Automatic metallurgical steel billet fishing control system based on PLC

Abstract

The invention belongs to the technical field of industrial automatic control systems, and particularly relates to an automatic metallurgical billet fishing control system based on a PLC (programmable logic controller). And the control module is responsible for executing preset sequence control logic, safety interlocking protection, emergency shutdown processing and control instruction interaction with the multi-degree-of-freedom fishing execution module. By fusing a multi-type sensing technology, real-time data processing and an AI decision algorithm, seamless integration with the deterministic control capability of the PLC is realized. According to the application, high-precision robust sensing and intelligent adaptive control of the steel billet in a severe environment can be realized, the PLC stability and reliability are realized, the energy consumption is reduced, and the efficiency is improved.

Description

Technical Field

[0001] This invention belongs to the field of industrial automation control system technology, specifically a PLC-based automated steel billet retrieval control system. Background Technology

[0002] With modern industrial production placing higher demands on efficiency, safety, and product quality, automation technology has become an indispensable core component of the metallurgical industry. In the billet production process, especially in continuous casting, furnace discharge, and subsequent processing, the precise and efficient automated retrieval and transfer of billets is crucial for ensuring production continuity, reducing labor intensity, and optimizing energy consumption. For a long time, to cope with the harsh conditions of high temperatures, dust, and heavy loads in the metallurgical environment, the development of industrial control systems in this field has consistently focused on stability, reliability, and anti-interference capabilities.

[0003] In the current field of industrial automation, Programmable Logic Controllers (PLCs) have become the core control component of various automated control systems, especially automated billet handling control systems in metallurgy, due to their excellent stability, strong anti-interference capabilities, modular design, and deterministic operating mechanism. Traditional PLC-based billet handling systems typically acquire signals from discrete sensors such as limit switches and encoders, combined with analog signals such as temperature and pressure, and use pre-programmed ladder diagrams, instruction lists, or structured text to achieve sequential and positional control of hydraulic or electric actuators (such as grippers, robotic arms, and gantry cranes). Specifically, this system can precisely control a series of actions such as extension, rotation, gripping, lifting, and placement of the handling equipment, ensuring that the billet is safely and stably removed from a designated location (such as inside a heating furnace) and accurately transported to the next process. Its design philosophy lies in decomposing complex industrial processes into a series of discrete logical judgments and action sequences, which are periodically executed by the PLC's scanning, thereby achieving automated management of the entire handling operation. Such systems effectively replace intensive manual operations, significantly improving operational efficiency and safety, and for a considerable period of time, they have played a key role in meeting the basic automation needs of metallurgical production.

[0004] However, as the metallurgical industry continues to evolve towards greater efficiency, flexibility, and intelligence, and with increasingly stringent requirements for energy efficiency and product quality consistency, the inherent limitations of the aforementioned automated billet handling systems based on traditional PLC architecture are becoming increasingly apparent in addressing current and future challenges. The reason for this lies in the fact that modern metallurgical billet handling operations are far more complex than a simple "grab-move-place" sequence. Specifically, firstly, the actual working conditions of the billets to be handled are often complex and variable. For example, the billet's position, orientation, or geometry may deviate from preset values ​​due to uneven heating or deformation under stress within the heating furnace. Secondly, the harsh environment inside the furnace, including high temperatures, smoke, and steam, severely interferes with the accuracy and reliability of traditional sensors during the handling process. Furthermore, to maximize furnace utilization and discharge efficiency, billets are increasingly arranged more closely, placing extremely high demands on the trajectory planning and obstacle avoidance capabilities of the handling equipment, especially in complex scenarios involving multiple intersecting handling paths or multiple devices operating collaboratively. Against this backdrop, traditional PLCs, primarily focused on deterministic logic and sequential control, have inherent limitations in their ability to handle complex continuous data streams and execute advanced algorithms (such as real-time billet recognition and posture correction based on machine vision, adaptive path optimization, multi-axis linkage precision motion control, and force feedback intelligent grasping strategies). Although some modern PLC products have expanded their capabilities by integrating dedicated function modules or communication interfaces, these expansions often increase system complexity, introduce additional communication latency, and struggle to achieve efficient fusion and real-time processing of massive amounts of high-frequency, heterogeneous sensor data at the core processor level. In other words, facing the demand for highly adaptive, high-precision, highly flexible, and real-time decision-making intelligent grasping requirements, the traditional PLC architecture is insufficient in terms of computing power, data throughput, and algorithm support. Especially when the system is required to dynamically adjust the grasping strategy based on real-time environmental changes and billet status, and even perform autonomous learning and optimization, its inherent discretization and scanning execution mechanism makes achieving true "intelligence" extremely difficult. This deep-seated contradiction leads to a slow response of the system when dealing with abnormal operating conditions, which can easily cause misoperation or low efficiency, thereby affecting the stable operation of the entire production line and product quality, and even bringing potential safety hazards. To address this, the present invention provides a PLC-based automated billet retrieval control system. Summary of the Invention

[0005] In order to overcome the shortcomings of the prior art, at least one technical problem raised in the background art is solved.

[0006] The technical solution adopted by this invention to solve its technical problem is as follows: The PLC-based automated billet retrieval control system of this invention includes: a programmable logic controller (PLC) module, an intelligent sensing and decision-making module, a multi-degree-of-freedom retrieval execution module, a sensor fusion module, and a high-speed industrial communication network. This system seamlessly integrates multiple advanced sensing technologies, real-time data processing and analysis, and artificial intelligence-based intelligent decision-making algorithms with the deterministic control capabilities of the PLC, thereby achieving high-precision, high-flexibility, and adaptive automated retrieval of metallurgical billets in complex and harsh environments.

[0007] Specifically, the programmable logic controller (PLC) module serves as the central deterministic control unit of the entire system, responsible for executing preset sequential control logic, safety interlock protection, emergency stop processing, and direct real-time command interaction with the underlying servo drives and hydraulic proportional valves of the retrieval execution module. The PLC module is equipped with a high-performance processor unit, large-capacity memory, high-speed digital input / output (DI / DO) modules, and analog input / output (AI / AO) modules. The PLC module exchanges bidirectional real-time data with the intelligent sensing and decision-making module and the multi-degree-of-freedom retrieval execution module through an integrated high-speed industrial Ethernet interface, such as a communication port supporting PROFINETIRT or EtherCAT protocols. The preset program logic within the PLC module includes, but is not limited to: a zero-position calibration program for the retrieval equipment, safety area restriction logic, an emergency stop response program, and conversion logic for receiving high-level motion commands from the intelligent sensing and decision-making module and converting them into underlying actuator drive signals. The PLC module also includes diagnostic functions, capable of real-time monitoring of its own status, communication link status, and the operating status of each connected module, and triggering corresponding alarms or safety stop procedures when an anomaly is detected.

[0008] The intelligent sensing and decision-making module is configured as the core intelligent unit for real-time perception of the metallurgical billet's state, environmental modeling, intelligent path planning, and grasping strategy decision-making. The intelligent sensing and decision-making module consists of an industrial-grade high-performance computing unit equipped with a multi-core central processing unit (CPU), a graphics processing unit (GPU), and dedicated artificial intelligence acceleration hardware. The intelligent sensing and decision-making module is interconnected with the PLC module via a high-speed industrial Ethernet interface and receives pre-processed fused sensing data from the sensor fusion module. The intelligent sensing and decision-making module internally deploys multiple intelligent algorithm sub-units, including but not limited to: a 3D visual perception and attitude estimation sub-unit, an adaptive path planning sub-unit, a force feedback intelligent grasping sub-unit, and a multi-sensor data fusion processing sub-unit. The industrial-grade high-performance computing unit runs a real-time operating system (RTOS) or a Linux operating system with real-time expansion capabilities to ensure deterministic execution and response time of key intelligent algorithms.

[0009] The sensor fusion module is configured to integrate and process heterogeneous data from multiple types of sensors, thereby providing the intelligent sensing and decision-making module with comprehensive, accurate, and robust environmental and target information. The sensor fusion module includes, but is not limited to: 3D Vision Sensor Unit: This unit consists of at least two high-resolution industrial-grade stereo cameras or one 3D LiDAR sensor. The stereo cameras employ a global shutter CMOS image sensor with a pixel resolution of 2048x1536 and a frame rate of 50 frames per second. They are equipped with an industrial-grade high-temperature and dust-resistant protective cover with an IP67 protection rating and active cooling. The LiDAR sensor uses a 905nm wavelength laser with a scanning frequency of 50Hz, an angular resolution of 0.1°, and a maximum measurement distance of 50 meters. It is also equipped with an environmental protection cover and a self-cleaning system. The 3D vision sensor unit is installed above or to the side of the retrieval equipment, enabling it to acquire 3D point cloud data and high-precision image data of the furnace interior and the steel billet.

[0010] Force / Torque Sensor Unit: This unit consists of a six-axis force / torque sensor, which is mounted at the end joint of the gripper of the multi-degree-of-freedom retrieval execution module. The measurement range of the six-axis force / torque sensor is ±500 Newtons for the X, Y, and Z axis mechanical components and ±50 Nm for the torque component, with a resolution of 0.1 Newtons and 0.01 Nm respectively. The force / torque sensor unit transmits real-time force / torque data to the sensor fusion module via a digital interface supporting EtherCAT or PROFINET protocols.

[0011] High-temperature infrared temperature sensor unit: This unit consists of at least two industrial-grade non-contact infrared thermometers with a spectral response range of 8μm to 14μm, a measurement temperature range of 400°C to 1200°C, and a measurement accuracy of ±1% of the reading. The infrared thermometers are installed on the forearm of the retrieval equipment and can monitor the surface temperature of the steel billet to be retrieved in real time, providing thermal deformation reference data for the intelligent grasping strategy.

[0012] Environmental parameter sensor unit: This unit includes a dust concentration sensor and a vapor concentration sensor. The dust concentration sensor has a measurement range of 0.1 mg / m³ to 1000 mg / m³ with an accuracy of ±5%. The vapor concentration sensor has a measurement range of 0% to 100%RH with an accuracy of ±3%RH. This environmental parameter sensor unit is used to monitor the severity of the harvesting environment in real time and provide environmental correction parameters to the vision system, such as adjusting the parameters of the image processing algorithm to compensate for the impact of dust and vapor on image quality.

[0013] The sensor fusion module integrates a data acquisition and preprocessing unit, responsible for synchronously acquiring, noise filtering, data calibration, and format conversion of raw data from the aforementioned sensor units. The data fusion processing subunit within the sensor fusion module employs extended Kalman filtering (EKF) or unscented Kalman filtering (UKF) algorithms to perform multi-source fusion of encoder data from 3D vision, force / torque, infrared thermometry, and the retrieval execution module. This results in the generation of a high-precision, robust model of the billet's 3D position, orientation, geometry, surface temperature, and the real-time status of the retrieval equipment's end effector.

[0014] The 3D visual perception and pose estimation subunit receives fused perception data from the sensor fusion module and performs the following functions: Billet identification and segmentation: A deep learning-based convolutional neural network (CNN) model, such as a model based on U-Net or Mask R-CNN architecture, is used to perform real-time analysis of 3D point cloud data or processed image data to accurately identify the billets to be retrieved inside the furnace and segment them from the background. The model is trained offline and fine-tuned online on a billet dataset containing various shapes, sizes, surface features, and environmental interferences to adapt to the complexity of the metallurgical environment.

[0015] 3D attitude and geometric shape estimation of steel billet: Using point cloud registration algorithms, such as the Iterative Closest Point (ICP) algorithm or feature-based registration algorithms, the identified steel billet point cloud is matched with a pre-set steel billet CAD model, thereby accurately estimating the 3D position (X, Y, Z coordinates), attitude (roll angle, pitch angle, yaw angle) of the steel billet, and detecting any possible deformation or misalignment. The algorithm can tolerate a certain degree of occlusion and noise.

[0016] Furnace environment modeling and obstacle detection: Based on point cloud data acquired by 3D vision sensors, a real-time 3D environment model of the retrieval area is constructed, and the position and geometric information of the furnace wall, adjacent billets, furnace equipment or other potential obstacles are detected in real time.

[0017] The adaptive path planning subunit receives the billet's 3D pose, geometric shape, and environmental obstacle information from the 3D vision perception and pose estimation subunit. Combined with the current pose and kinematic model of the multi-degree-of-freedom retrieval execution module, it generates or updates the optimal motion trajectory of the retrieval device in real time. The adaptive path planning subunit employs sampling-based methods, such as the Fast Random Tree (RRT*) algorithm or heuristic search algorithms (e.g., A* algorithm), to plan multiple collision-free trajectories from the current position to the target billet grasping point and then to the placement point, considering multiple optimization objectives such as robotic arm kinematics and dynamics constraints, collision avoidance, motion smoothness, and time or energy optimization. The path planning algorithm can perform rapid online replanning or trajectory adjustment based on real-time detected billet position deviations, pose changes, or new environmental obstacles.

[0018] The force feedback intelligent grasping subunit receives real-time force / torque data from the sensor fusion module and target grasping point information and target grasping force parameters from the adaptive path planning subunit. The force feedback intelligent grasping subunit achieves intelligent grasping in the following ways: Contact detection and position fine-tuning: As the gripper approaches the billet, force / torque sensor data is monitored in real time. Once the contact force with the billet exceeds a preset threshold, the position and orientation of the gripper are adaptively fine-tuned at the micrometer level based on the direction and magnitude of the contact force, using impedance control or admittance control strategies. This ensures that the gripper can cover the billet at the optimal angle and position, avoiding gripping failures caused by uneven billet surfaces or initial orientation estimation errors.

[0019] Adaptive gripping force control: During the gripping process, the clamping force of the gripper is dynamically adjusted based on the weight of the billet, its surface friction coefficient, temperature (from an infrared temperature sensor), and the sliding trend detected during actual gripping. This gripping force control employs a PID control algorithm combined with fuzzy logic or an adaptive gain scheduling strategy to ensure a firm grip on the billet without damaging its surface or wasting energy due to over-gripping. The gripping force setting is preset based on the billet's material properties and maximum allowable stress, and is corrected through real-time feedback.

[0020] Grasping Status Confirmation and Anomaly Handling: Continuously monitor force / torque data to determine whether the gripping was successful and whether the billet slipped or fell. If an anomaly is detected, such as the gripping force failing to reach the set value for an extended period or suddenly decreasing, an anomaly signal is immediately sent to the PLC module, triggering a preset anomaly handling procedure, such as repositioning the gripper or safely releasing it.

[0021] The multi-degree-of-freedom (DOF) grasping execution module is configured to execute high-precision motion commands issued by the PLC module. The grasping execution module includes a multi-joint robotic arm or a gantry robot system, possessing at least six degrees of freedom to ensure flexible grasping and transfer of the steel billet. Each joint of the robotic arm is equipped with a high-precision servo motor and a corresponding servo driver, which communicates with the PLC module in real time via EtherCAT or PROFINET protocols. The grasping execution module also includes a custom-designed electric or hydraulic gripper with high-temperature protection and adaptive clamping functions. This gripper can adaptively adjust according to the geometry and shape of the steel billet to ensure stable and reliable grasping. The gripper integrates independent displacement and pressure sensors to monitor the clamping status.

[0022] The high-speed industrial communication network is configured to enable efficient, low-latency, deterministic data transmission between the programmable logic controller module, the intelligent sensing and decision-making module, the sensor fusion module, and the multi-degree-of-freedom acquisition and execution module. The network employs industrial Ethernet technology based on the IEEE 802.3 standard, specifically implemented using protocols such as PROFINETIRT (isochronous real-time) or EtherCAT. The network's data transmission rate is 100 Mbps or 1 Gbps, and the communication cycle is configurable from 1 ms to 4 ms to meet the system's stringent real-time requirements. The network topology adopts a star or ring configuration and supports network redundancy to enhance system reliability and resistance to single points of failure. All network components, including industrial switches, communication cables, and connectors, comply with EMC immunity standards and protection level requirements for industrial environments.

[0023] The beneficial effects of this invention are as follows: 1. The PLC-based automated billet retrieval control system of the present invention, through multi-sensor fusion technology, combines three-dimensional vision, force / torque, infrared temperature measurement and environmental parameter sensors, effectively overcomes the interference of harsh environments such as high temperature, dust and steam in metallurgy on the accuracy and reliability of single sensors, and realizes high-precision and high-robust real-time perception of billet position, posture, shape and environmental obstacles.

[0024] 2. The PLC-based automated billet handling control system of this invention integrates advanced algorithms such as billet recognition and attitude estimation based on deep learning, adaptive path planning, and force feedback intelligent grasping through an intelligent sensing and decision-making module. The system can autonomously and in real-time adjust and optimize its strategy based on the real-time changes in the billet's state, dynamic changes in the environment, and actual force feedback during the grasping process, thereby achieving highly flexible and efficient handling operations and significantly reducing the error rate.

[0025] 3. The PLC-based automated billet handling control system of this invention, by introducing advanced intelligent algorithms and sensing capabilities, uses the PLC as the core deterministic control unit, retaining the inherent stability, reliability, and anti-interference capabilities of the PLC in the industrial control field, ensuring the safe and stable operation of the system under extreme conditions. The hierarchical control architecture avoids piling all complex logic into a single PLC controller, effectively balancing real-time performance and intelligence.

[0026] 4. The PLC-based automated billet handling control system of this invention, by introducing advanced intelligent algorithms and sensing capabilities, uses the PLC as the core deterministic control unit, retaining the inherent stability, reliability, and anti-interference capabilities of the PLC in the industrial control field, ensuring the safe and stable operation of the system under extreme conditions. The hierarchical control architecture avoids piling all complex logic into a single PLC controller, effectively balancing real-time performance and intelligence.

[0027] 5. The PLC-based automated billet retrieval control system of the present invention helps to reduce the operating energy consumption of the retrieval equipment by optimizing the energy consumption planning of the retrieval trajectory, which meets the requirements of modern industry for green manufacturing and sustainable development. Attached Figure Description

[0028] The invention will now be further described with reference to the accompanying drawings.

[0029] Figure 1 This is a system flowchart of the present invention; In the diagram: 1. Programmable Logic Controller (PLC) module; 11. High-speed industrial Ethernet interface; 2. Intelligent sensing and decision-making module; 21. Industrial-grade high-performance computing unit; 22. 3D vision perception and attitude estimation subunit; 23. Adaptive path planning subunit; 24. Force feedback intelligent grasping subunit; 3. Multi-degree-of-freedom grasping execution module; 31. Robotic arm; 32. Servo driver; 33. Gripper; 4. Sensor fusion module; 41. 3D vision sensor unit; 42. Force / torque sensor unit; 43. High-temperature infrared temperature measurement sensor unit; 44. Environmental parameter sensor unit; 45. Data acquisition and preprocessing unit; 46. Data fusion processing subunit; 5. High-speed industrial communication network. Detailed Implementation

[0030] This invention provides a PLC-based automated billet handling control system for metallurgical applications, addressing the challenges of high-precision, high-flexibility automated handling of high-temperature, heavy-load billets in the metallurgical industry. The system seamlessly integrates advanced sensing technology, real-time data processing and analysis, and AI-based intelligent decision-making algorithms with the inherent stability, reliability, and deterministic operating mechanism of the programmable logic controller (PLC), thereby achieving high-precision, adaptive automated billet handling in harsh metallurgical environments. The specific embodiments of this invention will be described in detail below with reference to the accompanying drawings to ensure that those skilled in the art can fully understand and implement this invention.

[0031] Reference Figure 1 The present invention relates to a PLC-based automated billet retrieval control system, the main components of which include a programmable logic controller (PLC) module 1, an intelligent sensing and decision-making module 2, a multi-degree-of-freedom retrieval execution module 3, a sensor fusion module 4, and a high-speed industrial communication network 5. The modules interact with each other via the high-speed industrial communication network 5, forming a closely collaborative intelligent automated system.

[0032] Specifically, the core programmable logic controller (PLC) module 1 functions as the central deterministic control unit of the entire system. This module is responsible for executing preset sequential control logic, processing emergency stop commands, implementing safety interlock protection functions, and interacting directly with the underlying servo drivers 32 and hydraulic proportional valves in the multi-degree-of-freedom retrieval execution module 3 in real time. The PLC module 1 is equipped with a high-performance processor unit, such as an industrial-grade microprocessor with a multi-core architecture, whose clock frequency is typically between hundreds of megahertz and several gigahertz, capable of supporting complex floating-point operations and high-speed logic processing. This module also integrates a large-capacity memory, for example, at least 32MB of non-volatile memory for program storage and at least 16MB of random access memory (RAM) for data processing and caching, ensuring stable and reliable storage and high-speed access to programs and data. The PLC module 1 is equipped with high-speed digital input / output (DI / DO) modules and analog input / output (AI / AO) modules. The digital input module typically supports high-speed counter functionality for accurately capturing pulse signals from encoders or limit switches; the digital output module drives relays or solid-state relays for controlling discrete actions such as motor start / stop and valve switching. The analog input module features high resolution (e.g., 16-bit or higher) and a high sampling rate for accurately acquiring continuous signals from force / torque sensors, temperature sensors, etc.; the analog output module outputs standard signals such as 0-10V or 4-20mA to control actuators such as hydraulic proportional valves or frequency converters. The PLC module 1 communicates bidirectionally in real-time with the intelligent sensing and decision-making module 2 and the multi-degree-of-freedom acquisition and execution module 3 via an integrated high-speed industrial Ethernet interface 11, such as a communication port supporting PROFINETIRT or EtherCAT protocols. The physical layer of this communication interface typically uses 100Mbps or Gigabit Ethernet standards and features RJ45 or M12 industrial connectors to ensure reliable connections even in harsh environments. The pre-set program logic within the PLC module 1 is developed using programming languages ​​defined by the IEC 61131-3 international standard, such as ladder diagrams (LD), function block diagrams (FBD), structured text diagrams (ST), or sequential function charts (SFC). This program logic encompasses the zero-position calibration procedure for the retrieval equipment, safety zone restriction logic to ensure the equipment operates within a specified space, emergency stop response procedures (compliant with EN ISO 13849 or IEC 62061 safety standards), and conversion logic for receiving high-level motion commands from the intelligent sensing and decision-making module 2 and converting them into low-level actuator drive signals. This conversion logic typically includes inverse kinematics calculations from joint space to Cartesian space, and the decomposition of the target pose into target position, velocity, and acceleration curves for each servo axis.The PLC module 1 also includes comprehensive diagnostic functions, capable of real-time monitoring of its own hardware status, CPU load, memory usage, communication link status, and the operating status of all connected modules. When any anomaly is detected, such as I / O module failure, communication link interruption, or program execution error, the PLC module 1 will immediately trigger a corresponding alarm or execute a preset safety shutdown procedure, and send the diagnostic information to the host computer or human-machine interface (HMI) for display via the high-speed industrial communication network 5. As a preferred embodiment of the present invention, the core programmable logic controller (PLC) module 1 integrates motion control function blocks conforming to the IEC61131-3 standard, such as MC_Power (for enabling / de-enabling servo axes), MC_MoveAbsolute (for absolute position motion), MC_MoveRelative (for relative position motion), and MC_Halt (for motion stop), etc. These function blocks greatly simplify the integration and programming work with the servo driver 32 in the multi-degree-of-freedom retrieval execution module 3. The scanning cycle of the PLC module 1 is optimized and is usually set to no more than 5 milliseconds. In a metallurgical environment, it can achieve a faster response speed and more precise control accuracy, ensuring real-time response and control of various key signals during the retrieval process. This is especially important for applications involving safety interlocks and emergency shutdowns.

[0033] The core function of the intelligent perception and decision-making module 2 is to realize real-time perception of the state of metallurgical steel billets, environmental modeling, intelligent path planning, and grasping strategy decision-making. The intelligent perception and decision-making module 2 consists of an industrial-grade high-performance computing unit 21. This computing unit 21 is equipped with a multi-core central processing unit (CPU), such as an Intel Xeon E3 series or AMD Ryzen Embedded series industrial-grade processor, with at least 8 physical cores and a high clock speed to provide powerful general-purpose computing capabilities. In addition, the computing unit 21 also integrates a high-performance graphics processing unit (GPU), such as an NVIDIA Quadro or Tesla series professional graphics card, or an embedded AI computing platform such as NVIDIA Jetson AXX Avier / Orin, which has hundreds to thousands of CUDA cores to accelerate the training and inference process of deep learning models. To further improve the execution efficiency of AI algorithms, the computing unit 21 can also be configured with dedicated NPU (Neural Processing Unit) or FPGA (Field Programmable Gate Array) and other artificial intelligence acceleration hardware. The intelligent sensing and decision-making module 2 is interconnected with the PLC module 1 via a high-speed industrial Ethernet interface and receives preprocessed fused sensing data from the sensor fusion module 4. The intelligent sensing and decision-making module 2 internally deploys multiple intelligent algorithm sub-units, including a 3D visual perception and attitude estimation sub-unit 22, an adaptive path planning sub-unit 23, and a force feedback intelligent grasping sub-unit 24. The industrial-grade high-performance computing unit 21 typically runs a real-time operating system (RTOS), such as VxWorks or QNX, or a Linux operating system with real-time extension capabilities (e.g., a Linux kernel with the PREEMPT_RT patch), to ensure deterministic execution and response time of key intelligent algorithms, meeting the stringent real-time requirements of industrial automation.

[0034] The sensor fusion module 4 integrates and processes heterogeneous data from various types of sensors, thereby providing the intelligent sensing and decision-making module 2 with comprehensive, accurate, and robust environmental and target information. The physical implementation of the sensor fusion module 4 is typically an industrial-grade embedded computer or a PC / 104 architecture controller with multiple high-speed data acquisition interfaces. It integrates a data acquisition and preprocessing unit 45, responsible for synchronously acquiring, digitally filtering, amplifying, calibrating, and converting the raw data from each sensor unit. The sensor fusion module 4 includes the following main sensor units: 3D Vision Sensor Unit 41: This unit consists of at least two high-resolution industrial-grade stereo cameras or one 3D LiDAR sensor. When using stereo cameras, each camera employs a global shutter CMOS image sensor with a pixel resolution of 2048x1536 and a frame rate of 50 frames per second to ensure clear image capture even in fast-moving and high-temperature environments. The camera's optics utilize low-distortion, high-transmittance industrial lenses and are equipped with industrial-grade high-temperature and dust-resistant protective covers with an IP67 protection rating. Internal active cooling functions, such as air cooling or water cooling systems, are integrated to prevent sensor performance degradation or damage due to high temperatures. When using a LiDAR sensor, it employs a 905nm wavelength laser, providing high-precision, long-distance measurement capabilities. Its scanning frequency is 50Hz, angular resolution is 0.1°, and the maximum measurement distance is up to 50 meters, making it particularly suitable for large furnace environments. The LiDAR sensor is also equipped with an environmental protection cover and a self-cleaning system, such as an air curtain blowing or scraping device, to address the effects of dust and steam on the optical window. The three-dimensional vision sensor unit 41 is typically installed above or to the side of the retrieval equipment. It can acquire three-dimensional point cloud data and high-precision image data of the furnace interior and the steel billet to be retrieved, for three-dimensional reconstruction of the environment and target recognition.

[0035] Force / torque sensor unit 42: This unit consists of a six-axis force / torque sensor with a compact and robust physical structure, typically operating on the principles of strain gauges or piezoelectric effect. The six-axis force / torque sensor is mounted at the end joint of the gripper 33 of the multi-degree-of-freedom grasping execution module 3, enabling real-time measurement of the mechanical components in the X, Y, and Z directions and the torque components around the X, Y, and Z axes. Its measurement range is set to ±500 Newtons for each of the X, Y, and Z axis mechanical components and ±50 Nm for each torque component, with resolutions of 0.1 Newtons and 0.01 Nm, ensuring accurate sensing of minute changes in contact force during the grasping process. The force / torque sensor unit 42 transmits real-time force / torque data to the sensor fusion module 4 via a digital interface supporting EtherCAT or PROFINET protocols, ensuring low latency and high reliability of data transmission.

[0036] High-temperature infrared temperature sensor unit 43: This unit consists of at least two industrial-grade non-contact infrared thermometers with a spectral response range of 8μm to 14μm. It is suitable for measuring the radiation temperature of high-temperature objects, especially in environments with flames or high-temperature steam interference. The thermometers have a measurement temperature range of 400°C to 1200°C, a measurement accuracy of ±1% of the reading, and a repeatability of ±0.5%FS, providing accurate data on the surface temperature of the steel billet. The infrared thermometers are typically mounted on the forearm of the retrieval equipment, maintaining a certain distance from the gripper 33, to monitor the surface temperature of the steel billet in real time. This provides crucial thermal deformation reference data for the intelligent gripping strategy, preventing improper gripping due to changes in the material properties of the high-temperature steel billet.

[0037] Environmental parameter sensor unit 44: This unit includes a dust concentration sensor and a steam concentration sensor. The dust concentration sensor typically employs the laser scattering principle, with a measurement range of 0.1 mg / m³ to 1000 mg / m³, an accuracy of ±5%, and a response time of less than 2 seconds. The steam concentration sensor employs a capacitive or impedance principle, with a measurement range of 0% to 100% RH (relative humidity), an accuracy of ±3% RH, and a response time of less than 5 seconds. The environmental parameter sensor unit 44 is used to monitor the severity of the retrieval environment in real time, such as assessing the air quality and humidity level inside the furnace, and providing environmental correction parameters for the vision system. For example, when the dust or steam concentration increases, the image processing algorithm of the visual perception and attitude estimation subunit 22 adaptively adjusts parameters, such as enhancing contrast, performing defogging or deblurring, to compensate for the impact of dust and steam on image quality, point cloud density, and measurement accuracy.

[0038] The data fusion processing subunit 46 in the sensor fusion module 4 employs advanced estimation algorithms, such as Extended Kalman Filter (EKF) or Unscented Kalman Filter (UKF), to perform multi-source fusion of data from the 3D vision sensor unit 41, force / torque sensor unit 42, high-temperature infrared temperature sensor unit 43, and the encoders of each joint in the multi-degree-of-freedom retrieval execution module 3. The Extended Kalman Filter linearizes the nonlinear system model and uses an iterative approach for state prediction and correction, while the Unscented Kalman Filter approximates the mean and covariance of the nonlinear function through an unscented transformation, avoiding the calculation of the Jacobian matrix. These fusion algorithms can fully utilize the complementary information of each sensor, effectively suppressing noise and errors from a single sensor, thereby generating high-precision, highly robust models of the billet's 3D position (X, Y, Z coordinates), attitude (roll angle, pitch angle, yaw angle), geometry, surface temperature, and the real-time state of the retrieval equipment's end effector. The state model will be transmitted to the intelligent sensing and decision-making module 2 via the high-speed industrial communication network 5 in a unified data format, such as ROS (Robot Operating System) messages or custom data packets.

[0039] The 3D visual perception and pose estimation subunit 22 receives fused perception data from the sensor fusion module 4 and performs the following core functions: Billet Recognition and Segmentation: This sub-unit employs a deep learning-based convolutional neural network (CNN) model, such as a model based on U-Net or MaskR-CNN architecture, to perform real-time analysis of 3D point cloud data or preprocessed image data. The U-Net model, with its symmetrical encoder-decoder structure, excels in medical image segmentation and can be used in this invention to accurately segment the billets to be retrieved from the background inside the furnace. MaskR-CNN adds pixel-level instance segmentation capabilities to object detection, enabling simultaneous identification of the billet's location and generation of its precise contour mask. The model is trained offline on a billet dataset containing various shapes, sizes, surface features (e.g., oxide scale, roughness), and environmental disturbances (e.g., uneven lighting, smoke). To adapt to the complexity and dynamic changes of the metallurgical environment, the model also supports online fine-tuning or incremental learning to improve generalization ability and robustness under actual working conditions.

[0040] 3D attitude and geometric shape estimation of steel billets: Point cloud registration algorithms, such as the Iterative Closest Point (ICP) algorithm or feature-based registration algorithms (e.g., FPFH feature matching combined with RANSAC), are used to accurately match the identified steel billet point cloud data with a pre-set steel billet CAD model. The ICP algorithm iteratively calculates the rigid body transformation between point clouds, minimizing the distance between corresponding points, thereby accurately estimating the 3D position (X, Y, Z coordinates) and attitude (roll angle, pitch angle, yaw angle) of the steel billet. This process can also detect possible deformation, bending, or misalignment of the steel billet, providing a basis for correcting subsequent grasping strategies. The algorithm has been optimized to tolerate a certain degree of occlusion and sensor noise, ensuring accuracy inside complex furnaces.

[0041] Furnace environment modeling and obstacle detection: Based on point cloud data acquired by the 3D vision sensor 41, the sub-unit constructs a high-precision 3D environment model of the retrieval area in real time. This model includes not only the static furnace wall and furnace bottom structure, but also the dynamic position and geometric information of adjacent billets, furnace equipment, maintenance tools, or other potential obstacles. The point cloud is represented by a data structure based on voxel grid or octree, and a collision detection algorithm (such as the GJK algorithm or a distance field-based algorithm) is used to monitor in real time whether there are potential collision risks in the movement space of the retrieval equipment, providing a real-time obstacle map for adaptive path planning.

[0042] The adaptive path planning subunit 23 receives the billet's three-dimensional posture, geometric shape, and environmental obstacle information from the three-dimensional vision perception and posture estimation subunit 22, and combines this information with the current pose and kinematic model (including forward and inverse kinematic models, as well as dynamic constraints such as joint limits, velocity, and acceleration limits) of the multi-degree-of-freedom retrieval execution module 3 to generate or update the optimal motion trajectory of the retrieval device in real time. The adaptive path planning subunit 23 employs a sampling-based method, such as the Fast Random Tree (RRT*) algorithm or a heuristic search algorithm (such as the A* algorithm). The RRT algorithm constructs an exploration tree through random sampling space and optimizes the path through a rewiring mechanism. The A* algorithm searches for the shortest path in a graph search using an evaluation function (f(n) = g(n) + h(n)). These algorithms comprehensively consider multiple optimization objectives during the planning process, including the kinematic and dynamic constraints of the robotic arm, collision avoidance with environmental obstacles, motion smoothness, and time or energy optimization. By introducing spline interpolation or Bézier curve smoothing techniques into the algorithm, the generated trajectory is ensured to be continuous and smooth, avoiding impacts during the robotic arm's movement. The path planning algorithm possesses a high degree of online replanning capability, capable of rapid online replanning or trajectory adjustment within milliseconds based on real-time detected billet position deviations, posture changes, or new environmental obstacles, thereby ensuring the flexibility and robustness of the retrieving operation. In a preferred embodiment of the invention, the adaptive path planning subunit 23 additionally considers energy consumption optimization factors when generating the retrieving trajectory. By introducing a detailed robotic arm dynamics model and motor efficiency curve into the path planning algorithm, energy consumption is taken as one of the optimization objectives, planning a motion trajectory with minimal energy consumption while meeting time constraints, accuracy requirements, and collision-free conditions. For example, the algorithm tends to select smoother acceleration / deceleration curves and fewer directional changes to reduce motor energy consumption during high-frequency start-stop and acceleration / deceleration. This energy consumption optimization strategy helps to significantly reduce operating costs in the metallurgical production process, meeting the requirements of modern industry for green manufacturing and sustainable development.

[0043] The force feedback intelligent grasping subunit 24 receives real-time force / torque data from the sensor fusion module 4 and target grasping point information and target grasping force parameters from the adaptive path planning subunit 23. The force feedback intelligent grasping subunit 24 achieves intelligent grasping through the following refined methods: Contact Detection and Position Fine-Tuning: As the gripper 33 approaches the billet, the subunit continuously monitors the force / torque data transmitted by the force / torque sensor unit 42 in real time. Once the contact force with the billet exceeds a preset micro-threshold (e.g., 0.5 Newtons or 0.05 Nm) in one or more axes, it is determined that the gripper has made contact with the billet surface. At this time, the subunit will adaptively fine-tune the position and attitude of the gripper at the micrometer level according to the direction and magnitude of the contact force through impedance control or admittance control strategies. The impedance control strategy establishes a virtual mechanical impedance relationship between force control and position control, enabling the robotic arm to exhibit the expected compliance when in contact with the environment. Admittance control calculates and adjusts the end effector speed of the robotic arm based on the contact force measured by the force sensor, allowing it to adapt to the external environment. These strategies ensure that the gripper can cover the billet at the optimal angle and position, avoiding gripping failures caused by uneven billet surface, oxide scale, or initial attitude estimation errors, effectively improving the gripping success rate.

[0044] Adaptive gripping force control: During the gripping process, the clamping force of the gripper 33 is dynamically adjusted based on the billet weight estimation, surface friction coefficient (obtainable through experiments or table lookup), real-time surface temperature from the high-temperature infrared temperature sensor unit 43 (high temperature reduces material strength and changes frictional characteristics), and the billet sliding trend detected during actual gripping (judged by the rate of change of force / torque sensor data or internal displacement sensor). The gripping force control employs a PID (proportional-integral-derivative) control algorithm combined with fuzzy logic or an adaptive gain scheduling strategy. The PID controller performs closed-loop control based on the deviation between the actual gripping force and the target gripping force; fuzzy logic can be used to handle nonlinear and uncertain gripping conditions, dynamically adjusting PID parameters or directly outputting control quantities based on multiple input variables (such as temperature and sliding trend). The adaptive gain scheduling strategy automatically adjusts the controller gain according to the system operating status to optimize control performance. These control strategies ensure that the billet is firmly gripped without damaging the billet surface (e.g., causing indentations or cracks) or wasting energy due to over-gripping. The gripping force setting will be preset based on the material characteristics, size, weight, and maximum allowable stress of the steel billet, and will be corrected through real-time force feedback and sliding trend judgment to adapt to different steel grades and working conditions.

[0045] Grasping Status Confirmation and Anomaly Handling: The subunit continuously monitors force / torque data, as well as data from the displacement and pressure sensors inside the gripper, to determine whether the gripping was successful and whether the billet slipped or fell during the gripping process. If an anomaly is detected, such as the gripping force failing to reach the set value for an extended period, suddenly decreasing (potentially indicating billet slippage), or the displacement sensors inside the gripper showing an abnormal billet position, an anomaly signal is immediately sent to the PLC module 1, triggering a preset anomaly handling procedure, such as safely repositioning the gripper, adjusting the gripping posture, or safely releasing the billet and issuing an alarm.

[0046] The multi-degree-of-freedom grasping execution module 3 functions to execute high-precision motion commands issued by the PLC module 1. The grasping execution module 3 can be a multi-joint robotic arm 31, such as a SCARA robot or a six-axis articulated robot, possessing at least six degrees of freedom to ensure flexible grasping and complex posture transfer of steel billets. Alternatively, it can be a gantry robot system, typically possessing three Cartesian coordinate axes of motion freedom, and achieving multi-degree-of-freedom operation through additional rotary joints or telescopic mechanisms. Each joint of the robotic arm is equipped with a high-precision servo motor and a corresponding servo driver 32. The servo motor is a high-performance permanent magnet synchronous motor, possessing high torque density, high dynamic response, and high positioning accuracy. The servo driver 32 typically supports multiple control modes, such as position mode, speed mode, and torque mode, and communicates with the PLC module 1 in real time via EtherCAT or PROFINET protocols to achieve high-precision synchronous motion control. The grasping execution module 3 also includes a custom-designed electric or hydraulic gripper 33 with high-temperature protection and adaptive clamping functions. The gripper 33 is constructed from high-temperature resistant alloy materials with a special surface coating and an integrated active cooling system. Its clamping mechanism is designed to adaptively adjust to the geometry and shape of the steel billet (e.g., square, round, or rectangular cross-section billets), ensuring stable and reliable clamping and gripping of billets of different sizes. The gripper 33 integrates independent displacement sensors (e.g., linear encoders) and pressure sensors for real-time monitoring of the clamping status, clamping force distribution, and the relative position of the billet within the gripper.

[0047] The high-speed industrial communication network 5 is designed to enable efficient, low-latency, and deterministic data transmission between the core programmable logic controller (PLC) module 1, the intelligent sensing and decision-making module 2, the sensor fusion module 4, and the multi-degree-of-freedom acquisition and execution module 3. The network employs industrial Ethernet technology based on the IEEE 802.3 standard, specifically implemented using protocols such as PROFINETIRT (isochronous real-time) or EtherCAT. PROFINETIRT uses dedicated ASIC hardware to implement frame scheduling and priority control, ensuring real-time transmission in the switching network. EtherCAT employs an "on-the-fly" processing mechanism, where data frames are directly read and written as they pass through each node, significantly reducing protocol stack processing time. The network's data transmission rate can be configured to 100Mbps or 1Gbps, and the communication cycle can be configured between 1ms and 4ms according to application requirements to meet the stringent real-time requirements of the system, especially for closed-loop control of motion control and force feedback. The network topology can be either star or ring. Star topology facilitates fault isolation and expansion, while ring topology supports network redundancy, such as PROFINET's Media Redundancy Protocol (MRP), which automatically switches paths in the event of a single point of failure (such as a broken network cable) to enhance system reliability and resilience against single points of failure. All network components, including industrial switches, communication cables (such as CAT5e or CAT6 shielded cables), and connectors (such as M12 or RJ45 industrial connectors), comply with industrial EMC (electromagnetic compatibility) immunity standards (such as the IEC61000 series) and protection level requirements (such as IP65 or IP67) to ensure long-term stable operation in the harsh environment of a metallurgical plant.

[0048] The specific working principle of this invention is as follows: After the system starts, the three-dimensional vision sensor unit 41 and the high-temperature infrared temperature sensor unit 43 continuously acquire three-dimensional environmental point cloud data, high-precision image data of the steel billet to be retrieved, and surface temperature data of the steel billet inside the furnace at a high frame rate and a high sampling rate. The environmental parameter sensor unit 44 simultaneously monitors the dust and steam concentration inside the furnace. All raw data from these sensors, including point clouds, image frames, temperature values, dust and steam concentrations, are aggregated to the data acquisition and preprocessing unit 45 of the sensor fusion module 4 through their respective high-speed interfaces. This unit performs time synchronization, noise filtering (e.g., median filtering or Gaussian filtering), sensor data calibration (e.g., camera intrinsic and extrinsic parameter calibration, thermometer emissivity correction), and format conversion on the raw data to ensure data quality and consistency. Subsequently, the data fusion processing subunit 46 of the sensor fusion module 4 uses the extended Kalman filter (EKF) or unscented Kalman filter (UKF) algorithm to fuse the preprocessed heterogeneous data in real time. The fusion inputs include a coarse pose of the billet from 3D vision, force on the end effector from a force / torque sensor, surface temperature of the billet from infrared thermography, and precise position data from the encoders of each joint in the multi-degree-of-freedom retrieval execution module 3. Through multi-source fusion, the system generates a high-precision real-time perception model that includes the billet's precise 3D position, orientation, geometry, surface temperature, and the distribution of obstacles in the furnace environment (such as other billets and the furnace wall).

[0049] The real-time perception model is transmitted to the intelligent perception and decision-making module 2 via a high-speed industrial communication network 5 in the form of structured data packets. The 3D visual perception and pose estimation subunit 22 in the intelligent perception and decision-making module 2 first performs deep analysis on the fused perception data. It utilizes a pre-trained deep learning model (such as Mask R-CNN) to accurately identify the target steel billet to be retrieved inside the furnace and segment it from the complex background. Subsequently, using a point cloud registration algorithm (such as ICP), the segmented steel billet point cloud is matched with a preset CAD model to estimate the precise 3D pose of the steel billet (including position and pose represented by Euler angles or quaternions). Simultaneously, this subunit also constructs a real-time 3D environment model of the furnace and detects all potential obstacles in real time. After acquiring the precise pose of the target billet and information about environmental obstacles, the adaptive path planning subunit 23 combines the current pose, kinematics, and dynamics model of the multi-degree-of-freedom retrieval execution module 3 with algorithms such as RRT or A to generate a multi-segment collision-free trajectory in real time from the current position to the target billet grabbing point and then to the safe placement point, while ensuring collision-free, smooth, and optimal motion (e.g., optimal time or optimal energy consumption). This trajectory data typically includes a series of target joint angles, end effector poses, or velocity profiles, which are encapsulated into high-level motion instructions and sent to the core programmable logic controller (PLC) module 1 via the high-speed industrial communication network 5.

[0050] Upon receiving the high-level motion commands, the core programmable logic controller (PLC) module 1 immediately initiates its internal motion control algorithms. These algorithms decompose the high-level trajectory commands into low-level control signals for each servo driver 32 in the multi-degree-of-freedom grasping execution module 3, such as periodic position, speed, or current commands. Upon receiving the commands, the servo drivers 32 drive the corresponding servo motors, thereby enabling the robot arm 31 or gantry system to move precisely along the planned trajectory. When the gripper 33 approaches the billet during its movement, the force / torque sensor unit 42 begins transmitting real-time force / torque data to the sensor fusion module 4 at a high frequency (e.g., 1 kHz). This data is then rapidly transmitted to the force feedback intelligent grasping subunit 24 of the intelligent sensing and decision-making module 2 for processing. Based on the real-time contact force feedback, the force feedback intelligent grasping subunit 24 fine-tunes the approach posture of the gripper 33 using impedance control or admittance control strategies to ensure that the gripper contacts the billet in the optimal posture. Once contact is stable, the subunit adaptively controls the clamping force of the gripper 33 to ensure the billet is gripped securely and without damage, while continuously monitoring the gripping status to prevent slippage or drop. After the billet is successfully gripped and its stability is confirmed, the adaptive path planning subunit 23 replans the transfer trajectory from the gripping point to the designated placement point and sends new instructions to the PLC module 1. The PLC module 1 then controls the retrieval equipment to safely and smoothly transfer the billet to the designated location and release it precisely.

[0051] Throughout the retrieval process, the PLC module 1 continuously executes safety interlocking and diagnostic functions according to its deterministic control cycle. If any abnormality is detected, such as any sensor failure (data loss or out-of-range), actuator overload (current or temperature exceeding limits), communication link interruption, or program logic execution error, the PLC module 1 will immediately trigger the preset emergency shutdown procedure and perform corresponding safety actions according to the fault type (e.g., moving the robotic arm to a safe position or immediately cutting off the power). It will also send detailed alarm information to the intelligent sensing and decision-making module 2 and the human-machine interface (HMI) through the high-speed industrial communication network 5 to ensure the safety of the system, equipment, and on-site operators.

[0052] Example In one specific embodiment, the PLC-based automated billet retrieval control system proposed in this invention is applied to the automatic billet unloading and transfer process of a continuous casting production line.

[0053] System configuration details: Core Programmable Logic Controller (PLC) Module 1: Employs a Siemens S7-1518F-4PN / DP CPU, equipped with an ET200SP distributed I / O module. This PLC features a high-performance multi-core processor with a scan cycle set to 1 millisecond, ensuring real-time response to motion control commands. It integrates a PROFINETIRT interface with a communication cycle of 2 milliseconds. Safety functions comply with IEC61508 SIL3 level.

[0054] Intelligent Sensing and Decision-Making Module 2: Utilizes an Advantech IPC-610H industrial computer, equipped with an Intel Xeon E3-1505Mv5 processor and an NVIDIA Quadro P5000 GPU. The operating system is Ubuntu 18.04 LTS with PREEMPT_RT kernel. Internally, Docker containers are used to deploy 3D visual perception, path planning, and force feedback algorithms.

[0055] Sensor fusion module 4: It adopts a custom NI CompactRIO controller with a built-in FPGA for high-speed data synchronization and preprocessing.

[0056] 3D Vision Sensor Unit 41: Two Baslerace2ProacA2040-180gc global shutter cameras, with a pixel resolution of 2048x1536, a frame rate of 100fps, equipped with 25mm industrial lenses, and an IP67-rated active water-cooled protective cover. Installed above the furnace, approximately 3 meters from the steel billet.

[0057] Force / torque sensor unit 42: ATII Industrial Automation Axia80 six-axis force / torque sensor, measurement range ±800N / ±80Nm, resolution 0.05N / 0.005Nm. Connects to NI CompactRIO via EtherCAT interface.

[0058] High-temperature infrared temperature sensor unit 43: Two OptrisCTlaser3M non-contact infrared thermometers, measuring range 50°C-1800°C, spectral response 2.3μm, accuracy ±1% of reading. Installed 1 meter in front of the robotic arm gripper.

[0059] Environmental parameter sensor unit 44: a dust concentration sensor from "SenseAir" and a vapor concentration sensor from "Vaisala", both with industrial-grade protection and measurement accuracy.

[0060] Multi-DOF Grabbing Execution Module 3: KUKAKR60-3L30-2SP 6-axis heavy-duty robot arm, maximum load 60kg, repeatability ±0.06mm, working range 3000mm. Each joint is equipped with a KUKAKCP4 controller and SEWEURODRIVE servo drive, communicating with PLC module 1 via PROFINETIRT. The custom gripper 33 is made of high-temperature alloy, with double-sided jaw plates, and integrates a Hall effect displacement sensor and a piezoelectric pressure sensor, with an adjustable gripping force range of 500N~3000N.

[0061] High-speed industrial communication network 5: Utilizes CAT6a shielded industrial Ethernet cables throughout, Siemens SCALANCEX M408 industrial switches, and supports PROFINETIRT and MRP redundancy. Communication cycle is set to 2 milliseconds.

[0062] Billet retrieval operation process: Initial state: The steel billet to be retrieved from the furnace measures 200mm x 200mm x 1200mm, weighs approximately 380kg, and has a surface temperature of approximately 950°C. The robotic arm 31 is in the safe docking position.

[0063] Perception and Modeling: After system startup, the Basler stereo camera 41 and Optris infrared thermometer 43 begin real-time acquisition of images, point clouds, and billet surface temperatures inside the furnace. SenseAir and Vaisala sensors 44 monitor the furnace dust concentration at approximately 150 mg / m³ and the steam concentration at approximately 40% RH. This data is aggregated to the NI CompactRIO controller (sensor fusion module 4) for time synchronization, noise reduction, and calibration. The data fusion processing subunit 46 uses the UKF algorithm to fuse data from the camera, infrared thermometer, and robotic arm encoder to generate the real-time three-dimensional pose of the billet (e.g., relative to the furnace reference coordinate system, the center point is at (1500, 800, 200) mm, with a roll angle of 5°, a pitch angle of -2°, and a yaw angle of 10°) and a precise geometric model.

[0064] Intelligent Decision-Making: The real-time perception model is transmitted to the Advantech industrial computer (Intelligent Perception and Decision-Making Module 2). The 3D visual perception and posture estimation subunit 22 uses the MaskR-CNN model to perform real-time inference on the GPU, identifying the target billet and estimating its precise pose. The adaptive path planning subunit 23, based on the kinematic model of the KUKA robotic arm, its current pose, and information on obstacles in the furnace (such as other billets and furnace walls), combined with the RRT* algorithm and energy consumption optimization objectives, plans a collision-free, smooth, and energy-optimized trajectory from the safe stopping position to the target billet gripping point (expected gripping center point), and then to the designated placement point. The planned trajectory is completed within 25 milliseconds, generating a series of joint angle trajectory points.

[0065] Precise Execution and Force Feedback: Joint angle trajectory points are sent as high-level motion commands to the Siemens S7-1518 PLC module 1. The PLC converts these commands into low-level motion commands for the SEW servo drive 32 in the KUKA robot KCP4 controller. The robotic arm 31 begins to move along the planned trajectory. When the gripper 33 is approximately 50mm away from the billet, data from the ATI six-axis force / torque sensor 42 begins to be monitored in real time by the force feedback intelligent gripping subunit 24. When a weak contact force (e.g., 2N) with the billet is detected, the subunit immediately initiates impedance control, fine-tuning the gripper's posture within a range of ±1mm and ±0.5° to ensure precise centering and coverage of the billet. Subsequently, based on the billet's weight, the surface temperature of 950°C (considering thermal deformation and material softening), and the sliding trend monitored by the gripper's internal displacement sensor, the subunit dynamically adjusts the gripper's clamping force using PID control combined with fuzzy logic. The initial clamping force is set to 2500N. If a sliding tendency is detected, the clamping force will automatically increase by 100N until it reaches a maximum of 3000N.

[0066] Transfer and Release: After the billet is firmly grasped, the adaptive path planning subunit 23 replans the trajectory from the grasping point to the designated placement position. The PLC module 1 controls the robotic arm 31 to smoothly transfer the billet to the designated cooling bed at a maximum speed of 0.8 m / s according to the new planned trajectory. Upon reaching the placement position, the gripper 33, under the control of the force feedback subunit, gradually reduces the clamping force to below 100 N and fully opens, safely releasing the billet. The entire retrieval and transfer process takes approximately 45 seconds.

[0067] Safety and Diagnostics: Throughout the process, PLC module 1 continuously monitors the status of all sensors and actuators. For example, if the lidar experiences data anomalies or communication interruptions, the PLC will immediately trigger an emergency stop, move the robotic arm to a preset safe area, and issue an alarm to the operator.

[0068] Comparative Example As a comparative example of the present invention, a billet retrieval system based on a traditional PLC and a single two-dimensional vision system is used.

[0069] System configuration details: The core programmable logic controller (PLC) module uses a Siemens S7-300 series PLC. It lacks an integrated real-time Ethernet motion control interface and communicates with the servo drive via PROFIBUS-DP. The scan cycle is 10 milliseconds.

[0070] Vision system: An industrial-grade 2D camera with a pixel resolution of 1280x960 and a frame rate of 30fps, equipped with an IP65 protective cover and without active cooling. It is mounted above the furnace chamber and transmits image data to an industrial PC for processing via an RS-232 serial port.

[0071] Sensors: Only one non-contact infrared thermometer is provided (measuring range 100°C-1200°C, accuracy ±2%). No force / torque sensor, no environmental parameter sensor.

[0072] Retrieval Execution Module: A three-axis electric gantry robot arm, driven by a stepper motor, with a repeatability of ±0.5mm. Equipped with a fixed-size hydraulic gripper, the clamping force is not adjustable and is controlled by a hydraulic pump station and solenoid valves.

[0073] Communication network: The PLC communicates with the vision PC via Modbus TCP / IP, and the PLC communicates with the servo drive via PROFIBUS-DP. The data transmission rate is limited and the response time is relatively long.

[0074] Billet retrieval operation process: Initial state: Same as the previous example, the steel billet to be retrieved in the furnace is 200mm x 200mm x 1200mm in size, weighs 380kg, and has a surface temperature of approximately 950°C. The gantry robotic arm is in the safe docking position.

[0075] Perception and Modeling: A 2D camera acquires top-view images of the steel billet inside the furnace. An industrial PC performs edge detection and geometric matching on the images to estimate the 2D center position and approximate orientation of the billet, but lacks 3D pose information. An infrared thermometer measures the surface temperature of the billet. Due to the lack of environmental parameter sensors, environmental correction cannot be performed on the images.

[0076] Decision-making and execution: The industrial PC sends two-dimensional position and temperature information to the S7-300 PLC via Modbus TCP / IP. The PLC controls the gantry robot arm to move above the steel billet according to a preset fixed path program. Due to the lack of three-dimensional posture information and online path planning capabilities, the system cannot identify the billet's tilt or misalignment, nor can it detect dynamic obstacles within the furnace. The motion trajectory is linear, with no energy consumption optimization.

[0077] Grasping: The robotic arm lowers the hydraulic gripper to a preset height. Due to force feedback, the gripper holds the billet with a fixed clamping force (e.g., 2800N). If the billet surface is uneven or the initial positioning is off, the gripper may not completely cover the billet, easily leading to unstable gripping or slippage. The gripping force for high-temperature billets cannot be dynamically adjusted according to temperature.

[0078] Transfer and Release: After the billet is gripped, the robotic arm transfers it to the placement position along a preset path. Due to the stepper motor drive, the movement is not very smooth and may vibrate. Upon reaching the placement position, the hydraulic gripper releases the billet in a fixed manner.

[0079] Safety and Diagnostics: The PLC only performs basic hard-wired interlocking and alarm functions and cannot diagnose complex sensor faults or anomalies at the intelligent decision-making level.

[0080] Comparison of experimental data To quantify the performance advantages of the PLC-based automated billet handling control system (example) provided by this invention compared to a traditional system (comparative example), we conducted multiple batches of handling experiments under different working conditions and collected the following key performance indicators. In the experiments, each handling operation was performed on a standard billet with dimensions of 200mm x 200mm x 1200mm and a temperature between 900°C and 1000°C. Each system underwent 100 repeated handling tests.

[0081] Performance indicators Embodiments of the present invention Comparison System Increase in magnitude (relative to the comparative ratio) Remark Average retrieval success rate 99.5% 85.0% +14.5% Success rate is defined as the steel billet being stably grasped and successfully released at the target position. Average time per retrieval 45.3 seconds 78.5 seconds -42.4% The time required from the start of the movement to the complete release of the billet and return to the safe position Target billet positioning accuracy (end point) Average error < 1.0 mm Average error 5-8 mm Significant improvement Positional deviation when the gripper is precisely aligned with the target steel billet, especially the Z-axis (depth) deviation. End effector attitude accuracy Average error < 0.5 degrees Average error 2-3 degrees Significant improvement Angle alignment deviation between the gripper and the billet surface Energy consumption optimization effect The average energy consumption per retrieval is 2.8 kWh. The average energy consumption per retrieval is 4.1 kWh. -31.8% Only calculate the energy consumption of the execution module and auxiliary equipment. Robustness (dust / vapor effects) Success rate decrease <1% (at 200 mg / m³ dust and 60% RH vapor) Success rate decreases by >10% (under 100mg / m³ dust and 30%RH vapor conditions). Significant improvement Changes in the success rate of operations under simulated harsh environments Online path replanning response time Average 25 milliseconds Not supported N / A The time required for the system to replan the path when obstacles or target positions change. Surface damage rate of steel billet ＜0.1% Approximately 5% Significantly reduced Indentations or scratches on the surface of the steel billet caused by excessive gripping force or improper handling Data Analysis and Technical Results: As can be seen from the above experimental data, the PLC-based automated billet retrieval control system provided by this invention is significantly superior to the traditional comparative system in all key performance indicators.

[0082] First, the average success rate of retrieval increased from 85.0% to 99.5%, which directly reflects the effectiveness of the invention's high-precision and robust sensing capabilities achieved through multi-sensor fusion in complex metallurgical environments, as well as the effectiveness of the adaptive grasping strategy achieved through intelligent decision-making and force feedback. Traditional systems, lacking three-dimensional attitude perception and force feedback, are prone to grasping failures due to billet position deviations, attitude tilts, or surface unevenness.

[0083] Secondly, the average time for a single retrieval was significantly reduced from 78.5 seconds to 45.3 seconds, an improvement of 42.4%. This is mainly due to the high-speed industrial communication network, the rapid response capability of the high-performance PLC, and the rapid calculation of the optimal trajectory by the adaptive path planning subunit of this invention. Traditional systems suffer from low overall operational efficiency due to communication delays, insufficient computing power, and fixed paths.

[0084] Furthermore, this invention achieves significant improvements in both the positioning accuracy of the target billet and the attitude accuracy of the end effector, with average errors controlled below 1.0 mm and 0.5 degrees, respectively. This high precision is achieved through a combination of a 3D visual perception and attitude estimation subunit and a point cloud registration algorithm, ensuring perfect alignment between the gripper and the billet and effectively preventing mis-gripping or slippage caused by angular deviations.

[0085] In terms of energy consumption optimization, the average energy consumption per retrieval in this invention is reduced by 31.8%. This confirms the effectiveness of introducing an energy consumption optimization algorithm into the adaptive path planning subunit. By planning a smoother and more efficient motion trajectory, the energy loss of the robotic arm during high-frequency start-stop and acceleration / deceleration is reduced, which is in line with the concept of green manufacturing.

[0086] Furthermore, this invention demonstrates superior robustness to harsh environments. In simulated environments with high dust and high vapor concentrations, the success rate of this invention drops by less than 1%, while that of conventional systems exceeds 10%. This is thanks to the real-time correction data provided by the environmental parameter sensor unit, which enables the 3D vision system to adaptively adjust image processing parameters, effectively compensating for environmental interference.

[0087] Finally, this invention significantly reduces the damage rate to the steel billet surface from approximately 5% to below 0.1% through the refined control of the force feedback intelligent gripping subunit. When retrieving high-temperature steel billets, the adaptive clamping force control can dynamically adjust according to real-time temperature and sliding trends, avoiding indentations or damage to the steel billet caused by excessive traditional fixed clamping force, thereby improving product quality.

[0088] In summary, the PLC-based automated billet retrieval control system provided by this invention, through its unique system architecture and deep integration of intelligent algorithms, has not only achieved a qualitative leap in retrieval efficiency, accuracy, robustness, and energy consumption, but also demonstrated unparalleled advantages in ensuring billet quality and operational safety, fully reflecting its non-obviousness and practicality.

[0089] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A PLC-based automated billet retrieval control system, characterized in that: include: The programmable logic controller module (1), as the central control unit of the entire system, is responsible for executing the preset sequential control logic, safety interlock protection, emergency shutdown processing, and interacting with the control instructions of the multi-degree-of-freedom retrieval execution module; The intelligent perception and decision-making module (2) serves as the core intelligent unit for realizing real-time perception of the state of metallurgical steel billets, environmental modeling, intelligent path planning, and grasping strategy decision-making. The multi-degree-of-freedom grabbing execution module (3) is used to execute high-precision motion commands issued by the programmable logic controller module (1) to realize flexible grabbing and transfer of steel billets; The sensor fusion module (4) is used to integrate and process heterogeneous data from sensors, providing comprehensive, accurate and robust environmental and target information for the intelligent sensing and decision-making module (2). A high-speed industrial communication network (5) is used to realize efficient, low-latency, deterministic data transmission between the programmable logic controller module (1), the intelligent sensing and decision-making module (2), the sensor fusion module (4), and the multi-degree-of-freedom retrieval and execution module (3).

2. The PLC-based automated billet retrieval control system as described in claim 1, characterized in that: The programmable logic controller module (1) is equipped with a high-performance processor unit, a large-capacity memory, a high-speed digital input / output module and an analog input / output module; The programmable logic controller module (1) exchanges bidirectional real-time data with the intelligent sensing and decision-making module (2) and the multi-degree-of-freedom retrieval and execution module (3) through the integrated high-speed industrial Ethernet interface (11).

3. The PLC-based automated billet retrieval control system as described in claim 2, characterized in that: The program logic preset inside the programmable logic controller module (1) includes the zero-position calibration program of the retrieval device, the safety area restriction logic, the emergency stop response program, and the conversion logic that receives high-level motion instructions from the intelligent sensing and decision-making module (2) and converts them into low-level actuator drive signals. The programmable logic controller module (1) also includes diagnostic functions, which can monitor its own hardware status, communication link status and the operating status of each connected module in real time, and trigger corresponding alarms or safety shutdown procedures when an abnormality is detected.

4. The PLC-based automated billet retrieval control system as described in claim 1, characterized in that: The intelligent sensing and decision-making module (2) is composed of an industrial-grade high-performance computing unit (21), which is equipped with a multi-core central processing unit, a graphics processor and dedicated artificial intelligence acceleration hardware. The intelligent sensing and decision-making module (2) is interconnected with the programmable logic controller module (1) through a high-speed industrial Ethernet interface and receives preprocessed fused sensing data from the sensor fusion module (4). The intelligent perception and decision-making module (2) is internally deployed with a three-dimensional visual perception and attitude estimation subunit (22), an adaptive path planning subunit (23), and a force feedback intelligent grasping subunit (24); the industrial-grade high-performance computing unit (21) runs a real-time operating system or a Linux operating system with real-time expansion capabilities to ensure the deterministic execution and response time of key intelligent algorithms.

5. The PLC-based automated billet retrieval control system as described in claim 1, characterized in that: The sensor fusion module (4) includes: The three-dimensional vision sensor unit (41) consists of at least two high-resolution industrial-grade stereo cameras or one three-dimensional lidar (LiDAR) sensor, used to acquire three-dimensional point cloud data and high-precision image data of the furnace interior and steel billet. The force / torque sensor unit (42) consists of a six-axis force / torque sensor and is installed at the end joint of the gripper (33) of the multi-degree-of-freedom retrieval execution module (3) for real-time measurement of force / torque data. The high-temperature infrared temperature sensor unit (43) consists of at least two industrial-grade non-contact infrared thermometers, which are installed on the forearm of the retrieval equipment to monitor the surface temperature of the steel billet to be retrieved in real time. The environmental parameter sensor unit (44) includes a dust concentration sensor and a steam concentration sensor, which are used to monitor the severity of the harvesting environment in real time and provide environmental correction parameters for the vision system.

6. The PLC-based automated billet retrieval control system as described in claim 5, characterized in that: The sensor fusion module (4) integrates a data acquisition and preprocessing unit (45), which is responsible for synchronously acquiring, noise filtering, data calibration and format conversion of the raw data from the three-dimensional vision sensor unit (41), force / torque sensor unit (42), high temperature infrared temperature sensor unit (43) and environmental parameter sensor unit (44). The data fusion processing subunit (46) in the sensor fusion module (4) uses the extended Kalman filter (EKF) or unscented Kalman filter (UKF) algorithm to perform multi-source fusion on encoder data from the three-dimensional vision sensor unit (41), force / torque sensor unit (42), high-temperature infrared temperature sensor unit (43) and multi-degree-of-freedom retrieval execution module (3) to generate a high-precision, high-robust three-dimensional position, attitude, geometry, surface temperature of the billet and the real-time state model of the end effector of the retrieval equipment.

7. The PLC-based automated billet retrieval control system according to claim 1, characterized in that: The multi-degree-of-freedom retrieval execution module (3) includes a multi-joint robotic arm (31) or a gantry robot system, which has at least six degrees of freedom; Each joint of the robotic arm (31) is equipped with a high-precision servo motor and a corresponding servo driver (32), which communicates with the programmable logic controller module (1) in real time via EtherCAT or PROFINET protocol; The retrieval execution module (3) also includes an electric or hydraulic gripper (33), which integrates an independent displacement sensor and a pressure sensor to monitor the gripping status.

8. The PLC-based automated billet retrieval control system according to claim 1, characterized in that: The high-speed industrial communication network (5) adopts a star or ring topology and supports network redundancy.

9. The PLC-based automated billet retrieval control system according to claim 4, characterized in that: The 3D visual perception and pose estimation subunit (22) receives fused perception data from the sensor fusion module (4) and performs the following functions: The billet identification and segmentation uses a deep learning-based convolutional neural network (CNN) model to perform real-time analysis on 3D point cloud data or processed image data, accurately identify the billets to be retrieved inside the furnace, and segment them from the background. The three-dimensional attitude and geometric shape estimation of the billet is achieved by using a point cloud registration algorithm to match the identified billet point cloud with the preset billet CAD model, thereby accurately estimating the three-dimensional position and attitude of the billet and detecting any possible deformation or misalignment. Furnace environment modeling and obstacle detection: Based on point cloud data acquired by 3D vision sensors, a real-time 3D environment model of the retrieval area is constructed, and the position and geometric information of furnace walls, adjacent steel billets, furnace equipment or other potential obstacles are detected in real time.

10. The PLC-based automated billet retrieval control system according to claim 4, characterized in that: The adaptive path planning subunit (23) receives the billet's three-dimensional posture, geometric shape, and environmental obstacle information from the three-dimensional visual perception and posture estimation subunit (22), and combines the current pose and kinematic model of the multi-degree-of-freedom retrieval execution module (3) to generate or update the optimal motion trajectory of the retrieval device in real time. The adaptive path planning subunit (23) adopts a sampling-based method or heuristic search algorithm to plan multiple collision-free trajectories from the current position to the target billet grab point and then to the placement point, under the premise of considering multiple optimization objectives such as the kinematic and dynamic constraints of the robotic arm, collision avoidance, motion smoothness, and time optimization or energy optimization. The path planning algorithm can perform rapid online replanning or trajectory adjustment based on real-time detected billet position deviations, attitude changes, or new environmental obstacles. When generating the retrieval trajectory, the adaptive path planning subunit (23) also considers the energy consumption optimization factors of the retrieval equipment. By introducing an energy consumption model into the path planning algorithm, it plans a motion trajectory with the minimum energy consumption under the premise of meeting time constraints and accuracy requirements.

Images