomings of single detection methods—insufficient dimensionality and susceptibility to interference—are significantly overcome, greatly improving the representativeness, accuracy, and reliability of the detection results. This provides real-time and intuitive data support for the endpoint control of converter steelmaking, thereby improving production efficiency and product quality.

[0018] In one possible implementation, S1 includes:

[0019] Multi-band spectral data is acquired using an optical probe positioned above the converter, which integrates visible and near-infrared sensors.

[0020] By adopting the above technical solution, visible light and near-infrared sensors are integrated into a single optical probe, enabling synchronous and simultaneous acquisition of data from two key spectral bands. This design ensures the temporal and spatial consistency of multi-band spectral data, avoiding problems such as signal asynchrony and field-of-view misalignment caused by separate sensor placement. This provides a high-quality data foundation for subsequent data fusion and further improves the accuracy and reliability of the carbon content inversion model.

[0021] In one possible implementation, the step S2 is followed by the following steps:

[0022] S204. Monitor the environmental status and obtain real-time temperature data of the optical probe's sealed chamber.

[0023] S205. Implement active protection: when the temperature is ≥300℃, activate the water cooling circulation and compressed air curtain; when the temperature is ≥400℃, activate the backup cooling circuit.

[0024] By adopting the above technical solution, a graded active protection mechanism based on temperature threshold triggering is established, rather than a single or low-threshold protection strategy. This graded protection strategy is highly practical for industrial applications. It ensures effective protection of the probe under most operating conditions and can activate stronger cooling under extreme conditions, significantly improving the stability and durability of the detection system in the long-term operation of the high-temperature, high-dust converter environment, and solving the common problem of online detection equipment being unable to operate reliably for a long time in extreme industrial environments.

[0025] In one possible implementation, S3 further includes the following steps:

[0026] S302. Accelerate data processing by compressing computation through model pruning techniques, reducing response time to less than 20ms.

[0027] The S303 is a parallel optimization node that uses a multi-core processor to process data from multiple sensors simultaneously.

[0028] By adopting the above technical solution, model pruning and multi-core parallel processing were introduced to accelerate the data processing stage. The total response time from data acquisition to result output was controlled within 20 milliseconds, achieving true real-time dynamic detection. This speed meets the extreme requirements of the control system for rapid carbon content changes at the end of converter blowing, breaking through the data processing speed bottleneck and enabling online analysis results to be directly used for closed-loop real-time control of the production process.

[0029] In one possible implementation, the step S1 is followed by the following steps:

[0030] S103. Acquire multi-channel spectral data at time intervals using an FPGA parallel architecture;

[0031] S104. Use a hardwired filter to deduct background radiated noise in real time;

[0032] S105. Control the effective signal extraction delay to within 5ms.

[0033] By adopting the above technical solution, the hardware parallel architecture of the FPGA chip and hardwired filters are used to process the raw spectral signal at the lowest level. Utilizing the hardware parallel characteristics of the FPGA, synchronous high-speed acquisition and noise suppression of multi-channel data are achieved from the source, keeping the signal preprocessing latency extremely low, within 5ms. This allows ample time for subsequent complex algorithm processing, ensuring the real-time foundation of the entire system at the hardware level, and greatly enhancing the system's anti-interference capability and signal stability in industrial environments with strong electromagnetic fields and high radiation.

[0034] In one possible implementation, the following is included before S3:

[0035] S304, control the rotary optical probe to centrifuge at a speed of 1200° / min to remove dust, and use pulsed compressed air to blow it;

[0036] S305, generate an argon curtain of 0.1MPa to 0.15MPa to isolate metal vapor;

[0037] S306. When the optical transmittance is <95% or the signal-to-noise ratio of a specific band is lower than the preset threshold, the self-cleaning procedure is triggered.

[0038] By adopting the above technical solutions, a composite self-cleaning system was constructed, integrating mechanical centrifugal cleaning, pulsed air blowing assistance, constant pressure air curtain isolation, and intelligent triggering. Through the combination of mechanical and pneumatic methods, and intelligent triggering based on signal quality, efficient, proactive, and adaptive cleaning and maintenance of the optical window was achieved. This effectively combats severe dust and metal vapor contamination in the converter environment, maintains high transmittance of the optical system over a long period, thereby ensuring the stability and measurement accuracy of the detection signal and significantly reducing the frequency of system maintenance.

[0039] Secondly, this application provides an online real-time dynamic detection system for carbon content in molten steel produced in converter steelmaking, including a central control host, and further comprising:

[0040] An infrared imaging unit, which is connected to the host, includes an infrared thermal imager installed on the furnace top;

[0041] A multispectral sensing unit, which is connected to the host, includes sensors for acquiring visible and near-infrared spectral data;

[0042] The protection control unit is driven by the host and includes a water-cooled module, a compressed air curtain, and a compressed air purging device.

[0043] An edge computing unit runs machine learning algorithms deployed on the host computer to optimize the carbon content model in real time.

[0044] By adopting the above technical solution, multispectral sensing, infrared imaging, multiple active protection mechanisms, and edge-side intelligent computing are integrated and coordinated through a central host. This ensures stable, accurate, and rapid online carbon content detection even in the extreme environment of the converter. It provides a crucial hardware foundation and technical guarantee for realizing intelligent and real-time control of the steelmaking process.

[0045] Thirdly, this application provides an electronic device including a memory and a processor, wherein the memory is used to store computer program code, and the processor is used to execute the computer program code stored in the memory to implement the methods in the first aspect and any one of the first aspects, or in the second aspect and any possible implementation of the second aspect.

[0046] Fourthly, this application provides a computer-readable storage medium storing a computer program or instructions that, when executed, implement the methods described in the first aspect and any one thereof, or the second aspect and any possible implementation thereof. Attached Figure Description

[0047] Figure 1 This is a schematic flowchart illustrating an online real-time dynamic detection method for carbon content in molten steel produced in a converter, provided in an embodiment of this application.

[0048] Figure 2 This is a schematic diagram of the structure of an online real-time dynamic detection system for carbon content in molten steel produced in a converter, provided in an embodiment of this application.

[0049] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0050] The technical solutions in this application will now be described with reference to all the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them.

[0051] In the description of the embodiments of this application, unless otherwise stated, " / " means "or". For example, A / B can mean A or B. "And / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Furthermore, in the description of the embodiments of this application, "plural" or "multiple" refers to two or more than two.

[0052] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this embodiment, unless otherwise stated, "a plurality of" means two or more.

[0053] The terminology used in the following embodiments is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification and appended claims of this application, the singular expressions “a,” “an,” “the,” “the,” “the,” and “this” are intended to also include expressions such as “one or more,” unless the context clearly indicates otherwise. It should also be understood that in the following embodiments of this application, “at least one” and “one or more” refer to one, two, or more than two.

[0054] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "one embodiment," "some embodiments," "another embodiment," "other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0055] This application provides a method for online real-time dynamic detection of carbon content in molten steel produced in converter steelmaking. The method is executed by an electronic device, which can be a standalone physical electronic device, a cluster of multiple physical electronic devices, a distributed system, or a cloud electronic device providing cloud computing services. This application does not impose any limitations on this method. Figure 1 As shown, the method includes the following steps:

[0056] S1. Acquire multi-source sensor data.

[0057] Specifically, the following steps are included:

[0058] S101. Acquire the surface temperature image of molten steel collected by an infrared thermal imager located above the converter.

[0059] S102. Acquire multi-band spectral data, including controlling the visible band sensor to acquire radiation intensity in the range of 500-600nm and controlling the near-infrared band sensor to acquire reflectivity characteristics in the range of 850-1100nm.

[0060] Specifically, the infrared thermal imager uses a mid-wave infrared detector connected to the central control unit. The multi-band spectral acquisition unit employs a beam-splitting prism structure to split the incident light into a visible light sensor and a near-infrared sensor, with both sensors sharing the same optical path. It is also equipped with a dustproof and waterproof housing and an active cooling system to ensure stable operation in high-temperature and high-dust environments.

[0061] Furthermore, the optical system adopts a coaxial design, and the visible light and near-infrared channels are separated by a dichroic mirror to ensure that the two bands acquire the same field of view.

[0062] Meanwhile, the sensor synchronization adopts a hardware triggering method, with the FPGA generating a synchronization pulse signal to simultaneously start the exposure and reading process of the two sensors, and its time synchronization accuracy reaches the microsecond level.

[0063] In summary, by using optical coaxial design and hardware synchronous triggering mechanism, spatiotemporal consistency of multi-source data was achieved, eliminating the field-of-view deviation and timing jitter problems existing in traditional multi-sensor separate placement schemes. This breaks through the technical limitations of traditional separate measurement, provides high-quality input for subsequent data fusion, and improves the accuracy of carbon content inversion.

[0064] In some embodiments, to ensure the temporal and spatial consistency of multi-band spectral data and avoid problems such as signal asynchrony and field-of-view misalignment caused by separate sensor placement, the step of acquiring multi-source sensor data includes:

[0065] Multi-band spectral data is acquired using an optical probe positioned above the converter, which integrates visible and near-infrared sensors.

[0066] Specifically, the optical probe uses a fully sealed stainless steel housing, which integrates a beam splitting system, a sensor array, and a temperature control unit.

[0067] Furthermore, the optical path design adopts a common optical path scheme, which uses a dichroic mirror to separate 500-600nm visible light and 850-1100nm near-infrared light to the corresponding sensors.

[0068] Furthermore, through integrated optical-mechanical design, fully synchronous acquisition of dual-band data is achieved, and perfect integration of multi-band sensors is realized, significantly improving the accuracy of carbon content inversion.

[0069] In summary, integrating visible light and near-infrared sensors into a single optical probe enables synchronous and simultaneous acquisition of data from two key spectral bands. This design ensures the temporal and spatial consistency of multi-band spectral data, avoiding issues such as signal asynchrony and field-of-view misalignment caused by separate sensor placement. This provides a high-quality data foundation for subsequent data fusion and further improves the accuracy and reliability of the carbon content retrieval model.

[0070] In some embodiments, in order to achieve synchronous high-speed acquisition and noise suppression of multi-channel data from the source, after the step of acquiring multi-source sensor data, the following steps are further included:

[0071] S103: Acquire multi-channel spectral data at time intervals using an FPGA parallel architecture.

[0072] Specifically, the FPGA chip is configured with a multi-channel data acquisition architecture, which includes 12 independent data acquisition channels, each equipped with a dedicated ADC interface and a FIFO buffer.

[0073] The acquisition timing is precisely controlled by a hardware state machine, and a ring buffer management strategy is adopted to support continuous data stream processing. The hardware interface uses LVDS differential signal transmission to enhance anti-interference capabilities.

[0074] Furthermore, this parallel acquisition architecture employs a pipelined design, with the ADC conversion, data buffering, and transmission stages executed concurrently. Each channel is equipped with an independent clock management unit, supporting dynamic adjustment of the sampling rate, and its time interval counter uses a 32-bit high-precision timer to ensure that the sampling interval accuracy reaches the nanosecond level.

[0075] By adopting a multi-channel parallel acquisition architecture and hardware timing control, it can achieve synchronous acquisition of 12-channel spectral data, breaking through the speed limitations of traditional acquisition systems, realizing high-speed and high-precision data acquisition, and providing a stable data stream for real-time processing.

[0076] S104. A hardwired filter is used to remove background radiation noise in real time.

[0077] Specifically, the hardwired filter adopts an FIR digital filter architecture. The filtering calculation uses a fully parallel structure, with 256 multiply-accumulate units operating simultaneously.

[0078] The background noise model is based on prior knowledge and includes environmental temperature compensation and electromagnetic interference characteristic parameters.

[0079] Furthermore, the filtering algorithm employs adaptive noise cancellation technology to estimate the background radiation intensity in real time and generate a compensation signal.

[0080] The filter coefficients are dynamically adjusted based on sensor characteristics and environmental conditions, and can be updated online.

[0081] Furthermore, by using hardware filters and adaptive noise cancellation technology, real-time and efficient suppression of background noise is achieved, significantly improving signal quality and solving signal quality problems in industrial environments with strong interference, thus meeting real-time processing requirements.

[0082] S105. Control the effective signal extraction delay to within 5ms.

[0083] Specifically, the effective signal extraction pipeline comprises three processing units: the first stage performs data preprocessing and format conversion, the second stage performs feature extraction and signal recognition, and the third stage completes data encapsulation and transmission. A double buffering mechanism is employed between each stage to ensure continuous data flow.

[0084] Furthermore, this delay control employs a hardware timestamp mechanism, marking the acquisition time and processing status of each data packet.

[0085] The system monitors and processes latency in real time, and automatically activates a simplified algorithm mode when the latency approaches a threshold.

[0086] Experimental results show that by optimizing the hardware pipeline and implementing a dynamic delay control mechanism, extremely low-latency signal processing can be achieved, meeting the requirements of real-time control systems and providing sufficient time margin for subsequent processing.

[0087] In summary, a hardware parallel architecture using FPGA chips and hardwired filters is employed to process the raw spectral signal at the lowest level. Leveraging the hardware parallel capabilities of FPGAs, synchronous high-speed acquisition and noise suppression of multi-channel data are achieved from the source, keeping the signal preprocessing latency extremely low, within 5ms. This allows ample time for subsequent complex algorithm processing, ensuring the real-time foundation of the entire system at the hardware level and significantly enhancing the system's anti-interference capability and signal stability in industrial environments with strong electromagnetic interference and high radiation levels.

[0088] In this embodiment, the method further includes:

[0089] S2. Data analysis and fusion.

[0090] Specifically, the following steps are included:

[0091] S201. Analyze the temperature image and generate a two-dimensional temperature field for the molten pool.

[0092] Specifically, the temperature field analysis uses a mid-wave infrared focal plane detector to collect radiation from the surface of molten steel, and corrects for the dust absorption effect through the radiation transfer equation to generate a two-dimensional temperature field.

[0093] S202. Analyze multi-band spectral data and extract spectral feature vectors related to carbon content.

[0094] Specifically, this spectral feature extraction integrates a dual-band sensor for visible light (500-600nm) and near-infrared light (850-1100nm). It separates the characteristic wavelengths using a narrowband filter, captures the radiation intensity using a photodiode array, and converts it into a digital signal via an ADC module.

[0095] S203, by integrating the two-dimensional temperature field and spectral feature vector, and based on a pre-trained carbon content machine learning model, generates a carbon content distribution map on the surface of molten steel.

[0096] Specifically, the fusion process involves an industrial computer equipped with a GPU running a pre-trained neural network model. The input layer receives the temperature field matrix and spectral feature vectors, the hidden layer performs nonlinear feature mapping, and the output layer generates a carbon content distribution map.

[0097] Furthermore, a multi-physical coupling mechanism is constructed. Its temperature field provides a baseline for the thermal state of the molten pool, correcting the background noise of thermal radiation in the spectral data; its visible light band captures the luminescence intensity of Fe-C compounds, directly correlated with carbon activity; and its near-infrared band analyzes surface reflectivity characteristics. Carbon concentration is inverted by a neural network that extracts spatial features through convolutional layers and establishes a cross-domain mapping relationship between temperature, spectrum, and carbon content through fully connected layers.

[0098] In some embodiments, to improve the stability and durability of the detection system during long-term operation in a high-temperature, high-dust converter environment, the following steps are included after the data parsing and fusion step:

[0099] S204. Monitor the environmental status and obtain real-time temperature data of the optical probe's sealed chamber.

[0100] S205. Implement active protection: when the temperature is ≥300℃, activate the water cooling circulation and compressed air curtain; when the temperature is ≥400℃, activate the backup cooling circuit.

[0101] Specifically, the temperature monitoring uses a K-type thermocouple embedded in the probe's sealed chamber wall and connected to the PLC controller via a shielded twisted-pair cable.

[0102] Specifically, the protection system is designed with a tiered mechanism. The first level of protection consists of a copper-based heat pipe attached to the sensor housing, an external heat dissipation fin assembly, and forced air cooling. The second level of protection consists of a compressed air curtain that outputs a laminar flow air curtain through an annular nozzle. The third level of protection consists of a backup semiconductor cooling module embedded in the probe base.

[0103] Furthermore, the protection system establishes dynamic control logic, employing a PLC-built-in fuzzy PID algorithm. When the temperature is between 300℃ and 400℃, it activates the primary water cooling and secondary air curtain; when the temperature is ≥400℃, it adds semiconductor cooling; and when the temperature is <280℃ and remains below 30 seconds, it shuts down all protection measures.

[0104] Among them, the airflow organization is optimized by pre-cooling compressed air to 50°C through a vortex tube and setting the air curtain tilt angle to 30° to form a three-dimensional isolation barrier.

[0105] The system automatically switches protection levels based on temperature changes to resolve conflicts between fixed protection strategies and variable operating conditions; it also utilizes air curtain kinetic energy to disperse dust, achieving both cooling and isolation functions.

[0106] In summary, a graded active protection mechanism based on temperature threshold triggering was established, rather than a single or low-threshold protection strategy. This graded protection strategy is highly practical for industrial applications. It ensures effective protection of the probe under most operating conditions and can activate stronger cooling under extreme conditions, significantly improving the stability and durability of the detection system in the long-term operation of the high-temperature, high-dust converter environment. This solves the common problem of online detection equipment being unable to operate reliably for extended periods in extreme industrial environments.

[0107] In this embodiment, the method further includes:

[0108] S3. Output optimized detection results.

[0109] Specifically, the following steps are included:

[0110] S301. Process multi-node data through a distributed computing framework and output a dynamic carbon content detection report.

[0111] Specifically, the edge node deployment involves configuring an industrial edge computing box in each sensor group, equipped with a built-in multi-core processor and pre-installed with a lightweight carbon content inference model. A central gateway architecture is established, employing a star network topology built with 5G industrial routers. The central server is equipped with a processor and FPGA acceleration card to run model aggregation algorithms. The edge nodes upload encrypted local carbon content matrices, and the central gateway performs global weighted average fusion periodically.

[0112] Furthermore, this distributed collaborative mechanism employs Precision Time Protocol (PTP) for time synchronization; fault-tolerant design ensures that when an edge node fails, the central gateway automatically uses an interpolation algorithm to compensate for data loss; and dynamic load balancing automatically adjusts the data upload frequency of edge nodes based on network latency.

[0113] The computing tasks are decomposed to edge nodes and central gateways to avoid the bottleneck of original data transmission bandwidth.

[0114] It adaptively adjusts the data transmission strategy based on network conditions to ensure system real-time performance; and enables carbon content detection covering most areas of the molten pool surface.

[0115] In some embodiments, in order to enable the online analysis results to be directly used for closed-loop real-time control of the production process, the step of outputting optimized detection results further includes the following steps:

[0116] S302 accelerates data processing by compressing computation through model pruning techniques, reducing response time to less than 20ms.

[0117] The S303 is a parallel optimization node that uses a multi-core processor to process data from multiple sensors simultaneously.

[0118] Specifically, the model pruning engine implements a structured pruning algorithm on an FPGA, which includes weight pruning to remove convolutional kernel parameters with low absolute values; channel pruning to remove feature channels with low contribution; and knowledge distillation to guide the training of the pruned model using the unpruned model.

[0119] Specifically, a parallel processing architecture is established, for example, cores 0-1 process infrared data, cores 2-3 process visible spectrum, and cores 4-5 process near-infrared spectrum; at the same time, zero-copy technology is used to avoid repeated data migration between memory.

[0120] Furthermore, the pruning rate is dynamically adjusted in a coordinated manner; for example, when the carbon content change rate is greater than 0.02% / s, the pruning threshold is automatically lowered. Simultaneously, the three-stage pipeline of data acquisition, preprocessing, and inference calculation is executed in an overlapping manner.

[0121] In summary, model pruning and multi-core parallel processing techniques were introduced to accelerate the data processing stage. The total response time from data acquisition to result output was controlled to within 20 milliseconds, achieving true real-time dynamic detection. This speed meets the extreme requirements of the control system for rapid carbon content changes at the end of converter blowing, breaking through the data processing speed bottleneck and enabling online analysis results to be directly used for closed-loop real-time control of the production process.

[0122] In some embodiments, in order to achieve efficient, proactive, and adaptive cleaning and maintenance of the optical window, the step of outputting optimized detection results further includes:

[0123] S304: Control the rotary optical probe to centrifuge at a speed of 1200° / min to remove dust, and combine with pulsed compressed air purging.

[0124] S305, generate an argon curtain of 0.1MPa to 0.15MPa to isolate metal vapor.

[0125] S306. When the optical transmittance is <95% or the signal-to-noise ratio of a specific band is lower than the preset threshold, the self-cleaning procedure is triggered.

[0126] Specifically, a rotary dust removal mechanism is constructed, for example, using a stepper motor driven by a magnetic coupling to power the sapphire window, with an adjustable speed of 0-1200 rpm and an acceleration of 100 rad / s². A pneumatic system is provided, employing pulse purging, for example, using a solenoid valve to control 0.5 MPa compressed air with a pulse width of 50 ms and an interval of 2 s; an argon curtain is generated, for example, using a mass flow controller to adjust the argon output to 0.1-0.15 MPa, while the vortex tube is pre-cooled to 40°C.

[0127] Specifically, the central control unit connects to the intelligent triggering module, which measures the transmittance in real time through a laser diode and a photodetector; the FPGA chip calculates the standard deviation of the 500-600nm band signal in real time.

[0128] Furthermore, the composite cleaning power is as follows: centrifugal cleaning, for example, at a rotation speed of 1200 rpm, generates a centrifugal acceleration of ≥8g to overcome dust adhesion; gas-solid synergy, for example, pulsed airflow forms vortices in the rotational tangential direction to enhance dust removal efficiency; and gas curtain isolation, for example, when the density of argon gas is greater than that of air, forms an inert gas barrier that covers downwards.

[0129] In summary, a composite self-cleaning system was constructed, integrating mechanical centrifugal cleaning, pulsed air blowing assistance, constant pressure air curtain isolation, and intelligent triggering. By combining mechanical and pneumatic methods and intelligently triggering based on signal quality, efficient, proactive, and adaptive cleaning and maintenance of the optical window was achieved. This effectively combats severe dust and metal vapor contamination in the converter environment, maintains high transmittance of the optical system over a long period, thereby ensuring the stability and measurement accuracy of the detection signal and significantly reducing the frequency of system maintenance.

[0130] Based on the technical solution of this embodiment, infrared thermal imaging technology is integrated with multi-band spectral analysis technology, and a pre-trained machine learning model is used to process and correlate these two different types of physical information, ultimately generating a carbon content distribution map of the molten steel surface. This method completely changes the traditional lagging mode that relies on manual sampling and offline analysis, realizing non-contact, online, and full-surface carbon content detection. Through multi-source information fusion, the shortcomings of single detection methods, such as insufficient dimensionality and susceptibility to interference, are significantly overcome, greatly improving the representativeness, accuracy, and reliability of the detection results. This provides real-time and intuitive data support for the endpoint control of converter steelmaking, thereby improving production efficiency and product quality.

[0131] The following describes the online real-time dynamic detection system for carbon content in molten steel for converter steelmaking provided in the embodiments of this application. The online real-time dynamic detection system for carbon content in molten steel for converter steelmaking described below can be referred to in correspondence with the online real-time dynamic detection method for carbon content in molten steel for converter steelmaking described above.

[0132] refer to Figure 2The online real-time dynamic monitoring system for carbon content in molten steel produced in converter steelmaking includes a central control host 1, and:

[0133] Infrared imaging unit 2 is connected to the central control host 1 to transmit raw infrared data streams. Infrared imaging unit 2 uses a mid-wave infrared thermal imager and is mounted on a heavy-duty pan-tilt unit 3-5 meters above the converter opening. The pan-tilt unit is equipped with a dual-degree-of-freedom adjustment mechanism with ±30° pitch and 360° rotation, and integrates a nitrogen purging interface to prevent mirror contamination.

[0134] Multispectral sensing unit 3 is connected to the multichannel spectrometer of the central control host 1 and synchronously triggers data acquisition. Multispectral sensing unit 3 integrates an optical probe and is deployed on the same platform as the infrared thermal imager. Its optical axis coincides with the infrared field of view. Internally, it includes a visible light sensor for acquiring visible light and a near-infrared sensor for acquiring near-infrared spectral data, arranged in parallel.

[0135] The visible light sensor uses a 500-600nm band CMOS photosensitive chip and is equipped with a narrowband filter; the near-infrared sensor uses an 850-1100nm band InGaAs photodiode array and is equipped with a switchable filter wheel.

[0136] The protection control unit 4 is driven by the central control host 1 through a PLC control cabinet and receives feedback signals from the temperature sensor and the light intensity sensor. The protection control unit 4 includes a water-cooling module, a compressed air curtain, and a compressed air purging device.

[0137] The water-cooling module uses a copper-based heat pipe array that fits tightly against the sensor housing, and is externally connected to a circulating pump and a plate heat exchanger. The compressed air curtain uses annular stainless steel nozzles surrounding the optical window, with an adjustable air source pressure of 0.7-1.0 MPa; the compressed air purging device uses a solenoid valve to control a 0.5 MPa pulse airflow, with the nozzles aligned with the center of the window.

[0138] Edge computing unit 5, housed within a protective enclosure, runs a lightweight convolutional neural network model optimized through model pruning, deployed on the central control host 1. It processes sensor data from the local area in real time and outputs a local carbon content matrix. Edge computing unit 5 utilizes an industrial-grade embedded computer and is equipped with an FPGA coprocessor to accelerate model inference.

[0139] Furthermore, the central control host 1 coordinates the work of each unit, runs a global data fusion algorithm, generates a final carbon content distribution report, and displays the molten pool temperature field, carbon content distribution cloud map, and system status parameters in real time.

[0140] In summary, the system integrates multispectral sensing, infrared imaging, multiple active protection mechanisms, and edge-side intelligent computing, all coordinated and controlled by a central host. This ensures stable, accurate, and rapid online carbon content detection even in the extreme environment of the converter. It provides a crucial hardware foundation and technical guarantee for achieving intelligent and real-time control of the steelmaking process.

[0141] This application provides an electronic device, such as... Figure 3 As shown, Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 3 The illustrated electronic device 300 includes a processor 301 and a memory 303. The processor 301 and the memory 303 are connected, for example, via a bus 302. Optionally, the electronic device 300 may also include a transceiver 304. It should be noted that in practical applications, the transceiver 304 is not limited to one type, and the structure of this electronic device 300 does not constitute a limitation on the embodiments of this application.

[0142] Processor 301 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in connection with the embodiments of this application. Processor 301 may also be a combination that implements computing functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0143] Bus 302 may include a pathway for transmitting information between the aforementioned components. Bus 302 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Bus 302 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 3 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0144] The memory 303 may be a ROM (Read-Only Memory) or other type of static storage device capable of storing static information and instructions, RAM (Random Access Memory) or other type of dynamic storage device capable of storing information and instructions, or it may be an EEPROM (Electrically Erasable Programmable Read-Only Memory), a CD-ROM (Compact Disc Read-Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto.

[0145] The memory 303 is used to store application code that executes the scheme of the embodiments of this application, and its execution is controlled by the processor 301. The processor 301 is used to execute the application code stored in the memory 303 to implement the content shown in the foregoing method embodiments.

[0146] Among them, electronic devices include, but are not limited to: mobile terminals such as mobile phones, laptops, digital radio receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), and in-vehicle terminals (such as in-vehicle navigation terminals), as well as fixed terminals such as digital TVs and desktop computers. Figure 3 The electronic device shown is merely an example and should not be construed as limiting the functionality or scope of the embodiments described in this application.

[0147] This application provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the steps of the above-described method for online real-time dynamic detection of carbon content in molten steel for converter steelmaking.

[0148] Since the embodiments of the computer-readable storage medium portion correspond to the embodiments of the method portion, please refer to the description of the embodiments of the method portion for the embodiments of the computer-readable storage medium portion.

[0149] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0150] The above are only some embodiments of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for online real-time dynamic detection of carbon content in molten steel produced in a converter steelmaking process, characterized in that, Includes the following steps: S1. Acquire multi-source sensor data, including: S101. Acquire the surface temperature image of molten steel collected by an infrared thermal imager. The infrared thermal imager is located above the converter and uses a mid-wave infrared focal plane detector to collect the surface radiation of molten steel. S102. Acquire multi-band spectral data, including controlling the visible band sensor to acquire radiation intensity in the range of 500-600nm and controlling the near-infrared band sensor to acquire reflectivity characteristics in the range of 850-1100nm. S103. Multi-channel spectral data is acquired at time intervals through FPGA parallel architecture. The FPGA chip is configured with 12 independent data acquisition channels. The acquisition timing is precisely controlled by the hardware state machine. The hardware interface adopts LVDS differential signal transmission. Each channel is equipped with an independent clock management unit. Its time interval counter adopts a 32-bit high-precision timer to ensure that the sampling interval accuracy reaches the nanosecond level. S104. A hardwired filter is used to subtract background radiation noise in real time. The hardwired filter adopts an FIR digital filter architecture and the filtering calculation adopts a fully parallel structure, which includes 256 multiply-accumulate units working simultaneously. The filtering algorithm adopts adaptive noise cancellation technology to estimate the background radiation intensity in real time and generate a compensation signal. S105. Control the effective signal extraction delay to within 5ms; S2. Data parsing and fusion, specifically including: S201. Analyze the temperature image and generate a two-dimensional temperature field of the molten pool. Specifically, the two-dimensional temperature field is generated by correcting the smoke absorption effect through the radiation transfer equation. S202. Analyze multi-band spectral data, extract spectral feature vectors related to carbon content, separate feature wavelengths through narrowband filters, capture radiation intensity using a photodiode array, and convert it into a digital signal via an ADC module; S203, by integrating two-dimensional temperature field and spectral feature vector, and based on a pre-trained carbon content machine learning model, a carbon content distribution map of molten steel surface is generated. The machine learning model is a neural network. The neural network extracts spatial features through convolutional layers and establishes cross-domain mapping relationships between temperature, spectrum, and carbon content through fully connected layers. A multi-physical coupling mechanism is constructed, the temperature field provides the baseline of the thermal state of the molten pool, the thermal radiation background noise of the spectral data is corrected, the luminescence intensity of Fe-C compounds is captured in the visible light band and directly correlated with carbon activity, the surface reflectivity characteristics are analyzed in the near-infrared band, and the carbon concentration is inverted. S204. Monitor the environmental status and obtain real-time temperature data of the optical probe's sealed chamber. S205. Implement graded active protection. When the temperature is ≥300℃, activate the first-level protection and the second-level protection. The first-level protection is a copper-based heat pipe attached to the sensor shell, with an external heat dissipation fin group and forced air cooling. The second-level protection is a compressed air curtain outputting a laminar air curtain through an annular nozzle. The compressed air is pre-cooled to 50℃ through a vortex tube, and the air curtain tilt angle is 30°. When the temperature is ≥400℃, an additional level 3 protection is activated, which is a backup semiconductor cooling module embedded in the probe base; When the temperature is below 280℃ for 30 seconds, turn off all protection. S3. Output optimized detection results. include, S301. Process multi-node data through a distributed computing framework and output a dynamic carbon content detection report.

2. The method according to claim 1, characterized in that, S1 includes: Multi-band spectral data is acquired using an optical probe positioned above the converter, which integrates visible and near-infrared sensors.

3. The method according to claim 1, characterized in that, S3 also includes the following steps: S302. Accelerate data processing by compressing computation through model pruning techniques, reducing response time to less than 20ms. The S303 is a parallel optimization node that uses a multi-core processor to process data from multiple sensors simultaneously.

4. The method according to claim 1, characterized in that, Before S3, the following also applies: S304, control the rotary optical probe to centrifuge at a speed of 1200° / min to remove dust, and use pulsed compressed air to blow it; S305, generate an argon curtain of 0.1MPa to 0.15MPa to isolate metal vapor; S306. When the optical transmittance is <95% or the signal-to-noise ratio of a specific band is lower than the preset threshold, the self-cleaning procedure is triggered.

5. An electronic device, characterized in that, include: One or more processors; One or more memory units; And one or more computer programs, wherein the one or more computer programs are stored in the one or more memories, the one or more computer programs including instructions that, when executed by the one or more processors, cause the electronic device to perform the method as described in any one of claims 1 to 4.

6. A computer-readable storage medium, characterized in that, The storage medium stores a program or instructions that, when executed, implement the method as described in any one of claims 1 to 4.

Images