Intelligent repairing method and system for lower tank of RH vacuum chamber

Abstract

This invention discloses an intelligent repair method and system for the lower tank of an RH vacuum chamber, belonging to the field of intelligent maintenance technology for steel smelting equipment. The system includes a mobile support module, a multi-degree-of-freedom actuator, an execution end unit integrating a multimodal vision unit and a spray nozzle, a feeding system, and a central processing system. The method simultaneously acquires two-dimensional images, three-dimensional point clouds, and temperature data of the furnace lining and fuses them to construct a three-dimensional erosion model with a temperature field. Based on this model, it intelligently identifies and quantifies erosion defects; generates adaptive spraying commands related to erosion depth and temperature based on the quantification results and executes the operation; after spraying, it calculates the deposition thickness through a secondary scan and decides on further spraying, forming a closed-loop quality control. This invention achieves a fundamental shift in repair operations from manual experience-based decision-making to data-driven intelligent decision-making, solving the challenges of erosion quantification and adaptive control.

Description

Technical Field

[0001] This invention relates to the field of intelligent maintenance technology for steel smelting equipment, and in particular to an intelligent repair method and system for the lower tank of an RH vacuum chamber. Background Technology

[0002] The lining of the lower tank of the RH vacuum refining furnace is subject to wear and tear due to the erosion caused by high-temperature molten steel and slag, making regular repairs crucial. Currently, repair work relies primarily on manual experience to determine the erosion area and the amount of repair needed, which has two fundamental drawbacks: firstly, the working environment is highly dangerous, seriously threatening personnel safety; secondly, it cannot accurately quantify parameters such as the location, depth, and volume of erosion, leading to blind repair decisions, large quality fluctuations, and significant material waste. Existing mechanized repair devices only achieve long-distance material feeding and do not solve the core problem of "how to automatically and accurately quantify and identify erosion and make intelligent decisions based on this"; while standalone visual inspection devices typically only provide two-dimensional image information, lacking comprehensive data on three-dimensional morphology and temperature fields, and cannot support accurate quantitative repair decisions. Therefore, there is an urgent need in this field for a repair technology that can achieve unmanned operation and automatically complete the quantitative identification and intelligent decision-making of erosion. Summary of the Invention

[0003] The purpose of this invention is to overcome the shortcomings of the prior art and provide an intelligent repair method and system for the lower tank of an RH vacuum chamber, so as to solve the problems of existing technologies relying on manual labor, being unable to quantitatively identify corrosion and make intelligent decisions, and to realize the quantitative, automated and closed-loop quality control of repair operations.

[0004] To address the aforementioned problems, according to one aspect of this application, an embodiment of the present invention provides a smart repair method for the lower groove of an RH vacuum chamber, comprising the following steps: Two-dimensional images, three-dimensional point clouds, and temperature distribution data of the lower trough lining of the RH vacuum chamber are collected and fused to construct a three-dimensional erosion model with a temperature field. The three-dimensional erosion model with a temperature field is a seven-dimensional fused data model that includes spatial coordinates, color information, and temperature values. Based on the three-dimensional erosion model, the geometric parameters and temperature state of the erosion defects are identified and quantified by machine learning algorithms. The geometric parameters include at least erosion depth, erosion area, and erosion volume, and the temperature state is the real-time temperature and temperature gradient of each region of the furnace lining. An adaptive spraying path and process parameters are generated based on the geometric parameters and temperature conditions, driving the actuator and feeding system to perform the spraying operation. The process parameters include at least the spray gun moving speed and the feeding frequency. After spraying, three-dimensional data of the repaired area are collected again and the material deposition thickness is calculated. Based on the thickness assessment results, a decision is made on whether to perform additional spraying, forming a closed-loop quality control. The thickness assessment results include at least the average deposition thickness, thickness standard deviation, and thickness compliance rate.

[0005] In some implementations, the synchronous acquisition and fusion includes: The system controls a rigidly integrated high-temperature resistant industrial camera, a 3D laser scanner, and a long-wave infrared thermal imager to synchronously trigger data acquisition. Using a pre-calibrated sensor space transformation matrix, the two-dimensional image pixel coordinates, temperature pixel coordinates, and three-dimensional point cloud coordinates are unified into the same world coordinate system, with the base of the multi-degree-of-freedom actuator as the origin. The iterative nearest point algorithm is used to register multi-source data and generate a fused data volume containing coordinate, color and temperature information.

[0006] In some implementations, the identification and quantification includes: The two-dimensional image is input into a trained convolutional neural network to obtain a pixel-level erosion classification map. The three-dimensional point cloud is input into the trained point cloud processing network to calculate the normal erosion depth of each point relative to the standard model and generate a full-field erosion depth cloud map. The standard model is a three-dimensional CAD model established based on the design drawings of the lower tank of the RH vacuum chamber or a reference surface model fitted based on the uneroded area in the scan data. The classification map, erosion depth map, and temperature distribution map are spatially overlaid. Areas that simultaneously meet the criteria of being classified as severely eroded, having an erosion depth greater than a preset depth threshold, and having a temperature higher than a preset temperature threshold are marked as the highest priority repair areas. The preset depth threshold and preset temperature threshold are pre-calibrated based on the furnace lining material of the lower tank of the RH vacuum chamber and the actual smelting conditions.

[0007] In some implementations, the generation of adaptive spraying paths and process parameters includes: The spraying task is modeled as a decision-making process, and the agent is trained using reinforcement learning algorithms. The intelligent agent takes the erosion depth and wall temperature as state inputs and outputs the spray gun movement speed and feeding frequency. For areas where both erosion depth and temperature exceed a specific threshold, the agent outputs a lower spray gun movement speed and a higher feeding frequency compared to the base value, realizing an adaptive spraying strategy of "deep pit, slow speed, multiple materials, and high temperature adapted to the process".

[0008] In some embodiments, the calculation of material deposition thickness and closed-loop quality control includes: The 3D point cloud acquired after spraying is precisely registered with the reference point cloud before spraying, and the registration error is controlled within 0.5 mm. For each point in the reference point cloud, search for the nearest neighbor point in the post-spraying point cloud along its normal direction and calculate the Euclidean distance as the thickness of that point to generate a thickness distribution map of the repaired area. The average thickness, standard deviation of thickness, and thickness compliance rate of the entire repair area are statistically analyzed. The thickness compliance rate is the percentage of the area whose deposition thickness reaches or exceeds the preset thickness threshold to the total area of ​​the repair area. If any indicator fails to meet the preset threshold, a respraying instruction for the non-compliant area is automatically generated and executed until all indicators meet the preset compliance threshold.

[0009] This invention also provides an intelligent repair system for the lower tank of an RH vacuum chamber for implementing the method described in any of the preceding claims, comprising: The mobile carrier module includes a trolley, a platform lifting mechanism and a slag-blocking platform connected in sequence. The trolley is a rail-type trolley that can move along a preset track between different RH refining stations. The platform lifting mechanism can adjust the working height of the slag-blocking platform to match the lower tank inlet of the RH vacuum chamber. The multi-degree-of-freedom actuator, installed on the slag-blocking platform, is a serial robotic arm structure with multiple degrees of freedom of motion, including rotation, extension, and pitch. The execution module, installed at the end of the multi-degree-of-freedom actuator, is a high-temperature resistant sealed structure. The execution module integrates a spray nozzle, a multimodal vision unit, and a high-temperature protection component. The multimodal vision unit includes at least a rigidly integrated high-temperature resistant industrial camera, a long-wave infrared thermal imager, and a 3D laser scanner. The feeding system is connected to the spray nozzle via pipeline; The central processing system communicates and controls the mobile carrier module, multi-degree-of-freedom actuator, multimodal vision unit, and feeding system via industrial Ethernet. The central processing system is configured to: control the vision unit to collect data and fuse it to construct a three-dimensional erosion model; generate adaptive spraying instructions based on the model to drive the spraying operation; and evaluate the quality and make decisions on re-spraying based on the secondary scanning data after spraying, thereby achieving closed-loop control of the entire process.

[0010] In some embodiments, the multi-degree-of-freedom actuator includes a rotation mechanism, a high-precision hydraulic telescopic mechanism, and a pitch adjustment mechanism connected sequentially from the base to the end; the rotation mechanism enables a 360° wide-range rotation in the horizontal plane, covering the circumferential area of ​​the lower tank furnace lining of the RH vacuum chamber; the actuator module is installed at the end of the pitch adjustment mechanism; the multimodal vision unit also includes a high-temperature resistant industrial camera and a long-wave infrared thermal imager rigidly integrated into the actuator module.

[0011] In some embodiments, the central processing system is further configured to: perform spatial overlay analysis on the high-temperature anomaly area detected by the thermal imager and the geometric erosion area identified by the three-dimensional laser scanner to determine the repair priority; and based on the overlay analysis results, dynamically plan differentiated spray nozzle movement speed and feeding frequency of the feeding system for areas with different priorities.

[0012] In some implementations, the central processing system includes a data fusion module, an erosion identification module, a path planning module, a motion control module, and a quality feedback module connected in sequence. The quality feedback module calculates the thickness by comparing the three-dimensional data before and after spraying and makes a decision on spraying, and outputs a signal to the path planning module and the motion control module to form a closed-loop software logic.

[0013] In some embodiments, the high-precision hydraulic telescopic mechanism integrates a magnetostrictive displacement sensor; the front end of the actuation module is provided with a high-temperature protection component, including an annular nozzle for forming a cooling air curtain and a miniature rotating brush for cleaning the lens.

[0014] Compared with the prior art, the intelligent repair method and system for the lower groove of the RH vacuum chamber of the present invention has at least the following beneficial effects: Completely achieving unmanned operation ensures personnel safety; through multimodal data fusion and quantitative analysis, repair decisions are elevated from "qualitative" to "quantitative," greatly improving the scientific nature of decisions and the consistency of repair quality; the adaptive spraying strategy enables "on-demand spraying," significantly improving material utilization; the online closed-loop quality control mechanism ensures the verifiability and reliability of repair results; the fully automated process significantly shortens maintenance downtime, improving equipment operating rate and economic benefits.

[0015] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, the preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 A flowchart of the intelligent repair method for the lower tank of the RH vacuum chamber provided in this embodiment of the invention; Figure 2This is a schematic diagram of the overall structure of the intelligent repair system for the lower tank of the RH vacuum chamber provided in an embodiment of the present invention.

[0018] Explanation of reference numerals in the attached figures: 1. Execution module; 2. Pitch adjustment mechanism; 3. High-precision hydraulic telescopic mechanism; 4. Rotation mechanism; 5. Piping; 6. Motor; 7. Feeding system; 8. Trolley; 9. Platform lifting mechanism; 10. Slag blocking platform; 11. Lower tank of vacuum chamber; 12. High-temperature resistant industrial camera; 13. Thermal imager; 14. 3D laser scanner; 15. Nozzle; 16. Central processing system. Detailed Implementation

[0019] To further illustrate the technical means and effects adopted by the present invention to achieve the intended purpose, the specific embodiments, structures, features, and effects according to the present invention will be described in detail below with reference to the accompanying drawings and preferred embodiments. In the following description, different "embodiments" or "embodiments" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0020] In the description of this invention, it should be clearly stated that the terms "first," "second," etc., in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence; the terms "vertical," "lateral," "longitudinal," "front," "rear," "left," "right," "up," "down," "horizontal," etc., indicate orientation or positional relationships based on the orientation or positional relationships shown in the accompanying drawings, and are merely for the convenience of describing this invention, and do not mean that the device or element referred to must have a specific orientation or position, and therefore should not be construed as a limitation of this invention.

[0021] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections or electrical connections; they can refer to direct connections or indirect connections through an intermediate medium. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0022] like Figure 1-2 As shown, this embodiment of the invention provides a smart repair method for the lower tank of an RH vacuum chamber, including the following steps: Two-dimensional images, three-dimensional point clouds, and temperature distribution data of the lower trough lining of the RH vacuum chamber are collected and fused to construct a three-dimensional erosion model with a temperature field. The three-dimensional erosion model with a temperature field is a seven-dimensional fused data model that includes spatial coordinates, color information, and temperature values. Based on the three-dimensional erosion model, the geometric parameters and temperature state of the erosion defects are identified and quantified by machine learning algorithms. The geometric parameters include at least erosion depth, erosion area, and erosion volume, and the temperature state is the real-time temperature and temperature gradient of each region of the furnace lining. An adaptive spraying path and process parameters are generated based on the geometric parameters and temperature conditions, driving the actuator and feeding system 7 to perform the spraying operation. The process parameters include at least the spray gun moving speed and the feeding frequency. After spraying, three-dimensional data of the repaired area are collected again and the material deposition thickness is calculated. Based on the thickness assessment results, a decision is made on whether to perform additional spraying, forming a closed-loop quality control. The thickness assessment results include at least the average deposition thickness, thickness standard deviation, and thickness compliance rate.

[0023] The intelligent repair system and method provided by this invention aim to fundamentally change the traditional repair mode that relies on manual experience. Its core lies in constructing a closed-loop intelligent system of "perception-decision-execution-verification". The entire system begins with the precise quantitative perception of the furnace lining condition, simultaneously collecting multi-dimensional data through multiple integrated sensors and fusing it into a unified digital model. Based on this high-fidelity model, the system can automatically diagnose the severity of erosion defects and generate a matching personalized repair strategy. Finally, through immediate quality verification and feedback after execution, it ensures that each repair action achieves the predetermined goal, thereby realizing a fundamental shift from qualitative judgment to quantitative decision-making, and from open-loop operation to closed-loop control.

[0024] In this embodiment, the scheme defines the core process of intelligent repair. First, the system needs to acquire comprehensive status information of the furnace lining. This is achieved by simultaneously acquiring three key data: a two-dimensional surface image acquired by a high-temperature industrial camera 12, a high-precision three-dimensional contour point cloud acquired by a three-dimensional laser scanner 14, and a temperature distribution map acquired by a long-wave infrared thermal imager 13. The central processing system 16 performs spatiotemporal registration and fusion of these data from different physical dimensions and with different coordinate systems. For example, it accurately attaches the texture information and temperature values ​​of the image to the corresponding three-dimensional point cloud coordinates, thereby constructing a three-dimensional erosion model that simultaneously includes geometry, visual features, and temperature field. This model is the basis for all subsequent intelligent decisions.

[0025] Based on this fusion model, the system employs machine learning algorithms for analysis. The algorithm can automatically identify abnormal areas in the model, such as comparing point cloud data with a standard lining model to calculate the erosion depth of each point. Simultaneously, it can integrate image texture and temperature information to determine the type and severity of erosion. Finally, the system outputs a quantitative report, precisely identifying the location, depth, volume, area, and corresponding temperature value of each erosion defect.

[0026] Next, the decision-making system begins its work. It reads the aforementioned quantitative report and, based on its built-in process knowledge base or adaptive algorithm, generates a unique set of spraying instructions for each area to be repaired. These instructions are not fixed but adaptive. For example, for a deep, high-temperature erosion pit, the system automatically plans a slower spray gun movement speed and a higher feed rate to ensure sufficient repair material deposition and proper sintering. Conversely, for shallow wear, a faster speed and lower feed rate are used to improve efficiency. After generating the instructions, the system coordinates the multi-degree-of-freedom actuators and the feeding system 7 to drive the spray nozzle 15 along the planned path for precise spraying.

[0027] Finally, the system incorporates a closed-loop quality control mechanism. After a patching operation is completed, the system does not immediately stop but immediately restarts the 3D laser scanner 14 to scan the newly repaired area. By precisely comparing the point cloud data of the new surface after patching with the original baseline model before patching, the actual deposition thickness of the repair material at each point can be calculated. The system analyzes the thickness distribution of the entire area. If indicators such as average thickness and uniformity do not meet the preset quality standards, it automatically determines that the area needs further patching and immediately generates a new, more targeted patching instruction, initiating a new round of operations until the quality standards are met. This process ensures the verifiability and reliability of the repair results.

[0028] To ensure the recognition accuracy and decision-making reliability of the convolutional neural network, point cloud processing network, and reinforcement learning agent, the central processing system 16 of this invention pre-installs a model library built during the offline training phase. The training dataset of this model library comes from multiple sets of multimodal data of the lower trough furnace lining of the RH vacuum chamber collected under historical operating conditions, including two-dimensional images, three-dimensional point clouds, and corresponding temperature distributions under various states such as normal lining, minor cracks, severe erosion, and slag layer coverage.

[0029] During the data annotation phase, domain experts perform pixel-level annotation of the eroded areas in the images, and simultaneously register the 3D point cloud with a standard uneroded CAD model. The normal erosion depth of each point is calculated as the ground truth. The reinforcement learning agent is trained in a physics-based simulation environment that simulates the adhesion effect of sprayed material at different erosion depths and temperatures. Through extensive trial-and-error training, the agent learns the optimal strategy of "deep pit, slow speed, and multiple materials." This offline training process ensures that the model can directly output accurate recognition results and optimized process parameters during online operations.

[0030] In some implementations, the synchronous acquisition and fusion includes: The rigidly integrated high-temperature resistant industrial camera 12, 3D laser scanner 14 and long-wave infrared thermal imager 13 are controlled to synchronously trigger acquisition; Using a pre-calibrated sensor space transformation matrix, the two-dimensional image pixel coordinates, temperature pixel coordinates, and three-dimensional point cloud coordinates are unified into the same world coordinate system, with the base of the multi-degree-of-freedom actuator as the origin. The iterative nearest point algorithm is used to register multi-source data and generate a fused data volume containing coordinate, color and temperature information. The preset depth threshold and preset temperature threshold are pre-calibrated based on the furnace lining material of the lower tank of the RH vacuum chamber and the actual smelting conditions.

[0031] In this embodiment, the central processing system 16 synchronously controls the high-temperature resistant industrial camera 12, the 3D laser scanner 14, and the long-wave infrared thermal imager 13, which are rigidly integrated at the front end of the execution module 1, by sending a unified hardware trigger signal. This hardware-level synchronization ensures that the image, topography, and temperature data of the same physical point are captured within the same millisecond-level time window, eliminating data misalignment caused by device response time or mechanical movement.

[0032] At the spatial level, the system relies on a pre-calibrated sensor spatial transformation matrix. This matrix, obtained through a high-precision calibration process after system assembly, accurately describes the mathematical transformation relationships between the pixel coordinate system of the high-temperature industrial camera 12, the temperature pixel coordinate system of the thermal imager 13, and the point cloud 3D coordinate system of the 3D laser scanner 14. During data processing, the central processing system 16 uses this matrix to transform and map each pixel in each frame of the 2D image and each temperature pixel in the temperature distribution map to a unified world coordinate system (usually with the base of the execution module 1 as the origin).

[0033] After initial coordinate transformation, the system employs an iterative nearest-point algorithm for refined multi-source data registration. This algorithm automatically identifies and optimizes the correspondence between feature points from different data sources, performing fine-tuning to eliminate calibration and mechanical errors. Ultimately, a fused data volume is generated, where each data unit contains spatial coordinates (XYZ), color information (RGB, from the high-temperature industrial camera 12), and temperature value (T, from the thermal imager 13), forming a complete "RGB-XYZ-T" seven-dimensional data set. This provides a perfect data foundation for constructing a realistic and comprehensive three-dimensional erosion model.

[0034] To ensure accurate spatial fusion of multimodal data, this invention performs an offline joint calibration after the execution module 1 is assembled. The calibration process uses a specially designed three-dimensional calibration board. The surface of the calibration board has both a checkerboard pattern (for camera calibration) and geometrically distinct pits or protrusions (for lidar calibration), and a heatable constant temperature source (for infrared thermal imager calibration) is attached to the back.

[0035] By controlling the actuator, the execution module 1 observes the calibration board from multiple angles, while the central processing system 16 simultaneously acquires data from three sensors. The system detects checkerboard corner points in the image, geometric feature centers in the point cloud, and high-temperature point centers in the thermal image. Using a perspective N-point algorithm and least squares optimization, it solves for the optimal rigid body transformation matrices (rotation matrix R and translation vector t) relative to the coordinate system of the high-temperature industrial camera 12, the long-wave infrared thermal imager 13, and the 3D laser scanner 14. This calibration process controls the registration error of the three-source data to the sub-pixel level, laying the physical foundation for subsequent accurate erosion depth calculation and thickness assessment.

[0036] In some implementations, the identification and quantification includes: The two-dimensional image is input into a trained convolutional neural network to obtain a pixel-level erosion classification map. The three-dimensional point cloud is input into the trained point cloud processing network to calculate the normal erosion depth of each point relative to the standard model and generate a full-field erosion depth cloud map. The standard model is a three-dimensional CAD model established based on the design drawings of the lower tank of the RH vacuum chamber or a reference surface model fitted based on the uneroded area in the scan data. The classification map, erosion depth map, and temperature distribution map are spatially overlaid, and the areas that simultaneously meet the criteria of being classified as severely eroded, having an erosion depth greater than a preset depth threshold, and having a temperature higher than a preset temperature threshold are marked as the highest priority repair areas.

[0037] This embodiment details the specific algorithmic implementation path for intelligent recognition and quantization. First, at the two-dimensional image analysis level, the system inputs the acquired high-resolution images into a pre-trained convolutional neural network, such as the U-Net network. This network, trained on a large number of labeled furnace lining images, is capable of pixel-level semantic segmentation of the input images and automatically outputs a classification map. On this map, each pixel is classified into different categories such as "normal lining," "slight erosion," "severe erosion," and "slag layer coverage," thereby quickly identifying suspicious areas based on visual texture.

[0038] Secondly, at the 3D geometric analysis level, the system inputs the point cloud data acquired by the 3D laser scanner 14 into another deep learning network specifically designed for processing point sets, such as the PointNet++ network. This network can directly process unordered point cloud data and, through hierarchical feature learning, accurately calculate the normal distance of each point in the point cloud relative to the surface of a standard intact 3D furnace lining model. This distance value is the erosion depth of that point, and the system can generate a full-field erosion depth cloud map based on this, visually displaying where the pits are deep and where the wear is significant.

[0039] Finally, multi-dimensional information fusion is performed for decision-making. The central processing system 16 overlays and analyzes the pixel-level erosion classification map, the full-field erosion depth map, and the temperature distribution map provided by the long-wave infrared thermal imager 13 in a unified spatial coordinate system. The system sets a comprehensive decision-making logic: only when a spatial location simultaneously meets the three conditions of "classified as severely eroded by the convolutional neural network," "its calculated normal erosion depth is greater than 20 mm," and "the temperature at that point is higher than 600 degrees Celsius," is it marked as a high-priority repair area. This decision-making mechanism based on multi-dimensional evidence cross-validation greatly improves the accuracy and reliability of defect identification and effectively avoids erroneous decisions caused by false alarms from a single sensor (such as misjudging a shadow as a crack or a hot spot as erosion).

[0040] When the design drawings or CAD model of the lower tank of the RH vacuum chamber are unavailable, the standard model can be automatically generated by fitting the uneroded area in the current scanned point cloud. The specific implementation steps are as follows: First, the point cloud data of the entire field of view acquired by the 3D laser scanner 14 is initially segmented using a random sampling consensus algorithm to identify local regions that conform to planar or specific curvature characteristics. These regions are typically assumed to be the original furnace lining surface that has not been significantly eroded. Subsequently, a B-spline surface or non-uniform rational B-spline surface fitting algorithm is used to optimally approximate the point cloud of these selected uneroded regions, constructing a continuous and smooth reference surface model.

[0041] For points in the point cloud located in eroded areas, the system calculates the shortest distance (i.e., normal distance) from the point to the fitted reference surface as its erosion depth value. This method does not rely on external design data, enabling the system to adaptively establish a high-precision geometric comparison benchmark when facing RH vacuum chambers with different furnace ages and design specifications, thus ensuring the versatility and robustness of the erosion quantification algorithm.

[0042] The standard model is a three-dimensional computer-aided design model established based on the design drawings of the lower tank of the RH vacuum chamber, or a reference surface model fitted based on the un-eroded area in the current scan data.

[0043] In some implementations, the generation of adaptive spraying paths and process parameters includes: The spraying task is modeled as a decision-making process, and the agent is trained using reinforcement learning algorithms. The intelligent agent takes the erosion depth and wall temperature as state inputs and outputs the spray gun movement speed and feeding frequency. For areas where both erosion depth and temperature exceed a specific threshold, the agent outputs a lower spray gun movement speed and a higher feeding frequency compared to the base value, realizing an adaptive spraying strategy of "deep pit, slow speed, multiple materials, and high temperature adapted to the process".

[0044] In this embodiment, the system models the entire spraying operation as a sequential decision problem, namely a Markov decision process. The agent needs to make continuous decisions (adjusting spray gun speed and material feeding frequency) under complex environments (different erosion shapes, depths, and temperatures) to maximize long-term rewards (high repair quality, low material consumption, and short repair time). To this end, the system employs reinforcement learning algorithms such as deep deterministic policy gradients to train the agent extensively in a simulation environment.

[0045] The trained agent is embedded in the path planning module of the central processing system 16. During actual operation, for each point to be sprayed on the planned path, the agent uses the "erosion depth" and "wall temperature" of that point as state inputs. After calculation by the internal policy network, the agent directly outputs the optimal action for that state, namely the "moving speed" of the spray nozzle 15 at that point and the "feeding frequency" of the feeding system 7.

[0046] This mechanism endows the system with extremely high flexibility. For example, when the agent determines that the current point is in an extremely harsh area with an erosion depth greater than 30 mm and a wall temperature exceeding 800 degrees Celsius, it will output instructions to adjust the spray gun movement speed to 0.2 m / s and increase the feeding frequency to 80 Hz, based on the optimal strategy learned during training. For other areas with different depths and temperature combinations, the agent will output different speed and frequency combinations, achieving true "one-site-one-policy" solutions. This method surpasses systems based on fixed rules, better handling the complex and varied irregular erosion morphology and temperature gradients on the furnace lining surface, further optimizing overall operational efficiency and material consumption while ensuring repair quality.

[0047] In some embodiments, the calculation of material deposition thickness and closed-loop quality control includes: The 3D point cloud acquired after spraying is precisely registered with the reference point cloud before spraying, and the registration error is controlled within 0.5 mm. For each point in the reference point cloud, search for the nearest neighbor point in the post-spraying point cloud along its normal direction and calculate the Euclidean distance as the thickness of that point to generate a thickness distribution map of the repaired area. The average thickness, standard deviation of thickness, and thickness compliance rate of the entire repair area are statistically analyzed. The thickness compliance rate is the percentage of the area whose deposition thickness reaches or exceeds the preset thickness threshold to the total area of ​​the repair area. If any indicator fails to meet the preset threshold, a respraying instruction for the non-compliant area is automatically generated and executed until all indicators meet the preset compliance threshold.

[0048] In this embodiment, immediately after the spraying operation is completed, the system drives the 3D laser scanner 14 to perform a secondary scan of the repaired area to obtain the 3D point cloud data of the surface after spraying. The first step in quality assessment is high-precision registration: the central processing system 16 uses a sophisticated registration algorithm to strictly align this new point cloud with the original reference point cloud of the area before spraying, which is stored in the system.

[0049] After alignment, the system performs thickness calculations. The algorithm traverses each point in the original reference point cloud, searching for the nearest corresponding point in the new point cloud after spraying, along the surface normal direction of that point's location. The Euclidean distance between these two points is calculated; this distance value represents the deposition thickness of the repair material at that location. By processing all points, the system generates a thickness distribution map of the entire repair area.

[0050] Finally, quantitative assessment and decision-making are conducted. The system statistically analyzes several key indicators: the average deposition thickness of the entire area, reflecting whether the overall repair volume is sufficient; the standard deviation of the thickness distribution, reflecting the uniformity of the repair; and the thickness compliance rate, which is the percentage of the area within the repair region whose deposition thickness reaches or exceeds the preset thickness threshold, reflecting the coverage of qualified areas. The central processing system 16 presets the qualification thresholds for each indicator. If the calculation results show that any indicator fails to meet the standard, the system does not simply issue an alarm, but automatically analyzes the thickness distribution map, accurately locates those local sub-regions with insufficient thickness, and generates new, precisely defined respray paths and parameter instructions for these sub-regions. Subsequently, the system automatically coordinates the equipment and returns to execute the respray operation. This "assessment-decision-execution" cycle can be automatically repeated until all quality indicators meet the requirements, thus forming a powerful and fully automatic closed-loop quality control loop to ensure the final repair result is flawless.

[0051] When a sensor (such as thermal imager 13) temporarily fails, the central processing system 16 can automatically switch to degraded mode, making decisions based solely on the three-dimensional geometric erosion depth and two-dimensional image classification, and can still complete basic repair work.

[0052] This invention provides an intelligent repair system for the lower tank of an RH vacuum chamber for implementing the method described in any of the preceding claims, comprising: The mobile carrier module includes a trolley 8, a platform lifting mechanism 9 and a slag-blocking platform 10 connected in sequence. The trolley 8 is a rail-type trolley that can move along a preset track between different RH refining stations. The platform lifting mechanism 9 can adjust the working height of the slag-blocking platform 10 to match the inlet of the lower tank of the RH vacuum chamber. The multi-degree-of-freedom actuator, installed on the slag-blocking platform 10, is a serial robotic arm structure with multiple degrees of freedom of motion, including rotation, extension, and pitch. The execution module, installed at the end of the multi-degree-of-freedom actuator, is a high-temperature resistant sealed structure. The execution module integrates a spray nozzle 15, a multimodal vision unit, and a high-temperature protection component. The multimodal vision unit includes at least a rigidly integrated high-temperature resistant industrial camera, a long-wave infrared thermal imager, and a three-dimensional laser scanner 14. The feeding system 7 is connected to the spray nozzle 15 via the pipeline 5; The central processing system 16 communicates and controls the mobile carrier module, multi-degree-of-freedom actuator, multimodal vision unit and feeding system 7 via industrial Ethernet. The central processing system 16 is configured to: control the vision unit to collect data and fuse it to construct a three-dimensional erosion model; generate adaptive spraying instructions based on the model to drive the spraying operation; and evaluate the quality and make spraying decisions based on the secondary scanning data after spraying, thereby achieving closed-loop control of the entire process.

[0053] In this embodiment, the entire system is built on a mobile support module, which includes a trolley 8 that can travel on a track, a platform lifting mechanism 9 mounted on the trolley 8, and a slag-blocking platform 10 connected to the top of the lifting mechanism. The trolley 8 is responsible for transferring between different RH refining stations, and the platform lifting mechanism 9 is responsible for raising and lowering the upper mechanism to a working height that matches the inlet of the lower tank 11 of the vacuum chamber.

[0054] A multi-degree-of-freedom actuator is mounted on the slag-blocking platform 10, providing flexible movement capabilities for the end effector. An integrated actuator module 1 is mounted at the end of this actuator, compactly integrating a multimodal vision unit (including at least a 3D laser scanner 14) and a spray nozzle 15. The feeding system 7 is independently mounted on the trolley 8 and connected to the remote spray nozzle 15 via high-temperature and wear-resistant spray pipes 5, achieving stable material transport.

[0055] The central processing system 16 is the control hub of the entire system. It first coordinates the vision unit to complete data acquisition and fusion modeling, providing the system with its "eyes." Next, it analyzes the model and generates adaptive spraying commands, acting as the "brain." Then, it precisely controls the actuators and feeding system 7 to coordinate their actions, driving the spraying nozzles 15 to perform the work, equivalent to the "nerves and limbs." Finally, it restarts the vision unit for quality verification and decides whether to initiate re-spraying based on the results, completing the "perception-action-verification" closed loop. This tight integration of hardware and software logic enables unmanned, intelligent quantitative repair.

[0056] In some embodiments, the multi-degree-of-freedom actuator includes a rotating mechanism 4, a high-precision hydraulic telescopic mechanism 3, and a pitch adjustment mechanism 2 connected sequentially from the base to the end; the rotating mechanism enables a 360° wide-range rotation in the horizontal plane, covering the circumferential area of ​​the lower tank furnace lining of the RH vacuum chamber; the actuator module 1 is installed at the end of the pitch adjustment mechanism 2; the multimodal vision unit also includes a high-temperature resistant industrial camera 12 and a long-wave infrared thermal imager 13 rigidly integrated into the actuator module 1.

[0057] In this embodiment, the multi-degree-of-freedom actuator adopts a tandem robotic arm structure, starting from the fixed slag-blocking platform 10, and sequentially connecting the rotating mechanism 4, the high-precision hydraulic telescopic mechanism 3, and the pitch adjustment mechanism 2. The rotating mechanism 4 achieves a large range of rotation in the horizontal plane, covering the circumference of the furnace lining; the high-precision hydraulic telescopic mechanism 3 provides radial linear telescopic motion, controlling the distance between the spray nozzle 15 and the furnace lining; the pitch adjustment mechanism 2 performs final attitude fine-tuning to ensure that the nozzle is always aligned with the curved furnace lining at the optimal angle. The actuator module 1 is fixedly installed at the end of the pitch adjustment mechanism 2 and moves with it.

[0058] In terms of the multimodal vision unit, in addition to the core 3D laser scanner 14, the execution module 1 also rigidly integrates a high-temperature resistant industrial camera 12 and a long-wave infrared thermal imager 13. "Rigid integration" means that these sensors are securely mounted within a rigid frame, and their relative positions are determined and remain unchanged after installation through calibration. This design is crucial, ensuring the consistency of the observation perspectives of different sensors and providing a physical foundation for the spatial synchronous fusion of multi-source data, avoiding registration errors caused by individual sensor shaking or displacement.

[0059] In some embodiments, the central processing system 16 is further configured to: perform spatial overlay analysis on the high-temperature abnormal area detected by the thermal imager 13 and the geometric erosion area identified by the three-dimensional laser scanner 14 to determine the repair priority; and based on the overlay analysis results, dynamically plan differentiated spray nozzle 15 movement speed and feeding frequency of the feeding system 7 for areas with different priorities.

[0060] In this embodiment, two key functions of the central processing system 16 in intelligent decision-making and adaptive control are defined. The first is the function of determining repair priorities. The system does not treat all identified defects equally, but rather performs intelligent classification. It independently analyzes temperature data from the long-wave infrared thermal imager 13 to delineate "high-temperature anomaly areas"; simultaneously, it independently analyzes geometric data from the 3D laser scanner 14 to delineate "geometric erosion areas." Subsequently, the system performs spatial overlay analysis on these two area layers. Those areas that overlap in space, i.e., areas that are both high-temperature and severely eroded, are determined to be the areas with the most dangerous lining condition and the most prone to perforation accidents, and are therefore assigned the highest repair priority. This priority determination method based on cross-validation of multi-source information is more scientific and reliable than relying solely on geometric shape or temperature information.

[0061] The second feature is dynamic parameter planning. Based on the aforementioned priority division and the specific quantitative parameters of each erosion point, the central processing system 16 dynamically adjusts when generating spraying instructions. Its core rule is: for areas with greater depth or higher temperature, the system automatically plans a lower movement speed for the spray nozzle 15 and may plan a higher feeding frequency for the material feeding system 7. For example, for the highest priority overlapping area, the system plans a movement speed that is 30% slower than in ordinary areas, and a feeding frequency that is 50% higher. This differentiated dynamic planning ensures that more material is deposited and adheres well at the most critical working conditions where repair material is needed, thus achieving optimal allocation of repair resources.

[0062] In some embodiments, the central processing system 16 includes a data fusion module, an erosion identification module, a path planning module, a motion control module, and a quality feedback module connected in sequence. The quality feedback module calculates the thickness by comparing the three-dimensional data before and after spraying and makes a decision on spraying, and outputs a signal to the path planning module and the motion control module to form a closed-loop software logic.

[0063] In this embodiment, the dedicated software of the central processing system 16 is structured into multiple sequentially connected functional modules. The data fusion module is responsible for receiving raw sensor data, synchronizing, registering, and fusing it, and outputting a unified multidimensional model. The erosion identification module receives this model, calls the algorithm to identify and quantify defects, and outputs a diagnostic report. The path planning module reads the diagnostic report, combines it with process rules or intelligent algorithms, and generates the optimal spraying path and parameter sequence. The motion control module solves the path planning into specific motion commands for each actuator and coordinates the feeding system 7 to execute them in conjunction.

[0064] The quality feedback module is the key component of the closed loop. It receives data from a rescan after the initial spraying, calculates the repair thickness by comparison, and performs a quality assessment. If the assessment is satisfactory, the process ends; if it fails, the quality feedback module generates a respraying decision signal. This signal is not output to the user but is directly fed back to the path planning and motion control modules. The path planning module generates a new respraying path for the substandard area based on this signal; the motion control module then drives the mechanism to perform the respraying. In this way, the data flow forms a complete loop within the software, with each module connected sequentially. This allows the entire process of "detection-planning-execution-verification-replanning" to proceed automatically and cyclically without manual intervention, achieving true software closed-loop self-control.

[0065] In some embodiments, the high-precision hydraulic telescopic mechanism 3 integrates a magnetostrictive displacement sensor; the front end of the execution module 1 is provided with a high-temperature protection component, including an annular nozzle for forming a cooling air curtain and a miniature rotating brush for cleaning the lens.

[0066] In this embodiment, the high-precision hydraulic telescopic mechanism 3 is not an ordinary hydraulic cylinder; it integrates a magnetostrictive displacement sensor. This sensor employs a non-contact measurement principle, enabling real-time, high-resolution detection of the absolute position of the piston rod and feeding the signal back to the control system. This allows the mechanism to achieve extremely high repeatability, such as ±0.1 mm, thereby ensuring precise control of the distance between the spray nozzle 15 and the furnace lining surface, which is fundamental to achieving uniform spraying.

[0067] The front end of execution module 1 is equipped with a high-temperature protection component, mainly consisting of an annular air curtain nozzle and a miniature rotating cleaning brush. Inside the high-temperature, dusty RH vacuum chamber, the annular air curtain nozzle continuously sprays cooling gas (such as compressed air or nitrogen) in front of the lenses of the high-temperature resistant industrial camera 12, thermal imager 13, and 3D laser scanner 14, forming a dynamic isolation air curtain. This air curtain effectively blocks high-temperature heat radiation, preventing overheating of the lenses and sensors, while simultaneously blowing away most of the floating dust. The miniature rotating cleaning brush is located inside the air curtain nozzle and can be driven by a miniature motor to rotate during operation intervals, mechanically removing stubborn stains that may adhere to the lens protective lens. The combination of these two components actively protects the precision optical sensors, ensuring their long-term stable operation and data acquisition quality under harsh conditions.

[0068] The following are some specific examples: The intelligent repair system of this invention is installed on a movable track trolley 8. A platform lifting mechanism 9 is installed on the trolley 8, and a slag-blocking platform 10 is provided on its top. A multi-degree-of-freedom actuator is installed on the slag-blocking platform 10, which includes, from the base to the end, a rotation mechanism 4, a high-precision hydraulic telescopic mechanism 3, and a pitch adjustment mechanism 2. The actuator module 1 is fixedly installed at the end of the pitch adjustment mechanism 2.

[0069] The execution module 1 is a high-temperature resistant sealed structure, with a multimodal vision unit and a spray nozzle 15 rigidly integrated at its front end. The multimodal vision unit includes: a high-temperature resistant, high-resolution industrial camera 12 (such as a 5-megapixel global shutter camera) for acquiring surface texture, a long-wave infrared thermal imager 13 (such as a 640x480 pixel uncooled type) for measuring temperature distribution, and a 3D laser scanner 14 (based on the laser line scanning principle) for capturing 3D topography. The spatial relative positions of the three components are precisely optically calibrated to obtain a unified spatial transformation matrix.

[0070] The feeding system 7 (such as a dual-cylinder alternating dry powder conveying pump) is mounted on the trolley 8 and connected to the spray nozzle 15 on the execution module 1 via a high-temperature resistant spraying pipe 5. The central processing system 16 (an industrial computer equipped with a high-performance GPU) is integrated in the control cabinet and communicates and controls all sensors, actuator drivers, and the feeding system 7 via an industrial Ethernet.

[0071] A specific embodiment of the intelligent repair method of the present invention is as follows: Step S1: System positioning and terminal placement.

[0072] The central processing system 16 controls the control vehicle 8 to move directly below the lower slot 11 of the target vacuum chamber, and raises the slag-blocking platform 10 to the working height via the platform lifting mechanism 9. Subsequently, it coordinates the control of the rotation mechanism 4, the high-precision hydraulic telescopic mechanism 3, and the pitch adjustment mechanism 2 to smoothly send the execution module 1 into the lower slot 11 of the vacuum chamber and to the predetermined scanning starting point.

[0073] Step S2: Synchronous acquisition and fusion modeling of multimodal data.

[0074] The central processing system 16 plans a scanning path (such as a spiral path) covering the inner wall. While driving the actuators along the path, it sends unified hardware trigger signals to the high-temperature industrial camera 12, thermal imager 13, and 3D laser scanner 14 to achieve millisecond-level time-synchronized data acquisition. The acquired raw data is transmitted to the central processing system 16 in real time. The data fusion module uses a pre-calibrated sensor transformation matrix to map the two-dimensional image pixels and temperature pixels at each moment to a three-dimensional point cloud coordinate system. It then uses the Iterative Closest Point (ICP) algorithm for fine registration, ultimately generating an "RGB-T-XYZ" fusion model where each three-dimensional point is associated with both color (RGB) and temperature (T) information—a three-dimensional erosion model with a temperature field.

[0075] Step S3: Intelligent erosion identification and adaptive spraying planning.

[0076] The erosion identification module analyzes the fusion model: a) Geometric analysis: Compare the current 3D geometric model with the standard un-eroded CAD model, calculate the normal offset distance of each point on the surface, identify the "geometric erosion zone" with a depression depth greater than 5mm, and calculate the area, depth and volume of each region.

[0077] b) Thermal analysis: Analyze the temperature data and mark areas with temperatures more than 150°C higher than the average temperature of the scanned area as "high temperature anomaly areas".

[0078] c) Intelligent Decision Making: The "Geometric Erosion Zone" and the "High Temperature Anomaly Zone" are spatially superimposed. The overlapping area indicates severe furnace lining thinning and abnormal temperature, and is classified as the highest priority "Class A Repair Zone". The system outputs a quantitative diagnostic report containing all defect types, three-dimensional coordinates, dimensions, and temperature.

[0079] The path planning module adaptively plans based on the diagnostic report: generating a zigzag spraying path covering the surface of each repair area. A base spray gun speed V0 (e.g., 0.5 m / s) and a base feed frequency F0 (e.g., 50 Hz) are set. The planning algorithm traverses each point on the path, querying its geometric depth (D) and temperature (T) in the fusion model. For example, the rule is set: if a point is located in a Class A area, its parameters are dynamically adjusted to (0.7V0, 1.5F0); if the point temperature T > 800℃, the speed is further reduced by 20% based on its existing parameters. This achieves a personalized spraying strategy of "deep pits, slow speed, multiple materials, and high-temperature process adjustment."

[0080] Step S4: Precise and coordinated spraying execution.

[0081] The motion control module converts the optimized path and parameter sequence into coordinated motion commands for each actuator. The central processing system 16 starts the feeding system 7 and controls the multi-degree-of-freedom actuators to drive the spray nozzle 15 to move strictly according to the plan, performing fully automatic spraying operations.

[0082] Step S5: Online quality assessment and closed-loop feedback.

[0083] After spraying is completed in a sub-region, the system immediately controls the 3D laser scanner 14 to perform a rapid secondary scan of that area. The quality feedback module performs an online evaluation. a) Thickness calculation: Perform high-precision ICP registration between the sprayed point cloud and the reference point cloud before spraying. For each point in the reference point cloud, search for the nearest neighbor point in the sprayed point cloud along its normal direction, and calculate the Euclidean distance between the two points as the deposition thickness at that point.

[0084] b) Quality analysis: Statistically analyze the thickness of all points in the area, and calculate the average thickness (H_avg), thickness standard deviation (H_std), and thickness compliance rate (P, i.e., the percentage of points with a thickness ≥15mm).

[0085] c) Closed-loop decision-making: The preset acceptable thresholds are H_avg≥20mm, H_std≤4mm, and P≥95%. If all evaluation results meet the standards, the system is deemed acceptable. If any one of the standards is not met, the system automatically analyzes the thickness distribution map, extracts the non-compliant areas, generates a new respray instruction for them, and then returns to step S4 to perform targeted respray. This "scan-evaluate-respray" cycle can be performed multiple times until the quality fully meets the standards, forming a strong closed-loop quality control.

[0086] Step S6: Job completion and data archiving.

[0087] After all areas have been repaired and deemed satisfactory, the system control actuator retracts, the platform descends, and the trolley returns to a safe position. All data throughout the process (scan data, models, diagnostic reports, process parameters, quality reports) is automatically stored and archived for production traceability and process optimization.

[0088] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0089] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for intelligent repair of the lower groove of an RH vacuum chamber, characterized in that, Includes the following steps: Two-dimensional images, three-dimensional point clouds, and temperature distribution data of the lower trough lining of the RH vacuum chamber are collected and fused to construct a three-dimensional erosion model with a temperature field. The three-dimensional erosion model with a temperature field is a seven-dimensional fused data model that includes spatial coordinates, color information, and temperature values. Based on the three-dimensional erosion model, the geometric parameters and temperature state of the erosion defects are identified and quantified by machine learning algorithms. The geometric parameters include at least erosion depth, erosion area, and erosion volume, and the temperature state is the real-time temperature and temperature gradient of each region of the furnace lining. Based on the geometric parameters and temperature conditions, an adaptive spraying path and process parameters are generated to drive the actuator and feeding system (7) to perform the spraying operation. The process parameters include at least the spray gun moving speed and the feeding frequency. After spraying, three-dimensional data of the repaired area are collected again and the material deposition thickness is calculated. Based on the thickness assessment results, a decision is made on whether to perform additional spraying, forming a closed-loop quality control. The thickness assessment results include at least the average deposition thickness, thickness standard deviation, and thickness compliance rate.

2. The intelligent repair method for the lower groove of the RH vacuum chamber according to claim 1, characterized in that, The synchronous acquisition and fusion includes: The rigidly integrated high-temperature resistant industrial camera (12), three-dimensional laser scanner (14) and long-wave infrared thermal imager (13) are controlled to perform synchronous trigger acquisition; Using a pre-calibrated sensor space transformation matrix, the two-dimensional image pixel coordinates, temperature pixel coordinates, and three-dimensional point cloud coordinates are unified into the same world coordinate system, with the base of the multi-degree-of-freedom actuator as the origin. The iterative nearest point algorithm is used to register multi-source data and generate a fused data volume containing coordinate, color and temperature information.

3. The intelligent repair method for the lower groove of the RH vacuum chamber according to claim 2, characterized in that, The identification and quantification include: The two-dimensional image is input into a trained convolutional neural network to obtain a pixel-level erosion classification map. The three-dimensional point cloud is input into the trained point cloud processing network to calculate the normal erosion depth of each point relative to the standard model and generate a full-field erosion depth cloud map. The standard model is a three-dimensional CAD model established based on the design drawings of the lower tank of the RH vacuum chamber or a reference surface model fitted based on the uneroded area in the scan data. The classification map, erosion depth map, and temperature distribution map are spatially overlaid. Areas that simultaneously meet the criteria of being classified as severely eroded, having an erosion depth greater than a preset depth threshold, and having a temperature higher than a preset temperature threshold are marked as the highest priority repair areas. The preset depth threshold and preset temperature threshold are pre-calibrated based on the furnace lining material of the lower tank of the RH vacuum chamber and the actual smelting conditions.

4. The intelligent repair method for the lower groove of the RH vacuum chamber according to claim 1, characterized in that, The generated adaptive spraying path and process parameters include: The spraying task is modeled as a decision-making process, and the agent is trained using reinforcement learning algorithms. The intelligent agent takes the erosion depth and wall temperature as state inputs and outputs the spray gun movement speed and feeding frequency. For areas where both erosion depth and temperature exceed a specific threshold, the agent outputs a lower spray gun movement speed and a higher feeding frequency compared to the base value, realizing an adaptive spraying strategy of "deep pit, slow speed, multiple materials, and high temperature-adaptive process".

5. The intelligent repair method for the lower groove of the RH vacuum chamber according to claim 1, characterized in that, The calculation of material deposition thickness and closed-loop quality control includes: The 3D point cloud acquired after spraying is precisely registered with the reference point cloud before spraying, and the registration error is controlled within 0.5mm. For each point in the reference point cloud, search for the nearest neighbor point in the post-spraying point cloud along its normal direction and calculate the Euclidean distance as the thickness of that point to generate a thickness distribution map of the repaired area. The average thickness, standard deviation of thickness, and thickness compliance rate of the entire repair area are statistically analyzed. The thickness compliance rate is the percentage of the area whose deposition thickness reaches or exceeds the preset thickness threshold to the total area of ​​the repair area. If any indicator fails to meet the preset threshold, a respraying instruction for the non-compliant area is automatically generated and executed until all indicators meet the preset compliance threshold.

6. An intelligent repair system for the lower RH vacuum chamber tank, used to implement the intelligent repair method for the lower RH vacuum chamber tank according to any one of claims 1-5, characterized in that, include: The mobile carrier module includes a trolley (8), a platform lifting mechanism (9) and a slag-blocking platform (10) connected in sequence. The trolley (8) is a rail-type trolley that can move along a preset track between different RH refining stations. The platform lifting mechanism (9) can adjust the working height of the slag-blocking platform (10) to match the lower tank inlet of the RH vacuum chamber. The multi-degree-of-freedom actuator is installed on the slag-blocking platform (10) and is a serial robotic arm structure with multiple degrees of freedom of motion, including rotation, extension, and pitch. The execution module (1) is installed at the end of the multi-degree-of-freedom actuator and is a high-temperature resistant sealed structure. The execution module integrates a spray nozzle (15), a multimodal vision unit and a high-temperature protection component. The multimodal vision unit includes at least a rigidly integrated high-temperature resistant industrial camera, a long-wave infrared thermal imager and a three-dimensional laser scanner (14). The feeding system (7) is connected to the spray nozzle (15) via the pipeline (5); The central processing system (16) communicates with the mobile carrier module, multi-degree-of-freedom actuator, multi-modal vision unit and feeding system (7) via industrial Ethernet and realizes centralized control; The central processing system (16) is configured to: control the vision unit to collect data and fuse it to construct a three-dimensional erosion model, generate adaptive spraying instructions based on the model to drive the spraying operation, and evaluate the quality and make a decision on the spraying based on the secondary scanning data after spraying, so as to realize the closed-loop control of the whole process.

7. The intelligent repair system for the lower tank of the RH vacuum chamber according to claim 6, characterized in that, The multi-degree-of-freedom actuator includes a rotating mechanism (4), a high-precision hydraulic telescopic mechanism (3), and a pitch adjustment mechanism (2) connected sequentially from the base to the end. The rotating mechanism enables a 360° wide-range rotation in the horizontal plane, covering the circumferential area of ​​the lower tank furnace lining of the RH vacuum chamber. The actuator module (1) is installed at the end of the pitch adjustment mechanism (2). The multimodal vision unit also includes a high-temperature resistant industrial camera (12) and a long-wave infrared thermal imager (13) rigidly integrated into the actuator module (1).

8. The intelligent repair system for the lower tank of the RH vacuum chamber according to claim 7, characterized in that, The central processing system (16) is further configured to: perform spatial overlay analysis on the high temperature abnormal area detected by the thermal imager (13) and the geometric erosion area identified by the three-dimensional laser scanner (14) to determine the repair priority; and dynamically plan the movement speed of the spray nozzle (15) and the feeding frequency of the feeding system (7) for areas with different priorities based on the overlay analysis results.

9. The intelligent repair system for the lower tank of the RH vacuum chamber according to claim 6, characterized in that, The central processing system (16) includes a data fusion module, an erosion identification module, a path planning module, a motion control module and a quality feedback module connected in sequence. The quality feedback module calculates the thickness by comparing the three-dimensional data before and after spraying and makes a decision on spraying. It outputs a signal to the path planning module and the motion control module to form a closed-loop software logic.

10. The intelligent repair system for the lower groove of the RH vacuum chamber according to claim 7, characterized in that, The high-precision hydraulic telescopic mechanism (3) integrates a magnetostrictive displacement sensor; the front end of the execution module (1) is provided with a high-temperature protection component, including an annular nozzle for forming a cooling air curtain and a miniature rotating brush for cleaning the lens.

Images