Semiconductor device high-precision positioning and welding device
By combining laser vision and infrared thermal imaging into a sensing layer, microscopic defects are identified and thermal deformation is predicted, solving the problems of microscopic gaps and thermal deformation in semiconductor equipment welding and achieving high-precision and reliable welding process control.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- WELL TECH ELECTRONIC TECH CHANGZHOU CO LTD
- Filing Date
- 2026-03-10
- Publication Date
- 2026-05-26
AI Technical Summary
Existing semiconductor manufacturing equipment cannot effectively identify micro-gaps and dynamic thermal deformation during the welding process, resulting in unstable welding quality and affecting accuracy and reliability.
A perception layer combining laser vision and infrared thermal imaging is used to identify microscopic defects and predict thermal deformation through a thermal excitation module. Combined with robots and fixtures, dynamic compensation is performed to build a closed-loop control system for the entire process.
It achieves highly sensitive detection and precise positioning of micro-defects, dynamically tracks thermal deformation during welding, improves welding accuracy and reliability, and is suitable for complex environments.
Smart Images
Figure CN121798150B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of precision welding technology, and more specifically to a high-precision positioning welding device for semiconductor equipment. Background Technology
[0002] The structural framework of semiconductor manufacturing equipment is the mechanical and thermal foundation of the entire equipment, and its welding quality is directly related to the precision and stability of the equipment during long-term operation.
[0003] Precision welding in existing semiconductor manufacturing equipment mostly employs laser welding. After assembly, the mating surfaces of the workpieces may have microscopic gaps (typically less than 0.05 mm) or incomplete fits that are difficult to detect with the naked eye or even laser scanning. These "invisible" defects can cause heat flow distortion and protective atmosphere disturbance during welding, easily leading to internal defects such as porosity and lack of fusion. Traditional visual positioning based on laser triangulation cannot effectively identify such non-geometric defects, creating hidden dangers for welding quality. Although laser welding is a low-heat-input process, under conditions of long welds, multiple passes, or constraint, the accumulated thermal stress can still cause complex, dynamic, transient thermal deformation of the workpiece. This deformation changes the relative position of the welding torch and the weld in real time during the welding process. However, traditional robotic welding paths are preset and rigid, unable to follow this dynamic change, resulting in centering deviations in the latter half of the weld bead, seriously affecting the consistency and overall dimensional accuracy of long welds.
[0004] In summary, there is an urgent need for a high-precision positioning and welding device for semiconductor equipment that can achieve precise positioning of the thermal field before welding and predictive compensation for thermal deformation during welding, so as to fundamentally improve the welding accuracy and reliability of semiconductor equipment structural components. Summary of the Invention
[0005] The purpose of this invention is to provide a high-precision positioning and welding device for semiconductor equipment, so as to solve the technical problems of the inability to precisely position micro-defects and dynamically compensate for thermal deformation.
[0006] To address the aforementioned technical problems, this invention provides a high-precision positioning and welding device for semiconductor equipment, comprising an execution layer, a sensing layer, and a decision control layer.
[0007] The execution layer includes:
[0008] The first robot serves as the loading and rough adjustment unit, used to transport workpieces to the welding station.
[0009] Workpiece positioning stage: Equipped with rigid clamps to support and ultimately fix the workpiece, and can perform fine adjustments.
[0010] The second robot, serving as a high-precision welding execution unit, is mounted on a mobile base to expand the workspace, with a laser welding head installed at its end.
[0011] The perception layer includes:
[0012] Integrated sensing module: Mounted on the laser welding head, it includes a laser structured light vision sensor and an infrared thermal imaging sensor. The former is used for weld geometry tracking, and the latter is used for thermal field monitoring throughout the process.
[0013] Independent thermal excitation module: This is a focused laser source, a high-frequency induction heating coil, or a hot air gun. It can be installed on the first robot, the workpiece positioning table, or the second robot to apply controllable, localized thermal excitation before welding to stimulate the thermal response of defects.
[0014] The decision control layer is an intelligent monitoring unit that communicates with all hardware. Its core innovation lies in executing a phased, closed-loop intelligent workflow:
[0015] Pre-welding positioning stage: Laser vision is used for initial workpiece positioning and macroscopic weld positioning. For critical areas, a combined "thermal excitation-infrared imaging" detection is initiated. By analyzing the evolution of the surface thermal field under thermal excitation (e.g., calculating the heat transfer coefficient distribution map), areas of abnormal heat flow caused by microscopic gaps are identified, achieving "visualized" defect positioning. Geometric and thermal field positioning data are fused to drive the first robot and workpiece positioning stage to perform final pose compensation and rigid fixation of the workpiece.
[0016] Pre-welding planning stage: Based on the precisely positioned workpiece model, an initial welding path is planned. Thermal deformation parameter calibration is performed by acquiring the actual temperature rise-deformation response of the workpiece through low-power preheating scanning. Key parameters of the thermal deformation prediction model are calibrated online, significantly improving the model's prediction accuracy for the current workpiece. The thermal deformation prediction model is then run to predict potential deformations throughout the welding process, and the welding path is proactively optimized accordingly, generating a "preferred welding path" that can preemptively offset some deformations.
[0017] During the welding process: Real-time dual-loop compensation for geometry and thermal field is implemented. Based on laser vision, the weld seam is tracked in real-time, generating high-frequency path compensation commands. The temperature field is inverted in real-time based on infrared thermal imaging, and a calibrated thermal deformation prediction model is input to generate low-frequency thermal deformation feedforward compensation commands. High-frequency geometric compensation and low-frequency thermal deformation compensation are vector-superimposed at the motion command level to drive the second robot and mobile base for comprehensive pose compensation, achieving dynamic following. Real-time analysis of the molten pool's infrared image characteristics (such as temperature, symmetry, and tail isotherm morphology) enables determination of the penetration state and adaptive adjustment of process parameters.
[0018] Post-weld optimization stage: Record data throughout the entire process to form a traceable "welding digital twin". Compare actual measurement results with model predictions to automatically update and optimize the thermal deformation prediction model and process parameter library, enabling the system to learn and continuously improve itself.
[0019] The beneficial effects of this invention are: it provides a dual-mode pre-welding positioning technology that integrates laser vision and active thermally excited infrared imaging, achieving high-sensitivity detection and precise positioning of micro-assembly defects; it constructs an online thermal deformation prediction model based on a real-time infrared temperature field to guide the correction of rigid fixtures and robot motion feedforward compensation for thermal deformation, maintaining dynamic alignment accuracy throughout the welding process; and it establishes a full-process data closed-loop and intelligent decision-making system covering pre-welding planning, in-welding execution, and post-welding optimization, particularly through the full mining of infrared thermal field data to achieve adaptive optimization of process parameters and prediction models. Infrared imaging is not easily affected by electromagnetic and strong light interference, the system operates stably and reliably, and it is suitable for complex environments such as strong light, darkness, smoke, and dust. Attached Figure Description
[0020] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:
[0021] Figure 1 This is a schematic diagram of the structure of a high-precision positioning and welding device for semiconductor equipment according to an embodiment of the present invention;
[0022] Figure 2 This is a partially enlarged view of the first robot and the thermal excitation module according to an embodiment of the present invention;
[0023] Figure 3 This is a partial enlarged view of the second robot, laser welding head, and sensing module according to an embodiment of the present invention;
[0024] Figure 4 This is a flowchart illustrating the specific workflow of a high-precision positioning and welding device for semiconductor equipment according to an embodiment of the present invention.
[0025] In the picture:
[0026] 1. The First Robot;
[0027] 2. Workpiece positioning table;
[0028] 3. The second robot;
[0029] 4. Laser welding head;
[0030] 5. Sensing module; 51. Laser structured light vision sensor; 52. Infrared thermal imaging sensor.
[0031] 6. Thermal excitation module. Detailed Implementation
[0032] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments implemented by those skilled in the art without creative effort are within the protection scope of the present invention.
[0033] Example
[0034] This embodiment provides a high-precision positioning and welding device for semiconductor equipment, such as... Figure 1 , 2 As shown in Figure 3, the hardware configuration includes:
[0035] First Robot 1: Used to transport workpieces to the welding station; a six-axis medium-load articulated robot can be selected, equipped with an adaptive gripper at the end for handling workpieces; for large workpieces, a large three-axis mobile base can also be equipped for First Robot 1 to increase the working range.
[0036] Workpiece positioning stage 2: used to support and fix the workpiece; multiple high-rigidity, micron-level adjustable pneumatic-hydraulic hybrid fixtures can be arranged on it.
[0037] The second robot 3 and the mobile base: are used to provide the motion degrees of freedom of the end effector; a high-precision (e.g., repeatability ±0.02mm) six-axis articulated robot can be selected; it can be further mounted on a mobile base equipped with a high-rigidity motor slide, thereby expanding to a seven-axis to nine-axis system.
[0038] Laser welding head 4: Installed at the end of the second robot 3, used to output welding laser beam;
[0039] Sensing module 5: Mounted on the laser welding head 4, it includes a laser structured light vision sensor 51 and an infrared thermal imaging sensor 52. The laser structured light vision sensor 51 projects a structured light pattern and acquires reflected light images to obtain three-dimensional geometric information of the workpiece and weld. The infrared thermal imaging sensor 52 acquires infrared radiation images to obtain thermal field distribution information. The laser structured light vision sensor 51 (which can be a line laser, cross laser, or grid laser) can be used to track fillet welds and analyze macroscopic three-dimensional information. The infrared thermal imaging sensor 52 is a high-speed mid-wave infrared camera, which is aimed at the molten pool and the area behind it at a certain tilt angle during welding.
[0040] The thermal excitation module 6 is used to apply controllable thermal excitation to a specific area of the workpiece. It can be an independent three-axis gantry motion mechanism with an adjustable-power semiconductor laser installed at its end as a focused active thermal excitation source. Optionally, the thermal excitation module 6 can be one of a focused laser source, a high-frequency induction heating coil, or a hot air gun, and its excitation power, application time, and application area can be programmably controlled. The thermal excitation module 6 can be integrated and installed on the first robot 1, the workpiece positioning stage 2, or the second robot 3. For example, the workpiece positioning stage 2 can be equipped with an array of high-frequency induction heating coils to thermally excite the workpiece at the corresponding position.
[0041] The intelligent monitoring unit is communicatively connected to the first robot 1, the workpiece positioning stage 2, the second robot 3, the laser welding head 4, the sensing module 5, and the thermal excitation module 6. The intelligent monitoring unit is a high-performance industrial server that runs customized control software based on a real-time operating system.
[0042] like Figure 4 As shown, the specific workflow is as follows:
[0043] Pre-welding positioning stage A includes:
[0044] Initial positioning and loading A1: The laser structured light vision sensor 51 is used to acquire the laser positioning data of the workpiece, and the first robot 1 is controlled to transport the workpiece to the welding station on the workpiece positioning table 2. The first robot 1 grabs the workpiece, and the intelligent monitoring unit controls the sensing module 5 (only the laser vision 71 is used at this time) to quickly scan several reference features on the workpiece to obtain rough pose data, and guides the first robot 1 to place the component in the approximate position of the workpiece positioning table 2.
[0045] Laser positioning of weld seam A2: Laser positioning data of weld seam is obtained using the laser structured light vision sensor 51; the second robot 3 carries the sensing module 5 and moves along the theoretical weld seam path, and the laser structured light vision sensor 51 performs high-precision three-dimensional scanning of all fillet weld seams to be welded to obtain macroscopic and continuous weld seam centerline point cloud data.
[0046] Infrared thermal field positioning A3: For areas where laser scanning identifies doubtful areas or preset key mating surfaces, the thermal excitation module 6 is activated to apply local thermal excitation, and the infrared thermal imaging sensor 52 is used simultaneously to monitor the surface thermal field evolution of the area, analyze and identify areas with abnormal heat flow, and obtain infrared thermal field positioning data of the workpiece or micro-defects. Taking fillet welds as an example, for the root mating surfaces of all fillet welds (which are prone to micro-gaps), a special detection program is activated; the focused laser beam of the thermal excitation module 6 performs a low-speed scan (e.g., power 50W, spot size 1mm) along the path on one side of the mating surface to form a thermal excitation zone; at the same time, the infrared thermal imaging sensor 52 simultaneously observes this area. When the thermal excitation sweeps over a defect-free area, the heat is uniformly conducted, and the infrared image shows a smooth temperature gradient; when encountering a micro-gaps, the heat flow is blocked, causing the area on the back side of the gap to heat up slowly or even not at all, forming a sharp contrast with the high-temperature area on the excitation side, producing a clear temperature gradient abrupt change line on the image (corresponding to the gap position). Taking spliced welds as an example, for mating surfaces with extremely high precision, laser scanning struggles to distinguish weld locations. Similarly, thermal excitation combined with infrared thermography is used. When encountering a seam, heat flow is blocked, causing the area behind the seam to heat up slowly or not at all, creating a stark contrast with the high-temperature area on the excitation side. This produces a clear temperature gradient abrupt change line on the image (corresponding to the seam location). The intelligent monitoring unit analyzes the image sequence, automatically identifying and labeling all such abnormal areas, and quantifying their location and approximate gap. Specifically, by analyzing the evolution of the surface thermal field under thermal excitation, the heat transfer coefficient distribution map or temperature gradient distribution map of the inspected area can be calculated. Areas with abrupt changes in heat transfer coefficient or temperature gradient are identified as defect locations with micro-gaps or abrupt changes in material properties, thereby obtaining infrared thermal field location data for the workpiece or micro-defects.
[0047] Workpiece pose compensation A4: By fusing laser positioning data and infrared thermal field positioning data, a workpiece pose compensation command is generated to drive the first robot 1 and the workpiece positioning stage 2 to perform pose compensation and fixation on the workpiece. For areas with out-of-tolerance gaps, the system first attempts to generate a command to drive specific fixtures on the workpiece positioning stage 2 to apply local pressure to eliminate or reduce the gaps. After pressure is applied, a rapid thermal excitation scan can be performed again for confirmation. Finally, all fixtures are locked, and the workpiece is rigidly fixed in its optimal pose.
[0048] Pre-welding planning phase B includes:
[0049] Initial Welding Path Planning B1: By fusing laser positioning data and infrared thermal field positioning data, an actual weld path model and a workpiece initial state model are generated, and the initial welding path is planned. Based on the final locked, precise 3D model of the workpiece (including weld geometry and known micro-gap information), the initial welding path and basic process parameters (including power, speed, wire feed, etc.) of the second robot 3 are planned.
[0050] Thermal Deformation Parameter Calibration B2: The thermal excitation module 6 is activated to preheat the workpiece with low power, and the infrared thermal imaging sensor 52 is used simultaneously to scan and obtain deformation response data with temperature rise, calibrating the preset thermal deformation parameters. This is a key step to improve prediction accuracy. The system controls the thermal excitation module 6 to perform rapid scanning preheating along a representative weld seam at extremely low power (e.g., 10% of the welding power). At the same time, the infrared thermal imaging sensor 52 records the temperature rise process of the workpiece surface at a high frame rate, and simultaneously measures the minute thermal deformation of specific features of the workpiece through the laser structured light vision sensor 51 or the infrared thermal imaging sensor 52. The laser structured light vision sensor 51 can use boundaries such as outer contours and holes as specific features, while the infrared thermal imaging sensor 52 can use boundaries, reduced gaps, or micro-defects as specific features. The intelligent monitoring unit 9 uses the temperature rise process data of the workpiece surface and the thermal deformation data of specific characteristics of the workpiece to invert and calibrate the key material parameters (such as the coefficient of thermal expansion) and boundary conditions (such as the equivalent stiffness of the fixture) in the thermal deformation prediction model, so that the model is more in line with the actual physical characteristics of the current workpiece-fixture system.
[0051] Pre-welding thermal deformation prediction B3: Based on the initial welding path and preset welding process parameters, temperature field evolution prediction data is obtained. A thermal deformation prediction model is established based on the modal superposition method, and the displacement prediction values of key feature points of the workpiece are output. Using the calibrated thermal deformation prediction model, the initial path and process parameters are input to perform a full-sequence simulation of the entire welding process, predicting the final displacement of key feature points of the frame (such as the four corner points and specific sealing surfaces) after welding.
[0052] Optimal Welding Path Planning (B4): Based on displacement prediction values, the welding path is replanned to obtain an optimal welding path where the thermal deformation of key feature points on the workpiece meets the preset range. If the predicted displacement of key feature points exceeds the tolerance, the system will automatically iteratively optimize the welding path and sequence. For example, by adjusting the welding direction, introducing reverse pre-deformation, or using a segmented skip welding strategy, a new optimal welding path is generated, so that under the same process, the predicted final deformation is reduced to meet the preset range. This path will be used as the reference path for actual welding.
[0053] Welding process stage C includes:
[0054] Online weld path compensation C1: Receives and processes image data from the laser structured light vision sensor 51, continuously calculates weld position deviation, and generates online weld path compensation commands. The laser structured light vision sensor 51 operates continuously, providing high-frequency weld position deviation data to generate high-frequency geometric compensation commands.
[0055] Online thermal deformation compensation C2: Receives and processes image data from the infrared thermal imaging sensor 52, extracts temperature field data, and uses the thermal deformation prediction model to predict the real-time thermal deformation of the workpiece, generating online thermal deformation compensation commands. The infrared thermal imaging sensor 52 provides the temperature field at a low frame rate. The calibrated thermal deformation prediction model operates as a state observer, receiving temperature field data in real time and predicting the amount of local elastic deformation that will be caused by the current heat input within a very short time (e.g., within hundreds of milliseconds), thereby generating low-frequency thermal compensation commands.
[0056] Specifically, the thermal deformation prediction model can be a machine learning model trained based on finite element simulation data and historical welding experimental data. The machine learning model can be a neural network model or a Gaussian process regression model. The image data of the infrared thermal imaging sensor 52 is received and processed to extract temperature gradient features including the molten pool area, the heat-affected zone and the far-end substrate. The temperature gradient features are input into the thermal deformation prediction model to predict the local deformation of the workpiece caused by the non-uniform thermal field at the current and future times. The deformation includes at least the warping displacement perpendicular to the weld direction.
[0057] Another alternative is that the thermal deformation prediction model is a finite element analysis subroutine; a dynamic heat source moving coordinate system is established with the weld point as the origin; the temperature field inverted from the infrared image is used as the boundary condition and input into a real-time finite element analysis subroutine running in the intelligent monitoring unit; based on the principles of elasticity and thermodynamics, the temperature field is mapped into angular deformation and lateral shrinkage compensation amounts that cause changes in the relative position of the weld point and the workpiece; after each calculation cycle, the subroutine outputs the predicted displacement of the preset monitoring point on the workpiece in the next short time domain, which is used to generate feedforward compensation instructions.
[0058] Comprehensive online pose compensation C3: The thermal deformation compensation command and the weld path tracking compensation command are fused to generate a comprehensive compensation command, which drives the second robot 3 to perform comprehensive pose compensation. The intelligent monitoring unit 9 performs bandpass filtering and coordinate transformation on the high-frequency geometric compensation command and the low-frequency thermal compensation command, and then superimposes them to obtain the comprehensive compensation command. This command is sent to the controllers of the second robot 3 and the moving base in real time. Among them, the large-scale, slowly changing overall deformation component is compensated by the moving base; the high-frequency, small-scale local deformation and tracking error are compensated by the joint motion of the second robot 3 itself. When the laser structured light vision sensor 51 is interfered with by the strong light of the welding laser beam or welding fume, the infrared thermal imaging sensor 52 can temporarily take over without affecting the overall online pose compensation.
[0059] Furthermore, the online compensation command for the weld path is set as the main high-frequency compensation quantity, and its response frequency is matched with the motion control frequency of the second robot 3; the online compensation command for thermal deformation is set as the auxiliary low-frequency compensation quantity, and its response frequency is related to the physical time constant of thermal accumulation and is lower than the response frequency of the high-frequency compensation quantity; the high-frequency compensation quantity and the low-frequency compensation quantity are vector-superimposed at the motion command level of the second robot 3.
[0060] Furthermore, a rigid fixture is provided on the workpiece positioning stage 2; the second robot 3 is mounted on a movable base, which is used to provide three-dimensional linear motion for the second robot 3; the overall deformation component of the workpiece in the online thermal deformation compensation command is corrected for thermal deformation by clamping with the rigid fixture; the corrected thermal deformation is compensated by controlling the movement of the movable base.
[0061] Online thermal monitoring of process quality C4: The infrared thermal imaging sensor 52 is used to monitor the state of the molten pool. When the molten pool state data exceeds the preset process parameter threshold, the process parameters are adjusted and the associated weld position information is recorded.
[0062] Specifically, based on the data from the infrared thermal imaging sensor 52, the characteristic state parameters of the molten pool are calculated in real time. These characteristic state parameters include at least one of the following: maximum temperature, average temperature, temperature distribution uniformity, and thermal field symmetry. The characteristic state parameters of the molten pool are compared with a preset process quality threshold. When the parameters continuously deviate from the threshold range, process parameter adjustment is triggered and the associated weld position information is recorded. After the process parameters are adjusted, the characteristic state parameters of the molten pool are compared with the preset process quality threshold. If the parameters still deviate from the threshold range, an alarm is triggered and the associated weld position information is recorded.
[0063] Furthermore, special attention is paid to the shape and radius of curvature of the isotherm at the tail of the molten pool (this radius of curvature is strongly correlated with the penetration depth), and it is used as an indirect indicator of the penetration state. When the shape of the isotherm changes abruptly, it is determined that the welding process may have incomplete penetration or burn-through defects, and process parameter adjustments or alarms are immediately triggered. When the system detects an abnormal decrease in the molten pool temperature or radius of curvature (indicating possible incomplete penetration), it will automatically fine-tune and increase the laser power; when it abnormally increases (indicating possible burn-through), it will fine-tune and decrease the power or increase the speed. All adjustments and corresponding position information are recorded.
[0064] Post-weld optimization stage D includes:
[0065] Welding process data recorder D1: Records sensor data, pose data and compensation command data throughout the entire process, forming a traceable welding database.
[0066] D2 optimization of the thermal deformation prediction model: The actual deformation measurement data after welding is compared with the model's predicted data to update and optimize the thermal deformation prediction model. After welding, the workpiece is sent to a coordinate measuring machine for final dimensional inspection. The measured actual deformation data is transmitted back to the intelligent monitoring unit and compared with the final predicted value of the thermal deformation prediction model. The deviation data is used to automatically update the thermal deformation prediction model's database, optimizing its internal parameters to make its predictions more accurate for similar workpieces in the future.
[0067] Process Parameter Optimization D3: This function compares post-weld quality data with molten pool state data and process parameter adjustment data during the welding process to update and optimize preset process parameters. All successful process parameter adjustments during the welding process are also recorded in the process parameter library to enrich the system's process knowledge.
[0068] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A high-precision positioning and welding device for semiconductor equipment, characterized in that, include: The first robot (1) is used to transport the workpiece to the welding station; The workpiece positioning stage (2) is used to support and fix the workpiece; The second robot (3) is used to provide the motion degrees of freedom of the execution end; The laser welding head (4) installed at the end of the second robot (3) is used to output a welding laser beam; The sensing module (5) installed on the laser welding head (4) includes a laser structured light vision sensor (51) and an infrared thermal imaging sensor (52); the laser structured light vision sensor (51) is used to project a structured light pattern and acquire reflected light images to obtain three-dimensional geometric information of the workpiece and the weld; the infrared thermal imaging sensor (52) is used to acquire infrared radiation images to obtain thermal field distribution information. The thermal excitation module (6) is used to apply controllable thermal excitation to a specific area of the workpiece; The intelligent monitoring unit, which is communicatively connected to the first robot (1), the workpiece positioning stage (2), the second robot (3), the laser welding head (4), the sensing module (5), and the thermal excitation module (6), is configured as follows: Pre-welding positioning stage (A) includes: Initial positioning and loading (A1): The laser structured light vision sensor (51) is used to obtain the laser positioning data of the workpiece, and the first robot (1) is controlled to transport the workpiece to the welding station on the workpiece positioning table (2); Laser positioning of weld seam (A2): Laser positioning data of weld seam is obtained using the laser structured light vision sensor (51); Infrared thermal field positioning (A3): For areas where laser scanning identifies doubtful areas or preset key mating surfaces, the thermal excitation module (6) is activated to apply local thermal excitation, and the infrared thermal imaging sensor (52) is used simultaneously to monitor the surface thermal field evolution of the area, analyze and identify areas with abnormal heat flow, and obtain infrared thermal field positioning data of the workpiece or micro-defects. Workpiece pose compensation (A4): By fusing laser positioning data and infrared thermal field positioning data, a workpiece pose compensation command is generated to drive the first robot (1) and the workpiece positioning stage (2) to perform pose compensation and fixation on the workpiece. The pre-welding planning phase (B) includes: Initial welding path planning (B1): Integrate laser positioning data and infrared thermal field positioning data to generate an actual weld path model and a workpiece initial state model, and plan the initial welding path; Thermal deformation parameter calibration (B2): The thermal excitation module (6) is activated to preheat the workpiece with low power, and the infrared thermal imaging sensor (52) is used to scan simultaneously to obtain deformation response data with temperature rise, and the preset thermal deformation parameters are calibrated. Pre-welding thermal deformation prediction (B3): Based on the initial welding path and preset welding process parameters, obtain temperature field evolution prediction data, establish a thermal deformation prediction model based on the modal superposition method, and output the displacement prediction values of key feature points of the workpiece. Preferred welding path planning (B4): Based on the displacement prediction value, the welding path is replanned to obtain the preferred welding path where the thermal deformation of the key feature points of the workpiece meets the preset range. Welding process stage (C) includes: Online weld path compensation (C1): Receives and processes the image data of the laser structured light vision sensor (51), continuously calculates the weld position deviation, and generates an online weld path compensation command; Online thermal deformation compensation (C2): Receive and process the image data of the infrared thermal imaging sensor (52), extract the temperature field data, and use the thermal deformation prediction model to predict the real-time thermal deformation of the workpiece and generate an online thermal deformation compensation command. Comprehensive pose online compensation (C3): The thermal deformation compensation command and the weld path tracking compensation command are fused together to generate a comprehensive compensation command, which drives the second robot (3) to perform comprehensive pose compensation; Online thermal monitoring of process quality (C4): The infrared thermal imaging sensor (52) is used to monitor the state of the molten pool. When the molten pool state data exceeds the preset process parameter threshold, the process parameter adjustment is triggered, and the associated weld position information is recorded. Post-weld optimization phase (D) includes: Welding process data recording (D1): Records sensor data, pose data and compensation command data throughout the entire process to form a traceable welding database; Optimization of the thermal deformation prediction model (D2): The actual deformation measurement data after welding is compared with the model prediction data to update and optimize the thermal deformation prediction model; Process parameter optimization (D3): The welding quality data after welding is compared with the molten pool state data and process parameter adjustment data during the welding process, in order to update and optimize the preset process parameters.
2. The high-precision positioning and welding device for semiconductor equipment according to claim 1, characterized in that, The thermal excitation module (6) is one of a focused laser source, a high-frequency induction heating coil, or a hot air gun, and its excitation power, action time, and action area can be programmed and controlled. The thermal excitation module (6) can be integrated and installed on the first robot (1), the workpiece positioning stage (2), or the second robot (3).
3. The high-precision positioning and welding device for semiconductor equipment according to claim 2, characterized in that, The infrared thermal field positioning (A3) specifically includes: By analyzing the evolution of the surface thermal field under thermal excitation, the heat transfer coefficient distribution map or temperature gradient distribution map of the inspected area is calculated. The abrupt change area of heat transfer coefficient or temperature gradient is identified as the location of defect with micro gaps or abrupt changes in material properties, thereby obtaining infrared thermal field positioning data of the workpiece or micro defect.
4. The high-precision positioning and welding device for semiconductor equipment according to claim 3, characterized in that, The heat distortion parameter calibration (B2) specifically includes: The thermal excitation module (6) is activated to preheat the workpiece with a preset power, and the infrared thermal imaging sensor (52) is used to perform continuous scanning simultaneously. By analyzing the thermal field data at different times, the temperature rise rate distribution on the workpiece surface is obtained; Analyze the data of the same feature location at different times to obtain the deformation of the corresponding feature; By coupling the temperature rise rate distribution and the corresponding characteristic deformation, deformation response data with temperature rise is obtained, and preset thermal deformation parameters are calibrated.
5. The high-precision positioning and welding device for semiconductor equipment according to claim 4, characterized in that, The online thermal deformation compensation (C2) specifically includes: Receive and process the image data from the infrared thermal imaging sensor (52), and extract the temperature gradient features including the molten pool region, the heat-affected zone, and the distal substrate; The temperature gradient features are input into the thermal deformation prediction model to predict the local deformation of the workpiece caused by the non-uniform thermal field at the current and future times. The deformation includes at least the warping displacement perpendicular to the weld direction. The thermal deformation prediction model is a machine learning model trained based on finite element simulation data and historical welding experimental data. The machine learning model is a neural network model or a Gaussian process regression model.
6. The high-precision positioning and welding device for semiconductor equipment according to claim 5, characterized in that, The online thermal deformation compensation (C2) specifically includes: Establish a dynamic heat source movement coordinate system with the solder joint as the origin; The temperature field retrieved from the infrared image is used as the boundary condition and input into a real-time finite element analysis subroutine running in the intelligent monitoring unit. Based on the principles of elasticity and thermodynamics, the temperature field is mapped to the angular deformation and lateral shrinkage compensation amount that cause changes in the relative position of the weld point and the workpiece. Each time this subroutine completes a calculation cycle, it outputs the predicted displacement of the preset monitoring point on the workpiece within a short time domain in the future, which is used to generate feedforward compensation instructions.
7. The high-precision positioning and welding device for semiconductor equipment according to claim 6, characterized in that, The comprehensive pose online compensation (C3) specifically includes: The online compensation command for the weld path is set as the main high-frequency compensation quantity, and its response frequency is matched with the motion control frequency of the second robot (3). The online thermal deformation compensation command is set as an auxiliary low-frequency compensation amount, whose response frequency is related to the physical time constant of thermal accumulation and is lower than the response frequency of the high-frequency compensation amount. The high-frequency compensation amount and the low-frequency compensation amount are vector-superimposed at the motion command level of the second robot (3).
8. The high-precision positioning and welding device for semiconductor equipment according to claim 7, characterized in that, A rigid clamp is provided on the workpiece positioning table (2); The second robot (3) is mounted on a mobile base, which provides three-dimensional linear motion for the second robot (3); The overall workpiece deformation component in the online thermal deformation compensation command is corrected for thermal deformation by clamping with a rigid fixture. The corrected thermal deformation is compensated for by controlling the movement of the movable base.
9. The high-precision positioning and welding device for semiconductor equipment according to claim 8, characterized in that, The online thermal monitoring of process quality (C4) specifically includes: Based on the data from the infrared thermal imaging sensor (52), the characteristic state parameters of the molten pool are calculated in real time. The characteristic state parameters of the molten pool include at least one of the following parameters: maximum temperature, average temperature, temperature distribution uniformity, and thermal field symmetry. The molten pool characteristic state parameters are compared with preset process quality thresholds. When the parameters continuously deviate from the threshold range, process parameter adjustments are triggered and the associated weld position information is recorded. After the process parameters are adjusted, the characteristic state parameters of the molten pool are compared with the preset process quality threshold. If they still deviate from the threshold range, an alarm is triggered and the associated weld position information is recorded.
10. The high-precision positioning and welding device for semiconductor equipment according to claim 9, characterized in that, The online thermal monitoring of process quality (C4) also includes: The shape and curvature of the isotherm at the tail of the molten pool are identified and used as an indirect indicator of the penetration state. When the shape of the isotherm changes abruptly, it is determined that the welding process may have defects such as incomplete penetration or burn-through, and process parameter adjustment or alarm is triggered immediately.