A production line high-performance connection process adaptive control system and method

Abstract

The application discloses a production line high-performance connection process adaptive control system and method, and belongs to the field of automobile manufacturing and advanced material connection technology. The system integrates a 3D vision sensor, a load sensor and an absolute value encoder on a servo-driven blank holder to obtain the image, point cloud and force-displacement data of a region to be riveted; a calculation center generates a pure geometric feature vector according to the image and the point cloud, and generates a real dynamic stiffness response according to the force-displacement; the pure geometric feature vector is used as an independent observation to separate the geometric and material contributions by inversion, and an abnormality index of the working condition is constructed according to the above, and the vision and mechanical compensation are weighted and fused to output a target blank holding force; a servo execution controller differentiates the control of the whole riveting stroke based on the index and the target blank holding force. The application solves the problems of connection quality dispersion and production line stop caused by parameter solidification and incoming material fluctuation, and realizes adaptive regulation of riveting parameters and improvement of joint quality consistency.

Description

Technical Field

[0001] This invention relates to the field of automotive manufacturing and advanced materials bonding technology, and in particular to an adaptive control system and method for high-performance bonding processes on a production line. Background Technology

[0002] Currently, the self-piercing riveting (SPR) process in automotive body-in-white production lines generally adopts an open-loop or semi-open-loop control mode of "fixed blank holder force + fixed stroke". Process engineers calibrate a set of nominal parameters based on the first piece trial production, which are directly reused in the same batch of production. The riveting station itself does not have the ability to identify and compensate for fluctuations in incoming materials piece by piece. With the widespread use of high-strength steel, aluminum alloys, and composite materials in hybrid car bodies, upstream stamping, drawing, and trimming processes cause geometric defects such as wavy edges, springback gaps, and corner collapses on sheet metal edges, and the wear degree of different batches of incoming materials and molds varies significantly. At the same time, the random variation in the degree of work hardening of materials further exacerbates the uncertainty of riveting conditions.

[0003] In existing production lines, force sensors and displacement encoders are only used for post-riveting NOK / OK quality assessment. They cannot actively acquire the true fit of the panels before riveting, nor can they distinguish the contribution of geometric and material factors to stiffness within a millisecond window. Some solutions introduce visual inspection, but this is limited to post-riveting visual inspection or pre-riveting geometric shape extraction. It cannot penetrate the surface to obtain the internal hardening state of the material, thus leading to directional misjudgments of stiffness differences caused by "small gaps" and "material pre-hardening." This results in improper adjustment of the blank holder force or stroke, causing quality defects such as gaps between panels, surface indentations, and rivet penetration. In addition, facing the working conditions of multiple materials alternating in mixed-body construction, the production line needs to maintain hundreds of parameter formulas, resulting in long changeover and debugging times and limited overall line cycle time.

[0004] In summary, existing SPR control technology lacks the ability to actively perceive, decouple, and make adaptive decisions across the entire geometry-material state of incoming materials. There is an urgent need for a closed-loop control system and method that can actively perceive the deformation and hardening degree of the sheet metal before riveting, achieve geometry-material decoupling, and adaptively adjust the blank holder force and servo parameters accordingly. Summary of the Invention

[0005] The present invention aims to at least partially solve one of the technical problems in the related art.

[0006] To address this, this invention proposes a high-performance adaptive control system for riveting processes on production lines. It integrates a 3D vision sensor, a load sensor, and an absolute encoder to construct a multimodal perception architecture. A computational hub generates pure geometric feature vectors from visual data and a true dynamic stiffness response from force-displacement data. The pure geometric feature vectors are treated as independent observations for inversion calculation to separate geometric and material hardening contributions. Based on these two factors, a condition anomaly index is constructed. This index is then used to weightedly fuse visual and mechanical compensation components to output the target blank holder force. The servo execution controller receives the condition anomaly index and the target blank holder force and divides the entire riveting stroke into multiple stages for segmented differentiated control. This invention achieves physical decoupling and adaptive compensation of geometric and material factors during self-piercing riveting, solving the problems of inconsistent connection quality and production line downtime caused by parameter fixation and material fluctuations in existing technologies.

[0007] Another objective of this invention is to propose an adaptive control method for high-performance connection processes in production lines.

[0008] To achieve the above objectives, the present invention provides a high-performance adaptive control system for connection processes in production lines, comprising: Servo-driven pressure ring; The sensing components include a 3D vision sensor mounted on the front end of the servo-driven pressure ring, and a load sensor and an absolute encoder mounted on the servo-driven pressure ring. The computing center is communicatively connected to the sensing components; A servo execution controller is connected to both the computing center and the servo-driven pressure ring control. The 3D vision sensor acquires images and point cloud data of the area to be riveted and sends them to the computing center. The computing center generates a pure geometric feature vector based on the image and point cloud data, generates a real dynamic stiffness response based on the force-displacement data collected by the load sensor and the absolute encoder during the detection pulse loading process, performs inversion calculation on the pure geometric feature vector as an independent observation value to separate the geometric contribution and the material hardening contribution, constructs a working condition anomaly index based on the pure geometric feature vector and the real dynamic stiffness response, and outputs the target blank holder force by weighted fusion of visual compensation component and mechanical compensation component based on the working condition anomaly index. The servo actuator receives the working condition anomaly index and the target blank holder force, and divides the entire riveting stroke into multiple stages to perform segmented control: in the piercing stage, the piercing speed is given based on the working condition anomaly index and the blank holder force is locked with the target blank holder force; in the upsetting stage, a pressure holding pause is selectively inserted according to the working condition anomaly index; and in the pressure holding stage, the pressure holding duration is given based on the working condition anomaly index.

[0009] An embodiment of the high-performance connection process adaptive control system for a production line according to an invention may also have the following additional technical features: In one embodiment of the present invention, the computing center generates a pure geometric feature vector based on the image and point cloud data, including: the target detection network built into the computing center receives two-dimensional image data as input, and outputs the coordinates of the region of interest (ROI) defining the riveting area after target detection inference; the computing center performs spatial cropping and registration of the three-dimensional point cloud data based on the ROI coordinates to generate an elevation map representing the local surface morphology as an intermediate output; the elevation map is input into a built-in feature regression network for forward inference, and the feature regression network outputs a pure geometric feature vector. Pure geometric eigenvectors Each component is used to quantify the wavy edge height, collapse depth, and out-of-plane warping of the plate in the current riveting area.

[0010] In one embodiment of the present invention, the computational center generates a true dynamic stiffness response based on the force-displacement data, comprising: a stiffness extraction module built into the computational center receiving force-displacement time-domain response data synchronously acquired by the load sensor and the absolute encoder throughout the entire process of the detection pulse loading as input; the stiffness extraction module sequentially performing bandpass filtering and linear regression fitting on the force-displacement time-domain response data to extract the slope of the pulse loading segment as the system-level measured stiffness. Output: The stiffness extraction module queries a pre-calibrated equipment stiffness mapping table based on the spatial coordinates of the current riveting point to obtain the equipment stiffness at the corresponding riveting point. And based on the following formula for series stiffness decoupling, the system-level measurement stiffness is subjected to equipment compliance deduction:

[0011] Output the true dynamic stiffness response of the plate side .

[0012] In one embodiment of the present invention, the computational hub performs inversion calculations using pure geometric feature vectors as independent observations to separate geometric contributions and material hardening contributions, including: the decoupling and fusion module built into the computational hub receives pure geometric feature vectors respectively. and true dynamic stiffness response As a dual-path input; the decoupling fusion module will use the pure geometric feature vector As independent observations of the geometric components, they are substituted into the pre-constructed composite response model for inversion solution:

[0013] The composite response model will reflect the true dynamic stiffness response. Characterized as geometric contribution components Contribution to material hardening The superposition of the components; after inversion and solution, the decoupling and fusion module outputs the separated geometric contribution components. and the contribution of material hardening .

[0014] In one embodiment of the present invention, the computational hub constructs a condition anomaly index based on pure geometric feature vectors and the actual dynamic stiffness response, including: the decoupling and fusion module built into the computational hub receives the pure geometric feature vectors. and true dynamic stiffness response As input, the decoupling and fusion module normalizes the pure geometric feature vectors to obtain the geometric normalization ratio. The normalized stiffness ratio is obtained by normalizing the offset of the actual dynamic stiffness response relative to the nominal reference stiffness. The decoupling and fusion module calculates and outputs the operating condition anomaly index according to the following formula. :

[0015] in, The preset allowable deformation threshold, The nominal reference stiffness of the plate is given. The preset nonlinear sensitivity index, To meet The preset weighting coefficients.

[0016] In one embodiment of the present invention, the calculation center outputs the target edge clamping force by weighted fusion of visual compensation components and mechanical compensation components based on the working condition anomaly index, including: the decoupling fusion module built into the calculation center receiving the working condition anomaly index. As input, the decoupling and fusion module performs a nonlinear transformation on the working condition anomaly index using the following Sigmoid function to smoothly output the mechanical trust weight. and visual trust weight :

[0017]

[0018] in, The preset critical threshold, The preset Sigmoid steepness coefficient, and These are the lower and upper limits of the mechanical trust weight, respectively; The decoupling and fusion module calculates the visual compensation component according to the following formula. :

[0019] in, The preset visual compensation gain, The first nonlinear exponent is preset. The preset nominal reference blank holder force; The decoupling and fusion module calculates the mechanical compensation component according to the following formula. :

[0020] in, This is the preset mechanical compensation gain. This is a preset second nonlinear index; The decoupling and fusion module calculates and outputs the target blank holder force according to the following weighted fusion formula. :

[0021] In one embodiment of the present invention, the servo execution controller determines the puncture speed based on the working condition anomaly index during the puncture phase, including: the servo execution controller receiving the working condition anomaly index. and preset baseline puncture speed As input, the servo actuator dynamically calculates the puncture speed according to the following formula. :

[0022] in, The preset speed adjustment coefficient; the servo actuator will calculate the puncture speed. The output is sent to the servo-driven blank holder ring, and the target blank holder force is applied. Lock the blank holder force.

[0023] In one embodiment of the present invention, the servo execution controller selectively inserts a pressure holding pause during the upsetting stage based on an anomaly index, comprising: the servo execution controller receiving the anomaly index. and preset critical threshold As input, the servo actuator controller will use the operating condition anomaly index. With the critical threshold Perform real-time numerical comparison: when At that time, the servo execution controller automatically inserts an intermediate pressure holding pause with a preset duration of 50ms to 200ms during the upsetting section of the stroke; when Do not insert an intermediate pressure holding pause.

[0024] In one embodiment of the present invention, the servo execution controller determines the pressure holding duration based on the operating condition anomaly index during the pressure holding phase, including: the servo execution controller receiving the operating condition anomaly index. and the preset baseline holding time As input, the servo actuator dynamically calculates the pressure holding time according to the following formula. :

[0025] in, The servo actuator controller will calculate the pressure holding time as a preset abnormal operating condition extension factor. The output is sent to the servo-driven pressure ring, and a gradual linear unloading process is performed from 50ms to 100ms after the pressure holding period ends.

[0026] To achieve the above objectives, another aspect of the present invention proposes an adaptive control method for high-performance connection processes in production lines, comprising: Acquire visual data of the area to be riveted and generate pure geometric feature vectors; Execute probe pulse loading and extract the true dynamic stiffness response; The pure geometric eigenvectors are used as independent observations for inversion calculation to separate the geometric contribution and the material hardening contribution. The working condition anomaly index is constructed based on the pure geometric eigenvectors and the real dynamic stiffness response. The target blank holder force is output after fusing visual compensation and mechanical compensation based on the working condition anomaly index. Segmented riveting control is performed based on the anomaly index of working conditions and the target blank holder force: during the piercing stage, the piercing speed is given according to the anomaly index of working conditions and the blank holder force is locked with the target blank holder force; during the upsetting stage, a pressure holding pause is selectively inserted according to the anomaly index of working conditions; and during the pressure holding stage, the pressure holding time is given according to the anomaly index of working conditions.

[0027] This invention discloses an adaptive control system and method for high-performance connection processes in production lines. It constructs a dual-perception collaborative sensing system of "visual appearance + physical reaction force" using a 3D vision sensor, a load sensor, and an absolute encoder, overcoming the limitations of traditional SPR riveting guns that rely on preset force-displacement control. For the first time, it proposes using visually extracted pure geometric feature vectors as independent observations, substituting them into a dynamic stiffness composite response model for inversion calculation. This separates the geometric fit state from the contribution of material work hardening, solving the problem of misjudgment in compensation caused by the conflation of geometric and material factors in traditional methods, ensuring clear physical interpretability for each compensation decision. Furthermore, the entire riveting stroke is divided into an idle stroke segment, a piercing segment, an upsetting segment, and a final riveting pressure holding segment. The piercing speed is reduced according to the anomaly index to lower the peak impact stress. In the upsetting segment, when the anomaly index exceeds a critical threshold, an intermediate pressure holding segment is automatically inserted to utilize stress relaxation to reduce the radial tensile stress of the bottom plate. The final riveting pressure holding time is dynamically extended to fully release elastic strain energy, achieving adaptive control throughout the entire process. This invention effectively solves the problems of inconsistent connection quality and production line downtime caused by parameter fixation and material fluctuations in the prior art, and significantly improves the adaptability and connection quality consistency of the self-piercing riveting process under different material conditions.

[0028] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0029] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is an architecture diagram of a high-performance connection process adaptive control system for a production line according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the process steps of an adaptive control method for high-performance connection technology in a production line according to an embodiment of the present invention. Detailed Implementation

[0030] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0031] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0032] The following description, with reference to the accompanying drawings, describes an adaptive control system and method for high-performance connection processes in a production line, according to an embodiment of the present invention.

[0033] Example 1 like Figure 1 As shown, a high-performance connection process adaptive control system for production lines according to the present invention includes: Servo-driven clamping ring, which has a servo-driven clamping ring for clamping rivets and applying riveting force.

[0034] The sensing components include a 3D vision sensor mounted on the front end of the servo-driven pressure ring, and a load sensor and an absolute encoder mounted on the servo-driven pressure ring.

[0035] Further explanation is needed regarding the hardware selection of the sensing components: the 3D vision sensor employs a structured light binocular camera, mounted on the front end of the clamping ring, with a field of view of 60×45mm, a resolution of 8μm, and a frame rate of 30fps; the load sensor uses a piezoelectric ring force sensor with a range of 10kN, a sampling rate of 20kHz, and an accuracy of ±0.3%FS; the absolute encoder uses a multi-turn absolute photoelectric encoder with a resolution of 0.5μm and a repeatability of ±1μm. The 3D vision sensor is used to scan the surface of the sheet metal in the riveting area before the riveting gun reaches the riveting point, acquiring local 3D point cloud data and synchronous 2D image data. The load sensor and absolute encoder are used to synchronously acquire force-displacement time-domain response data during the application of the detection pulse.

[0036] The computing center is communicatively connected to the sensing components.

[0037] It should be further noted that the computing hub uses an AI edge computing box equipped with an NVIDIA Jetson AGX Orin (64GB), deploying a hybrid network of YOLOv8n object detection and ResNet-34 feature regression, with a single-point end-to-end inference latency of no more than 35ms. The computing hub integrates an object detection network, a feature regression network, a stiffness extraction module, and a decoupling fusion module.

[0038] The servo execution controller is connected to both the computing center and the servo drive pressure ring control.

[0039] It should be further explained that the servo actuator controller is used to receive the working condition anomaly index and target pressing force output by the calculation center, and divides the entire riveting stroke into a rapid idle stroke segment, a puncture segment, an upsetting segment and a final riveting pressure holding segment based on the dynamic characteristics of the real-time force-displacement curve to perform segmented differentiated control.

[0040] Specifically, the SPR riveting gun is a C-type riveting gun with a maximum riveting force of 80kN; the blank holder pressure range is 500-8000N; the servo motor has a rated power of 35kW and a maximum torque of 22N·m; the rivet specification is a semi-hollow rivet C5.3×6.5 with a hardness of H4 (approximately 480HV).

[0041] It should be noted that the working process of the system in this embodiment of the invention is as follows: The 3D vision sensor acquires images and point cloud data of the area to be riveted and sends them to the computing center.

[0042] The computational center generates pure geometric feature vectors based on the image and point cloud data. Specifically, the target detection network built into the computational center receives two-dimensional image data as input, and outputs the coordinates of the region of interest (ROI) defining the riveting area after target detection inference; the computational center performs spatial cropping and registration of the three-dimensional point cloud data based on the ROI coordinates to generate an elevation map representing the local surface morphology as an intermediate output; the elevation map is input into the built-in feature regression network for forward inference, and the feature regression network outputs pure geometric feature vectors. Pure geometric eigenvectors Each component is used to quantify the wavy edge height, collapse depth, and out-of-plane warping of the plate in the current riveting area.

[0043] The load sensor and the absolute encoder collect force-displacement data during the detection pulse loading process and send it to the calculation center. Specifically, after the pressure ring contacts the plate and establishes initial contact, the servo motor outputs a micro-amplitude trapezoidal detection pulse. The force value rises from the pre-contact force of about 100N to the detection force of about 500N with a controlled slope, is maintained for about 30ms, and then falls back to the pre-contact force. The total pulse duration is about 100-150ms.

[0044] The computational center generates a true dynamic stiffness response based on the force-displacement data. Specifically, the stiffness extraction module built into the computational center receives force-displacement time-domain response data synchronously acquired by the load sensor and the absolute encoder throughout the entire pulse loading process as input; the stiffness extraction module sequentially performs bandpass filtering (50-2000Hz) and linear regression fitting on the force-displacement time-domain response data, extracting the slope of the pulse loading segment as the system-level measured stiffness. Output: The stiffness extraction module queries a pre-calibrated equipment stiffness mapping table based on the spatial coordinates of the current riveting point to obtain the equipment stiffness at the corresponding riveting point. And according to the series stiffness decoupling formula The system-level measured stiffness is processed by equipment compliance subtraction to output the true dynamic stiffness response on the plate side. The true dynamic stiffness response This comprehensively reflects the coupling effect between the local geometric fit state of the current riveting area and the elastic-plastic response characteristics of the material.

[0045] The computational hub treats pure geometric feature vectors as independent observations for inversion calculation to separate geometric contributions from material hardening contributions. Specifically, the decoupling and fusion module built into the computational hub receives pure geometric feature vectors respectively. and true dynamic stiffness response As a dual-path input; the decoupling fusion module will use the pure geometric feature vector As independent observations of the geometric components, they are substituted into the pre-constructed composite response model. Invert the solution to obtain the actual dynamic stiffness response. Characterized as geometric contribution components Contribution to material hardening The superposition of the components; after inversion and solution, the separated geometric contribution components are output. and the contribution of material hardening .

[0046] The computational center constructs a condition anomaly index based on pure geometric feature vectors and the actual dynamic stiffness response. Specifically, the decoupling and fusion module normalizes the pure geometric feature vectors to obtain a geometric normalization ratio. The normalized stiffness ratio is obtained by normalizing the offset of the actual dynamic stiffness response relative to the nominal reference stiffness. The output operating condition anomaly index is calculated according to the following formula. :

[0047] in, The preset allowable deformation threshold, The nominal reference stiffness of the plate is given. The preset nonlinear sensitivity index, To meet The preset weighting coefficients.

[0048] The calculation center outputs the target edge clamping force by weighted fusing visual compensation components and mechanical compensation components based on the working condition anomaly index. Specifically, the decoupled fusion module receives the working condition anomaly index. As input, the abnormality index of the working condition is nonlinearly transformed using the following Sigmoid function to smoothly output the mechanical trust weight. and visual trust weight :

[0049]

[0050] in, This is a preset critical threshold (typically 0.5). The preset Sigmoid steepness coefficient (values ​​range from 10 to 15). and These are the lower and upper limits of the mechanical trust weight (0.2 and 0.9 respectively). This ensures that when the operating condition is normal... When the value is low, compensation is primarily based on visual geometric information; when the operating condition is abnormal... When the value is high, it smoothly switches to a system dominated by mechanical response information to avoid servo force oscillations caused by step switching.

[0051] The decoupling and fusion module calculates the visual compensation component according to the following formula. :

[0052] in, The preset visual compensation gain (values ​​range from 0.3 to 0.6). The first nonlinear exponent is preset (with a value of 1.0-1.5). The preset nominal reference blank holder force.

[0053] The decoupling and fusion module calculates the mechanical compensation component according to the following formula. :

[0054] in, The preset mechanical compensation gain (value 0.2-0.5) is used. It is a preset second nonlinear index (with a value of 1.0-1.2). Used to compensate for poor fit caused by geometric deformations such as wavy edges and collapsed corners. The sign function enables bidirectional adjustment of pressure, increasing when the material is too hard and decreasing when it is too soft.

[0055] The decoupling and fusion module calculates and outputs the target blank holder force according to the following weighted fusion formula. :

[0056] And through the clip function Limiting amplitude within a safe range This prevents indentations on the board surface and overloading of equipment caused by insufficient or excessive clamping.

[0057] The decoupling and fusion module will calculate the operating condition anomaly index. and target blank holder force The control command data is sent to the servo execution controller.

[0058] The servo execution controller receives the operating condition anomaly index. and target blank holder force The entire riveting process is divided into multiple stages for segmented control. Specifically: During the puncture phase, the servo actuator controller applies the target clamping force. The blank holder ring is controlled by a constant blank holder force locking value, and the puncture speed is dynamically calculated according to the following formula. :

[0059] in, The preset baseline puncture speed, The preset speed adjustment coefficient; adjusts the puncture speed according to the abnormality index of the working conditions. The pressure is increased and decreased to reduce the peak impact stress and the risk of over-shear cracking of the base plate.

[0060] During the upsetting stage, the servo actuator controller will use the abnormality index of the working condition. With the preset critical threshold Perform real-time numerical comparison: when During the upsetting process, an intermediate pressure holding pause with a preset duration of 50ms to 200ms is automatically inserted to actively reduce the peak radial tensile stress of the bottom plate using the stress relaxation effect; when Do not insert an intermediate pressure holding pause.

[0061] During the pressure holding phase, the servo actuator dynamically calculates the pressure holding time according to the following formula. :

[0062] in, The preset baseline holding time is 300-500ms. The preset abnormal operating condition extension factor is (0.5-1.5); that is... The greater the pressure, the longer the holding time, in order to fully release the elastic strain energy. The servo actuator controller will calculate the holding time. The output is sent to the servo-driven pressure ring, and a gradual linear unloading process is performed from 50ms to 100ms after the pressure holding period ends.

[0063] At the same time, the system will complete each riveting step. , , , , , , Once the data is fed back to the parameter library, the ResNet convolutional neural network is continuously optimized using the Bayesian optimization algorithm, enabling the model to adaptively cope with batch fluctuations in materials.

[0064] The system of this invention effectively solves the problems of inconsistent connection quality and production line stoppage caused by parameter solidification and material fluctuations in existing self-piercing riveting processes. It achieves adaptive control of riveting parameters at a specific point and improves the consistency of joint quality.

[0065] Example 2 As an example, the following implementation case of self-piercing riveting adaptive compensation using DC01-1.6mm / 6061-T6-3.0mm dissimilar plates is presented to illustrate the invention: This embodiment selects a typical steel-aluminum dissimilar connection condition in automotive body-in-white. The upper plate is cold-rolled low-carbon steel DC01, with a thickness of 1.6mm, tensile strength of 410MPa, yield strength of 280MPa, and elongation of 28%. The lower plate is pre-aged aluminum alloy 6061-T6, with a thickness of 3.0mm, tensile strength of 310MPa, yield strength of 276MPa, and elongation of 10%. The total laminate thickness is 4.6mm. Due to the excellent plasticity of DC01 and the high yield strength and obvious work hardening tendency of 6061-T6, the traditional fixed-parameter SPR strategy is prone to typical defects such as excessively deep steel plate indentation and insufficient rivet leg expansion under this combination.

[0066] Actuator configuration: SPR riveting gun model is C-type riveting gun, with a maximum riveting force of 80kN; the blank holder pressure range is 500-8000N; the servo motor has a rated power of 35kW and a maximum torque of 22N·m; the rivet specification is a semi-hollow rivet C5.3×6.5, with a hardness of H4 (approximately 480HV).

[0067] Sensing component configuration: The 3D vision sensor is a structured light binocular camera, mounted on the front end of the pressure ring, with a field of view of 60×45mm, a resolution of 8μm, and a frame rate of 30fps; the load sensor is a piezoelectric ring force sensor with a range of 10kN, a sampling rate of 20kHz, and an accuracy of ±0.3%FS; the absolute encoder is a multi-turn absolute photoelectric encoder with a resolution of 0.5μm and a repeatability of ±1μm.

[0068] Computing hub configuration: An AI edge computing box equipped with NVIDIA Jetson AGX Orin (64GB), deploying a YOLOv8n object detection network and a ResNet-34 feature regression hybrid network, with a single-point end-to-end inference latency of no more than 35ms.

[0069] Calibrated parameters: N, N / 4500N, kN / mm, kN / mm, mm, , , , , , , .

[0070] It needs to be further explained that, such as Figure 2 As shown, the detailed method flow of this example is as follows: S1. Visual and Feedforward Predictive Control Stage: The rivet gun is positioned at a distance from the point to be riveted. The system hovers at a height of mm, and the 3D vision sensor initiates scanning, simultaneously acquiring one frame of 2D image and one frame of point cloud (approximately 120,000 points) in 33ms. The YOLOv8n network performs inference on the 2D image, outputting the bounding box of the riveting region ROI with a detection confidence of 0.94, corresponding to the physical region. mm. The point cloud is cropped based on ROI back projection and then subjected to statistical filtering ( , After removing outliers, register to the rivet gun coordinate system to generate... A pixel elevation map with a z-axis resolution of 0.05 mm. ResNet-34 performs inference on the elevation map and outputs feature vectors. mm, where mm (height of the wavy edge, medium to large). mm (collapse depth, normal range) mm (out-of-plane warping, quite noticeable). Normalization treatment The current panel is determined to have a moderate geometric fit defect.

[0071] S2. Pre-compression and dynamic stiffness response coefficient K extraction stage: The pressure ring moves downwards at 80 mm / s, and the force sensor detects this. Contact is determined at time N, and the contact displacement is recorded. As a baseline, the servo motor outputs probe pulses according to the following timing sequence: 0-45ms, linearly increasing from 100N to 500N at 8.9N / ms; 45-75ms, maintaining 500N (for 30ms); 75-120ms, linearly decreasing back to 100N. Load sensors at 20kHz and encoders at 10kHz synchronously acquire data, resulting in 900 sets of force-displacement data points during the pulse loading phase. After bandpass filtering (50-2000Hz) and linear regression fitting of the increasing phase (100N→500N interval):

[0072] Fitting The data quality is acceptable. A query of the equipment stiffness mapping table yields... kN / mm, according to the series stiffness decoupling formula:

[0073] Substituting D of S1 as geometrically independent observations into the composite response model The geometric contribution term was calibrated offline. kN / mm, inverted material hardening contribution kN / mm, offset from nominal value The 6061-T6 material batch was determined to be slightly harder. relatively Overall offset .

[0074] S3. Formal dynamic compensation calculation stage for blank holder force: Operating Condition Anomaly Index calculate:

[0075] determination The current operating conditions are dominated by geometric factors.

[0076] Trust weight allocation:

[0077]

[0078] Visual geometric information dominated (77.6%), while mechanical response was secondary (22.4%).

[0079] Visual compensation component:

[0080] Mechanical compensation components:

[0081] Target edge clamping force fusion:

[0082] Clip limiter check: Final output N, compared to The N value is increased by approximately 25%, primarily to overcome the poor fit caused by the 1.82mm wavy edge and 1.36mm warpage.

[0083] S4. Servo segmented riveting control optimization stage: Phase I – Rapid Idle Stroke Section: Control mode is speed control, speed mm / s, punch distance 25mm from nail head, termination criterion is force sensor detection. N, the actual measurement time is about 130ms, and the smooth cutting stage II is pre-decelerated to 50mm / s within 50ms.

[0084] Phase II – Puncture Phase: The control mode is speed-force composite control, with dynamic setting of puncture speed. mm / s, blank holder force locking N. The measured peak puncture force is approximately 22.5 kN (at the moment of DC01 penetration). The inflection point criterion dF / dS is determined by the peak value. kN / mm decreased to The puncture is considered complete when the pressure is below kN / mm; the actual puncture displacement is measured. mm, taking approximately 26ms.

[0085] Stage III – Upsetting Section: Speed ​​limit 20-50 mm / s (35 mm / s in this example), trajectory shape factor Initial thrust kN, termination thrust kN, upset termination displacement mm. Rivet thrust trajectory:

[0086] Intermediate pressure holding determination: Without any pauses, the measured time was approximately 72ms. The thrust increased monotonically without any abnormal steps, indicating that the rivet leg had expanded sufficiently.

[0087] Phase IV – Final Riveting and Pressure Holding Section: Control mode is load holding, pressure holding kN, holding time dynamically given:

[0088] Force decay monitoring relaxation completion threshold N / ms, measured at 480ms The pressure holding time is determined to be N / ms. Within 80ms, the linear ramp decreases from 52kN to 0, and the pressure ring simultaneously decreases from 250kN to 0 within 50ms.

[0089] It is worth mentioning that the riveting results are as follows: the measured height of the rivet head protrusion is +0.08mm (target value -0.1mm +0.3mm), and the self-locking value is 0.52mm (target value). 0.40mm), remaining bottom thickness 0.48mm (target value) 0.35mm), upper plate indentation depth 0.09mm (target value) (0.15mm), no visible cracks on the base plate. All quality indicators are qualified.

[0090] It is worth mentioning that, through comparative verification: the same board material was combined using the scheme of this invention and the traditional fixed parameter scheme respectively ( Comparison of 50 points of each riveting method (N fixed, single-segment force control): Traditional method has a self-locking value deficiency rate of 18% and an indentation excess rate of 8%; The method of this invention has a self-locking value deficiency rate of 2% (improvement -89%) and an indentation excess rate of 0% (improvement -100%).

[0091] The practical application of this invention effectively solves the problems of inconsistent connection quality and production line downtime caused by parameter fixation and material fluctuations in the prior art. It realizes complete closed-loop control from multi-source sensing, geometry-material physics decoupling, target blank holder force adaptive fusion to full-process servo segmented execution, which significantly improves the adaptability and connection quality consistency of the self-piercing riveting process under different material conditions.

[0092] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0093] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

Claims

1. A production line high performance joining process adaptive control system, characterized by, include: Servo-driven pressure ring; The sensing components include a 3D vision sensor mounted on the front end of the servo-driven pressure ring, and a load sensor and an absolute encoder mounted on the servo-driven pressure ring. The computing center is communicatively connected to the sensing components; A servo execution controller is connected to both the computing center and the servo-driven pressure ring control. The 3D vision sensor acquires images and point cloud data of the area to be riveted and sends them to the computing center. The computing center generates a pure geometric feature vector based on the image and point cloud data, generates a real dynamic stiffness response based on the force-displacement data collected by the load sensor and the absolute encoder during the detection pulse loading process, performs inversion calculation on the pure geometric feature vector as an independent observation value to separate the geometric contribution and the material hardening contribution, constructs a working condition anomaly index based on the pure geometric feature vector and the real dynamic stiffness response, and outputs the target blank holder force by weighted fusion of visual compensation component and mechanical compensation component based on the working condition anomaly index. The servo actuator receives the working condition anomaly index and the target blank holder force, and divides the entire riveting stroke into multiple stages to perform segmented control: in the piercing stage, the piercing speed is given based on the working condition anomaly index and the blank holder force is locked with the target blank holder force; in the upsetting stage, a pressure holding pause is selectively inserted according to the working condition anomaly index; and in the pressure holding stage, the pressure holding duration is given based on the working condition anomaly index.

2. The system of claim 1, wherein, The computational center generates pure geometric feature vectors based on the image and point cloud data, including: The target detection network built into the computing center receives two-dimensional image data as input, and outputs the coordinates of the region of interest that defines the riveting area after target detection and inference. The computing center performs spatial cropping and registration of the 3D point cloud data based on the coordinates of the region of interest, and generates an elevation map representing the local surface morphology as an intermediate output. inputting the elevation map into an embedded feature regression network for forward inference, outputting a pure geometric feature vector by the feature regression network ; pure geometric feature vector The components of the pure geometric feature vector are respectively used for quantifying the wave edge height, the collapse depth and the out-of-plane warping amount of the plate in the current riveting area.

3. The system of claim 1, wherein, The calculation center generates a true dynamic stiffness response based on the force-displacement data, including: The stiffness extraction module built into the computing center receives force-displacement time-domain response data synchronously acquired by the load sensor and the absolute encoder throughout the entire process of probe pulse loading as input; The rigidity extraction module sequentially performs band-pass filtering processing and linear regression fitting on the force-displacement time domain response data, and extracts the slope of the pulse loading section as the system level measured rigidity output; The rigidity extraction module queries a pre-calibrated equipment rigidity mapping table according to the spatial coordinates of the current riveting point to obtain the equipment rigidity under the corresponding riveting point And according to the series rigidity decoupling formula, the equipment flexibility deduction processing is performed on the system-level measured rigidity. With the real dynamic stiffness response on the output plate side .

4. The system according to claim 1, characterized in that, The computational center uses pure geometric eigenvectors as independent observations for inversion calculation to separate geometric contributions and material hardening contributions, including: The decoupling and fusion modules built into the computing center receive pure geometric feature vectors respectively. and true dynamic stiffness response As a dual-path input; the decoupling fusion module will use the pure geometric feature vector As independent observations of the geometric components, they are substituted into the pre-constructed composite response model for inversion solution: The composite response model will reflect the true dynamic stiffness response. Characterized as geometric contribution components Contribution to material hardening The superposition; After inversion calculation, the decoupling and fusion module outputs the separated geometric contribution components. and the contribution of material hardening .

5. The system according to claim 1, characterized in that, The calculation center constructs a condition anomaly index based on pure geometric eigenvectors and the actual dynamic stiffness response, including: The decoupling and fusion module built into the computing center receives pure geometric feature vectors. and true dynamic stiffness response As input; The decoupling and fusion module normalizes the pure geometric feature vectors to obtain the geometric normalization ratio. The normalized stiffness ratio is obtained by normalizing the offset of the actual dynamic stiffness response relative to the nominal reference stiffness. ; The decoupling and fusion module calculates and outputs the operating condition anomaly index according to the following formula. : in, The preset allowable deformation threshold, The nominal reference stiffness of the plate is given. The preset nonlinear sensitivity index, To meet The preset weighting coefficients.

6. The system according to claim 1, characterized in that, The calculation center outputs the target edge clamping force based on a weighted fusion of visual compensation components and mechanical compensation components according to the working condition anomaly index, including: The decoupling and fusion module built into the computing center receives the operating condition anomaly index. As input; The decoupling and fusion module performs a nonlinear transformation on the working condition anomaly index using the following Sigmoid function to smoothly output the mechanical trust weight. and visual trust weight : in, The preset critical threshold, The preset Sigmoid steepness coefficient, and These are the lower and upper limits of the mechanical trust weight, respectively; The decoupling and fusion module calculates the visual compensation component according to the following formula. : in, The preset visual compensation gain, The first nonlinear exponent is preset. The preset nominal reference blank holder force; The decoupling and fusion module calculates the mechanical compensation component according to the following formula. : in, This is the preset mechanical compensation gain. This is a preset second nonlinear index; The decoupling and fusion module calculates and outputs the target blank holder force according to the following weighted fusion formula. : 。 7. The system according to claim 1, characterized in that, The servo actuator controller determines the puncture speed based on the abnormality index of the working condition during the puncture phase, including: The servo execution controller receives the operating condition anomaly index. and preset baseline puncture speed As input; The servo actuator dynamically calculates the puncture speed according to the following formula. : in, The preset speed adjustment coefficient; The servo execution controller will calculate the puncture speed. The output is sent to the servo-driven blank holder ring, and the target blank holder force is applied. Lock the blank holder force.

8. The system according to claim 1, characterized in that, The servo actuator selectively inserts a pressure-holding pause during the upsetting stage based on the abnormality index of the working conditions, including: The servo execution controller receives the operating condition anomaly index. and preset critical threshold As input; The servo actuator controller will use the abnormality index of the working condition. With the critical threshold Perform real-time numerical comparison: when At that time, the servo execution controller automatically inserts an intermediate pressure holding pause with a preset duration of 50ms to 200ms during the upsetting section of the stroke; when Do not insert an intermediate pressure holding pause.

9. The system according to claim 1, characterized in that, The servo actuator controller specifies the holding time based on the operating condition anomaly index during the holding phase, including: The servo execution controller receives the operating condition anomaly index. and the preset baseline holding time As input; The servo actuator dynamically calculates the pressure holding time according to the following formula. : in, This is a preset extension factor for abnormal operating conditions; The servo actuator will calculate the pressure holding time. The output is sent to the servo-driven pressure ring, and a gradual linear unloading process is performed from 50ms to 100ms after the pressure holding period ends.

10. An adaptive control method for high-performance connection processes in a production line, characterized in that, include: Acquire visual data of the area to be riveted and generate pure geometric feature vectors; Execute probe pulse loading and extract the true dynamic stiffness response; The pure geometric eigenvectors are used as independent observations for inversion calculation to separate the geometric contribution and the material hardening contribution. The working condition anomaly index is constructed based on the pure geometric eigenvectors and the real dynamic stiffness response. The target blank holder force is output after fusing visual compensation and mechanical compensation based on the working condition anomaly index. Segmented riveting control is performed based on the anomaly index of working conditions and the target blank holder force: during the piercing stage, the piercing speed is given according to the anomaly index of working conditions and the blank holder force is locked with the target blank holder force; during the upsetting stage, a pressure holding pause is selectively inserted according to the anomaly index of working conditions; and during the pressure holding stage, the pressure holding time is given according to the anomaly index of working conditions.

Images