An AI vision-based continuous casting machine fire cutting function dynamic closing system and method

Abstract

The application discloses a continuous casting machine fire cutting function medium dynamic closing system and method based on AI vision and belongs to the technical field of metallurgical continuous casting automation control. The system comprises an image acquisition module, an AI image analysis module, an edge computing gateway and a programmable logic controller. The image acquisition module acquires flame images in real time, the AI image analysis module extracts flame shape features based on a pre-trained convolutional neural network model and outputs a confidence score of a cutting completion state; the edge computing gateway generates a trigger signal when consecutive N frame scores are greater than or equal to a preset threshold; and the programmable logic controller closes a gas valve and an oxygen valve after receiving the signal and synchronously starts a roller bed to move a casting blank away from a cutting area. The application takes a morphological mutation of flame scattering from a fan shape to a concentrated beam shape as a recognition basis, can shorten a closing response time by 5-8 seconds compared with a traditional gun swing limiting control mode, can improve a gas utilization rate by about 30%, and significantly reduces energy consumption and equipment loss.

Description

Technical Field

[0001] This invention belongs to the field of automation control technology for continuous casting in metallurgy, and particularly relates to a dynamic shut-off system and method for the fire cutting machine of a continuous casting machine based on AI vision. Background Technology

[0002] In continuous casting production, the flame cutting machine is a key piece of equipment for cutting billets to length. Its working principle is to use the high-temperature flame generated by the combustion of a mixture of fuel gas and oxygen to melt the billet to the required length according to the process.

[0003] Traditional flame cutting machines typically rely on mechanical gun swing limit switches for energy medium (gas and oxygen) shut-off control. When the cutting gun head swings to a preset end position during the cutting process, the striking pin touches the limit switch, and the control system determines that the cutting is complete and issues a command to close the gas and oxygen valves. However, this control method has the following inherent drawbacks:

[0004] Firstly, the cutting speed needs to be dynamically adjusted based on various process factors such as steel grade, billet temperature, and cross-sectional dimensions. However, the mechanical limit switch has a fixed position and cannot adaptively match the real-time changing cutting speed. When the cutting speed is lower than the preset value, the limit switch has not yet been triggered, but the billet has already been cut off, causing the system to continue supplying gas for 5 to 10 seconds after cutting, resulting in serious energy waste. It is estimated that a single ineffective gas supply during a cut can waste approximately 0.8 standard cubic meters of gas.

[0005] Secondly, if the high-temperature flame cannot be extinguished in time after cutting, it will continue to burn the cutting torch tip. The torch tip wears out faster in the high-temperature oxidizing environment, resulting in a shorter service life, increased replacement frequency, and directly increased equipment maintenance costs and spare parts consumption.

[0006] Third, the entire control process relies on a single mechanical signal, lacks direct confirmation of the cutting endpoint state, has insufficient system robustness, and cannot meet the needs of modern intelligent production lines for refined and low-energy-consumption control. Summary of the Invention

[0007] The purpose of this invention is to provide a dynamic shut-off system and method for the fire cutting machine of a continuous casting machine based on AI vision, so as to solve the problems existing in the prior art.

[0008] The technical solution adopted by this invention to solve its technical problem is:

[0009] A dynamic shut-off system for the heat cutter of a continuous casting machine based on AI vision, comprising:

[0010] The image acquisition module is set at the flame cutting station and is used to acquire real-time image data of the flame status during the cutting process of multiple flame cutting machines;

[0011] The AI ​​image analysis module communicates with the image acquisition module to receive flame state image data and performs real-time inference on the flame state image data based on a pre-trained AI recognition model, outputting a confidence score representing the completion of the cutting process.

[0012] An edge computing gateway, connected to an AI image analysis module, is used to generate a power-off trigger signal when the confidence score meets preset conditions;

[0013] The programmable logic controller (PLC) communicates with the edge computing gateway to receive energy shut-off trigger signals and controls the gas valve and oxygen valve to close based on these signals.

[0014] Furthermore, the image acquisition module is a multispectral industrial camera, whose spectral response range covers the visible light band and the near-infrared band.

[0015] Furthermore, the AI ​​recognition model is a binary classification model built on a convolutional neural network. It uses flame morphology features as the basis for recognition to distinguish between two flame states: during cutting and after cutting. The training dataset contains no fewer than 2,000 flame image samples labeled with cutting status.

[0016] A method for dynamic shut-off of the heat-cutting mechanism in a continuous casting machine based on AI vision includes the following steps:

[0017] S1. Image Acquisition: Real-time acquisition of flame status image data during the cutting process is achieved through an image acquisition module installed above the fire cutting machine, with an acquisition frame rate ≥100 frames / second;

[0018] S2. Real-time inference: Input the flame state image data into the pre-trained AI recognition model for inference. The inference cycle is configured to be ≤0.01 seconds / inference, and the confidence score representing the cutting completion state is output.

[0019] S3. Status determination: Determine whether the confidence score is greater than or equal to the preset threshold for N consecutive frames. If so, determine that the segmentation is complete, where N is a positive integer greater than or equal to 3, and the preset threshold is 0.90~0.98.

[0020] S4, Energy Discharge Closure: In response to the determination that the cutting is complete, an energy discharge closure trigger signal is generated and sent to the programmable logic controller (PLC), which then controls the closure of the gas valve and the oxygen valve.

[0021] Furthermore, N=5, and the preset threshold=0.95.

[0022] Furthermore, the training process for the AI ​​recognition model includes:

[0023] Collect flame cutting image samples under different working conditions, and label each sample with a corresponding cutting status label. The cutting status labels include two categories: cutting in progress and cutting completed.

[0024] Training and validation datasets were constructed based on the labeled samples;

[0025] A convolutional neural network is used to train the training dataset, and the model accuracy is verified by using a validation dataset until the model's accuracy on the validation set reaches a preset standard.

[0026] Furthermore, in step S4, while the programmable logic controller controls the closure of the gas valve and the oxygen valve, it sends a billet conveying start signal to the roller conveyor control system, driving the roller conveyor to move the cut billet away from the cutting area.

[0027] The present invention has the following beneficial effects:

[0028] 1. This invention uses AI vision to monitor the flame shape in real time, and uses the sudden change in flame shape from "fan-shaped scattering" to "concentrated bundle" as the core criterion for completion of cutting. Compared with the traditional gun swing limit control method, it can shorten the energy medium shut-off response time by 5 to 8 seconds, save about 0.8 cubic meters of gas per cut, increase the energy medium utilization rate by about 30%, and significantly reduce production costs.

[0029] 2. The instant shut-off function prevents the cutting head from being continuously burned by the flame after cutting, significantly reducing high-temperature oxidation loss of the cutting head, extending the replacement cycle by about 40%, and significantly reducing equipment maintenance costs.

[0030] 3. The system automatically starts the roller conveyor to move the billet away from the cutting area while shutting off the energy medium, avoiding the billet from stagnating in the cutting area and burning the gun head. It forms a complete closed-loop control from "image acquisition → AI inference → status determination → energy medium shutdown → roller conveyor linkage", which requires no manual intervention throughout the process and improves the consistency of the cutting rhythm by about 25%.

[0031] 4. This invention uses flame morphology changes as the identification basis. During cutting, the flame is blocked as it passes through the billet and scatters in a fan shape. After cutting, the flame loses its obstruction and converges into a bundle shape. This morphological change is determined by the physical process of cutting and is independent of the type of gas (acetylene, propane, coke oven gas, etc.) and the air-fuel ratio, exhibiting strong versatility under various operating conditions. Even if the type of gas or the air-fuel ratio changes on-site, only a small number of new operating condition samples need to be collected through transfer learning to quickly fine-tune and adapt the model, without the need to recollect a large-scale dataset for full training. Attached Figure Description

[0032] Figure 1 This is a schematic diagram of the overall architecture of the dynamic shut-off system for the continuous casting machine fire cutter based on AI vision, according to the present invention.

[0033] Figure 2 This is an image of the flame during cutting, acquired by the image acquisition module of this invention.

[0034] Figure 3 This is an image of the flame after it has been cut, acquired by the image acquisition module of this invention.

[0035] Figure 4 This is a schematic diagram of the process of the dynamic shut-off method for the heat cutter of a continuous casting machine based on AI vision, according to the present invention.

[0036] The components include: 1. Image acquisition module; 2. AI image analysis module; 21. AI recognition model; 3. Edge computing gateway; 4. Programmable logic controller; 5. Gas valve; 6. Oxygen valve; 7. Roller conveyor control system. Detailed Implementation

[0037] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments and accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0038] Example 1:

[0039] like Figure 1-3 As shown, this embodiment provides a dynamic shut-off system for a continuous casting machine's fire cutting mechanism based on AI vision. The system includes an image acquisition module 1, an AI image analysis module 2, an edge computing gateway 3, and a programmable logic controller 4.

[0040] Image acquisition module 1 is a multispectral industrial camera with a spectral response range of 400nm~1100nm and an adjustable frame rate. In this embodiment, it is set to 100 frames / second and a resolution of 1920×1080 pixels, enabling it to clearly capture the state of the flame, such as the scattering of the flame during cutting. The installation method of image acquisition module 1 is as follows: small billets are conveyed via roller conveyors. Taking an eight-flow roller conveyor as an example, one image acquisition module 1 is shared by every four flows of the conveyor. Image acquisition module 1 is installed at a high position in the flame cutting station, such as on a wall or bracket.

[0041] AI image analysis module 2 is an industrial-grade embedded device with a built-in NVIDIA Jetson AGX Xavier AI computing module. This module connects to image acquisition module 1 via a gigabit Ethernet cable to receive image data streams in real time. The module contains a pre-trained AI recognition model 21. AI recognition model 21 uses a lightweight MobileNetV3-SSD network architecture, and its training process is as follows: First, 2000 flame cutting images under different process conditions at the continuous casting site are acquired; then, process experts label each image with "cutting in progress" or "cutting completed"; next, the labeled images are divided into training and validation sets at an 8:2 ratio; finally, the model is trained on a training server using transfer learning techniques until its recognition accuracy on the validation set reaches over 99%.

[0042] Specifically, the training involves identifying flames by extracting their morphological features, specifically the aspect ratio of the flame profile. This is because the flame scatters in a fan shape after penetrating the cast billet, such as... Figure 2 As shown, the flame within the red box spreads in a fan shape, interspersed with sparks; while the flame after cutting is in a bundle shape, without any scattering phenomenon, as shown... Figure 3 As shown, the flames within the red box are bundled. The aspect ratios of the flame outlines are significantly different, and their flame shapes are quite distinct. The verification process involves inputting a validation set image into the AI ​​recognition model 21. After recognition, the AI ​​recognition model 21 outputs a predicted value (e.g., 0.9). This predicted value is then compared with the actual label (e.g., 1) of the image. Since the predicted value of 0.9 is close to 1, the prediction is considered accurate; conversely, if the predicted value is 0.2, the prediction is considered incorrect. This process is repeated until the prediction accuracy reaches over 99%, completing the training.

[0043] After training, the model is converted to TensorRT engine format and deployed to AI image analysis module 2. In actual operation, this module performs inference on the input image every 0.01 seconds and outputs a floating-point number between 0 and 1, which is the Softmax probability value of the "cutting complete" category, also known as the confidence score S.

[0044] Edge computing gateway 3 adopts an industrial-grade edge computing gateway (Advantech UNO-2271G), which establishes communication with AI image analysis module 2 and programmable logic controller 4 via Ethernet / IP protocol. Internally, the gateway runs a decision logic service that receives the score S from AI image analysis module 2 in real time and uses a sliding window mechanism for decision-making: it caches the score values ​​of the most recent 5 frames; once 5 consecutive score values ​​are detected to be ≥0.95, the segmentation is considered complete, and a shutdown trigger data packet is constructed and sent to programmable logic controller 4.

[0045] Programmable Logic Controller 4 is an existing Siemens S7-1500 series controller on the production line. Specific ports of its digital output module are connected to the solenoid pilot valves of gas valve 5 and oxygen valve 6, respectively. When Programmable Logic Controller 4 receives a trigger signal from edge computing gateway 3, its CPU immediately executes an interrupt subroutine, sets the corresponding output port to a low level, and drives the solenoid valves to rapidly depressurize, thereby cutting off the gas and oxygen supply in a very short time.

[0046] Example 2:

[0047] like Figure 4 As shown, this embodiment provides a method for dynamic shut-off of the energy medium of a continuous casting machine's heat cutter based on AI vision. The method steps are implemented based on the AI ​​vision-based dynamic shut-off system for the energy medium of a continuous casting machine's heat cutter in Embodiment 1, and include the following steps:

[0048] Step S1: Image Acquisition. The small billet is transported to the cutting position of the flame cutter via the roller conveyor control system 7 and then stops. After cutting starts, the image acquisition module 1 begins acquiring flame images at a frame rate of 100 frames per second. The image data is pushed in real time to the receiving buffer of the AI ​​image analysis module 2 via the GigEVision protocol.

[0049] Step S2: Real-time inference. The AI ​​image analysis module 2 retrieves images sequentially from the cache and inputs them into the AI ​​recognition model 21. After forward calculation, the AI ​​recognition model 21 outputs the probability that the current frame belongs to "segmentation completed", which is the confidence score S. This process is executed cyclically with a period of 10ms.

[0050] Step S3: Status Determination. The edge computing gateway 3 continuously monitors the output port of the AI ​​image analysis module 2 and uses sliding window logic for determination. Each received score S value is stored in a circular buffer of length 5. Upon receiving a new confidence score S, the system performs a first-in, first-out (FIFO) queue update: the new score is written to the tail of the queue; if the queue is full, the earliest score data at the head of the queue is automatically discarded. After each queue update, the determination logic unit iterates through all 5 score values ​​in the current queue and checks whether all 5 values ​​in the buffer are ≥0.95. If the result is yes, the segmentation is determined to be complete, and the process jumps to step S4; if the result is no, the system is determined to be "segmenting in progress" and then continues to wait for the next frame score.

[0051] Step S4: Power off. After the cutting is completed, the edge computing gateway 3 encapsulates a Modbus command and sends it to the data register of the programmable logic controller 4. Upon receiving the command, the programmable logic controller 4 triggers a hardware interrupt, directly operating the physical output point to shut off the gas valve 5 and the oxygen valve 6. Simultaneously, the programmable logic controller 4 sends a billet conveying start signal to the roller conveyor control system 7, driving the roller conveyor to move the cut billet away from the cutting area, and then waits for the next cutting position to be reached and cut.

[0052] The above embodiments are merely descriptions of preferred embodiments of the present invention and are not intended to limit the concept and scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the design concept of the present invention should fall within the protection scope of the present invention.

[0053] The technologies, shapes, and structures not described in detail in this invention are all known technologies.

Claims

1. A dynamic shut-off system for the heat-cutting mechanism of a continuous casting machine based on AI vision, characterized in that, include: The image acquisition module is set at the flame cutting station and is used to acquire real-time image data of the flame status during the cutting process of multiple flame cutting machines; The AI ​​image analysis module communicates with the image acquisition module to receive flame state image data and performs real-time inference on the flame state image data based on a pre-trained AI recognition model, outputting a confidence score representing the completion of the cutting process. An edge computing gateway, connected to an AI image analysis module, is used to generate a power-off trigger signal when the confidence score meets preset conditions; The programmable logic controller (PLC) communicates with the edge computing gateway to receive energy shut-off trigger signals and controls the gas valve and oxygen valve to close based on these signals.

2. The AI ​​vision-based dynamic shut-off system for the heat-cutting mechanism of a continuous casting machine according to claim 1, characterized in that, The image acquisition module is a multispectral industrial camera, whose spectral response range covers the visible light band and the near-infrared band.

3. The AI ​​vision-based dynamic shut-off system for the heat-cutting mechanism of a continuous casting machine according to claim 1, characterized in that, The AI ​​recognition model is a binary classification model built on a convolutional neural network. It uses flame morphology features as the basis for recognition to distinguish between two flame states: during cutting and after cutting. The training dataset contains no fewer than 2,000 flame image samples labeled with cutting status.

4. A method for dynamic shut-off of the heat-cutting mechanism in a continuous casting machine based on AI vision, characterized in that, The AI ​​vision-based dynamic shut-off system for the heat cutter of a continuous casting machine, as described in any one of claims 1-3, includes the following steps: S1. Image Acquisition: Real-time acquisition of flame status image data during the cutting process is achieved through an image acquisition module installed above the fire cutting machine, with an acquisition frame rate ≥100 frames / second; S2. Real-time inference: Input the flame state image data into the pre-trained AI recognition model for inference. The inference cycle is configured to be ≤0.01 seconds / inference, and the confidence score representing the cutting completion state is output. S3. Status determination: Determine whether the confidence score is greater than or equal to the preset threshold for N consecutive frames. If so, determine that the segmentation is complete, where N is a positive integer greater than or equal to 3, and the preset threshold is 0.90~0.

98. S4, Energy Discharge Closure: In response to the determination that the cutting is complete, an energy discharge closure trigger signal is generated and sent to the programmable logic controller (PLC), which then controls the closure of the gas valve and the oxygen valve.

5. The AI ​​vision-based dynamic shut-off method for the heat-cutting mechanism of a continuous casting machine according to claim 4, characterized in that, The value of N is 5, and the preset threshold is 0.

95.

6. The AI ​​vision-based dynamic shut-off method for the heat-cutting mechanism of a continuous casting machine according to claim 4, characterized in that, The training process of the AI ​​recognition model includes: Collect flame cutting image samples under different working conditions, and label each sample with a corresponding cutting status label. The cutting status labels include two categories: cutting in progress and cutting completed. Training and validation datasets were constructed based on the labeled samples; A convolutional neural network is used to train the training dataset, and the model accuracy is verified by using a validation dataset until the model's accuracy on the validation set reaches a preset standard.

7. The AI ​​vision-based dynamic shut-off method for the heat-cutting mechanism of a continuous casting machine according to claim 4, characterized in that, In step S4, while the programmable logic controller controls the closure of the gas valve and oxygen valve, it sends a billet conveying start signal to the roller conveyor control system, driving the roller conveyor to move the cut billet away from the cutting area.

Images