A machine vision-based SMT mounting plate card adaptive clamping system and device

Abstract

The present application relates to the technical field of electronic manufacturing equipment, and discloses a SMT mounting plate card adaptive clamping system and equipment based on machine vision, which comprises a track transmission module, a driving execution module, a visual acquisition module, a force feedback module and a central control module. The visual acquisition module collects two-dimensional surface images and three-dimensional point cloud data of the plate card to be mounted; the central control module uses an algorithm to analyze the material type, actual thickness and warping degree characteristics of the plate card, and calculates a unique target clamping force based on a multi-factor weighted model. The driving execution module combines the force feedback signal and drives the track edge to apply precise clamping force to the plate card through a position and torque hybrid control strategy. The present application integrates a dynamic thermal compensation mechanism, can fine-tune the track width according to temperature changes, realizes flexible adaptive clamping of plate cards of different specifications and materials, effectively eliminates plate card warping, slipping and thermal deformation, and improves the positioning accuracy and yield of SMT mounting.

Description

Technical Field

[0001] This invention relates to the field of electronic manufacturing equipment technology, and in particular to an adaptive clamping system and device for SMT placement boards based on machine vision. Background Technology

[0002] Surface Mount Technology (SMT) is a core process in the modern electronics manufacturing industry. The positioning accuracy and flatness of the printed circuit board (PCB) within the mounting area directly determine the component mounting quality. In existing SMT production equipment, the track transport and clamping mechanism typically employs mechanical limits or pneumatic drive. This involves adjusting the track to a fixed position based on a preset board width and applying lateral pressure to the board edges using cylinders or simple spring mechanisms to achieve fixation.

[0003] However, this traditional rigid mechanical clamping method has technical limitations. Because existing clamping systems generally lack the ability to sense the physical properties of the circuit board, their control logic is usually based on open-loop settings, unable to adaptively adjust according to the material characteristics, thickness tolerances, or initial deformation state of the actual workpiece. When facing the flexible production needs of multiple varieties and small batches, this single constant clamping force often struggles to balance the contradiction between stable fixation and prevention of deformation. For example, when handling thin circuit boards with weak rigidity (such as FR-4 thin boards), in order to overcome the risk of displacement caused by the high-speed movement of the placement head, the system often needs to apply a large lateral friction force. This can easily cause the circuit board to arch upwards or twist due to lateral compression, severely damaging the process coplanarity of the circuit board and leading to quality defects such as poor soldering, bridging, or tombstoning in fine-pitch components. Conversely, if the clamping force is reduced to protect the circuit board, slight slippage may occur during high-speed placement due to insufficient friction, resulting in placement coordinate shift.

[0004] Furthermore, existing technologies are inadequate for handling complex operating conditions. On one hand, substrates of different materials (such as ceramic substrates, aluminum substrates, and fiberglass boards) have vastly different elastic moduli and coefficients of thermal expansion. Traditional equipment cannot automatically identify the material and match corresponding mechanical control strategies, easily causing brittle materials to break or flexible materials to deform excessively. On the other hand, SMT operating environments or equipment interiors often experience temperature fluctuations, and circuit boards undergo thermal expansion and contraction after reflow soldering preheating or prolonged operation. Existing fixed tracks cannot sense and adapt to these micro-dimensional changes, leading to internal stress accumulation due to thermal expansion and boundary constraints, further exacerbating the nonlinear deformation of the circuit boards. For large-size or thick heavy-duty circuit boards, relying solely on edge clamping faces the problem of sagging in the central area due to its own weight. Existing technologies lack effective dynamic auxiliary support methods, making it difficult to ensure high consistency across the entire board area. Therefore, a clamping system with sensing and feedback capabilities and adaptive flexible control is urgently needed. Summary of the Invention

[0005] The purpose of this invention is to provide an adaptive clamping system and equipment for SMT placement boards based on machine vision. This solves the problems of existing SMT equipment using fixed clamping parameters due to a lack of perception of the material and geometric features of the boards, which leads to boards being prone to deformation under stress, positioning slippage, and loss of process coplanarity due to thermal expansion.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] The first aspect of this invention provides an adaptive clamping system for SMT (Surface Mount Technology) board placement based on machine vision, comprising a track transmission module, a drive execution module, a vision acquisition module, a force feedback module, and a central control module. The track transmission module constitutes the mechanical load-bearing body of the system, including a first track edge and a second track edge that are parallel to each other. The second track edge is configured to displace relative to the first track edge along a direction perpendicular to the board transfer direction to adjust the width of the transmission channel. The drive execution module is mechanically connected to the second track edge and is used to drive the second track edge to perform linear motion. The vision acquisition module is located in the entrance area of ​​the track transmission module and is configured to acquire two-dimensional surface image data and three-dimensional point cloud data of the board before it enters the placement working area. The force feedback module is associated with the drive execution module and is used to monitor the actual drive torque output by the drive execution module in real time.

[0008] The central control module establishes communication connections with the drive execution module, vision acquisition module, and force feedback module. The central control module receives and processes two-dimensional surface image data and three-dimensional point cloud data, analyzing the material type, actual thickness, and warpage characteristics of the board to be mounted. Based on these analyzed physical characteristics, the central control module calculates the target clamping force for the board using preset decision logic. Subsequently, the central control module generates control commands based on the target clamping force and sends them to the drive execution module. It also uses feedback signals from the force feedback module to perform closed-loop control on the second track edge, thereby applying a precise target clamping force to the edge of the board.

[0009] In a preferred embodiment, the vision acquisition module employs a multi-dimensional perception architecture. A high-resolution industrial camera, with its optical axis perpendicular to the board's transmission plane, acquires two-dimensional surface image data containing the board's surface texture features. A line laser profile sensor, combined with a linear encoder, performs cross-sectional profile scanning on the moving board to obtain three-dimensional point cloud data reflecting the board's surface height distribution. This configuration enables optical digital characterization of the board's physical properties.

[0010] In a preferred embodiment, the central control module pre-stores a mechanical property database, which establishes a mapping relationship between different material types and material stiffness coefficients. When calculating the target clamping force, the system employs a multi-factor weighted calculation model. This model indexes the material stiffness coefficient corresponding to the material type; multiplies the actual thickness by a thickness influence weighting factor to determine the friction holding force component, which is used to overcome the weight of the plate and the shear force during processing; and multiplies the warpage characteristic by a correction force weighting factor to determine the correction force component, which is used to flatten the warped plate to the allowable range of the process. The final target clamping force is determined as the product of the sum of the friction holding force component and the correction force component and the material stiffness coefficient, plus a preset basic preload constant.

[0011] In a preferred embodiment, the drive execution module adopts a dual-end synchronous drive architecture, including dual-axis synchronous servo motor sets and two sets of parallel precision ball screw assemblies located at both ends of the track transmission module. The servo motors drive the ball screw assemblies to move the two ends of the second track edge synchronously, thereby eliminating the torque imbalance caused by single-sided drive. In addition, the track edge sidewalls are integrated with reinforcing rib structures, and the contact surfaces are made of high-hardness alloy materials to improve mechanical rigidity under high clamping force conditions.

[0012] In a preferred embodiment, the feature recognition processing of the central control module includes: processing two-dimensional surface image data using a convolutional neural network model to extract the surface texture features of the board to output the material type; simultaneously, performing region segmentation on the three-dimensional point cloud data to extract an effective point cloud set, determining the actual thickness by calculating the average height of the effective point cloud set, and determining the warping feature by calculating the difference between the maximum and minimum height values ​​in the effective point cloud set.

[0013] In a preferred embodiment, the system employs a segmented hybrid control strategy. During the rapid approach phase, the drive execution module operates in position control mode until the actual drive torque exceeds a preset contact threshold. Upon contact detection triggering, the system switches to the flexible locking phase, operating in torque control mode. In this phase, the controller dynamically adjusts the motor current using a PID control algorithm, stabilizing the actual drive torque at a reference value corresponding to the target clamping force, thereby achieving impact-free flexible clamping.

[0014] In a preferred embodiment, the system integrates a dynamic thermal compensation mechanism. An ambient temperature sensor monitors the mounting environment temperature in real time, and, combined with the material's coefficient of thermal expansion, temperature difference data, and the initial width of the board, calculates the thermal expansion compensation amount for the board. During mounting, the system dynamically updates the target position setpoint of the second track edge based on this compensation amount, driving the second track edge to perform a slight outward retraction movement to release the internal stress generated by the board's thermal expansion and prevent the board from arching.

[0015] In a preferred embodiment, the system is further configured with a bottom auxiliary support module located below the track. When the actual thickness of the card exceeds a preset heavy-load threshold, the central control module controls the bottom auxiliary support module to rise, using a flexible pin array to contact the bottom surface of the card and provide support reaction force to eliminate the sagging deformation of the central area of ​​the thick card caused by its own weight.

[0016] A second aspect of the present invention provides a computer device including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the operation steps performed by the central control module in any of the above embodiments. This computer device enables automated conversion and control from visual perception data to mechanically executed actions.

[0017] In summary, the present invention has at least one of the following beneficial technical effects:

[0018] 1. This invention acquires the three-dimensional features of the circuit board's material, thickness, and warpage through a visual acquisition module, and uses a multi-factor weighted calculation model in the central control module to transform this unstructured data into a unique target clamping force for each circuit board. This technical feature enables the system to automatically adjust the clamping force according to the differences in circuit board rigidity. It can apply sufficient corrective force to restore flatness to flexible circuit boards with warpage, while avoiding excessive stress on brittle substrates such as ceramics that could lead to breakage. This effectively solves the problem of circuit board slippage or mechanical damage caused by fixed clamping force in traditional equipment.

[0019] 2. This invention employs a force feedback module combined with a segmented hybrid control strategy, seamlessly switching from position control to torque closed-loop control the instant the track edge contacts the board. This control method can compensate for friction loss and clearance error of the lead screw drive in real time, achieving micron-level flexible locking; combined with a dual-axis synchronous servo drive structure, it ensures strict parallelism of the second track edge during movement, preventing board rotational offset caused by imbalance of single-sided drive torque, thus meeting the stringent requirements for substrate positioning accuracy in high-density, small-pitch component mounting.

[0020] 3. This invention integrates a dynamic thermal compensation mechanism and a bottom auxiliary support module. The dynamic thermal compensation mechanism dynamically fine-tunes the track width based on real-time temperature difference and the material's thermal expansion coefficient, effectively releasing the internal stress generated by the thermal expansion of the circuit board and preventing the board from arching. The bottom auxiliary support module provides active support for thick circuit boards, eliminating center sagging caused by its own weight. These two technical features work together to ensure that the circuit board maintains an ideal horizontal state during reflow preheating and long-term suspended operation, avoiding component soldering defects or tombstoning defects caused by substrate deformation. Attached Figure Description

[0021] Figure 1This is a schematic diagram of the overall system structure and electrical connections of the present invention;

[0022] Figure 2 This is a block diagram of the control logic of the expert decision-making system of the present invention;

[0023] Figure 3 This is a flowchart illustrating the adaptive clamping method of the present invention. Detailed Implementation

[0024] The following is in conjunction with the appendix Figure 1 -Appendix Figure 3 The present invention will be further described in detail below.

[0025] This invention provides a machine vision-based adaptive clamping system and device for SMT placement boards.

[0026] See attached document Figure 1 This invention provides a machine vision-based adaptive clamping system and device for SMT (Surface Mount Technology) board mounting. The system is configured to perform board transfer and positioning operations in the SMT mounting process. The system mainly includes a track transfer module, a drive execution module, a vision acquisition module, a force feedback module, and a central control module. These modules are physically connected and interact with each other via electrical buses or signal lines, collaboratively completing the adaptive clamping of boards of different thicknesses.

[0027] The track transport module constitutes the mechanical support of the system, used to support and transport the PCB boards to be mounted. The track transport module includes a first track edge and a second track edge arranged parallel to each other. A transport channel for accommodating the boards is defined between the first and second track edges. In this embodiment, the first track edge is set as a fixed reference edge, and the second track edge is set as a movable edge that can be displaced perpendicular to the transport direction to adjust the width of the transport channel. The contact surfaces of both the first and second track edges are made of a high-hardness alloy material, which possesses high yield strength and wear resistance.

[0028] To enhance the load-bearing rigidity of thick circuit boards ranging from 5mm to 8mm, reinforcing ribs are integrated into the non-contact sidewalls of the first and second track edges. These reinforcing ribs are arrayed along the track length to increase the moment of inertia of the track cross-section and suppress bending deformation under load. An embedded buffer strip is provided on the inner sidewall of the second track edge. The buffer strip is made of a material with a lower hardness than the high-hardness alloy material, providing slight elastic deformation at the rigid contact interface. The mechanical travel of the track transmission module is configured to allow a maximum opening distance between the first and second track edges to accommodate circuit boards up to 8mm thick.

[0029] The drive execution module is mechanically connected to the track transmission module and is used to drive the second track edge to generate precise linear motion. The drive execution module includes a dual-axis synchronous servo motor set and a precision ball screw assembly. The dual-axis synchronous servo motor set is respectively located at the inlet and outlet ends of the transmission channel, and drives two sets of parallel precision ball screw assemblies through couplings. The second track edge engages with the precision ball screw assembly through a screw-nut pair. This dual-end drive physical architecture is used to eliminate the torque imbalance and track figure-eight opening deformation caused by single-point drive when clamping large mass plates.

[0030] The servo motor in the drive execution module converts rotational motion into linear displacement of the second track edge. This motion conversion follows kinematic relationships, where the displacement of the second track edge relative to the first track edge... (Unit: mm) and rotation angle of the servo motor The functional relationship between (unit: radians) is expressed as follows:

[0031] ;

[0032] In the above formula, Lead (unit: millimeters) of a precision ball screw assembly is the linear distance the nut moves in one revolution of the screw. Pi is a constant. This indicates the reduction ratio of the reduction mechanism between the motor output shaft and the lead screw (if a direct drive is used, then...). This formula is used to establish a deterministic mapping between the motor control quantity and the physical clamping width.

[0033] The vision acquisition module is installed above the entrance area of ​​the track transport module, located on the path before the board enters the placement area. The vision acquisition module includes a high-resolution industrial camera and a line laser profile sensor. The optical axis of the high-resolution industrial camera is perpendicular to the board transport plane and is configured to acquire two-dimensional surface image data of the board entering the system. This two-dimensional surface image data includes the board's surface texture features and identification code information. The line laser profile sensor is configured to project a linear laser beam onto the board surface and receive the reflected light signal to generate three-dimensional point cloud data of the board surface. This three-dimensional point cloud data reflects the board's thickness distribution and surface flatness profile.

[0034] The force feedback module is associated with the drive execution module and is used to monitor the force parameters during the clamping process in real time. In this embodiment, the force feedback module includes a dynamic torque sensor connected in series between the servo motor output shaft and the lead screw. The dynamic torque sensor is configured to output an electrical signal reflecting the actual clamping resistance torque. In another embodiment, the force feedback module can also be integrated into the servo driver, indirectly calculating the output torque by monitoring the motor's current loop load (q-axis current).

[0035] The central control module establishes communication connections with the vision acquisition module, drive execution module, and force feedback module. It receives 2D surface image data and 3D point cloud data from the vision acquisition module, and torque feedback signals from the force feedback module. The central control module sends position and speed commands to the drive execution module via its internal bus. Using its built-in processor, the central control module performs calculations, analyzes the board parameters based on the received data, and generates corresponding control strategies, thereby achieving closed-loop motion control of the second track edge.

[0036] See attached document Figure 2 This embodiment further discloses the specific structure of the vision acquisition module and its implementation method for acquiring physical characterization data of the board. The vision acquisition module is constructed as a multi-dimensional optical detection unit, and its physical architecture includes a two-dimensional image acquisition unit and a three-dimensional contour scanning unit. The two-dimensional image acquisition unit mainly consists of a high-resolution industrial camera and its matching light source components.

[0037] The light source assembly is configured to provide a uniform and high-contrast illumination field to the surface of the board under test. Its specific form may include a coaxial light source, a ring light source, or a strip light source, to accommodate the reflective characteristics of boards made of different materials. The high-resolution industrial camera includes an optical lens assembly and a photoelectric conversion sensor (such as a CMOS or CCD chip), whose optical axis is orthogonal to the track transport plane.

[0038] The two-dimensional image acquisition unit operates according to the principle of perspective projection imaging. When the board transmits data to the center of the field of view, the high-resolution industrial camera exposes under the control of a trigger signal, converting the light intensity distribution on the board surface into a digital image matrix. Let a point in the physical coordinate system of the board surface be... Its corresponding point in the image pixel coordinate system is The mapping relationship between the two is determined by the camera's intrinsic and extrinsic parameter matrices. This two-dimensional image data is transmitted to the back-end processing unit as the basic data source for subsequent texture analysis and material classification.

[0039] The 3D contour scanning unit consists of a line laser contour sensor and a linear encoder. The line laser contour sensor operates on the principle of laser triangulation. A line laser emitter projects a laser line onto the board surface, which generates diffuse reflection light on the surface. The receiving lens group inside the line laser contour sensor forms an angle with the emitted optical axis. It collects reflected light in the direction of the object and focuses it onto the internal position-sensitive detector (PSD) or linear image sensor.

[0040] When the height of the board surface changes, the imaging position of the laser line on the internal image sensor will shift laterally. In this embodiment, the line laser profile sensor calculates the height value of each point on the board surface relative to the reference plane based on this lateral shift. Height value The calculation follows the following geometric trigonometric relationship formula:

[0041] ;

[0042] In the above formula, Indicates the optical baseline distance (in millimeters) from the laser emission point to the center of the receiving lens. The angle between the laser emission axis and the receiving axis is expressed in radians; ω represents the angle of incidence of the reflected light entering the receiving lens. This angle ω varies with the height of the board surface and is directly derived from the pixel position coordinates of the light spot on the internal sensor. Using this formula, the sensor converts the pixel displacement of the light spot into physical height data.

[0043] To construct the complete 3D shape of the board, a linear encoder is introduced into the system. The linear encoder is mechanically coupled to the conveying mechanism of the track transfer module, providing real-time feedback of the board's longitudinal displacement pulses. A line laser profile sensor is configured to trigger a profile scan upon receiving an equally spaced pulse signal from the linear encoder, thereby continuously acquiring a series of cross-sectional profile data as the board moves along the track.

[0044] The final dataset output by the vision acquisition module includes: two-dimensional grayscale or color images characterizing the optical properties of the board surface. and a 3D point cloud dataset characterizing the geometry of the board. 3D point cloud dataset A series of coordinate points The set consists of, where The thickness fluctuations and warping deformation of the board at different locations were accurately recorded. The aforementioned two-dimensional and three-dimensional data were strictly aligned in time and space using system clock and encoder pulses to ensure the consistency of texture and height information at the same physical location.

[0045] This embodiment details how the central control module utilizes algorithms to analyze visual data and extract key physical parameters for subsequent clamping force decisions. The feature recognition and deep learning module is configured to run within the high-performance computing unit of the central control module, and its processing flow is divided into an image feature analysis path and a geometric shape analysis path. These two paths process the two-dimensional image data from the visual acquisition module in parallel. and 3D point cloud data .

[0046] In the image feature parsing path, the system integrates a pre-trained convolutional neural network (CNN) model. To meet the real-time requirements of industrial environments and prevent overfitting, the CNN model in this embodiment is specifically constructed as a lightweight five-layer sequence structure. The input layer of this model is configured to receive a pre-processed 256×256 pixel single-channel grayscale image matrix. The front end of the model contains three consecutive feature extraction units. Each feature extraction unit sequentially contains a convolutional layer, a batch normalization layer, and a max pooling layer. The first convolutional layer uses 32 3×3 kernels with a stride of 1 to extract primary edge and texture features; the second and third convolutional layers use 64 and 128 3×3 kernels, respectively, to extract high-dimensional semantic features. All convolutional layers use Rectified Linear Units (ReLU) as the activation function to increase the model's non-linear expressiveness and accelerate convergence. The back end of the model consists of a classifier composed of fully connected layers. After feature extraction, the data is flattened into a one-dimensional vector through a flattening layer, then input into a hidden layer containing 256 neurons, and connected to the output layer. The output layer contains the same number of neurons as the preset material categories (e.g., 3 nodes, corresponding to FR-4 fiberglass board, ceramic substrate, and aluminum substrate, respectively), and uses the Softmax function to calculate the probability distribution value for each material category. The training process of this CNN model is completed on an offline server. The training set contains thousands of images of labeled board surfaces taken under different lighting conditions. During training, the cross-entropy loss function is used to measure the difference between the predicted probability distribution and the true labels, and the Adam optimizer is used to iteratively update the network weight parameters based on the backpropagation algorithm until the loss function converges to a preset threshold.

[0047] During the system's online operation phase, the trained model parameters are embedded into the inference engine of the central control module. When a real-time image is input, the model outputs a probability vector, and the system logic selects the material category corresponding to the index with the highest probability value as the recognition result. This recognition result not only includes the material type but also associates the material's physical property parameters, particularly the material's stiffness coefficient (elastic modulus), through a database index, thus providing a physical basis for subsequent clamping force calculations. This process achieves a digital mapping from optical images to physical mechanical properties, solving the technical problem that traditional devices cannot perceive differences in board stiffness.

[0048] In the geometric topography analysis path, the system first processes the input 3D point cloud data. Perform region segmentation, remove background tracks and edge noise, and extract the effective point cloud set containing only the board surface. ,in Indicates the first The vertical height of each sampling point relative to the reference plane.

[0049] Regarding the actual thickness of the circuit board The calculation in this module does not rely on single-point measurements, but instead employs a statistical filtering algorithm. The system calculates the effective point cloud set. The arithmetic mean of all height values ​​is used to eliminate measurement noise caused by surface roughness. Actual thickness. The computational logic is expressed as follows:

[0050] ;

[0051] in, The total number of valid sampling points. The known reference height is the bearing surface of the track. This is for the board warpage. The calculation of warpage involves analyzing the extreme span of point cloud data in the vertical direction to evaluate the overall flatness of the board. Defined as the difference between the maximum and minimum height values ​​in the valid point cloud set, minus the thickness fluctuation tolerance under theoretically flat conditions. The calculation formula is:

[0052] ;

[0053] If the calculation yields If the flatness value exceeds the preset flatness threshold, the system will determine that the board has warping characteristics. At this time, the warping value will be transmitted as a key variable to the subsequent mechanical model for calculating additional correction components. Through the combination of deep learning classification and three-dimensional geometric operations, the system realizes three-dimensional holographic digital reconstruction of the material, thickness, and deformation of each board entering the mounting area.

[0054] This embodiment describes in detail how the central control module, through its built-in decision logic, converts the physical features acquired by the aforementioned vision module into specific operation instructions that drive the execution module.

[0055] The expert decision-making system architecture is logically divided into a knowledge base unit and an inference and computation engine. The knowledge base unit stores a pre-defined database of material mechanical properties, which establishes the material identification results. The mapping relationship between these parameters and mechanical control parameters. Specifically, for each material identifier (such as FR-4, ceramic, aluminum substrate, etc.), the database associates a set of attribute coefficients, including the material stiffness coefficient. Thickness influence weighting factor and corrective force weighting factor .

[0056] The inference engine is configured to receive the actual board thickness from the feature recognition module. Circuit board warpage and material identification results First, the inference calculation engine is based on... Retrieve the corresponding [item] from the knowledge base unit. , and Value. Subsequently, in order to determine the optimal normal clamping force applied to the second track edge. The inference engine performs calculations based on a multi-factor weighted mechanical model.

[0057] The mechanical model calculation aims to balance the contradiction between stable clamping and non-destructive protection, especially for thick boards ranging from 5mm to 8mm, where the risk of gravitational slippage and warping elastic potential energy must be overcome simultaneously. Target clamping force The formula for calculating (unit: Newton) is as follows:

[0058] ;

[0059] In the above formula, It characterizes the compressive force density required for a material under unit thickness and unit deformation, and its value is determined by the elastic modulus of the material. The measured board thickness value for the vision system. This reflects the frictional holding force component that needs to increase nonlinearly with increasing thickness; The measured warpage value of the vision system, item This reflects the corrective force component required to flatten the warped plate to the allowable flatness of the process. The basic preload constant set for the system is used to ensure the minimum static friction between the plate and the track contact surface under ideal flat and load-free conditions.

[0060] In addition to generating force control commands, the expert decision-making system also includes safety threshold limitation logic. The inference calculation engine will calculate the... The maximum permissible pressure of this material stored in the knowledge base Compare the results. If the calculation results would cause the contact surface pressure to exceed [a certain value], then [the calculation will be performed]. The system will force the output of a safety clamp value to prevent damage to the physical structure of the board due to algorithm calculation overflow.

[0061] Ultimately, the inference engine outputs two key control variables: the target clamping position. (based on (Calculated from the preset track reference width) and the target clamping torque (Depend on (Derived through lead screw transmission ratio conversion). These two variables are encapsulated into a control frame and sent to the servo controller of the drive execution module via the internal bus as reference inputs for closed-loop control.

[0062] This embodiment details how the drive execution module, under the command of the central control module, achieves high-precision flexible clamping through real-time torque monitoring. The core logic of the torque closed-loop feedback control system lies in constructing a negative feedback adjustment loop based on torque error. This loop is configured to eliminate nonlinear friction interference and inertial overshoot in the mechanical transmission system, ensuring that the actual physical pressure applied to the edge of the board strictly approximates the target clamping force calculated by the expert decision system. .

[0063] First, the motion controller within the central control module performs reference value conversion. Since the servo motor cannot directly control the contact force of the end effector, the controller first utilizes the physical parameters of the mechanical transmission system to convert the target clamping force... Converted to target torque reference value for motor output shaft This conversion follows the principles of energy conservation and mechanical efficiency loss, and its calculation formula is as follows:

[0064] ;

[0065] In the above formula, Lead of the ball screw (unit: meters). It is the overall mechanical efficiency coefficient of the transmission system (dimensionless, usually taken as 0.85-0.95). Let be the reduction ratio of the reduction mechanism. This formula establishes a quantitative mapping between the electrical control domain and the mechanical physical domain.

[0066] The control process employs a segmented hybrid control strategy, specifically divided into a rapid approach phase and a flexible locking phase:

[0067] During the rapid approach phase, the servo motor operates in position control mode. The controller drives the second track edge to move towards the target position at a relatively high speed. During this period, the force feedback module continuously monitors the real-time torque. The system has a set contact threshold torque. When detected A step increase and exceeding When the system determines that the track clamping edge has made physical contact with the board, it immediately triggers a control mode switch.

[0068] During the flexible locking phase, the servo motor seamlessly switches to torque control mode. At this time, the controller activates the PID control algorithm. This algorithm calculates the target torque reference value in real time. Compared with the actual torque value fed back by the sensor Deviation between The PID controller generates a voltage regulation command based on this deviation. This is used to adjust the current vector of the motor stator coils. The discretized control equation for PID control is expressed as:

[0069] ;

[0070] in, , , These are the proportional gain, integral gain, and differential gain coefficients, respectively. Through the cumulative effect of the integral term, the system can eliminate steady-state error, ensuring that the final applied clamping force remains constant. Horizontal; through the predictive action of the differential term, the system can suppress torque overshoot and prevent damage to the board edges due to inertial impact at the moment of contact.

[0071] Furthermore, this embodiment also introduces a position and torque safety window monitoring mechanism. During closed-loop adjustment, the controller simultaneously monitors the motor's position feedback. If the position encoder indicates that the track has not yet reached the expected plate width position, the real-time torque... That has been achieved The system will determine if there is a jam or discrepancy in the size of the plate, and immediately stop the motor output to trigger an alarm. Conversely, if the track position has exceeded the theoretical width but the torque has not yet been established, the system will determine that the plate is missing or there is an empty clamping abnormality. This dual verification mechanism ensures the safety of automated operations.

[0072] See attached document Figure 1 and attached Figure 2 This embodiment further discloses how the central control module actively intervenes in the dimensional deformation caused by changes in ambient temperature or heat accumulation of the circuit board itself during the mounting process, so as to prevent the circuit board from arching or warping due to obstructed thermal expansion.

[0073] The dynamic thermal compensation mechanism is constructed as a time-series-based, slowly varying signal feedforward control loop. A high-precision ambient temperature sensor is integrated into the operating area of ​​the track transfer module. This sensor is configured to acquire real-time temperature data of the mounting environment at a preset sampling frequency (e.g., 1Hz). .

[0074] The triggering basis of this mechanism lies in the material type determined by the aforementioned visual recognition module. The knowledge base within the central control module stores the linear thermal expansion coefficients (CLTE) of different substrate materials, denoted as... For example, for FR-4 epoxy resin glass cloth laminate, its Typically between 14 and 17 ppm / ℃; while for aluminum substrates, its That is approximately 23 ppm / ℃. The system is based on... Index out the specific board corresponding to the current board value.

[0075] When the clamping action is initially established, the system records the initial ambient temperature at that time. And the physical width of the board in the horizontal dimension at this time. As the placement process progresses, if the circuit board is subjected to heat radiation from the reflow preheating zone or an increase in the internal temperature of the equipment, the board will experience physical dimensional expansion. The central control module calculates the temperature difference relative to the initial moment in real time. Based on the principles of materials thermodynamics, the system calculates the required track width correction compensation. (Unit: millimeters). The calculation formula is expressed as:

[0076] ;

[0077] In conclusion Subsequently, the central control module no longer locks the second track edge to a single fixed geometric coordinate, but dynamically updates the position loop setpoint of the servo system. The updated target position... Indicates the initial clamping position With compensation amount Algebraic sum:

[0078] ;

[0079] The driver execution module is based on the updated This drives the second track edge to perform a micrometer-level retraction action. The direction of this action is strictly defined as the outward expansion direction along the track width. Through this real-time dynamic retraction, the system maintains effective lateral restraint on the board while reserving free space for microscopic thermal expansion imperceptible to the human eye. This eliminates the accumulation of compressive stress inside the board caused by thermal expansion and boundary constraints, ensuring that the board maintains ideal flatness throughout the entire mounting cycle.

[0080] See attached document Figure 3 This embodiment connects the aforementioned hardware architecture, visual perception, expert decision-making, and closed-loop control logic to form a complete and time-sensitive automated workflow. The system's workflow begins with the initialization and board entry detection phase. After the equipment is powered on, the central control module first executes a self-test program to verify the communication status and zero-point position of the dual-axis synchronous servo motor, industrial camera, line laser sensor, and torque sensor. When the upstream transmission equipment delivers the board to be mounted to the equipment's inlet buffer, the linear encoder detects the motion pulses of the conveyor belt, triggering the vision acquisition module to enter the working state. At this time, the first and second track edges of the track transmission module are maintained at a preset maximum opening position, which is configured to be greater than the maximum allowable board thickness (e.g., greater than 8mm) to ensure that the board can enter the field of view without interference.

[0081] The process then proceeds to the visual perception and parameter analysis stage. As the circuit board moves at a constant speed across the visual acquisition area on the conveyor belt, a high-resolution industrial camera and a line laser contour sensor are simultaneously triggered. The industrial camera captures the surface texture image of the circuit board, while the line laser sensor scans the cross-sectional contour of the circuit board. These two data streams are transmitted in real time to the deep learning processing unit of the central control module. The processing unit performs material classification and geometric measurement in parallel: on the one hand, it analyzes texture features through a convolutional neural network to determine the material properties of the circuit board (such as whether it is a high-rigidity ceramic substrate or a easily deformable thin FR-4); on the other hand, it accurately calculates the actual thickness value of the circuit board through point cloud analysis. and surface warpage If the system detects a board with a thickness between 5mm and 8mm, it will automatically mark the board as an overloaded object.

[0082] Based on the above analysis results, the process enters the decision calculation stage. The central control module invokes the built-in expert system algorithm, combined with the identified material stiffness coefficient. The measured actual thickness and warpage data are substituted into the aforementioned multi-factor weighted formula to calculate the optimal target clamping force for this single board. Simultaneously, the system plans the self-calibrated travel position of the second track edge based on the actual width of the board. This process achieves a parameterized mapping from what is seen to what is calculated, ensuring the uniqueness and specificity of the control strategy.

[0083] Next, the dynamic clamping and force closed-loop control phase is executed. The central control module sends motion commands to the drive execution module. The servo motor drives the second track edge to approach the board at high speed to shorten non-operation time; when the second track edge reaches a preset safe distance (e.g., 1mm) from the edge of the board, the system automatically switches to a low-speed creep mode to prevent inertial impact. When the second track edge contacts the edge of the board, the torque value monitored by the force feedback module changes abruptly, and the system immediately activates the PID torque control loop. The controller dynamically adjusts the motor current so that the actual applied clamping force rises smoothly and is stably locked at the calculated value. Value. For boards with warping, the clamping force includes a pressure component sufficient to correct the deformation; for heavy boards with a thickness of 8mm, the clamping force includes a frictional support component sufficient to overcome gravitational slippage.

[0084] After the clamping state is established, the placement operation and status monitoring phase begins. The system sends a ready signal to the SMT placement head, which then begins picking up and placing components onto the board. During this period, the dynamic thermal compensation mechanism continues to operate. The ambient temperature sensor samples in real time. If a change in ambient temperature is detected that causes a microscopic thermal expansion trend in the board, the central control module will drive the second track edge to perform a micrometer-level retraction action based on the thermal expansion coefficient of the material. This action releases the planar compressive stress inside the board while maintaining the effectiveness of lateral limiting, preventing the board from warping due to heat during the placement process.

[0085] Finally, once all components are mounted, the system enters the release and board ejection stage. The central control module receives the mounting completion signal and controls the servo motor to reverse, driving the second track edge to quickly retract to its initial maximum opening position, releasing the physical constraints on the board. The track conveyor belt then restarts, transporting the mounted PCBA board to downstream inspection or reflow soldering equipment. Simultaneously, the system resets the vision acquisition module, preparing to receive the next board. The entire process achieves adaptive and flexible handling of continuously arriving boards of different specifications without human intervention.

[0086] This embodiment further discloses an auxiliary reinforcement scheme to prevent the central area of ​​the circuit board from sagging due to its own weight, specifically for the special working conditions of ultra-thick (e.g., 5mm to 8mm) or large-size wide circuit boards mentioned in the previous embodiments.

[0087] In this embodiment, a bottom auxiliary support module is integrated in the lower space between the first and second track edges of the track transmission module. This module is not always active for all work objects, but is configured as a condition-triggered interventional subsystem. Its physical components include a lifting platform that moves along the vertical direction (Z-axis) and a flexible pin array mounted on the platform. The lifting platform is driven by a precision linear motor or electric actuator, possessing micron-level position control resolution and torque sensing capability. The contact tips of the flexible pin array are made of a soft polymer material with electrostatic dissipation (ESD) properties to prevent scratching the bottom surface of the circuit board or damaging the mounted bottom components during contact.

[0088] The intervention logic of the bottom auxiliary support module is determined by the central control module based on the data acquired by the vision acquisition module. This is when the vision system analyzes the actual thickness value of the board. When the load exceeds the preset heavy load threshold (e.g., 5mm), or when the calculated aspect ratio of the board shows that its span exceeds the physical limit of maintaining flatness by edge clamping, the central control module will generate an auxiliary support activation command.

[0089] At the execution level, the auxiliary support action and the lateral edge clamping action are synchronized through a collaborative control strategy. When the drive execution module controls the second track edge to complete the initial positioning of the board edge but has not yet reached the final locking torque, the bottom auxiliary support module initiates the Z-axis feed action. To prevent over-lifting that could cause the board to arch upwards (reverse deformation), the lifting platform adopts a contact-stop force control mode. The system monitors the load current of the Z-axis driver in real time. When the pin array contacts the bottom surface of the board and generates a support reaction force sufficient to balance the gravity component of the board's central area, the driver immediately locks the current height position.

[0090] At this point, the mechanical constraint state of the circuit board changes from a simply supported beam model constrained only by the two side edges to a composite constraint model constrained by clamping from both sides and multiple evenly distributed supports at the bottom. This composite support structure eliminates the elastic deflection of the circuit board due to its own weight, ensuring that the upper surface of the circuit board maintains a strict horizontal reference throughout the entire mounting working area. After the mounting operation is completed, the lifting platform performs a descent and retraction action before the lateral clamping mechanism, ensuring that the circuit board will not slide or shift due to the loss of lateral friction at the moment of release. This auxiliary mechanism ensures that even thick circuit boards with a large self-weight can meet the coplanarity requirements for mounting high-density, small-pitch components.

Claims

1. A machine vision-based adaptive clamping system for SMT placement boards, characterized in that, include: A track transport module is used to carry and transport the board to be mounted. The track transport module includes a first track edge and a second track edge arranged parallel to each other. The second track edge is configured to move relative to the first track edge in a direction perpendicular to the board transport direction. The drive execution module is mechanically connected to the second track edge and is used to drive the second track edge to generate linear displacement to adjust the track spacing. A visual acquisition module is located in the entrance area of ​​the track transmission module and is used to acquire two-dimensional surface image data and three-dimensional point cloud data of the board to be mounted entering the system. A force feedback module, associated with the drive execution module, is used to monitor the actual drive torque output by the drive execution module in real time. The central control module is communicatively connected to the drive execution module, the vision acquisition module, and the force feedback module, respectively, and is configured to perform the following operations: Receive the two-dimensional surface image data and the three-dimensional point cloud data, and parse the material type, actual thickness and warpage characteristics of the board to be mounted accordingly; A multi-factor weighted calculation model is used to calculate the target clamping force for the board to be mounted. The multi-factor weighted calculation model is configured as follows: indexing the corresponding material stiffness coefficient from a pre-stored mechanical property database according to the material type, multiplying the actual thickness by a thickness influence weighting factor to obtain the friction holding force component, and multiplying the warpage feature by a correction force weighting factor to obtain the correction force component; the target clamping force is calculated as the product of the sum of the friction holding force component and the correction force component and the material stiffness coefficient, plus a preset basic preload constant; The target clamping force is generated and a control command is sent to the drive execution module. The feedback signal from the force feedback module is used to perform closed-loop control on the second track edge to apply the target clamping force.

2. The machine vision-based adaptive clamping system for SMT placement boards according to claim 1, characterized in that, The visual acquisition module includes: A high-resolution industrial camera, whose optical axis is perpendicular to the transmission plane formed by the first track edge and the second track edge, is configured to acquire the two-dimensional surface image data containing the surface texture features of the board. A line laser profile sensor, combined with a linear encoder, is configured to perform cross-sectional profile scanning on a moving board to acquire the three-dimensional point cloud data reflecting the height distribution on the board's surface.

3. The machine vision-based adaptive clamping system for SMT placement boards according to claim 1, characterized in that, The drive execution module includes a dual-axis synchronous servo motor group and two sets of parallel precision ball screw assemblies. The dual-axis synchronous servo motor sets are respectively installed at the inlet and outlet ends of the track transmission module, and drive the two sets of precision ball screw assemblies respectively, thereby driving the two ends of the second track to move synchronously. The non-contact sidewalls of the first and second track edges are integrated with reinforcing ribs distributed along the length direction, and the contact surface of the second track edge is made of a high-hardness alloy material.

4. The machine vision-based adaptive clamping system for SMT placement boards according to claim 2, characterized in that, The central control module includes a feature recognition unit, which is configured to: The two-dimensional surface image data is processed using a convolutional neural network model to extract the surface texture features of the board and output the material type. The three-dimensional point cloud data is segmented into regions to extract an effective point cloud set. The average height of the effective point cloud set is calculated to determine the actual thickness. The difference between the maximum and minimum height values ​​in the effective point cloud set is calculated to determine the warping feature.

5. The machine vision-based adaptive clamping system for SMT placement boards according to claim 1, characterized in that, The control process of the central control module over the drive execution module includes a segmented hybrid control strategy: During the rapid approach phase, the drive execution module is controlled to operate in position control mode until the actual drive torque detected by the force feedback module exceeds the preset contact threshold. After the contact detection is triggered, the system switches to the flexible locking stage and controls the drive execution module to work in torque control mode. The motor current is adjusted through a PID adjustment algorithm so that the actual drive torque is stabilized at the reference value corresponding to the target clamping force.

6. The machine vision-based adaptive clamping system for SMT placement boards according to claim 1, characterized in that, It also includes an ambient temperature sensor for real-time monitoring of the mounting environment temperature; The central control module also includes a dynamic thermal compensation mechanism, which is configured as follows: Determine the corresponding coefficient of thermal expansion based on the material type; Based on the temperature difference data collected by the ambient temperature sensor, the coefficient of thermal expansion, and the initial width of the board, the thermal expansion compensation amount of the board is calculated. During the mounting process, the target position setting point of the second track edge is dynamically updated according to the thermal expansion compensation amount, driving the second track edge to perform an outward retraction action.

7. The machine vision-based adaptive clamping system for SMT placement boards according to claim 1, characterized in that, The system includes a bottom auxiliary support module, which is disposed in the space below the first track edge and the second track edge, and includes a lifting platform and a flexible pin array. The central control module is configured as follows: When the actual thickness obtained from the analysis exceeds the preset heavy load threshold, the bottom auxiliary support module is controlled to rise until the flexible pin array contacts the bottom surface of the board to be mounted and provides support reaction force, thereby eliminating the gravity-induced sagging deformation in the central area of ​​the board.

8. The machine vision-based adaptive clamping system for SMT placement boards according to claim 5, characterized in that, The force feedback module includes a dynamic torque sensor connected in series between the output shaft of the drive execution module and the transmission mechanism; Alternatively, the force feedback module is integrated inside the servo driver and calculates the output torque by monitoring the current loop load component of the motor; The central control module is also configured to perform security window monitoring: If the torque reaches the target value when the position fed back by the position encoder has not reached the preset board width, it is determined that the material jamming is abnormal and the drive is stopped.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the operational steps performed by the central control module in the system as described in any one of claims 1-8.

Images