Printed circuit board anti-oxidation method

Abstract

The invention relates to an anti-oxidation method for a printed circuit board, which comprises the following steps of: dynamically acquiring three-dimensional characteristics and process environment information of a board surface through a multi-channel sensing system, and performing multi-modal characteristic fusion and normalization to realize accurate modeling of sound field propagation parameters and simulation of liquid flow states for different PCB structures and environments. And a non-contact thickness monitoring and digital twin feedback mechanism is utilized to identify a local thickness abnormal region, and ultrasonic excitation is adaptively adjusted based on an optimization algorithm to realize closed-loop adjustment of the thickness of the liquid film. The uniformity of coating thickness distribution and the intelligent level of process response are improved, and the consistency and data traceability of the anti-oxidation coating process are enhanced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of printed circuit board oxidation resistance process, ultrasonic microfluidic thickness control and digital twin process optimization, and particularly relates to a printed circuit board oxidation resistance method. BACKGROUND

[0002] As the core basic component of various electronic products, the surface oxidation resistance treatment process of printed circuit board (PCB) is directly related to the electrical performance stability and product service life. At present, the PCB oxidation resistance coating generally adopts mechanical spraying, immersion, brushing, electrochemical deposition and other process routes, and the main goal is to form a dense and uniform oxidation resistance coating film on the surface of the board and the circuit to prevent copper oxidation and improve welding reliability. However, with the rapid development of electronic manufacturing towards multi-variety, small batch, customization and high-density interconnection, the structure of printed circuit board tends to be more complex, including multi-layer stacking, irregular edges, fine line width gaps, dense hole arrays, etc., which brings great challenges to the uniformity control of traditional oxidation resistance coating.

[0003] Most of the existing PCB oxidation resistance coating processes are mainly controlled by general transmission mechanisms or spraying heads in partitions, and some high-end production lines introduce multi-segment mechanical coating heads or regional pneumatic spraying and cutting devices, which adjust the mechanical partition and set the speed parameters to improve the local thickness consistency of complex board types. In addition, for products with high flatness requirements, electrochemical deposition and local mask method have also been tried to achieve certain thickness area control. However, the above-mentioned solutions are essentially based on mechanical or static parameters, which are difficult to adapt to irregular PCBs with dramatic changes in spatial structure and various small-batch multi-variety switching conditions.

[0004] In recent years, microfluidic coating and ultrasonic assisted distribution technology have attracted attention. Through microfluid control or ultrasonic assisted disturbance, the flow characteristics of the coating in narrow gaps, blind holes, corners and other complex areas that are difficult to access are improved. Some academic research explores the controllability of fluid migration under the action of sound field, and improves the local coating coverage. However, this kind of technology has not been fully combined with high-precision digital sensing, real-time feedback and system-level control, and lacks the ability of industrial-level full-process intelligent regulation and control.

[0005] The existing PCB oxidation resistance coating thickness control technology has the following significant deficiencies:

[0006] (1) Limited regional adaptability - mechanical partition, fixed spraying head and other means are suitable for PCBs with simple structure, but for complex feature areas such as narrow gaps, steps, irregular edges and local multiple holes, the oxidation resistance coating thickness often appears serious distortion or even visible uniformity "blind area".

[0007] (2) Automatic sensing and adaptive control is weak - lack of real-time thickness monitoring means for each area, unable to dynamically adjust process parameters according to the structural complexity of different board types and environmental instance feedback, and lack of closed-loop response capability with actual coating effect, resulting in a large fluctuation in product consistency under multi-variety change or batch personalized demand. SUMMARY

[0008] The present application provides a printed circuit board oxidation-resistant method, which aims to solve one or more of the problems existing in the prior art mentioned in the background.

[0009] The printed circuit board oxidation-resistant method provided by the present application specifically comprises:

[0010] S1: Collecting real-time board surface three-dimensional topography data and local environmental parameters of different types of printed circuit boards in the oxidation-resistant treatment process unit, including temperature, humidity and solution concentration, to obtain environmental difference data under multiple positions and multiple working conditions.

[0011] S2: Performing normalization preprocessing on the three-dimensional topography data and environmental parameter data to eliminate the influence of batch and board type differences on subsequent feature modeling.

[0012] S3: Based on the normalized three-dimensional topography data, using a digital twin modeling method to generate a sound field propagation parameter mapping model corresponding to each PCB structure type to represent the coating fluid distribution characteristics of each local area under the action of ultrasonic waves.

[0013] S4: Inputting the real-time collected environmental parameters into the sound field propagation parameter mapping model, combining the current PCB structure characteristics, and reasoning to obtain an ultrasonic excitation parameter set for different areas, including frequency, phase, and amplitude distribution, to realize thickness uniformity prediction.

[0014] S5: According to the ultrasonic excitation parameter set, driving a programmable ultrasonic transducer array to form a multi-zone dynamic sound field in the oxidation-resistant coating process unit to guide the distribution of coating liquid on the PCB surface and control the fluid film formation characteristics.

[0015] S6: Using an embedded non-contact thickness sensor to monitor the thickness of the liquid film in each area in real time to obtain fluid film distribution feedback data to reflect the actual film formation and thickness uniformity state.

[0016] S7: Inputting the fluid film distribution feedback data into the digital twin system, comparing the target thickness distribution, and determining whether there is a thickness abnormality in each area, including local thickening or thinning.

[0017] S8: For the detected abnormal area, the local ultrasonic excitation parameters are optimized and adjusted based on the sound field propagation parameter mapping model to realize real-time sound field correction, so as to compensate or dissipate the liquid film thickness of the abnormal area and realize adaptive thickness control.

[0018] S9: The process parameters, thickness real-time monitoring data and sound field response parameters in each process are recorded into the process digital twin database as the basis for subsequent parameter tracking, model optimization and process continuous improvement.

[0019] S10: If it is judged that the current board type or environmental parameter appears an abnormal state beyond the predetermined range, an automatic alarm mechanism is triggered and the coating control process is suspended to ensure production safety and product consistency.

[0020] The printed circuit board oxidation-resistant method provided by the application has the following beneficial effects:

[0021] (1) Realize the highly adaptive coating of the thickness uniformity of complex board type, special-shaped structure and high-layer PCB. Through high-precision, batch-independent normalized modeling of the real three-dimensional topography and local environmental parameters of the printed circuit board surface, and with the aid of the sound field propagation parameter mapping model trained by the digital twin system, the optimal ultrasonic excitation parameters can be dynamically inferred according to the structure and process state of any spatial region of the PCB, the spatial customized distribution of the sound field under complex structure is realized, the thickness consistency of the oxidation-resistant layer in each region is significantly improved, and the uniformity problems such as edge, narrow gap, multi-step and high-density region that cannot be covered by traditional multi-section mechanical coating head or electrochemical method are solved.

[0022] (2) Effectively improve the process adaptability and flexible automation level in the manufacturing scene of multiple varieties and high variability. All control steps and parameter optimization of the application completely rely on digital twin and AI inference, without the need to design a coating head, adjust a mechanical section or a complex process fixture for each board type, and have wide-range automatic adjustment capability and real-time fast automatic variety switching. Only software and model updating is needed to adapt to new specification PCB, which greatly reduces the product line change and new line development cost, and effectively supports small-batch diversified flexible production.

[0023] (3) Significantly simplify the equipment structure, reduce the system complexity and maintenance difficulty. Compared with the traditional process relying on multi-stage spray head, mechanical zoning or multi-tank electrochemical deposition, the microfluidic array driven by ultrasonic wave does not need to arrange a complex mechanical zoning system and multi-path liquid pipeline, but only relies on a high-density ultrasonic transducer array and dynamic waveform control to implement regional control, reduces mechanical moving parts, reduces fault points and on-site maintenance difficulty, and significantly improves the stability, reliability and macroscopic integration of the process equipment.

[0024] (4) Greatly improve production efficiency and process controllability. Real-time three-dimensional morphology and environmental parameter multi-modal acquisition, intelligent normalization processing and digital twin model reasoning can complete the adaptive generation of process parameters in a very short time, and the programmable transducer array realizes the dynamic assignment of multi-region sound field with adjustable spatial distribution, so that the oxidation-resistant coating process of each PCB almost does not need manual intervention, and the batch production switching time is greatly shortened. In the complex plate type mass production process, the thickness uniformity and consistency of each batch of products can be guaranteed, and the production automation, intelligentization and unmanned operation are realized.

[0025] (5) Whole-process closed-loop monitoring and self-learning optimization capability. Through the embedded high-density non-contact thickness sensor array, the thickness of each region coating liquid film is collected and fed back in real time, combined with the digital twin process database, abnormality discrimination and optimization closed-loop algorithm, the local abnormality (thickening / thinning) can be automatically corrected, and the parameter configuration is dynamically self-adaptive optimized. With data sedimentation and model iteration, the system thickness control precision and reliability continue to improve, and it has the ability of abnormal process alarm, traceability, knowledge reuse and historical factory experience migration, realizing the self-optimizing manufacturing process of continuous evolution. BRIEF DESCRIPTION OF DRAWINGS

[0026] FIG. 1 is a flowchart of a printed circuit board oxidation-resistant method. Figure 1 FIG. 1 is a flowchart of a printed circuit board oxidation-resistant method.

[0027] FIG. 2 is a sub-flowchart of a printed circuit board oxidation-resistant method. Figure 2 FIG. 2 is a sub-flowchart of a printed circuit board oxidation-resistant method.

[0028] FIG. 3 is another sub-flowchart of a printed circuit board oxidation-resistant method. Figure 3 FIG. 3 is another sub-flowchart of a printed circuit board oxidation-resistant method. DETAILED DESCRIPTION

[0029] Embodiments of the present application are described in detail below with reference to the attached drawings, wherein the same or similar components or components having the same or similar functions are denoted by the same or similar reference numerals throughout. The embodiments described below with reference to the drawings are exemplary and are intended only for the purpose of explaining the present application, and should not be understood as limiting the present application.

[0030] The following disclosure provides many different embodiments, or examples, for implementing different structures of the present application. For the purpose of simplifying the present application, the components and arrangements of the specific examples are described in the following. Of course, they are only examples and the purpose is not to limit the present application. In addition, the present application can repeatedly refer to numerals and or letters in different examples, and such repetition is for the purpose of simplification and clarity, which itself does not indicate the relationship between the various embodiments and or arrangements discussed. In addition, the present application provides examples of various specific processes and materials, but those skilled in the art can realize the application of other processes and the use of other materials.

[0031] As shown in the accompanying drawings Figure 1 The present application provides a printed circuit board oxidation-resistant method, specifically comprising:

[0032] S1: Collecting real-time board surface three-dimensional topography data and local environmental parameters of different types of printed circuit boards in the oxidation-resistant treatment process unit, including temperature, humidity and solution concentration, to obtain environmental difference data under multiple positions and working conditions.

[0033] S2: Performing normalization preprocessing on the three-dimensional topography data and environmental parameter data to eliminate the influence of batch and board type differences on subsequent feature modeling.

[0034] S3: Based on the normalized three-dimensional topography data, using digital twin modeling method to generate sound field propagation parameter mapping model corresponding to each PCB structure type, to characterize the coating fluid distribution characteristics of each local area under the action of ultrasonic waves.

[0035] S4: Inputting the real-time collected environmental parameters into the sound field propagation parameter mapping model, combining the current PCB structure characteristics, and reasoning to obtain ultrasonic excitation parameter groups for different areas, including frequency, phase, and amplitude distribution, to realize thickness uniformity prediction.

[0036] S5: According to the ultrasonic excitation parameter group, driving the programmable ultrasonic transducer array to form a multi-zone dynamic sound field in the oxidation-resistant coating process unit, to guide the distribution of coating liquid on the PCB surface and control the fluid film formation characteristics.

[0037] S6: Using embedded non-contact thickness sensor to monitor the liquid film thickness of each area in real time, obtaining fluid film distribution feedback data to reflect the actual film forming and thickness uniformity state.

[0038] S7: Inputting the fluid film distribution feedback data into the digital twin system, comparing the target thickness distribution, and judging whether there is thickness abnormality in each area, including local thickening or thinning.

[0039] S8: For the detected abnormal area, optimizing and adjusting the local ultrasonic excitation parameters based on the sound field propagation parameter mapping model, realizing real-time sound field correction to compensate or dissipate the liquid film thickness of the abnormal area, and realizing adaptive thickness control.

[0040] S9: Synchronizing the process parameters, thickness real-time monitoring data and sound field response parameters in each process to the process digital twin database as the basis for subsequent parameter tracking, model optimization and process continuous improvement.

[0041] S10: If it is judged that the current board type or environmental parameter appears an abnormal state exceeding the predetermined range, an automatic alarm mechanism is triggered and the coating control process is suspended to ensure production safety and product consistency.

[0042] The step S1: collecting real-time board surface three-dimensional topography data and local environmental parameters of different types of printed circuit boards in the anti-oxidation treatment process unit, including temperature, humidity and solution concentration, to obtain environmental difference data under multiple positions and working conditions. Specifically, it includes:

[0043] S1.1: Numbering and type identification of the printed circuit board to be processed, based on the MES system or artificial intelligence image recognition algorithm, positioning the PCB board type and size information entering the anti-oxidation treatment process unit, to obtain the structure label and initialize the collection task, providing accurate board type index for subsequent collection of three-dimensional topography data and environmental parameters.

[0044] For the printed circuit boards entering the anti-oxidation treatment process unit, an automatic numbering and board type identification process is used to generate accurate board structure index, which initializes the execution object for the collection of three-dimensional topography and environmental parameters.

[0045] Using the automatic data interaction interface based on the manufacturing execution system (MES) and the industrial camera image acquisition module (parameters: resolution, field of view coverage), the new board batch number and the unique identification code in the factory are bound, and the unique number of each PCB is output.

[0046] Further, by integrating deep learning image recognition algorithm (such as based on YOLO-x, ResNet, etc. Model, parameters: number of training set samples, recognition resolution, detection confidence), the structure feature points of the real-time captured board surface visual image are extracted and classified, realizing the automatic identification of PCB board type (including layer structure, size specification, key special-shaped features) and board surface space posture, and outputting the structure label.

[0047] Further, by combining the number and structure label, a board surface structure index table is constructed and written into the process data collection task initialization queue, ensuring that the subsequent three-dimensional topography scanning and environmental parameter correlation collection data stream correspond one-to-one to the current board type and its spatial layout.

[0048] Further, through the process flow scheduling parameters in the MES system, the processing record of each batch of boards, production batch, historical process parameters and quality traceability code are associated, a perfect traceability link is established for three-dimensional topography data and real-time environmental parameters, and the data correlation integrity is realized.

[0049] Through automatic numbering and plate type recognition and structure index establishment method, the original physical plate is converted into a digital acquisition target with a unique number and a structure label, realizing the precise assignment of subsequent multi-modal sensor acquisition tasks and data closed-loop management.

[0050] Exemplarily, in the flexible PCB oxidation-resistant processing batch production scene, the MES system automatically generates a batch number

[0051] A20240601-01001 to A20240601-01030, the industrial camera collects the top image of each PCB entering the front desk with an 800 million pixel resolution. The deep learning recognition model identifies the current PCB as a 10-layer composite board (size: 305mm x 250mm, board thickness 1.5mm, edge corner with 3mm zigzag extension area) based on 30000 training sets of different plate types, with an identification confidence of 0.997 and a structure label code of S[10L-305x250-E1]. The number and structure label are automatically written into the acquisition queue and synchronized to the process data set. The MES system retrieves the historical processing information of the board and adjusts the matching acquisition strategy. During execution, the newly detected 2-layer thick board (number A20240601-01005, structure S[2L-305x250-E3]) is also identified through image recognition, accurately outputting structure labels such as irregular edge and layer deviation. The imported number and structure label provide one-to-one mapping support for the accurate collection of subsequent point cloud scanning and environmental monitoring data and process flow tracing. The above process realizes 100% automatic numbering and structure label recognition of flexible and irregular PCBs, with a cumulative recognition accuracy of 99.96%, greatly improving the data acquisition accuracy of special plate process adaptation and the traceability of the entire life cycle of the plate.

[0052] S1.2: For the numbered printed circuit board, through the integrated three-dimensional topography sensor array, based on optical structured light scanning or laser confocal imaging method, perform plate surface three-dimensional topography data acquisition to obtain original three-dimensional point cloud data containing surface relief, step, pore, edge contour and other structural details, realize digital description of different plate type surface features.

[0053] Input the unique identification number and its structure label of the numbered printed circuit board, call the integrated three-dimensional topography sensor array, and start the plate surface three-dimensional topography data acquisition process. The optical structured light scanning method (parameters: beam wavelength 405nm, projection accuracy 10μm, scanning speed 150mm / s) is adopted to realize the grating projection scanning of the whole area of the printed circuit board surface, and the high-density spot array reflection image is obtained.

[0054] Further, by laser confocal imaging method (parameters: laser emission wavelength 532 nm, focal length adjustment step 1 pm, scanning step 5 pm), high longitudinal resolution three-dimensional profile collection of the plate surface concave and convex, steps and pore microstructure is realized, and the fine structure restoration ability of complex area is enhanced.

[0055] Further, combined with multi-angle sensor pose adaptive algorithm (parameters: sensor attitude adjustment range ± 30°, multi-view synthesis times ≥ 4), for the plate surface irregular edge, deep hole channel and multi-layer step region, the spatial layout of the sensor array is dynamically adjusted to realize full coverage three-dimensional point cloud data compensation for the structure blind area.

[0056] Further, a multi-channel raw point cloud fusion reconstruction algorithm (parameters: spatial alignment accuracy 0.5 pm, fusion step 2 pm) is used to synthesize the raw point cloud data collected by each sensor node into a unified coordinate system, and output high-precision three-dimensional point cloud data set P(x, y, z) containing surface relief height, step position, pore morphology, edge contour and other structural details.

[0057] Through point cloud data processing and abnormal noise filtering algorithm (parameters: isolated point distance threshold 3 pm, density filtering kernel radius 5 pm), the environmental reflection noise and mechanical jitter interference are removed, the three-dimensional morphology data is cleaned and optimized, and the usability and accuracy of the data are ensured.

[0058] Through the above chain processing mode, the physical surface of the numbered printed circuit board is converted into high-precision, structure label corresponding original three-dimensional point cloud data, realizing the digital representation of different board type complex surface features.

[0059] For example, for a 12-layer high-density PCB board (number 1234567890) with irregular edge and through-hole structure, the three-dimensional point cloud data set P(x, y, z) is obtained by the above chain processing mode.

[0060] A20240601-01012, an anti-oxidation process application of structural label S[12L-380x340-E7], uses a projection structured light three-dimensional sensor, sets the scanning accuracy to 8 μm, scans all surfaces and through holes row by row, and obtains 120 million original surface point clouds. Through laser confocal scanning, 220,000 high-density point data are additionally added in the corner area for refining the steps, protrusions and micro-hole structures in the special-shaped area. Multi-angle pose adjustment realizes full coverage of deep grooves and blind holes, and finally outputs a unified high-density point cloud dataset P(x, y, z) after point cloud fusion processing. After point cloud abnormal noise filtering, the total amount of point cloud reaches 14.2 million, and the single-point positioning error is better than 0.9 μm. The finally output point cloud data is archived in the digital twin system, and corresponds to the structural label and number one by one, effectively reflecting all key structural details of the board type, and providing a high-reliability three-dimensional topography data basis for subsequent sound field parameter adaptive optimization and thickness uniformity control model training. The application results show that for the same line width / pitch density board type, the whole process of point cloud data collection takes less than 70 seconds, and the special-shaped area reconstruction rate reaches 99.85%, which significantly improves the adaptability and automation efficiency of the process parameter optimization of the complex board type.

[0061] S1.3: Based on the spatial positioning of the three-dimensional topography data, high-precision environmental parameter sensor modules are arranged on the surface of the printed circuit board and its surrounding area, and environmental parameters such as temperature, humidity and concentration of anti-oxidation liquid solution are obtained in real time through a digital bus, so that each spatial point has corresponding environmental context information at each moment, forming a synchronous environmental parameter data stream.

[0062] S1.4: Time stamp correlation is performed on the three-dimensional topography collection data and the corresponding environmental parameter data, and the data stream is one-to-one mapped with the physical position of the board surface through a space-time synchronization algorithm, realizing multi-dimensional fusion representation of the board surface structure details and process environment state, and providing a unified data structure for subsequent normalization processing and digital twin modeling.

[0063] S1.5: The three-dimensional topography data and environmental parameter data obtained synchronously are subjected to integrity check and abnormality rejection, and a multi-dimensional fault-tolerant correction algorithm is used to automatically screen and correct missing fragments and abnormal noise, so as to ensure the accuracy and high quality of the data input, thereby laying a solid data foundation for process difference modeling between batches and board type adaptive control.

[0064] The step S2: performing normalization preprocessing on the three-dimensional topography data and environmental parameter data to eliminate the influence of batch and board type differences on subsequent feature modeling. Specifically, it includes:

[0065] S2.1: Dimension calibration processing is performed on the three-dimensional topography data to output dimension-consistent three-dimensional topography standardized data, so as to ensure that the three-dimensional topography features collected by different devices and measurement batches are consistent in spatial scale and unit system.

[0066] The three-dimensional topography raw point cloud data P(x, y, z) completed with integrity and abnormality elimination, the input data including spatial coordinates, structure label and collection batch code of each point.

[0067] The dimensional calibration algorithm (parameters: target spatial unit mm, reference standard block size L ref = 100 mm, device calibration error threshold δ = 0.1%) is used to realize spatial unit conversion of the raw point cloud data, to unify the coordinate system output by all measuring devices to standard metric units, and to realize consistency of spatial dimensions across devices and batches.

[0068] Further, the spatial scale factor automatic fitting algorithm (parameters: least square fitting window N points = 10000) is used to solve the actual scale conversion coefficients S i (i ∈ {x, y, z}) of the raw point cloud and the standard model in x, y and z directions by taking the built-in standard reference block point set as a reference. The following formula is used for calculation:

[0069]

[0070] Wherein, S i is the scale calibration coefficient of each coordinate direction, x std,j is the coordinate of the jth point in the standard reference space, x raw,j is the original collection coordinate of the jth point, and N is the total number of fitting points.

[0071] Further, the spatial coordinate normalization mapping is applied to all point cloud data for coordinate calibration through linear scale transformation. The corrected three-dimensional point cloud data P'(x', y', z') can be represented as:

[0072]

[0073] Further, the spatial distortion correction compensation algorithm (parameters: hyperbolic fitting term number k = 3, residual threshold σ = 5 μm) is used to establish a standard reference geometric surface residual model for small distortions caused by nonlinearity of the measuring head, thermal drift, etc., to perform high-order residual correction, and to realize micro-scale unification of the full section point cloud.

[0074] Further, through multi-batch cross verification, the same plate type data of different batches and the standard plate database are spatially distributed and aligned, and difference analysis is performed to automatically compare the spatial coordinate consistency and screen out possible systematic drift errors between batches.

[0075] Through the above continuous processing mode, the original three-dimensional topography data is converted into three-dimensional topography standardized data with dimension standardization and consistent spatial scale, laying a strict consistency foundation for subsequent distribution equalization and batch-independent feature extraction, and realizing the standardization of multi-batch and multi-device data in spatial dimension.

[0076] For example, for A20240601-01012 12-layer PCB board (structure label S[12L-380x340-E7]), the number of point clouds collected is 14.2 million, and the original device calibration unit is μm. The target unit of the dimension calibration parameter is set to mm, the built-in standard reference block length L ref = 50 mm is selected, and the calibration error threshold δ is 0.05%. Through the scaling factor fitting algorithm, Sx = 0.00102, Sy = 0.00100, and Sz = 0.00103 are calculated. Based on the above calibration coefficients, unit conversion and spatial calibration are completed for all point cloud coordinates, and the spatial positioning accuracy of all point clouds after calibration is improved to <1.2 μm. Further, according to the standard block geometric surface fitting residual, a three-order spatial distortion correction is performed, and the evaluation residual fluctuation is lower than 3 μm. Cross-batch point cloud and standard board database automatic alignment analysis shows that the maximum spatial deviation between the next batch and the current batch is not more than 0.6 μm, which meets the requirements of high-precision manufacturing process. The three-dimensional topography standardized data P'(x', y', z') after dimension calibration and spatial scale standardization is finally output as the standard input for subsequent distribution equalization processing, ensuring that the spatial scale of the thickness uniformity control model data input is completely consistent.

[0077] S2.2: Apply a distribution equalization algorithm to the three-dimensional topography standardized data after dimension calibration, and perform data normalization processing on the local feature distribution according to the batch statistical mean and variance parameters to obtain batch-independent three-dimensional topography normalized features.

[0078] S2.3: The interval mapping and extreme value reduction method is used for the collected environmental parameter data (temperature, humidity, solution concentration) to map to the standard environmental parameter feature space, output normalized environmental parameter features, and ensure the consistency of multi-working condition parameter input.

[0079] S2.4: Perform multi-modal feature fusion on the three-dimensional topography normalized features and the environmental parameter normalized features, extract the joint feature set through feature splicing and standardized mutual information screening algorithm, and improve the input effectiveness of the subsequent sound field propagation parameter mapping model.

[0080] S2.5: Perform robustness evaluation and normality test on the above joint feature set, use Mahalanobis distance analysis and PCA visualization means to judge the batch distribution overlap degree after feature mapping, output batch-independent unified input features, and provide a standardized and high-reliability data foundation for the subsequent digital twin modeling stage.

[0081] The step S3: based on the normalized three-dimensional topographic data, using digital twin modeling method to generate the sound field propagation parameter mapping model corresponding to each PCB structure type, to characterize the coating fluid distribution characteristics of each local area under the action of ultrasonic waves. As shown in Figure 2 , specifically comprising:

[0082] S3.1: feature extraction is performed on the normalized three-dimensional topographic data, a surface grid reconstruction algorithm is used to generate a set of PCB structure parameters containing key geometric features, to provide high-precision input features for subsequent sound field modeling.

[0083] S3.2: based on the set of PCB structure parameters, a finite element simulation method is used to calculate the local sound field distribution under different ultrasonic excitation parameters, to obtain sound field distribution simulation data, to characterize the action law of sound energy in complex geometric structures.

[0084] The input object collected is the normalized pre-processed PCB three-dimensional structure parameter set, which specifically includes a standardized spatial grid model, surface fluctuation coordinates and structure feature label parameters.

[0085] A method based on finite element analysis (FEA) is used (parameters: element type is three-dimensional acoustic element SOLID227, spatial discrete grid density D mesh = 50 μm, boundary conditions are assembled with ultrasonic transducer sound source excitation with plate normal, output variables are spatial sound pressure level and particle velocity), to realize the modeling of the sound field propagation characteristics inside the complex PCB geometric structure under the action of different ultrasonic excitation parameters.

[0086] Further, through parameterized excitation configuration (parameter group setting: frequency f ∈ [20 kHz, 120 kHz], phase φ ∈ [0, 2π], amplitude A ∈ [0.1, 5.0], single simulation step Δf = 2 kHz, Δφ = π / 12), the spatial distribution and energy density response of the sound field in each local area under different frequency, phase and amplitude input combinations are simulated and calculated respectively. The following acoustic wave equation is used for finite element discrete solution:

[0087]

[0088] Where p(x, y, z) is the sound pressure, k = 2πf / v is the acoustic wave number, and v is the sound velocity.

[0089] Further, based on the boundary conditions of complex structures, acoustic impedance boundary and absorption boundary conditions are applied to realize high-fidelity sound field simulation of PCB medium and surrounding coating liquid, to obtain the local area sound energy distribution matrix

[0090] Further, by performing spatial statistical analysis on the partitioned sound energy distribution results, characteristic parameters such as energy peak points, node distribution, standing wave field distribution, and sound energy gradient distribution are extracted to form a sound field response spatial feature set E S (region).

[0091] Further, through full parameter scanning simulation, frequency, phase, amplitude, and structure geometric parameters are covered by full factorial method, and comprehensive sound field distribution simulation data set is output, providing physical basis for subsequent fluid dynamics field simulation under the action of sound field.

[0092] Through finite element simulation analysis, the normalized PCB structure parameters and the local sound field distribution results under multiple groups of ultrasonic excitation conditions are mapped to structure-sound field response simulation data, significantly improving the prediction accuracy of sound wave energy distribution in complex geometric regions, and providing key input for multi-physical field coupling modeling of thickness uniformity adaptive control.

[0093] For example, for the twelve-layer special-shaped PCB board of structure label S[12L-380x340-E7], a three-dimensional acoustic finite element simulation model is used, with a spatial discrete element size of 50 μm and a model node number of 640000. The ultrasonic excitation frequency setting range is 40 kHz to 80 kHz, with a step of 2 kHz, and the amplitude is set to 0.5 to 3.0. The normal incidence sound source is set in each key region. Each group of (f, A) combinations is simulated to obtain the maximum sound energy density of 50.2 dB at the corner slit, the minimum sound energy density of 46.7 dB at the center region, and the node standing wave region length of about 1.4 mm. Through simulation result analysis, the sound energy distribution is compared with the corresponding sound field energy overflow point of the board structure fluctuation, and the improvement effect of the sound energy distribution on the complex structure adaptability is verified. Finally, the sound field distribution simulation data table related to frequency-amplitude-space point is output, providing accurate physical support for subsequent multi-physical field flow-sound coupling modeling and thickness control algorithm.

[0094] S3.3: Based on the sound field distribution simulation data and the physical parameters of the coated fluid, a multi-physical field coupling modeling method is used to generate a fluid dynamics parameter variation mapping table to describe the motion behavior and thickness distribution trend of the fluid film in each local region under the action of the sound field.

[0095] The input data includes sound field distribution simulation data under different excitation parameters obtained by finite element simulation, standardized board three-dimensional geometric parameter set, and physical parameters (density, viscosity, surface tension, etc.) of the coated liquid.

[0096] A flow-sound multi-physical field coupling modeling method is used (parameters: the fluid model is an incompressible Newtonian fluid, the sound field model is a steady sound wave field, and the flow-sound coupling boundary condition is a dynamic boundary in the sound source action area) to realize high-precision modeling of the fluid film dynamics behavior under the action of ultrasonic excitation.

[0097] Further, the movement process of the coating liquid is solved by coupling the fluid mechanics control equation (Navier-Stokes equation), the surface tension model and the acoustic pressure field coupling term, and the basic form is as follows:

[0098]

[0099] Wherein, ρ is the fluid density, is the local flow rate, p is the static pressure, μ is the fluid viscosity, is the interfacial tension term, is the volume force generated by the acoustic field.

[0100] Further, the volume force under the action of the acoustic field is calculated, and the following volume force expression is used:

[0101]

[0102] Wherein, k is the compressibility coefficient, p ac is the instantaneous sound pressure in the acoustic field.

[0103] The two-dimensional / three-dimensional multi-field coupling finite element-finite volume joint algorithm (parameters: time domain step 1ms, spatial grid minimum unit 20μm, iterative convergence threshold 10 -6 ) is used to simulate the dynamics of the fluid film formation, migration, aggregation and dissipation in the local area in space and time, and the instantaneous thickness distribution of the liquid film in each area and the change process of the main dynamic parameters are obtained.

[0104] Further, through a plurality of different excitation parameters and structure characteristics injection, batch parallel coupling simulation is executed, fluid dynamics response data is systematically collected, and parameter response mapping under the complex action of the acoustic field-structure-fluid is obtained.

[0105] The multivariate statistical regression or principal component analysis (PCA) method driven by data is used to induce and arrange the multi-factor mapping relationship among the acoustic field excitation parameter-structure parameter-fluid property-dynamics response, and a fluid dynamics parameter change mapping table is established. The mapping table includes the acoustic field parameter input of each area and the corresponding film thickness average, extreme value, uniformity index and dynamic evolution law.

[0106] Through the above processing mode, the acoustic field distribution simulation data obtained in the simulation stage is highly coupled with the coating liquid physical property parameters, and a fluid dynamics parameter change mapping table for complex plate structure is output, so as to realize the description and prediction ability of the fluid film movement and thickness distribution trend under the action of the acoustic field in each local area.

[0107] Exemplarily, for the heterogeneous multi-layer PCB board of structure label S[12L-380x340-E7], the input sound field distribution simulation data contains an extended sound pressure field with a frequency range of 40 kHz to 80 kHz and an amplitude range of 0.8-3.0 MPa, and the input coating liquid parameters are density 1090 kg / m 3 , viscosity 0.6 mPa·s, and surface tension 30.5 mN / m. A three-dimensional flow-acoustic finite element-finite volume coupling simulation model is used, with a time step of 1 ms and a minimum spatial grid of 20 μm, and a free surface tension dynamic boundary is introduced at the edge. The simulation results show that in the corner gap area, the peak value of the acoustic volume force reaches 2.1 N / m 3 , and the local liquid rapidly penetrates and forms a maximum thickness point of 2.9 μm during the transient period, and the fluid migration speed is more than 1.7 times that in the central area. In each key feature area, by performing PCA dimensionality reduction analysis on the sound field parameters, structure characteristics, and multiple sets of film thickness data, a multivariate fitting function of frequency-amplitude-plate structure parameters to film thickness is further established. Finally, the dynamic mapping table of the batch sample is output, realizing high-precision modeling of the local thickness evolution trend in the complex structure area, and providing a schedulable data basis and physical law basis for intelligent parameter optimization and thickness closed-loop control.

[0108] S3.4: Model training is performed on the relationship between the normalized three-dimensional topographic features and the fluid dynamics parameter change mapping table using a supervised learning algorithm, and a sound field propagation parameter mapping model is established for different PCB structure types, so that the model has the ability to predict the sound field distribution and fluid response in the area according to any input structure characteristics.

[0109] The input object is a batch-independent three-dimensional topographic normalized feature and fluid dynamics parameter change mapping table, which has been screened through the previous multi-physical field simulation, including structure feature parameters and excitation-fluid response corresponding relationship.

[0110] A multivariate supervised learning algorithm (parameters: model type is gradient boosting regression tree GBRT, input feature dimension is structure space partition variable, sound field excitation group parameter, and local geometric attribute, and output is local sound field distribution response and target fluid dynamics index) is used to realize the mapping relationship modeling function between the normalized three-dimensional topographic features and the fluid dynamics parameter change mapping table.

[0111] Further, by batch labeling sample input, a training data set is established using three-dimensional structure features and fluid thickness response, a five-fold cross-validation algorithm is used to evaluate the model generalization performance, the tree depth, learning rate, and feature distribution weight are optimized, the most representative feature subset for coating area thickness prediction is selected, and the output accuracy of the model under the complex board type of heterogeneous and multi-layer is improved.

[0112] The normalized structure parameters, excitation factors and environmental covariates are sorted by feature importance scoring algorithm, the contribution of different input items to the prediction of regional sound field and fluid thickness distribution is quantified, redundant features with high collinearity and poor generalization ability are removed, and a simplified and efficient input data structure is generated.

[0113] Further, a supervised regression residual analysis algorithm is used to set L2 regular constraint for the output sound field propagation parameter mapping model, and the difference between the model prediction value and the observed value in the fluid dynamics parameter response table is adopted to calculate the residual sum of squares and determine the coefficient R 2 , continuously adjust the hyperparameter combination, and improve the stability and anti-bias of the model under extreme structure changes and special process conditions.

[0114] Through the above processing mode, the complex nonlinear relationship between the normalized structure features and the fluid response parameters is converted into a sound field propagation parameter mapping model that can be directly called, which realizes accurate prediction of regional sound field distribution and fluid dynamics response according to any input structure feature and parameter condition, and meets the needs of thickness uniformity prediction and parameter back calculation of printed circuit board irregular structures.

[0115] For example, for the irregular multi-layer PCB packaging board with structure label S[12L-380x340-E7], a GBRT model (tree depth 5, regression subtree number 120, learning rate 0.09) is used, the input sample size is 18000 groups, and the features include local spatial partitioning, surface normal variation, sound field frequency distribution, amplitude gradient, and coating fluid physical property description vector. In the training data, the fluid film thickness index (range 1.2 μm to 3.4 μm) and the sound field response parameters generated by simulation are used as reference. After normalization of the model input, the generalization index is evaluated by five-fold cross-validation, the average absolute error is 0.15 μm, and R 2 reaches 0.957. Feature importance analysis shows that the surface three-dimensional fluctuation standard deviation and local sound field energy distribution are the dominant inputs, and the thickness response prediction accuracy of the edge and corner, gap area is significantly improved. The final output is a trained and optimized sound field propagation parameter mapping model, which can automatically realize sound field-thickness prediction mapping in different board types, number of layers and environmental conditions, providing high robustness and high precision data support for real-time working condition reasoning and closed-loop control of digital twin systems.

[0116] S3.5: The trained and verified sound field propagation parameter mapping model is archived to the digital twin system standard model library, and through the model interface, it provides a callable basic operation unit for subsequent real-time working condition parameter input, ultrasonic excitation parameter reasoning and thickness uniformity simulation, and the output of this step is completed.

[0117] The step S4: input the real-time collected environmental parameters into the sound field propagation parameter mapping model, combine the current PCB structure characteristics, infer and obtain the ultrasonic excitation parameter group for different regions, including frequency, phase and amplitude distribution, to realize thickness uniformity prediction. Figure 3 As shown in the figure, specifically including:

[0118] S4.1: The local temperature, humidity and solution concentration raw data obtained by real-time detection of the embedded environmental sensor are subjected to data synchronous acquisition and time sequence alignment processing to obtain the environmental parameter feature sequence which can be directly input into the sound field propagation parameter mapping model, ensuring that the data are matched one by one with the PCB three-dimensional structure number to be processed.

[0119] The local temperature, humidity and anti-oxidation solution concentration raw data obtained by the embedded environmental sensor in the printed circuit board anti-oxidation processing unit in real time are subjected to data synchronous acquisition and processing, and the data acquisition frequency is set according to the process rhythm and dynamic environmental change rate. The typical sampling period can be configured to 100ms to 500ms, ensuring that the dynamic fluctuations in the whole process are covered.

[0120] A multi-channel data synchronous acquisition method (parameters: multi-channel parallel acquisition interface, time resolution ≤1ms, synchronization error ≤10ms) is used to realize high-precision and low-delay data synchronization of the original signals generated by the spatially distributed sensor nodes.

[0121] Further, a unified time reference signal is used to mark the time stamp of the raw data stream of all acquisition channels, and the real-time clock system of the process control unit is used for global time sequence locking, ensuring that the data collected at different spatial positions have a strict consistent time sequence reference.

[0122] Further, a grouping buffer and batch data buffer mechanism (parameters: single batch data window capacity of 64-512 sampling points) is used to sort the data blocks collected in a short time in a first-in-first-out manner, eliminate asynchronous fragments caused by signal delay, frame loss or occasional acquisition pause, and realize stable data stream processing.

[0123] Further, for each frame of synchronous collected data, the PCB three-dimensional structure number and the spatial node mapping table are used for structure label alignment, and the board type identification code and physical space index are used to uniquely bind the collected environmental parameters and the PCB structure to be processed, ensuring that the environmental parameter sequence corresponds one by one with the normalized PCB three-dimensional structure, and no cross-board type misplacement occurs.

[0124] Further, the anomaly detection and missing point filling algorithm (parameters: sliding window constraint length of 3-5 sampling frames, extreme value elimination threshold of 2.5 times the standard deviation) is used to screen abnormal mutations and missing points in the data synchronization alignment process and perform linear interpolation or Lagrange interpolation filling, outputting complete, continuous and physically consistent environmental parameter feature time series.

[0125] Through the above multi-level synchronous acquisition, time series alignment and structure prompt label binding process, the original environmental monitoring data is converted into a standardized environmental parameter feature sequence that can be directly input into the sound field propagation parameter mapping model, realizing the high-reliability data input basis required for thickness uniformity adaptive prediction.

[0126] For example, in the printed circuit board oxidation resistance process unit, a 20-channel embedded high-precision thermistor, humidity capacitor and ion-selective solution concentration sensor array is deployed, and the collection period is set to 200 ms. The single-channel collection error is ≤0.1℃ or ≤1%RH, and the solution concentration accuracy is better than 0.05 mol / L. Every 20-channel sensor data is stamped with a unified time reference, grouped using a ring buffer and recorded in parallel to the process control host through a PCIe high-speed acquisition card. The PCB structure ID is automatically aligned with the sampling timestamp, and the output data frame structure is: board type ID, spatial node index, [temperature, humidity, concentration] three-tuple at t0…tn time. For occasional missing points, a 3-frame sliding window range statistics is enabled, and after detecting the overrun, it is filled back with linear interpolation to ensure uninterrupted data flow. The above processing outputs a 20×n time series feature matrix, which accurately reflects the multi-node parameter distribution of each PCB in the real-time process environment, provides high-quality input for the digital twin sound field parameter model, and realizes fine thickness uniformity prediction and subsequent adaptive control for multiple boards.

[0127] S4.2: Perform multi-modal feature fusion processing on the time series aligned environmental parameter feature sequence and the normalized PCB structure characteristic parameters, use feature splicing and normalization mapping algorithm to form a multi-dimensional joint input feature vector as the joint input condition of the sound field propagation parameter mapping model.

[0128] The input object is the environmental parameter feature sequence (including multi-dimensional data of temperature, humidity and solution concentration of each spatial node) processed by data synchronization and time series alignment, and the normalized PCB structure characteristic parameters (including normalized features of three-dimensional topography of each sub-region, local surface normal, board space partition index, etc.) corresponding to these environmental features.

[0129] The multi-modal feature splicing algorithm (parameters: feature dimension alignment strategy is spatial node linkage, splicing order is structural features first and environmental features second, and dimension consistency is forced to check) is adopted to splice the structural characteristic vector of each space region and the corresponding environmental parameter feature vector to form a composite feature unit. Further, the unified scale normalization processing is performed on each component of the preliminary multi-modal feature matrix formed after splicing by the normalization mapping method (parameters: minimum-maximum normalization lower bound is 0 and upper bound is 1, and distribution mean standardization reference historical batch feature statistics), so as to eliminate the statistical distribution difference between different dimensions and batches, and improve the comparability of feature distribution and the stability of input data.

[0130] Further, the correlation analysis is performed on the spliced multi-modal feature set by using the mutual information test algorithm (parameters: mutual information calculation window width is 5-9, and correlation screening threshold is 0.12-0.18), the low contribution or strong collinearity components are screened out by measuring the explanation ability of each component to the downstream sound field response parameters, and only the high correlation input features are reserved to optimize the model calculation performance. Further, for the high-dimensional multi-modal feature matrix, the principal component analysis (PCA) dimension reduction method (PCA parameters: cumulative contribution rate is greater than 95%, and the number of reserved principal components is generally not less than 10) is applied to compress and map the multi-dimensional structural and environmental joint features into a low-redundancy high-information-density feature vector, so as to effectively reduce the redundancy between features and the input dimension series.

[0131] Further, the feature label verification algorithm is used to map the low-dimensional high-expression joint feature vector to a plate type number, a space partition and a time stamp, so as to realize the correctness verification of one-to-one mapping relationship and ensure that each group of input features can be traced back to the physical space node and the environmental measurement instantaneous working condition.

[0132] Through the above multi-modal feature fusion and normalization mapping chain processing mode, the original time sequence aligned environmental parameters and normalized structural features are converted into a multi-dimensional joint input feature vector which can be directly input into the sound field propagation parameter mapping model, so as to realize the fusion of local area physical description under different environmental and structural conditions, and provide high confidence and high adaptability algorithm input for subsequent sound field response prediction and thickness uniformity regulation.

[0133] Exemplarily, in the anti-oxidation process example of the eight-layer complex-shaped printed circuit board labeled S[8L-240x210-C4], the input parameters include the time sequence alignment characteristic sequence of the temperature (acquisition range 22.1℃ to 26.5℃), humidity (38% RH to 51% RH) and solution concentration (0.74 mol / L to 0.98 mol / L) of 16 spatial nodes, and the normalized features of the three-dimensional topography in the same region (mean 0.59, standard deviation 0.14, maximum surface fluctuation 1.5 mm, spatial partitioning label 1-16). Using the multi-modal feature splicing algorithm, the 10-dimensional normalized vector of the three-dimensional topography of each spatial node and the corresponding 3-dimensional environmental feature vector are spliced in sequence to obtain a 16x13-dimensional preliminary composite feature matrix. Further, the normalized mapping method is used to perform minimum-maximum normalization and mean standardization on each column of features to eliminate batch differences. Then, mutual information screening is performed to remove 4 components with mutual information values below 0.15, leaving 9 highly expressed features. Next, the PCA method is applied to select the first 5 principal components, and the cumulative variance explanation rate reaches 95.7%. The output of 16 sets of 5-dimensional joint feature vectors is accurately mapped to each spatial node and board type number after label verification. Finally, the above 5-dimensional joint input feature vector set is input into the digital twin sound field propagation parameter mapping model to support subsequent sound field response prediction and provide data support for dynamic adaptive regulation of regional thickness uniformity. The measured process performance indicators show that the regional thickness uniformity is improved to ±0.18μm, and the corner is greatly optimized by 23%.

[0134] S4.3: Based on the obtained multi-dimensional joint input feature vector, the sound field propagation parameter mapping model trained and verified in the digital twin modeling stage is called to perform multi-region physical reasoning and obtain the sound field coupling response output for different PCB structure regions, i.e., the preliminary prediction results of frequency, phase and amplitude at the regional level.

[0135] The input object is a multi-dimensional joint input feature vector, including the normalized parameters of the PCB structure and the environmental parameters (temperature, humidity, solution concentration, etc.) of each spatial partition after multi-modal feature fusion and normalization, each spatial node corresponds to a specific board type number, spatial partition index and timestamp.

[0136] The sound field propagation parameter mapping model calling method (parameters: digital twin system standard model library, recall model type: region-level sound field parameter GBRT regression model trained and verified, input feature dimension: 5-20) is used to realize multi-region physical reasoning and large-scale vectorized batch operation functions.

[0137] Further, through the model input interface, the joint input feature vector is batched into the sound field propagation parameter mapping model, the regional physical inference is simultaneously performed on each spatial partition, the parallel calculation among multiple regions is realized, and the sound field coupling response prediction results of each local region are output.

[0138] The physical inference algorithm is internally set, the preliminary response of frequency, phase and amplitude of each region is calculated by analyzing the input spatial structure parameters (such as three-dimensional normalized features of the plate surface, variation of the surface normal and environmental covariants) and the model kernel function weight, and the multi-region sound field parameter output is obtained.

[0139] In the model, according to the regression node feature splitting and weight optimization processing, the quantitative mapping of the acoustic energy distribution, standing wave node and energy gradient under each type of structure-environment composite working condition is realized, so that the following prediction parameter set is formed for each spatial node:

[0140]

Block formula

[0141]

[0142] Where, f i * is the optimal excitation frequency of the i-th region sound field, is the optimal phase, is the optimal amplitude distribution, N region is the total number of partitions. The above parameters are obtained according to the multivariate regression and physical feature prediction framework.

[0143] Further, through the normalization constraint of the sound field propagation parameter model, it is ensured that the region excitation parameter output meets the physical boundary conditions and the equipment constraint range (such as frequency 20-120 kHz, amplitude 0.1-5.0, spatial distribution continuity constraint), and the unreasonable prediction is automatically optimized to the reasonable interval.

[0144] Through the parallel output operation, the full-region prediction parameter results are batched as the preliminary prediction matrix of the regional level frequency, phase and amplitude, which are used for subsequent multi-target parameter optimization and dynamic sound field modulation.

[0145] Through the above sound field propagation parameter mapping model inference chain processing mode, the multi-modal multi-dimensional joint input feature vector is converted into the preliminary prediction results of the full-region sound field excitation parameter group, the sound field response adaptive inference based on complex structure and real-time environment is realized, and the benchmark data support is provided for the thickness uniformity prediction and subsequent process dynamic regulation.

[0146] For example, for a multi-layer complex profile PCB board with a structure label of S[8L-240x210-C4], in an actual process running environment, the input parameters are 5-dimensional joint feature vectors of 16 groups of spatial nodes (features include area normalized three-dimensional relief, normal, local temperature, humidity, concentration), and the input vector of the group batch is sent to the standard sound field propagation parameter GBRT mapping model (model tree depth is set to 5, the number of sub-trees is 80, and input normalization processing is performed). The model outputs the sound field preliminary response parameters of each spatial node within 0.3 seconds. The output result is 16 groups of parameters Frequency distribution f i * Range 42.5kHz-74.2kHz, phase Distribution 0.31π-1.8π, amplitude Distribution 1.2-2.8 (normalized amplitude). For each group of parameters described above, automatic checking is performed by device constraints to remove physically over-limit parameters to form a final 16x3 area-level prediction matrix. After measuring the above preliminary response input, the local area (corner, narrow gap) thickness uniformity prediction deviation is better than ±0.21μm, and the batch-to-batch repeatability difference is less than 5%. The above results show that this step can efficiently complete the accurate reasoning and parameter preliminary distribution of the multi-area sound field response, and lay a solid data foundation for intelligent dynamic sound field formation and adaptive optimization of uniformity.

[0147] S4.4: Multi-objective parameter optimization processing is performed on the sound field coupling response output, an adaptive multivariate constraint optimization algorithm is used, the target thickness distribution requirement and historical similar cases are combined to generate ultrasonic excitation parameter groups for refined areas, and the feasibility thereof is confirmed to obtain key control parameters for thickness uniformity prediction and subsequent dynamic sound field modulation.

[0148] S4.5: The finally determined area-level ultrasonic excitation parameter groups (including frequency distribution, phase distribution and amplitude distribution) are output as input data for driving the programmable ultrasonic transducer array to form a dynamic sound field, and provide a parameter basis for subsequent area adaptive thickness adjustment and thickness uniformity prediction.

[0149] The step S5: according to the ultrasonic excitation parameter group, driving the programmable ultrasonic transducer array to form a multi-area dynamic sound field in the anti-oxidation coating process unit to guide the local distribution of the coating liquid on the PCB surface and control the fluid film formation characteristics. Specifically, it includes:

[0150] S5.1: The ultrasonic excitation parameter groups output by the sound field propagation parameter mapping model are subjected to multi-area distribution processing to generate ultrasonic transducer array driving parameters corresponding to each area, so as to ensure that the parameter groups accurately cover the local areas of the complex printed circuit board structure and realize professional and consistent mapping of the sound field excitation parameter groups to the ultrasonic transducer array driving parameters.

[0151] S5.2: Based on the ultrasonic transducer array driving parameters, using programmable waveform control technology, each ultrasonic transducer unit in the array is executed to generate independent frequency, phase and amplitude excitation signals to obtain spatially adjustable multi-channel excitation signals, generate driving waveform signals and ensure full expression of the excitation parameter group.

[0152] The input object is the multi-zone ultrasonic transducer array driving parameters output by the previous S5.1 step, and the parameter content includes the target set values of the frequency f i , phase and amplitude A i of each spatial partition and its corresponding physical space index information.

[0153] Using programmable waveform control method (parameters: multi-channel parallel driving, signal bandwidth 20-120kHz, phase resolution 0.01π, amplitude resolution 0.01), independent driving signal generation is performed on each unit in the ultrasonic transducer array, ensuring differential and high-precision sound field excitation for each spatial region in the array.

[0154] Further, through the waveform generation control algorithm (parameters: sine / square wave selection, waveform smoothness priority level 2, edge steepness suppression coefficient less than 0.05), the number of sampling points of the signal output DAC is automatically configured according to the input driving parameters f A i , and the output waveform of each channel is:

[0155]

Block formula

[0156]

[0157] Where v i (t) is the instantaneous output voltage of the i-th channel, A i is the amplitude setting, f i is the frequency parameter, is the initial phase.

[0158] Further, relying on multi-channel synchronous waveform buffer technology (parameters: ring buffer queue length is 1024 sampling points, refresh period is not greater than 1ms), each group of independent driving waveform signals generated is real-time buffered and synchronized distributed, ensuring that the output signals of all channels have strict time alignment and global synchronization capability, supporting precise empowerment of dynamic sound field.

[0159] Further, a dynamic calibration method based on parameter mapping is adopted (parameters: output self-check loop, feedback accuracy better than 0.5% FS), and periodic deviation detection is performed on the actual driving signal and the target set value of each channel to automatically correct the driving mismatch caused by DAC nonlinearity, channel crosstalk or temperature rise, and output the final compensated and optimized multi-channel high-precision excitation waveform signal data set.

[0160] Through the above chain processing mode, the regional level ultrasonic excitation parameter group of the previous step is efficiently and accurately converted into full-quantity independent driving waveform signals of each transducer unit in the array, realizing spatial distribution adjustable and time sequence strictly synchronous multi-channel excitation signal supply, and establishing a physical basis for the subsequent complex configuration of sound field distribution.

[0161] For example, on the 12x12 array of antioxidant treated ultrasonic transducer plates, the input parameters are output through the sound field parameter mapping model, respectively: frequency matrix f i，j distributed in 28kHz-84kHz, phase matrix distributed in 0.12pi-1.87pi, amplitude matrix A i，j distributed in 1.1-3.8V, a total of 144 groups of parameters. The programmable waveform control module takes 64MHz external clock as reference, configures each channel sine wave output, DAC single channel sampling rate is 512Ksps, and real-time generates each channel driving signal. The waveform self-check loop real-time calibrates the output amplitude and target deviation less than 0.3%, and the maximum synchronization error between signals is less than 9us. All 144 output signals are pushed to the transducer array through the array control interface, realizing spatial adjustable sound field energization. Through spatial acoustic field measurement, the correlation coefficient of sound field energy distribution and the set target is better than 0.98, supporting the uniformity improvement of the subsequent fluid film in the narrow gap and corner area, and finally the actual measured thickness uniformity is improved to ±0.21um, and the mechanical section partition requirement is effectively eliminated, verifying the high-precision and multi-region adaptive driving effect of the present step.

[0162] S5.3: Use the array control interface to synchronously issue excitation waveform signals to the ultrasonic transducer array, dynamically energize each region transducer, form a complex directional sound field distribution, realize real-time generation of multi-region dynamic sound field, and provide a physical field environment for subsequent fluid control.

[0163] The input is the multi-channel ultrasonic transducer array excitation waveform signal set output by the previous S5.2 step, and the signal has been independently generated and calibrated according to the target region frequency f i , phase amplitude A i respectively, with spatial distribution and time sequence alignment information.

[0164] The array control interface communication method (parameters: number of channels, interface type is SPI or PCIe composite link, communication bandwidth is not less than 10Gbps, transmission delay is less than or equal to 2ms) is adopted to realize the synchronous delivery of multi-channel excitation waveform signals. The array control interface maps and distributes each group of regional driving signals to the corresponding transducer units in the physical array layout through parallel channels, obtains a time and space synchronized signal grouping structure, and realizes one-to-one correspondence between a single channel and a spatial partition.

[0165] Further, by using the synchronous sending control algorithm (parameters: synchronous trigger flag, global clock signal source, maximum starting deviation is not more than 5μs), a global synchronous starting pulse is applied to the excitation waveform signals of the regional transducers, and the driving signals of all regional channels are activated at the same time, so that the clock consistency and phase accurate alignment of the excitation signals in the physical space are ensured.

[0166] Further, by using the redundancy check and real-time feedback mechanism (parameters: error code detection threshold 10 -8 , number of packet retransmission times 3), the excitation waveform data delivered is checked in real time, and for occasional communication errors or signal loss, the data retransmission and synchronization recovery process is automatically started to ensure that the enabling signals are not distorted and have no frame loss during delivery.

[0167] Further, by using the multi-region parameter dynamic enabling mapping algorithm (parameters: one-to-one mapping of spatial region label and channel index, variable spatial partition matrix dimension, typically 12x12 or 16x16), the signal distribution is automatically adjusted on the array control interface side, which flexibly adapts to different structure plates and regional allocation requirements, optimizes the directivity and complexity of the sound field spatial distribution, and meets the uniform coating requirements of complex PCB structure surfaces.

[0168] By using the dynamic enabling parameter recording and tracking method, the frequency, phase and amplitude of the excitation signal of each region at each time step are recorded in real time to form an enabling parameter stream, and abnormal waveforms (such as amplitude deviation exceeding threshold, phase mutation anomaly, etc.) are identified and automatically corrected online. The above mechanism ensures that all regions realize the full-quantity accurate expression of the sound field enabling.

[0169] Through the comprehensive chain technology processing of array control interface communication, synchronous sending control, real-time redundancy check and signal distribution optimization, the multi-channel excitation signal set generated by waveform control is accurately and synchronously delivered to the transducer array, realizing the real-time generation of a multi-region directivity complex sound field in the physical space, and providing an accurate and stable physical field basis for the orderly flow and thin film fine control of the liquid in the subsequent process.

[0170] For example, in an ultrasonic transducer-based coating unit applied to a certain type of 12x12 spatial array, the driving frequency f i，jdistributed in 30 kHz to 90 kHz, phase covering 0.1π to 1.9π, amplitude A i，j 1.2 to 4.0 V. The array control interface adopts a PCIe x8 bus, and the actual test communication bandwidth is 13.2 Gbps, and the global synchronization deviation is less than 3μs. The control interface is used to batch push the excitation signal queue, and 12x12 channels are respectively mapped to the corresponding transducer nodes in the physical space. After transmission verification, the detection error rate is better than 10 -9 , the mismatch waveform is automatically returned and reissued to ensure error-free energization. Multi-region sound field distribution measurement shows that the maximum relative error of spatial energy distribution is less than 1.6%, and the correlation coefficient of sound field configuration and digital twin target is 0.984. After the optimization of directivity, the uniform distribution of the coating fluid is realized at the edge of the region and the narrow gap. The actual measurement of the film thickness uniformity index of the PCB surface is better than ±0.18μm, which significantly improves the energization effect of the complex area that cannot be covered by the traditional mechanical section method, and supports the subsequent closed-loop control and global uniformity improvement of the fluid film.

[0171] S5.4: Drive the multi-region dynamic sound field to act on the coating liquid in the oxidation-resistant coating process unit, and realize the ordered flow, local accumulation and film extension of the coating liquid according to the spatial sound field distribution generated by the ultrasonic transducer array, form the local guiding distribution characteristics of the coating liquid, and effectively match the target area of the printed circuit board.

[0172] S5.5: On the basis of the local guiding distribution of the coating liquid, the spatial field type of the ultrasonic transducer array excitation is dynamically corrected through the sound field distribution and flow field response model, the liquid film thickness formation path is optimized in real time, and the adaptive adjustment of the liquid film formation characteristics is realized. Lay a data foundation for subsequent liquid film thickness feedback closed-loop control.

[0173] The step S6: using an embedded non-contact thickness sensor to monitor the liquid film thickness of each region in real time, obtaining fluid film distribution feedback data to reflect the actual film forming and thickness uniformity state. Specifically includes:

[0174] S6.1: The embedded non-contact thickness sensor array is arranged in the fluid film forming area to obtain the fluid film thickness original signal of each key position of the PCB board with high spatial resolution as the real-time monitoring object.

[0175] S6.2: Based on the original signal collected by the thickness sensor, the signal denoising and multi-point sampling interpolation algorithm are used to perform data preprocessing operation on the fluid film thickness original signal, so as to obtain the time and space consistent fluid film thickness preliminary data set.

[0176] The input is the multi-region fluid film thickness raw signal collected and output by the embedded non-contact thickness sensor array. The signal form is time sequence sampling data covering key positions of the printed circuit board, including sensor number, spatial position index, and corresponding measurement voltage or frequency shift value at the time.

[0177] A multi-stage signal denoising method is used to realize spectral cleaning and asynchronous noise suppression of the thickness raw signal. The parameters are: low-pass filter cutoff frequency of 5 kHz, wavelet threshold denoising wavelet basis of db4, and soft threshold coefficient λ = 3σ. This method removes high-frequency environmental interference, system noise, and transient pulse interference, ensuring the authenticity of subsequent data.

[0178] Further, a multi-point sampling interpolation algorithm is used to interpolate and complete the signal segments with spatial blind areas or missing time sampling. The parameters are: spatial minimum sampling interval of 0.8 mm, time minimum sampling step of 20 ms, and cubic spline interpolation method. This method fills in the missing measurement points caused by factors such as sensor miniaturization layout, process clamping shielding, and realizes the balanced distribution of thickness signals in space and time.

[0179] Further, an outlier detection and outlier correction algorithm is used to locate and correct occasional jump points in the signal. The parameters are: Z-score discrimination threshold of 2.5, and the mean values of the same spatial neighborhood and adjacent time points as references. This method adaptively replaces abnormal sampling points with multi-dimensional spatial and temporal neighborhood data, reducing the influence of occasional interference on the consistency of subsequent data sets.

[0180] Further, a space-time consistency calibration algorithm is applied (parameters: maximum resampling residual error less than 0.02 μm). The multi-channel thickness signal is mapped one-to-one in spatial index and time sequence label, and automatically aligned according to the physical size of the board and the process flow beat, ensuring that the sampling points of each region thickness data have strict position information and time label.

[0181] Through the above denoising, sampling completion, anomaly correction, and space-time calibration processing methods, the thickness sensor raw collection signal is converted into a space-time consistent fluid film thickness preliminary data set, realizing high-quality integration of each region thickness data, and laying a unified data foundation for subsequent thickness standardization, boundary correction, and spatial distribution analysis.

[0182] For example, in a multilayer printed circuit board (PCB) process with a production line width of 220mm, a board length of 350mm, and a minimum line width and spacing of 70μm, the thickness sensor array is configured with a 17×28 dot matrix layout, totaling 476 sampling points, with a spatial resolution of 1mm×1.2mm. The sampling rate per channel is 50Hz, and the signal acquisition time is 9s. After low-pass filtering (5kHz) and wavelet denoising, the average noise amplitude of the raw signal is reduced from 0.028μm to 0.008μm. The cubic spline method is used to automatically complete the 14 spatial data points missing due to fixture obstruction and the 3 occasional sampling omission cycles, achieving a 100% sampling integrity rate. Z-score discrimination and neighborhood mean correction correct 74 occasional outlier points, with a maximum correction amount of 0.07μm. After spatiotemporal consistency calibration, the spatial error of the entire board thickness dataset is less than 0.013μm, and the sampling points correspond one-to-one with the physical board surface. The final output preliminary thickness dataset covers the entire region, with complete spatial and temporal labels, significantly improving the reliability of subsequent thickness standardization and distribution analysis. This processing significantly optimizes the spatiotemporal consistency of data acquisition under complex structures of batch-produced panels, effectively ensuring the accuracy and stability of downstream links in the data chain, such as thickness uniformity correction and feedback loops.

[0183] S6.3: Call the boundary correction and non-uniformity discrimination algorithm on the initial thickness dataset to dynamically verify the measurement error caused by edge effects in the detection area, sensor drift or environmental interference, and output the calibrated fluid film thickness standard data.

[0184] S6.4: Input the standardized fluid film thickness data into the time series dynamic tracking module, and perform historical subdivision by combining the preceding process parameters (including ultrasonic excitation parameters, three-dimensional morphology of the plate surface, etc.) to generate fluid film thickness change trend data in each region.

[0185] S6.5: Based on the fluid thin film thickness variation trend data, the thickness distribution mapping algorithm is invoked to realize the spatial distribution visualization of regional thin film thickness and the calculation of uniformity index, and output fluid thin film distribution feedback data to provide high-dimensional dynamic input for the real-time process feedback link of the digital twin system.

[0186] Step S7: Input the fluid thin film distribution feedback data into the digital twin system, compare it with the target thickness distribution, and determine whether there are thickness anomalies in each region, including local thickening or excessive thinning. Specifically, this includes:

[0187] S7.1: Based on the fluid thin film distribution feedback data obtained by the embedded non-contact thickness sensor, the data is spatially aligned and mapped according to the three-dimensional topography parameters of the printed circuit board surface to obtain a spatially consistent set of three-dimensional distribution parameters of the fluid thin film.

[0188] S7.2: Using the fluid film three-dimensional distribution parameter set as input, multi-region thickness target distribution reasoning is performed using the thickness target model in the digital twin system to generate expected thickness distribution benchmark data for the current board type and process parameters.

[0189] S7.3: Comparing the obtained fluid film three-dimensional distribution parameter set with the expected thickness distribution benchmark data, a difference analysis algorithm based on region segmentation is used to calculate the thickness deviation vector of each region point by point to form the thickness abnormality preliminary discrimination parameter.

[0190] The input conditions include: the spatially aligned fluid film three-dimensional distribution parameter set, the expected thickness distribution benchmark data output by the thickness target model of the digital twin system for the current board type and process conditions.

[0191] The region segmentation algorithm (parameters: adaptive region division based on spatial block, minimum region edge length 3mm, maximum region boundary and three-dimensional topography change rate threshold 0.12) is used to divide the printed circuit board surface in space to realize accurate partitioning of multiple regions of complex or special-shaped boards, forming a spatial index matrix corresponding one-to-one to the fluid film three-dimensional data and the expected thickness distribution.

[0192] Further, by the regional difference analysis method (parameters: thickness deviation discrimination threshold Δt = ±0.15 μm), the real-time thickness mean value of the fluid film in each segmented region is calculated point by point with the expected thickness benchmark to obtain the regional thickness deviation array. The difference operation can be specifically expressed as:

[0193]

[0194] Where, ΔT i is the thickness deviation of the i-th region, is the average thickness of the region monitored in real time, is the target thickness of the region.

[0195] Further, by the thickness deviation vectorization algorithm (parameters: spatial distribution weight coefficient ω = 0.85, topography complexity adjustment coefficient κ = 1.18), for each regional thickness deviation, a full-board thickness abnormality vector is constructed to provide input for subsequent multi-dimensional abnormality feature discrimination and spatial distribution modeling.

[0196] Further, by the principal component residual analysis algorithm (parameters: the cumulative variance contribution rate of retained principal components is not less than 98%), principal component analysis is performed on the thickness abnormality vector data to eliminate periodic global bias and other overall trends, only the sensitive components of regional abnormalities are retained, and the spatial features of abnormal deviations are highlighted.

[0197] Through the chain processing of algorithms such as spatial region segmentation, region-by-region difference calculation, thickness anomaly vectorization and principal component residual analysis, the spatially aligned three-dimensional fluid film thickness distribution is compared with the expected value point by point, and finally the preliminary thickness anomaly discrimination parameters are output, so as to realize sensitive modeling and preliminary layering of abnormal behavior of local thickening or thinning in various regions of PCB.

[0198] For example, in a multi-layer irregular PCB mass production scenario, assuming a single board is divided into 36 irregular spatial partitions, with the smallest unit size of the region being 3mm × 4mm, a real-time thickness dataset is used. With digital twin models generated The average thickness of each region was calculated, and for each region, Δt = ±0.15 μm was used as the criterion to segment it into 7 regions with excessive thickness, with a maximum deviation of 0.36 μm and a minimum deviation of 0.17 μm. Thickness anomaly vectorization was applied to obtain a 36-dimensional thickness offset vector for the entire board. After principal component residual analysis, only 4 regions showed significant principal component shifts, and the locations of these regions overlapped with detected complex morphological structures such as holes and three-dimensional steps. After implementing this step, the accuracy of the preliminary thickness anomaly discrimination parameters was higher than 99.6%, laying a precise data foundation for subsequent statistical filtering and robust anomaly region extraction, effectively improving the response capability and data reliability of thickness uniformity anomaly detection under complex board shapes.

[0199] S7.4: For the initial thickness anomaly discrimination parameters, statistical filtering and robustness evaluation methods are used to eliminate occasional anomalies that may be caused by sensor errors or microscopic defects on the board surface, and output the thickness anomaly distribution parameters after robust processing.

[0200] S7.5: Input the robustly processed thickness anomaly distribution parameters into the digital twin system, and combine them with the sound field propagation parameter mapping model to identify and mark specific areas of local thickening or thinning, providing locational basis and quantitative indicators for subsequent local ultrasonic excitation parameter optimization and sound field correction control.

[0201] Step S8: For the detected abnormal area, optimize and adjust the local ultrasonic excitation parameters based on the sound field propagation parameter mapping model to achieve real-time sound field correction, thereby compensating for or dissipating the liquid film thickness in the abnormal area and achieving adaptive thickness control. Specifically, this includes:

[0202] S8.1: Locate local thickness anomalies in the fluid thin film distribution feedback data, and calculate the spatial distribution mask of the anomaly region based on the digital twin system to determine the local regions that need to be optimized and corrected.

[0203] S8.2: For the abnormal area spatial distribution mask, combined with the sound field propagation parameter mapping model, the current local ultrasonic excitation parameter group (including the existing frequency, phase, amplitude distribution) is obtained as the basic input of the subsequent adaptive optimization algorithm.

[0204] S8.3: Using the parameter adjustment module based on optimization algorithm (such as differential evolution or deep reinforcement learning), according to the local thickness abnormal feature, target thickness distribution and sound field propagation parameter mapping model, the existing local ultrasonic excitation parameter group is iteratively optimized in real time, and the optimized local ultrasonic excitation parameter is output.

[0205] The local thickness abnormal spatial distribution mask area obtained in the previous S8.2 step, and the current local ultrasonic excitation parameter group, thickness abnormal distribution parameter, target thickness distribution reference data and sound field propagation parameter mapping model are used as input conditions.

[0206] A multi-objective constrained optimization algorithm (parameter settings: fitness function is the square sum of regional liquid film thickness deviation, constraint condition is local excitation parameter change range ΔF / ΔA / ΔP, maximum iteration step is 150, convergence threshold ε = 0.01 μm) is used to minimize the difference between the local area real-time thickness and the target thickness, and the ultrasonic excitation parameter group of each local abnormal area is searched and optimized.

[0207] Further, through the differential evolution algorithm (parameters: population size is 30, mutation factor F = 0.7, crossover probability

[0208] CR = 0.9), the frequency, phase and amplitude distribution of the current local abnormal area are used as the initial population, and the optimal excitation parameter group that makes the liquid film thickness change closest to the target value under the action of the sound field propagation parameter mapping model is iteratively searched, and the parameter search step is dynamically adjusted.

[0209] Further, the deep reinforcement learning parameter adjustment module integrated by the digital twin system (network structure: 3-layer fully connected neural network, loss function is the mean square error of local thickness and target thickness, weighted reward factor λ = 0.75) is used to perform reinforcement learning training on the optimization process, dynamically collect reward feedback in the parameter adjustment process, and update the excitation strategy, so that the parameter adjustment has stronger dynamic response and generalization ability.

[0210] Further, the differential evolution algorithm optimization result and the deep reinforcement learning output result are multi-model fused, and the performance of each optimization result in the historical case library and the current real-time feedback is evaluated through weight fusion (fusion weight ratio η = 0.6:0.4) to output the local ultrasonic excitation parameter group with the best comprehensive performance.

[0211] By the above multi-objective constraint parameter optimization, differential evolution search, deep reinforcement learning strategy training and multi-model fusion processing mode, the local abnormal area thickness deviation, sound field response model and historical parameter obtained in the previous step are integrated and converted into the optimized local ultrasonic excitation parameter set, realizing the self-adaptive sound field driving correction capability for the local abnormal area.

[0212] For example, in the oxidation-resistant coating process of a six-layer PCB board containing a complex ring-shaped via structure and a multi-step edge, two local thin areas (area numbers A1 and B3, with areas of 12mm 2 and 9mm 2 respectively) are detected. The existing frequency distribution 470kHz / 445kHz, amplitude distribution 22Vpp / 19Vpp and phase difference π / 5 ultrasonic excitation parameter set is input. The differential evolution algorithm is used, the parameter configuration is F=0.7, CR=0.9, the search space boundary is set to ±15%, and the iteration is 120 steps. The deep reinforcement learning module loads the initial network weight from the historical 100 board similar case library, λ=0.75, and completes 300 rounds of training. The historical cases show that the use of differential evolution algorithm optimization alone can reduce the liquid film thickness mean square error of the target area to 0.038μm 2 , and the deep reinforcement learning optimization can further reduce it to 0.028μm 2 . The fusion output is weighted with η=0.6:0.4, and the preferred local parameter set is finally obtained, with the frequency adjusted to 477kHz (A1) and 452kHz (B3), the amplitude corrected to 24Vpp and 20.5Vpp respectively, and the phase difference adjusted to π / 4.2 and π / 5.7 respectively. The thickness closed-loop monitoring results show that the thickness of A1 and B3 areas is increased to 0.31μm and 0.28μm respectively, the thickness uniformity index is increased to more than 98.2%, and the process setting tolerance is met.

[0213] S8.4: Input the optimized local ultrasonic excitation parameters to the programmable ultrasonic transducer array, dynamically apply the adjusted sound field to the local abnormal area, realize the migration, accumulation or dissipation of the fluid film in the area, compensate or reduce the thickness abnormality, and complete the sound field correction behavior.

[0214] S8.5: Perform closed-loop real-time monitoring on the corrected fluid film thickness distribution, if the local area thickness abnormality does not meet the preset tolerance, automatically feed back the fluid film distribution feedback data to the digital twin system, re-enter the parameter optimization cycle until the thickness uniformity target is achieved.

[0215] The step S9: synchronously record the process parameters, thickness real-time monitoring data and sound field response parameters in each process to the process digital twin database as the basis for subsequent parameter tracking, model optimization and process continuous improvement. Specifically, it includes:

[0216] S9.1: Periodically collect and integrity check fluid film distribution feedback data, process parameters, and acoustic field response parameters to ensure real-time completeness and temporal coherence of mechanism attribute data required throughout the entire anti-oxidation coating process, including liquid film thickness in each region, coating unit temperature, solution concentration, ultrasonic frequency, phase, amplitude distribution, etc.

[0217] S9.2: Based on industrial data hierarchical standards, format standardization conversion is performed on multi-source heterogeneous process parameters, real-time thickness monitoring data, and acoustic field response parameters (such as normalization, structured nesting, and redundant field elimination) to obtain a unified structured format of industrial digital signal multi-dimensional data set.

[0218] S9.3: Use data meta-information tag generation algorithm to extend the structured process data set with meta-data tags, including plate type identification code, process batch number, collection timestamp, region identifier, etc. to facilitate subsequent typing, tracing, horizontal comparison, and vertical process evolution tracking.

[0219] S9.4: Perform data integrity verification and anomaly detection on the data after label extension, identify data collection omissions or sensor abnormal feedback in a timely manner, set a verification mark for abnormal records, and ensure the accuracy and traceability of data in the digital twin database.

[0220] S9.5: Upload and store the formatted, verified, and labeled data packets to the process digital twin database through the data synchronization interface, and maintain ordered association with historical data in real time in a closed-loop data link management manner to form a structured, high-confidence process parameter time series data warehouse.

[0221] S9.6: Perform preliminary statistics and feature aggregation on the newly stored multi-dimensional process data set, including thickness uniformity distribution statistics and acoustic field parameter section aggregation analysis, to provide structured raw data for parameter tracking, model training, and process continuous optimization in the next stage, and realize the mapping of data process value.

[0222] The step S10: If it is judged that the current plate type or environmental parameter appears an abnormal state beyond the specified range, an automatic alarm mechanism is triggered and the coating control process is suspended to ensure production safety and product consistency. Specifically, it includes:

[0223] S10.1: Perform dynamic judgment analysis on the thickness monitoring data, acoustic field response parameters, and environmental parameter data collected in real time by the embedded non-contact thickness sensor and the process digital twin database, use interval discrimination algorithm to judge whether the current plate type parameters and environmental parameters are within the set threshold range to obtain preliminary criteria for abnormal state.

[0224] S10.2: Based on the above abnormal state preliminary criterion, the specific attributes of the current abnormality are identified qualitatively and quantitatively by calling the abnormal type discrimination model, including environmental variable drift, thickness uniformity abnormality or sound field distortion, etc., to output the abnormal type label of the standardized abnormal attribute.

[0225] S10.3: According to the abnormal type label and the standardized abnormal attribute, the corresponding alarm priority and process suspension instruction are generated through the expert rule base or adaptive decision logic, to ensure immediate response to high-risk abnormalities, and output the automatic alarm trigger signal and suspension instruction set.

[0226] S10.4: Receive the automatic alarm trigger signal, drive the coating process PLC (Programmable Logic Controller) system and digital twin master control unit to execute process unit shutdown and safety locking operation, to ensure the safety of the equipment and product under the current plate type and environmental parameter state.

[0227] S10.5: The thickness monitoring data of this abnormal event, environmental parameter abnormality detailed information, alarm response process operation log, etc. are structured and stored to the process digital twin database, for subsequent cause tracing, abnormal mode learning and anti-oxidation process continuous optimization.

[0228] For those skilled in the art, other various corresponding changes and deformations can be made according to the above described technical solutions and concepts, and all of these changes and deformations should belong to the protection scope of the claims of the present application.

[0229] Unless otherwise defined, the technical terms or scientific terms used herein should be understood as the usual meaning understood by a person with ordinary skills in the field to which the present application belongs. The "first", "second", "third" and similar words used in the patent application specification and claims of the present application do not represent any order, quantity or importance, but are used to distinguish different components. Similarly, "one" or "a" and similar words do not represent a quantity limit, but represent the existence of at least one. "Including" or "containing" and similar words mean that the elements or objects appearing before "including" or "containing" cover the elements or objects listed after "including" or "containing" and their equivalents, and do not exclude other elements or objects. The plurality of the embodiments of the present application refers to two or more. A and / or B means that there are three cases: A; B; and A and B.

[0230] The above description is only exemplary embodiments of the present application, and the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present application, and these modifications or replacements should be encompassed in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A printed circuit board oxidation-resistant method, specifically comprising: S1: Collecting real-time board surface three-dimensional topography data and local environmental parameters of different types of printed circuit boards in the oxidation-resistant treatment process unit, including temperature, humidity and solution concentration; S2: Performing normalization preprocessing on the three-dimensional topography data and the local environmental parameters; S3: Based on the normalized three-dimensional topography data, using digital twin modeling method to generate sound field propagation parameter mapping model corresponding to each type of PCB structure; S4: Input the real-time collected environmental parameters into the sound field propagation parameter mapping model, combine the current PCB structure characteristics, and infer to obtain ultrasonic excitation parameter group for different regions; S5: According to the ultrasonic excitation parameter group, drive programmable ultrasonic transducer array to form a multi-region dynamic sound field in the oxidation-resistant coating process unit; S6: Use embedded non-contact thickness sensor to monitor the liquid film thickness of each region in real time, and obtain fluid film distribution feedback data; S7: Input the fluid film distribution feedback data into the digital twin system, compare the target thickness distribution, and judge whether there is thickness abnormality in each region; S8: For the detected abnormal area, optimize and adjust the local ultrasonic excitation parameters based on the sound field propagation parameter mapping model, and perform real-time sound field correction; S9: Record the process parameters, thickness real-time monitoring data and sound field response parameters in each process to the process digital twin database; S10: If it is judged that the current board type or environmental parameters appear abnormal state beyond the established range, trigger the automatic alarm mechanism and suspend the coating control process.

2. The printed circuit board oxidation-resistant method according to claim 1, wherein The step S1 specifically comprises: Numbering and type identification of the printed circuit board to be processed, positioning the PCB board type and size information entering the oxidation-resistant treatment process unit based on the MES system or artificial intelligence image recognition algorithm, obtaining the structure label and initializing the collection task; For the numbered printed circuit board, through the integrated three-dimensional topography sensor array, based on optical structured light scanning or laser confocal imaging method, perform three-dimensional topography data collection of the board surface, obtain the original three-dimensional point cloud data containing surface relief, step, pore, edge contour structure details; Based on the spatial positioning of three-dimensional topography data, the printed circuit board surface and its nearby area are laid out with high-precision environmental parameter sensor modules, and the temperature, humidity and oxidation-resistant liquid solution concentration environmental parameters are obtained in real time through digital bus, forming a synchronous environmental parameter data stream; Perform time stamp association on three-dimensional topography collection data and corresponding environmental parameter data, and map the data stream and the board surface physical position one by one through space-time synchronization algorithm, to obtain the multi-dimensional fusion representation of board surface structure details and process environment state; Perform integrity check and abnormality rejection on the synchronously obtained three-dimensional topography data and environmental parameter data, and automatically screen and correct missing fragments and abnormal noise using multi-dimensional fault tolerance correction algorithm.

3. The printed circuit board oxidation-resistant method of claim 1, wherein, The step S2 specifically comprises: Dimension calibration processing is performed on the three-dimensional topography data, and dimension-consistent three-dimensional topography standardized data is output; The distribution equalization algorithm is applied to the dimensionally calibrated three-dimensional topography standardized data, and the local feature distribution is normalized according to the batch statistical mean and variance parameters; The interval mapping and extreme value reduction method is used to map the collected environmental parameter data to the standard environmental parameter feature space, and the normalized environmental parameter features are output; The three-dimensional topography normalized features and the environmental parameter normalized features are fused, and the joint feature set is extracted through feature splicing and standardized mutual information screening algorithm; The robustness evaluation and normalization test of the joint feature set are performed to determine the batch-independent unified input features.

4. The printed circuit board oxidation-resistant method of claim 1, wherein, The step S3 specifically includes: Feature extraction is performed on the normalized three-dimensional topography data, and a PCB structure parameter set containing key geometric features is generated by using a curved surface grid reconstruction algorithm; Based on the PCB structure parameter set, the local sound field distribution under different ultrasonic excitation parameters is calculated, and the sound field distribution simulation data is obtained; The multi-physics field coupling modeling method is used to generate a fluid dynamics parameter variation mapping table by combining the sound field distribution simulation data with the physical property parameters of the coating fluid; A supervised learning algorithm is used to train the relationship between the normalized three-dimensional topography features and the fluid dynamics parameter variation mapping table, and a sound field propagation parameter mapping model for different PCB structure types is established, and the trained and verified sound field propagation parameter mapping model is archived to the digital twin system standard model library.

5. The printed circuit board oxidation-resistant method of claim 1, wherein, The step S4 specifically includes: The local temperature, humidity, and solution concentration raw data obtained by real-time detection of the embedded environmental sensor are synchronously collected and time-aligned; The time-aligned environmental parameter feature sequence and the normalized PCB structure characteristic parameters are fused to form a multi-dimensional joint input feature vector; Based on the obtained multi-dimensional joint input feature vector, the sound field propagation parameter mapping model trained and verified in the digital twin modeling stage is called to perform multi-region physical reasoning and obtain the sound field coupling response output for different PCB structure regions; The sound field coupling response output is subjected to multi-objective parameter optimization processing, and an adaptive multivariate constraint optimization algorithm is used to generate a refined regional ultrasonic excitation parameter group based on the target thickness distribution requirements and historical similarity cases, and output the final determined regional ultrasonic excitation parameter group.

6. The printed circuit board oxidation-resistant method of claim 2, wherein: The number and structure label of the printed circuit board are automatically recognized based on the MES system and deep learning algorithm, and the structure label includes the number of layers, size, irregular edge, and key structure features.

7. The printed circuit board oxidation-resistant method of claim 1, wherein: The three-dimensional topography data is collected by using structure light scanning and laser confocal imaging, supplemented by multi-angle sensor pose adaptation and multi-channel point cloud fusion algorithm, and the surface three-dimensional point cloud data is output.

8. The printed circuit board oxidation-resistant method of claim 1, wherein: The environmental parameters include temperature, humidity, and antioxidant liquid solution concentration, and are one-to-one mapped with the three-dimensional topography data through a space-time synchronization algorithm.

9. The printed circuit board oxidation-resistant method of claim 1, wherein: The digital twin modeling is performed by training the mapping model of regional ultrasonic driving and thickness response based on finite element simulation and multi-physics field coupling for different PCB structure types.

10. The printed circuit board oxidation-resistant method of claim 1, wherein: The real-time environmental parameters and the PCB structure parameters are processed by a multi-modal feature fusion algorithm into high-confidence joint features, which are input into a sound field propagation parameter mapping model after normalization, mutual information screening and PCA dimension reduction.

Images