Online intelligent monitoring system and method in electronic product production process

Abstract

The invention discloses an online intelligent monitoring system and method in an electronic product production process. The online intelligent monitoring system comprises a multi-sensor fusion monitoring network, a multi-channel data acquisition module, an edge calculation module, a process linkage regulation and control module and a historical database module, the multi-sensing fusion monitoring network is used for synchronously acquiring multi-physical field signals of components and substrates in the mounting, welding and curing processes of electronic products, and the multi-physical field signals comprise micro-stress distribution signals, micron-sized dynamic deformation signals and process whole-flow acoustic signals; the multi-sensor fusion monitoring network comprises a miniature fiber grating sensing array which is arranged in an embedded manner, a laser speckle interference scanning module which is arranged in a non-contact manner, and a high-sensitivity acoustic sensing network which is arranged in an array manner. A defect diagnosis result with high confidence can be converted into a specific process adjustment instruction in real time, and the specific process adjustment instruction is automatically issued to a production line for execution.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of intelligent monitoring, and particularly relates to an online intelligent monitoring system and method in the production process of electronic products. BACKGROUND

[0002] Currently, in the field of precision manufacturing of electronic products, especially in the surface mount (SMT) and advanced packaging production lines, in order to achieve quality control, various offline or online detection and monitoring technologies are mainly relied on. In the post-welding quality inspection link, automatic optical detection (AOI) and automatic X-ray detection (AXI) are widely used to identify the appearance defects (such as bridging, displacement) and internal cavities of the solder joints. For the solder paste printing process, a three-dimensional solder paste inspection (SPI) system is used to monitor the printing volume, area and height. At the process monitoring level, some high-end production lines will deploy temperature curve testers to periodically verify the reflow soldering furnace, or use the sensors built-in the equipment to monitor the motor current, air pressure and other basic parameters. In addition, destructive sampling methods such as acoustic microscope (SAM) are used to analyze the interface delamination and other reliability problems inside the package. These technologies together constitute the main technical system for ensuring production quality in the industry, each of which plays a role in a specific detection dimension.

[0003] However, the above existing technologies have systematic limitations and cannot meet the needs of full-process and intelligent quality control for high-reliability manufacturing. First, the existing technologies are mostly static and local inspections of production results (such as solder joint morphology), which cannot perform real-time and full-field monitoring of physical fields such as micro-stress distribution and dynamic deformation of components and substrates in the process, which directly reflect the assembly consistency and structural health, resulting in a large number of "hidden defects" such as stress concentration and microscopic warping, which become blind areas of detection, corresponding to the beneficial effects of the present application "comprehensively sensing hidden defects”. Secondly, various detection devices are information silos, lacking deep fusion and real-time intelligent analysis of multi-source information, and the defect judgment relies on fixed thresholds, which is easily disturbed and cannot output high-confidence root cause diagnosis, and cannot form decisions, which highlights the lack of existing technologies in "improving identification accuracy and reliability" and "implementing intelligent closed-loop regulation". Thirdly, the data fragmentation of the existing methods makes it difficult to support precise data tracing and process knowledge sedimentation from single products to the whole process, hindering the continuous improvement of quality and the self-learning evolution of the system, which is the key deficiency to be solved by the present application "building an evolvable data system”. SUMMARY

[0004] In order to overcome the above-mentioned defects of the prior art, the present application provides an online intelligent monitoring system and method in the production process of electronic products, which solves the problem that the prior art cannot realize real-time sensing of the changes of multi-physical fields in the process, and the detection, analysis and regulation links are mutually isolated.

[0005] To achieve the above object, the present application provides the following technical solutions: An online intelligent monitoring system and method in an electronic product production process, comprising a multi-sensor fusion monitoring network, a multi-channel data acquisition module, an edge computing module, a process linkage control module, and a historical database module; The multi-sensor fusion monitoring network is used for synchronously collecting multi-physical field signals of components and substrates in electronic product mounting, welding, and curing processes, and the multi-physical field signals include micro stress distribution signals, micron-level dynamic deformation signals, and process full-flow acoustic signals. The multi-channel data acquisition module is connected with the multi-sensor fusion monitoring network and is used for synchronously collecting the multi-physical field signals and transmitting the collected original signals to the edge computing module. The edge computing module is used for preprocessing the original signals, time-space synchronous alignment, and multi-source data fusion analysis, and outputs defect recognition results and confidence levels. The process linkage control module is connected with the edge computing module and a production line PLC system through an industrial Ethernet bus and is used for triggering corresponding process adjustment instructions according to the defect recognition results to realize accurate calibration of process parameters. The historical database module is connected with the edge computing module and is used for storing original signal data, defect recognition results, process adjustment parameters, and fusion model parameters to realize data tracing and model iterative optimization.

[0006] Preferably, the miniature fiber grating sensing array is embedded and deployed on a jig clamping surface and a transmission track bearing area, the grating period of the miniature fiber grating is 100-500 nm, the diameter is ≤5 μm, and the miniature fiber grating has an anti-electromagnetic interference characteristic and an electromagnetic shielding effectiveness ≥40 dB.

[0007] Preferably, the laser speckle interference scanning module is deployed above each process station, the laser wavelength is 632.8 nm, the scanning frame rate is ≥30 fps, the measurement accuracy is ≤0.1 μm, and a phase unwrapping algorithm is used to extract component deformation field data.

[0008] Preferably, the high-sensitivity acoustic sensing network is arranged around an equipment actuator and a material interaction area, the frequency response range is 20 Hz-20 kHz, the signal-to-noise ratio is ≥80 dB, an adaptive noise cancellation algorithm is built-in, and a reference signal is constructed by collecting environmental background noise to eliminate environmental interference.

[0009] Preferably, the sampling rate of the multi-channel data acquisition module is ≥1 MHz, and the time-space synchronization algorithm in the edge computing module realizes a time synchronization accuracy ≤10 μs.

[0010] Preferably, the fusion machine learning model is a CNN-LSTM hybrid architecture, and the dynamic weight calculation model introduces the material parameter, size parameter and process type feature of the component, and the material parameter of the component includes the elastic modulus and the thermal expansion coefficient.

[0011] Preferably, an online intelligent monitoring method in an electronic product production process comprises the following steps: S1: synchronously collecting micro stress distribution signals, micron-level dynamic deformation signals and process whole-process acoustic signals in electronic product mounting, welding and curing processes through a multi-sensing fusion monitoring network; S2: synchronously collecting the collected multi-physical field original signals through a multi-channel data acquisition module, and transmitting the original signals to an edge computing module; S3: the edge computing module pre-processes the original signals to obtain stress wavelength shift data, component deformation field data and acoustic spectrum feature data, respectively; S4: time stamp matching and spatial position association of the stress wavelength shift data, the component deformation field data and the acoustic spectrum feature data are realized through a time-space synchronization algorithm; S5: adjusting the feature weights of the three types of data based on a dynamic weight calculation model, and outputting defect recognition results and confidence through fusion analysis of the aligned multi-source data based on a fusion machine learning model; S6: if the confidence of the defect recognition result exceeds a preset threshold, a process linkage control module triggers a process adjustment instruction, and the process parameter is accurately calibrated through a production line PLC system; S7: storing the original signal data, defect recognition results, process adjustment parameters and fusion model parameters in steps S1 to S6 to a historical database for model iteration optimization and data traceability.

[0012] Preferably, the pre-processing of step S3 comprises: converting the micro stress distribution signals into wavelength shift data through a fiber grating demodulation module, and eliminating the influence of environmental temperature drift through a temperature compensation algorithm; filtering and Fourier transforming the acoustic signals to obtain acoustic spectrum feature data.

[0013] Preferably, the multi-source data fusion analysis of step S5 adopts a three-level fusion algorithm framework of time-space alignment, feature enhancement and weight distribution, and the attention mechanism is adopted in the feature enhancement stage to screen key features in each dimension.

[0014] Preferably, the process adjustment instruction of step S6 includes jig clamping force adjustment range, welding temperature curve correction parameter and equipment component replacement warning; the inducing process and key process parameters of the implicit defect are deduced through time sequence correlation analysis of multi-source data, and the process parameter is accurately adjusted in reverse.

[0015] The technical effects and advantages of the online intelligent monitoring system and method in the production process of electronic products of the present application are as follows: 1. The present application builds a multi-physical field synchronous monitoring network containing micro-stress, micro-deformation and full-process acoustics, and the system can capture the internal physical state changes that cannot be found by traditional visual detection and are closely related to the long-term reliability of the product, thereby fundamentally expanding the coverage and depth of defect detection.

[0016] 2. The edge computing module of the present application performs spatio-temporal synchronous alignment and fusion analysis on multi-source data, effectively overcoming the shortcomings of single signal being easily disturbed and information being one-sided. Combined with a dynamic weight calculation model, the system can adaptively optimize the analysis strategy, thereby outputting a defect recognition result with higher confidence, and significantly reducing the false positive and false negative rates.

[0017] 3. The present application can convert high-confidence defect diagnosis results into specific process adjustment instructions in real time through the process linkage control module, and automatically issue them to the production line for execution. This online and automatic closed-loop control changes the traditional process adjustment mode that relies on manual experience and lags behind in batches to real-time and precise individualized process calibration, greatly improving the adaptive ability and process stability of the production line.

[0018] 4. The historical database module of the present application stores the full-chain information from raw data, analysis results to control actions, not only establishing a "manufacturing process digital archive" for each product to realize accurate traceability of quality problems, but also providing a continuous iterative optimization data basis for the fusion of machine learning models, so that the detection and decision-making ability of the system can continuously evolve independently with the accumulation of production data, forming a continuously improved intelligent quality control ecosystem.

[0019] 5. The present application designs each sensing unit specifically for industrial scenarios, combined with high-precision spatio-temporal synchronous acquisition, ensuring the stability and reliability of signal acquisition and analysis in a real production environment with multiple disturbances and fast rhythms, laying a solid foundation for the large-scale and industrial deployment of the system. BRIEF DESCRIPTION OF DRAWINGS

[0020] Fig. 1 is a system flowchart of an online intelligent monitoring system and method in the production process of electronic products proposed by the present application; Fig. 2 is a method flowchart of an online intelligent monitoring system and method in the production process of electronic products proposed by the present application. DETAILED DESCRIPTION

[0021] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present application, and not all the embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0022] It should be noted that, in this document, relational terms such as first and second and the like can be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. The terms including, including, or any other variation thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that includes a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article, or apparatus. Without more limitations, the elements defined by the statement including... are not excluded from the process, method, article, or apparatus including the elements.

[0023] Reference Figs. 1-2 The present application provides an online intelligent monitoring system and method in the production process of electronic products. The system includes a multi-sensor fusion monitoring network, a multi-channel data acquisition module, an edge computing module, a process linkage control module, and a historical database module. The multi-sensor fusion monitoring network integrates an embedded miniature fiber grating sensor array, a non-contact laser speckle interference scanning module, and an arrayed high-sensitivity acoustic sensing network for synchronous acquisition of micro-stress distribution, micron-level dynamic deformation, and full-process acoustic signals in the processes of mounting, welding, and curing. The multi-channel data acquisition module is responsible for synchronous acquisition and transmission of these multi-physical field original signals. The edge computing module has built-in fusion machine learning models and dynamic weight calculation models for signal preprocessing, time-space synchronization alignment, and multi-source data fusion analysis, outputting defect recognition results and confidence. The process linkage control module triggers the process parameter precision calibration of the production line PLC system through industrial Ethernet according to the defect results. The historical database module stores full-process data and model parameters, supporting traceability and iterative optimization. The corresponding method covers complete steps from signal synchronous acquisition, fusion analysis to process adjustment and data storage. The present application realizes real-time, accurate, and closed-loop intelligent control of hidden defects in the production process.

[0024] Embodiment 1 The embodiment provides an online intelligent monitoring system and method in the production process of electronic products for real-time detection and prevention of component damage caused by abnormal mounting pressure. The specific implementation content includes: Purpose of implementation: To solve transient faults such as excessive pressure and tilting caused by nozzle wear, vacuum pressure fluctuations or program deviations during the mounting of precision chips (such as 0.4mm pitch QFP), and to prevent component cracks, solder paste bridging or cold solder joints.

[0025] Implementation System: This embodiment involves all modules of the system described in claim 1. Its unique feature lies in the collaborative perception and millisecond-level response to microsecond-level mounting events under high-precision spatiotemporal synchronization.

[0026] Implementation steps: S1: Placement process triggered. A miniature fiber optic grating sensor array (embedded in a dedicated QFP fixture) acquires dynamic micro-stress distribution signals in the clamping area at a rate of 1MHz, sensitively capturing the impact force and static clamping force at the moment of chip contact. A laser speckle interferometry scanning module (running in 120fps high-speed mode) simultaneously acquires interference fringes between the chip pin plane and the PCB pad plane, monitoring changes in flatness and parallelism during the bonding process. A high-sensitivity acoustic sensor network (beam focused on the nozzle) simultaneously acquires continuous acoustic signals during component contact and vacuum release.

[0027] S2: The synchronous clock card of the multi-channel data acquisition module ensures that the sampling clocks of the three signals are from the same source, and the timestamp synchronization accuracy reaches 2μs.

[0028] S3: Edge computing module preprocessing: The stress signal is compensated for temperature drift and demodulated to obtain a nanometer-level wavelength offset curve; the deformation image sequence is solved in real time to generate the height difference time series data of the four corners of the chip relative to the center; the acoustic signal is denoised and STFT (short time Fourier transform) to obtain a time spectrum diagram, and the clarity index of the "pop" sound feature point of the vacuum release is extracted.

[0029] S4: The spatiotemporal synchronization algorithm aligns the "force peak moment" of the stress curve, the "stable fit moment" of the deformation data, and the "vacuum release moment" of the acoustic spectrum according to a unified trigger signal, and maps them to the same spatial coordinate system (with the chip center as the origin).

[0030] S5: The dynamic weighting calculation model calculates the weight allocation based on the QFP chip's size (10mm × 10mm), material (plastic body, copper alloy leads), and "precision mount" process type: deformation weight. Stress weight Acoustic weight The integrated machine learning model analyzes the aligned data accordingly. The CNN branch identifies "the lower right corner made contact first, with an approximately 0.7μm tilt" from the deformation data; "the peak impact force exceeds the limit by 15%" from the stress data; and "the vacuum release sound is weak, suggesting incomplete release" from the acoustic data. The LSTM network integrates the temporal relationships and determines that "slight blockage of the nozzle caused poor vacuum release, and the component was forcibly pressed, resulting in tilt and overstress," outputting the defect result with a confidence level of 98.5%.

[0031] S6: The process linkage control module makes a decision within 50ms. First, it sends an emergency stop command to the pick-and-place machine PLC via OPCUA to stop the subsequent actions of the current nozzle and calls the backup nozzle to complete the remaining placement of this board. Second, it generates a maintenance work order, prompting "Clean or replace the 3rd nozzle on the X-axis". Finally, for the chip that has been placed, it adds a "Marked, X-Ray recheck recommended" flag to its MES record.

[0032] S7: The complete waveform raw data (compressed), all intermediate features, final diagnostic results, trigger commands and device response status of this event are completely stored in the historical database module, and the data packets are associated with a unique board serial number and event ID.

[0033] Implementation Results: This system successfully intercepted a potential batch of placement defects caused by a minor equipment defect. Compared to traditional methods (which rely on subsequent AOI inspection of appearance and cannot detect internal stress damage and minor tilting), this system moves the inspection process forward and deepens it, achieving online, in-situ, and non-destructive process compliance verification. Statistics show that after implementing this system, the early failure rate caused by abnormal placement pressure decreased by more than 85%.

[0034] Example 2 This embodiment provides an online intelligent monitoring system and method for electronic product manufacturing processes, used for BGA cold solder joint detection and reverse optimization of process parameters after reflow soldering. Specific implementation methods include: Objective: To overcome the limitations of traditional AOI and sampling X-Ray in that they cannot see the bottom solder joints of BGA and have large blind spots. By monitoring the non-uniform deformation during the soldering cooling process, it can detect local cold solder joints / poor solder joints caused by mismatched furnace temperature profiles and uneven heat capacity of PCB layout online and in full inspection, and realize the automated and precise reverse adjustment of process parameters.

[0035] Implementation System: This embodiment highlights the value of combining the multiphysics field signal (especially the micrometer-level dynamic deformation signal) in claim 1 with the dynamic weight calculation model (introducing the coefficient of thermal expansion) in claim 6.

[0036] Implementation steps: S1: In the forced air cooling zone at the reflow oven exit, a laser speckle interferometry scanning module (deployed above) performs a full-board scan on each PCB exiting the oven, simultaneously acquiring its dynamic deformation signal throughout the cooling process (within approximately 60 seconds). Meanwhile, a miniature fiber optic grating sensor array deployed on the cooling zone support synchronously monitors the evolution of the overall thermal stress experienced by the PCB during cooling.

[0037] S2-S4: Deformation video streams and stress time-series streams are acquired, transmitted, preprocessed, and spatiotemporally aligned simultaneously. Deformation data is processed into "deformation field cube" data with time as the third dimension.

[0038] S5: The dynamic weighting calculation model obtains the specifications (ball diameter, pitch), material (SAC305 solder balls, FR-4 substrate), and position on the board of the BGA from the MES. Based on the difference in thermal expansion coefficients between SAC305 and FR-4 and the BGA's position on the board edge, the model assigns extremely high weights to the deformation signal. The fusion model (whose deformation branch has been trained on numerous BGA soldering cases) analyzes the "deformation field cube," focusing on the deformation (shrinkage) rate and final displacement difference between the BGA region and the surrounding substrate, as well as between the corners and the center of the BGA. The model identifies the deformation hysteresis phenomenon in the upper left corner region of the BGA during the mid-cooling stage, and ultimately there is a residual height difference of 1.5μm between it and the center, which is consistent with the deformation characteristics of "local cold soldering," outputting defect results with a confidence level of 97%.

[0039] S6: The process linkage control module performs a composite operation. First, the PCB is marked as "quality control interception" in the MES and guided to the re-inspection station. The key action is reverse optimization: the module calls the historical database to calculate the real-time temperature field of the PCB as it passed through each temperature zone in the oven, based on deformation data. Through comparative analysis, it is found that when the board passes through the 5th temperature zone (corresponding to the upper left corner of the BGA on the board), the measured peak temperature in this area is about 8°C lower than the set value. The module immediately sends a process correction command to the reflow oven's temperature control system via OPCUA: "For PCB type XXXX, increase the set temperature of the heater in the 5th temperature zone by 10°C and fine-tune the airflow speed in this zone to improve heat convection uniformity." This correction automatically takes effect for subsequent production of the same model of PCB.

[0040] S7: Store all data from this event and add the causal relationship "BGA local cold solder joint - low temperature in zone 5" as a new knowledge entry to the case library. The system automatically starts a batch analysis task to retrieve recent production data for the same PCB model and assess the general necessity of this process correction.

[0041] Implementation Results: The system represents a leap from "detecting individual defects" to "diagnosing and automatically correcting systemic process deviations." It not only acts as a "quality inspector" but also plays the role of a "process engineer." By directly linking online detection data with process equipment, the closed-loop cycle for process optimization has been shortened from "days to weeks" of traditional manual analysis, experimentation, and verification to "minutes," significantly improving the production line's adaptability and process stability.

[0042] Example 3 This embodiment provides an online intelligent monitoring system and method for the electronic product manufacturing process, used for early identification and model self-learning of colloidal void defects during the curing process. Specific implementation details include: Objective: To address internal void defects caused by residual air bubbles, poor flowability, or improper curing rates during the curing process of underfill adhesives. These defects are invisible after curing and traditionally rely on expensive scanning acoustic microscopes (SAMs) for random sampling, making online full inspection impossible. This embodiment aims to achieve early online screening through acoustic-stress correlation analysis and drive the autonomous evolution of the model using continuous production data.

[0043] Implementation System: This embodiment highlights the value of fusing acoustic signals and micro-stress distribution signals throughout the entire process in claim 1, as well as the application of the idea of ​​inferring latent defects through multi-source data in claim 10, and ultimately demonstrates the ability of the historical database module to support iterative optimization of the model.

[0044] Implementation steps: S1: During the heating and curing stage of the curing oven, a high-sensitivity acoustic sensor network deployed at specific locations inside the oven chamber synchronously collects the acoustic signals emitted by the colloid during the cross-linking reaction and bubble bursting. Simultaneously, a miniature fiber optic grating sensor attached to the back of the PCB near the corresponding chip location synchronously monitors changes in the micro-stress distribution signal generated by the shrinkage of the colloid on the substrate.

[0045] S2-S4: Acoustic and stress signals are acquired and preprocessed simultaneously (the acoustic signal is analyzed primarily for high-frequency components above 20kHz to capture acoustic emission events from the rupture of tiny bubbles). The system uses physical location correlation to ensure that the analyzed acoustic events and stress changes originate from the same colloid under the same chip.

[0046] S5: For this type of novel defect, the initial fusion model has limited discriminative ability. Model analysis revealed abnormal, sparse pulse events in the acoustic signal at specific frequency bands (suspected microbubble rupture), and the stress release curve showed slight fluctuations at the corresponding time points, but the characteristics were not significant. The model output a warning of "suspected colloidal microvoids," but the confidence level was only 85%, which did not reach the threshold (95%) for automatically triggering process adjustments.

[0047] S6: The process linkage control module adopts a conservative strategy: it does not immediately adjust the curing oven process parameters (to avoid introducing new problems due to misadjustment), but adds a "process data abnormal, it is recommended to strengthen the final test" mark to the MES record of the PCB.

[0048] S7: This board failed during post-processing functional testing, and offline SAM (Self-Aware Acoustic Machining) confirmed the presence of a colloidal void with a diameter of approximately 200μm. The quality engineer located this warning record in the historical database system's human-machine interface, marked it as a confirmed "colloidal void" positive sample, and associated it with the SAM image as evidence. The system's backend active learning module was triggered, retrieving all recent "suspicious" data with similar acoustic-stress characteristic patterns but not yet confirmed, as well as a large amount of normal curing process data, automatically constructing an incremental training dataset. One week later, the system used the incremental data to fine-tune the fusion model. The new model enhanced its ability to identify the correlation pattern between "high-frequency acoustic pulses and micro-stress fluctuations." When encountering similar situations again, the new model's output confidence increased to 96%, reaching a level where it could automatically alarm and suggest adjustments to the dispensing amount or preheating temperature.

[0049] Implementation Results: This embodiment demonstrates the system's ability to handle unknown or rare defects. It is not a static expert system designed from the ground up, but a "living system" with continuous learning and evolution capabilities. Through human-in-the-loop annotation and automatic incremental learning, the system can continuously expand its knowledge boundaries, incorporating new defect patterns into its reliable automatic identification range. This achieves autonomous growth in quality control capabilities, perfectly supporting the design goal of "achieving iterative model optimization" in the claims.

[0050] Example 4 This embodiment provides an online intelligent monitoring system and method for electronic product manufacturing processes, used for full-process stress history traceability of high-value circuit boards. Specific implementation details include: Objective: To construct an objective, continuous, and tamper-proof "manufacturing process stress digital twin" for high-reliability, high-value multilayer PCBs used in aerospace, medical equipment, and other fields. When intermittent failures occur during later testing or use, the exact process, time, and magnitude of the failure cause can be precisely located by tracing back the history of all mechanical stresses experienced throughout the entire production process. This enables root cause analysis (RCA) for quality accountability, process improvement, and supplier management.

[0051] Implementation System: This embodiment is a model of the combined application of the multi-sensor fusion monitoring network (especially the micro fiber grating sensor array) and the historical database module in claim 1, embodying the core value of "data-driven traceability".

[0052] Implementation steps: S1: From the moment the PCB enters the SMT line until it leaves the line, fiber optic grating sensor arrays embedded in every node that may have mechanical contact with the PCB, such as the transport track, printer stops, pick-and-place machine fixtures, reflow oven guides, in-circuit test (ICT) bed of pins, and connector crimping fixtures, continuously and synchronously collect micro-stress distribution signals of every contact, clamping, and crimping the PCB undergoes. Each signal carries a high-precision timestamp and position code.

[0053] S2-S4: Multiple data acquisition modules distributed throughout the production line upload stress data streams from each node to the edge computing module in real time. The edge computing module not only performs real-time over-limit judgment, but its core task is to integrate and stitch together the data from all discrete nodes into a continuous "stress-time-location" three-dimensional data curve covering the entire manufacturing cycle, i.e., the "stress history," based on the PCB serial number and a unified global timeline.

[0054] S5: At the real-time monitoring level, the edge computing module performs real-time analysis of the "stress history". If any transient stress peak is found to exceed the safety threshold set for different processes (such as mounting pressure > 30N, test pressure > 50N), an alarm will be triggered immediately.

[0055] S6: Suppose a military communication board experiences a channel failure after a 72-hour aging test. Traditional analysis methods (such as electrical performance testing and visual inspection) are ineffective. In this case, the process linkage control module does not directly activate, but the quality engineer can input the serial number of the faulty board through the dedicated traceability interface of the historical database module.

[0056] S7: The system immediately retrieved the complete "stress history" and presented it visually. Engineers analyzed the data and discovered a sudden, sharp spike of 80N on the stress history curve. The timestamp corresponded to the board's testing cycle on the automated test equipment (ATE) station. Further analysis of the ATE station's high-definition video log at that moment revealed that a test probe had been mechanically jammed and pressed down excessively. Thus, the root cause of the fault was precisely pinpointed to the abnormal operation of the test equipment, rather than a process issue in the SMT production process. This "stress history" became irrefutable objective evidence.

[0057] Implementation Results: Quality management has been transformed from a "black box" model based on statistical sampling to a transparent "white box" model based on objective data throughout the entire process. The "stress history" is like a product's "manufacturing process electrocardiogram," providing a unique "digital fingerprint" for high-value products. It's not only used for post-production traceability, but the accumulated big data can also be used to analyze the sensitivity of different designs and different supplier materials to manufacturing stress, providing feedback to R&D and procurement departments to achieve design optimization and upgrade supply chain quality control.

[0058] Example 5 This embodiment provides an online intelligent monitoring system and method for the manufacturing process of electronic products, used for comprehensive monitoring and prevention of assembly stress in multi-chip modules. Specific implementation details include: Purpose of implementation: In the final assembly stage of system-in-package (SiP) or complex multi-chip module (MCM), especially during the mechanical assembly of heat sinks or housings (such as screw fastening and fastener installation), to monitor in real time the stress distribution and deformation caused by uneven assembly forces throughout the module (especially on chips of different sizes and materials), and to prevent hidden reliability risks such as chip microcracks, substrate warping, or thermal interface material (TIM) crushing caused by assembly stress concentration.

[0059] Implementation System: This embodiment is a comprehensive application of the system described in claim 1 in scenarios involving multiple processes, multiple physical fields, and complex components, fully demonstrating the powerful capabilities of multi-sensor fusion and edge intelligent analysis.

[0060] Implementation steps: S1: At the workstation of the automatic screw fastening machine, when the robotic arm executes the fastening program, the system simultaneously starts: 1) a fiber optic grating array embedded in the bottom of the module carrier and the contact surface of the heat sink base to simultaneously collect the evolution signal of macroscopic assembly stress distribution throughout the fastening process; 2) a laser speckle interferometry scanning module deployed above the workstation to simultaneously scan the microscopic dynamic deformation field on the surface of key large chips such as CPU, GPU, and Memory; 3) a high-precision acoustic sensor integrated on the fastening bit to simultaneously collect the torque-angle related sound characteristics during the screwing process of each screw.

[0061] S2-S4: The three signals are acquired synchronously with ultra-high precision (thanks to the TSN network). In the preprocessing stage, the acoustic signals are used to accurately segment the tightening stage of each screw (contact, fit, torque increase, stop).

[0062] S5: The dynamic weight calculation model assigns differentiated sensor weights to different spatial locations based on the known attributes of each chip (e.g., GPU has the largest size and silicon is brittle; Memory chips are small but sensitive to flatness). The fusion model performs real-time analysis. During the torque increase phase of tightening the fourth screw (located in a corner of the module), the model found: a) the fiber optic sensor showed a sharp increase in local pressure at that corner with uneven distribution; b) the speckle interferometer detected a slight upward warping (0.9 μm) at the corner of a nearby small Memory chip; c) acoustic feature analysis showed an abnormal frictional spectrum of the screw, suggesting possible resistance or misalignment of the threads. The model's overall judgment was "assembly sequence leading to stress concentration and local warping risk," with a confidence level of 99%.

[0063] S6: The process linkage control module immediately sends an emergency stop command to the fastening machine PLC and highlights an alarm on the HMI, indicating "Stress concentration was detected due to the fastening of the 4th screw. It is recommended to switch to diagonal fastening sequence or reduce the torque at this point." At the same time, the module is marked as "Stress relaxation monitoring is required before proceeding to the next process."

[0064] S7: All multiphysics data, analysis processes, interruption decisions, and recommended solutions for this event have been meticulously recorded in the historical database. This data serves as invaluable information for optimizing the product's assembly process (such as determining the optimal locking sequence and torque curve) and can be extended to similar products.

[0065] Implementation Results: This extends quality control from chip-level "manufacturing" to system-level "assembly." Through multiphysics fusion, the system can "sense" how assembly forces are transmitted and distributed within complex heterogeneous systems, thus providing early warning and intervention before irreversible damage occurs. This significantly improves the long-term reliability and yield of high-end module products and reduces the risk of later field failures due to assembly problems.

[0066] Comparative Example 1 This comparison model provides both traditional single-eye visualization testing and manual parameter adjustment modes, including: System Implementation: Typical traditional production line configuration. Several automated optical inspection (AOI) devices perform 2D visual inspection of solder joints after soldering and cleaning; one offline 3DX-ray inspection device is used for the first piece of the day or for sampling inspection of hidden solder joints such as BGAs according to AQL standards; one flying probe tester is used for sampling electrical performance testing. Each device operates as an information silo, and process parameter adjustments rely on quality engineers summarizing defect rate reports from each station daily, combined with experience-based manual judgment and manual modifications on the device HMI.

[0067] The implementation process and problems became apparent: Taking the production of a batch of motherboards for smartwatches as an example. AOI inspection showed "good" solder joint appearance, and sampling X-rays revealed no solder bridging or voids; flying probe testing passed. However, when the products were assembled into complete devices and subjected to drop tests, approximately 3% of the specific batches experienced sensor malfunction. After tedious rework analysis, dye penetration testing and microscopic sectioning revealed that the root cause was insufficient placement pressure during the mounting of the 01005-sized micro-bead inductors due to slight wear on the nozzle, leading to a "tombstoning" tendency and poor wetting on one side of the solder joint. The drop impact completely severed the already fragile connection.

[0068] Implementation effect analysis: Inspection capability dimension: Traditional methods are completely ineffective. AOI is a top-down 2D inspection, which cannot determine the micron-level tilt of components or the wetting of the solder base. X-ray is a side-view 2D projection, which is not sensitive to the tiny standing defects of 01005 components, and the sampling rate is extremely low. This defect is a typical latent defect of "critical electrical contact, insufficient mechanical strength".

[0069] Response and adjustment time: From the occurrence of the problem (production day) to its final location (after the whole machine drop test), the cycle can take several weeks. The adjustment process is even slower: engineers need to speculate on possible causes (it could be solder paste, reflow soldering, or mounting), and then conduct DOE experiments to verify them one by one. The whole process relies on personal experience and is full of uncertainty, making it impossible to achieve accurate and rapid process recovery.

[0070] Data traceability and prevention: There is no data to prove that the defect occurred during the placement process, by which placement head, or at what time. Precise traceability is impossible, let alone data-driven predictive maintenance (it is impossible to predict nozzle wear trends).

[0071] Compared with this system: This system, through the technical path described in Example 1, can identify "micro-standing" risks in real time during production by integrating micro-deformation (detecting tilt) and micro-stress (detecting insufficient pressure) signals, and immediately calibrate or deactivate the faulty placement head. Example 4's capability ensures that any problem can be traced back to a specific workstation and moment. The process linkage capabilities demonstrated in Examples 2 and 5 fundamentally optimize process parameters and prevent problem recurrence. Therefore, this system fundamentally solves the four major pain points of traditional methods: large detection blind spots, delayed feedback, lack of traceability, and inefficient adjustments. It achieves a paradigm shift in quality control from "post-inspection, offline sampling, and experience-driven" to "process monitoring, online full inspection, and data-driven."

[0072] Compared with Examples 1-5 and Comparative Example 1, the online intelligent monitoring system shown in Examples 1-5 forms a fundamental paradigm opposition with the traditional model represented by Comparative Example 1. The core differences are reflected in three aspects: detection dimension, response mechanism, and data value.

[0073] Comparative Example 1 clearly reveals the inherent limitations of traditional methods relying on Optical Inspection (AOI) and Destructive Sampling (X-Ray). These methods can only capture macroscopic appearance defects, completely failing to perceive the intrinsic process physical states that determine product reliability—such as mounting stress, solder deformation, and assembly stress distribution. This results in latent defects becoming blind spots in detection and significant feedback lag. In stark contrast, Example 1, by fusing micro-stress and micro-deformation signals, intercepted abnormal mounting pressure and component tilting in real time; Example 2 utilized full-field deformation analysis to diagnose localized BGA solder joint defects online and automatically optimized the oven temperature process; Example 3, through acoustic-stress correlation analysis, achieved early screening of colloidal void defects and model self-evolution. These examples collectively demonstrate that this system, through multi-physics fusion sensing, extends the detection range from "appearance" to "process essence," achieving in-situ, holographic perception of defect formation mechanisms.

[0074] Furthermore, Comparative Example 1 represents an open-loop, delayed "firefighting" management approach, where process adjustments rely entirely on manual experience and time-consuming trials. Examples 1 to 5, on the other hand, systematically demonstrate millisecond-level autonomous closed-loop capabilities from "perception" to "optimization." Examples 1 and 5 achieve real-time intervention and shutdown warnings for placement and assembly anomalies; Example 2 elevates the system's role from "quality inspector" to "process engineer," achieving self-correction of process parameters based on data back-reasoning. More importantly, Example 4 constructs a "stress history" covering the entire manufacturing cycle, providing a precisely traceable "digital twin" for high-value products, whereas Comparative Example 1 is completely powerless in fault tracing. In summary, through the technical path verified in Examples 1-5, this system completely upgrades quality control from the old paradigm of sampling-based "statistical inference" and "manual intervention" represented by Comparative Example 1 to a new paradigm of "intelligent proactive assurance" based on full perception, real-time closed-loop, and continuous learning.

[0075] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of protection of the claims.

[0076] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An online intelligent monitoring system for the manufacturing process of electronic products, characterized in that, include: Multi-sensor fusion monitoring network, multi-channel data acquisition module, edge computing module, process linkage control module, and historical database module; The multi-sensor fusion monitoring network is used to synchronously acquire multi-physical field signals of components and substrates in the mounting, soldering and curing processes of electronic products. The multi-physical field signals include micro-stress distribution signals, micron-level dynamic deformation signals and acoustic signals throughout the process. The multi-sensor fusion monitoring network includes an embedded micro fiber grating sensor array, a non-contact laser speckle interferometry scanning module, and an array-structured high-sensitivity acoustic sensor network. The multi-channel data acquisition module is connected to the multi-sensor fusion monitoring network to synchronously acquire multi-physics field signals and transmit the acquired raw signals to the edge computing module. The edge computing module is used for preprocessing the raw signal, spatiotemporal synchronization alignment, and multi-source data fusion analysis, outputting defect identification results and confidence levels; the edge computing module has a built-in fusion machine learning model and dynamic weight calculation model; The process linkage control module is connected to the edge computing module and the production line PLC system via an industrial Ethernet bus. It is used to trigger corresponding process adjustment commands based on defect identification results to achieve precise calibration of process parameters. The historical database module is connected to the edge computing module to store raw signal data, defect identification results, process adjustment parameters, and fusion model parameters, enabling data traceability and model iterative optimization.

2. The online intelligent monitoring system for the production process of electronic products as described in claim 1, characterized in that, The miniature fiber grating sensor array is embedded in the fixture clamping surface and the transmission track bearing area. The grating period of the miniature fiber grating is 100-500nm, the diameter is ≤5μm, and it has anti-electromagnetic interference characteristics and electromagnetic shielding effectiveness ≥40dB.

3. The online intelligent monitoring system for the manufacturing process of electronic products as described in claim 1, characterized in that, The laser speckle interferometry scanning module is deployed above each process station. The laser wavelength is 632.8nm, the scanning frame rate is ≥30fps, the measurement accuracy is ≤0.1μm, and the phase unwrapping algorithm is used to extract the component deformation field data.

4. The online intelligent monitoring system for the manufacturing process of electronic products as described in claim 1, characterized in that, A high-sensitivity acoustic sensor network is deployed around the device actuator and in the material interaction area. The frequency response range is 20Hz-20kHz, the signal-to-noise ratio is ≥80dB, and the built-in adaptive noise cancellation algorithm constructs a reference signal by collecting environmental background noise to eliminate environmental interference.

5. The online intelligent monitoring system for the production process of electronic products as described in claim 1, characterized in that, The sampling rate of the multi-channel data acquisition module is ≥1MHz, and the spatiotemporal synchronization algorithm in the edge computing module achieves a time synchronization accuracy of ≤10μs.

6. The online intelligent monitoring system for the manufacturing process of electronic products as described in claim 1, characterized in that, The integrated machine learning model is a CNN-LSTM hybrid architecture. The dynamic weight calculation model incorporates component material parameters, size parameters, and process type features. Component material parameters include elastic modulus and coefficient of thermal expansion.

7. An online intelligent monitoring method for the electronic product manufacturing process based on the system of claim 1, characterized in that, Includes the following steps: S1: Simultaneously collect micro-stress distribution signals, micron-level dynamic deformation signals, and acoustic signals throughout the electronic product mounting, soldering, and curing processes via a multi-sensor fusion monitoring network; S2: The multi-physics field raw signals are synchronously acquired through the multi-channel data acquisition module and transmitted to the edge computing module; S3: The edge computing module preprocesses the original signal to obtain stress wavelength offset data, component deformation field data and acoustic spectrum characteristic data respectively; S4: Realize the timestamp matching and spatial location association of stress wavelength offset data, component deformation field data and acoustic spectrum characteristic data through a spatiotemporal synchronization algorithm; S5: Adjust the feature weights of the three types of data based on the dynamic weight calculation model, perform fusion analysis on the aligned multi-source data through the fusion machine learning model, and output defect identification results and confidence scores; S6: If the confidence level of the defect identification result exceeds the preset threshold, the process linkage control module triggers the process adjustment command and completes the precise calibration of process parameters through the production line PLC system; S7: Store the original signal data, defect identification results, process adjustment parameters and fusion model parameters from steps S1 to S6 into the historical database for model iteration optimization and data traceability.

8. The online intelligent monitoring method for the manufacturing process of electronic products as described in claim 7, characterized in that, The preprocessing in step S3 includes: converting the micro-stress distribution signal into wavelength offset data through the fiber optic demodulation module, eliminating the influence of environmental temperature drift through the temperature compensation algorithm; filtering and Fourier transforming the acoustic signal to obtain acoustic spectrum characteristic data.

9. The online intelligent monitoring method for the manufacturing process of electronic products as described in claim 7, characterized in that, The multi-source data fusion analysis in step S5 adopts a three-level fusion algorithm framework of spatiotemporal alignment, feature enhancement, and weight allocation. In the feature enhancement stage, an attention mechanism is used to screen key features in each dimension.

10. The online intelligent monitoring method for the manufacturing process of electronic products as described in claim 7, characterized in that, The process adjustment instructions in step S6 include the adjustment range of fixture clamping force, welding temperature curve correction parameters, and equipment component replacement warnings; by using time-series correlation analysis of multi-source data, the inducing process and key process parameters of hidden defects are deduced, and the process parameters are adjusted in reverse with precision.

Images