A flexible circuit board mounting precision visual guidance and correction system

Abstract

The present application belongs to the technical field of electronic component assembly manufacturing, and relates to a flexible circuit board mounting precision visual guidance and correction system, which comprises: a front positioning module that obtains an instruction to drive a mounting head to hover and generates a trigger signal; a physical detection module that applies a physical air pulse to a target flexible circuit board; a dynamic response acquisition module that monitors exhaust to trigger the shooting of a stress deformation image to generate a dynamic image sequence; a mechanical parameter identification module that inputs the image sequence into a neural network to analyze a local mechanical parameter set; a safety check module that compares a stiffness coefficient with a preset lower limit to generate a trajectory modeling enable signal; a predicted trajectory generation module that inputs the set and a preset descending trajectory into a dynamics model to obtain a position prediction trajectory; and a feedforward compensation control module that deconstructs the prediction trajectory to generate a control instruction that is injected into a driver to perform compensation. The present application solves the problem of how to identify and predict the dynamic mechanical properties of a flexible substrate online, and effectively compensates for dynamic deformation errors in the mounting process.

Description

Technical Field

[0001] This invention belongs to the technical field of electronic component manufacturing, and relates to a visual guidance and correction system for flexible circuit board mounting accuracy. Background Technology

[0002] In modern electronics manufacturing, flexible printed circuit boards (PCBs) are widely used in many smart terminals and precision instruments due to their thinness and flexibility. Compared with traditional rigid circuit boards, flexible substrates have lower mechanical stiffness and are more susceptible to localized deformation due to external disturbances during automated high-speed component placement. In particular, when the placement head descends at high speed, the airflow from the nozzles causes aerodynamic impact on the substrate. This impact can easily lead to micron-level dynamic positional displacement of the target pads, resulting in placement misalignment or rotation defects and affecting overall placement accuracy.

[0003] Currently, the industry mainly uses machine vision static position compensation technology and enhanced mechanical fixation solutions to address the aforementioned aerodynamic impact and localized deformation issues. Static vision compensation technology captures images of the target pads and reference marks before the placement head descends, calculating the static position and angular deviation of the substrate in a two-dimensional plane. The motion control system then uses this information to make a one-time correction to the target coordinates of the placement head. Enhanced mechanical fixation solutions, on the other hand, use complex vacuum adsorption platforms or added pressing clamps to fix the flexible circuit board, aiming to physically limit its dynamic deformation during the placement process.

[0004] Existing static visual compensation techniques acquire compensation values ​​based on static presets, making it difficult to anticipate and offset dynamic deformation errors induced by the downward airflow of the placement head. Therefore, they exhibit limitations in addressing dynamic disturbances during high-speed placement. Meanwhile, while reinforced mechanical fixation can reduce macroscopic substrate displacement to some extent, continuous mechanical stress poses a risk of damaging the substrate material. Furthermore, for the small localized area directly beneath the placement head, conventional suction or clamping forces are often insufficient to effectively resist sudden airflow impacts, leaving microscopic vibrations. In addition, given the objective differences in stiffness and damping within flexible circuit boards, fixed compensation models struggle to achieve universal and accurate error compensation. Summary of the Invention

[0005] To address the aforementioned problems, this invention provides a visual guidance and correction system for flexible circuit board mounting accuracy.

[0006] A visual guidance and correction system for flexible circuit board mounting accuracy includes: The front positioning module is used to obtain the placement instructions, extract the global coordinates of the target pad, drive the placement head to hover above the target flexible circuit board within a preset height range, and generate a hovering trigger signal. The physical detection module is used to receive hovering trigger signals, activate the air pulse piezoelectric nozzle unit, and apply physical air pulses to the target flexible circuit board; The dynamic response acquisition module is used to monitor the exhaust action of physical air pulses, trigger the image acquisition array to capture images of the stress deformation of the target flexible circuit board, and generate a dynamic deformation image sequence. The mechanical parameter identification module is used to acquire a preset physical information neural network. The dynamic deformation image sequence is input into the physical information neural network for analysis, and the local mechanical parameter set is output. The safety verification module is used to extract the equivalent dynamic stiffness coefficient from the local mechanical parameter set, compare the equivalent dynamic stiffness coefficient with the preset stiffness safety lower limit value, and generate a trajectory modeling enable signal. The predicted trajectory generation module is used to receive the trajectory modeling enable signal, obtain the preset spatial dynamics model and the preset descent trajectory, input the local mechanical parameter set and the preset descent trajectory into the spatial dynamics model for solution, and obtain the full time domain position prediction trajectory. The feedforward compensation control module is used to deconstruct the full-time-domain position prediction trajectory, generate a feedforward control command sequence, and inject the feedforward control command sequence into the horizontal servo driver to perform alignment compensation actions.

[0007] In a further embodiment of the present invention, the front positioning module is specifically used for: Extract the global coordinates of the target pad from the placement command; Monitor the dynamic Z-axis position data during the downward movement of the placement head; Compare the dynamic Z-axis position data with the preset height range; A hovering trigger signal is generated when the dynamic Z-axis position data reaches the upper boundary value of the preset height range.

[0008] In a further embodiment of the present invention, the physical detection module is specifically used for: Extract the hovering trigger signal and close the underlying hardware drive circuit of the air pulse piezoelectric nozzle unit; Control the underlying hardware driver circuit to release the pulse drive signal; A pulsed drive signal is used to drive a piezoelectric ceramic array to undergo directional deformation. Physical air pulses are expelled by directional deformation and extrusion of the built-in air cavity, and then applied vertically to the target flexible circuit board.

[0009] In a further embodiment of the present invention, the dynamic response acquisition module is specifically used for: The rising edge waveform of the intercepted pulse drive signal; A hardware-triggered synchronization signal is generated using the rising edge waveform. Input the hardware trigger synchronization signal into the image acquisition array to activate the continuous exposure acquisition mode; Images of the target flexible circuit board are captured continuously using a continuous exposure acquisition mode, and the consecutive image frames are stacked according to the timestamp to generate a dynamic deformation image sequence.

[0010] In a further embodiment of the present invention, the mechanical parameter identification module is specifically used for: Extract feature point clouds from each consecutive image frame of a dynamic deformation image sequence; Track the cross-frame motion trajectory of feature point clouds and generate a pixel displacement mapping field with timestamp information; The pixel displacement mapping field is input into the physical information neural network, and the deformation law is extracted through the data-driven loss term and the physical law constraint loss term of the physical information neural network. The equivalent dynamic stiffness coefficient and equivalent damping coefficient of the target flexible circuit board are solved in reverse, and the set of local mechanical parameters is output by combination.

[0011] A further aspect of this invention involves inversely solving for the equivalent dynamic stiffness coefficient and equivalent damping coefficient of the target flexible circuit board, specifically including: The data-driven loss term and the physical law constraint loss term are used to acquire physical information from the neural network. Obtain the preset balance weight hyperparameters; The weighted physical constraint term is obtained by multiplying the physical constraint loss term with the balance weight hyperparameter. The total loss function value is obtained by summing the data-driven loss term and the weighted physical constraint term. The internal parameters are iteratively adjusted with the goal of minimizing the total loss function value until convergence is obtained to obtain the equivalent dynamic stiffness coefficient and the equivalent damping coefficient.

[0012] In a further embodiment of the present invention, the security verification module is specifically used for: Extract the equivalent dynamic stiffness coefficient from the local mechanical parameter set; Compare the numerical values ​​of the equivalent dynamic stiffness coefficient and the lower limit of stiffness safety; If the equivalent dynamic stiffness coefficient is less than the lower limit of stiffness safety, a dangerous shutdown warning sign will be output. If the equivalent dynamic stiffness coefficient is greater than or equal to the lower limit of stiffness safety, generate a trajectory modeling enable signal.

[0013] In a further embodiment of the present invention, the trajectory prediction generation module is specifically used for: Extract the equivalent dynamic stiffness coefficient and equivalent damping coefficient from the local mechanical parameter set; The equivalent dynamic stiffness coefficient and equivalent damping coefficient are transformed into equivalent horizontal plane mechanical parameters through a preset mapping transformation matrix, and the equivalent horizontal plane mechanical parameters are used as constitutive parameters of the spatial dynamics model to construct the local vibration equation in the horizontal plane. Separate the temporal descent velocity and temporal acceleration variables from the preset descent trajectory; By substituting the time-series downward velocity and time-series acceleration variables into the excitation terms of the local vibration equation and introducing the in-plane geometric coupling effect for solution, the full-time domain position prediction trajectory is obtained.

[0014] A further aspect of this invention involves substituting the time-series downward velocity and time-series acceleration variables into the excitation terms of the local vibration equation, and introducing in-plane geometric coupling effects for solution calculation. Specifically, this includes: Obtain the equivalent vibration mass of the target flexible circuit board; The time-series descent velocity and time-series acceleration variables are mapped to the equivalent in-plane coupled tensile force term induced by aerodynamic indentation; Using the two-dimensional horizontal plane displacement vector to be solved as a variable, the inertial force term is constructed from the equivalent vibrating mass and the second time derivative of the two-dimensional horizontal plane displacement vector; the damping force term is constructed from the horizontal damping coefficient in the equivalent horizontal plane mechanical parameters and the first time derivative of the two-dimensional horizontal plane displacement vector; and the elastic force term is constructed from the horizontal plane dynamic stiffness matrix in the equivalent horizontal plane mechanical parameters and the two-dimensional horizontal plane displacement vector. Let the sum of the inertial force term, damping force term, and elastic force term equal the equivalent horizontal plane coupled tension term, and construct a second-order differential equation; solve the second-order differential equation to obtain the time-varying sequence of the two-dimensional horizontal plane displacement vector, which serves as the full-time domain position prediction trajectory.

[0015] In a further embodiment of the present invention, the feedforward compensation control module is specifically used for: The trajectory decomposition algorithm is applied to deconstruct the full-time domain position prediction trajectory, and the X-axis position deviation sequence and the Y-axis position deviation sequence are solved separately. Based on the preset gain coefficient, the X-axis position deviation sequence and the Y-axis position deviation sequence are proportionally amplified, and after rounding and remainder accumulation, discrete integer digital instruction values ​​are generated. The discrete integer digital instruction values ​​are encoded into discrete electronic control information packets to generate a feedforward control instruction sequence. The feedforward control command sequence is fed forward into the horizontal servo driver to synthesize the XY plane compensation trajectory curve.

[0016] In summary, the present invention has the following beneficial technical effects: 1. By actively exciting a flexible circuit board with physical air pulses and simultaneously acquiring deformation images, and then using a physical information neural network for inverse solving, the observed two-dimensional optical image sequence can be converted into the equivalent dynamic stiffness and damping coefficient of the substrate's local area. This processing method, by introducing constraints from the physical equations of thin-film vibration, reduces the dependence on large-scale labeled data and achieves online quantification of the local mechanical properties of the flexible circuit board, thus providing reliable basic data for the subsequent construction of a dynamic mounting error prediction model.

[0017] 2. By updating the local mechanical parameters identified online to the spatial dynamics model and combining them with the preset descent trajectory of the placement head, a full-time-domain position prediction trajectory is calculated. This is then used to generate feedforward control commands during the placement head's hovering gap. This feedforward compensation mechanism based on mechanical characteristics can offset the estimated dynamic offset during the synchronous phase of the descent action, reducing the response delay in traditional feedback control. This helps reduce end-position placement deviations caused by aerodynamic impact and flexible substrate coupling, improving placement accuracy in complex dynamic operating environments.

[0018] 3. By introducing safety verification logic before precision trajectory modeling, the measured equivalent dynamic stiffness coefficient is compared with the preset safety lower limit value, which can assess the local physical load-bearing state of the substrate before the mounting operation is executed. When a stiffness feature below the threshold is identified, the operation is stopped in advance and an alarm is issued. This helps to detect macroscopic mounting defects such as vacuum adsorption failure, severe local warping or wrinkles in a timely manner, reducing the risk of component damage or mounting failure due to continued operation with poor substrate physical condition, thereby ensuring the overall stability of the mounting process. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. The drawings are used to provide a further understanding of the present invention.

[0020] Figure 1 This is a schematic diagram of the framework in the embodiments of this application.

[0021] Figure 2 This is a flowchart illustrating an embodiment of this application.

[0022] Figure 3 This is a graph showing the monitoring of the Z-axis downward movement of the mounting head and the response curve of its hovering position in an embodiment of the present invention.

[0023] Figure 4 This is a comparison chart of the forced deformation observation and the reverse solution curve of the physical network in an embodiment of the present invention.

[0024] Figure 5This is a full-time-domain position prediction and compensation trajectory diagram under aerodynamic impact in an embodiment of the present invention. Detailed Implementation

[0025] The following is in conjunction with the appendix Figures 1-5 A preferred description of the present invention is provided below.

[0026] See attached document Figures 1-5 This invention proposes a visual guidance and correction system for flexible circuit board mounting accuracy, comprising the following modules: The front positioning module is used to obtain the placement instructions, extract the global coordinates of the target pad, drive the placement head to hover above the target flexible circuit board within a preset height range, and generate a hovering trigger signal. The physical detection module is used to receive hovering trigger signals, activate the air pulse piezoelectric nozzle unit, and apply physical air pulses to the target flexible circuit board; The dynamic response acquisition module is used to monitor the exhaust action of physical air pulses, trigger the image acquisition array to capture images of the stress deformation of the target flexible circuit board, and generate a dynamic deformation image sequence. The mechanical parameter identification module is used to acquire a preset physical information neural network. The dynamic deformation image sequence is input into the physical information neural network for analysis, and the local mechanical parameter set is output. The safety verification module is used to extract the equivalent dynamic stiffness coefficient from the local mechanical parameter set, compare the equivalent dynamic stiffness coefficient with the preset stiffness safety lower limit value, and generate a trajectory modeling enable signal. The predicted trajectory generation module is used to receive the trajectory modeling enable signal, obtain the preset spatial dynamics model and the preset descent trajectory, input the local mechanical parameter set and the preset descent trajectory into the spatial dynamics model for solution, and obtain the full time domain position prediction trajectory. The feedforward compensation control module is used to deconstruct the full-time-domain position prediction trajectory, generate a feedforward control command sequence, and inject the feedforward control command sequence into the horizontal servo driver to perform alignment compensation actions.

[0027] In one embodiment of the present invention, the front positioning module is used to perform the following steps: Extract the global coordinates of the target pads from the placement command; monitor the dynamic Z-axis position data during the downward movement of the placement head; compare the dynamic Z-axis position data with the preset height range; generate a hover trigger signal when the dynamic Z-axis position data reaches the upper boundary value of the preset height range.

[0028] Specifically, the placement head motion control system receives placement instructions from the main control computing unit via an industrial bus interface. The main control computing unit is an industrial computer or embedded controller that directs the entire placement equipment, generating work instructions based on a pre-loaded circuit board layout file and production task list. The placement instruction is a structured data packet transmitted via industrial communication protocols such as Controller Area Network (CLAN) bus or Ethernet control automation technology, encapsulating all parameters required to guide a single placement action. This instruction includes data such as the identifier of the component to be placed, the global coordinates of the target pad, and the rotation angle. The system immediately parses the placement instruction, extracting the global coordinates of the target pad corresponding to the component to be placed, denoted as... .

[0029] It should be noted that the global coordinates of the target pads are two-dimensional Cartesian coordinates defined based on the inherent mechanical coordinate system of the entire device, with a numerical accuracy down to the micrometer level. Subsequently, the motion control system instructs the horizontal axis of the placement head... shaft and The axis servo system drives the placement head to move directly above the target pad's global coordinates.

[0030] As horizontal positioning is completed, the system activates the mounting head's built-in... The axis servo closed-loop motor is instructed to move downwards from its current safe height. The axis servo driver continuously reads the number of pulses fed back by the motor encoder, and the controller converts this into dynamic values. Axis position data.

[0031] In this embodiment, dynamic Axis position data is The shaft motor encoder continuously generates real-time position feedback values ​​during operation, with an update frequency typically no less than 1kHz, ensuring real-time position monitoring. This data stream is fed in real-time into a digital comparator logic unit to dynamically... The axis position data and the upper boundary value of the system's preset height descent range Perform continuous comparisons.

[0032] Specifically, the preset height drop range is set based on a trade-off in engineering practice. The upper boundary value is set to 3.0mm to 6.0mm to avoid substrate warping and ensure response time; its lower boundary value... This defines the termination plane for active physical detection and fine measurement, which is usually set between 0.3mm and 0.8mm. This height is very close to physical contact and can maximize the signal-to-noise ratio of the detection signal.

[0033] Once dynamic The axis position data first reaches or falls below the upper boundary value of the preset height exploration range. This comparison logic is triggered, indicating that the front end of the placement head has entered the starting position for subsequent fine-tuning operations. At this time, the logic unit immediately generates a hover trigger signal, which is a standard transistor-to-transistor logic level signal, serving as the hardware synchronization reference for subsequent startup. This signal is output as a high-level pulse to the designated input / output port of the control system, and simultaneously to... The axis servo driver issues a position holding command, causing the placement head to hover at a preset height above the target flexible circuit board.

[0034] In one example of the present invention, the main control computing unit issues a placement instruction, which includes the global coordinates of the target pad as follows: , The system's preset height drop range is... That is, the upper boundary value The thickness is 5.0mm. The motion control system first drives the placement head to... Move to coordinates on the plane Directly above. After positioning, The axis servo closed-loop motor starts and descends from a safe height of 20.0 mm. The controller monitors the dynamic data transmitted back from the encoder in real time. The Z-axis position data stream is sequentially 12.5mm, 9.8mm, 7.3mm, 5.1mm, 4.9mm, ... When the digital comparator input receives the value 4.9mm, the current Z-axis position data first meets the set condition of being less than or equal to 5.0mm. This logical state transition immediately triggers two actions: the controller sends a position hold command to the Z-axis servo driver, locking it in the current position, allowing the placement head to hover stably at a height of 4.9mm above the flexible circuit board surface; the control system's specific output pin level jumps from 0V to 5V, generating a valid hover trigger signal to notify the subsequent air pulse piezoelectric nozzle unit to enter standby mode. See also Figure 3 , Figure 3 The following diagram shows the Z-axis downward movement monitoring and hovering position response curves of the mounting head in this embodiment.

[0035] In one embodiment of the present invention, the physical detection module is configured to perform the following steps: Extract the hovering trigger signal and close the underlying hardware drive circuit of the air pulse piezoelectric nozzle unit; control the underlying hardware drive circuit to release the pulse drive signal; use the pulse drive signal to drive the piezoelectric ceramic array to undergo directional deformation; squeeze the built-in air cavity through directional deformation to discharge the physical air pulse, and apply the physical air pulse vertically to the target flexible circuit board.

[0036] Specifically, the control module continuously polls or monitors the hardware input port bound to the hover trigger signal via interrupts. When the voltage at this port rises from a logic low level to a logic high level, the control module confirms receipt of the hover trigger signal. This signal capture serves as a logic trigger condition, immediately closing the underlying hardware drive circuit of the air pulse piezoelectric nozzle unit. The air pulse piezoelectric nozzle unit is a microelectromechanical system (MEMS) device that integrates a piezoelectric ceramic actuator, a microfluidic gas chamber, and a nozzle structure, enabling the efficient conversion of electrical energy into the kinetic energy of trace amounts of gas.

[0037] The underlying hardware driver circuit is a high-voltage amplifier module. Its function is to convert low-voltage logic signals from the control module, such as transistor-to-transistor logic levels or complementary metal-oxide-semiconductor logic levels, into high-voltage pulses required to drive the piezoelectric ceramic. This closing action is specifically manifested in the control module sending a high-level signal to the enable pin of the underlying hardware driver circuit, thereby switching the driver circuit from standby mode to operating mode.

[0038] Once in working mode, the underlying hardware driver circuit generates and releases pulse drive signals in real time according to preset pulse parameters.

[0039] Furthermore, the transient characteristics of the pulse-driven signal are reflected in its extremely short pulse duration and extremely fast voltage change rate, rather than continuous high-frequency oscillations; its typical voltage amplitude ranges from 80V to 200V, and its pulse width ranges from 20μs to 150μs. The timing and amplitude of this signal directly determine the energy of the subsequently generated physical air pulse. This signal is essentially a high-voltage, narrow-pulse-width square wave. Its rapid voltage rise and fall edges force the piezoelectric ceramic array connected to the output of the underlying hardware drive circuit to undergo instantaneous and precise directional deformation. The directional deformation of the piezoelectric array compresses the volume of the micro-air cavity. The air within the cavity is expelled through micro-orifice nozzles, forming a physical air pulse with specific momentum.

[0040] In this embodiment, the air volume carried by the physical air pulse is typically between 0.2 μL and 1.5 μL. The impact force of the air pulse is configured to induce only microscopic deformation of the substrate surface without causing physical damage or macroscopic displacement. The air pulse piezoelectric nozzle unit is fixed to the side of the placement head and mechanically calibrated, with its jet axis aligned with the placement head. The axes are parallel. Therefore, when the placement head is hovering at the target position, the physical air pulse is applied vertically to the surface of the target flexible circuit board. The specific area of ​​action is the annular substrate area within a preset range around the target pads, rather than the pads themselves.

[0041] It should be noted that the substrate within the preset range surrounding the target pad specifically refers to an annular area with a radius between 0.8mm and 2.5mm centered on the geometric center of the pad. This area is selected to avoid direct impact on the pad that may be coated with solder paste, while effectively reflecting the local mechanical properties of the substrate in the area where the pad is located.

[0042] In one example of the invention, after the hover trigger signal is generated, the 5V high-level pulse is sent to the control interface of the air pulse piezoelectric nozzle unit and captured by its internal microcontroller. The microcontroller immediately outputs a 3.3V high-level signal to the enable pin of the DRV2700 piezoelectric drive integrated circuit, activating the drive circuit. Then, the microcontroller generates and sends a 5V logic pulse to the input of the piezoelectric drive integrated circuit through its pulse width modulation output channel. Assuming the gain of the drive circuit is set to 30, it will immediately generate a high-voltage square wave with an amplitude of 150V and a pulse width consistent with the input logic pulse, for example, 80μs, at the output. This 150V voltage acts on the piezoelectric ceramic array, causing it to undergo a rapid bending deformation, expelling approximately 0.8μL of compressed air from the built-in air chamber through a nozzle with a diameter of 0.1mm. Since the mounting head has previously hovered at the coordinate... Directly above, the emitted physical air pulse strikes the target flexible circuit board vertically downwards. Within the central annular region, a localized downward concave deformation is induced in the substrate to facilitate image acquisition by external optical equipment.

[0043] In one embodiment of the present invention, the dynamic response acquisition module is used to perform the following steps: The rising edge waveform of the pulse drive signal is intercepted; a hardware trigger synchronization signal is generated using the rising edge waveform; the hardware trigger synchronization signal is input into the image acquisition array to activate the continuous exposure acquisition mode; images of the target flexible circuit board are continuously captured according to the continuous exposure acquisition mode, and the continuous image frames are stacked according to the timestamp to generate a dynamic deformation image sequence.

[0044] Specifically, to monitor the air pulse exhaust action and synchronously generate a dynamic deformation image sequence recording the microscopic deformation of the substrate under stress, the system executes the following electromechanical synchronization control process: A high-speed level monitoring circuit is connected in parallel to the control signal line of the air pulse piezoelectric nozzle unit, specifically used to acquire the pulse drive signal driving the nozzle action. The high-speed level monitoring circuit can be a hardware circuit based on a high-speed comparator and a monostable multivibrator, used to detect the edge of the pulse signal and output a shaped hardware trigger synchronization signal.

[0045] When the pulse drive signal exhibits a rising edge waveform from low to high, the edge detection logic unit within the high-speed level monitoring circuit immediately identifies it and uses this rising edge waveform to instantaneously generate a standard hardware trigger synchronization signal. The hardware trigger synchronization signal is a common external synchronization signal in industrial vision, typically a square wave pulse with transistor-to-transistor logic levels. Its rising or falling edge is used as the starting point for camera sensor exposure. This synchronization signal undergoes no software processing and is directly input via a coaxial cable to the external trigger port of the image acquisition array fixed to the side of the mounting head.

[0046] In this embodiment, the image acquisition array is an industrial camera module that integrates a high-speed, high-sensitivity image sensor. Its core is a global shutter complementary metal-oxide-semiconductor sensor. This type of sensor can ensure that there is no image blur or rolling shutter effect when shooting high-speed moving objects. In order to capture millisecond-level deformation events caused by air pulses, the predetermined frame rate of the array is usually set in the range of 1000fps to 5000fps, and the exposure time is set between 50μs and 200μs depending on the ambient light conditions.

[0047] Upon receiving the pulse, the control timing circuit of the image acquisition array is immediately interrupted by hardware, triggering the camera to exit sleep mode and activate continuous exposure acquisition mode. Specifically, continuous exposure acquisition mode defines the total number of frames that the camera needs to continuously capture after being triggered, as well as the time interval between frames. For example, it can be set to capture 15 frames of images continuously after being triggered.

[0048] Based on this exposure sequence, the image acquisition array begins to continuously capture images of the target pad and the substrate within a predetermined range surrounding it, which is under the influence of physical air pulses, at a predetermined frame rate. Each successfully exposed image is immediately transmitted to the onboard buffer and appended with a timestamp relative to the arrival time of the hardware trigger synchronization signal. After a predetermined number of image frames have been captured, the system integrates and stacks all the consecutive timestamped image frames in the buffer in chronological order, ultimately generating a dynamic deformation image sequence that characterizes the entire process of the substrate from stress, deformation to recovery. This dynamic deformation image sequence is a structured data set, which can be logically regarded as a three-dimensional data volume. Two dimensions are the pixel coordinates of the image, and the third dimension is the timestamp. It fully records the morphological evolution history of the target area within a small time window and stores it in high-speed memory for subsequent analysis.

[0049] In one example of this invention, at the instant the 150V, 80μs pulse drive signal output by the underlying hardware driver circuit is generated, its voltage jumps from 0V to 150V within 100ns. A high-speed level monitoring circuit connected in parallel to this line captures this rising edge waveform and immediately generates a 5V, 1μs pulse width hardware trigger synchronization signal. This signal is sent to the image acquisition array via the hardware line. Assume the camera is preset to capture 10 frames continuously at a frame rate of 2000fps after triggering. The moment the trigger signal is received is denoted as... The camera immediately activates its continuous exposure acquisition mode and begins image acquisition. The exposure center time of the first frame is... The second frame is And so on, until the exposure center time of the tenth frame is... Each frame is a record. The grayscale image of the preset area surrounding the coordinate point at the corresponding time. These 10 frames, each with its own timestamp, are packaged together to form a complete dynamic deformation image sequence. Features are extracted and displayed from this sequence, for example, in... and At that time, the deformation of the flexible circuit board substrate reached its maximum value, while In subsequent images, the substrate has largely recovered its flatness. This sequence file is stored in system memory, awaiting further analysis by the physical information neural network.

[0050] In one embodiment of the present invention, the mechanical parameter identification module is used to perform the following steps: Feature point clouds are extracted from each consecutive image frame of a dynamic deformation image sequence; the motion trajectory of the feature point clouds across frames is tracked to generate a pixel displacement mapping field with timestamp information; the pixel displacement mapping field is input into a physical information neural network, and the deformation law is extracted through the data-driven loss term and the physical law constraint loss term of the physical information neural network; the equivalent dynamic stiffness coefficient and equivalent damping coefficient of the target flexible circuit board are solved in reverse, and the set of local mechanical parameters is output by combination; the data-driven loss term and the physical law constraint loss term of the physical information neural network are obtained. Obtain the preset balance weight hyperparameters; multiply the physical law constraint loss term with the balance weight hyperparameters to obtain the weighted physical constraint term; sum the data-driven loss term with the weighted physical constraint term to obtain the total loss function value; iteratively adjust the internal parameters with the goal of minimizing the total loss function value until convergence is obtained to obtain the equivalent dynamic stiffness coefficient and the equivalent damping coefficient.

[0051] Specifically, the system loads the physical information neural network, pre-trained with dynamic physical knowledge, from memory into the memory of the computing unit and completes the initialization configuration of its operating environment. Then, the system reads the dynamic deformation image sequence generated in the previous implementation. For the first frame image in the dynamic deformation image sequence, i.e., the reference image before deformation, the system uses feature detection methods such as the Carnard-Lucas-Thomas algorithm to extract features from the image region of the substrate within a preset range surrounding the target pad. A feature point cloud composed of high-contrast, texture-stable pixels.

[0052] In this embodiment, the feature point cloud is a set of tracking markers virtually defined on the substrate surface, used to quantify the surface motion.

[0053] For each subsequent image frame in the sequence, the system uses optical flow to track the pixel coordinate changes of each feature point in the feature point cloud, thus forming a set of motion trajectories spanning all timestamps. By calculating the pixel offset of each feature point in each frame relative to its initial position in the reference frame, the system constructs a pixel displacement mapping field with timestamp information.

[0054] It should be noted that the pixel displacement mapping field is a discretized mathematical expression of the deformation process, providing direct and quantitative observational evidence for the neural network. This mapping field is formatted as a dataset containing spatial coordinates, timestamps, and corresponding displacement vectors. This dataset serves as the core input and is fed into a pre-configured physical information neural network. Specifically, the physical information neural network employs a multi-layer fully connected network architecture, with its input layer containing three neurons, each corresponding to a spatial coordinate. With timestamp The hidden layers are set to 4 to 6 layers, each containing 64 to 128 neurons, and use activation functions with high-order continuous differentiability, such as Tanh or Swish, to support subsequent partial differential operations; the output layer has 1 neuron, used to output the predicted vertical displacement in the corresponding spatiotemporal coordinates. During backpropagation and reverse solution, the model not only compares the predicted displacements using the loss function, but also... More crucially than the difference in observed displacement is the use of automatic differentiation techniques within deep learning frameworks to analyze the network output. Directly regarding input parameters The first and second time derivatives are obtained. The model uses these derivative results and the mechanical parameters to be identified as variables, and substitutes them into the partial differential equation describing the local vibration of the substrate to construct a physical loss term. The network weights and internal physical parameters are jointly optimized through the gradient descent algorithm until the total loss function fits and converges, so that the solution of the partial differential equation describing the local vibration of the substrate fits the observed displacement data.

[0055] When the fitting error converges to below the preset threshold, the inverse solution process ends. The parameter values ​​output by the model at this point are the equivalent dynamic stiffness coefficient and the associated equivalent damping coefficient of the target flexible circuit board in that local region.

[0056] Specifically, the equivalent dynamic stiffness coefficient is denoted as... The unit is It quantifies the ability of flexible circuit boards to resist dynamic deformation and is a key parameter characterizing their "rigidity" or "support"; the equivalent damping coefficient is denoted as... The unit is It quantifies how quickly a material dissipates energy during vibration, characterizing its toughness or ability to absorb impact.

[0057] Finally, the system encapsulates and combines the two solved absolute coefficients into a set of local mechanical parameters. This set of local mechanical parameters is a vector containing the physical absolute values ​​of stiffness and damping, which directly reflects the actual deformation resistance and energy dissipation capacity of the substrate in the target region, providing direct constitutive parameters for subsequent dynamic calculations.

[0058] The specific calculation formula is as follows: ; The analysis is based on consecutive image frames within the dynamic deformation image sequence. The physical information neural network is a special deep learning architecture whose loss function includes not only data-driven terms but also physical constraint terms. (Data-driven terms...) Used to measure the degree of agreement between the displacement predicted by the network and the pixel displacement mapping field obtained through image tracking. Physical constraint term. This is based on the partial differential equations describing the local dynamic response of flexible materials, such as the forced vibration equation based on the discretized distributed spring equivalent model. ,in It is a vertical displacement. It is the equivalent vibrational mass. It is the equivalent area of ​​the deformed region. This is the known air pulse pressure load. This term is used to penalize any predictions that violate this physical law. It is a hyperparameter that balances the two losses. By minimizing the total loss... The physical information network performs parameter iterative convergence based on the constraints of observation data and physical laws.

[0059] In one example of this invention, after receiving a sequence of dynamically deformed images, the system detects 150 feature points in the first frame using the Canard-Lucas-Tommasi algorithm, forming an initial feature point cloud. Subsequently, the system tracks these 150 points over the next nine time steps. to The pixel displacements within the dataset were used to generate a pixel displacement mapping field containing 150×9 spatiotemporal displacement vectors. This dataset was then input into a pre-loaded physical information neural network for inverse solving. In the initial stage of the iteration, the model was set with initial parameters of... , The initial loss function is obtained by calculating the difference between the predicted deformation and the observed data. The gradient is calculated using the backpropagation algorithm, and the gradient is iteratively adjusted. and The value. After dozens of iterations, when the model's internal parameters were adjusted to... , At that time, the total loss was calculated. Upon reaching the minimum and converging, the system stops solving. Ultimately, the system generates a set of local mechanical parameters, directly including the equivalent dynamic stiffness coefficient of 132.5 N / m and the equivalent damping coefficient of 0.28 N⋅s / m. The process of the aforementioned physical information neural network performing inverse partial differential constraint solving and convergence fitting based on timestamp sequence observation data can be found in [reference needed]. Figure 4 As shown.

[0060] In one embodiment of the present invention, the security verification module is configured to perform the following steps: Extract the equivalent dynamic stiffness coefficient from the local mechanical parameter set; compare the equivalent dynamic stiffness coefficient with the lower limit of stiffness safety; if the equivalent dynamic stiffness coefficient is less than the lower limit of stiffness safety, output a dangerous shutdown warning sign; if the equivalent dynamic stiffness coefficient is greater than or equal to the lower limit of stiffness safety, generate a trajectory modeling enable signal.

[0061] Specifically, to unpack the local mechanical parameter set and execute security verification logic, the system first locates and parses the local mechanical parameter set file generated by the previous implementation method in the running memory. The processing logic unit directly extracts the equivalent dynamic stiffness coefficient, which characterizes the material's resistance to normal deformation, from this data structure. .

[0062] At the same time, the system accesses the device’s local non-volatile memory to obtain the stiffness safety lower limit value that has been pre-calibrated and debugged.

[0063] Specifically, the lower limit of stiffness safety is a critical process threshold. Its setting is not arbitrary but based on statistical results from active physical testing of a large number of qualified and unqualified flexible printed circuit board (FPCB) samples. This lower limit is determined by recording the equivalent dynamic stiffness coefficients of numerous samples that failed to mount due to warping, wrinkling, or poor adhesion, taking the upper limit of their statistical distribution, and combining this with a certain safety margin. For example, this value can be set to 85% of the average stiffness measurement of all qualified samples. In actual production, the typical range of this threshold is 80 N / m to 110 N / m.

[0064] After acquiring the two values, the system performs a direct comparison to determine if the equivalent dynamic stiffness coefficient is less than the lower limit of stiffness safety. If the equivalent dynamic stiffness coefficient is less than this lower limit, it indicates that the flexible circuit board has not been completely flattened by the suction cup or that abnormal wrinkles exist. The system immediately generates a dangerous shutdown warning sign. The dangerous shutdown warning sign is a system interrupt signal or a specifically coded data message with the highest processing priority, ensuring that actions that may damage the equipment or render the product unusable can be stopped in the shortest possible time. This sign is output to the motion control system via the internal control bus, controlling all servo motors on all axes to stop moving and enter a locked state, while simultaneously displaying an alarm message on the human-machine interface.

[0065] Conversely, if the equivalent dynamic stiffness coefficient is greater than or equal to the lower limit, it indicates that the mechanical state of the target area meets the requirements of high-precision mounting. The system then generates and pushes a trajectory modeling enable signal to the task queue of the main control computing unit. In this embodiment, the trajectory modeling enable signal is a logical signal, usually implemented as a variable with a Boolean value of true or a specific message queue notification, which is used to trigger subsequent spatial dynamics model calculations after the physical characteristics verification is passed.

[0066] In one example of the present invention, the system processing unit unpacks the mechanical parameter results from memory and extracts the absolute value of the equivalent dynamic stiffness coefficient. Subsequently, the system accesses its configuration file and reads the preset lower limit of stiffness safety as 120 N / m. Next, the system performs a comparison operation: 132.5 N / m ≥ 120 N / m. The comparison result is true. Therefore, the system determines that the current state of the flexible circuit board is suitable for mounting and immediately generates a trajectory modeling enable signal. Specifically, this is implemented by setting a semaphore to an available state, waking up the task process in the main control computing unit that is waiting for this semaphore, thus successfully starting the subsequent spatial dynamics model. For comparison, suppose in another measurement, due to blockage of a micro-hole in the vacuum chuck, the target pad area slightly tilts up; the measured equivalent dynamic stiffness coefficient in this case is... During the comparison, 105.7 N / m < 120 N / m, so the result is true. The system will immediately generate a danger shutdown warning sign and send it as an emergency message on the controller area network bus, with an ID number. issue, Upon receiving this flag, the axis driver stops its downward movement to prevent component placement failure.

[0067] In one embodiment of the present invention, the trajectory prediction generation module is configured to perform the following steps: The equivalent dynamic stiffness coefficient and equivalent damping coefficient are extracted from the local mechanical parameter set; the equivalent dynamic stiffness coefficient and equivalent damping coefficient are transformed into equivalent horizontal plane mechanical parameters through a preset mapping transformation matrix, and the equivalent horizontal plane mechanical parameters are used as constitutive parameters of the spatial dynamics model to construct the local vibration equation in the horizontal plane; the time-series descent velocity and time-series acceleration variables are separated from the preset descent trajectory; the time-series descent velocity and time-series acceleration variables are substituted into the excitation terms of the local vibration equation, and the in-plane geometric coupling effect is introduced for solution to obtain the full-time domain position prediction trajectory; the equivalent vibration mass of the target flexible circuit board is obtained. The time-series downward velocity and acceleration variables are mapped to an equivalent horizontal plane coupled tension term induced by aerodynamic depression. Using the two-dimensional horizontal plane displacement vector to be solved as a variable, an inertial force term is constructed from the equivalent vibrating mass and the second-order time derivative of the two-dimensional horizontal plane displacement vector; a damping force term is constructed from the horizontal damping coefficient in the equivalent horizontal plane mechanical parameters and the first-order time derivative of the two-dimensional horizontal plane displacement vector; and an elastic force term is constructed from the horizontal plane dynamic stiffness matrix in the equivalent horizontal plane mechanical parameters and the two-dimensional horizontal plane displacement vector. The sum of the inertial force term, damping force term, and elastic force term is set equal to the equivalent horizontal plane coupled tension term, thus constructing a second-order differential equation. Solving the second-order differential equation yields the time-varying sequence of the two-dimensional horizontal plane displacement vector, which serves as the full-time-domain position prediction trajectory.

[0068] Specifically, upon receiving the trajectory modeling enable signal pushed by the previous implementation method, the main controller immediately mounts a preset spatial dynamics model in the real-time operating environment. This spatial dynamics model is a simplified model of a mechanical vibration system based on Newton's second law, used to simulate the dynamic behavior of an object with mass, elasticity, and damping characteristics under external forces. Mathematically, this model is described using a second-order mass-spring-damped system model. The system reads the identified out-of-plane equivalent dynamic stiffness coefficient and equivalent damping coefficient, and transforms them into in-plane values ​​using a preset material constitutive mapping transformation matrix. The equivalent horizontal shear / tensile stiffness and equivalent horizontal damping coefficient of the plane are used as constitutive parameters of the spatial dynamics model to establish the local vibration equation in the horizontal plane.

[0069] The system obtains information from the task planning module of the motion controller that the placement head will be in operation soon. The preset descent trajectory is executed along the axis. It should be noted that the preset descent trajectory is the initial command for the mounting action, describing the ideal motion. This trajectory is a high-density time series, defining in detail the target position, velocity, and acceleration for each minute time step from the current hovering height to the mounting plane. The system extracts the time-series descent velocity variable and the time-series acceleration variable from this trajectory data.

[0070] Subsequently, the system uses a preset gas-solid deformation coupling conversion function, which includes a preset column vector of vertical-horizontal two-dimensional displacement-tension conversion coefficients, to calculate the equivalent distortion tension vector of the aerodynamic indentation acting on the horizontal plane of the target pad. This equivalent distortion tension vector is then used as the external excitation term of the local vibration equation. After completing the model parameterization and excitation settings, the system deterministically solves the second-order ordinary differential equation using numerical integration methods, such as the fourth-order Runge-Kutta method, to obtain a description of the positional displacement of the target pad on the horizontal plane over time under the impact of the downward airflow from the mounting head.

[0071] The calculated position offset is vector-superimposed with the initial global coordinates of the target pad to obtain the final full-time-domain position prediction trajectory, which includes the position at all future moments. Furthermore, the full-time-domain position prediction trajectory is a two-dimensional vector time series, which gives the predicted coordinate flow of the target pad on the horizontal plane from the current moment to the placement completion moment, and its accuracy is much higher than that of static positioning coordinates.

[0072] The second-order mass-spring-damped system model clarifies the mathematical structure of the model, and the calculation formula is as follows: ; In the formula, It is the two-dimensional horizontal displacement vector that needs to be solved for, which varies with time, i.e. ; It is the equivalent vibration mass of the flexible circuit board in the area impacted by airflow. This value is preset based on the material density and thickness of the flexible circuit board. It is the equivalent dynamic stiffness matrix of the transformed horizontal plane. This is the equivalent horizontal damping coefficient. These two parameters are determined online. and Obtained through spatial mapping transformation to match the order of magnitude difference between the out-of-plane bending stiffness and in-plane tensile stiffness of thin film materials. This is applicable to flexible circuit boards with anisotropic trace structures. It can be extended and configured as a second-order diagonal matrix to represent respectively shaft and Differentiated horizontal mechanical properties of the shaft. It is the horizontal geometric coupling tension generated by the asymmetric indentation of the substrate induced by the pneumatic impact of the mounting head protruding downwards. This tension, as an equivalent excitation term in the model, is... virtual downward movement speed of the axis and acceleration The specific coupling transformation form of the mapping function depends on the offline physical calibration program obtained before the system leaves the factory.

[0073] The specific calibration steps are as follows: First, a calibration platform containing a miniature two-dimensional force sensor and a laser vibrometer is built, and a standard test flexible circuit board is fixed above the force sensor. Second, the mounting head is controlled to perform a downward movement with different pre-set timing velocity and acceleration vectors, inducing substrate deformation through air pulses. Simultaneously, the actual aerodynamic coupling tension data on the horizontal surface of the substrate is collected through the two-dimensional force sensor. Finally, multiple sets of tension data under different velocity and acceleration conditions are collected, and a binary linear polynomial fitting is performed using the least squares method to solve for the exact transformation coefficient column vector. and The value of is then stored in the system as an empirical constant matrix. The local vibration equation is obtained by substituting . The differential equation constructed with three parameters.

[0074] In one example of this invention, after receiving a trajectory modeling enable signal, the system activates the spatial dynamics model. It then reads the calculated out-of-plane equivalent dynamic stiffness coefficients. and equivalent damping coefficient Based on thin-film mechanics theory, the in-plane tensile stiffness and out-of-plane bending stiffness of a flexible circuit board exhibit a proportional relationship based on its thickness and effective width. The system applies a preset geometric coupling amplification factor. (Set here) Perform a physical mapping transformation to obtain the equivalent horizontal local tensile stiffness. Equivalent horizontal damping Simultaneously, the equivalent vibration mass of the target area is retrieved from the system configuration library. The system acquires the preset descent trajectory along the Z-axis, which plans for the placement head to descend uniformly from a height of 0.5mm to 0mm within 10ms. Therefore, the timing descent speed is constant. acceleration The system consults the stored gas-solid calibration coefficient table and obtains the equivalent distortion tensile force function as follows: ,in and These are two-dimensional transformation coefficient column vectors representing the coupled tensile force mapping from vertical deformation induced by aerodynamic impact to horizontal displacement. (Assuming the values ​​obtained during calibration...) Value And because of the uniform descent Therefore, the constant excitation forces generated in the X and Y directions are both Therefore, a system of symmetrical second-order vibration equations to be solved along the X and Y axes is constructed. as well as The system uses a numerical solver in... Solve this system of equations within the interval to obtain the displacement trajectory. and The steady-state deviation of this set of equations under high damping approximates... That is, 2.83 μm. Assuming the initial global coordinates of the target pad are (152.450000 mm, 76.320000 mm), the calculated displacement trajectory... and exist The time-averaged steady-state value reached 0.002830 mm. Therefore, the coordinates of the endpoint of the full-time-domain position prediction trajectory were predicted as (152.452830 mm, 76.322830 mm), and the entire sequence contained the complete dynamic path from (152.450000, 76.320000) to this endpoint. See also Figure 5 , Figure 5 The full-time-domain position prediction compensation trajectory and steady-state deviation approximation trend under the aforementioned aerodynamic impact-induced in-plane coupled deformation are visually demonstrated.

[0075] In one embodiment of the present invention, the feedforward compensation control module is configured to perform the following steps: The trajectory decomposition algorithm is applied to deconstruct the full-time domain position prediction trajectory, and the X-axis position deviation sequence and the Y-axis position deviation sequence are solved separately. According to the preset gain coefficient, the X-axis position deviation sequence and the Y-axis position deviation sequence are proportionally amplified, and after rounding and remainder accumulation, discrete integer digital command values ​​are generated. The discrete integer digital command values ​​are encoded into discrete electronic control information packets to generate a feedforward control command sequence. The feedforward control command sequence is fed forward into the horizontal servo driver to synthesize the XY plane compensation trajectory curve.

[0076] Specifically, to segment the full-time-domain position prediction trajectory and generate a feedforward control command sequence, the system first calls the built-in trajectory decomposition algorithm module to deconstruct the full-time-domain position prediction trajectory generated in the previous implementation method. The trajectory decomposition algorithm is an application of inverse kinematics, aiming to decompose the target's motion trajectory in the task space into control commands for each joint space of the robot; in this scenario, it is equivalent to decomposing the compensated motion in a two-dimensional plane into... shaft and Independent displacement commands for two orthogonal drive axes.

[0077] The predicted trajectory for full-time location is a two-dimensional coordinate sequence that evolves over time. This includes the ideal movement path of the target pad in the reference coordinate system. The trajectory decomposition algorithm compares this trajectory with the default linear motion trajectory of the placement head under uncompensated conditions. By performing point-by-point vector subtraction, the solution at each time step can be obtained separately. Required position deviation sequence in axial direction and in the orthogonal direction Required position deviation sequence in axial direction .

[0078] It should be noted that the position deviation sequence is time-series data, and its sampling rate is consistent with the generation rate of the full-time-domain position prediction trajectory, ensuring the real-time nature of the compensation. After obtaining these two one-dimensional deviation sequences, the system, based on the preset gain coefficient, performs... Axial position deviation sequence and Each data point in the axial position deviation sequence is proportionally amplified. Here, the gain coefficient, set according to the servo drive's resolution and electronic gear ratio, ensures that the physical displacement calculated by the front end can be correctly converted into the pulse equivalent required by the motor. For example, if the motor moves 2.5 nm for each pulse received by the servo system, the gain coefficient is 400 pulses / μm, and then rounded and the remainders are accumulated. This operation converts the physical displacement into a discrete integer digital instruction value that the servo system recognizes and executes.

[0079] The amplified deviation sequence is then fed into the encoder module, which packages the discrete integer digital instruction values ​​into a series of independent, timestamped discrete electronic control information packets according to the communication protocol format required by the servo drive, and finally generates a feedforward control instruction sequence.

[0080] In this embodiment, the high-frequency update characteristic of the feedforward control command sequence is reflected in its data update rate, which must match or exceed the basic control frequency of the servo loop, typically above 1kHz, in order to achieve smooth and jitter-free micro-trajectory tracking.

[0081] The feedforward control command sequence is input to the feedforward control register of the horizontal servo driver via the data bus. Specifically, feedforward input is an advanced servo control technique where an external controller directly injects compensation signals into the driver's underlying control loop to reduce the response latency of the command parsing stage. While executing its main feedback control loop, the horizontal servo driver superimposes the received feedforward value with the feedback error, thereby forcing the placement head to... A compensation trajectory curve is fitted onto the horizontal plane of the axis.

[0082] The compensation trajectory curve is relative to the macroscopic straight-line descent path, and its offset is usually at the micrometer level, but it plays a decisive role in high-precision placement. The trajectory of this curve is completely synchronized with the predicted movement trajectory of the target pad, and ultimately, at the moment of placement, the center of the placement head nozzle is aligned with the center of the dynamically offset pad.

[0083] In one example of this invention, the system receives a full-time-domain position prediction trajectory with endpoint coordinates of (152.452830mm, 76.322830mm). The initial target point is (152.450000mm, 76.320000mm). The trajectory decomposition algorithm calculates the total compensation travel of the X-axis and Y-axis during the entire 10ms downward compression process. and Assuming the sampling rate of the predicted trajectory is 1kHz, this will generate a position deviation sequence along the X-axis and a position deviation sequence along the Y-axis for 10 data points. For example, in When, the deviation of the X-axis is Assuming a gain coefficient of 400,000 pulses / mm, the deviation at that moment is converted into 600 pulses. The encoder packages this integer value and its corresponding timestamp into an electronic control information packet. Thus, the system generates a feedforward control command sequence containing 10 data packets and sends it to the horizontal servo drive via the EtherCAT bus in feedforward mode. Upon receiving these feedforward values, the horizontal servo drive, in its own position loop control, in addition to processing the main commands from the host computer and the errors fed back by the encoder, also adds these 600 pulse commands. After the aforementioned feedforward commands are generated and injected, the system wakes up the hovering Z-axis servo drive to resume executing the downward, straight macroscopic motion. Simultaneously, the placement head synchronously executes a feedforward compensation trajectory in the XY plane, aligning with the center of the pads that have shifted due to airflow disturbance, i.e., (152.452830, 76.322830), to complete the placement alignment action.

[0084] Each of the modules can be implemented in whole or in part through software, hardware, or a combination thereof. It supports hardware embedded in or independent of the processor in the computer device, and also supports software stored in the memory of the computer device, so that the processor can call and execute the operations corresponding to each of the above modules.

[0085] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A visual guidance and correction system for flexible circuit board mounting accuracy, characterized in that, include: The front positioning module is used to obtain the placement instructions, extract the global coordinates of the target pad, drive the placement head to hover above the target flexible circuit board within a preset height range, and generate a hovering trigger signal. The physical detection module is used to receive hovering trigger signals, activate the air pulse piezoelectric nozzle unit, and apply physical air pulses to the target flexible circuit board; The dynamic response acquisition module is used to monitor the exhaust action of physical air pulses, trigger the image acquisition array to capture images of the stress deformation of the target flexible circuit board, and generate a dynamic deformation image sequence. The mechanical parameter identification module is used to acquire a preset physical information neural network. The dynamic deformation image sequence is input into the physical information neural network for analysis, and the local mechanical parameter set is output. The safety verification module is used to extract the equivalent dynamic stiffness coefficient from the local mechanical parameter set, compare the equivalent dynamic stiffness coefficient with the preset stiffness safety lower limit value, and generate a trajectory modeling enable signal. The predicted trajectory generation module is used to receive the trajectory modeling enable signal, obtain the preset spatial dynamics model and the preset descent trajectory, input the local mechanical parameter set and the preset descent trajectory into the spatial dynamics model for solution, and obtain the full time domain position prediction trajectory. The feedforward compensation control module is used to deconstruct the full-time-domain position prediction trajectory, generate a feedforward control command sequence, and inject the feedforward control command sequence into the horizontal servo driver to perform alignment compensation actions.

2. The flexible circuit board mounting accuracy visual guidance and correction system according to claim 1, characterized in that, The front positioning module is specifically used for: Extract the global coordinates of the target pad from the placement command; Monitor the dynamic Z-axis position data during the downward movement of the placement head; Compare the dynamic Z-axis position data with the preset height range; A hovering trigger signal is generated when the dynamic Z-axis position data reaches the upper boundary value of the preset height range.

3. The flexible circuit board mounting accuracy visual guidance and correction system according to claim 1, characterized in that, The physical detection module is specifically used for: Extract the hovering trigger signal and close the underlying hardware drive circuit of the air pulse piezoelectric nozzle unit; Control the underlying hardware driver circuit to release the pulse drive signal; A pulsed drive signal is used to drive a piezoelectric ceramic array to undergo directional deformation. Physical air pulses are expelled by directional deformation and extrusion of the built-in air cavity, and then applied vertically to the target flexible circuit board.

4. The flexible circuit board mounting accuracy visual guidance and correction system according to claim 1, characterized in that, The dynamic response acquisition module is specifically used for: The rising edge waveform of the intercepted pulse drive signal; A hardware-triggered synchronization signal is generated using the rising edge waveform. Input the hardware trigger synchronization signal into the image acquisition array to activate the continuous exposure acquisition mode; Images of the target flexible circuit board are captured continuously using a continuous exposure acquisition mode, and the consecutive image frames are stacked according to the timestamp to generate a dynamic deformation image sequence.

5. The flexible circuit board mounting accuracy visual guidance and correction system according to claim 1, characterized in that, The mechanical parameter identification module is specifically used for: Extract feature point clouds from each consecutive image frame of a dynamic deformation image sequence; Track the cross-frame motion trajectory of feature point clouds and generate a pixel displacement mapping field with timestamp information; The pixel displacement mapping field is input into the physical information neural network, and the deformation law is extracted through the data-driven loss term and the physical law constraint loss term of the physical information neural network. The equivalent dynamic stiffness coefficient and equivalent damping coefficient of the target flexible circuit board are solved in reverse, and the set of local mechanical parameters is output by combination.

6. The flexible circuit board mounting accuracy visual guidance and correction system according to claim 5, characterized in that, The equivalent dynamic stiffness coefficient and equivalent damping coefficient of the target flexible circuit board are solved in reverse, specifically including: The data-driven loss term and the physical law constraint loss term are used to acquire physical information from the neural network. Obtain the preset balance weight hyperparameters; The weighted physical constraint term is obtained by multiplying the physical constraint loss term with the balance weight hyperparameter. The total loss function value is obtained by summing the data-driven loss term and the weighted physical constraint term. The internal parameters are iteratively adjusted with the goal of minimizing the total loss function value until convergence is obtained to obtain the equivalent dynamic stiffness coefficient and the equivalent damping coefficient.

7. The flexible circuit board mounting accuracy visual guidance and correction system according to claim 1, characterized in that, The security verification module is specifically used for: Extract the equivalent dynamic stiffness coefficient from the local mechanical parameter set; Compare the numerical values ​​of the equivalent dynamic stiffness coefficient and the lower limit of stiffness safety; If the equivalent dynamic stiffness coefficient is less than the lower limit of stiffness safety, a dangerous shutdown warning sign will be output. If the equivalent dynamic stiffness coefficient is greater than or equal to the lower limit of stiffness safety, generate a trajectory modeling enable signal.

8. A flexible circuit board mounting accuracy visual guidance and correction system according to claim 5 or 6, characterized in that, The predicted trajectory generation module is specifically used for: Extract the equivalent dynamic stiffness coefficient and equivalent damping coefficient from the local mechanical parameter set; The equivalent dynamic stiffness coefficient and equivalent damping coefficient are transformed into equivalent horizontal plane mechanical parameters through a preset mapping transformation matrix, and the equivalent horizontal plane mechanical parameters are used as constitutive parameters of the spatial dynamics model to construct the local vibration equation in the horizontal plane. Separate the temporal descent velocity and temporal acceleration variables from the preset descent trajectory; Substituting the time-series downward velocity and time-series acceleration variables into the excitation terms of the local vibration equation, and introducing the in-plane geometric coupling effect for solution, the full-time domain position prediction trajectory is obtained.

9. A flexible circuit board mounting accuracy visual guidance and correction system according to claim 8, characterized in that, The temporal downward velocity and temporal acceleration variables are substituted into the excitation terms of the local vibration equation, and the in-plane geometric coupling effect is introduced for solution. Specifically, this includes: Obtain the equivalent vibration mass of the target flexible circuit board; The time-series descent velocity and time-series acceleration variables are mapped to the equivalent in-plane coupled tension term induced by aerodynamic indentation; Using the two-dimensional horizontal plane displacement vector to be solved as a variable, the inertial force term is constructed from the equivalent vibrating mass and the second time derivative of the two-dimensional horizontal plane displacement vector; the damping force term is constructed from the horizontal damping coefficient in the equivalent horizontal plane mechanical parameters and the first time derivative of the two-dimensional horizontal plane displacement vector; and the elastic force term is constructed from the horizontal plane dynamic stiffness matrix in the equivalent horizontal plane mechanical parameters and the two-dimensional horizontal plane displacement vector. Let the sum of the inertial force term, damping force term, and elastic force term equal the equivalent horizontal plane coupled tension term, and construct a second-order differential equation; solve the second-order differential equation to obtain the time-varying sequence of the two-dimensional horizontal plane displacement vector, which serves as the full-time domain position prediction trajectory.

10. A flexible circuit board mounting accuracy visual guidance and correction system according to claim 1, characterized in that, The feedforward compensation control module is specifically used for: The trajectory decomposition algorithm is applied to deconstruct the full-time domain position prediction trajectory, and the X-axis position deviation sequence and the Y-axis position deviation sequence are solved separately. Based on the preset gain coefficient, the X-axis position deviation sequence and the Y-axis position deviation sequence are proportionally amplified, and after rounding and remainder accumulation, discrete integer digital instruction values ​​are generated. The discrete integer digital instruction values ​​are encoded into discrete electronic control information packets to generate a feedforward control instruction sequence. The feedforward control command sequence is fed forward into the horizontal servo driver to synthesize the XY plane compensation trajectory curve.

Images