CN-122028332-A - High multilayer board processing method oriented to inter-board glue filling uniformity

CN122028332ACN 122028332 ACN122028332 ACN 122028332ACN-122028332-A

Abstract

The invention provides a high multilayer board processing method for the uniformity of glue filling among boards, which comprises the steps of integrating a high-temperature stable annular piezoelectric transducer array to collect multi-region sound wave signals, improving the signal quality through wavelet denoising and normalization, extracting sound velocity, attenuation coefficient, nonlinear parameter and other time sequence characteristics, establishing a mapping relation between sound parameters and colloid modulus by means of a graph neural network to realize the real-time generation of a pressure compensation quantity matrix, and adopting a distributed robust predictive controller to optimally distribute driving signals of a flexible piezoelectric execution unit to realize dynamic redistribution and uniform glue filling of sub-millimeter pressure.

Inventors

LI XIN
CHEN YUYIN
Dai Jiluo
Huang Nengdiao
Liao su
PAN CHIQIN
Jia Bobo
LIU HUIFENG
LI YU

Assignees

龙宇电子(梅州)有限公司

Dates

Publication Date: 20260512
Application Date: 20260302

Claims (10)

1. The method for processing the high multilayer board facing to the uniformity of the filling glue among boards is characterized by comprising the following steps of: s1, acquiring transmission acoustic wave original data from a plurality of areas in the lamination process to acquire the spatial distribution differences of acoustic propagation parameters at different positions; S2, denoising and normalizing the transmission acoustic wave original data, and outputting the preprocessed transmission acoustic wave data; S3, extracting acoustic parameter time sequence characteristics based on the preprocessed transmission acoustic wave data; s4, constructing a graph neural network mapping model based on the corresponding relation between the acoustic parameter time sequence characteristics and the local colloid dynamic shear modulus and loss modulus in the off-line calibration data; S5, inputting the real-time extracted acoustic parameter time sequence characteristics into the graphic neural network mapping model to obtain a current pressure compensation quantity matrix; s6, inputting the current pressure compensation quantity matrix into a distributed robust prediction controller, and generating a driving signal of a flexible piezoelectric execution unit by the distributed robust prediction controller according to the thermodynamic state of the pressing process; S7, based on the driving signal, controlling the flexible piezoelectric execution unit to apply dynamic pressure distribution so as to realize the optimization of the uniformity of the inter-plate glue filling; and S8, monitoring the change of the acoustic propagation parameter in the pressing process, and triggering the self-adaptive adjustment of the pressure compensation parameter if the parameter fluctuation exceeds the acoustic propagation parameter fluctuation threshold value.
2. The method for processing a high multilayer board facing to inter-board paste uniformity according to claim 1, wherein the transmitted acoustic raw data comprises transmitted acoustic time domain waveform data, phase shift data, and envelope decay rate data.
3. The method for processing a high multilayer board facing to the uniformity of filling glue between boards according to claim 1, wherein the step S3 specifically comprises: Performing cross-correlation operation processing on the preprocessed transmitted sound wave data output in the step S2 to obtain sound wave propagation delay of each region, and further calculating sound velocity; Based on the sound velocity, carrying out path length compensation processing on the preprocessed envelope attenuation rate data, and calculating an attenuation coefficient; band-pass filtering processing and Hilbert transformation processing are applied to the preprocessed acoustic time domain waveform data, a fundamental frequency component and a second harmonic component are extracted, and nonlinear parameters are calculated based on component amplitude ratios; integrating the sound velocity, the attenuation coefficient, the nonlinear parameter pressing region coordinates and the time sequence to generate an acoustic parameter time sequence feature matrix; And carrying out moving average filtering processing on the acoustic parameter time sequence feature matrix, and outputting smooth and stable time sequence feature data.
4. The method for processing a high multilayer board facing to the uniformity of filling glue between boards according to claim 3, wherein the passband range in the bandpass filtering process is 0.9-1.5MHz, the filter type is an FIR linear phase structure, and the order is 128.
5. The method for processing the high multilayer board facing to the uniformity of the inter-board paste according to claim 1, wherein the step S4 specifically comprises: Based on the space topological relation of the acoustic parameter time sequence characteristics in the off-line calibration data, carrying out graph structure modeling processing on the time sequence characteristic data, and outputting graph neural network model structure parameters; based on the structure parameters of the graph neural network model, carrying out standardized pretreatment on the offline calibration data set, carrying out data enhancement operation to balance dynamic shear modulus and loss modulus distribution, and carrying out data set division on the enhanced standardized data set into a standardized training set and a verification set; Performing batch input processing on the standardized training set, performing forward propagation calculation to generate a dynamic shear modulus and a loss modulus predicted value, calculating the mean square error loss of the dynamic shear modulus and the loss modulus predicted value and a calibration actual value, performing reverse propagation to optimize network weight, and iteratively updating model parameters until the loss converges to obtain a trained graph neural network model; Based on the trained graph neural network model, carrying out prediction processing on the verification set data, calculating absolute errors of a predicted dynamic shear modulus, a predicted loss modulus and an actual value, evaluating model precision indexes, judging whether the precision meets a preset threshold value or not, and outputting a model performance evaluation report; and carrying out serialization processing on the verified graph neural network model to generate a mapping model file, and storing the mapping model file into a system memory.
6. The method for processing a high multilayer board facing to inter-board glue filling uniformity according to claim 5, wherein the graph structure modeling process specifically comprises: The method comprises the steps of defining press-fit region coordinates as nodes based on a space topological relation of acoustic parameter time sequence characteristics in offline calibration data, taking an acoustic propagation correlation matrix as an edge weight to construct a graph structure, extracting local space characteristics through a 3-layer graph rolling operation (64 convolution kernels in each layer), enhancing global representation capability by adopting a mean pooling and maximum pooling mixed strategy (weight coefficient 0.6:0.4), introducing a space-time attention mechanism (weight dynamic adjustment range 0.1-0.9) to optimize key region weight distribution, and finally outputting graph neural network model structural parameters.
7. The method for processing the high multilayer board facing to the inter-board glue filling uniformity according to claim 5, wherein the training stage input standardized training set of the graph neural network model is normalized by adopting Z-score, and the training set and the verification set are divided according to 70% to 30% of spatial partition, and the batch size is 64.
8. The method for processing a high multilayer board facing to uniformity of paste filling between boards according to claim 1, wherein the step S5 specifically comprises: Based on the acoustic parameter time sequence feature vector, acquiring normalized sound velocity, attenuation coefficient and nonlinear parameter standardized data, and executing feature dimension alignment operation on the normalized sound velocity, attenuation coefficient and nonlinear parameter standardized data to generate a feature tensor meeting the input dimension requirement of the graph neural network; inputting the feature tensor into a graph neural network mapping model constructed in the step S4, executing node feature aggregation operation based on the pre-training weight, and performing nonlinear transformation on the spatial topological relation between the areas by using a graph convolution layer to generate an implicit layer state matrix; Based on the hidden layer state matrix, performing region importance weighted calculation through a space-time attention mechanism, and applying a dynamic weight coefficient to a high-influence region to generate a preliminary pressure compensation vector; performing sub-pixel level spatial interpolation processing on the preliminary pressure compensation vector, and performing grid mapping operation based on a physical distribution topological structure of a piezoelectric execution unit to generate a pressure compensation matrix; And performing boundary check comparison on the pressure compensation quantity matrix and a process safety threshold, performing dynamic range clipping processing based on thermodynamic state feedback, and outputting a calibration pressure compensation quantity matrix.
9. The method for processing the high multilayer board facing the uniformity of the inter-board glue filling according to claim 1, wherein the distributed robust predictive controller is constructed based on parameters of a thermodynamic model of a pressing process and real-time state feedback data.
10. The method for processing the high multilayer board facing to the inter-board glue filling uniformity according to claim 1, wherein the acoustic propagation parameter fluctuation threshold is generated by performing multi-region statistical fluctuation range analysis based on acoustic parameter time sequence characteristic distribution in offline calibration data.

Description

High multilayer board processing method oriented to inter-board glue filling uniformity Technical Field The invention relates to the technical field of electronic manufacturing process control, in particular to a high multilayer board processing method facing to inter-board glue filling uniformity. Background At present, the high multilayer Printed Circuit Board (PCB) lamination process is widely applied to the field of high-end electronic manufacturing, and the core aim is to successfully press multilayer copper foil and prepreg resin material (PP) into an integral board with excellent electrical performance and structural consistency by means of the synergistic effect of pressure and temperature. In the prior art, the pressure distribution in the pressing process is mainly regulated by means of mechanical control, heating parameter optimization, die rigidity compensation, material thickness homogenization and the like. Common industry measures include local padding by top rigidizer/buffer, adjusting vacuum low pressure and rate of heating, selecting PP sheets of different brands to improve resin flow boundaries, optimizing vent groove structure, inhibiting copper foil wave shape deformation, and the like. The problems of poor resin filling caused by inconsistent overall pressure distribution are relieved to a certain extent by the means; Currently, traditional pressure uniformity improvement relies on static measures such as increasing the rigidity of a die, optimizing the copper foil/glass cloth laying design, adjusting the formula or thickness of PP prepreg cloth, and the like, and the methods essentially determine the subsequent pressure distribution characteristics before pressing, and have limited response capability to dynamic disturbance (material viscoelasticity change, local gassing, curing rate difference, and the like) in the pressing process. In addition, the high multilayer PCB lamination has extremely complex internal material distribution, the copper foil, the glass cloth, the core plate and the thickness differential result in obvious non-uniformity of a pressure field, the conventional constant pressure lamination process is difficult to adapt to timely compensation and fine regulation of local thin layer colloid, and a series of quality hidden troubles such as excessively low (adhesive shortage), excessively high (adhesive overflow) or pore formation of the partial area colloid layer thickness easily occur, so that the subsequent product has fluctuation of electrical performance, interlayer short circuit, structural failure and the like; The existing multi-region pressure feedback compensation technology is limited by sensor deployment, high temperature resistance of a signal cable, high temperature insulation reliability of a dielectric scheme, dynamic calibration complexity and the like. Dielectric type, thickness type and temperature response type sensing methods are generally faced with the problems of precision attenuation, system complexity improvement, poor long-term working stability and the like in a high-temperature environment, and are mostly dependent on the intrinsic characteristic change of materials, and are required to be calibrated independently for different stacking structures and material systems, so that the on-site operation and maintenance burden is heavy. In addition, in the traditional mode based on mechanical parameter adjustment (such as changing a pressing plate, adjusting exhaust or copper foil structure), in the process of lag of adjustment effect, the instant self-adaptive correction in the pressing process cannot be realized, the actual pressure distribution deviation of different areas of the circuit board is difficult to dynamically compensate, and the increasingly complicated high multilayer PCB quality management and control requirement is obviously insufficient. Disclosure of Invention The invention aims to solve the technical problems and provides a high multilayer board processing method facing to the uniformity of filling glue between boards. The technical scheme of the invention is realized in that the method for processing the high multilayer board facing to the uniformity of the filling glue among boards comprises the following steps: s1, acquiring transmission acoustic wave original data from a plurality of areas in the lamination process to acquire the spatial distribution differences of acoustic propagation parameters at different positions; s2, denoising and normalizing preprocessing are carried out on the transmission acoustic wave original data so as to eliminate background noise interference and improve signal-to-noise ratio; S3, extracting time sequence characteristics of acoustic parameters, including sound velocity, attenuation coefficient and nonlinear parameters, based on the preprocessed acoustic wave data; s4, constructing a graph neural network mapping model based on the corresponding relation between the acoustic parameter time s