Continuous casting robot device for clamping automatic measuring rod and AI identification method
Through the continuous casting robot device and AI recognition method of automated measuring rod clamping, the problems of cumbersome connection methods and the influence of the steel slag layer are solved, efficient and accurate liquid level height detection is achieved, and the automation of the continuous casting casting process is improved. level.
Patent Information
- Application Number
- CN202410814111.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-24
- Publication Date
- 2024-11-15
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
In the existing continuous casting cast steel technology, the connection method between the robot and the connector is cumbersome, resulting in low measurement efficiency; the steel slag layer and rolled slag affect the stability of the liquid measuring rod and reduce the detection accuracy.
A continuous casting robot device for automatic measurement rod clamping is designed, using robotic arm assembly and clamping hand assembly, combining infrared imager and AI recognition method to realize automatic clamping of liquid measuring rod and liquid level height detection, through servo The parallel components of the electric cylinder and pneumatic jaws are adjusted in universal angle to meet different environments and size requirements.
Improves measurement flexibility and accuracy, reduces manual intervention, enhances the stability of the system and the ability to adapt to complex environments, and ensures the accuracy and efficiency of measurement.
Smart Images

Figure CN118950971A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of continuous casting steel pouring, specifically to measuring tundishes of different heights or sizes, and particularly to a continuous casting robot device and an AI recognition method for automatic measuring rod clamping. Background Art
[0002] The control accuracy of the molten steel level has a direct impact on the quality of the ingot. If there is an error in the control of the liquid level, the molder slag and inclusions may be brought into the molten steel, which will not only reduce the quality of the ingot, but may also cause major production accidents such as steel leakage and overflow. By accurately controlling the weight of the tundish to maintain the stability of the molten steel level, the continuity and stability of the continuous casting process can be ensured, thereby improving production efficiency.
[0003] At present, the detection technology of continuous casting steel pouring technology using robots combined with liquid measuring rods does have two major defects, which have a significant impact on the efficiency and accuracy of measurement:
[0004] (1) Limitation of the connection method between the robot and the connector: In the existing robot device for measuring the liquid level of the tundish, the connection method between the robot and the connector is mainly fixed connection, or the disassembly process is relatively cumbersome. When it is necessary to measure tundishes of different heights or sizes, this connection method requires technicians to spend more time replacing parts or disassembling and reassembling, thereby reducing the efficiency of measurement.
[0005] (2) The influence of slag layer and slag curling on the liquid measuring rod: During the continuous casting process, a slag layer or slag curling may form on the surface of the molten steel. These slag layers or slag curling will affect the stability of the liquid measuring rod in the molten steel, causing the liquid measuring rod to be unevenly stressed and tilted. The tilt of the liquid measuring rod will directly affect the accuracy of its measurement results. Because the tilt will cause the liquid level measured by the liquid measuring rod to deviate from the actual liquid level, thereby reducing the accuracy and effect of the detection.
[0006] Therefore, a continuous casting robot device and an AI recognition method for automated measuring rod clamping are proposed. Summary of the invention
[0007] In view of this, the embodiment of the present invention hopes to provide a continuous casting robot device and an AI recognition method for automatic measuring rod clamping to solve or alleviate the technical problems existing in the prior art, that is, how to solve the connection mode limitation between the robot and the connector, and solve the influence of the molten steel slag layer and slag roll on the liquid measuring rod, and at least provide a beneficial option for this; the technical solution of the present invention is implemented as follows:
[0008] First, a continuous casting robot device for automated measuring rod clamping:
[0009] It consists of a robot arm assembly with a gripper assembly at the end for holding a dipstick, a traditional tool used to detect the level of the steel ladle in the continuous casting process.
[0010] During operation, the robot arm assembly is first driven to the top of the placement rack where liquid measuring rods of different specifications are stored. The gripper assembly picks up the specified liquid measuring rod from the placement rack according to the preset task requirements and moves it to the top of the tundish. The tundish contains the covering agent and molten steel.
[0011] Next, the robotic arm assembly adjusts the position of the gripper assembly and its liquid measuring rod to ensure that the liquid measuring rod remains perpendicular to the liquid surface of the molten steel. Then, the gripper assembly slowly inserts the liquid measuring rod into the molten steel. During the insertion process, if the liquid measuring rod is subjected to uneven force due to the slag layer and slag curl of the molten steel, the gripper assembly will perform a reverse universal angle adjustment compensation according to the change in the force direction to ensure that the liquid measuring rod is always perpendicular to the liquid surface of the molten steel.
[0012] In one embodiment, the infrared imager can capture the temperature field distribution and temperature pixels of the molten steel in the tundish. The infrared imager receives the infrared radiation emitted by the molten steel in real time through its highly sensitive infrared detector and converts it into image data of the temperature field distribution and temperature pixels.
[0013] In one embodiment, the design of the robotic arm assembly adopts a structure combining a multi-axis rotary actuator and an arm body to achieve multi-dimensional flexible operation. Specifically, the robotic arm assembly consists of the following parts:
[0014] 1) The first rotary actuator 1 and the first arm body: The first rotary actuator is arranged along the Y-axis direction and is used to drive the first arm body to rotate on the horizontal plane. This design enables the robot arm assembly to achieve a wide range of movement and positioning in the Y-axis direction.
[0015] 2) Second rotary actuator and second arm: The second rotary actuator is mounted on the first arm and arranged along the X-axis. It is responsible for driving the second arm to rotate on the extension line of the first arm. This design increases the flexibility of the robot assembly in the X-axis direction, enabling more precise positioning and operation.
[0016] 3) The third rotary actuator and the third arm: The third rotary actuator is mounted on the second arm and arranged along the Z-axis. It is responsible for driving the third arm to perform vertical rotational motion. This design enables the robotic arm assembly to achieve full and flexible operation in three-dimensional space.
[0017] 4) End effector: The end effector is installed at the end of the third arm and belongs to a frame-like structure. By cooperating with the above-mentioned rotary actuator and arm, the end effector can perform precise positioning and operation in a complex working environment.
[0018] In one embodiment, the clamping hand assembly includes a pneumatic clamp for clamping the liquid measuring rod, and an upper frame and a lower frame mounted on the lower surface of the end effector and the upper surface of the pneumatic clamp;
[0019] A parallel assembly is provided between the upper frame and the lower frame, and the parallel assembly is used to control the lower frame and the pneumatic clamp to adjust the universal angle. The pneumatic clamp is connected to an external compressed air source by an electric-controlled air valve for air supply.
[0020] The lower frame is evenly provided with three level sensors in the form of a circular array. When the liquid measuring rod is inserted into the molten steel, the slag layer and slag roll of the molten steel will cause the liquid measuring rod to be unevenly stressed, which will in turn affect the angle deviation of the lower frame and the pneumatic clamp. The level sensor at the nearest neighboring position will sense its own tilt state, and then feedback the parallel component to perform universal angle adjustment on the lower frame and the pneumatic clamp to achieve compensation.
[0021] In one embodiment, the parallel assembly is composed of six linear actuators, which are arranged in a circular array between the upper frame and the lower frame.
[0022] Linear actuators are devices that convert electrical energy or other forms of energy into linear motion. They are evenly distributed between the upper and lower frames to form a circular array, ensuring that the lower frame and its pneumatic gripper can achieve precise universal angle adjustment.
[0023] By arranging the linear actuators in a circular array, they can evenly apply force to the lower frame, thus achieving smoother and more precise angle adjustment. This design also improves the structural strength and stability of the entire gripper assembly.
[0024] In one embodiment, the linear actuator is a servo electric cylinder, and there are six of them in total; the cylinder body and piston rod of the servo electric cylinder are universally hinged to the mutually opposite sides of the upper frame and the lower frame through universal joint couplings. When any servo electric cylinder is actuated, the universal joint coupling converts the linear motion into rotational motion, so as to achieve the angle adjustment of one position of the lower frame.
[0025] In one embodiment, three adjacent servo electric cylinders are arranged in an N-shaped arrangement. This inclined arrangement ensures that when any one servo electric cylinder reaches the limit stroke position, the remaining servo electric cylinders still have stroke space to achieve stress compensation.
[0026] Second, AI recognition method for automated measuring rod clamping:
[0027] In the above device, the infrared imager measures the temperature field of the liquid measuring rod that has reached the measuring position, and the data is processed by the image processing computer and transmitted to the PLC controller. This method aims to introduce an AI recognition method to the infrared imager in the device, and use the U-Net semantic segmentation model to identify the position and range of the molten steel level in the image, so as to achieve more intelligent management and application. It includes the following steps S1 to S4.
[0028] Step S1, data preparation:
[0029] Collect infrared image data i containing the molten steel surface, and mark the position a and range b of the surface to form a training data set D and sample i i (a i ,b i ):
[0030] D=[i 1 (a 1 ,b 1 ),i 2 (a 2 ,b 2 ),...,i 1 (a n ,b n )]
[0031] Where n is the sample i i (a i ,b i )
[0032] We can also use the sample i i (a i ,b i ) to perform necessary preprocessing, such as contrast enhancement and noise removal, to improve image quality. At the same time, according to the input requirements of the U-Net model, the image is adjusted to a uniform size and format.
[0033] Step S2, read the U-Net semantic segmentation model:
[0034] Use the TensorFlow deep learning framework to read the U-Net semantic segmentation model.
[0035] Then you need to train the U-Net semantic segmentation model, which includes the following steps:
[0036] P1. Divide the labeled training data set D into training set, validation set and test set.
[0037] P2. Set the cross entropy loss function and Adam optimizer.
[0038] The cross-entropy loss function is used to measure the difference between the probability distribution predicted by the model and the true distribution. In the semantic segmentation task, the category of each pixel is predicted and the difference between it and the true label is calculated:
[0039]
[0040] Where N is the total number of pixels, C is the total number of categories, and y i,c is the true label (1 if the i-th pixel belongs to category c, 0 otherwise), p i,c is the probability that the model predicts that the pixel belongs to category c.
[0041] The Adam optimizer (Adaptive Moment Estimation) is an optimization algorithm based on gradient descent. It adjusts the learning rate of each parameter by calculating the first-order moment estimate and the second-order moment estimate of the gradient. The Adam optimizer combines the advantages of the AdaGrad and RMSProp optimization algorithms, and can apply different learning rates to different parameters, while having good convergence speed and robustness. The update rule of the Adam optimizer is:
[0042] m t =β 1 m t-1 +(1-β 1 ) t ;
[0043] v t =β 2 v t-1 +(1-β 2 ) t 2 ;
[0044]
[0045] Among them, m t and v t are the first-order moment estimate and the second-order moment estimate of the gradient, and is its updated value; g t is the current gradient, β 1 and β 2 is a hyperparameter (close to 1), η is the learning rate, ∈ is a small number (used to prevent division by zero), and θ t is the current parameter value, θ t+1 is the updated parameter value.
[0046] P3. Train the U-Net model on the training set and validate it on the validation set to adjust hyperparameters and prevent overfitting.
[0047] By continuously iterating the training process, the model's hyperparameters (such as learning rate, weight decay, etc.) can be gradually adjusted to find the parameter combination that optimizes the model's performance on the validation set. At the same time, overfitting can be prevented by using regularization techniques (such as Dropout, L1 / L2 regularization, etc.) or early stopping methods.
[0048] Step S3, deploying the U-Net semantic segmentation model: deploying the trained U-Net model to the system where the infrared imager 5 is located.
[0049] The image processing computer receives the actual samples collected by the infrared imager i (a i ,b i )' and input it into the U-Net model for inference.
[0050] The model outputs the segmentation result of the liquid surface area, that is, identifies the position a' and range b' of the molten steel surface in the image.
[0051] Step S4, system integration: Integrate the inference results of the U-Net model with the PLC controller to achieve real-time monitoring and control of the molten steel level.
[0052] Compared with the prior art, the present invention has the following beneficial effects:
[0053] 1. High flexibility and adaptability: The parallel assembly consisting of the servo electric cylinder and the pneumatic gripper of the present invention allows the gripper assembly to achieve precise angle adjustment in multiple directions, thereby adapting to the requirements of different working environments and connector sizes. The N-shaped arrangement of the servo electric cylinder design makes the system more flexible. Even when a servo electric cylinder reaches its limit stroke, other electric cylinders can still provide stress compensation to ensure the stable operation of the system.
[0054] 2. High precision and stability: The servo cylinder of the present invention, as a linear actuator, has high-precision telescopic control capability, which can ensure that the angle adjustment of the lower frame and the liquid measuring rod reaches extremely high precision. The introduction of the horizontal sensor monitors the tilt state of the lower frame in real time, and with the precise control of the servo cylinder, it can quickly and accurately adjust the lower frame to the correct position to maintain the stability and verticality of the liquid measuring rod.
[0055] 3. Intelligence and Automation: The present invention can automatically calculate and drive the servo electric cylinder to adjust the angle according to the feedback signal of the horizontal sensor, realizing intelligent and automated operation, reducing the need for manual intervention, improving work efficiency, and reducing the possibility of human error.
[0056] 4. Ability to cope with complex environments: The technology of the present invention is particularly suitable for complex environments such as molten steel slag layers and slag rolls. Through real-time monitoring and automatic compensation mechanisms, it can effectively cope with the impact of these environmental factors on the liquid measuring rod and ensure the accuracy and stability of the measurement. BRIEF DESCRIPTION OF THE DRAWINGS
[0057] Figure 1 It is a schematic diagram of the working arrangement of the whole device of the present invention (from a top view); the figure shows multiple working positions of the mechanical arm assembly;
[0058] Figure 2 It is a schematic diagram of the working arrangement (main viewing angle) of the whole device of the present invention;
[0059] Figure 3 It is a schematic diagram of the working arrangement of the whole device of the present invention (main viewing angle) and the inclusive trajectory of the robot arm assembly;
[0060] Figure 4 It is a three-dimensional detailed schematic diagram of the mechanical arm assembly and the gripping hand assembly of the present invention;
[0061] Figure 5 It is a schematic diagram of the composition of the mechanical arm assembly of the present invention;
[0062] Figure 6 The components of the clamping hand assembly of the present invention are as follows ( Figure 5 Schematic diagram of the enlarged view of area A;
[0063] Figure 7 The figure is a schematic diagram of the method flow of embodiment 3.
[0064] Figure numerals: 1. Robot arm assembly; 101. First rotary actuator; 102. First arm body; 103. Second rotary actuator; 104. Second arm body; 105. Third rotary actuator; 106. Third arm body; 106. End actuator; 2. Gripping hand assembly; 201. Pneumatic gripper; 202. Upper frame; 203. Lower frame; 204. Linear actuator; 205. Universal joint coupling; 206. Level sensor; 3. Liquid measuring rod; 4. Placement rack; 5. Infrared imager; 6. Computer; 7. PLC controller; 8. Electric-controlled gas valve; 9. Tundish; 10. Covering agent; 11. Molten steel. DETAILED DESCRIPTION
[0065] In order to make the above-mentioned purposes, features and advantages of the present invention more understandable, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below;
[0066] Embodiment 1:
[0067] In the prior art, the connection between the robot and the connector is mainly fixed, or the disassembly process is cumbersome. When it is necessary to measure tundishes of different heights or sizes, this connection method requires technicians to spend more time replacing parts or disassembling and reassembling, thereby reducing the efficiency of measurement. At the same time, during the continuous casting process, slag layers or slag curling may form on the surface of the molten steel. These slag layers or slag curling will affect the stability of the liquid measuring rod in the molten steel, causing the liquid measuring rod to tilt due to uneven force; for this reason, please refer to Figure 1-6 , this specific implementation will provide relevant technical solutions to solve the above technical problems: a continuous casting robot device for automatic measuring rod clamping and an AI recognition method, including a liquid measuring rod 3 for detecting the liquid level height of a molten steel ladle 11 in the traditional technology; the improvement of this solution is: introducing a mechanical arm assembly 1, and a clamping hand assembly 2 for clamping the liquid measuring rod 3 is installed on the end effector 106 of the mechanical arm assembly 1;
[0068] When the liquid measuring rod 3 is inserted into the molten steel 11 , based on the force resistance in three directions of the molten steel 11 , the clamping hand assembly 2 performs corresponding universal angle adjustment to ensure that the liquid measuring rod 3 is always perpendicular to the molten steel 11 .
[0069] In this scheme: when in use, the mechanical arm assembly 1 drives the clamping hand assembly 2 to move to the placement rack 4 where liquid measuring rods 3 of different specifications are stored, and then the clamping hand assembly 2 clamps the specified liquid measuring rod 3 according to the corresponding task requirements and moves it to the top of the tundish 9, in which the covering agent 10 and molten steel 11 are contained. At this time, the mechanical arm assembly 1 allows the clamping hand assembly 2 and its liquid measuring rod 3 to remain perpendicular to the liquid surface of the molten steel 11, and then slowly inserts the liquid measuring rod 3 into the molten steel. During this period, the liquid measuring rod 3 will be unevenly stressed due to the influence of the slag layer and slag roll of the molten steel, and the hand assembly 2 changes the force direction accordingly, and performs reverse universal angle adjustment compensation, so that the liquid measuring rod 3 is always perpendicular to the molten steel 11; thereby achieving more accurate height detection. The trajectory control of the above-mentioned mechanical arm assembly 1 and the clamping hand assembly 2 is achieved by the PLC controller 7 issuing preset instructions to its corresponding components.
[0070] In this solution, all electrical components of the device as a whole rely on AC power for energy supply; specifically, the electrical components of the device as a whole are conventionally electrically connected to the AC power output port through devices such as relays, transformers and button panels to meet the energy supply requirements of all electrical components of the device.
[0071] Specifically: The device realizes automatic clamping and precise insertion of the liquid measuring rod 3 through the coordinated work of the mechanical arm assembly 1 and the clamping hand assembly 2. The design of the clamping hand assembly 2 takes into account the special environment of the molten steel 11. Through the universal angle adjustment mechanism, when the liquid measuring rod 3 is subjected to uneven force, its angle can be adjusted in real time to keep the liquid measuring rod 3 in a vertical state.
[0072] At the same time, a PLC controller 7 is integrated in the device to control the trajectory of the mechanical arm assembly 1 and the gripping hand assembly 2. Through preset instructions, the PLC controller 7 can ensure that the mechanical arm assembly 1 and the gripping hand assembly 2 operate according to a predetermined path and speed, thereby improving the accuracy and efficiency of the measurement.
[0073] It can be understood that in the above scheme: the device realizes the automatic clamping and precise insertion of the liquid measuring rod 3 by introducing the mechanical arm assembly 1 and the clamping hand assembly 2, and solves the problems of errors and inefficiency caused by manual operation in traditional technology. The universal angle adjustment mechanism of the clamping hand assembly 2 can keep the liquid measuring rod 3 in the vertical state under the special environment of molten steel 11, thereby improving the accuracy of measurement. In addition, the PLC controller 7 integrated in the device controls the trajectory of the mechanical arm assembly 1 and the clamping hand assembly 2 through preset instructions, realizing the automation and intelligence of measurement. This not only improves the efficiency of measurement, but also reduces the labor intensity and safety risks of operators.
[0074] In this specific implementation, please refer to Figures 1 to 3 : Also include an infrared imager 5 for counting the temperature field distribution and temperature pixels of the molten steel 11. The infrared imager 5 captures the temperature field distribution and temperature pixels and transmits them to a computer 6 for visualization, which is then viewed by the staff.
[0075] The computer 6 is installed with conventional image processing software, which can visualize the received image data and generate an intuitive temperature distribution diagram. The staff can clearly view the temperature field distribution and temperature pixels of the molten steel 11 through the display screen of the computer 6, so as to more accurately understand the temperature state of the molten steel. During the measurement process, the mechanical arm assembly 1 drives the clamping hand assembly 2 and the liquid measuring rod 3 to perform automated operations, while the infrared imager 5 continuously captures the temperature information of the molten steel 11. This parallel working mode enables temperature measurement and liquid level measurement to be performed simultaneously, improving the comprehensiveness and efficiency of the measurement.
[0076] During the continuous casting process, the molten steel 11 will emit strong infrared radiation. The infrared imager 5 can capture this radiation and convert the radiation intensity into image data of temperature field distribution and temperature pixels through its internal image processing algorithm. These data can intuitively reflect the temperature state of the molten steel 11 and provide important reference information for the staff.
[0077] In this specific implementation, please refer to Figures 1 to 5 The design of the robot arm assembly 1 adopts a structure combining a multi-axis rotary actuator and an arm body to achieve multi-dimensional flexible operation. Specifically, the robot arm assembly 1 consists of the following parts:
[0078] 1) The first rotary actuator 101 and the first arm 102: The first rotary actuator 101 is arranged along the Y-axis direction, and is used to drive the first arm 102 to perform a rotary motion on a horizontal plane. This design enables the robot arm assembly 1 to achieve a wide range of movement and positioning in the Y-axis direction.
[0079] 2) The second rotary actuator 103 and the second arm 104: The second rotary actuator 103 is mounted on the first arm 102 and arranged along the X-axis direction. It is responsible for driving the second arm 104 to perform a rotary motion on the extension line of the first arm 102. This design increases the flexibility of the robot arm assembly 1 in the X-axis direction, enabling it to be positioned and operated more accurately.
[0080] 3) The third rotary actuator 105 and the third arm 106: The third rotary actuator 105 is mounted on the second arm 104 and arranged along the Z-axis direction. It is responsible for driving the third arm 106 to perform vertical rotational motion. This design enables the robot arm assembly 1 to achieve full-range and flexible operation in three-dimensional space.
[0081] 4) End effector 106: The end effector 106 is installed at the end of the third arm 106 and is a frame-like structure. By cooperating with the above-mentioned rotary actuator and arm, the end effector 106 can be accurately positioned and operated in a complex working environment.
[0082] Specifically: The design principle of the robot arm assembly 1 is based on the combination of a multi-axis rotary actuator and an arm body. By controlling the rotary actuators in different axes, the robot arm can move flexibly in three-dimensional space. Each rotary actuator has independent drive and control capabilities, and its rotation angle and speed can be adjusted as needed. This design enables the robot arm assembly 1 to adapt to various complex operating requirements and play an important role in processes such as continuous casting and steel pouring.
[0083] It should be noted that the first rotary actuator 101, the second rotary actuator 103 and the third rotary actuator 105 can all be servo motors used in existing robotic arm devices, and their control (ie, their trajectory control) can be achieved through existing robotic arm control algorithms, which will not be described in detail here.
[0084] In this specific implementation, please refer to Figure 6 : The clamping hand assembly 2 includes a pneumatic clamp 201 for clamping the liquid measuring rod 3, and an upper frame 202 and a lower frame 203 mounted on the lower surface of the end effector 106 and the upper surface of the pneumatic clamp 201;
[0085] A parallel assembly is provided between the upper frame 202 and the lower frame 203, and the parallel assembly is used to control the lower frame 203 and the pneumatic clamp 201 to perform universal angle adjustment. The pneumatic clamp 201 is connected to an external compressed air source by an electric-controlled air valve 8 for air supply.
[0086] Three level sensors 206 are evenly arranged in a circular array on the lower frame 203. When the liquid measuring rod 3 is inserted into the molten steel, the slag layer and slag roll of the molten steel will cause the liquid measuring rod 3 to be unevenly stressed, which will in turn affect the angle deviation of the lower frame 203 and the pneumatic clamp 201. The level sensor 206 at the nearest position will sense its own tilt state, and then the feedback parallel component will perform universal angle adjustment on the lower frame 203 and the pneumatic clamp 201 to achieve compensation.
[0087] Specifically: The working principle of the clamping hand assembly 2 is based on the kinematic and dynamic principles of the parallel mechanism. When the liquid measuring rod 3 is subjected to uneven force due to the slag layer and slag curl of the molten steel, the lower frame 203 and the pneumatic clamp 201 will have an angular displacement. At this time, the horizontal sensor 206 at the nearest position will sense the tilt state of the lower frame 203 and feed this information back to the parallel assembly. The parallel assembly calculates the required angle adjustment amount through a complex algorithm based on the feedback signal from the sensor, and drives the corresponding actuator to act. Finally, through the precise control of the parallel assembly, the lower frame 203 and the pneumatic clamp 201 can quickly return to a vertical state to ensure that the liquid measuring rod 3 is always perpendicular to the molten steel surface.
[0088] It can be understood that the design of the clamping hand assembly 2 fully considers various factors in the measurement process, especially the uneven force problem that the liquid measuring rod 3 may encounter when inserted into the molten steel. By adopting advanced technologies and components such as pneumatic clamps 201, upper frame 202, lower frame 203, parallel components and horizontal sensors 206, the clamping hand assembly 2 can achieve accurate clamping and stable control of the liquid measuring rod 3. During the measurement process, even if the liquid measuring rod 3 is offset in angle due to uneven force, the clamping hand assembly 2 can quickly sense and make corresponding angle adjustments to ensure that the liquid measuring rod 3 is always perpendicular to the molten steel surface. This design not only improves the accuracy and stability of the measurement, but also reduces the labor intensity and safety risks of the operator.
[0089] In this specific implementation, please refer to Figure 6 The parallel assembly is composed of six linear actuators 204, which are arranged in a circular array between the upper frame 202 and the lower frame 203.
[0090] The linear actuators 204 are devices that can convert electrical energy or other forms of energy into linear motion. They are evenly distributed between the upper frame 202 and the lower frame 203 to form a ring array, ensuring that the lower frame 203 and the pneumatic gripper 201 thereon can achieve precise universal angle adjustment.
[0091] Through the arrangement of the circular array, the linear actuator 204 can evenly apply force to the lower frame 203, thereby achieving a more stable and precise angle adjustment. This design also improves the structural strength and stability of the entire clamping hand assembly 2.
[0092] Specifically, when the liquid measuring rod 3 is subjected to uneven force due to the slag layer and slag roll of the molten steel, the angle of the lower frame 203 and the pneumatic clamp 201 will be offset. At this time, the level sensors 206 evenly distributed on the lower frame 203 will sense this change and transmit the relevant information to the control system. The control system calculates the required angle adjustment amount based on the feedback signal of the level sensor and drives the corresponding linear actuator 204 to act.
[0093] Since the linear actuators 204 are arranged in a circular array, they can simultaneously act on different positions of the lower frame 203, thereby achieving more uniform and precise angle adjustment. In this way, the lower frame 203 and the pneumatic clamp 201 can quickly return to a vertical state, ensuring that the liquid measuring rod 3 is always perpendicular to the molten steel surface.
[0094] It can be understood that: by introducing the parallel assembly and the six linear actuators 204, the functionality of the clamping hand assembly 2 has been significantly improved. This design not only improves the stability and accuracy of the clamping hand assembly 2, but also enables it to better adapt to complex working environments. Even in the case where the angle of the liquid measuring rod 3 is offset due to uneven force, the clamping hand assembly 2 can quickly sense and make corresponding angle adjustments to ensure the smooth progress of the liquid measuring process. In addition, the arrangement of the annular array also improves the structural strength and durability of the entire clamping hand assembly 2, allowing it to maintain a stable working state for a longer period of time.
[0095] In this specific embodiment, preferably, the linear actuator 204 is a servo electric cylinder, and there are six of them in total; the cylinder body and piston rod of the servo electric cylinder are universally hinged on the mutually opposite sides of the upper frame 202 and the lower frame 203 through the universal joint coupling 205. When any servo electric cylinder is actuated, the universal joint coupling 205 converts the linear motion into rotational motion, so as to realize the angle adjustment of one position of the lower frame 203.
[0096] Specifically, when any servo electric cylinder performs a telescopic action, the lower frame 203 will adjust the angle at the position corresponding to the servo electric cylinder through the action of the universal joint coupling 205. Since there are six servo electric cylinders corresponding to different positions, the lower frame 203 can adjust the angle at multiple positions, thereby achieving more precise and flexible control.
[0097] It can be understood that by using a servo electric cylinder as the linear actuator 204 and realizing the conversion from linear motion to rotational motion through the universal joint coupling 205, the functionality of the clamping hand assembly 2 is greatly improved. This design enables the lower frame 203 to be accurately adjusted in angles in multiple directions, thereby ensuring that the liquid measuring rod 3 always remains perpendicular to the molten steel surface. Even in the case where the liquid measuring rod 3 is offset in angle due to uneven force, the clamping hand assembly 2 can quickly sense and make corresponding angle adjustments to ensure the accuracy and stability of the liquid measuring process. In addition, since the servo electric cylinder has the advantages of high precision, high response speed and programmable control, the clamping hand assembly 2 can also achieve more flexible and intelligent control, thereby improving the automation level of the entire continuous casting process.
[0098] It should be pointed out that the linear actuator 204 is in the form of a servo electric cylinder and its circular array arrangement, which actually makes the above-mentioned "parallel component" belong to the mechanical sense (Parallel Mechanism); therefore, the specific control form of the servo electric cylinder can use the traditional parallel robot PID controller algorithm, where the error term of the PID controller algorithm can be set to the difference between the feedback signal of the above-mentioned horizontal sensor 206 and the horizontal data.
[0099] In this specific embodiment, preferably, three adjacent servo electric cylinders are arranged in an N-shape. This form of mutually inclined arrangement ensures that when any servo electric cylinder reaches the limit stroke position, the remaining servo electric cylinders still have stroke space to achieve stress compensation.
[0100] Specifically: when the lower frame 203 tilts due to uneven force, the control system will calculate the required angle adjustment amount based on the feedback signal of the horizontal sensor 206. Then, the control system will drive the corresponding servo electric cylinder to extend and retract, and convert the linear motion into the rotational motion of the lower frame 203 through the universal joint coupling 205. Since the servo electric cylinders are arranged in an N shape, when one of the servo electric cylinders reaches its limit stroke position, the other servo electric cylinders can still continue to work, providing stress compensation for the system. In this way, the clamping hand assembly 2 can maintain stable performance in a complex working environment.
[0101] It is understandable that this arrangement allows the servo cylinders to have a larger space for movement within their respective travel ranges, thereby improving the flexibility and adaptability of the system. Secondly, when one of the servo cylinders reaches the limit of travel, the other servo cylinders can still continue to work, providing stress compensation for the system, which ensures that the clamping hand assembly 2 can maintain stable performance in a complex working environment. Finally, this design also improves the reliability and durability of the clamping hand assembly 2 and reduces maintenance and replacement costs.
[0102] In summary, the operating logic of the device provided in this embodiment is:
[0103] S1: The robot arm assembly 1 moves from the standby position to the storage position of the gripper hand assembly 2;
[0104] S2: The robot arm assembly 1 moves from the storage position of the clamping hand assembly 2 to the storage position of the liquid measuring rod 3, and the clamping hand assembly 2 automatically clamps the liquid measuring rod 3.
[0105] S3: The robot arm assembly 1 moves from the storage position of the liquid measuring rod 3 to the measuring port on the upper part of the tundish 9, and inserts the liquid measuring rod 3 into the molten steel 11, and immediately lifts it out of the tundish 9 after reaching the bottom surface of the tundish 9 to reach the measuring rod abandonment position.
[0106] S4: The infrared imager 5 measures the temperature field of the liquid measuring rod 3 that has reached the measuring position, and the data is processed by the image processing computer 6 and then transmitted to the PLC controller 7.
[0107] S5: The clamping hand assembly 2 releases the liquid measuring stick 3 and automatically falls into the stick discarding hole. The robot arm assembly 1 returns to the storage position of the clamping hand assembly 2, and returns to the standby position after releasing.
[0108] Embodiment 2: Based on Embodiment 1, this embodiment further provides an adaptive control method for six servo electric cylinders, using a PID controller algorithm:
[0109] The PID (Proportional-Integral-Derivative) controller algorithm is a widely used control algorithm for adjusting the output of the system to match the desired reference input. In the application of the clamping hand assembly 2, the PID controller can correct the tilt state of the lower frame 203 by adjusting the extension and contraction amount of the servo electric cylinder, thereby keeping the liquid measuring rod 3 always perpendicular to the molten steel 11.
[0110] e(t) = set value - actual value;
[0111]
[0112] Where: e(t) is the error signal, that is, the difference between the desired vertical state (set value) and the currently measured tilt state (actual value). u(t) is the output of the PID controller, which will be used as the control signal for the extension and retraction of the servo electric cylinder. K p is the proportionality factor, used to control the amplitude of the response. i is the integral coefficient used to eliminate steady-state errors. d is the differential coefficient, which is used to predict the changing trend of the error and make adjustments in advance to improve the response speed. τ is the time variable.
[0113] Specifically, proportional control (P): Proportional control outputs a control signal based on the current error value. If the error is large, the control signal is also large to speed up the adjustment. Proportional control can respond to errors quickly, but may cause system overshoot or oscillation.
[0114] Specifically, integral control (I): Integral control takes into account the accumulation of errors over a period of time. Even if the error is zero for a short period of time, if there is an accumulation of errors before, the integral term will still produce an output, thereby pushing the system toward the set point. Integral control helps eliminate steady-state errors, but it may also cause the system to respond too slowly or integral saturation.
[0115] Specifically, Derivative Control (D): Derivative control predicts future errors based on the rate of change of the error. If the error is increasing rapidly, the derivative term will generate an early adjustment signal, thereby reducing the possibility of overshoot or oscillation. Derivative control helps to speed up the response and improve the stability of the system.
[0116] It is understandable that in the application of the clamping hand assembly 2, the PID controller will continuously calculate the error between the tilt state and the vertical state of the lower frame 203, and output a control signal according to the PID algorithm to adjust the angle of the lower frame 203 through the servo electric cylinder to keep the liquid measuring rod 3 vertical. p , Ki and K d The three coefficients can control the response speed, stability and accuracy of the system.
[0117] Embodiment 3: Based on Embodiment 1, this embodiment further introduces an AI recognition method to the infrared imager 5 in the device, and uses the U-Net semantic segmentation model to identify the position and range of the molten steel surface in the image, so as to achieve more intelligent management and application. It includes the following steps S1 to S4. Figure 7 :
[0118] Step S1, data preparation:
[0119] Collect infrared image data i containing the molten steel surface, and mark the position a and range b of the surface to form a training data set D and sample i i (a i ,b i ):
[0120] D=[i 1 (a 1 ,b 1 ),i 2 (a 2 ,b 2 ),...,i 1 (a n ,b n )]
[0121] Where n is the sample i i (a i ,b i )
[0122] We can also use the sample i i (a i ,b i ) to perform necessary preprocessing, such as contrast enhancement and noise removal, to improve image quality. At the same time, according to the input requirements of the U-Net model, the image is adjusted to a uniform size and format.
[0123] Step S2, read the U-Net semantic segmentation model:
[0124] Use the TensorFlow deep learning framework to read the U-Net semantic segmentation model.
[0125] Then you need to train the U-Net semantic segmentation model, which includes the following steps:
[0126] P1. Divide the labeled training data set D into training set, validation set and test set.
[0127] P2. Set the cross entropy loss function and Adam optimizer.
[0128] The cross-entropy loss function is used to measure the difference between the probability distribution predicted by the model and the true distribution. In the semantic segmentation task, the category of each pixel is predicted and the difference between it and the true label is calculated:
[0129]
[0130] Where N is the total number of pixels, C is the total number of categories, and y i,c is the true label (1 if the i-th pixel belongs to category c, 0 otherwise), p i,c is the probability that the model predicts that the pixel belongs to category c.
[0131] The Adam optimizer (Adaptive Moment Estimation) is an optimization algorithm based on gradient descent. It adjusts the learning rate of each parameter by calculating the first-order moment estimate and the second-order moment estimate of the gradient. The Adam optimizer combines the advantages of the AdaGrad and RMSProp optimization algorithms, and can apply different learning rates to different parameters, while having good convergence speed and robustness. The update rule of the Adam optimizer is:
[0132] m t =β 1 m t-1 +(1-β 1 ) t ;
[0133] v t =β 2 v t-1 +(1-β 2 ) t 2 ;
[0134]
[0135] Among them, m t and v t are the first-order moment estimate and the second-order moment estimate of the gradient, and is its updated value; g t is the current gradient, β 1 and β 2 is a hyperparameter (close to 1), η is the learning rate, ∈ is a small number (used to prevent division by zero), and θ t is the current parameter value, θ t+1 is the updated parameter value.
[0136] P3. Train the U-Net model on the training set and validate it on the validation set to adjust hyperparameters and prevent overfitting.
[0137] By continuously iterating the training process, the model's hyperparameters (such as learning rate, weight decay, etc.) can be gradually adjusted to find the parameter combination that optimizes the model's performance on the validation set. At the same time, overfitting can be prevented by using regularization techniques (such as Dropout, L1 / L2 regularization, etc.) or early stopping methods.
[0138] Step S3, deploying the U-Net semantic segmentation model: deploying the trained U-Net model to the system where the infrared imager 5 is located.
[0139] The image processing computer receives the actual samples collected by the infrared imager i (a i ,b i )' and input it into the U-Net model for inference.
[0140] The model outputs the segmentation result of the liquid surface area, that is, it identifies the position a' and range b' of the molten steel surface in the image:
[0141] S300, data capture: using an infrared imager to capture image data in an actual scene, that is, actual sample ii(ai,bi)'. ii represents an infrared image, and ai and bi are metadata related to the image.
[0142] S301, data preprocessing: (optional) preprocess the captured image data, such as scaling, normalization, etc., to meet the requirements of U-Net model input.
[0143] S302, model reasoning: input the preprocessed image data into the U-Net model. The U-Net model performs forward propagation and outputs the segmentation result of the liquid surface area.
[0144] S303, result analysis: parse the information of the liquid surface area from the output of the U-Net model.
[0145] The output is a probability map or segmentation map of the same size as the input image, where the value of each pixel represents the probability or category of the pixel belonging to the liquid surface area. The exact location and range of the liquid surface area are determined by threshold processing or post-processing (morphological operations).
[0146] The position a' and range b' of the liquid surface in the image are extracted. Here a' is the coordinate of a center point or the coordinate of a bounding box, and b' may be the size of a bounding box or the number of pixels / area of the liquid surface area.
[0147] S304, result output: output the analyzed liquid level position and range information to the system or user interface.
[0148] Step S4, system integration: Integrate the inference results of the U-Net model with the PLC controller to achieve real-time monitoring and control of the molten steel level.
[0149] Furthermore, the Python execution program of the U-Net semantic segmentation model of this embodiment is as follows:
[0150]
[0151]
[0152]
[0153] The principle of the above procedure is:
[0154] 1) Step S1, data preparation: collect a large number of infrared images and corresponding labels (marking of liquid surface areas). Then load these images and labels into the Python environment, resize and normalize them to ensure the consistency of model input.
[0155] Use the tool (sklearn.model_selection.train_test_split) to split the dataset into training, validation, and test sets in order to verify the performance of the model during training and prevent overfitting.
[0156] 2) Step S2: Read the U-Net semantic segmentation model and train it: Based on the TensorFlow deep learning framework, load the predefined U-Net model structure. Use the optimizer (Adam) and loss function (cross entropy loss) to compile the model. After compilation, use the training set to train the model, and use the backpropagation algorithm and optimizer to update the model's weights to minimize the loss function. At the same time, use the validation set to evaluate the performance of the model, and adjust hyperparameters (such as learning rate, batch size) as needed. This training process will iterate multiple times (i.e., multiple epochs) until the performance of the model on the validation set reaches the preset standard or the maximum number of iterations is reached.
[0157] 3) Step S3: Deploy the U-Net semantic segmentation model: Load the trained model (i.e., the U-Net model with the best weights saved). Then, when a new infrared image is received, perform the same preprocessing steps as during training and input it into the model for prediction. The output of the model is a probability map of the same size as the input image, where the value of each pixel represents the probability that the pixel belongs to the liquid surface area. The probability map is converted into a binary segmentation result by setting a threshold or using other post-processing methods. This segmentation result can be used for subsequent liquid level extraction or control decisions.
[0158] 4) Step S4: System Integration: Extract the liquid level information from the prediction results of the model (by calculating the number of pixels or the centroid position of the liquid level area in the segmentation result). Then send this liquid level information to the PLC controller to adjust the corresponding control parameters according to the liquid level. This process requires communication with the PLC or other control system, and uses the interface protocol (Modbus) to realize data transmission and reception. The procedure is:
[0159]
[0160] / / The liquid level is close to the target value and the change is within the tolerance range, maintain the current state
[0161] / / Tolerance is the allowable range of liquid level fluctuation
[0162] }else{
[0163] / / Send the control valve status to the actual electric valve
[0164] SendControlSignalToValve(ControlValveState); / / Send control signal
[0165] }
[0166] / / Call OnReceiveDataFromUNet() function regularly, once per second
[0167] TimerTask(){
[0168] OnReceiveDataFromUNet();
[0169] }
[0170] In this way, deep learning technology can be combined with traditional industrial automation technology to achieve more efficient and accurate control of molten steel level.
[0171] The above-mentioned embodiments only express the implementation methods of the relevant practical applications of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be based on the attached claims.
Claims
1. A continuous casting robot device for automatic measuring rod clamping, comprising a liquid measuring rod (3) for detecting the liquid level of a molten steel (11) ladle, characterized in that: It also comprises a mechanical arm assembly (1), wherein a clamping hand assembly (2) for clamping the liquid measuring rod (3) is mounted on the end effector (106) of the mechanical arm assembly (1); When the liquid measuring rod (3) is inserted into the molten steel (11), based on the force resistance in at least three directions of the molten steel (11), the clamping hand assembly (2) performs corresponding universal angle adjustment to ensure that the liquid measuring rod (3) is always perpendicular to the molten steel (11).
2. The continuous casting robot device according to claim 1, characterized in that: It also includes an infrared imager (5) for counting the temperature field distribution and temperature pixels of the molten steel (11).
3. The continuous casting robot device according to claim 1, characterized in that: The robot arm assembly (1) comprises a first rotary actuator (101) arranged in the Y-axis direction and a first arm body (102) driven by the first rotary actuator (101), a second rotary actuator (103) installed in the X-axis direction and a second arm body (104) driven by the first rotary actuator (103), a third rotary actuator (105) installed in the Z-axis direction and a third arm body (106) driven by the second arm body (104), and the end actuator (106) installed on the third arm body (106).
4. The continuous casting robot device according to claim 1, characterized in that: The clamping hand assembly (2) comprises a pneumatic clamp (201) for clamping the liquid measuring rod (3), and an upper frame (202) and a lower frame (203) mounted on the lower surface of the end effector (106) and the upper surface of the pneumatic clamp (201); A parallel component is provided between the upper frame (202) and the lower frame (203), and the parallel component is used to control the lower frame (203) to perform universal angle adjustment.
5. The continuous casting robot device according to claim 4, characterized in that: At least three level sensors (206) are evenly arranged on the lower frame (203) in the form of a ring array.
6. The continuous casting robot device according to claim 4, characterized in that: The parallel assembly comprises at least three linear actuators (204), and the linear actuators (204) are arranged in a circular array between the upper frame (202) and the lower frame (203).
7. The continuous casting robot device according to claim 6, characterized in that: The linear actuator (204) is a servo electric cylinder, and the cylinder body and piston rod of the servo electric cylinder are universally hinged on the mutually opposite sides of the upper frame body (202) and the lower frame body (203) through a universal joint coupling (205); three adjacent servo electric cylinders are arranged in an N shape.
8. AI recognition method for automated measuring rod clamping, characterized in that: The AI recognition method is implemented by performing the following steps based on the continuous casting robot device as described in any one of claims 1 to 7 as a hardware carrier: S1, data preparation: collect infrared image data i containing the molten steel surface, and mark the position a and range b of the liquid surface to form a data set D and sample i i (a i ,b i ); S2, read U-Net semantic segmentation model: use TensorFlow deep learning framework to read U-Net semantic segmentation model; S3, deploying the U-Net semantic segmentation model: deploying the trained U-Net model to the control system where the infrared imager 5 is located; receiving the actual samples i collected by the infrared imager through the image processing computer; i (a i ,b i )', and input it into the U-Net model for inference; the model outputs the segmentation result of the liquid surface area, that is, identifying the position a' and range b' of the molten steel surface in the image; S4, system integration: Integrate the obtained position a' and range b' with the PLC controller to achieve real-time monitoring and control of the molten steel level.
9. The AI recognition method according to claim 9, characterized in that: In S1, the dataset D: <h2 style=";text-align:left;direction:ltr">D=[i1(a1,b1),i2(a2,b2),...,i1(a<h2 style=";text-align:left;direction:ltr"> n <h2 style=";text-align:left;direction:ltr"> ,b<h2 style=";text-align:left;direction:ltr"> n <h2 style=";text-align:left;direction:ltr"> )] Where n is the sample i i (a i ,b i ) 10. The AI recognition method according to claim 9, characterized in that: In S2, the training method of the U-Net semantic segmentation model includes: P1, divide the labeled training data set D into training set, validation set and test set; P2. Set the cross entropy loss function and Adam optimizer; P3. Train the U-Net model on the training set.
Citation Information
Patent Citations
Portrait contour detection method in video based on lightweight deep neural network
CN111145195A
Embedded terminal reservoir water level early warning method based on semantic segmentation
CN112163496A
Continuous casting tundish molten steel liquid level measuring device and using method thereof
CN116475392A
Atmospheric pressure wafer conveying manipulator based on parallel mechanism
CN220807384U
Instrument for measuring variation of molten metal surface level in continuous casting mold
JP1994234051A
