Virtual debugging method for SMT production line based on digital twinning

Abstract

The application relates to the field of intelligent manufacturing and industrial automation, and discloses an SMT production line virtual debugging method based on digital twinning. The method comprises the following steps: constructing a high-fidelity digital twin body which is fused with geometric, kinematic, dynamic, control logic and process models; collecting physical production line data in real time through a multi-source sensor and driving the twin body to perform synchronous simulation; when an abnormality is detected, performing fault tracing based on a multivariate causal diagram, and inversely calculating a root cause parameter set; performing parameter sensitivity scanning and multi-objective optimization in the twin body, generating a corrected parameter combination; automatically issuing the corrected parameter combination to physical equipment through a security authentication channel, and verifying the effect through feedback data to form a "perception-diagnosis-decision-execution-verification" closed loop. The application improves the debugging efficiency and mounting precision stability, shortens the debugging period, and reduces the dependence on artificial experience.

Description

Technical Field

[0001] This invention belongs to the field of intelligent manufacturing and industrial automation, and specifically relates to a virtual debugging method for SMT production lines based on digital twins. Background Technology

[0002] With the widespread adoption of surface mount technology (SMT) in the electronics manufacturing industry, production line debugging efficiency and accuracy have become key bottlenecks restricting the upgrading of intelligent manufacturing. Traditional physical debugging relies on repeated trial and error on-site, which is costly and time-consuming, making it difficult to meet the needs of flexible, multi-variety, and small-batch production.

[0003] Virtual debugging technology has emerged to address this need by constructing digital models of equipment to simulate its operation and expose potential anomalies in advance. However, current virtual debugging methods generally suffer from the limitation of "visible symptoms but difficult to trace the root cause": the system can only identify surface-level faults such as placement misalignment and abnormal reflow soldering temperature, but cannot penetrate the internal structure and process flow of the equipment for in-depth attribution. It often simply attributes complex problems to a single factor, ignoring the multi-source coupling effects such as transmission clearance, motor step angle drift, and thermal field coupling effects. This results in fault diagnosis being highly dependent on the experience of senior engineers, making the debugging process inefficient and unreproducible.

[0004] Virtual commissioning based on digital twins is considered a crucial path to improving the intelligence level of SMT production lines. This approach aims to achieve real-time mapping between physical equipment and virtual models through high-fidelity modeling, supporting non-intrusive commissioning and predictive maintenance. Its core lies in integrating multi-dimensional information such as equipment geometry, kinematics, thermodynamics, and process parameters to construct a calculable and predictable virtual entity, thereby verifying the rationality of control logic and process parameters before production. However, current digital twin applications mostly remain at the level of visual monitoring or static simulation, and a commissioning mechanism with autonomous diagnostic and closed-loop optimization capabilities has not yet been formed.

[0005] In existing technologies, there is a "parameter gap" between virtual debugging and physical equipment: even if anomalies are identified and correction suggestions are made in the virtual environment, the adjusted parameters still need to be manually entered into the PLC or equipment controller. This not only easily introduces input errors, but also makes it impossible to provide timely feedback on the correction effect, resulting in a break in the "virtual correction - physical execution - effect verification" chain.

[0006] The lack of a knowledge system that links anomalies, equipment degradation, and process constraints makes it difficult for the system to automatically deduce the root cause from massive amounts of operational data and generate targeted parameter compensation strategies. This deficiency is particularly pronounced in high-density, high-cycle SMT production lines, easily leading to batch soldering defects or equipment shutdowns. It severely restricts the evolution of virtual debugging from an "auxiliary tool" to an "autonomous decision-making system," urgently requiring a closed-loop debugging method that deeply integrates anomaly detection, multi-dimensional tracing, and adaptive correction through virtual-physical linkage. Summary of the Invention

[0007] This invention provides a virtual debugging method for SMT production lines based on digital twins, aiming to solve core problems in traditional virtual debugging technology such as unclear fault root cause location, disconnect between virtual and physical equipment parameter adjustment, and broken debugging closed loop.

[0008] This invention constructs a high-fidelity multi-physics coupled digital twin, integrating equipment structural dynamics models, motion control logic models, and process models to achieve full-element, full-process, and full-dimensional mapping of the SMT production line's operating status;

[0009] Based on this, a fault tracing mechanism based on causal reasoning and a virtual-real two-way synchronous parameter control architecture are introduced, which enables virtual debugging to not only identify abnormal phenomena, but also accurately invert the underlying mechanical, electrical or control parameter deviations that cause the phenomena, and automatically send the optimized parameter instructions to the physical equipment execution unit. At the same time, physical feedback is collected in real time to verify the correction effect, thus forming a complete closed-loop debugging process of "perception-diagnosis-decision-execution-verification".

[0010] This invention provides a virtual debugging method for SMT production lines based on digital twins, comprising:

[0011] Construct a high-fidelity digital twin of an SMT production line, the digital twin including geometric models, kinematic models, dynamic models, control logic models and process parameter models of the pick-and-place machine, printer, reflow oven and conveyor system;

[0012] The equipment operation data is collected in real time by a multi-source sensor network deployed on the physical SMT production line. The equipment operation data includes servo motor encoder position signal, spindle vibration acceleration signal, pneumatic actuator pressure signal, vision positioning system coordinate deviation data, and PCB board surface temperature distribution data.

[0013] The device's operating data is synchronized to the digital twin in real time via an industrial Ethernet network, driving the digital twin to reproduce its state and simulate its behavior.

[0014] When the digital twin detects a process anomaly, it activates a fault tracing analysis module based on a multivariate cause-effect graph. The fault tracing analysis module, based on the preset physical dependencies between equipment components and signal transmission paths, combined with the current anomaly feature vector, reversely derives the set of root cause parameters most likely to cause the anomaly.

[0015] Based on the root cause parameter set, a parameter sensitivity scan and multi-objective optimization algorithm are executed inside the digital twin to generate a set of modified parameter combinations that meet the process quality constraints.

[0016] The modified parameter combination is automatically sent to the corresponding programmable logic controller or motion control card of the physical SMT production line through a secure and certified command channel;

[0017] After the physical device executes the corrected parameters, new operating data is continuously collected and fed back to the digital twin to verify the effectiveness of the correction. Based on the verification results, a decision is made on whether to terminate the debugging process or proceed to the next round of iterative optimization.

[0018] As one embodiment of the present invention, the construction of a high-fidelity digital twin of an SMT production line specifically includes:

[0019] The geometry of the XYZ three-axis motion platform, rotary placement head, feeder array, and vision camera of the chip mounter is accurately modeled using a non-uniform rational B-spline surface.

[0020] A rigid-flexible coupled dynamic model of a three-axis motion platform is established based on the Lagrange equation, which takes into account the elastic deformation of the ball screw, the nonlinear frictional characteristics of the linear guide, and the electromagnetic torque fluctuation of the servo motor.

[0021] The pick-up-rotate-place action sequence of the placement head is encoded as a finite state machine, and its state transition conditions are jointly triggered by the visual recognition result, the vacuum negative pressure sensor threshold, and the cylinder positioning signal.

[0022] For reflow ovens, a three-dimensional thermal field model based on computational fluid dynamics is established, and its boundary conditions are determined by the power setpoint of the heating zone, the speed of the cooling fan, and the nitrogen flow rate in the oven.

[0023] All sub-models are co-simulated using a unified time base and event-driven mechanism, with the simulation step size set at 0.5 milliseconds.

[0024] As one embodiment of the present invention, the multi-source sensor network includes: an incremental photoelectric encoder mounted on the rear end of the X-axis servo motor, with a resolution of 400,000 pulses per revolution; and a triaxial MEMS accelerometer attached to the Z-axis ball screw nut seat, with a range of ±50 times the gravitational acceleration and a sampling frequency of 5000 Hz.

[0025] The miniature piezoresistive pressure sensor integrated inside the mounting nozzle has a measurement range of 0 to -100 kPa and a response time of 0.1 milliseconds.

[0026] A high-resolution industrial camera, with a pixel size of 2.2 micrometers and a frame rate of 120 frames per second, is positioned above the pick-and-place machine to capture the relative pose of components and pads; a K-type thermocouple array is arranged in each temperature zone of the reflow oven, with a temperature measurement accuracy of ±0.5 degrees Celsius.

[0027] As one embodiment of the present invention, the method for constructing the fault tracing analysis module based on multivariate causal graphs is as follows:

[0028] All key components and their interactions in the SMT production line are predefined to form a directed acyclic graph structure, where nodes represent physical components or control variables, and edges represent energy flow, information flow, or force transmission paths.

[0029] Assign a transfer function model to each edge to describe the dynamic effect of input perturbation on the output response;

[0030] When the mounting offset is detected to be greater than the preset threshold, the offset is used as observation evidence, and the posterior probability of each potential root cause node is calculated using the Bayesian inversion algorithm.

[0031] The parameters corresponding to nodes with a posterior probability greater than the threshold of 0.8 are selected as the root cause parameter set. The parameters include at least one of the following: X-axis guide rail preload, Y-axis servo gain, mounting head rotational inertia compensation coefficient, and visual calibration matrix offset.

[0032] As one embodiment of the present invention, the parameter sensitivity scanning and multi-objective optimization algorithm specifically performs the following steps:

[0033] In the digital twin, a Latin hypercube sampling space is constructed with the root parameter set as the center and within its allowable range of variation;

[0034] Perform a complete mounting process simulation for each sampling point and record three indicators: mounting accuracy, cycle time, and equipment energy consumption.

[0035] A non-dominated sorting genetic algorithm is used to perform multi-objective optimization on the sampling results and select the Pareto optimal solution set.

[0036] From the Pareto optimal solution set, the final combination of correction parameters is selected according to the preset weight coefficients. The weight coefficients are dynamically adjusted according to the priority of the current production task. If it is a high-precision product, the mounting accuracy weight is set to 0.7, the cycle time weight is set to 0.2, and the energy consumption weight is set to 0.1.

[0037] As one embodiment of the present invention, the instruction channel for security authentication adopts a two-way authentication mechanism based on digital certificates, and the physical device only accepts parameter modification instructions from the registered digital twin.

[0038] Before issuing instructions, formal verification must be performed to ensure that parameter values ​​are within the safe operating envelope of the equipment;

[0039] Parameter modifications are performed in an atomic transaction manner. If any parameter write fails, all parameters are rolled back to their original values.

[0040] After the parameters take effect, the physical device returns an acknowledgment signal and the actual written value to the digital twin for comparison to check for communication errors.

[0041] As one embodiment of the present invention, the verification of the effectiveness of the correction specifically includes:

[0042] The standard deviation of the mounting offset is collected over three consecutive production cycles after the physical equipment executes the correction parameters.

[0043] If the standard deviation is less than the specified value, i.e. 80% of the preset tolerance bandwidth, then the correction is deemed effective.

[0044] Otherwise, the input parameters and output results of this correction attempt will be stored in the historical case library to update the cause-effect graph weights in the fault tracing analysis module and improve the accuracy of subsequent diagnosis.

[0045] This invention also provides a virtual debugging system for SMT production lines based on digital twins, comprising:

[0046] High-fidelity digital twin building blocks are used to construct digital twins of SMT production lines that include geometric, kinematic, dynamic, control logic, and process parameter models.

[0047] The multi-source sensor data acquisition and synchronization unit is used to acquire equipment operation data of the physical production line in real time via industrial Ethernet and drive the status update of the digital twin.

[0048] The anomaly detection and fault tracing unit is used to reverse deduce the root cause parameter set based on a multivariate cause-effect graph when a process anomaly is detected.

[0049] The parameter optimization and instruction generation unit is used to perform sensitivity scanning and multi-objective optimization within the digital twin, generate modified parameter combinations, and send them out through a secure channel.

[0050] The virtual-real closed-loop verification unit is used to receive feedback data after the physical device has performed the operation, evaluate the effect of the correction, and decide whether to terminate or iterate the debugging process.

[0051] The high-fidelity digital twin building unit further includes a geometric modeling subunit, a rigid-flexible coupling dynamics modeling subunit, a finite state machine control logic modeling subunit, and a computational fluid dynamics thermal field modeling subunit.

[0052] The multi-source sensing data acquisition and synchronization unit is connected to an incremental photoelectric encoder, a three-axis MEMS accelerometer, a miniature piezoresistive pressure sensor, a high-resolution industrial camera, and a K-type thermocouple array.

[0053] The anomaly detection and fault tracing unit has a built-in directed acyclic graph storage module, a Bayesian inversion calculation engine, and a root cause parameter screening threshold setting module.

[0054] The parameter optimization and instruction generation unit integrates a Latin hypercube sampler, a non-dominated sorting genetic algorithm solver, and an atomic transaction instruction wrapper.

[0055] The virtual-real closed-loop verification unit is equipped with a mounting accuracy statistical analyzer, a safety envelope checker, and a historical case knowledge update interface.

[0056] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0057] 1. By constructing a high-fidelity digital twin that integrates multiple physics fields, an accurate mirror image of the SMT production line's operating status was achieved, providing a reliable simulation basis for fault diagnosis;

[0058] 2. The introduction of a fault tracing mechanism based on cause-effect graphs breaks through the limitation of traditional methods that can only identify superficial anomalies. It can directly locate the root cause parameters such as specific mechanical clearance, control gain or calibration error, reducing the dependence on engineers' experience.

[0059] 3. Establish a parameter control architecture that is synchronous between virtual and physical environments, so that the optimization results in the virtual environment can be automatically, accurately and securely sent to physical devices, and the correction effect can be verified through real-time feedback, eliminating the problems of manual input errors and broken debugging loops.

[0060] 4. The entire debugging process forms a complete automated closed loop, which shortens the time required for a single debugging cycle from that required by traditional methods, while improving the stability of mounting accuracy, thus ensuring the manufacturing yield and production efficiency of high-end electronic products. Attached Figure Description

[0061] Figure 1 This is a schematic diagram of the overall technical solution architecture of the SMT production line virtual debugging method based on digital twin proposed in this invention;

[0062] Figure 2 This is a schematic diagram of the core principle framework of the fault tracing mechanism based on multivariate causal graphs in this invention;

[0063] Figure 3 This is a logical flow diagram of the construction of a high-fidelity digital twin in this invention;

[0064] Figure 4 This is a flowchart illustrating the logical flow of the parameter sensitivity scanning and multi-objective optimization algorithm in this invention.

[0065] Figure 5 This is a flowchart illustrating the logical flow of virtual-real bidirectional synchronous parameter control and closed-loop verification in this invention.

[0066] Figure 6This is a schematic diagram of the multi-level interaction relationship and data flow between the physical SMT production line and the digital twin in this invention. Detailed Implementation

[0067] Please refer to Figures 1 to 6 This invention provides a virtual debugging method for SMT production lines based on digital twins. Its core lies in constructing a high-fidelity multi-physics field coupled digital twin, integrating equipment structure dynamics model, motion control logic model and process model, to achieve full-element, full-process and full-dimensional mapping of the SMT production line's operating status.

[0068] Based on this, a fault tracing mechanism based on causal reasoning and a virtual-real two-way synchronous parameter control architecture are introduced, which enables virtual debugging to not only identify abnormal phenomena, but also accurately invert the underlying mechanical, electrical or control parameter deviations that cause the phenomena, and automatically send the optimized parameter instructions to the physical equipment execution unit. At the same time, physical feedback is collected in real time to verify the correction effect, thus forming a complete closed-loop debugging process of "perception-diagnosis-decision-execution-verification".

[0069] The virtual debugging method for SMT production lines based on digital twins includes the following steps:

[0070] S1, Construct a high-fidelity digital twin of the SMT production line. The digital twin includes geometric models, kinematic models, dynamic models, control logic models, and process parameter models of the pick-and-place machine, printer, reflow oven, and conveyor system.

[0071] S2, real-time acquisition of equipment operation data through a multi-source sensor network deployed on the physical SMT production line, including servo motor encoder position signal, spindle vibration acceleration signal, pneumatic actuator pressure signal, vision positioning system coordinate deviation data, and PCB board surface temperature distribution data;

[0072] S3, The device operation data is synchronized to the digital twin in real time via industrial Ethernet, driving the digital twin to perform state reproduction and behavior simulation;

[0073] S4. When the digital twin detects a process anomaly, it activates the fault tracing analysis module based on a multivariate cause-effect graph. The fault tracing analysis module, based on the preset physical dependencies between equipment components and signal transmission paths, combined with the current anomaly feature vector, reversely derives the set of root cause parameters most likely to cause the anomaly.

[0074] S5. Based on the root cause parameter set, perform parameter sensitivity scanning and multi-objective optimization algorithm inside the digital twin to generate a set of modified parameter combinations that meet the process quality constraints.

[0075] S6, The modified parameter combination is automatically sent to the programmable logic controller or motion control card corresponding to the physical SMT production line through a secure and certified instruction channel;

[0076] S7. After the physical device executes the correction parameters, new operating data is continuously collected and fed back to the digital twin to verify the effectiveness of the correction effect, and a decision is made on whether to terminate the debugging process or enter the next round of iterative optimization based on the verification results.

[0077] In step S1, constructing a high-fidelity digital twin of the SMT production line specifically includes:

[0078] The geometry of the XYZ three-axis motion platform, rotary placement head, feeder array, and vision camera of the chip mounter is accurately modeled using a non-uniform rational B-spline surface.

[0079] A rigid-flexible coupled dynamic model of a three-axis motion platform is established based on the Lagrange equation, which takes into account the elastic deformation of the ball screw, the nonlinear frictional characteristics of the linear guide, and the electromagnetic torque fluctuation of the servo motor.

[0080] The pick-up-rotate-place action sequence of the placement head is encoded as a finite state machine, and its state transition conditions are jointly triggered by the visual recognition result, the vacuum negative pressure sensor threshold, and the cylinder positioning signal.

[0081] For reflow ovens, a three-dimensional thermal field model based on computational fluid dynamics is established, and its boundary conditions are determined by the power setpoint of the heating zone, the speed of the cooling fan, and the nitrogen flow rate in the oven.

[0082] All sub-models are co-simulated using a unified time base and event-driven mechanism, with a simulation step size set at 0.5 milliseconds. The construction process of this digital twin begins with the CAD drawings of the physical device, extracting the geometric topology of key moving parts, and using non-uniform rational B-spline surfaces to fit complex surface contours, ensuring that the geometric model and the physical entity are consistent at the millimeter level of accuracy.

[0083] Subsequently, for the dynamic behavior of the motion platform, flexible body modeling technology was introduced, and the ball screw was regarded as a distributed parameter system. Its elastic modulus, damping coefficient and inertial parameters were all obtained through experimental modal analysis.

[0084] The Coulomb friction and viscous friction terms of the linear guide rail are modeled using Stribeck curves, and their parameters are obtained by fitting low-speed crawling experimental data.

[0085] The electromagnetic torque fluctuation of the servo motor is described by the motor phase current harmonic components and the rotor position function.

[0086] In terms of control logic, the action sequence of the mounting head is abstracted into a finite state machine with 13 state nodes. Each state transition is bound to multiple Boolean conditions. For example, the "placement completed" state can only be triggered when "visual recognition is successful, the vacuum negative pressure is less than -80 kPa, and the Z-axis positioning signal is high".

[0087] The thermal field model of the reflow oven uses the Reynolds-averaged Navier-Stokes equations to solve for the gas flow inside the oven. The energy equation is coupled with the radiation heat transfer term. The thermal conductivity and specific heat capacity of the oven wall material are looked up in tables according to temperature segments to ensure that the accuracy of the thermal field simulation is within ±1 degree.

[0088] All sub-models are integrated through a federated simulation framework based on HLA (High-Level Architecture). The time management service ensures that each sub-model is strictly synchronized in 0.5 millisecond steps, and the event scheduler handles asynchronous control signals such as emergency stop commands or material shortage alarms.

[0089] In step S2, the multi-source sensor network includes:

[0090] An incremental photoelectric encoder mounted on the rear end of the X-axis servo motor has a resolution of 400,000 pulses per revolution.

[0091] A triaxial MEMS accelerometer is attached to the Z-axis ball screw nut seat, with a range of ±50 times the gravitational acceleration and a sampling frequency of 5000 Hz;

[0092] The miniature piezoresistive pressure sensor integrated inside the mounting nozzle has a measurement range of 0 to -100 kPa and a response time of 0.1 milliseconds.

[0093] A high-resolution industrial camera, with a pixel size of 2.2 micrometers and a frame rate of 120 frames per second, is positioned above the pick-and-place machine to capture the relative pose of components and pads.

[0094] The K-type thermocouple arrays arranged in each temperature zone of the reflow oven have a temperature measurement accuracy of ±0.5 degrees Celsius.

[0095] Sensor data acquisition is accomplished through distributed I / O modules. Each module has a built-in FPGA to implement hardware-level timestamp marking, with a timestamp accuracy better than 10 microseconds.

[0096] The encoder signal is input to the high-speed counter after being quadrupled in frequency, and the position calculation uses a moving average filter to suppress high-frequency jitter.

[0097] The raw accelerometer data is filtered by anti-aliasing and then stored in a ring buffer at a sampling rate of 5000 Hz; the pressure sensor output is filtered by a second-order Butterworth low-pass filter to eliminate air path pulsation interference.

[0098] The industrial camera trigger signal and the placement head descent signal are hardwired to synchronize, ensuring that the image capture moment is strictly aligned with the placement action.

[0099] The thermocouple signal is transmitted at a rate of 100 Hz after cold junction compensation and linearization.

[0100] All sensor data is transmitted to the central data aggregation node via the PROFINET industrial Ethernet protocol, with a transmission cycle of 1 millisecond. Data packets are checked using CRC32 to ensure integrity.

[0101] In step S3, the equipment operation data is synchronized to the digital twin in real time via industrial Ethernet, driving the digital twin to perform state reproduction and behavior simulation.

[0102] The synchronization mechanism adopts a master-slave time alignment strategy, with the physical production line acting as the time master station, broadcasting a global timestamp every millisecond;

[0103] The digital twin acts as a slave station. After receiving the timestamp, it adjusts the phase of the local simulation clock to ensure that the simulation time deviates from the physical time by less than 50 microseconds.

[0104] Data injection employs a hybrid interpolation-extrapolation strategy:

[0105] For high-frequency signals such as accelerometer data, cubic spline interpolation is used to fill in the intermediate values ​​within the simulation step size;

[0106] For low-frequency signals such as thermocouple data, zero-order hold extrapolation is used to the next sampling point.

[0107] During the state reproduction process, the dynamic solver of the digital twin uses the received encoder position as the boundary condition to calculate the required joint driving torque in reverse and performs consistency verification with the electromagnetic torque output by the servo motor model; if the deviation is greater than the preset threshold, the online correction process of the model parameters is triggered.

[0108] Behavioral simulation, based on the current control logic state and sensor input, drives the finite state machine to perform state transitions and updates the attitude matrix of the geometric model.

[0109] The entire synchronization process is scheduled by a dedicated real-time operating system to ensure that data processing latency is less than 2 milliseconds.

[0110] In step S4, when the digital twin detects a process anomaly, it activates the fault tracing analysis module based on a multivariate cause-effect graph. Anomaly detection is achieved through a joint criterion of multiple indicators:

[0111] Any abnormal event is considered to be a placement offset greater than 25 micrometers, a Z-axis vibration acceleration RMS value greater than 1.5 times the gravitational acceleration, a nozzle negative pressure fluctuation greater than 10 kPa, or a reflow soldering peak temperature deviating from the set value by more than 3 degrees Celsius.

[0112] The method for constructing the fault tracing and analysis module is as follows:

[0113] All key components and their interactions in the SMT production line are predefined to form a directed acyclic graph structure, where nodes represent physical components or control variables, and edges represent energy flow, information flow, or force transmission paths.

[0114] Assign a transfer function model to each edge to describe the dynamic effect of input perturbation on the output response.

[0115] The cause-effect graph contains 127 nodes and 305 directed edges, covering the entire link parameters from servo driver gain to visual calibration matrix.

[0116] When the mounting offset is detected to be greater than a preset threshold, the offset is used as observation evidence, and the posterior probability of each potential root cause node is calculated using the Bayesian inversion algorithm.

[0117] The Bayesian inversion is implemented using the Markov chain Monte Carlo method. The prior distribution is set based on the equipment's factory calibration data and historical maintenance records, and the likelihood function is constructed from the forward simulation results of the digital twin.

[0118] The parameters corresponding to nodes with a posterior probability greater than the threshold of 0.8 are selected as the root cause parameter set. The parameters include at least one of the following: X-axis guide rail preload, Y-axis servo gain, mounting head rotational inertia compensation coefficient, and visual calibration matrix offset.

[0119] For example, if the X-axis guide rail preload is insufficient, it will cause the motion platform to experience elastic hysteresis during acceleration and deceleration, which in turn will cause the mounting position to shift. This causal path is represented in the figure as a directed link from the guide rail preload to the X-axis stiffness to the platform displacement error to the mounting offset. Its transfer function is obtained by identifying the finite element modal superposition method.

[0120] In step S5, the parameter sensitivity scan and multi-objective optimization algorithm specifically perform the following steps:

[0121] In the digital twin, a Latin hypercube sampling space is constructed with the root parameter set as the center and within its allowable range of variation;

[0122] Perform a complete mounting process simulation for each sampling point and record three indicators: mounting accuracy, cycle time, and equipment energy consumption.

[0123] A non-dominated sorting genetic algorithm is used to perform multi-objective optimization on the sampling results and select the Pareto optimal solution set.

[0124] From the Pareto optimal solution set, the final combination of correction parameters is selected based on the preset weight coefficients.

[0125] The number of samples for Latin hypercube sampling is set to 10 times the root cause parameter dimension to ensure sufficient coverage of the parameter space.

[0126] Each simulation run includes 50 consecutive placement cycles to eliminate the effects of transients.

[0127] The mounting accuracy index is defined as the standard deviation of the distance between the center of the component and the center of the pad.

[0128] The cycle time metric is the average time taken for a single placement cycle;

[0129] The equipment energy consumption index is the total electrical energy consumption of the servo motor and vacuum generator within the cycle.

[0130] The non-dominated sorting genetic algorithm has a population size of 200, a crossover probability of 0.9, a mutation probability of 0.1, and a maximum number of iterations of 50. The convergence of the Pareto front is monitored by the hypervolume index, and optimization is terminated when the hypervolume increment is less than 1‰ for five consecutive generations.

[0131] The weighting coefficients are dynamically adjusted according to the priority of the current production task. For high-precision products, the weighting of mounting accuracy is set to 0.7, the weighting of cycle time is set to 0.2, and the weighting of energy consumption is set to 0.1.

[0132] If it is a large batch of standard products, the weight allocation is 0.3, 0.6, and 0.1.

[0133] The final selection of the modified parameter combination is determined by the weighted Chebyshev distance minimization criterion, ensuring that the solution is closest to the ideal point in the weighted sense.

[0134] In step S6, the modified parameter combination is automatically sent to the programmable logic controller or motion control card corresponding to the physical SMT production line through a secure and authenticated instruction channel.

[0135] The secure authentication command channel adopts a two-way authentication mechanism based on digital certificates, and the physical device only accepts parameter modification commands from the registered digital twin;

[0136] Before issuing instructions, formal verification is required to ensure that parameter values ​​are within the safe operating envelope of the device; parameter modifications are executed in an atomic transaction manner, and if any parameter write fails, all parameters are rolled back to their original values;

[0137] After the parameters take effect, the physical device returns an acknowledgment signal and the actual written value to the digital twin for comparison to check for communication errors.

[0138] The digital certificate is issued by a production-level certificate authority, the public key infrastructure supports elliptic curve cryptography, and the key length is 256 bits.

[0139] Formal verification employs interval analysis, representing the safety envelope of each parameter as a set of closed intervals, where the correction value must strictly lie within the interval.

[0140] Atomic transactions are implemented through the transaction log of the device controller. Before a write operation, a snapshot of the original parameters is saved. After the write is completed, a commit command is sent. If a commit command is not received within a timeout period, the transaction is automatically rolled back.

[0141] Communication error detection employs a dual verification mechanism:

[0142] First, compare the value sent by the digital twin with the actual written value returned by the device. If the absolute error is greater than twice the parameter resolution, it is marked as a communication failure.

[0143] Secondly, monitor the device response after the parameters take effect. For example, after modifying the servo gain, observe whether the current loop bandwidth meets expectations. If not, trigger the secondary verification process.

[0144] In step S7, verifying the effectiveness of the correction specifically includes: collecting the standard deviation of the mounting offset within three consecutive production cycles after the physical equipment executes the correction parameters;

[0145] If the standard deviation is less than the specified value, i.e. 80% of the preset tolerance bandwidth, then the correction is deemed effective.

[0146] Otherwise, the input parameters and output results of this correction attempt will be stored in the historical case library to update the cause-effect graph weights in the fault tracing analysis module and improve the accuracy of subsequent diagnosis.

[0147] The tolerance bandwidth is set according to the product specifications; it is 20 micrometers for high-density packaged devices and 35 micrometers for standard devices.

[0148] The historical case library storage structure includes five types of fields: root cause parameter set, modified parameter combination, simulation prediction index, actual measurement index, and validity label.

[0149] The cause-effect graph weight update adopts an online learning strategy, which reduces the prior probability of nodes with high posterior probability in invalid amendment cases, and enhances the transfer function gain of related edges for valid cases.

[0150] This update process performs batch optimizations monthly to ensure that the cause-effect graph evolves dynamically as the equipment ages.

[0151] The SMT production line virtual debugging system based on digital twins includes:

[0152] High-fidelity digital twin building blocks are used to construct digital twins of SMT production lines that include geometric, kinematic, dynamic, control logic, and process parameter models.

[0153] The multi-source sensor data acquisition and synchronization unit is used to acquire equipment operation data of the physical production line in real time via industrial Ethernet and drive the status update of the digital twin.

[0154] The anomaly detection and fault tracing unit is used to reverse deduce the root cause parameter set based on a multivariate cause-effect graph when a process anomaly is detected.

[0155] The parameter optimization and instruction generation unit is used to perform sensitivity scanning and multi-objective optimization within the digital twin, generate modified parameter combinations, and send them out through a secure channel.

[0156] The virtual-real closed-loop verification unit is used to receive feedback data after the physical device has performed the operation, evaluate the effect of the correction, and decide whether to terminate or iterate the debugging process.

[0157] The high-fidelity digital twin building blocks further include geometric modeling sub-units, rigid-flexible coupling dynamic modeling sub-units, finite state machine control logic modeling sub-units, and computational fluid dynamics thermal field modeling sub-units.

[0158] The geometric modeling sub-unit receives a CAD model in STEP format and generates a non-uniform rational B-spline mesh through a surface reconstruction algorithm. The mesh resolution is adaptively adjusted to balance accuracy and computational load.

[0159] The rigid-flexible coupling dynamic modeling sub-unit integrates a finite element model reduction tool, which compresses a flexible body model with millions of degrees of freedom to a hundred-order modal space, while retaining the first 20 natural frequencies and mode shapes.

[0160] The finite state machine control logic modeling sub-unit provides a graphical state editing interface, supporting the configuration of Boolean expressions for state transition conditions and simulation debugging.

[0161] The computational fluid dynamics thermal field modeling subunit uses a GPU-accelerated solver, supporting transient thermal field simulations to be completed in minutes.

[0162] The multi-source sensor data acquisition and synchronization unit is connected to an incremental photoelectric encoder, a three-axis MEMS accelerometer, a miniature piezoresistive pressure sensor, a high-resolution industrial camera, and a K-type thermocouple array.

[0163] This unit has a built-in time synchronization engine that supports the IEEE 1588 precision time protocol, ensuring that the time alignment error of multi-sensor data is less than 20 microseconds. The data preprocessing module integrates multiple filtering algorithm libraries and can automatically select the optimal filtering strategy according to the signal type.

[0164] The anomaly detection and fault tracing unit has a built-in directed acyclic graph storage module, a Bayesian inversion calculation engine, and a root cause parameter screening threshold setting module.

[0165] The directed acyclic graph storage module adopts an adjacency list structure, supporting fast forward or backward traversal; the Bayesian inversion calculation engine utilizes GPU parallel computing to accelerate Markov chain sampling.

[0166] The root cause parameter filtering threshold is configurable, with a default value of 0.8, allowing users to dynamically adjust it based on their debugging experience.

[0167] The parameter optimization and instruction generation unit integrates a Latin hypercube sampler, a non-dominated sorting genetic algorithm solver, and an atomic transaction instruction wrapper.

[0168] The sampler supports stratified sampling to handle mixed-type parameters; the genetic algorithm solver provides a visualization interface for multi-objective optimization results.

[0169] The atomic transaction instruction wrapper generates parameter write instruction blocks that conform to the IEC61131-3 standard.

[0170] The virtual-real closed-loop verification unit is equipped with a mounting accuracy statistical analyzer, a safety envelope checker, and a historical case knowledge update interface.

[0171] The mounting accuracy statistical analyzer automatically calculates the CPK process capability index; the safety envelope checker monitors the equipment's operating status in real time to ensure it does not exceed safety boundaries.

[0172] The historical case knowledge update interface supports integration with enterprise maintenance management systems to achieve knowledge accumulation and sharing.

[0173] Through the above methods and systems, this invention achieves a fundamental transformation in virtual debugging of SMT production lines, from phenomenon identification to root cause localization, from manual intervention to automatic closed-loop, and from static parameters to dynamic optimization, thereby improving the debugging efficiency and product quality stability of high-end electronic manufacturing.

[0174] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0175] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A virtual debugging method for SMT production lines based on digital twins, characterized in that, include: Construct a high-fidelity digital twin of an SMT production line, the digital twin including geometric models, kinematic models, dynamic models, control logic models and process parameter models of the pick-and-place machine, printer, reflow oven and conveyor system; The equipment operation data is collected in real time by a multi-source sensor network deployed on the physical SMT production line. The equipment operation data includes servo motor encoder position signal, spindle vibration acceleration signal, pneumatic actuator pressure signal, vision positioning system coordinate deviation data, and PCB board surface temperature distribution data. The device's operating data is synchronized to the digital twin in real time via an industrial Ethernet network, driving the digital twin to reproduce its state and simulate its behavior. When the digital twin detects a process anomaly, it activates a fault tracing analysis module based on a multivariate cause-effect graph. The fault tracing analysis module, based on the preset physical dependencies between equipment components and signal transmission paths, combined with the current anomaly feature vector, reversely derives the set of root cause parameters most likely to cause the anomaly. Based on the root cause parameter set, a parameter sensitivity scan and multi-objective optimization algorithm are executed inside the digital twin to generate a set of modified parameter combinations that meet the process quality constraints. The modified parameter combination is automatically sent to the corresponding programmable logic controller or motion control card of the physical SMT production line through a secure and certified command channel; After the physical device executes the corrected parameters, new operating data is continuously collected and fed back to the digital twin to verify the effectiveness of the correction. Based on the verification results, it is decided whether to terminate the debugging process or enter the next round of iterative optimization. The high-fidelity digital twin for constructing the SMT production line includes: The geometry of the XYZ three-axis motion platform, rotary placement head, feeder array, and vision camera of the chip mounter is accurately modeled using a non-uniform rational B-spline surface. A rigid-flexible coupled dynamic model of a three-axis motion platform is established based on the Lagrange equation, which takes into account the elastic deformation of the ball screw, the nonlinear frictional characteristics of the linear guide, and the electromagnetic torque fluctuation of the servo motor. The pick-up-rotate-place action sequence of the placement head is encoded as a finite state machine, and its state transition conditions are jointly triggered by the visual recognition result, the vacuum negative pressure sensor threshold, and the cylinder positioning signal. For reflow ovens, a three-dimensional thermal field model based on computational fluid dynamics is established, and its boundary conditions are determined by the power setpoint of the heating zone, the speed of the cooling fan, and the nitrogen flow rate in the oven. All sub-models are co-simulated using a unified time base and event-driven mechanism; The method for constructing the fault tracing analysis module based on multivariate cause-effect graphs includes: All key components and their interactions in the SMT production line are predefined to form a directed acyclic graph structure, where nodes represent physical components or control variables, and edges represent energy flow, information flow, or force transmission paths. Assign a transfer function model to each edge to describe the dynamic effect of input perturbation on the output response; When the mounting offset is detected to be greater than the preset threshold, the offset is used as observation evidence, and the posterior probability of each potential root cause node is calculated using the Bayesian inversion algorithm. The parameters corresponding to nodes with posterior probabilities greater than a threshold are selected as the root cause parameter set. The parameters include at least one of the following: X-axis guide rail preload, Y-axis servo gain, mounting head rotational inertia compensation coefficient, and visual calibration matrix offset.

2. The virtual debugging method for SMT production lines based on digital twins according to claim 1, characterized in that, The multi-source sensor network includes: An incremental photoelectric encoder mounted on the rear end of the X-axis servo motor; A three-axis MEMS accelerometer is attached to the Z-axis ball screw nut seat; A miniature piezoresistive pressure sensor integrated inside the mounting nozzle; A high-resolution industrial camera positioned above the pick-and-place machine; K-type thermocouple arrays are arranged in each temperature zone of the reflow oven.

3. The virtual debugging method for SMT production lines based on digital twins according to claim 2, characterized in that, Synchronizing the physical device's operating data to the digital twin in real time via Industrial Ethernet includes: A master-slave time alignment strategy is adopted, with the physical production line as the time master station broadcasting a global timestamp every millisecond, and the digital twin as the slave station adjusting the phase of the local simulation clock; For high-frequency signals, cubic spline interpolation is used to fill the intermediate values ​​within the simulation step size; for low-frequency signals, zero-order hold extrapolation is used to the next sampling point. The joint driving torque is calculated in reverse using the received encoder position as the boundary condition, and the consistency is verified with the electromagnetic torque output by the servo motor model.

4. The virtual debugging method for SMT production lines based on digital twins according to claim 1, characterized in that, The parameter sensitivity scanning and multi-objective optimization algorithm specifically includes: In the digital twin, a Latin hypercube sampling space is constructed with the root parameter set as the center and within its allowable range of variation; Perform a complete mounting process simulation for each sampling point and record three indicators: mounting accuracy, cycle time, and equipment energy consumption. A non-dominated sorting genetic algorithm is used to perform multi-objective optimization on the sampling results and select the Pareto optimal solution set. From the Pareto optimal solution set, the final combination of correction parameters is selected according to the preset weight coefficients, which are dynamically adjusted according to the priority of the current production task.

5. The virtual debugging method for SMT production lines based on digital twins according to claim 4, characterized in that, The mounting accuracy is defined as the standard deviation of the distance between the component center and the pad center. The cycle time is the average time consumed in a single mounting cycle; The energy consumption index of the device is the total electrical energy consumption of the servo motor and vacuum generator within a cycle.

6. The virtual debugging method for SMT production lines based on digital twins according to claim 5, characterized in that, The security authentication command channel adopts a two-way authentication mechanism based on digital certificates, and the physical device only accepts parameter modification commands from the registered digital twin; Before issuing instructions, formal verification must be performed to ensure that parameter values ​​are within the safe operating envelope of the equipment; Parameter modifications are performed in an atomic transaction manner. If any parameter write fails, all parameters are rolled back to their original values. After the parameters take effect, the physical device returns an acknowledgment signal and the actual written value to the digital twin for comparison to check for communication errors.

7. The virtual debugging method for SMT production lines based on digital twins according to claim 6, characterized in that, The verification of the effectiveness of the correction includes: The standard deviation of the mounting offset is collected over three consecutive production cycles after the physical equipment executes the correction parameters. If the standard deviation is less than the specified value, the correction is deemed valid. Otherwise, the input parameters and output results of this correction attempt will be stored in the historical case library and used to update the cause-effect graph weights in the fault tracing analysis module.

8. The virtual debugging method for SMT production lines based on digital twins according to claim 7, characterized in that, The historical case library storage structure includes five types of fields: root cause parameter set, modified parameter combination, simulation prediction index, actual measurement index, and validity label. The causal graph weight update adopts an online learning strategy, which reduces the prior probability of nodes with high posterior probability in invalid amendment cases, and enhances the transfer function gain of related edges for valid cases.

Images