RH one-key refining system based on machine vision

Abstract

The invention discloses an RH one-key refining system based on machine vision, and belongs to the technical field of metallurgical engineering and intelligent manufacturing, the RH one-key refining system based on machine vision comprises a data acquisition module, a data processing module, a data storage module, a splashing prediction module and an automatic regulation and control module, the data acquisition module is used for acquiring images before and after splashing in a vacuum tank of the RH furnace and parameters of the RH furnace in an operation process, the data processing module is used for extracting texture features of the images of the vacuum tank of the RH furnace, and the data storage module is used for storing all acquired data. The splashing prediction module is used for recognizing and predicting the splashing phenomenon by building a splashing prediction model, the automatic regulation and control module is used for regulating the RH circulation flow, the oxygen supply flow and the vacuum degree, and real-time monitoring and prediction of splashing are achieved by building the splashing prediction model and replacing manual observation based on the actual working condition.

Description

Technical Field

[0001] The present invention relates to the technical field of metallurgical engineering and intelligent manufacturing, and in particular to a one-key RH refining system based on machine vision. Background Art

[0002] RH vacuum molten steel circulation degassing method is a secondary refining process, mainly used to produce high-quality molten steel. This method was jointly developed by Ruhr Steel and Hellas, both subsidiaries of Germany's Thyssen Company. Its basic principle is to circulate molten steel in a vacuum chamber, blow argon through the riser of the immersion tube, so that the molten steel enters the vacuum chamber and flows back to the steel ladle from the downcomer, and repeat this process to remove gases such as hydrogen and inclusions in the molten steel. The RH process is particularly suitable for modern smelting and continuous casting, and can produce ultra-low carbon high-quality steel grades. With the development and development of vacuum steelmaking technology, VD furnaces and RH furnaces eventually became the mainstream of this technology.

[0003] The current problems are as follows: First, during the vacuum decarburization process, splashed slag steel blocked the camera hole, and the operator could not see the metallurgical reaction in the vacuum tank; slag steel stuck to the top gun hole, which hindered the lifting of the top gun, making it difficult to blow oxygen, and slag accumulation at the gun head made it difficult to ignite the oxygen gun; slag steel accumulated in the hot bend pipe affected the vacuum function and decarburization effect, resulting in frequent replacement of hot bend pipes; secondly, most steel mills currently use manual monitoring of on-site detection instruments and monitoring cameras installed in vacuum tanks to determine the carbon-oxygen reaction rate and splashing of RH smelting, and control RH smelting by adjusting the RH oxygen supply, circulating lifting gas, vacuum degree, scrap steel addition intensity, etc., to achieve both control of excessive splashing of molten steel and guarantee of decarburization speed, and manual operation makes production The operation is not standardized, the smelting effect is unstable, and manual operation is affected by the operator's experience, technology and mental concentration. There are differences in the judgment and operation of each programmer. Manual detection and judgment are affected by subjective factors and fatigue. In addition, the labor intensity is high, and the operator needs to monitor and adjust the equipment for a long time, which increases the risk of operational errors. Finally, a few steel companies have developed a one-click RH refining system to control the splashing of the process by presetting the RH vacuum degree, oxygen gun position and oxygen flow rate, circulation flow rate, scrap steel alloy addition amount and addition time at different smelting stages. However, the preset mode often cannot keep up with the changes in the RH smelting process, and the smelting control has not formed a closed loop. Once splashing occurs during the smelting process, the operator needs to intervene manually. Summary of the invention

[0004] The object of the present invention is to provide a one-key RH refining system based on machine vision to solve the problems raised in the above background technology.

[0005] To achieve the above-mentioned purpose, the present invention provides the following technical solutions: a one-key RH refining system based on machine vision, the system comprising: a data acquisition module, a data processing module, a data storage module, a splash prediction module and an automatic control module, the data acquisition module, the data processing module and the splash prediction module are all connected to the data storage module, the automatic control module is connected to the splash prediction model, the data acquisition module is used to obtain images before and after splashing in the RH furnace vacuum tank and parameters of the RH furnace during operation, the data processing module is used to extract texture features of the RH furnace vacuum tank image, the data storage module is used to store all acquired data, the splash prediction module is used to construct a splash prediction model to identify and predict splashing phenomena, and the automatic control module adjusts the RH circulation flow, oxygen supply flow and vacuum degree according to the prediction results of the splash prediction module.

[0006] Furthermore, the data acquisition module includes an image acquisition unit, a physical parameter acquisition unit and a data preprocessing unit, the image acquisition unit is used to acquire images before and after splashing occurs in the vacuum tank of the RH furnace, the images before splashing of the molten steel in the vacuum tank of the RH furnace include images of the molten steel liquid level in the vacuum tank, the physical parameter acquisition unit is used to acquire the vacuum degree, circulation flow, oxygen lance oxygen blowing flow and time data of the RH furnace, and the data preprocessing unit is used to perform grayscale preprocessing on the images before and after splashing occurs in the vacuum tank, wherein the images before and after splashing in the vacuum tank of the RH furnace are acquired by an industrial camera, the camera is mounted on the outer wall of the vacuum tank and fixed by a flange connection, which enhances the stability of the camera, reduces image blurring caused by equipment vibration or collision, and ensures the quality of image data, in addition, the industrial camera is placed in a heat-insulating protective cover and its shooting orientation is ensured to be consistent with the orientation and angle of manual fire viewing, thereby improving the convenience and safety of operation, and through this configuration, the data acquisition system is automated and the need for manual operation is reduced.

[0007] Furthermore, the image of the RH furnace vacuum tank before splashing occurs is acquired frame by frame in the form of video, and a 32*32 pixel area in the middle of the captured image is taken as input.

[0008] Furthermore, the data processing module includes a feature extraction unit and a feature vector construction unit. The feature extraction unit convolves the molten steel level image in the vacuum tank to obtain a response matrix, and calculates specific statistics on the filter response matrix, including: mean, variance, maximum and minimum values. The statistics calculated for the filter response constitute local features of the image, and these features constitute texture features of the molten steel level image in the vacuum tank of the RH furnace. By applying convolution at different scales and directions, the texture features of the image can be more comprehensively analyzed, and the mean, variance, maximum and minimum values ​​are selected as statistics. These statistics can describe the distribution characteristics of the filter response and help capture the local features of the image.

[0009] Furthermore, the feature vector construction unit constructs the splash feature vector based on the texture features of the molten steel surface image in the vacuum tank of the RH furnace, the vacuum degree of the RH furnace, the circulation flow rate, the oxygen lance oxygen blowing flow rate and time data.

[0010] Furthermore, the splash prediction module includes a splash prediction model building unit and a splash prediction model training unit, the splash prediction model building unit is used to construct a splash prediction model, the splash prediction model includes: an input layer, an RNN layer, a fully connected layer and an output layer, wherein the input layer is used to receive time series data of image texture features and physical parameters, the RNN layer is used to process the time-dependent features of image texture features and physical parameters, the fully connected layer integrates the output features of the RNN layer and generates predictions, and the output layer is used to output the probability of splashing, and the probability value of splashing occurs ranges from 0 to 1. By constructing a splash prediction model and using the output layer to output the probability of splashing, an intuitive indicator is provided for assessing splashing risks, which helps to take corresponding preventive measures in advance.

[0011] Furthermore, the training splash prediction model unit divides the data set into a training set, a validation set and a test set. The data set includes historical splash data and its corresponding splash feature vector. First, the training set data is input into the splash prediction model for training. After passing through the RNN layer and the fully connected layer, the splash prediction value output by the model is obtained. Secondly, the splash prediction model training is iterated, and the splash label is set. The label of splash occurrence is set to 1, and the label of splash non-occurrence is set to 0, which is used to represent the binary classification problem. In the splash model training process, the loss function is defined, and the calculation formula of the loss function is as follows:

[0012]

[0013] Among them, Loss represents the difference between the splash probability predicted by the splash prediction model and the actual result, N represents the number of all training samples, and y i represents the splash label of the i-th sample, p irepresents the probability that the splash model predicts that the i-th sample belongs to splash;

[0014] Next, through back propagation, the BPTT algorithm is used to back propagate the error along the time dimension, and the weights of the RNN layer and the fully connected layer are updated through the gradient descent algorithm. The calculation formula for updating the weights is as follows:

[0015]

[0016] Among them, W new The weight value adjusted by the gradient descent algorithm, W old represents the current weight value, η represents the learning rate, which determines the step size of the weight update. Represents the gradient calculated by back propagation, the influence of reaction weight on loss function. The splash prediction model training adopts an iterative method. By defining the loss function to measure the difference between the predicted value and the actual value, the model training direction is guided. The weight value adjusted by the gradient descent algorithm is updated by giving the calculation method of updating the weight, thereby improving the prediction accuracy.

[0017] Finally, the validation set is used to evaluate the performance of the splash prediction model, and the splash prediction model is optimized by adjusting the learning rate hyperparameter to control the step size of each update. When the step size is too large, the splash prediction model will skip the optimal solution and fail to converge; when the step size is too small, the splash prediction model will converge very slowly. Gradient descent combines the learning rate and gradient value to ensure that the parameters can gradually approach the optimal value.

[0018] Furthermore, the automatic control module includes a unit for reducing the RH circulation flow rate, a unit for reducing the oxygen supply intensity, and a unit for reducing the vacuum degree. The automatic control module is used to send a control action start signal and the RH circulation target flow rate, target oxygen supply flow rate, and target vacuum degree to the PLC when the splashing prediction value exceeds the set threshold value. The PLC controls the equipment to complete the action according to the target value, including: reducing the RH circulation flow rate, reducing the oxygen supply intensity and reducing the vacuum degree to reduce the occurrence of splashing. According to the prediction results of the splashing prediction model, the circulation flow rate, oxygen supply intensity and vacuum degree are adjusted. By working in collaboration with the PLC, the automatic control module can realize automatic control of the operating parameters of the RH furnace, thereby reducing the occurrence of splashing.

[0019] Compared with the prior art, the beneficial effects achieved by the present invention are:

[0020] 1. The present invention proposes a one-key RH refining system based on machine vision. On the basis of the traditional one-key RH refining system, the system realizes intelligent recognition of abnormal splashing state of RH smelting by deep learning of high-definition image signals on the vacuum tank. On this basis, the oxygen supply, circulating lifting gas and vacuum smelting operations of RH are controlled to prevent splashing and realize closed-loop control of one-key RH refining.

[0021] 2. The present invention realizes intelligent and fully automated closed-loop control of RH smelting on the basis of intelligent perception, improves the full automation and intelligence of RH one-key refining, reduces the labor intensity of operators, and on this basis can improve the standardization of RH refining, reduce the influence of operator experience, technology and mental concentration on operation, improve smelting operation effect, and reduce splashing phenomenon of RH smelting.

[0022] 3. Monitor changes in the smelting process in real time and automatically adjust parameters through feedback, so that the system can dynamically optimize operations based on actual conditions, rather than relying on the initial set values ​​set by the preset mode. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 It is a structural diagram of a one-key RH refining system based on machine vision of the present invention. DETAILED DESCRIPTION

[0024] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0025] like Figure 1 As shown, the present invention provides a technical solution: a RH one-key refining system based on machine vision, the system comprising: a data acquisition module, a data processing module, a data storage module, a splash prediction module and an automatic control module, the data acquisition module, the data processing module and the splash prediction module are all connected to the data storage module, the automatic control module is connected to the splash prediction model, the data acquisition module is used to obtain images before and after splashing in the RH furnace vacuum tank and parameters of the RH furnace during operation, the data processing module is used to extract texture features of the RH furnace vacuum tank image, the data storage module is used to store all acquired data, the splash prediction module is used to construct a splash prediction model, identify and predict splashing phenomena, and the automatic control module adjusts the RH circulation flow, oxygen supply flow and vacuum degree according to the prediction results of the splash prediction module.

[0026] The data acquisition module includes an image acquisition unit, a physical parameter acquisition unit and a data preprocessing unit. The image acquisition unit is used to acquire images before and after splashing occurs in the vacuum tank of the RH furnace. The images before splashing of molten steel in the vacuum tank of the RH furnace include images of the molten steel liquid level in the vacuum tank. The physical parameter acquisition unit is used to acquire the vacuum degree, circulation flow rate, oxygen lance oxygen blowing flow rate and time data of the RH furnace. The data preprocessing unit is used to perform grayscale preprocessing on the images before and after splashing occurs in the vacuum tank. The images before and after splashing in the vacuum tank of the RH furnace are acquired by an industrial camera. The camera is mounted on the outer wall of the vacuum tank and fixed by a flange connection, which enhances the stability of the camera, reduces image blurring caused by equipment vibration or collision, and ensures the quality of image data. In addition, the industrial camera is placed in a heat-insulating protective cover and its shooting orientation is ensured to be consistent with the orientation and angle of manual fire viewing, which improves the convenience and safety of operation. Through this configuration, the data acquisition system is automated and the need for manual operation is reduced.

[0027] The image of the RH furnace vacuum tank before splashing occurs is acquired frame by frame in the form of video, and the 32*32 pixel area in the middle of the captured image is taken as input.

[0028] The data processing module includes a feature extraction unit and a feature vector construction unit. The feature extraction unit performs convolution on the molten steel level image in the vacuum tank to obtain response matrices at different scales and directions, and calculates specific statistics on the filter response matrix, including: mean, variance, maximum and minimum values. The statistics calculated for the filter response constitute the local features of the image, and these features constitute the texture features of the molten steel level image in the vacuum tank of the RH furnace. By applying convolution at different scales and directions, the texture features of the image can be more comprehensively analyzed, and the mean, variance, maximum and minimum values ​​are selected as statistics. These statistics can describe the distribution characteristics of the filter response and help capture the local features of the image.

[0029] The feature vector construction unit constructs a splash feature vector based on the texture features of the molten steel surface image in the vacuum tank of the RH furnace, the vacuum degree of the RH furnace, the circulation flow rate, the oxygen lance oxygen flow rate and time data.

[0030] The splash prediction module includes a splash prediction model building unit and a splash prediction model training unit. The splash prediction model building unit is used to build a splash prediction model. The splash prediction model includes: an input layer, an RNN layer, a fully connected layer and an output layer. The input layer is used to receive time series data of image texture features and physical parameters, the RNN layer is used to process the time-dependent features of image texture features and physical parameters, the fully connected layer integrates the output features of the RNN layer and generates predictions, and the output layer is used to output the probability of splashing, and the probability value of splashing occurs ranges from 0 to 1. By building a splash prediction model and using the output layer to output the probability of splashing, an intuitive indicator is provided for assessing splashing risks, which helps to take corresponding preventive measures in advance.

[0031] The splash prediction model training unit divides the data set into a training set, a validation set, and a test set. The data set includes historical splash data and its corresponding splash feature vector. First, the training set data is input into the splash prediction model for training. After passing through the RNN layer and the fully connected layer, the splash prediction value output by the model is obtained. Secondly, the splash prediction model training is iterated, and the splash label is set. The label of splash occurrence is set to 1, and the label of splash non-occurrence is set to 0, which is used to represent the binary classification problem. In the splash model training process, the loss function is defined. The calculation formula of the loss function is as follows:

[0032]

[0033] Among them, Loss represents the difference between the splash probability predicted by the splash prediction model and the actual result, N represents the number of all training samples, and y i represents the splash label of the i-th sample, p i represents the probability that the splash model predicts that the i-th sample belongs to splash;

[0034] Next, through back propagation, the BPTT algorithm is used to back propagate the error along the time dimension, and the weights of the RNN layer and the fully connected layer are updated through the gradient descent algorithm. The calculation formula for updating the weights is as follows:

[0035]

[0036] Among them, W new The weight value adjusted by the gradient descent algorithm, W old represents the current weight value, η represents the learning rate, which determines the step size of the weight update. Represents the gradient calculated by back propagation, the influence of reaction weight on loss function. The splash prediction model training adopts an iterative method. By defining the loss function to measure the difference between the predicted value and the actual value, the model training direction is guided. The weight value adjusted by the gradient descent algorithm is updated by giving the calculation method of updating the weight, thereby improving the prediction accuracy.

[0037] Finally, the validation set is used to evaluate the performance of the splash prediction model, and the splash prediction model is optimized by adjusting the learning rate hyperparameter to control the step size of each update. When the step size is too large, the splash prediction model will skip the optimal solution and fail to converge; when the step size is too small, the splash prediction model will converge very slowly. Gradient descent combines the learning rate and gradient value to ensure that the parameters can gradually approach the optimal value.

[0038] The automatic control module includes a unit for reducing the RH circulation flow rate, a unit for reducing the oxygen supply intensity, and a unit for reducing the vacuum degree. The automatic control module is used to send a control action start signal and the RH circulation target flow rate, the target oxygen supply flow rate, and the target vacuum degree to the PLC when the splashing prediction value exceeds the set threshold value. The PLC controls the equipment to complete the action according to the target value, including: reducing the RH circulation flow rate, reducing the oxygen supply intensity and reducing the vacuum degree to reduce the occurrence of splashing. According to the prediction results of the splashing prediction model, the circulation flow rate, oxygen supply intensity and vacuum degree are adjusted. By working in collaboration with the PLC, the automatic control module can realize automatic control of the operating parameters of the RH furnace, thereby reducing the occurrence of splashing.

[0039] In the embodiment: an industrial camera is used to monitor the molten steel level in the vacuum tank of the RH furnace in real time, and 5 frames of images are collected per second, each frame of which is 640×480 pixels. The collection time is 10 minutes before and after the molten steel treatment, and the physical parameters of the RH furnace are collected, including vacuum degree: from 0.1Pa to 1000Pa, circulation flow: from 50m 3 / h to 500m 3 / h, oxygen lance oxygen flow rate: from 1000Nm 3 / h to 5000Nm 3 / h and time data: physical parameters are collected once per second, for a total of 600 times, and the splash prediction value threshold is set to 0.8. The collected images are grayed and the image format is unified to 32×32 pixels. The convolutional neural network CNN is used to extract features from the images. CNN consists of 3 convolutional layers and 2 pooling layers: the first convolution kernel size is 3×3, which extracts low-level texture features and generates a feature map of size 30×30. The second pooling layer (2×2) is further downsampled to generate a feature map of size 15×15. The feature vector generated at last is a feature value of length 256, which is used as the input of the subsequent splash prediction model. The collected vacuum degree, circulation flow rate, and oxygen flow rate of the oxygen gun are used as time series input. The time series is processed by LSTM to capture the key features before splashing. The time step of LSTM is set to 600. The fully connected layer integrates the output features of LSTM and finally generates a splash probability value of 0.85. The splash prediction model uses the loss function and the update weight formula for Training, each training sample contains 10 minutes of data, and the training data contains 100 RH furnace operation data, of which 50 splashing occurred and 50 splashing did not occur. In each iteration, the model calculates the loss and adjusts the model parameters according to the gradient of the loss. The initial learning rate is set to 0.001. Through training iterations, it is determined that when the learning rate η is 0.0008, the training effect of the splashing prediction model is the best. At this time, during the system test, the model predicts 20 RH furnace operations, and the system predicts 18 splashings with an accuracy of 90%. It is judged that the splashing prediction value is greater than the set threshold of 0.8. At this time, when the model predicts that the probability of splashing is higher than 0.8, the circulation flow rate is automatically reduced, and the target value is 70% of the current flow rate; the oxygen supply intensity is reduced, and the target is 60% of the current flow rate; the vacuum degree is adjusted to reduce the risk of splashing, and the vacuum degree is adjusted from 1000Pa to 300Pa. The automatic control module sends the target flow rate and vacuum degree to the PLC, and the PLC control equipment makes corresponding adjustments.

[0040] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

Claims

1. A one-key RH refining system based on machine vision, characterized in that: The system includes: a data acquisition module, a data processing module, a data storage module, a splash prediction module and an automatic control module. The data acquisition module, the data processing module and the splash prediction module are all connected to the data storage module. The automatic control module is connected to the splash prediction model. The data acquisition module is used to obtain images before and after splashing in the RH furnace vacuum groove and parameters of the RH furnace during operation. The data processing module is used to extract texture features of the RH furnace vacuum groove image. The data storage module is used to store all acquired data. The splash prediction module is used to construct a splash prediction model to identify and predict splashing phenomena. The automatic control module adjusts the RH circulation flow, oxygen supply flow and vacuum degree according to the prediction results of the splash prediction module.

2. The one-key RH refining system based on machine vision according to claim 1, characterized in that: The data acquisition module includes an image acquisition unit, a physical parameter acquisition unit and a data preprocessing unit. The image acquisition unit is used to acquire images before and after splashing occurs in the vacuum tank of the RH furnace. The images before splashing of molten steel in the vacuum tank of the RH furnace include images of the molten steel level in the vacuum tank. The physical parameter acquisition unit is used to acquire the vacuum degree, circulation flow, oxygen lance oxygen blowing flow and time data of the RH furnace. The data preprocessing unit is used to perform grayscale preprocessing on the images before and after splashing occurs in the vacuum tank.

3. The one-key RH refining system based on machine vision according to claim 2, characterized in that: The image of the RH furnace vacuum tank before splashing occurs is acquired frame by frame in the form of video, and a 32*32 pixel area in the middle of the captured image is taken as input.

4. The one-key RH refining system based on machine vision according to claim 1, characterized in that: The data processing module includes a feature extraction unit and a feature vector construction unit. The feature extraction unit convolves the molten steel level image in the vacuum tank to obtain a response matrix, and calculates specific statistics on the filter response matrix, including: mean, variance, maximum value and minimum value. The statistics calculated for the filter response constitute local features of the image, and these features constitute the texture features of the molten steel level image in the vacuum tank of the RH furnace.

5. The one-key RH refining system based on machine vision according to claim 4, characterized in that: The feature vector construction unit constructs a splash feature vector based on the texture features of the molten steel surface image in the vacuum tank of the RH furnace, the vacuum degree of the RH furnace, the circulation flow rate, the oxygen lance oxygen blowing flow rate and time data.

6. The one-key RH refining system based on machine vision according to claim 1, characterized in that: The splash prediction module includes a splash prediction model building unit and a splash prediction model training unit. The splash prediction model building unit is used to build a splash prediction model. The splash prediction model includes: an input layer, an RNN layer, a fully connected layer and an output layer. The input layer is used to receive time series data of image texture features and physical parameters, the RNN layer is used to process the time-dependent features of image texture features and physical parameters, the fully connected layer integrates the output features of the RNN layer and generates predictions, and the output layer is used to output the probability of splashing, and the probability value of splashing occurs ranges from 0 to 1.

7. The one-key RH refining system based on machine vision according to claim 6, characterized in that: The training splash prediction model unit divides the data set into a training set, a validation set and a test set. The data set includes historical splash data and its corresponding splash feature vector. First, the training set data is input into the splash prediction model for training. After passing through the RNN layer and the fully connected layer, the splash prediction value output by the model is obtained. Secondly, the splash prediction model training is iterated, and the splash label is set. The label of splash occurrence is set to 1, and the label of splash non-occurrence is set to 0, which is used to represent the binary classification problem. In the splash model training process, the loss function is defined. The calculation formula of the loss function is as follows: Among them, Loss represents the difference between the splash probability predicted by the splash prediction model and the actual result, N represents the number of all training samples, and y i represents the splash label of the i-th sample, p i represents the probability that the splash model predicts that the i-th sample belongs to splash; Next, through back propagation, the BPTT algorithm is used to back propagate the error along the time dimension, and the weights of the RNN layer and the fully connected layer are updated through the gradient descent algorithm. The calculation formula for updating the weights is as follows: Among them, W new The weight value adjusted by the gradient descent algorithm, W old represents the current weight value, η represents the learning rate, which determines the step size of the weight update. Represents the gradient calculated by back propagation, reflecting the impact of weights on the loss function; Finally, the validation set is used to evaluate the performance of the splash prediction model, and the splash prediction model is optimized by adjusting the learning rate hyperparameter to control the step size of each weight update.

8. The one-key RH refining system based on machine vision according to claim 1, characterized in that: The automatic control module includes a unit for reducing the RH circulation flow rate, a unit for reducing the oxygen supply intensity, and a unit for reducing the vacuum degree. The automatic control module is used to send a control action start signal and the RH circulation target flow rate, the target oxygen supply flow rate, and the target vacuum degree to the PLC when the splashing prediction value exceeds the set threshold value. The PLC controls the equipment to complete the action according to the target value, including: reducing the RH circulation flow rate, reducing the oxygen supply intensity and reducing the vacuum degree to reduce the occurrence of splashing.

Images