A molten steel level control system and control method for amorphous ribbon manufacturing

Abstract

This invention relates to the field of amorphous ribbon manufacturing equipment technology, and more particularly to a molten steel level control system and method for an amorphous ribbon forming furnace. It includes a vision inspection module, a data calculation module, and an execution module, all electrically connected in sequence, as well as an alarm module electrically connected to the data calculation module; achieving automated, high-precision control of the molten steel level, reducing safety risks, and ensuring the quality of amorphous ribbon products.

Description

Technical Field

[0001] This invention relates to the field of amorphous ribbon manufacturing equipment technology, and in particular to a molten steel level control system and control method for an amorphous ribbon forming furnace. Background Technology

[0002] In the manufacturing process of amorphous ribbon, when molten steel is transported from the large furnace to the amorphous ribbon forming furnace, the molten steel level in the furnace must be controlled at a specified height with an error strictly controlled within ±2mm. This is a core prerequisite for ensuring the yield and high quality of amorphous ribbons. In existing technologies, the molten steel level is mostly controlled manually: operators manually adjust the depth of the tungsten steel rod inserted into the molten steel by observing the level gauge to control the flow rate. However, this manual control method has significant drawbacks: insufficient precision: subjective errors exist in manual observation and operation, resulting in large fluctuations in the liquid level and making it difficult to maintain a stable error range within ±2mm, leading to poor consistency in the quality of amorphous ribbon products; high safety risks: operators need to focus on observing the liquid level for extended periods, which can easily lead to fatigue and production accidents such as molten steel overflow and equipment damage due to slow reaction or operational errors; strong reliance on experience: significant differences in the technical level and experience of different operators further affect the stability of liquid level control. Therefore, developing a molten steel level control system that can replace manual labor and achieve high precision and automation has become a key requirement for solving the pain points of existing technologies. Summary of the Invention

[0003] To address the problems of low precision, high safety risks, and reliance on human experience in existing manual control of molten steel levels, this invention provides a molten steel level control system and method for amorphous ribbon forming furnaces, achieving automated and high-precision control of molten steel levels, reducing safety risks, and ensuring the quality of amorphous ribbon products.

[0004] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is a molten steel level control system for amorphous ribbon manufacturing, comprising a vision detection module, a data calculation module, an execution module, and an alarm module electrically connected in sequence to the data calculation module.

[0005] The visual inspection module is used to collect image information of the float inside the molten steel ladle and identify the float height to obtain the actual liquid level value;

[0006] The data calculation module is preset with a liquid level setpoint, a liquid level deviation threshold, an execution action correction number threshold, a maximum movement distance of the execution module, and a minimum movement distance. After receiving the actual liquid level value, it calculates the liquid level deviation: ΔY=Ys-Y, and calculates the target position of the execution module. Based on the liquid level deviation, the current position of the execution module, and the liquid level offset direction, it generates an execution action command and counts the number of actions of the execution module in real time. When the number of actions in the same direction exceeds the correction number threshold and the liquid level continues to deviate, it triggers a position offset adjustment.

[0007] The formula for calculating the target location of the execution module is as follows:

[0008] X = Mmax / 10 * (Ys - Y);

[0009] In the formula, X is the target position of the execution module, Mmax is the maximum moving distance, Ys is the liquid level set value, and Y is the actual liquid level value;

[0010] The execution module is used to receive instructions from the data calculation module and adjust the molten steel flow rate, with a response time of no more than 50ms;

[0011] The alarm module triggers corresponding levels of audible and visual warnings based on the liquid level deviation value and equipment status signals transmitted by the data calculation module.

[0012] The aforementioned molten steel level control system for amorphous ribbon manufacturing includes a vision detection module comprising an industrial camera and an image recognition unit. The industrial camera's shooting angle is aimed at the area where the float is located, and the image recognition unit uses a deep learning algorithm to identify the float height in real time with an accuracy of not less than 0.5mm.

[0013] The deep learning algorithm in the above-mentioned molten steel level control system for amorphous ribbon manufacturing includes:

[0014] S1, Preprocessing stage formula:

[0015] (1) Grayscale formula

[0016] Convert RGB images captured by industrial cameras to grayscale images to suppress color interference caused by molten steel reflections:

[0017] ;

[0018] Parameter definition:

[0019] G(x,y): The gray value at coordinates (x,y);

[0020] R(x,y), G(x,y), B(x,y): RGB three-channel pixel values ​​(0-255) at coordinates (x,y);

[0021] (2) Gaussian filtering formula

[0022] Remove image noise while preserving float contour features:

[0023] ;

[0024] Parameter definition:

[0025] : The gray value at coordinates (x, y) after filtering;

[0026] σ=0.8 (standard deviation of the filter kernel, adapted to the noise characteristics of molten steel surface);

[0027] (3) Normalization formula

[0028] Mapping grayscale values ​​to the [0,1] interval improves the convergence speed of model training.

[0029] ;

[0030] : Normalized grayscale value;

[0031] , The minimum and maximum grayscale values ​​in a single frame of an image;

[0032] S2: Core formula for feature extraction layer:

[0033] (1) Convolutional layer calculation formula (taking Conv1 as an example, the same applies to Conv2 / Conv3)

[0034] ;

[0035] Parameter definition:

[0036] : The value of the feature map output by the k-th convolutional kernel at coordinates (x, y) (k=1,2,...,32);

[0037] The weight parameters of the k-th 3×3 convolution kernel;

[0038] : The bias parameters of the k-th convolutional kernel;

[0039] ReLU: Activation function, ReLU(z)=max(0,z) (enhances the nonlinear expressive power of the model);

[0040] (2) Formula for calculating the maximum pooling layer (taking Pool1 as an example, the same applies to Pool2)

[0041] ;

[0042] Parameter definition:

[0043] The value of the k-th feature map at coordinates (x, y) after pooling;

[0044] Step size = 2, pooling kernel size = 2×2 (dimensionality reduction ratio 1:2);

[0045] (3) Formula for calculating the global average pooling layer

[0046] ;

[0047] : Global average pooling result of the k-th higher-order feature map (k=1,2,...,128);

[0048] H, W: Height and width of the Conv3 output feature map (80×60 after two pooling operations);

[0049] S3: Formulas for Fully Connected Layers and Output Layers

[0050] (1) Calculation of fully connected layer 1

[0051] ;

[0052] Parameter definition:

[0053] : Output of the m-th neuron in fully connected layer 1 (m=1,2,...,256);

[0054] The weight parameters from feature map k to neuron m;

[0055] : Bias parameters of neuron m;

[0056] Dropout (p=0.2): Randomly deactivates 20% of neurons to prevent overfitting;

[0057] (2) Output layer calculation (float exposed pixel length h) px )

[0058] ;

[0059] Parameter definition:

[0060] : The pixel length of the part of the float that is exposed above the liquid surface, as output by the model;

[0061] Weight parameters from neuron m in the fully connected layer to the output layer;

[0062] : The bias parameters of the output layer;

[0063] S4: Liquid level conversion formula (connecting with previous derivations to form a complete chain)

[0064] ;

[0065] Parameter definition:

[0066] Y: Actual molten steel level (unit: mm);

[0067] H0: Total height of the float = 150mm;

[0068] K: Pixel equivalent (unit: pixels / mm), calibrated using a 100mm standard liquid level block, K = pixel length corresponding to the standard liquid level block / 100mm;

[0069] δ: Recognition error correction value, δ∈[-0.5,+0.5] (ensuring recognition accuracy ≥0.5mm).

[0070] The aforementioned molten steel level control system for amorphous ribbon manufacturing includes an execution module comprising a servo motor, an electric cylinder, a tungsten steel rod, a position feedback sensor, and an anti-blocking unit. The servo motor receives action commands from the data calculation module and drives the electric cylinder to move the tungsten steel rod axially to adjust the molten steel flow rate. The position feedback sensor transmits the current position of the execution module to the data calculation module. The anti-blocking unit automatically triggers a reciprocating action when the execution module has been inactive for a set anti-blocking time. If the position feedback sensor has no signal for 100ms during the action, it is determined to be stuck, and a secondary alarm is immediately triggered and a forced reset command is output.

[0071] The alarm module of the aforementioned molten steel level control system for amorphous ribbon manufacturing includes the following warning levels: when the level deviation exceeds ±5mm, the red warning light flashes briefly and the buzzer sounds briefly; when the level deviation exceeds ±7.5mm, the red warning light flashes rapidly and the buzzer sounds rapidly; when a communication failure or electric cylinder overtravel occurs, the red warning light remains on and the buzzer sounds continuously, simultaneously triggering the molten steel delivery valve to shut off and the execution module to reset.

[0072] In the aforementioned molten steel level control system for amorphous ribbon manufacturing, the anti-blocking unit is set to anti-blocking time of 25-30 seconds, and the automatically triggered reciprocating motion travels 5-10 mm. The anti-blocking action is triggered no more than 3 times within 2 minutes to avoid interfering with the stability of the liquid level.

[0073] The control method for the above-mentioned molten steel level control system for amorphous ribbon manufacturing includes the following steps:

[0074] (1) The float is vertically placed into the molten steel casting ladle of the amorphous ribbon forming furnace. The float is made of silicon nitride ceramic material and has a height of 150mm and a diameter of 30mm. It is fixed by a high-temperature resistant bracket (the distance between the bracket and the inner wall of the casting ladle is 50mm) to ensure that the bottom of the float is in contact with the molten steel surface and the top is exposed above the liquid surface. The industrial camera is selected with a resolution of 1920×1080, a lens focal length of 25mm, an exposure time of 1 / 1000s, and a protection level of IP68. It is installed 1.5m to the side of the casting ladle, with the lens axis flush with the top of the float, to collect the image information of the float in real time. The industrial camera is connected to the image recognition unit through a data cable. The image recognition unit has a built-in deep learning algorithm to process the collected float image, identify the height of the float exposed above the molten steel surface, and then convert it into the actual liquid level value Y of the molten steel. The recognition accuracy is not less than 0.5mm to ensure the accuracy of subsequent calculations. After installation, it is calibrated by a 100mm standard liquid level block to determine the pixel equivalent K value.

[0075] (2) Pre-stored liquid level setting value (Ys), liquid level deviation threshold (±2.5mm, ±5mm, ±7.5mm), execution action correction number threshold (3 times), execution module maximum movement distance (Mmax) and minimum movement distance (Mmin); Mmax is set according to the maximum deviation threshold, and the calculation formula is Mmax=10×maximum deviation adjustment amount (10% margin reserved).

[0076] (3) Calculate the liquid level deviation ΔY = Ys - Y based on the actual liquid level value Y transmitted in step (1), and calculate the target position X of the execution module using the formula X = Mmax / 10 * (Ys - Y).

[0077] (4) Generate execution action instructions based on the magnitude of liquid level deviation, the current position (Xd) of the execution module and the direction of liquid level offset, and count the number of actions of the execution module in real time. When the number of actions in the same direction exceeds the correction threshold and the liquid level continues to deviate, trigger position offset adjustment (adjustment amount = current X value × (1 + 0.01 × cumulative number of actions)) to avoid control failure due to mechanical error; the initial retraction position is defined as the mechanical origin of the electric cylinder (factory preset, calibrated by limit switch), and the active adjustment of the retraction distance is based on the initial retraction position, and is finely adjusted by ±0.5-1mm according to the liquid level rise indication value (positive fine adjustment when the rise indication value is >3mm, negative fine adjustment when <3mm);

[0078] (5) Trigger the corresponding level of audible and visual alarm based on the liquid level deviation value and equipment status signal.

[0079] In the control method of the above-mentioned molten steel level control system for amorphous ribbon manufacturing, in step (4), when the liquid level deviation is within ±2.5mm, the control execution module performs an upward or downward operation and retracts immediately after reaching the extension distance; if the liquid level continues to deviate in one direction and the number of actions of the execution module reaches the preset correction number, position offset adjustment is triggered; when the liquid level deviation is between ±2.5mm and ±5mm, the control execution module pauses briefly after completing the upward or downward operation and retracts back to its original position; if the liquid level continues to deviate and the number of actions reaches the preset correction number, position offset adjustment is triggered; when the liquid level deviation exceeds ±5mm, the control execution module moves rapidly at 1.5 times the speed, pauses continuously after reaching the extension distance and detects signs of liquid level recovery. When the recovery sign value is detected to be 4mm, it retracts immediately and actively adjusts the retraction distance after retraction; if it does not retract after more than 8 seconds, the extension distance is adjusted and the upward and downward operation is repeated; the recovery sign value is the number of millimeters by which the liquid level recovers from the deviation state to the set value.

[0080] The beneficial effects of this invention, a molten steel level control system and method for an amorphous ribbon forming furnace, are that it eliminates the need for manual intervention throughout the entire process, replacing traditional manual observation and operation, reducing reliance on operator experience, minimizing prolonged manual observation and operation, avoiding operational errors due to personnel fatigue, and reducing the risk of production accidents. The alarm module provides tiered alarms for different anomaly types and has a built-in anti-blocking function, comprehensively ensuring production safety. The execution module has a fast response time, quickly addressing level fluctuations and preventing deviations from escalating.

[0081] Through closed-loop control involving visual inspection, data calculation, and execution, the liquid level control error is reduced to ±1mm, which is better than the industry requirement of ±2mm. This significantly improves the consistency of the amorphous ribbon product quality and enables fully automatic detection and adjustment of the liquid level, reducing reliance on workers' operating experience, reducing labor costs, and facilitating standardized production.

[0082] By adjusting preset parameters and correcting position offsets, it can be adapted to different specifications of amorphous ribbon forming furnaces and process requirements, and has a wide range of applications. Attached Figure Description

[0083] Figure 1 This is a circuit structure block diagram of the present invention;

[0084] Figure 2 Here is the flowchart of the liquid level control logic;

[0085] Figure 3 This is a block diagram of the circuit structure of the alarm module. Detailed Implementation

[0086] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0087] like Figure 1-3 As shown, a molten steel level control system for amorphous ribbon manufacturing includes a vision inspection module, a data calculation module, an execution module, and an alarm module electrically connected in sequence.

[0088] The vision inspection module, the data calculation module (PLC), and the execution module are electrically connected in sequence. The alarm module is electrically connected to the data calculation module. The data calculation module communicates with the vision inspection module via Ethernet and is connected to the execution module via pulse signals.

[0089] The visual inspection module is used to collect image information of the float inside the molten steel ladle and identify the float height to obtain the actual liquid level value;

[0090] The data calculation module is preset with a liquid level setpoint, a liquid level deviation threshold, an execution action correction number threshold, a maximum movement distance of the execution module, and a minimum movement distance. These settings are configured according to the furnace specifications and process requirements to meet the needs of different liquid level adjustment ranges. After receiving the actual liquid level value, the module calculates the liquid level deviation: ΔY = Ys - Y, and calculates the target position of the execution module. Based on the magnitude of the liquid level deviation, the current position of the execution module, and the direction of liquid level offset, the module generates an execution action command and counts the number of actions of the execution module in real time. When the number of actions in the same direction exceeds the correction number threshold and the liquid level continues to deviate, a position offset adjustment is triggered.

[0091] The formula for calculating the target location of the execution module is as follows:

[0092] X = Mmax / 10 * (Ys - Y);

[0093] In the formula, X is the target position of the execution module, Mmax is the maximum moving distance, Ys is the liquid level set value, and Y is the actual liquid level value;

[0094] The execution module is used to receive instructions from the data calculation module and adjust the molten steel flow rate, with a response time of no more than 50ms, ensuring rapid feedback on changes in liquid level.

[0095] The alarm module triggers corresponding levels of audible and visual warnings based on the liquid level deviation value and equipment status signals transmitted by the data calculation module.

[0096] The visual inspection module includes an industrial camera and an image recognition unit. The shooting angle of the industrial camera is aimed at the area where the float is located, and the image recognition unit uses a deep learning algorithm to identify the height of the float in real time with an accuracy of not less than 0.5mm.

[0097] The aforementioned molten steel level control system for amorphous ribbon manufacturing includes a deep learning algorithm comprising:

[0098] S1, Preprocessing stage formula:

[0099] (4) Grayscale formula

[0100] Convert RGB images captured by industrial cameras to grayscale images to suppress color interference caused by molten steel reflections:

[0101] ;

[0102] Parameter definition:

[0103] G(x,y): The gray value at coordinates (x,y);

[0104] R(x,y), G(x,y), B(x,y): RGB three-channel pixel values ​​(0-255) at coordinates (x,y);

[0105] (5) Gaussian filtering formula

[0106] Remove image noise while preserving float contour features:

[0107] ;

[0108] Parameter definition:

[0109] : The gray value at coordinates (x, y) after filtering;

[0110] σ=0.8 (standard deviation of the filter kernel, adapted to the noise characteristics of molten steel surface);

[0111] (6) Normalization formula

[0112] Mapping grayscale values ​​to the [0,1] interval improves the convergence speed of model training.

[0113] ;

[0114] : Normalized grayscale value;

[0115] , The minimum and maximum grayscale values ​​in a single frame of an image;

[0116] S2: Core formula for feature extraction layer:

[0117] (1) Convolutional layer calculation formula (taking Conv1 as an example, the same applies to Conv2 / Conv3)

[0118] ;

[0119] Parameter definition:

[0120] : The value of the feature map output by the k-th convolutional kernel at coordinates (x, y) (k=1,2,...,32);

[0121] The weight parameters of the k-th 3×3 convolution kernel;

[0122] : The bias parameters of the k-th convolutional kernel;

[0123] ReLU: Activation function, ReLU(z)=max(0,z) (enhances the nonlinear expressive power of the model);

[0124] (2) Formula for calculating the maximum pooling layer (taking Pool1 as an example, the same applies to Pool2)

[0125] ;

[0126] Parameter definition:

[0127] The value of the k-th feature map at coordinates (x, y) after pooling;

[0128] Step size = 2, pooling kernel size = 2×2 (dimensionality reduction ratio 1:2);

[0129] (3) Formula for calculating the global average pooling layer

[0130] ;

[0131] : Global average pooling result of the k-th higher-order feature map (k=1,2,...,128);

[0132] H, W: Height and width of the Conv3 output feature map (80×60 after two pooling operations);

[0133] S3: Formulas for Fully Connected Layers and Output Layers

[0134] (1) Calculation of fully connected layer 1

[0135] ;

[0136] Parameter definition:

[0137] : Output of the m-th neuron in fully connected layer 1 (m=1,2,...,256);

[0138] The weight parameters from feature map k to neuron m;

[0139] : Bias parameters of neuron m;

[0140] Dropout (p=0.2): Randomly deactivates 20% of neurons to prevent overfitting;

[0141] (2) Output layer calculation (float exposed pixel length h) px )

[0142] ;

[0143] Parameter definition:

[0144] : The pixel length of the part of the float that is exposed above the liquid surface, as output by the model;

[0145] Weight parameters from neuron m in the fully connected layer to the output layer;

[0146] : The bias parameters of the output layer;

[0147] S4: Liquid level conversion formula (connecting with previous derivations to form a complete chain)

[0148] ;

[0149] Parameter definition:

[0150] Y: Actual molten steel level (unit: mm);

[0151] H0: Total height of the float = 150mm;

[0152] K: Pixel equivalent (unit: pixels / mm), calibrated using a 100mm standard liquid level block, K = pixel length corresponding to the standard liquid level block / 100mm;

[0153] δ: Recognition error correction value, δ∈[-0.5,+0.5] (ensuring recognition accuracy ≥0.5mm).

[0154] Model training data: The training adopts a combination of real-world industrial scene data and synthetic data. The real-world data includes float images with different molten steel temperatures (1300-1500℃) and different liquid levels (30-120mm). The synthetic data simulates extreme working conditions by adding reflections and noise. The total sample size is ≥5000 images.

[0155] Calibration process: After each device is deployed, it is calibrated at three different heights (50mm, 80mm, 110mm) using a 100mm standard liquid level block. The average value of the K value is taken to ensure consistency across devices.

[0156] Real-time performance guarantee: The model adopts TensorRT quantization (INT8 quantization) and is deployed on the GPU core of the embedded PLC. The single-frame inference time is ≤8ms, which meets the system's response requirement of ≤50ms for the entire chain of "visual detection, data calculation and action execution".

[0157] The execution module includes a servo motor, an electric cylinder, a tungsten steel rod, a position feedback sensor, and an anti-blocking unit. The servo motor receives pulse commands from the data calculation module and drives the electric cylinder to move the tungsten steel rod axially, adjusting its insertion depth into the molten steel, thereby controlling the flow rate of molten steel into the furnace. Performance parameters: Response time not exceeding 50ms ensures rapid feedback to changes in liquid level; the position feedback sensor transmits the current position of the execution module to the data calculation module; the anti-blocking unit automatically triggers a reciprocating motion when the execution module remains inactive for a set anti-blocking time; if the position feedback sensor shows no signal for 100ms during the action, it is determined to be stuck, immediately triggering a level two alarm and outputting a forced reset command.

[0158] The alarm module's warning levels include: when the liquid level deviation exceeds ±5mm, the red warning light flashes briefly once per second, and the buzzer sounds briefly once per second; when the liquid level deviation exceeds ±7.5mm, the red warning light flashes rapidly three times per second, and the buzzer sounds rapidly three times per second; when a communication failure or electric cylinder overtravel occurs, the red warning light remains constantly on, the buzzer sounds continuously, and the molten steel delivery valve is simultaneously cut off, and the execution module reset action is triggered.

[0159] The anti-blocking unit is set to anti-blocking time of 25-30 seconds, and the automatically triggered reciprocating movement distance is 5-10 mm. The anti-blocking action is triggered no more than 3 times within 2 minutes to avoid interfering with liquid level stability. A tiered alarm mechanism and anti-blocking function are set up to promptly trigger warnings or automatically handle issues such as excessive liquid level, communication failure, and equipment jamming, forming multiple safety safeguards. The anti-blocking unit operates as follows: a built-in timer automatically triggers a short-distance movement when the execution module has been inactive for a set time and the anti-blocking function has been activated, preventing the tungsten steel rod from jamming due to solidification of molten steel or adhesion of impurities.

[0160] The control method for the above-mentioned molten steel level control system for amorphous ribbon manufacturing includes the following steps:

[0161] (1) The float is vertically placed into the molten steel casting ladle of the amorphous ribbon forming furnace. The float is made of silicon nitride ceramic material, with a height of 150mm and a diameter of 30mm. It is fixed by a high-temperature resistant bracket (the distance between the bracket and the inner wall of the casting ladle is 50mm) to ensure that the bottom of the float is in contact with the molten steel surface and the top is exposed above the liquid surface. An industrial camera with a resolution of 1920×1080, a lens focal length of 25mm, an exposure time of 1 / 1000s, and an IP68 protection rating is selected. It is installed 1.5m to the side of the casting ladle, with the lens axis flush with the top of the float, to collect the image information of the float in real time. The industrial camera is connected to the ladle. The image is connected to the image recognition unit via a data cable. The image recognition unit has a built-in deep learning algorithm that processes the acquired float image, identifies the height of the float protruding from the molten steel surface, and then converts it into the actual liquid level value Y of the molten steel. The recognition accuracy is no less than 0.5mm, ensuring the accuracy of subsequent calculations. After installation, it is calibrated using a 100mm standard liquid level block to determine the pixel equivalent K value. Its working principle is to use an industrial camera to acquire visual images of the float in the molten steel, extract the float contour and position features through a deep learning algorithm, and establish a correspondence between the float height and the molten steel level to achieve non-contact and accurate detection of the liquid level.

[0162] (2) Pre-stored liquid level setting value (Ys), liquid level deviation threshold (±2.5mm, ±5mm, ±7.5mm), execution action correction number threshold (3 times), execution module maximum movement distance (Mmax) and minimum movement distance (Mmin); Mmax is set according to the maximum deviation threshold, and the calculation formula is Mmax=10×maximum deviation adjustment amount (10% margin reserved).

[0163] (3) Calculate the liquid level deviation ΔY = Ys - Y based on the actual liquid level value Y transmitted in step (1), and calculate the target position X of the execution module using the formula X = Mmax / 10 * (Ys - Y); (i.e., the distance the tungsten steel rod needs to move).

[0164] (4) Generate execution action instructions based on the magnitude of liquid level deviation, the current position (Xd) of the execution module and the direction of liquid level offset, and count the number of actions of the execution module in real time. When the number of actions in the same direction exceeds the correction threshold and the liquid level continues to deviate, trigger position offset adjustment (adjustment amount = current X value × (1 + 0.01 × cumulative number of actions)) to avoid control failure due to mechanical error; the initial retraction position is defined as the mechanical origin of the electric cylinder (factory preset, calibrated by limit switch), and the active adjustment of the retraction distance is based on the initial retraction position, and is finely adjusted by ±0.5-1mm according to the liquid level rise indication value (positive fine adjustment when the rise indication value is >3mm, negative fine adjustment when <3mm);

[0165] (5) Trigger the corresponding level of audible and visual alarm based on the liquid level deviation value and equipment status signal.

[0166] In step (4), when the liquid level deviation is within ±2.5mm, the control execution module performs an upward or downward operation, and retracts immediately after reaching the extension distance; if the liquid level continues to shift in one direction and the number of times the execution module operates reaches the preset correction number, a position offset adjustment is triggered; the principle is to achieve micro-correction of the liquid level through small-amplitude rapid adjustment, avoiding excessive movement that causes liquid level fluctuations; when the liquid level deviation is between ±2.5mm and ±5mm, the control execution module pauses briefly after completing the upward or downward operation, and then retracts back to its original position; if the liquid level continues to shift and the number of times the operation ... Once the preset number of corrections is reached, position offset adjustment is triggered. The principle is to observe the correction effect through brief pauses, balancing adjustment accuracy and response speed. When the liquid level deviation exceeds ±5mm, the control module moves rapidly to position X at 1.5 times the speed, pausing continuously and detecting liquid level recovery. When a recovery indicator value of 4mm is detected, the module immediately retracts, actively adjusting the retraction distance. If it does not retract within 8 seconds, the extension / retraction distance is adjusted, and the pull-up and pull-down operations are repeated. The principle is to quickly suppress the expansion of liquid level deviation through accelerated movement, combined with recovery detection to achieve precise correction and avoid correction lag. The recovery indicator value is the number of millimeters the liquid level rises from the deviation state to the set value.

[0167] Control logic: Based on the magnitude of the liquid level deviation ΔY, different control commands are output (the specific logic is shown in Table 1). At the same time, the number of actions of the execution module is counted in real time. When the number of actions reaches the preset correction number and the liquid level continues to deviate in one direction, the position offset adjustment is triggered to avoid control failure due to mechanical errors.

[0168] Table 1: Liquid Level Control Logic Table of Data Calculation Module

[0169] Liquid level deviation ΔY Execution module control instructions Within ±2.5mm Pull up / down to the target position X and immediately retract; if the liquid level continues to shift in one direction and the number of actions reaches the correction threshold, then perform a minor adjustment. Between ±2.5mm and ±5mm Pull up / down to the target position X, pause briefly, and then retract to the original position; if the liquid level continues to deviate and the number of actions reaches the correction threshold, perform a minor adjustment. More than ±5mm Move quickly to X at 1.5 times the speed, stay there and check for liquid level rise; when the rise indicator value is 4, retract immediately and actively adjust the retraction distance; if it does not retract after 8 seconds, readjust the extension distance and repeat the action.

[0170] The alarm module includes a red warning light and a buzzer, which are electrically connected to the data calculation module. It receives liquid level deviation signals and equipment status signals (such as communication failure and electric cylinder overtravel). Warning level: Based on different fault types and the degree of liquid level deviation, a graded warning is triggered (specific levels are shown in Table 2) to remind operators to intervene in a timely manner and avoid the accident from escalating.

[0171] Table 2: Alarm Module Tiered Warning Table

[0172] Triggering conditions Warning light status Phoenix sound instrument status Processing requirements Liquid level deviation ±5mm to ±7.5mm 1 flash per second One short blast per second Monitor the situation regularly and prepare for manual intervention. Liquid level deviation exceeds ±7.5mm 3 rapid flashes per second 3 rapid blasts per second Human intervention was immediately initiated. Communication failure / electric cylinder overtravel Chang Liang Long cry In case of emergency intervention by manual intervention, the system automatically shuts off the molten steel delivery valve and resets the execution module.

[0173] The system setup of this invention is as follows:

[0174] Visual inspection module deployment: The float is placed inside the molten steel casting ladle of the amorphous ribbon forming furnace and limited by a high-temperature resistant bracket to ensure that the float can rise and fall smoothly with the liquid level; the visual recognition unit (using an industrial camera and image recognition chip) is installed at a preset height on the side of the furnace, with the lens aimed at the float, and the height of the top of the float is identified by a pre-trained intelligent algorithm, with the sampling frequency set to 10Hz; after installation, the pixel equivalent K value is calibrated by a 100mm standard liquid level block.

[0175] Data calculation module settings: PLC is used as the core calculation unit and communicates with the vision recognition unit via Ethernet; the liquid level value Ys, the maximum moving distance Mmax (calculated as Mmax=10×maximum deviation adjustment amount+10% margin), the minimum moving distance Mmin, and the anti-blocking function setting time are preset in the PLC.

[0176] Execution module connection: The execution component uses a tungsten steel rod adjustment mechanism driven by a servo electric cylinder. The electric cylinder is connected to the PLC via pulse signals. The position feedback sensor of the electric cylinder transmits the current position Xd to the PLC in real time. The mechanical origin of the electric cylinder is calibrated by a limit switch and serves as the initial retraction position reference.

[0177] Alarm module configuration: A red warning light and a buzzer are installed on the operation panel and connected to the PLC output terminal via a relay. The duration and frequency of the audible and visual signals corresponding to different abnormalities can be set. In case of communication failure or overtravel of the electric cylinder, the molten steel delivery valve will be shut off.

[0178] The working principle of this invention is as follows: the visual detection module acquires float images in real time and converts them into actual liquid level values. The data calculation module compares the actual liquid level value with the preset value to calculate the liquid level deviation. Based on the magnitude of the deviation, the corresponding action command is output to the execution module. The execution module adjusts the insertion depth of the tungsten steel rod to change the molten steel flow rate to correct the liquid level. At the same time, the data calculation module monitors the liquid level correction effect and equipment status in real time. When the liquid level deviation exceeds the threshold or the equipment malfunctions, the alarm module is triggered to issue a warning, forming a closed-loop control process of "detection, calculation, execution, feedback, and warning".

[0179] The specific workflow is as follows: After the system starts, the data calculation module loads the preset liquid level setpoints Ys, Mmax, Mmin and other parameters. The vision detection module starts to collect float images and identify the actual liquid level value Y. The data calculation module calculates the liquid level deviation ΔY=Ys-Y, matches the control logic according to the magnitude of ΔY, and outputs the target position X and action command to the execution module. The execution module drives the tungsten steel rod to move to the target position X, adjusts the molten steel flow, and feeds back its own action status (such as the current position Xd, number of actions) to the data calculation module. The vision detection module updates the actual liquid level value Y in real time. The data calculation module judges whether the liquid level has returned to the set range according to the new Y value: if it has returned, the current state is maintained; if it has not returned and the number of actions has reached the correction number, the position offset adjustment is triggered. When the liquid level deviation exceeds the preset threshold or a communication failure or electric cylinder overtravel occurs, the alarm module triggers the corresponding level of audible and visual warning. In the event of a communication failure or electric cylinder overtravel, the molten steel delivery valve is simultaneously shut off. If the execution module does not move for a long time, the anti-blocking function automatically triggers an action to prevent the tungsten steel rod from getting stuck. The cumulative triggering time is no more than 3 times within 2 minutes.

[0180] Example 1

[0181] After receiving the Y value, the PLC calculates the liquid level deviation ΔY=Ys-Y in real time, and performs logical judgments based on the current electric cylinder position Xd (obtained by the position feedback sensor) and the action count counter data.

[0182] Parameter settings are based on: Ys=60mm (set according to the requirements of amorphous ribbon forming process), Mmax=80mm (maximum deviation adjustment amount 7.5mm, calculated according to Mmax=10×7.5+10% margin), execution action correction threshold=3 times, anti-blocking time=30s.

[0183] When Y=61.8mm, ΔY=60-61.8=-1.8mm (within ±2.5mm). Using the formula X=80 / 10*(60-61.8)=-14.4mm, the target position of the electric cylinder is determined to be pulled down by 14.4mm. At the same time, it is determined that the current action count counter is 1 (not reaching the correction count of 3), so there is no need to trigger offset adjustment.

[0184] When Y=64.2mm, ΔY=+4.2mm (between ±2.5mm and ±5mm), calculate X=80 / 10*(60-64.2)=-33.6mm, the action count counter accumulates to 2 times, but still has not reached the correction count.

[0185] When Y=53.5mm, ΔY=-6.5mm (exceeding ±5mm), calculate X=80 / 10*(60-53.5)=+52mm, trigger the 1.5 times speed command, and the action count counter is incremented to 1.

[0186] Normal deviation adjustment (ΔY=±2.5mm): The PLC outputs pulse commands to the servo driver, including parameters for movement direction (pull down), distance (14.4mm), and speed (standard speed); the servo motor drives the electric cylinder to pull down from 30mm (mechanical origin, initial retracted position) to 15.6mm. After reaching the position, the driver sends a "action completed" signal to the PLC, and the PLC immediately outputs a reset command. The electric cylinder retracts to the initial position of 30mm, and the action takes 32ms; the electric cylinder position sensor feeds back the current position Xd to the PLC in real time, and the action count counter increments by 1.

[0187] Medium deviation adjustment (ΔY = ±2.5mm - ±5mm): The PLC outputs a 33.6mm pull-down command. After the electric cylinder moves to the -3.6mm position, the PLC starts a 1.5s timer. During the pause, it continuously receives the Y value update from the vision inspection module (Y = 61.5mm). After the timer ends, a reset command is output, the electric cylinder retracts to 30mm, and the action count counter is incremented to 2. After the reset, Y = 60.3mm is acquired, and ΔY is reduced to +0.3mm.

[0188] Severe deviation adjustment (ΔY = exceeding ±5mm): The PLC outputs a 1.5 times speed pull-up command of 52mm, the electric cylinder moves to the 82mm position at a speed of 750mm / s, and at the same time starts the liquid level recovery detection program; continuously receive changes in the Y value, when it detects that Y rises from 53.5mm to 57.5mm (recovery indication value = 4mm), immediately output a reset command, the electric cylinder retracts to 30.5mm (recovery indication value > 3mm, positive fine adjustment 0.5mm); if no recovery is detected or retraction is not completed within 8 seconds, the PLC recalculates X = 51mm and repeats the pull-up action.

[0189] Offset adjustment trigger: If Y is still 62.0mm after 3 consecutive adjustments under medium deviation, the action count counter reaches 3 times and the liquid level continues to rise, the PLC will automatically correct X to -34.1mm (adjustment amount = -33.6 × (1 + 0.01 × 3) = -34.608, take -34.1mm), and Y will return to 60.1mm after the 4th adjustment.

[0190] Anti-blocking mechanism: The PLC's internal timer monitors the execution module's operation status in real time. When the electric cylinder does not receive an action command for 30 consecutive seconds, it automatically outputs a 5mm reciprocating action command. The electric cylinder completes the "pull down 5mm - retract to original position" action to prevent the tungsten steel rod from jamming. After the action is completed, the timer is reset. If the anti-blocking function is triggered 3 times within 120 seconds, the anti-blocking function is paused and the equipment is prompted to be checked.

[0191] Tiered alarm:

[0192] When Y=66.5mm (ΔY=+6.5mm, ±5mm-±7.5mm), the PLC controls the warning light to flash briefly once per second, and the buzzer to sound briefly once per second, triggering a level one alarm and prompting the operator to check the status.

[0193] When Y=68.0mm (ΔY=+8.0mm, exceeding ±7.5mm), the PLC-controlled warning light flashes rapidly 3 times / second, and the buzzer sounds rapidly 3 times / second, triggering a level two alarm and requiring immediate manual intervention.

[0194] When the PLC detects a communication timeout (>500ms) with the vision inspection module, it controls the warning light to stay on, the buzzer to sound continuously, triggers a level three alarm, records fault information, shuts off the molten steel delivery valve, and suspends the execution of actions.

[0195] System closed-loop maintenance and shutdown (continuous operation and termination phases)

[0196] The above steps 1-5 are executed cyclically. The visual detection module continuously collects liquid level data, the PLC calculates and adjusts in real time, the execution module responds dynamically, and the alarm module triggers warnings as needed, forming a closed-loop control of "detection, calculation, execution, feedback, and warning".

[0197] After the production task is completed, the operator sends a stop command through the console. The PLC control execution module retracts the tungsten steel rod to the initial position (mechanical origin), shuts down the visual inspection module's acquisition function, resets the alarm module, and finally cuts off the main power supply.

[0198] Performance Verification: This embodiment operated continuously for 72 hours without human intervention, using a laser level gauge (accuracy ±0.1mm) to sample every 5 seconds, collecting a total of 51,840 data points. 99.9% of the errors were within ±0.7mm, exceeding the design target of ±1mm and the industry requirement of ±2mm. The thickness tolerance of the amorphous ribbon was reduced from the manually controlled ±0.03mm to ±0.018mm, effectively improving the yield rate.

[0199] Example 2

[0200] The similarities between this embodiment and Embodiment 1 will not be repeated here. The differences are as follows:

[0201] Parameter adjustment: The PLC pre-stored parameters are updated to Ys=40mm (to adapt to the process requirements of the small material furnace), Mmax=50mm (maximum deviation adjustment amount 4.5mm, calculated according to Mmax=10×4.5+10% margin), anti-blocking time 25s, and the recovery sign value is reduced proportionally (0.8mm when ΔY=±5mm).

[0202] The vision inspection module uses a 1920×1080 resolution industrial camera with a lens focal length of 20mm and an exposure time of 1 / 800s; the execution module uses a 50mm stroke electric cylinder with a response time of 30ms, and the mechanical origin is calibrated through a limit switch.

[0203] When adjusting for severe deviations, the maximum speed of the electric cylinder is adjusted to 600 mm / s, and the dwell time is shortened to 6 seconds, adapting to the faster liquid level changes in small material furnaces.

[0204] This embodiment is adapted to a 100L small-scale molding furnace. After 48 hours of continuous operation, the liquid level control error is stable within ±0.6mm.

[0205] Of course, the above description is not a limitation of the present invention, and the present invention is not limited to the examples given above. Any changes, modifications, additions or substitutions made by those skilled in the art within the scope of the present invention should also fall within the protection scope of the present invention.

Claims

1. A molten steel level control system for non-crystalline ribbon manufacturing, characterized by: The visual detection module, the data calculation module, the execution module and the alarm module are sequentially electrically connected. The visual detection module is used for collecting image information of the float in the steel pouring ladle and identifying the float height to obtain an actual liquid level value. The data calculation module is provided with a liquid level setting value, a liquid level deviation threshold value, an execution action correction number threshold value, a maximum movement distance of the execution module and a minimum movement distance. After receiving the actual liquid level value, the data calculation module calculates a liquid level deviation: ΔY=Ys-Y and calculates a target position of the execution module. The execution action instruction is generated according to the liquid level deviation, the current position of the execution module and the liquid level deviation direction. The action number of the execution module is counted in real time. When the action number in the same direction exceeds the correction number threshold value and the liquid level continuously deviates, the position deviation adjustment is triggered. The operation formula for calculating the target position of the execution module is X=Mmax / 10*(Ys-Y).

2. The non-crystalline ribbon manufacturing molten steel level control system according to claim 1, characterized by, In the formula, X is the target position of the execution module, Mmax is the maximum movement distance, Ys is the liquid level setting value and Y is the actual liquid level value.

3. The non-crystalline ribbon manufacturing molten steel level control system according to claim 2, characterized by, The execution module is used for receiving the instruction of the data calculation module and adjusting the flow of the molten steel. The response time is not more than 50 ms. The alarm module triggers the corresponding level of sound and light warning according to the liquid level deviation value and the equipment state signal transmitted by the data calculation module. The visual detection module includes an industrial camera and an image recognition unit. ； The shooting angle of the industrial camera is aligned with the area where the float is located. The image recognition unit identifies the float height in real time through a deep learning algorithm. The recognition accuracy is not less than 0.5 mm. The deep learning algorithm includes the following steps. S1: Preprocessing stage formula ； (1) Grayscale formula : gray value at filtered coordinates (x, y); The RGB image collected by the industrial camera is converted into a grayscale image to suppress color interference caused by molten steel reflection. Parameter definition: G(x,y): Gray value at coordinate (x,y); ； : normalized gray value; , : minimum and maximum value of the gray value in a single frame image; R(x,y), G(x,y), B(x,y): RGB three-channel pixel values (0-255) at coordinate (x,y); (2) Gaussian filter formula ； Remove image noise and retain float contour features. : value of the feature map output by the kth convolution kernel at coordinate (x, y) (k = 1, 2,..., 32); : weight parameter of the kth 3x3 convolution kernel; : a bias parameter of the kth convolution kernel; Parameter definition: σ=0.8; ； (3) Normalization formula : value of the kth feature map at coordinate (x, y) after pooling; Map the gray value to the [0,1] interval to improve the model training convergence speed. S2: Feature extraction layer core formula ； : global average pooling result of the kth high-order feature map (k = 1, 2, …, 128); (1) Convolution layer calculation formula Parameter definition: ReLU: Activation function, ReLU(z)=max(0,z); ； (2) Max pooling layer calculation formula : output of the mth neuron of the fully connected layer 1 (m = 1, 2,..., 256); : weight parameter of feature map k to neuron m; : bias parameter of neuron m; Parameter definition: Step=2, Pooling kernel size=2x2 (dimension reduction ratio 1:2); ； (3) Global average pooling layer calculation formula : Pixel length of the float exposed to the liquid surface portion of the model output; : weight parameters of fully connected layer 1 neurons m to output layer; : bias parameter of the output layer; H, W: Height and width of the Conv3 output feature map (80x60 after two times of pooling); ； S3: Fully connected layer and output layer formula (1) Fully connected layer 1 calculation Parameter definition: Dropout(p=0.2): Randomly inactivate 20% of neurons to prevent overfitting; (2) Output layer calculation Parameter definition: S4: Liquid level value conversion formula Parameter definition: Y: Actual liquid level value of molten steel (unit: mm); H0: Total height of float = 150 mm; K: Pixel equivalent (unit: pixels / mm), calibrated using a 100mm standard liquid level block, K = pixel length corresponding to the standard liquid level block / 100mm; δ: Correction value for identification error, δ∈[-0.5,+0.5].

4. The non-crystalline ribbon manufacturing molten steel level control system according to claim 3, characterized by, The execution module includes a servo motor, an electric cylinder, a tungsten steel rod, a position feedback sensor, and an anti-blocking unit. The servo motor receives the action command from the data calculation module and drives the electric cylinder to move the tungsten steel rod axially to adjust the molten steel flow rate. The position feedback sensor transmits the current position of the execution module to the data calculation module. The anti-blocking unit automatically triggers a reciprocating action when the execution module has been inactive for a set anti-blocking time.

5. The non-crystalline ribbon manufacturing molten steel level control system according to claim 4, characterized by, The alarm module has the following warning levels: when the liquid level deviation exceeds ±5mm, the red warning light flashes briefly and the buzzer sounds briefly; when the liquid level deviation exceeds ±7.5mm, the red warning light flashes rapidly and the buzzer sounds rapidly; when a communication failure or electric cylinder overtravel occurs, the red warning light stays on and the buzzer sounds continuously.

6. The non-crystalline ribbon manufacturing molten steel level control system according to claim 5, characterized by, The anti-blocking unit is set to anti-blocking time of 25-30 seconds, and the automatically triggered reciprocating motion travels 5-10 mm.

7. The control method of the molten steel level control system for amorphous ribbon manufacturing steel according to any one of claims 1 to 6, characterized by: Includes the following steps: (1) Place the float vertically into the molten steel ladle of the amorphous ribbon forming furnace, ensuring that the bottom of the float contacts the molten steel surface and the top of the float is exposed above the surface; install an industrial camera 1.5m to the side of the ladle, with the lens axis aligned with the top of the float, to collect the image information of the float in real time. The industrial camera is connected to the image recognition unit via a data cable. The image recognition unit has a built-in deep learning algorithm to process the collected float image, identify the height of the float exposed above the molten steel surface, and then convert it into the actual liquid level value Y of the molten steel; the recognition accuracy is not less than 0.5mm to ensure the accuracy of subsequent calculations; (2) Pre-stored liquid level setting value (Ys), liquid level deviation threshold, execution action correction number threshold, execution module maximum moving distance (Mmax) and minimum moving distance (Mmin), wherein the liquid level deviation threshold includes ±2.5mm, ±5mm and ±7.5mm; (3) Calculate the liquid level deviation ΔY = Ys - Y based on the actual liquid level value Y transmitted in step (1), and calculate the target position X of the execution module using the formula X = Mmax / 10 * (Ys - Y). (4) Generate execution action instructions based on the magnitude of liquid level deviation, the current position of the execution module and the direction of liquid level offset, and count the number of actions of the execution module in real time. When the number of actions in the same direction exceeds the correction threshold and the liquid level continues to deviate, trigger position offset adjustment to avoid control failure due to mechanical error. (5) Trigger the corresponding level of audible and visual alarm based on the liquid level deviation value and equipment status signal.

8. The control method of the molten steel level control system for amorphous ribbon manufacturing steel according to claim 7, characterized by, In step (4), when the liquid level deviation is within ±2.5mm, the control execution module performs an upward or downward operation and retracts immediately after reaching the telescopic distance; if the liquid level continues to deviate in one direction and the number of actions of the execution module reaches the preset correction number, position offset adjustment is triggered; when the liquid level deviation is between ±2.5mm and ±5mm, the control execution module pauses briefly after completing the upward or downward operation and retracts back to its original position; if the liquid level continues to deviate and the number of actions reaches the preset correction number, position offset adjustment is triggered; when the liquid level deviation exceeds ±5mm, the control execution module moves rapidly at 1.5 times the speed, pauses continuously after reaching the telescopic distance and detects signs of liquid level recovery. When the recovery sign value is 4mm, it retracts immediately and actively adjusts the retraction distance after retraction; if it does not retract after more than 8 seconds, the telescopic distance is adjusted and the upward and downward operation is repeated; the recovery sign value is the number of millimeters the liquid level rises from the deviation state to the set value; the active adjustment of the retraction distance is based on the liquid level recovery sign value, with a slight adjustment of ±0.5-1mm on the basis of the initial retraction position.

Images