Multi-point pressure compensation pressing method for large-size high multi-layer board thickness uniformity

Abstract

The present application relates to a multi-point pressure compensation pressing method for large-size high multi-layer plate thickness uniformity, and relates to thickness uniformity intelligent control technology in the pressing process of large-size high multi-layer plate. In view of the problem that the local pressure response inertia distribution of the material in the existing pressing process is difficult to identify and dynamically compensate in real time, a micro thermal-force coupling marker unit composed of low-melting point phase change material and directional glass fiber bundle is embedded in the key area. Combined with multi-modal synchronous observation and space-time data analysis, a closed-loop regulation system for precise analysis and compensation of pressure response time delay characteristics is constructed. The system can quantitatively and calibrate the transient rheological behavior of the micro interface of the material in real time, realize asynchronous scheduling and feedforward compensation of the pressure pulse sequence, dynamically reconstruct the pressure field topology, continuously optimize the thickness uniformity of the plate, and effectively improve the quality stability and intelligent level of the high-layer pressing plate manufacturing process.

Description

Technical Field

[0001] This invention relates to the field of dynamic control technology for lamination processes, and in particular to a multi-point pressure compensation lamination method for achieving uniform thickness of large-size, multilayer boards. Background Technology

[0002] Currently, ensuring uniform thickness of large-size multilayer PCBs is a core aspect of manufacturing high-performance electronic products in the lamination field. Mainstream technologies typically rely on multi-point distributed pressure loading systems, coupled with external sensor networks (such as pressure-sensing membranes and temperature sensors) and finite element-based numerical prediction models, to monitor and compensate for pressure throughout the lamination process. These technologies have achieved adjustable regional pressure distribution and dynamic setting of lamination temperature and pressure parameters, meeting to some extent the increasingly demanding requirements of large-size multilayer PCB manufacturing. To further overcome issues such as board thickness deviation and uneven interlayer air gaps, some companies and research institutions have proposed integrated control strategies including online resin rheology monitoring, lamination cavity temperature field homogenization, and multi-region feedback closed-loop pressure compensation, resulting in a phased improvement in the automation level of the lamination process and the consistency of finished products.

[0003] With the industrial development trend of increasing electronic complexity, multilayer stacking, and breakthroughs in large size, existing technologies in multi-point pressure control are still constrained by both system response speed and accuracy. Specifically, most multi-point pressure compensation systems are based on feedback-type closed-loop pressure control, which relies on real-time measurement of current thickness or pressure deviation information, followed by adjustment of closed-loop control parameters based on the error. This model generally has the following limitations: First, the rheological response of different regions of the material during the pressing process is not synchronous, and its pressure-thickness response relationship exhibits significant dynamic time lag characteristics related to the temperature-pressure path. Existing feedback systems often fail to capture this nonlinear hysteresis behavior in time, resulting in a fundamental time mismatch between the compensation command and the actual material response. Second, the stress relaxation spatial distribution caused by temperature changes and microflow at the material interface is extremely complex. Even with refined partitioned pressing pressure gauges, traditional models cannot fully reproduce the actual stress transfer state within large-size, multilayer boards. Third, to improve response speed, some solutions attempt to introduce high-sampling-rate sensors and large-scale data-driven model predictive control, but this significantly increases system complexity and computational load, increasing the risk of hardware failure and misadjustment.

[0004] For controlling the thickness uniformity of large-size, multilayer boards during lamination, current representative technologies are mainly applicable to scenarios such as static monitoring of pressure / thickness distribution, boundary area feedback compensation, and simplified thermal field consistency adjustment. Typical methods include: embedding a flexible pressure membrane array in the lamination equipment and performing threshold-type compensation control based on the measured spatial pressure distribution; using pre-placed heat flow diffusion blocks to reduce the impact of center-edge temperature and pressure unevenness; or improving cross-layer thickness consistency through batch optimization of process parameters. While these technologies are effective in synchronously capturing static errors or gradual responses, they struggle to achieve real-time, accurate, and asynchronous dynamic pressure compensation for the multi-scale stress-rheological dynamic processes of interface materials during lamination, which occur in conjunction with temperature and pressure. In particular, they are prone to defects such as local thickness heterogeneity and air gap concentration due to response lag during material flow, temperature abrupt changes, and the dynamic evolution of multilayer interfaces.

[0005] The prominent problems currently existing are reflected in the following aspects: First, the feedback compensation process is generally lagging, and system units cannot perform advanced pressure compensation based on the inherent micro-rheological hysteresis characteristics of the material. Second, the existing process parameter scheduling mechanism relies heavily on preset models or experience bases, and its adaptability to actual product differences and real-time operating condition changes is limited, making it impossible to meticulously match the dynamic pressure response window on a piece-by-piece basis. Third, compensation actions are issued in a globally unified timing sequence, ignoring the differences in material stress-temperature memory in different regions, resulting in an inherent time delay between the actual control signal and local physical feedback, affecting the thickness uniformity effect and production consistency. Fourth, the complex algorithms and multimodal signal processing system introduce a large number of modeling, calibration, and data iteration steps, which not only increases the difficulty of process operation but also poses the risk of overall loss of control in the event of system drift or sudden anomalies.

[0006] From an industry perspective, addressing the aforementioned issues necessitates an urgent need for a dynamic pressure field compensation mechanism with high spatiotemporal resolution. This mechanism should leverage the material's inherent dynamic stress-deformation memory characteristics without relying on redundant external feedback, excessive modeling, or high-complexity hardware and software investment. This mechanism would truly solve the industry-wide challenges of slow response speed in pressing systems, mismatch between pressure commands and actual rheological responses, and difficulty in ensuring local thickness uniformity. By developing a dynamic pressure compensation method coupled with the material's intrinsic multi-scale dynamic behavior, it is expected to significantly improve the timing control accuracy and product consistency during the pressing process, breaking through the fundamental response bottlenecks and the technological limits of precise zonal compensation in existing technologies. Summary of the Invention

[0007] This application provides a multi-point pressure compensation lamination method for large-size multilayer boards to improve thickness uniformity, aiming to solve one of the problems or issues of the prior art mentioned in the background section.

[0008] The multi-point pressure compensation lamination method for improving the thickness uniformity of large-size multilayer boards provided in this application specifically includes: S1: Embed a micro thermo-mechanical coupling marker unit composed of a low-melting-point phase change material and a directionally arranged micron-sized glass fiber bundle in the key area of ​​the large-size high-multilayer board lamination structure, so as to construct a physical information carrier that can generate controllable micro-deformation and record the local flow state during the heating and pressurization process.

[0009] S2: Based on the thermal and pressure behavior of the micro thermo-mechanical coupling marker unit in the initial stage of heating and pressurization, the infrared thermal radiation offset and surface micro-deformation optical interferogram of each marker unit are collected simultaneously to obtain the original spatiotemporal dataset characterizing the micro-response trajectory of the material.

[0010] S3: Compare and analyze the deformation-thermal response trajectory clusters in the original spatiotemporal dataset to identify the characteristic relaxation time windows that characterize the material pressure response inertia at each location, so as to generate time-series characteristic parameters that reflect the coupling relationship between local stress relaxation rate and thermal history.

[0011] S4: Map the feature relaxation time windows corresponding to all marked units on the entire plate to a dynamic weight matrix of the pressure field to construct a spatiotemporal distribution model that does not represent the current pressure value but represents the actual response delay time after pressure disturbance.

[0012] S5: Based on the dynamic weight matrix of the pressure field, the preset multi-point pressure compensation command sequence is subjected to spatiotemporal decoupling processing, and a pressure pulse sequence with asynchronous start-up and gradient propagation characteristics is generated by combining the spatial neighborhood pressure diffusion constraint condition, so as to eliminate the response lag caused by the unified timing command.

[0013] The multi-point pressure compensation lamination method for improving the thickness uniformity of large-size multilayer boards provided in this application has the following beneficial effects: (1) The method proposed in this application effectively overcomes the problems of response lag, model mismatch and system redundancy caused by relying on external sensor feedback or complex constituent modeling in traditional thickness uniformity control technology by using the pressing material system itself as a physical information carrier with "pressure memory" capability. Existing technologies usually perform closed-loop adjustment based on preset rheological parameters or real-time monitoring of deviations, which is difficult to adapt to the dynamic evolution process of large-size multilayer boards with uneven local flow and significant differences in curing kinetics under thermo-mechanical coupling environment. This can easily cause time mismatch between compensation action and actual material response, leading to over-pressing or under-compensation. The present invention utilizes thermo-mechanical coupling marker units embedded in the core board to actively excite and capture the stress relaxation behavior of each region in the early stage of heating, identify the relaxation time window characterizing local "pressure response inertia" and convert it into a dynamic weight matrix of pressure field, realizing the pre-sensing and quantitative ranking of material response time sequence characteristics. This mechanism eliminates the need to introduce resin constitutive equations or fractional-order control algorithms, avoiding modeling errors caused by material batch differences and temperature control fluctuations. It significantly improves the timing matching accuracy between multi-point pressure regulation commands and real material behavior, making the compensation effect closer to the actual physical evolution path.

[0014] (2) The pressure pulse sequence generation strategy based on the relaxation time window realizes the spatiotemporal decoupling and gradient propagation control of the pressure compensation process, which greatly improves the spatial coordination and dynamic adaptability of thickness control. Unlike the traditional method of applying global pressure increments in a unified time sequence, this scheme allocates nonlinear time delays to the preset compensation commands based on the response inertia differences of each marker unit, and constructs a pressure pulse sequence with asynchronous start-up and step-by-step propagation characteristics by combining the neighborhood pressure diffusion constraint. The entire process does not rely on artificial intelligence model training or iterative learning mechanisms, has low computational burden and flexible deployment, and is particularly suitable for rapid switching and stable operation in multi-variety, small-batch production scenarios.

[0015] The combined effect of the aforementioned techniques enables this method to achieve highly precise intervention in the thickness evolution of key interfaces during the lamination process without increasing system hardware complexity or control algorithm hierarchy. Its core advantage lies in transforming the inherent microstructural evolution behavior of the material itself under specific process paths into measurable, sortable, and schedulable physical control parameters. This overcomes the technical bottleneck of the disconnect between "instruction-driven" and "material response" in traditional control logic, significantly improving the reliability and robustness of full-area thickness uniformity control in large-size, multilayer boards. It is particularly suitable for applications with stringent requirements for interlayer consistency, such as high-frequency, high-speed PCBs and packaging substrates. Attached Figure Description

[0016] Figure 1 This is a flowchart of the multi-point pressure compensation lamination method for large-size high-multilayer boards to ensure thickness uniformity according to the present invention; Figure 2 This is a sub-flowchart of the multi-point pressure compensation lamination method for achieving thickness uniformity of large-size multilayer boards according to the present invention. Figure 3 This is another sub-flowchart of the multi-point pressure compensation lamination method for achieving thickness uniformity of large-size multilayer boards according to the present invention. Detailed Implementation

[0017] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0018] The following disclosure provides many different embodiments or examples for implementing different structures of the invention. To simplify the disclosure, specific examples of components and arrangements are described below. Of course, these are merely examples and are not intended to limit the invention. Furthermore, reference numerals and / or letters may be repeated in different examples; such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.

[0019] like Figure 1 As shown, this application provides a multi-point pressure compensation lamination method for large-size, multilayer boards to ensure thickness uniformity, specifically including: S1: Embed micro thermo-mechanical coupling marker units composed of low-melting-point phase change materials and oriented micron-sized glass fiber bundles in key areas of large-size high-multilayer board lamination to construct a physical information carrier that can generate controllable micro-deformation and record local flow state during the heating and pressurization process. S2: Based on the thermal and pressure behavior of the micro thermo-mechanical coupling marker unit in the initial stage of heating and pressurization, the infrared thermal radiation offset and surface micro-deformation optical interferogram of each marker unit are collected simultaneously to obtain the original spatiotemporal dataset characterizing the micro-response trajectory of the material. S3: Analyze the original spatiotemporal dataset to identify characteristic relaxation time windows that characterize the inertia of material pressure response at each location, so as to generate time-series characteristic parameters that reflect the coupling relationship between local stress relaxation rate and thermal history. S4: Map the feature relaxation time window corresponding to all marked units of the entire plate into a dynamic weight matrix of the pressure field; S5: Based on the dynamic weight matrix of the pressure field, the preset multi-point pressure compensation command sequence is subjected to spatiotemporal decoupling processing, and a pressure pulse sequence with asynchronous start-up and gradient propagation characteristics is generated by combining the spatial neighborhood pressure diffusion constraint condition, so as to eliminate the response lag caused by the unified timing command.

[0020] Step S1: Embedding micro-thermal-mechanical coupling marker units, composed of low-melting-point phase change materials and directionally arranged micron-sized glass fiber bundles, in key areas of large-size high-multilayer laminated structures, to construct a physical information carrier capable of generating controllable micro-deformation and recording local flow states during heating and pressurization. Specifically, this includes: S1.1: Perform interfacial compatibility pretreatment on low-melting-point phase change material raw materials and oriented micron-sized glass fiber bundle raw materials to obtain a composite matrix material with stable interfacial bonding force, ensuring the reliability of subsequent deformation transfer.

[0021] When performing interfacial compatibility pretreatment on low-melting-point phase change material raw materials and oriented micron-sized glass fiber bundles, both types of raw materials are placed in a dry, constant-temperature environment to stabilize their initial moisture content and surface chemical state. The interfacial tension difference between the two types of raw materials is determined using a surface energy matching evaluation algorithm, and the required concentration of the interfacial modifier is determined through numerical calculation. The type of interfacial modifier selected is a bifunctional molecule compatible with the silane groups on the glass fiber surface and the molecular chains of the phase change material. In the surface treatment stage, the glass fiber bundles are first subjected to plasma etching to remove the hydrophobic contamination layer and generate reactive active sites. The etching parameters are set based on the interfacial tension difference and the target bonding force, with the product of etching time and power satisfying the required interfacial activation depth. Subsequently, a nano-coated interfacial modifier film is applied to the surface of the phase change material raw materials. The thickness of this film is precisely controlled by optical interferometry to ensure that the coating does not block the microscopic deformation transmission of the phase change material under heated and pressurized conditions. The shear force and peel force of the composite matrix under different temperature and pressure combinations were measured using an interfacial bonding force testing device. The bonding force value output by the interfacial bonding force calculation formula was compared with the target threshold, and the ratio of the interfacial modifier was adjusted until the bonding force stabilized within the design range. Through this process, the original material state of the previous step was transformed into a composite matrix material with stable interfacial bonding force, ensuring the reliability of deformation transfer of the subsequent micro-thermo-mechanical coupling marking unit during heating and pressurization.

[0022] For example, in the manufacturing process of a batch of large-size high-multilayer core boards, the moisture content of the low-melting-point phase change material raw material was controlled at 0.2% using a constant-temperature drying oven. The surface energy of the oriented micron-sized glass fiber bundle raw material was measured to be 38 mN / m, and the surface energy of the phase change material raw material was measured to be 42 mN / m. The calculated interfacial tension difference was 4 mN / m, and the concentration of the interface modifier was determined to be 1.5%. The glass fiber bundles were subjected to 100 W plasma etching for 8 minutes, resulting in a reactive site density of 3 × 10^6 sites / mm². The thickness of the interfacial modifier film coating on the phase change material surface was controlled at 150 nm. In the interfacial bonding strength test, the measured interfacial stress was 58 MPa under conditions of 120°C and 1.5 MPa pressure, while the reference sample interfacial stress was 55 MPa. The bonding strength calculated using the above formula was 9, exceeding the set minimum bonding strength threshold of 6, indicating that the interfacial bonding strength meets the design requirements. After this step, the composite matrix material exhibits significantly improved deformation synchronization in stress transfer tests during the actual pressing process, providing a reliable physical basis for subsequent marking unit encapsulation.

[0023] S1.2: Perform micro-nano scale encapsulation process based on the composite matrix material to generate a micro thermo-mechanical coupling marker unit with a specific phase transition temperature threshold and optical reflection characteristics, making it a physical information carrier that can be identified by infrared and optical systems.

[0024] For the composite matrix material obtained after interface compatibility pretreatment, a micro-nano scale encapsulation process is used to structurally process it to form a composite marking unit with a preset phase transition temperature threshold and recognizable by optical and infrared systems. First, the composite matrix material is slicing into micro-flakes with geometric tolerances within ±0.5 μm, and surface planarization and polishing are performed in a low-temperature inert atmosphere to reduce the interference of interface roughness on optical reflectivity. Second, a transparent protective layer is deposited in situ at the outer edge of the micro-flakes to achieve mechanical and chemical isolation. The thickness of this protective layer is controlled at 500 nm to ensure that the thermal conduction delay after encapsulation does not exceed a set value. Third, a holographic microstructure array is etched on the surface of the protective layer using micro-nano photolithography. Specific optical reflectivity characteristics are set by the diffraction peak positions; the reflectivity spectrum must show a detectable peak change near the phase transition temperature of the marking unit. Furthermore, a phase transition response enhancement layer is implanted in the core region of the microchip. This layer is composed of a mixture of uniformly distributed nano-metal particles and low-melting-point phase transition materials. The size distribution of the nanoparticles is controlled within 20–50 nm to improve thermal conductivity and ensure that the physical deformation during the phase transition process can be resolved by the optical interference system. Finally, the characteristics of the encapsulated marking unit are calibrated. Differential scanning calorimetry is used to measure its phase transition temperature range, and infrared spectroscopy is used to verify its absorption peak transfer characteristics at a preset temperature threshold. Multi-point optical reflectance measurements confirm that its reflectance coefficient in the target band is stable within ±2% error. Through the above process chain, the interface preprocessing results are transformed into a miniature thermo-mechanical coupling marking unit with a clear phase transition response and optical recognition capability, realizing the construction of a high-fidelity physical information carrier in an embedded sensor network.

[0025] For example, in the critical area of ​​the core board of a large-size multilayer board with dimensions of 1200mm × 600mm, a phase change temperature threshold of 95℃ and a thermal conductivity of 1.8W·m are selected. ·K A low-melting-point alloy was used as the phase change material, and oriented glass fiber bundles with a diameter of 5 μm were arranged along the length of the plate. The composite matrix material, after interface compatibility pretreatment, was cut into micro-sheets of 0.5 mm × 0.5 mm × 0.2 mm and precision polished at 300 rpm in a nitrogen atmosphere to reduce the surface roughness Ra to 10 nm. A 500 nm thick transparent silica protective layer was deposited on the outer surface of the micro-sheets. A holographic microstructure array with a period of 1.2 μm was generated by electron beam lithography, resulting in an intensity change in the reflection peak at a wavelength of 650 nm, with the reflection coefficient maintained within ±2%. Silver nanoparticles, accounting for 5% of the total volume, were uniformly dispersed in the phase change material in the central region of the micro-sheets. Differential scanning calorimetry verified that the phase change peak temperature was 95 °C ± 0.2 °C, and infrared spectroscopy confirmed a shift in the absorption peak at 95 °C, with a shift amount of 15 nm. The final output of the optical recognition parameters of the marking unit is used for the visual positioning system recognition in S1.3. The packaging unit can stably provide signal triggering at the target temperature in the pressing environment to realize the high-precision spatiotemporal reference required for dynamic pressure compensation during the pressing process.

[0026] S1.3: Use a high-precision visual positioning system to scan the key stress concentration areas on the surface of large-size multilayer laminated boards to obtain a precise spatial coordinate distribution map of the location to be embedded, providing a positional basis for the layout of the marking unit.

[0027] The input conditions are a large-size, multilayer board laminated prototype surface that has been encapsulated with micro thermo-coupling marker units and has specific phase transition temperature thresholds and optical reflection characteristics, as well as a high-precision visual positioning system and a preset key stress concentration area recognition algorithm.

[0028] A high-precision visual positioning system is invoked to establish a multi-scale resolution imaging task on the entire board surface to capture the micro-texture patterns and macro-layout structures of different regions, generating a basic image matrix containing the reflectivity, brightness gradient, and geometric feature vectors of each pixel on the board surface.

[0029] Based on the aforementioned basic image matrix, a stress concentration prediction model is executed to identify areas prone to thickness deviation during the pressing process through the correlation mapping between optical reflection patterns and plate structure layers, thereby forming a candidate set of regions with preliminary location labels.

[0030] The local magnification scanning module is invoked on the candidate region set to acquire the boundary topography data of the candidate region at nanometer-level pixel resolution, extract the contour coordinates and calculate the precise spatial coordinate value of its center point, forming a local feature dataset containing spatial index and coordinate accuracy index.

[0031] The local feature dataset is uniformly projected and transformed to the global coordinate system of the entire board to eliminate coordinate system drift caused by magnified scanning and generate a normalized and accurate spatial coordinate distribution map as the input basis for the layout of the marked units.

[0032] Through the above-mentioned chain scanning and coordinate transformation processing method, the embedded target area data of the encapsulated marker unit is transformed into a precise spatial coordinate distribution map with spatial index consistency and coordinate accuracy of ±5μm, thus achieving the expected technical effect of providing a reliable positional basis for subsequent embedding actions.

[0033] For example, when performing a visual positioning task on a large-size, high-multilayer board laminated and stacked structure with dimensions of 1800mm × 1200mm and 24 layers, this step first sets the full-board scanning resolution to 25μm / pixel to generate a global base image matrix. This matrix contains optical feature vector data of approximately 3.45 × 10^9 pixels. Based on a stress concentration prediction model, approximately 156 candidate regions are selected, whose center point optical reflection gradient values ​​are more than 2.5 times the average value of the entire board. When the local magnification scanning module is invoked, the resolution is increased to 2μm / pixel, and the boundary coordinates of each candidate region are extracted. The calculated spatial coordinates of the center point have an accuracy of ±3μm in the local coordinate system. During the projection transformation process, a matrix calculation formula for the spatial coordinate system is used. This formula is used to uniformly project the 156 center points. The final output accurate spatial coordinate distribution map has a coordinate deviation of no more than ±5μm in the full-board coordinate system, ensuring a significant improvement in the positional accuracy when embedding micro-thermal-mechanical coupling marker units and a substantial improvement in thickness uniformity control performance in subsequent embedding steps.

[0034] S1.4: Based on the precise spatial coordinate distribution map, the micro thermo-coupling marker unit is embedded into the resin semi-cured layer of the key area of ​​the core board to construct an embedded sensing network structure that generates controllable micro-deformation during the heating and pressurization process.

[0035] The input conditions are the precise spatial coordinate distribution map data generated in step S1.3 and the micro-thermal-coupling marker unit entity encapsulated in step S1.2. Based on these input conditions, a high-precision mechanical positioning platform is used to load the positioning data and generate the embedding path planning matrix corresponding to the resin semi-cured layer. This embedding path planning matrix drives a three-dimensional automatic point insertion system to position the marker unit to the predetermined embedding point of the resin semi-cured layer in the key area of ​​the core board, ensuring that the spatial coordinate error does not exceed 5 micrometers. Local resin softening treatment is performed at the embedding point, using a temperature-controlled heating unit to control the temperature of the embedding point within the range of 85-95 degrees Celsius, reducing the resin viscosity to a controllable range. A precision pressure loading mechanism is used to press the marker unit into the resin semi-cured layer with a transient pressure of 0.2-0.3 MPa, forming a bonding contact surface consistent with the main material interface. After embedding, three-dimensional optical contour scanning is used to obtain the surface contour deviation data of the embedding position. Spatial difference analysis is performed to determine the impact of the embedding process on the flatness of the stack, ensuring that the flatness deviation is controlled within a certain range. Within micrometers. Through the aforementioned positioning planning, local resin softening, pressure loading, and three-dimensional optical verification processing, the results of the previous step are transformed into an embedded sensor network structure with high-precision spatial distribution and a stable interface, achieving high-fidelity acquisition capabilities for controllable micro-deformation response and local resin flow state during heating and pressurization.

[0036] For example, on a large-size multilayer board core with dimensions of 1200×600 mm and a thickness of 2.8 mm, the precise spatial coordinate distribution map contains 256 key embedding points with a point coordinate error of 3 micrometers. After the embedding path planning matrix is ​​analyzed by a 3D automatic insertion system, the cumulative embedding path length is 15.6 meters. The local resin softening treatment temperature is set to 90℃, corresponding to a decrease in resin viscosity from 1200 Pa·s to 450 Pa·s, enabling smooth pressing of the marking units. During the pressure loading stage, the transient pressure is controlled at 0.25 MPa, and the indentation depth is 0.45 mm. After embedding, 3D optical contour scanning shows that the average surface deviation of the embedding position is 0.8 micrometers, and the maximum value is 0.95 micrometers, which does not exceed the preset value. The micrometer limit verifies that the embedded sensor network structure has significantly improved its ability to adapt to vacancies and maintain flatness, providing a stable physical basis for the acquisition of micro-deformation response during subsequent heating and pressurization processes.

[0037] S1.5: Perform a pre-compression stability check on the embedded sensor network structure after embedding to confirm that the micro thermo-coupling marker unit can accurately record the local flow state in the initial state without affecting the overall stack flatness.

[0038] Perform static appearance and geometric dimension scanning on the embedded embedded sensing network structure in the initial environment of constant temperature and humidity, obtain the baseline matrix of the flatness of the board surface in the area where the micro thermal-mechanical coupling marking unit is located, and use it as the reference value for subsequent stability indicators.

[0039] Use a low-load preloading device to uniformly load the entire board surface under a pressure condition not exceeding 0.1 MPa, and use a high-speed white light interferometer to collect three-dimensional surface topography data of the embedded area in real time to detect whether there are initial deformation sensitive points under low-load conditions.

[0040] Perform multi-scale surface fitting and differential calculation processing on the collected topography data, generate a differential topography matrix reflecting the topography change amount under low-load conditions, and perform element-level comparison and determination with the baseline matrix to screen out local areas exceeding the preset flatness deviation threshold.

[0041] Under the heating preloading condition, slowly raise the board surface temperature to the phase change sensitive interval of 80 - 120 °C, and synchronously record the changes in the infrared thermal radiation intensity and optical reflection characteristics of the marking unit to evaluate the response accuracy of recording the local glue flow state in the initial state.

[0042] Perform time series correlation calculation on the infrared and optical signals, and confirm the stability of the recording performance by calculating and determining whether the correlation coefficient reaches the preset high correlation threshold.

[0043] Through the above static flatness comparison, low-load topography detection, phase change sensitive area response verification and signal correlation calculation processing, convert the state of the embedded sensing network structure in the previous step into a quantifiable flatness pass judgment matrix and recording performance stability index, and achieve the goal of pre-compression stability verification that the micro thermal-mechanical coupling marking unit can accurately record the local glue flow state in the initial state without affecting the overall lamination flatness.

[0044] Exemplarily, on a large-size high-multi-layer board sample with a size of 1200×800 mm, the embedded sensing network includes 48 micro thermal-mechanical coupling marking units, and the maximum deviation value of the baseline flatness matrix is 0.015 mm. Under the action of a low load of 0.08 MPa, after rapid scanning detection by a white light interferometer, the absolute peak value of the differential topography matrix is 0.018 mm, and no deformation points exceeding the preset threshold of 0.02 mm appear. In the phase change sensitive interval when the temperature is raised to 100 °C, the correlation coefficient between the thermal radiation intensity sequence and the optical reflection change sequence of the marking unit is 0.92, which is higher than the preset high correlation threshold of 0.9, indicating stable recording performance. In this scenario, both the flatness pass judgment matrix and the recording performance stability index output are in a qualified state, ensuring high fidelity and repeatability of data collection in the subsequent initial stage of temperature and pressure increase.

[0045] Step S2: Based on the thermal and pressure behavior of the micro thermo-mechanical coupling marker units in the initial stage of heating and pressurization, the infrared thermal radiation shift and surface micro-deformation optical interferograms of each marker unit are simultaneously collected to obtain the original spatiotemporal dataset characterizing the microscopic response trajectory of the material. Specifically, this includes: S2.1: Perform multi-band infrared thermal imaging scanning on the surface of the large-size multilayer board lamination during the initial stage of heating and pressurization to obtain a real-time infrared thermal radiation intensity distribution map containing the location information of each micro thermo-mechanical coupling marker unit, thereby extracting the basic thermal data source reflecting the non-uniformity of the local temperature field.

[0046] S2.2: Based on the coordinate information of the marker unit in the real-time infrared thermal radiation intensity distribution map, drive the high-speed white light interferometer to perform nanoscale optical phase demodulation processing on the surface of each micro thermo-mechanical coupling marker unit to generate a high-resolution surface micro-deformation optical interference original spectrum characterizing local micro-displacement changes, thereby establishing the spatial geometric mapping relationship of the thermo-mechanical coupling response.

[0047] Based on the coordinate information of the marked units in the real-time infrared thermal radiation intensity distribution map, a high-speed white light interferometer scanning task parameter matrix is ​​constructed to clarify the spatial positioning and scanning priority of each marked unit.

[0048] The parameter matrix is ​​loaded into the control module of the high-speed white light interferometer, and the nanoscale optical phase demodulation mode initialization is performed to set the light source wavelength, reference mirror position and detector integration time of the interferometer.

[0049] During the scanning process, the coordinate index is invoked to drive the interferometer optical head to precisely align with the surface of the marked unit, and multi-angle incident interferometry is performed to capture displacement components in different directions.

[0050] The acquired light intensity interference fringe data sequence is subjected to phase expansion processing, and a phase demodulation algorithm based on quadratic polynomial fitting is used to convert the fringe phase information into an absolute displacement value matrix.

[0051] To improve demodulation accuracy, an interference noise filter is used to limit the frequency domain bandwidth and suppress higher-order harmonics of the phase data, in order to eliminate interference components from sources other than material deformation.

[0052] The filtered absolute displacement matrix is ​​rearranged according to the coordinates of the marked units, and a spatial geometric mapping table is established with the infrared thermal radiation intensity distribution data to form a high-resolution original optical interference pattern of surface micro-deformation.

[0053] By using the aforementioned optical phase demodulation and spatial matching processing methods, the infrared thermal data from the previous step is transformed into deformation spatial mapping data that can be directly used for thermo-mechanical coupling analysis, thereby achieving complete spatiotemporal acquisition accuracy of the material's microscopic response trajectory.

[0054] For example, in a multilayer laminated structure measuring 1200mm × 800mm, the coordinate accuracy of the micro-thermal-coupling marker unit is set to 0.05mm, the wavelength of the high-speed white light interferometer source is set to 632.8nm, the reference mirror position is adjusted to the interference zero point, and the detector integration time is 2ms. After initializing the scanning mode, the optical head performs five-directional incident illumination according to the marker unit coordinates. Phase demodulation in each direction is performed using quadratic polynomial fitting, the coefficient matrix is ​​solved using the least squares method, the noise filter bandwidth is limited to 0.5–5MHz, and the phase data is analyzed in the frequency domain using fast Fourier transform to remove components exceeding the bandwidth range. The final generated absolute displacement matrix corresponds to the deformation of each marker unit surface at the nanoscale, and a spatial geometric mapping table is established with the infrared thermal radiation intensity values. The output high-resolution optical interferometric original spectrum significantly improves the accuracy of hysteresis window identification in subsequent thermal-coupling analysis.

[0055] S2.3: A high-precision timestamp synchronization algorithm is used to perform millisecond-level time alignment processing on the real-time infrared thermal radiation intensity distribution map and the original high-resolution surface micro-deformation optical interference map to construct a multi-modal synchronous observation data frame with a unified spatiotemporal reference, thereby eliminating signal phase mismatch errors caused by differences in sensor sampling frequencies.

[0056] S2.4: Based on the continuous time series thermal radiation intensity values ​​and surface micro-deformation variables in the multimodal synchronous observation data frame, perform sliding window difference operation and gradient vector calculation processing to generate deformation-thermal response trajectory cluster data that characterizes the dynamic evolution of each marked unit with temperature and pressure changes, thereby quantifying the transient rheological behavior characteristics of the material micro-interface.

[0057] Based on continuous time-series thermal radiation intensity values ​​and surface micro-variables from multimodal synchronous observation data frames, a sliding window differential calculation module is invoked under a millisecond-level synchronous alignment reference. This module performs a first-order difference operation on the thermal radiation intensity of each marked unit at adjacent observation times to extract instantaneous temperature changes and form a thermal change time series. The gradient vector calculation module is then invoked to map the differential thermal radiation intensity time series to a spatial coordinate system and calculate its gradient distribution along the two-dimensional direction of the plate surface, quantifying the spatial rate of change of the local temperature field. For the surface micro-variable series, a first-order difference operation is performed using the same sliding window length to obtain the instantaneous micro-displacement changes at adjacent times. This is combined with the original spatial index to calculate the two-dimensional gradient vector of the displacement field, reflecting the rate of micro-morphological evolution. The thermal radiation intensity gradient vector and the surface micro-variable gradient vector are associated and paired in a unified coordinate system to form deformation-thermal gradient combination data corresponding to each marked unit, and a temporal index is introduced to maintain continuity. A trajectory cluster construction algorithm is performed on the deformation-thermal gradient combination data. Gradient pairings of the same labeled unit in different time slices are connected sequentially to generate a complete dynamic response trajectory curve, and physical parameter labels are attached. Through the above processing, the multimodal synchronous observation data frame of the previous step is transformed into deformation-thermal response trajectory cluster data covering all labeled units, realizing a quantitative description of the transient rheological behavior characteristics of the material's micro-interface.

[0058] For example, in a large-size multilayer board lamination experiment, the infrared thermal radiation intensity sampling frequency is set to 500Hz, the sliding window length is selected to be 5ms, and the window step size is 2ms. For a certain marker unit's thermal radiation intensity time series... The formula for calculating the first-order difference is as follows: ,in The window step size is used. The difference results are mapped to plate coordinates using the gradient calculation formula. Obtain the two-dimensional thermal gradient distribution. Consider surface micro-deformations. Perform the same steps to obtain the displacement gradient. In this experiment, deformation-thermal gradient combined data showed that when the local temperature gradient reached 0.12 K / mm, the displacement gradient simultaneously increased to 25 nm / mm, indicating significant transient rheological behavior at the material interface. The trajectory cluster construction results output a 250-point dynamic response curve for each marker unit, used for subsequent spatiotemporal topology reconstruction and noise filtering, effectively improving the accuracy of local pressure response inertia feature extraction.

[0059] S2.5: Perform full-plate spatial topology reconstruction and noise filtering on the deformation-thermal response trajectory cluster data to output a standardized original spatiotemporal dataset covering key areas of large-size multilayer boards, thereby forming the final input data entity that can completely record the inertial evolution process of material pressure response.

[0060] like Figure 2 As shown, step S3 involves comparing and analyzing the deformation-thermal response trajectory clusters in the original spatiotemporal dataset to identify characteristic relaxation time windows that characterize the material's pressure response inertia at each location, thereby generating time-series characteristic parameters that reflect the coupling relationship between local stress relaxation rate and thermal history. Specifically, this includes: S3.1: Perform multidimensional tensor reconstruction processing on the standardized original spatiotemporal dataset to convert the discrete deformation-thermal response trajectory cluster data into a four-dimensional spatiotemporal response tensor with a unified spatial index and time axis alignment. This eliminates the data dimension heterogeneity problem caused by uneven distribution of labeling units and generates a topologically consistent four-dimensional spatiotemporal response tensor as the input basis for subsequent analysis.

[0061] A multidimensional tensor reconstruction process is performed on the noise-removed deformation-thermal response trajectory clusters in the standardized original spatiotemporal dataset covering key areas of large-size multilayer boards to unify spatial indices and temporal axis references and generate a topologically consistent four-dimensional spatiotemporal response tensor. For the infrared thermal radiation intensity sequence and micro-deformation sequence of each labeled unit in the input dataset, a bimodal attribute mapping structure is established, grouping each modal data into the same data unit according to its spatial coordinate index. Interpolation resampling is performed on the temporal axis of each data unit to eliminate differences in sampling temporal resolution between different labeled units and ensure global temporal alignment. A spatial gridding reconstruction method is used to map the discretely distributed labeled unit measurements to physical grid nodes on the board surface, ensuring a one-to-one correspondence between all nodes and their actual physical locations. When constructing the multidimensional tensor architecture, the time dimension, spatial X / Y dimension, and modal dimension are used as four index axes. Tensor filling rules are used to fill in grid positions without measurement data according to a missing value completion strategy, thereby ensuring tensor integrity. The initial four-dimensional tensor after reconstruction is subjected to topological consistency verification, which checks the numerical differences and temporal continuity of physically adjacent nodes in the spatial neighborhood, and removes outlier data points that do not meet the topological consistency constraints. Through multidimensional tensor reconstruction processing, the standardized original spatiotemporal dataset from the previous step is transformed into a four-dimensional spatiotemporal response tensor with a unified index and time base, which can be directly correlated and calculated, thus providing a high-precision input basis for subsequent dynamic coupling coefficient calculation.

[0062] For example, in a large-scale multilayer board lamination experiment, a standardized original spatiotemporal dataset covering an 800×800mm board surface was collected. The infrared thermal radiation intensity of each labeled unit was sampled at 2ms intervals, and the micro-deformation was sampled at 5ms intervals, spatially distributed as 50×50 discrete points. During temporal interpolation resampling, the micro-deformation sequence was resampled to 2ms intervals using a spline interpolation function, aligned with the infrared data. A square grid was used to map the discrete points to a physical grid node coordinate set (Δx=16mm, Δy=16mm). During tensor reconstruction, thermal radiation intensity and micro-deformation were used as two modal dimensions, with a temporal dimension of 200 frames and spatial X / Y dimensions of 50 nodes each, forming a four-dimensional tensor T(X,Y,t,modal), where modal takes values ​​in the range {thermal radiation, micro-deformation}. When performing missing value imputation on all nodes, if a node in a certain frame lacked a thermal radiation value, the average value of the neighboring 3×3 nodes was used for imputation. When verifying topological consistency, the spatial gradient is calculated. If the gradient exceeds a preset threshold of 0.5 K / mm (thermal radiation) or 20 nm / mm (micro-deformation), it is considered an anomaly and discarded, finally yielding a complete four-dimensional spatiotemporal response tensor. This tensor is input into subsequent sliding window cross-correlation calculations, significantly improving the stability of the dynamic coupling coefficient matrix calculation and eliminating coupling coefficient fluctuations caused by sampling differences and spatial omissions.

[0063] S3.2: Perform sliding window cross-correlation operation based on the four-dimensional spatiotemporal response tensor to calculate the dynamic coupling coefficient between the thermal radiation intensity gradient vector and the surface micro-variable gradient vector of each spatial coordinate point on a continuous time slice, thereby quantifying the instantaneous driving efficiency of local temperature and pressure changes on the micro-rheological behavior of the material and generating a dynamic coupling coefficient matrix characterizing the transient driving efficiency.

[0064] Based on the reconstructed four-dimensional spatiotemporal response tensor as input, cross-correlation coefficients are calculated for the thermal radiation intensity gradient vector and surface microvariable gradient vector at each spatial coordinate point on continuous time slices to quantify the instantaneous driving efficiency of synchronous temperature and pressure changes on the material's micro-rheological behavior. A fixed-length sliding time window is selected, within which the thermal radiation intensity gradient vector sequence is centered to achieve a mean of zero to eliminate the influence of DC bias. The same centering process is performed on the surface microvariable gradient vector sequence within the same window to ensure that the cross-correlation calculation results only reflect the dynamically changing portion. Normalized variance processing is performed on the centered thermal radiation intensity gradient vector and surface microvariable gradient vector to obtain a unit variance sequence to improve the feasibility of comparison between data of different dimensions. The cross-correlation coefficients at different window positions are sorted according to the time lag, and the time lag position corresponding to the maximum value is selected as the peak response identifier of the local transient driving efficiency. The peak cross-correlation coefficients of all spatial coordinate points are recorded in matrix form to form a dynamic coupling coefficient matrix. The element values ​​of this matrix directly reflect the degree of instantaneous response of the material at that location under temperature and pressure changes. Through the above processing method, the four-dimensional spatiotemporal response tensor of the previous step is transformed into a dynamic coupling coefficient matrix that can be used to identify stress relaxation time windows, thus realizing the quantitative basis for the inertial analysis of material pressure response.

[0065] For example, during the lamination process of large-size multilayer boards, the thermal radiation intensity gradient vector sequence and the surface micro-variable gradient vector sequence at spatial coordinate points (125mm, 240mm) are selected. The sliding window length is set to 50ms, and the window step size is 10ms. Within a certain window, the thermal radiation intensity gradient value sequence, after centering and mean adjustment, ranges from [-0.08, 0.07], and the micro-variable gradient value sequence ranges from [-0.12, 0.09], both normalized to unit variance. Using the cross-correlation formula, when... At 15ms, the calculated cross-correlation coefficient was 0.86, which is the maximum value within this window. This value is filled into the corresponding position in the dynamic coupling coefficient matrix, indicating that the transient driving efficiency at this point is relatively high under temperature and pressure changes. In the full-plate matrix, the high-coefficient region at this position can be used as a key object for pressure response inertia analysis, for precise definition of the subsequent relaxation time window. In actual verification, micro-pressure compensation control can significantly improve the thickness uniformity of this region.

[0066] S3.3: The nonlinear hysteresis loop fitting algorithm is executed using the dynamic coupling coefficient matrix to extract the time span between the stress relaxation inflection point and the saturation equilibrium point of each marked unit under the heating and pressurization path, thereby defining the hysteresis interval of the material from being subjected to external excitation to producing substantial physical deformation and generating an initial relaxation time series characterizing the single-point hysteresis characteristics.

[0067] For the input conditions of the dynamic coupling coefficient matrix, the correlation values ​​between the thermal radiation intensity gradient vector and the surface micro-deformation gradient vector corresponding to each element of the matrix are loaded as fitting data points.

[0068] Data preprocessing is performed, adjusting each column of the time series in the dynamic coupling coefficient matrix to a zero-mean, unit-variance state through normalization and mean centering operations to eliminate the interference of amplitude differences on the fitting stability.

[0069] A nonlinear hysteresis loop model was constructed, a bidirectional mapping path between stress response variables and thermal driving variables was defined, and a state transition factor was introduced into the model to distinguish the response characteristics of the heating and pressurization stage from those of the unloading stage.

[0070] A combination strategy of piecewise polynomial fitting and power-law function fitting is adopted to identify the curve inflection point in the hysteresis curve. This inflection point is defined as the stress relaxation inflection point, and the index value of the time slice in which it is located is recorded as the lag start time.

[0071] The same combined fitting strategy is applied to continue searching along the curve for the boundary point of the response saturation equilibrium interval. This boundary point is defined as the relaxation saturation equilibrium point, and its time slice index value is recorded as the lag end time.

[0072] The relaxation time span of each marker unit is calculated using a formula. The calculated time span value is then correlated with the spatial coordinates of the corresponding marker unit, and the initial relaxation time series representing the single-point lag characteristics is output as the input basis for S3.4.

[0073] By using the aforementioned nonlinear hysteresis loop fitting algorithm, the dynamic coupling coefficient matrix result from the previous step is transformed into an initial relaxation time series that can quantify the pressure response inertia of each labeled unit, thereby accurately defining the hysteresis interval of the material pressure response and providing high-precision temporal feature parameters for subsequent spatial neighborhood smoothing and time window clustering.

[0074] For example, in the lamination of a 16-layer high-multilayer board measuring 1200mm × 800mm, the single-point data length of the dynamic coupling coefficient matrix is ​​500 time slices, with a sampling interval of 20ms. After normalization, a hybrid fitting of a third-order polynomial and a power-law decay model is used for the coupling coefficient curve of a certain marked unit to identify the time slice index where the inflection point is located. =85, Equilibrium Point Time Slice Index =240. Substituting both into the relaxation time span formula, we get... The corresponding actual time lag is = The lag time is recorded in the initial relaxation time series and then spatially weighted and fused with the results of neighboring marked units in subsequent steps. This significantly improves the time matching accuracy between the pressure compensation signal and the actual material response during the pressing process, resulting in a substantial improvement in thickness uniformity control performance.

[0075] S3.4: Perform spatial neighborhood weighted smoothing filtering based on the initial relaxation time sequence to fuse the initial relaxation time values ​​of adjacent marker units and remove abnormal jump points caused by sensor noise or local defects, thereby correcting the temporal discrepancies caused by measurement errors and generating a smooth relaxation time distribution map with spatial continuity.

[0076] Based on the obtained initial relaxation time series data, spatial neighborhood weighted smoothing filtering is performed on the time series values ​​of the marked units covering the key areas of the entire plate to achieve the fusion of local time values ​​and the suppression of anomalous jumps. The neighborhood coverage and weight allocation function of each marked unit are defined based on the physical grid coordinate system. A weighted model of the time values ​​of adjacent points is constructed using a weight coefficient matrix, and this model is applied to the initial relaxation time series to generate a weighted smoothing result. In the weighted calculation, the time value of each marked unit is averaged with the weight values ​​in its neighborhood. The weight decay in this function is set according to spatial distance and the consistency of local rheological properties of the material. Anomaly detection algorithms are used to perform abrupt change removal processing on the weighted results. The abrupt change judgment condition is set as the difference between the smoothed values ​​of adjacent nodes exceeding twice the standard deviation of the smoothed values ​​of the entire plate. Data points that meet the condition are retained, and anomalous points that do not meet the condition are removed. A second interpolation is performed on the smoothed matrix after removing anomalies to ensure spatial continuity. A bilinear interpolation model is used to supplement time values ​​at gaps, thereby forming a continuous smooth relaxation time distribution map covering the entire plate area. By using the chain-like processing method of weighted fusion, anomaly removal, and interpolation completion, the discrete initial relaxation time series obtained in the previous step is transformed into a smooth relaxation time distribution map with good spatial continuity and noise suppression characteristics, thus realizing a highly stable input basis for feature time window recognition.

[0077] For example, in the monitoring of the lamination process of large-size multilayer boards, the initial relaxation time series is obtained, with time values ​​ranging from 0.12s to 0.85s. The neighborhood coverage of the marker cell in the physical grid is set to a radius of 2 grid cells, and the weighting coefficient function is set to... in The distance between nodes. The distance decay factor was set to 0.5. After weighted smoothing, the smoothed value of each marked unit was averaged in its neighborhood. The anomaly detection threshold was set to 1.5 times the standard deviation of the smoothed value across the entire plate. Outliers were removed using a rejection rule, and missing values ​​were filled in using bilinear interpolation. After this processing, the rate of change of the temporal gradient of the smoothed relaxation time distribution map across the entire plate was significantly reduced compared to the initial data, and the stability of the thickness uniformity control model in the subsequent construction of the dynamic weight matrix for response delay prediction was greatly improved.

[0078] S3.5: Based on the smooth relaxation time distribution map, perform threshold segmentation and clustering mapping to divide the continuous relaxation time values ​​into discretized time window intervals representing different pressure response inertia levels, thereby completing the transformation from continuous physical quantities to controllable scheduling parameters and generating a set of time-series feature parameters for the final construction of the pressure field dynamic weight matrix.

[0079] Based on the spatial continuity of the smooth relaxation time distribution map, the relaxation time series values ​​of each node are loaded as input parameters into the threshold judgment module to select the cutting boundary value that distinguishes different pressure response inertia levels. A multi-segment adaptive threshold segmentation algorithm is used to perform interval partitioning on the entire relaxation time data. Specifically, after comparing the time value of each node with the mean and variance of the entire region, the corresponding inertia level boundary conditions are determined, and a level switch is marked when the value crosses the boundary. A phased clustering center is constructed using a mapping matrix. Cluster analysis is performed on the same level value range after threshold segmentation, calculating the internal time value distribution deviation of each clustering center and removing outliers to ensure the stability of the inertia level partitioning. Based on the clustering results, level number assignment is performed, mapping the time window interval of each node to a discretized inertia level index to form a schedulable level label matrix. Consistency verification is performed on the level label matrix to detect the continuity and physical rationality of the level distribution in the spatial grid, and excessively dispersed or isolated level nodes are adjusted to ensure that the level distribution conforms to the stress transfer law. Through the threshold segmentation and clustering mapping processes described above, the smooth relaxation time distribution map is transformed into a set of discrete time window interval parameters, realizing the transformation from continuous physical quantities to controllable scheduling parameters, and outputting a set of time-series characteristic parameters adapted to the dynamic weight matrix of the pressure field.

[0080] For example, in the lamination process of large-size multilayer boards, the numerical range of the smooth relaxation time distribution map is 2.5ms to 15.8ms. The initial limits of the adaptive threshold segmentation algorithm are set at two segmentation points: 8ms and 12ms, corresponding to low, medium, and high inertia levels, respectively. The relaxation time of each node is compared with the global mean of 9.6ms and the standard deviation of 2.1ms, which meets the requirements. The node is labeled as level L1, satisfying The node is labeled as level L2, satisfying The nodes are labeled as level L3. K-means clustering is performed on nodes within each level, with K set to 3. The maximum deviation of the time values ​​of nodes within each cluster center is calculated to be 1.5ms. Outliers exceeding this deviation are removed and reassigned to the nearest center. The clustered level label matrix is ​​loaded into the consistency verification module to check spatial continuity. It is found that isolated L3 nodes are surrounded by L1 nodes, which are then adjusted to L2 to conform to the stress transmission path. Finally, a set of discretized time window interval parameters containing L1, L2, and L3 levels is formed, successfully realizing the conversion from continuous time values ​​to scheduling levels. This set is used in the dynamic weight matrix of the pressure field to complete the differentiated configuration of command timing, significantly improving the matching accuracy of dynamic compensation response.

[0081] like Figure 3 As shown, step S4 involves mapping the feature relaxation time windows corresponding to all marked units on the entire plate to a dynamic weight matrix of the pressure field, thereby constructing a spatiotemporal distribution model that characterizes the actual response delay time after pressure disturbance but not the current pressure value. Specifically, this includes: S4.1: Perform spatial coordinate index reconstruction processing on the set of time-series feature parameters to map the discrete marker unit relaxation time data to the physical grid coordinate system of the large-size high-multilayer board laminated stack, thereby generating an initial relaxation time distribution matrix with a unified spatial reference as the input basis for subsequent weighted calculations.

[0082] Spatial index mapping initialization is performed on the relaxation time value of each marked unit in the generated time-series feature parameter set to clarify its corresponding position coordinates in the physical structure of large-size high-multilayer board laminate.

[0083] The physical position coordinates of the marked units obtained from the initial mapping are matched with the actual mesh division scheme of the plate surface. A unified mesh coordinate system is constructed based on the length, width and resolution requirements of the press-fit structure, and a unique mesh node index is assigned to each marked unit.

[0084] Interpolation is performed on the relaxation time values ​​of the marked cells with established grid node indexes. A two-way interpolation method based on distance weights is used to fill the irregularly distributed marked cell data into all node positions of the unified grid coordinate system, forming a complete spatial coverage of time parameters.

[0085] Coordinate reference unification processing is performed on the interpolated and filled grid time data. The geometric center of the plate surface is selected as the origin, and the spatial position vector of all nodes is redefined to achieve a unified spatial reference for time data measured in different coordinate systems.

[0086] Matrix reconstruction processing is performed on the grid time data of the normalized spatial reference, combining the coordinates of the two-dimensional spatial nodes with the corresponding relaxation time values ​​into a matrix data structure with a clear row-column correspondence, ensuring that each matrix element is a unique mapping between physical location and time parameter.

[0087] Through the above spatial coordinate index reconstruction process, the set of temporal feature parameters from the previous step is transformed into an initial relaxation time distribution matrix with a unified spatial reference, which serves as the input basis for subsequent anisotropic diffusion and dynamic weight coefficient calculation.

[0088] For example, consider a multilayer board with dimensions of 1200mm × 600mm and a grid resolution of 10mm. The time-series characteristic parameter set contains 95 irregularly distributed marker cells with relaxation times ranging from 0.15s to 0.42s. After mapping initialization, each marker cell is assigned a corresponding grid node index. For instance, the marker cell at the top left corner of the board has coordinates (x=20mm, y=15mm), an index of (2,2), and a corresponding relaxation time of 0.31s. A bidirectional interpolation method is used to fill in the time values ​​for uncovered nodes. The interpolation formula is as follows: in, Let be the interpolation relaxation time at grid node (i,j). For distance weighting function, The relaxation time values ​​are those of adjacent known marked units. The distance between nodes is the Euclidean distance. After normalization, the matrix elements clearly correspond to the physical positions on the board surface, ultimately forming an initial relaxation time distribution matrix with a size of 120×60. This matrix significantly improves spatial continuity in subsequent anisotropic diffusion filtering, ensuring the stability and accuracy of the dynamic weight coefficient calculation process.

[0089] S4.2: Based on the initial relaxation time distribution matrix, perform anisotropic diffusion filtering algorithm to fuse the relaxation time values ​​of adjacent grid nodes and suppress spatiotemporal jumps caused by local measurement noise, thereby generating modified relaxation time field data with spatial continuity and physical smoothness.

[0090] In an initial relaxation time distribution matrix with a unified spatial reference, the matrix data is used as the input for anisotropic diffusion filtering, with the relaxation time value of each grid node as the core diffusion variable. An anisotropic diffusion coefficient field is constructed, with coefficient values ​​set according to the gradient magnitude of the relaxation time difference between adjacent nodes. A larger gradient results in a smaller diffusion coefficient to preserve physical boundary characteristics, while a smaller gradient results in a larger diffusion coefficient to enhance continuity. A spatial discretization format is used to transform the physical grid coordinate system into a node index system, ensuring consistency between the diffusion calculation and the actual stacked geometry. In each iteration step, the relaxation time gradient field is calculated, and a weighted average operation is performed based on the gradient field and the diffusion coefficient field in the local direction to achieve anisotropic differential updates. Boundary conditions are introduced to lock the original time values ​​of key stress concentration areas to prevent distortion of physical characteristics. Differential iteration control parameters are used to set an iteration convergence threshold; the update process terminates when the change in relaxation time of all nodes on the plate falls below the set convergence limit. The output is a corrected relaxation time field data with spatial continuity and preserved physical boundary characteristics, serving as a reliable input reference for subsequent normalization mapping operations. Anisotropic diffusion filtering transforms the initial relaxation time distribution matrix from the previous step into spatially continuous and physically smooth corrected relaxation time field data, achieving the desired technical effects of noise suppression and feature preservation.

[0091] For example, in the physical mesh of a large-size multilayer laminated structure, the node spacing is set to 2 mm, and the reference coefficient of the anisotropic diffusion coefficient field is set to... When the gradient magnitude of the relaxation time difference between adjacent nodes exceeds At ms, the diffusion coefficient decreases to To protect boundary features, an iteration step size is adopted. During the iteration process, the boundary nodes maintain their original relaxation times, while the remaining nodes are updated synchronously according to the formula above. After 120 iterations, the change in relaxation time across all nodes is less than [ms]. The output modified relaxation time field data retains the original high gradient characteristics at the stress concentration boundary and significantly improves spatial continuity in the uniform region, laying a stable physical data foundation for the subsequent generation of dynamic weight tensors of the pressure field.

[0092] S4.3: Perform nonlinear normalization mapping operation using the modified relaxation time field data to convert the absolute time dimension into a dimensionless weight coefficient that characterizes the material response hysteresis under unit pressure disturbance, thereby generating the original pressure field dynamic weight tensor that reflects the local pressure response inertia level.

[0093] Based on the absolute time value of each grid node in the corrected relaxation time field data, a reference time interval for normalization mapping is set, and the upper and lower limits of the reference time interval are taken as the minimum and maximum values ​​of the relaxation time distribution of the entire plate, respectively.

[0094] The modified relaxation time field data is subjected to interval scaling, which converts the absolute time dimension of each grid node into a dimensionless intermediate variable between 0 and 1. This intermediate variable is used to characterize the relative lag.

[0095] The curvature of the above intermediate variables is adjusted by using a nonlinear mapping function, and a weighted exponential mapping is introduced in the region with high lag to enhance the function value resolution in the high lag region. Numerical stabilization is performed on the mapping results. The weight field data is smoothed by bilateral filtering, which suppresses isolated peaks and extremely low values ​​within a reasonable range, while avoiding abrupt changes at the edges of the weight field.

[0096] The processed weighting coefficients are assembled to form the original pressure field dynamic weighting tensor that reflects the inertia level of the pressure response at each location. This tensor maintains a one-to-one correspondence with the physical grid of the pressurized stack in the spatial dimension.

[0097] Through the above-mentioned nonlinear normalization mapping and stabilization processing, the modified relaxation time field data of the previous step is transformed into dimensionless weight coefficients that can directly participate in pressure compensation scheduling, thereby realizing the quantitative expression of the material response lag and forming the original pressure field dynamic weight tensor.

[0098] For example, in a large-size multilayer board laminate with dimensions of 1200mm × 800mm, the relaxation time distribution ranges from 8ms to 45ms, with the upper and lower limits of the reference time interval set to 8ms and 45ms respectively, and the nonlinear enhancement coefficient γ set to 1.5. During the mapping calculation, for example, if the absolute relaxation time of a certain grid node is 30ms, then after normalization, the intermediate variable (30ms) is obtained. 8) / (45 8) = 0.594, further performing nonlinear mapping operations. The weighting coefficient was 0.457. Bilateral filtering was performed on the weighting coefficient field generated across the entire plate, with a spatial filtering radius of 3 grid cells and an intensity weight threshold of 0.2 to eliminate local outliers. The final dynamic weighting tensor of the original pressure field showed a significant increase in weighting coefficients in the high-hysteresis region, accurately quantifying the response delay of the same pressure disturbance in this region. Verification showed that inputting this weighting tensor into the subsequent pressure compensation scheduling module can greatly improve the timing matching accuracy of pressure pulses, resulting in a significant improvement in the thickness deviation of the pressurized product.

[0099] S4.4: Perform neighborhood pressure diffusion constraint coupling processing based on the original pressure field dynamic weight tensor to introduce the lateral stress transfer effect in the resin rheological process and correct the weight deviation of isolated points, thereby generating an enhanced pressure field dynamic weight matrix that includes spatial interaction relationships.

[0100] When performing neighborhood pressure diffusion constraint coupling processing on the original pressure field dynamic weight tensor, the dimensionless weight coefficients of each grid node in the original tensor are used as the input reference.

[0101] For each grid node, a spatial interaction set containing all nodes within its specified radius neighborhood is established, and the weight coefficients of these neighborhood nodes are extracted from the modified relaxation time field data.

[0102] The coupling coefficients based on the lateral stress transfer model of resin rheological properties are calculated for the extracted neighborhood weight coefficient set. The calculation uses the center distance between nodes, resin viscosity and temperature and pressure state as parameters to generate the lateral pressure interference coefficient matrix between nodes.

[0103] The lateral pressure disturbance coefficient matrix is ​​used to perform weighted equalization processing on isolated high-weight or low-weight nodes in the original pressure field dynamic weight tensor, and the weight of each isolated node is adjusted to be a combination of the neighborhood weighted average value and the original value.

[0104] An iterative convergence algorithm is used to update the weights of all nodes on the board through multiple rounds of neighborhood diffusion until the weight distribution of the entire board meets the preset spatial smoothness and physical consistency thresholds.

[0105] Through the aforementioned neighborhood pressure diffusion constraint coupling process, the original pressure field dynamic weight tensor from the previous step is transformed into an enhanced pressure field dynamic weight matrix that includes spatial interaction relationships and eliminates isolated point bias, thereby achieving accurate modeling of lateral stress transmission effects and improving spatiotemporal consistency during pressure regulation.

[0106] For example, in a large-scale multilayer board laminate with physical dimensions of 1200mm × 600mm, the original pressure field dynamic weight tensor is nonlinearly normalized to obtain dimensionless weight coefficients, with values ​​ranging from 0.25 to 0.85. Assuming a neighborhood radius of 15mm, viscosity of 1200Pa·s, temperature of 130℃, and pressure of 1.0MPa, under these parameters, after three iterations, the weight deviation of isolated nodes is eliminated, and the weight distribution across the entire board satisfies the spatial smoothness threshold of 0.05. The enhanced pressure field dynamic weight matrix, when used for subsequent asynchronous scheduling of pressure pulse sequences, can significantly improve the synchronization consistency and local response matching accuracy of multi-point pressure compensation.

[0107] S4.5: Perform spatiotemporal topology encapsulation processing on the enhanced pressure field dynamic weight matrix to format it into a standard control data structure that can directly drive the multi-point pressure compensation instruction sequence for asynchronous scheduling, thereby completing the final construction of the spatiotemporal distribution model characterizing the actual response delay time after pressure disturbance.

[0108] Normalized topological configuration analysis is performed on the dynamic weight matrix of the enhanced pressure field, which includes spatial interaction relationships, to extract the weight coefficients of each spatial node in the matrix and their coupling relationship parameters with neighboring nodes.

[0109] The parsed node weight coefficients are processed by multidimensional index mapping to convert them into composite data tuples containing spatial coordinates, delay levels, and interaction coefficients, in order to establish a physical address linked list structure that can be directly called by the control system.

[0110] The composite data tuple is subjected to time-triggered association operations. By calculating the start time offset and delay compensation amount corresponding to each node, a weighted sequence index table with time-domain sortable characteristics is generated, which forms the basis for asynchronous scheduling of pressure compensation signals.

[0111] A hierarchical encapsulation algorithm is used to merge the spatial topology index table and the temporal trigger index table to form a multi-dimensional matrix control data framework, and a calibration flag is embedded in the framework for difference detection during subsequent dynamic updates.

[0112] The merged control data framework is formatted by encapsulating the spatial index, time delay parameter, weight coefficient and neighborhood interaction parameter of each node into a standardized control data structure according to the control system interface specification, thereby ensuring that the multi-point pressure compensation instruction sequence can be directly read and executed asynchronously.

[0113] Through the above processing method, the enhanced pressure field dynamic weight matrix of the previous step is transformed into a standard control data structure that can directly drive the multi-point pressure compensation command sequence for asynchronous scheduling, thereby realizing the executable spatiotemporal distribution model of the actual response delay time after pressure disturbance.

[0114] For example, on a large-size multilayer board lamination production line, the enhanced pressure field dynamic weight matrix contains 512 spatial nodes. Each node records a dimensionless weight coefficient ranging from 0.05 to 0.95 and an interaction coefficient ranging from 0.01 to 0.20. After performing standardized topology analysis on this matrix, the extracted spatial coordinate resolution is 0.5 mm, and the time delay level is divided into 5 levels. According to the control system interface specification, the node parameters are encapsulated into a four-tuple data structure <spatial coordinates, time delay level, weight coefficient, interaction coefficient>, and stored in a physical address linked list. For a node with a weight coefficient of 0.80, the calculated start-up time offset is 0.096 s. By merging the spatial index table and the timing trigger table and formatting them into a multidimensional matrix data file conforming to the control bus protocol, the control system successfully achieved asynchronous start-up of different nodes within the range of 0–0.12 s. In this embodiment, the implementation of multi-point pressure compensation controls the plate thickness deviation within ±6% and maintains the plate flatness at 0.015 mm / m, achieving the expected performance of significantly improving thickness uniformity and low warpage control.

[0115] Step S5: Based on the dynamic weight matrix of the pressure field, the preset multi-point pressure compensation command sequence is subjected to spatiotemporal decoupling processing. Combined with spatial neighborhood pressure diffusion constraints, a pressure pulse sequence with asynchronous initiation and gradient propagation characteristics is generated to eliminate response lag caused by unified timing commands. Specifically, this includes: S5.1: Perform time axis benchmark alignment processing on the preset multi-point pressure compensation instruction sequence to obtain a standard timing instruction set including the initial synchronization trigger moment, thereby providing a unified time reference benchmark for subsequent asynchronous scheduling.

[0116] The preset multi-point pressure compensation command sequence includes: First, constructing a pressing process database. This database is based on actual pressing data collected from a large number of historical production batches and the corresponding final thickness uniformity evaluation indicators. For each production batch, the following fields are recorded: board code (including total number of layers, core board thickness, copper foil type, and resin grade), actuation array number (spatial coordinates), actual applied pressure pulse sequence (time-pressure discrete points), and the thickness deviation value and local resin flow length measured in the vicinity of that point after pressing. The pressing process database is stored in a relational table, with the primary key being the batch number + actuator ID, and auxiliary indexes including board code, pressure amplitude range, and pulse width range. Second, the raw data is cleaned and tagged during database construction: the measured pressure sequence of each actuation unit is resampled into an equal-length pressure amplitude vector according to a fixed time step (e.g., 10ms), and the corresponding thickness deviation value is normalized into a dimensionless compensation benefit score (the higher the score, the closer the pressure sequence is to the ideal uniform state at that position). Next, for newly added large-size multilayer boards in the production plan, key features (number of layers N, glass transition temperature Tg of resin, and roughness grade of copper foil Rz) are extracted from their board codes. Similar batch searches are performed in the database using these features as a joint index. The feature Euclidean distance between the current board and historical batches is calculated, and the top K batches with the smallest distance (e.g., K=5) are selected as reference sources. Then, for the spatial coordinates of each actuation unit, the pressure amplitude vector corresponding to that coordinate is extracted from the reference source batches, and a weighted average is performed using the compensation benefit scores of each batch as weights to obtain the initial pressure amplitude vector for that coordinate. If a coordinate has no direct match in the reference source batches (due to different actuator mesh resolutions), an inverse distance interpolation method is used to calculate the pressure value at that location using the pressure amplitudes of the four adjacent actuator coordinates. Finally, the initial pressure amplitude vectors of all actuation units are packaged with a globally unified initial trigger time (e.g., 5000ms after system startup) to form a preset multi-point pressure compensation command sequence.

[0117] Input the preset multi-point pressure compensation command sequence, load the global spatiotemporal reference parameters in the enhanced pressure field dynamic weight matrix constructed in the previous step S4.5, and establish a unified time axis reference model.

[0118] Based on the time axis reference model, the high-precision clock synchronization module is invoked to convert control command trigger time information from different sources into serialized data frames under a single time scale, and the global synchronization trigger time index is uniformly identified in the data frames.

[0119] By using the global trigger time index to perform millisecond-level timestamp reconstruction operations, phase correction is performed on multi-point instruction signals with sampling interval differences or communication delays, so that each instruction has strict synchronization marking characteristics on a global scale.

[0120] A sequence consistency check algorithm is used to verify the consistency of the corrected time series, eliminating abnormal offsets in trigger time caused by network jitter or signal packet loss, and ensuring that the formed standard timing instruction set has stability and repeatability.

[0121] Through time base reshaping and consistency verification, the result of the previous step is transformed into a standard timing instruction set containing the initial synchronization trigger time, thereby realizing the unified time reference base required for subsequent asynchronous scheduling.

[0122] For example, in a pressure compensation system for the lamination process of large-size multilayer boards, the input multi-point pressure compensation command sequence includes eight synchronous command signals from the central controller, with an initial trigger time deviation of ±3ms for each signal. After loading the time base parameters of the enhanced pressure field dynamic weight matrix, a high-precision clock synchronization module is invoked to uniformly map all commands to a millisecond-level time scale, setting the global synchronization trigger time to 5000 milliseconds. A timestamp reconstruction operation is performed to converge the deviation range to ±0.5 milliseconds, and two abnormal commands with trigger delays exceeding 1 millisecond are eliminated using consistency checks. The reorganized standard timing command set is stored in array form, with each command starting from a global trigger index of 5000 milliseconds. It is repeatable and compatible with the nonlinear delay mapping operation in S5.2. The application effect is that the calculation accuracy of the start delay of asynchronous command fragments is significantly improved, supporting the dynamic response matching of the multi-point pressure compensation system.

[0123] S5.2: Based on the standard timing instruction set and the pressure field dynamic weight matrix, perform nonlinear time delay mapping operation to calculate the independent start-up delay corresponding to each spatial coordinate point and generate an asynchronous instruction fragment set carrying local response hysteresis characteristics, thereby transforming the unified synchronous instruction into a differentiated timing signal that adapts to the inertia of the material pressure response.

[0124] Using a standard timing instruction set containing the initial synchronization trigger moment and an enhanced pressure field dynamic weight matrix as input data objects, nonlinear time delay mapping operations are performed on each spatial coordinate point.

[0125] By establishing the correspondence between spatial coordinate points and weight coefficients, the coordinate index parser is called to extract the dimensionless weight value corresponding to the coordinate in the matrix, which serves as the coefficient benchmark for time delay calculation.

[0126] Based on the analyzed weight values, a time delay mapping function is constructed, which multiplies the synchronous trigger time parameter in the standard timing instruction set with the weight coefficient and introduces a nonlinear adjustment term. The nonlinear adjustment term is dynamically set according to the local material pressure response inertia level to ensure that different response levels form differentiated start-up times during the mapping process.

[0127] The independent startup delay at each spatial coordinate point is calculated using a formula: The calculated delays for each coordinate point... The numerical values ​​are paired with the original synchronous instruction set item by item to form an asynchronous instruction fragment set indexed by the independent start delay, ensuring that each fragment carries its corresponding local response lag characteristics.

[0128] Through the above processing method, the standard timing instruction set of the previous step is transformed into a differentiated timing signal that can adapt to the pressure response inertia level of different materials, thereby realizing the construction of asynchronous scheduling logic for a multi-point pressure compensation system.

[0129] For example, on a large-size multilayer board stack with a lamination area of ​​1200mm × 600mm, the synchronous triggering reference coefficient of the standard timing instruction set... Set to 4.5ms, the weighting coefficients for a specific region in the enhanced pressure field dynamic weighting matrix. The value is 0.72. Based on the material experimental data, the coefficients of the nonlinear adjustment model are set to... =1.1ms, = 0.5ms =0.3ms. The calculated coordinate point value is 3.24ms. This delay is embedded in the instruction segment corresponding to the coordinate point, forming an asynchronous instruction set containing a delay index. When running this asynchronous instruction set, the start-up times of different actuation units are staggered according to independent delays, significantly improving the timing matching accuracy of the local pressure field. Verification results show that the deviation in the plate thickness area is significantly reduced during the pressing process, and the flatness reaches the control target of ≤0.02mm / m.

[0130] S5.3: Construct a spatial neighborhood pressure diffusion constraint model using the asynchronous instruction fragment set to simulate the lateral stress transmission effect in the resin rheological process and calculate the pressure interference coefficient between adjacent actuation units, thereby generating a spatial coupling correction parameter set to suppress local pressure overshoot or undershoot.

[0131] The asynchronous instruction fragment set is partitioned into neighborhoods based on its spatial coordinate index to determine the set of direct neighboring cells for each actuation unit in the physical mesh of a large-size multilayer laminated plate, forming a spatial neighborhood mapping table that can be used for pressure diffusion analysis. For each node in the neighborhood mapping table, a lateral stress transfer coefficient template from the resin rheological property database is called. By comparing the distance, relative orientation, and thermo-pressure coupling state between neighboring cells, an initial raw interference coefficient matrix reflecting the intensity of local stress interaction is generated. The raw interference coefficient matrix is ​​normalized to convert the interference intensity under different physical locations and material response conditions into dimensionless coefficients, facilitating subsequent unified superposition calculation of multi-source data. Based on the normalized coefficient matrix, convolution integral operations are performed on the time axis. The propagation delay characteristics of pressure pulses in the resin medium are used to temporally modulate the lateral interference. The temporal modulation interference value of each neighboring cell is calculated using a formula. Spatial superposition smoothing filtering is performed on the temporal modulation interference value matrix, fusing the interference values ​​of adjacent nodes to correct abnormal fluctuations caused by measurement noise or local material defects, generating a well-continuous interference distribution field. The disturbance distribution field is multiplied element-wise with the original asynchronous command fragment set to obtain a spatial coupling correction parameter set containing a spatial pressure interrelation correction factor. This parameter set serves as input data to eliminate the risk of local pressure overshoot or undershoot. Through this processing method, the asynchronous command fragment data from the previous step is transformed into spatial coupling correction parameters that reflect the lateral transmission effect of resin rheology, achieving interactive balance control for multi-point pressure compensation.

[0132] For example, in a multilayer laminated structure with dimensions of 1200mm × 600mm, the actuation array comprises 64 elements, and the physical mesh is divided into 8×8 nodes. The neighborhood of each node is set to the four directly adjacent nodes, and the resin rheological delay decay coefficient is... Set to 0.15ms, asynchronous instruction start-up difference time. The interference coefficients vary within the range of 0ms to 40ms. The initial values ​​of the normalized interference coefficient matrix are distributed between 0.1 and 0.6. After calculation using the formula, the modulation values ​​of some high-interference nodes are significantly reduced to the range of 0.25 to 0.35, forming a balanced interference distribution field. After superimposing this interference distribution field with the asynchronous command fragment set, the generated spatial coupling correction parameter set is applied in real time during the pressing process. The verification results show that the peak value of local pressure overshoot is significantly reduced compared with the uncorrected state, the thickness uniformity control effect is improved, and the flatness is maintained within 0.018mm / m during long-term operation, and the thickness deviation is stable within ±8%, meeting the process control objectives.

[0133] S5.4: Perform gradient propagation characteristic superposition processing on the asynchronous instruction fragment set according to the spatial coupling correction parameter set, so as to fuse the asynchronous start-up logic in the time dimension and the pressure diffusion law in the spatial dimension and generate an original pressure pulse waveform sequence with smooth transition characteristics, thereby ensuring the continuity of stress distribution in the pressure field reconstruction process.

[0134] Based on the input conditions of the spatial coupling correction parameter set, a gradient propagation characteristic superposition processing module is constructed for the asynchronous instruction fragment set. A two-dimensional fusion mechanism needs to be established between the asynchronous startup logic on the time axis and the pressure diffusion law in the spatial neighborhood. The startup delay of each asynchronous instruction fragment is paired with the pressure interference coefficient of its physical grid node using a matrix operation to form a composite instruction parameter unit with both time offset and spatial coupling attributes. A gradient propagation kernel function is used to perform two-dimensional convolution processing on each composite instruction parameter unit, fusing the time delay gradient and spatial pressure gradient in a unified convolution window to generate a draft instruction waveform with coherent stress transmission directionality. Based on the output of the convolution window, a multi-scale smoothing algorithm is used to attenuate high-amplitude abrupt changes near the startup time, suppressing stress overshoot caused by sudden changes in the spatial gradient in the local pressure field, and ensuring the spatial continuity of the waveform. The smoothed draft instruction waveform is then weighted and superimposed with the original asynchronous delay signal. The weight coefficients are dynamically allocated according to the stress response hysteresis level of each node, forming a complete time-space gradient fusion waveform matrix. A time-frequency domain consistency correction algorithm is employed to suppress the frequency bands of the fused waveform matrix, weakening high-frequency, short-period pressure fluctuations while retaining low-frequency, long-period, stable transmission components. This yields an original pressure pulse waveform sequence with smooth transition characteristics that meets the continuity requirements of pressure field reconstruction. Through gradient propagation characteristic superposition and smoothing correction, the asynchronous instruction fragment set from the previous step is transformed into a data sequence that can directly drive the actuation array and possesses temporal and spatial coordination characteristics, thus achieving the goal of continuous stress distribution during pressure field reconstruction.

[0135] For example, in the lamination process of large-size multilayer boards, the asynchronous start-up delay of a certain region is set to 15 milliseconds, the pressure interference coefficient of adjacent nodes is 0.35, and the composite instruction parameter unit matrix is ​​constructed with a dimension of 32×32. A third-order gradient propagation kernel function is used, with a kernel matrix center value of 1.0 and a circumferential attenuation coefficient of 0.65. Two-dimensional convolution is performed to obtain a waveform draft with a time delay gradient of 4.2 milliseconds and a spatial pressure gradient peak of 0.42. Multi-scale smoothing is applied to the waveform draft, with window sizes of 3, 5, and 7 respectively. After smoothing, the reduction of high-amplitude abrupt changes significantly increases the proportion of low-frequency long-period components. During the superposition process, the weighting coefficients are divided according to the lag level. Nodes with high lag levels are assigned a time delay weight of 0.7 and a spatial pressure weight of 0.3, while nodes with low lag levels are assigned the opposite weights. Finally, after time-frequency domain consistency correction, pressure fluctuation components above 120Hz are truncated. The output original pressure pulse waveform sequence, in actuator loading verification, achieves improved board thickness uniformity within the deviation range, and the continuous distribution of the pressure field meets the preset stress transmission path requirements.

[0136] S5.5: Perform amplitude normalization and pulse width truncation optimization processing on the original pressure pulse waveform sequence to limit the duration and intensity range of a single micro-pressure pulse of each actuation unit and output the final pressure pulse sequence, thereby eliminating response lag and avoiding the risk of heat accumulation and interface slippage caused by continuous loading.

[0137] With the original pressure pulse waveform sequence having gradient propagation characteristics as input, the amplitude parameters of each actuation unit in the sequence are normalized to unify the pressure amplitude dimensions of different units into a comparable and scalable dimensionless scale, and to ensure that all amplitudes fall within a preset safe range.

[0138] Based on the normalized amplitude parameter and the physical location index corresponding to the actuation unit, the amplitude constraint mapping function is called to perform amplitude compression or gain compensation operations on amplitudes that exceed the safe range, so as to eliminate amplitude anomalies caused by neighborhood interference or cumulative effects.

[0139] The pulse width analysis module is used to calculate the original micro-pulse duration corresponding to each actuation unit for the normalized sequence amplitude parameters, and compare it with the preset upper and lower limits of pulse width to form a pulse width deviation detection dataset.

[0140] The pulse width deviation detection dataset is input into the pulse width truncation optimization algorithm. Pulse widths exceeding the upper limit are truncated to generate pulse width correction values ​​that meet the temperature accumulation control requirements. Pulse widths below the lower limit are compensated for time extension to ensure that each micro-pressure impulse can produce effective physical deformation within the rheological response time window.

[0141] A synchronous amplitude-pulse width optimization coupling strategy is adopted, which pairs and binds the modified amplitude parameter with the truncated optimized pulse width parameter to form the final micro-pressure action parameter set executed by the actuation unit.

[0142] By optimizing the amplitude normalization and pulse width truncation, the gradient propagation pressure pulse waveform of the previous step is transformed into a final pressure pulse sequence that meets the safety range and matches the material response hysteresis characteristics. This eliminates response hysteresis while avoiding the risks of heat accumulation and interface slippage caused by continuous loading.

[0143] For example, in the lamination process of large-size multilayer boards, the original amplitude of a certain actuating unit in the gradient propagation pressure pulse sequence is 0.45 MPa, the preset safe range amplitude upper limit is 0.40 MPa, and the lower limit is 0.25 MPa. After normalization, the amplitude of this unit is... =1.125, compressed to 0.40MPa through amplitude constraint mapping; the original pulse width was 350ms, with a preset upper limit of 300ms and a lower limit of 150ms, which was corrected to 300ms through truncation optimization. Another actuation unit had a normalized amplitude of 0.28MPa and an original pulse width of 120ms, which was extended to 150ms through extension compensation. After amplitude-pulse width coupling, a pressure pulse sequence was formed and applied to the corresponding actuation unit to execute short-duration micro-pressure pulses. The measured thermal accumulation index was significantly reduced, interface slippage was effectively suppressed in continuous pressing batches, and thickness deviation was controlled within ±10%.

[0144] Furthermore, the present invention also includes step S6: driving the actuation array to load the pressure pulse sequence, and controlling each actuation unit to execute a single short-duration micro-pressure pulse only within a specified time before the end of the corresponding characteristic relaxation time window, so as to achieve feedforward pressure field topology reconstruction for the nonlinear effects of local stress distribution. Specifically, this includes: S6.1: Perform actuation unit mapping analysis on the final pressure pulse sequence to extract the independent start-up delay, single micro-pressure pulse duration and intensity range parameters corresponding to each spatial coordinate point, thereby generating a distributed actuation drive instruction set carrying local response hysteresis characteristics as the input basis for subsequent physical execution.

[0145] S6.2: Based on the distributed actuation drive instruction set, execute the asynchronous timing scheduling algorithm to calculate the independent triggering time of each flexible piezoelectric actuation unit relative to the global clock reference, thereby generating an actuator timing control signal sequence with asynchronous start-up characteristics to match the material pressure response inertia difference.

[0146] Based on the input conditions of the distributed actuation drive instruction set, a globally unified time reference system is established for the start-up delay, duration, and intensity range parameters of each actuation unit in the instruction set, serving as the processing object of the asynchronous timing scheduling algorithm. Using the timestamp index of the global clock reference, a delay compensation matrix generation operation is performed. The difference between the independent start-up delay of each unit and the global reference is calculated as the absolute trigger time to eliminate the inconsistency in the start-up phase caused by differences in local response inertia. The absolute trigger times are sorted to generate a time priority queue, and the trigger times of adjacent actuation units in the queue are fine-tuned according to a preset spatial pressure diffusion weight to avoid local pressure overshoot caused by simultaneous start-up. Multi-segment grouping is performed on the time priority queue, mapping the start-up events of actuation units within the same group to asynchronous trigger signal clusters. A non-overlapping time window allocation strategy is adopted to ensure that the minimum safe interval is maintained between each signal cluster. The asynchronous trigger signal clusters are converted into a timing control signal sequence, with an added amplitude threshold constraint to ensure that the control signals generated within each corresponding characteristic relaxation time window can both match the differences in material response inertia and limit energy input to prevent heat accumulation. Through the above asynchronous timing scheduling algorithm, the distributed actuation drive instruction set in step S6.1 is transformed into an actuator timing control signal sequence with asynchronous start-up characteristics, so as to achieve precise matching between the triggering time of the actuation unit and the hysteresis characteristics of the local pressure response of the material.

[0147] For example, in a large-size multilayer board lamination structure with dimensions of 1200mm × 800mm, the actuation array contains 96 actuation units. Each unit has a start-up delay ranging from 5ms to 45ms, an action duration ranging from 80ms to 150ms, and an intensity ranging from 0.3MPa to 0.7MPa. With a global clock reference of 0ms, an asynchronous timing scheduling formula is used to sort the start-up delays according to the global reference difference to generate a time priority queue. The trigger interval between adjacent units in the queue is determined by... The timing was adjusted to be no less than 10ms, with 8 groups, each containing an average of 12 actuator units. In the scheduling results, the trigger times were distributed between 0ms and 60ms, ensuring that actuator starts within the same group did not overlap. The maximum control signal amplitude was constrained to 0.7MPa to match the optimal material response within the corresponding characteristic relaxation time window. After executing this scheduling, the actuator trigger timing and the material response inertia difference achieved high-precision synchronization. The fluctuation amplitude of the local pressure field in the target area during pressing was significantly reduced, the plate thickness uniformity was improved to a deviation ≤ ±10%, and the flatness was stabilized within 0.02mm / m.

[0148] S6.3: The actuator timing control signal sequence is used to drive the independent piezoelectric ceramic stack in the actuator array to perform voltage step modulation processing, so as to convert the electrical signal into a mechanical displacement output with gradient propagation characteristics, thereby generating a single short-time micro-pressure physical force source that acts only within a specified time before the end of the corresponding characteristic relaxation time window.

[0149] Using the actuator timing control signal sequence as input, the independent trigger time point and corresponding pressure pulse amplitude parameters of each actuator unit are extracted, and a multi-channel voltage loading task queue is generated in the controller drive interface module.

[0150] The voltage loading task queue is allocated to each independent piezoelectric ceramic stack channel of the actuation array in ascending order of time to ensure that no voltage interference coupling occurs during multi-channel parallel driving.

[0151] Voltage step modulation is performed on each piezoelectric ceramic stack. A high-speed D / A converter chip is used to generate a voltage step signal with controllable rising and falling edges, and the signal start time and pulse width are set with nanosecond-level resolution.

[0152] In the voltage step modulation process, the power relationship of the mechanical displacement output is generated.

[0153] Based on the specified action time before the end of the characteristic relaxation time window, the step signal is converted into a short-time mechanical displacement output with gradient propagation characteristics on the piezoelectric ceramic stack, forming a single short-time micro-impact physical force source.

[0154] By using the voltage step modulation method described above, the asynchronous timing control signal is transformed into an actual mechanical action that matches the inertia of the material's pressure response, thereby achieving the fixed-point application of the local pressure field during the pressing process.

[0155] For example, in the constant pressure holding stage of laminating large-size multilayer boards, the start-up delay of a certain actuation unit is set to 2.5ms, the duration of a single micro-pressure pulse is 0.8ms, the voltage amplitude is set to 85V, and the piezoelectric constant is... Pick m / V, structural stiffness coefficient Take 1.8 N / m. Substitute this into the formula to calculate the mechanical displacement output, and obtain the displacement as follows: m. Under this displacement, the local pressure field is significantly enhanced, and feedforward loading is achieved within the material rheological hysteresis range. The test results show that the plate thickness deviation is controlled within ±5μm, and the flatness is maintained at 0.015mm / m, achieving simultaneous optimization of thickness uniformity and low warpage.

[0156] S6.4: Perform local stress field superposition calculation based on the single short-time micro-pressure physical force source to integrate the pressure interference coefficients between adjacent actuation units and correct the pressure deviation of isolated points, thereby generating an enhanced local stress distribution field containing spatial interaction relationships to simulate the lateral stress transmission effect in the resin rheological process.

[0157] The input condition is a physical force source matrix of a single short-duration micro-pressure impulse generated by S6.3. Its spatial index corresponds to the position of each element in the actuation array and carries displacement amplitude data with gradient propagation characteristics. Based on the position, amplitude, and duration parameters of each physical force source in the matrix, a local stress field calculation model is established as the processing object of this step. The physical force source matrix is ​​input into the local stress field calculation module. Pressure interference coefficients are retrieved for the spatial neighborhood around each actuation element. Using the pressure interference coefficient group between adjacent actuation elements obtained by S5.3, bidirectional coupling calculation is performed on node pairs to obtain the preliminary superposition value of the pressure field in the neighborhood. Spatial weighting is performed on the superposition value matrix. The interference coefficient is used as a weighting factor and combined with the anisotropy correction of the physical force source amplitude distribution to eliminate the problem of uneven stress transmission caused by the difference in element spacing. Isolated point detection is performed on the corrected pressure field matrix. Nodes whose pressure values ​​differ from those in the neighborhood by more than a preset threshold are screened out. Based on the detection results, deviation compensation is performed to adjust the pressure values ​​of isolated points towards the neighborhood average to enhance the overall spatial continuity. A lateral stress transfer simulation algorithm is used to iteratively calculate the compensated pressure field matrix, constructing an enhanced local stress distribution field that includes the interaction of adjacent nodes. Resin rheological parameters are introduced during the calculation process to reflect the actual physical behavior of stress transfer under the temperature-pressure path. Through the above superposition, correction, compensation, and simulation processing methods, the physical force source of a single short-term micro-pressure impulse is transformed into an enhanced local stress distribution field with spatial interaction relationships, achieving an accurate characterization of the lateral stress transfer effect in the resin rheological process.

[0158] For example, in the constant pressure stage of a large-size multilayer board lamination process, the actuation array contains 64 actuation units with physical force source amplitudes ranging from 0.15 to 0.25 MPa and a duration of 300 ms. The neighborhood pressure interference coefficient was obtained through experimental calibration, with an average value of 0.08 and a fluctuation range of ±0.02. The input physical force source matrix is ​​used to generate a preliminary superposition matrix after interference coefficient retrieval. The weighted calculation uses the element form of the coefficient matrix W. With the corresponding physical force source amplitude matrix F element The products are multiplied and then superimposed to form a correction matrix R, where each element is calculated using the following formula: The isolated point detection threshold was set to ±0.03 MPa of the mean of adjacent nodes. The bias-compensated matrix incorporated resin viscosity η = 3.2 × 10³ Pa·s and thermal expansion coefficient α = 6.5 × 10⁻⁶ Pa·s in the iterative lateral stress transfer simulation. -5 / ℃ was used as the rheological parameter, and the iteration count was 50. After execution, the output enhanced local stress distribution field showed that the maximum difference in stress gradient between adjacent nodes was reduced to 0.012 MPa, verifying a significant improvement in spatial continuity and providing a stable input basis for the feedforward pressure field topology reconstruction in S6.5.

[0159] S6.5: Perform feedforward pressure field topology reconstruction processing on the enhanced local stress distribution field to offset the thickness deviation caused by the influence of material flow, temperature change and nonlinear local stress distribution, thereby outputting a dynamic compensation pressure field with spatiotemporal matching accuracy to achieve the thickness uniformity control target in the lamination process of large-size multilayer boards.

[0160] Based on the stress amplitude and distribution gradient data of each spatial node in the enhanced local stress distribution field, a spatiotemporal correlation matrix analysis is performed on the pressure vector of the key area of ​​the plate surface to remove transient high-frequency disturbance components and obtain a stable stress reference field that can be used for reconstruction.

[0161] By combining the characteristic relaxation time window parameters of each node, the shift operation of the pressure modulation function is performed, and the local stress reference field is phase-adjusted according to the specified trigger time before the end of the time window, forming a stress-driven prediction field with feedforward characteristics.

[0162] The predicted field is convolved with the nonlinear correction operator in the material rheological property model, which consists of the temperature change rate and the resin viscosity, to obtain the pressure pre-distribution matrix that takes into account the heat conduction hysteresis effect.

[0163] The spatial gradient balancing algorithm is applied to the pressure pre-distribution matrix to constrain the pressure difference between adjacent nodes to not exceed a preset limit, thereby avoiding interface slippage caused by local overshoot, and generating a balanced pressure quasi-topology map.

[0164] Based on the pressure quasi-topology diagram, pressure load mapping is performed in the full-plate grid coordinate system, and the modulation values ​​of each node are converted into an actuation command set. This drives the array to form a dynamic compensation pressure field that is highly matched with the actual response time of the material, thereby achieving the cancellation of thickness deviation.

[0165] By using the above processing method, the result of the previous step is transformed into a dynamic compensation pressure field with spatiotemporal matching accuracy, thereby achieving the thickness uniformity control target in the lamination process of large-size multilayer boards.

[0166] For example, during the lamination of a 1200mm × 800mm, 18-layer high-multilayer board, the average inter-node pressure difference in the enhanced local stress distribution field is 0.12 MPa, and the average characteristic relaxation time window is 0.85–1.05 seconds. Through spatiotemporal correlation matrix analysis, the maximum deviation of the stable stress reference field after filtering out high-frequency disturbances is reduced to 0.08 MPa. When performing phase shift calculations for the relaxation time window of each node, the shift amount is set to a fraction of the time window length. Approximately 0.25 seconds before the end, pressure modulation is triggered to generate the prediction field. When the prediction field is convolved with the material rheological property model, the resin viscosity is set to 450 Pa·s, and the temperature change rate is 2.5%. The pressure pre-distribution matrix output by the calculation applies pressure to the edge region 0.18 seconds earlier than the center region. The spatial gradient balancing algorithm sets the pressure difference limit between adjacent nodes to 0.05 MPa, and the maximum gradient value of the quasi-topology map after balancing is controlled at 0.048 MPa. After mapping the quasi-topology map to the array drive instruction set, a dynamic compensation pressure field is formed by a single short-duration micro-pressure pulse (pulse width 0.12 seconds, amplitude 0.06 MPa) at a specified time. After execution, the plate thickness deviation is reduced from ±0.11 mm to ±0.07 mm, and the flatness is improved to 0.015 mm / m, which significantly improves the thickness uniformity and finished product stability during the pressing process.

[0167] Furthermore, the present invention also includes step S7: within a specified time window after each single short-duration micro-pressure pulse execution, a high-speed white light interferometer is invoked to perform a three-dimensional topographic scan of the surrounding area, extracting the transient fluctuation rate of the surface to obtain real-time topographic feedback data for verifying the pressure field reconstruction effect. Specifically, this includes: S7.1: Trigger a high-speed white light interferometer scanning command on the local stress distribution field region after a single short-time micro-pressure pulse is executed, so as to obtain the original light intensity interference fringe image sequence containing microscopic surface height information, thereby establishing the basic optical data source for subsequent phase demodulation.

[0168] S7.2: Based on the original light intensity interference fringe image sequence, perform multi-wavelength phase demodulation algorithm processing to convert the light intensity signal into a nanoscale three-dimensional point cloud dataset representing absolute height values, thereby eliminating interference order ambiguity and generating a surface topography original matrix with high-precision spatial coordinates.

[0169] Based on the original intensity interference fringe image sequence, the multi-wavelength interference data preprocessing module is invoked to spatially register the intensity fringe groups corresponding to different wavelengths according to the pixel index of the imaging array, so as to ensure that the phase calculation of multi-band data can be performed under a unified spatial reference framework.

[0170] The registered multi-band light intensity values ​​are input into the core of the phase demodulation algorithm, and the absolute phase of the stripe phase of each pixel is solved using the multi-wavelength phase expansion formula.

[0171] Based on the obtained absolute phase matrix, and combined with the phase-height scaling factor calibrated by the interferometer system, a height inversion operation is performed to convert the phase information into an absolute height value matrix. This scaling factor is obtained through the relationship between the system's optical path length and the phase period.

[0172] Nanoscale spatial interpolation compensation is performed on the absolute height value matrix to eliminate the slight spatial distortion caused by sampling point misalignment during multi-wavelength combination, so that the point cloud maintains a uniform accuracy benchmark across the entire board.

[0173] By using the 3D coordinate reconstruction module, the compensated height matrix is ​​fused with the spatial position index of the corresponding pixel to form a 3D point cloud dataset with nanometer-level resolution. This dataset fully represents the spatial distribution characteristics of the local surface morphology.

[0174] Through the above-mentioned multi-wavelength phase demodulation and three-dimensional reconstruction processing, the light intensity signal of the fringe image is transformed into a surface morphology original matrix with high-precision spatial coordinates that eliminates interference order ambiguity, thus realizing the basic data preparation required for subsequent local surface fitting and deformation component separation.

[0175] For example, in a local area after the lamination of a large-size multilayer board, a high-speed white light interferometer is used to simultaneously acquire light intensity interference fringes at three wavelengths: 532nm, 632.8nm, and 850nm. The pixel array size is 1024×1024, and the spatial sampling interval is 5μm. After spatial registration, the phase expansion formula is input, and the scaling factor is set to... The phase matrix was converted to an absolute height matrix using a resolution of 1.25 μm. Nanoscale interpolation compensation was applied to unify the height matrix to a full-plate mesh resolution of 1 μm, resulting in an output 3D point cloud dataset containing approximately 1.05 × 10⁶ spatial coordinate points. This point cloud data was validated, showing complete elimination of interference-order ambiguity, a height resolution better than 2 nm, and significantly improved spatial consistency, providing a reliable original topography matrix for subsequent topography analysis.

[0176] S7.3: Perform adaptive surface fitting and background noise removal operations using the original surface topography matrix to separate the low-frequency component caused by the inherent bending of the laminated substrate from the local high-frequency deformation component caused by micro-impact, thereby outputting a pure local deformation elevation map that only reflects the effect of a single short-term micro-impact.

[0177] Based on the spatial coordinate index and absolute height value of the original surface topography matrix, a high-precision height data frame covering the local action area is selected as the fitting input benchmark. Multi-scale surface decomposition is performed on the fitting input benchmark, and an initial surface model is constructed using bicubic spline functions across the entire domain to capture the overall geometric trend of the laminated substrate. Adaptive residual calculation is performed on the initial surface model, mapping the difference between the original height value and the fitted surface to the local residual matrix, dynamically adjusting the fitting control node density to match the spatial scale of high-frequency local deformation. Background noise analysis is performed based on the local residual matrix, using a frequency domain bandpass filter to suppress the inherent low-frequency warping components of the substrate below the preset cutoff frequency, while retaining the micro-impact characteristic components above the preset high-frequency threshold. The filtered residual data is reconstructed into an elevation matrix, and outliers caused by measurement edge effects are removed through boundary constraints to form a pure local high-frequency deformation matrix. The pure high-frequency deformation matrix is ​​superimposed on the global surface model after removing low frequencies to generate a pure local deformation elevation map containing only the effect of a single short-term micro-impact.

[0178] By using adaptive surface fitting and background noise removal, the original surface morphology matrix generated in the previous step is transformed into a clean local deformation elevation map, which enables accurate extraction of the micro-impact effect and provides reliable input data for subsequent calculation of the transient surface undulation rate.

[0179] For example, in a large-scale multilayer board laminate with dimensions of 1200mm × 600mm, the spatial resolution of the original nanoscale three-dimensional point cloud matrix acquired by a white light interferometer is set to 0.5μm, and the radius of the coverage area is 5mm. When performing bicubic spline fitting, the initial control node spacing is set to 10mm. The residual between the original height matrix and the fitted surface is calculated. After fast Fourier transform analysis, the residual matrix is ​​selected with a cutoff frequency of 0.5Hz to suppress low-frequency components and retain high-frequency components above 5Hz. After filtering, the measurement points within a boundary radius of 0.2mm in the residual data are removed, resulting in a pure local high-frequency deformation matrix. When superimposed on the surface model after removing low-frequency components, the pure local deformation elevation map has a thickness fluctuation peak of approximately 0.85μm, which significantly improves the identifiability of micro-impact and retains all high-frequency information for dynamic rheological feature extraction.

[0180] S7.4: Perform spatiotemporal differential gradient calculation processing based on the pure local deformation elevation map to quantify the rate of change of the surface height vector per unit time and generate a surface transient fluctuation rate tensor characterizing the instantaneous flow response of the material after micro-pressure, thereby transforming the static morphology data into dynamic rheological characteristic parameters.

[0181] S7.5: Perform threshold determination and outlier filtering encapsulation processing on the surface transient fluctuation rate tensor to filter out effective feedback data fragments that conform to the physical response law and generate standardized real-time morphology feedback data packets, thereby completing the final input entity construction for verifying the pressure field reconstruction effect and driving the next round of cyclic calibration.

[0182] Furthermore, the present invention also includes step S8: calibrating and correcting the feature relaxation time window in the next cycle based on the real-time topography feedback data, so as to form a closed-loop control dynamic response regulation mechanism and continuously optimize the timing matching accuracy of the multi-point pressure compensation system. Specifically, this includes: S8.1: Perform spatiotemporal coordinate inverse mapping processing on the standardized real-time topography feedback data packet to extract the set of coordinates of abnormal deformation regions that exceed the preset thickness tolerance threshold in the surface transient fluctuation rate tensor, thereby generating a spatial error distribution map characterizing the deviation of the current pressure field reconstruction effect as the input benchmark for closed-loop correction.

[0183] S8.2: Perform residual gradient inversion operation based on the spatial error distribution map and the enhanced local stress distribution field to calculate the time offset between the actual stress relaxation rate and the theoretical prediction value corresponding to each abnormal deformation region, thereby generating a time-series error correction vector sequence that characterizes the prediction deviation of the characteristic relaxation time window.

[0184] S8.3: The smooth relaxation time distribution map generated in the previous cycle is updated by using the time error correction vector sequence to integrate the historical memory weights and the current measured deviation data, thereby outputting an updated smooth relaxation time distribution map containing dynamic correction factors to eliminate model drift.

[0185] S8.4: Re-execute threshold segmentation and clustering mapping processing based on the updated smooth relaxation time distribution map to divide the corrected continuous relaxation time values ​​into discretized time window intervals representing the latest pressure response inertia level, thereby generating a set of calibrated time-series characteristic parameters for the next compression cycle.

[0186] S8.5: Based on the calibrated time-series characteristic parameter set, perform full reconstruction and topology encapsulation processing on the dynamic weight matrix of the pressure field to replace the original hysteresis model and generate a new generation of spatiotemporal distribution model with self-evolution capability, thereby completing the iteration of the dynamic response control mechanism of closed-loop control and establishing the time-series benchmark for the next round of multi-point pressure compensation commands.

[0187] For those skilled in the art, various other corresponding changes and modifications can be made based on the technical solutions and concepts described above, and all such changes and modifications should fall within the protection scope of the claims of this invention.

[0188] Unless otherwise defined, the technical or scientific terms used herein shall have the ordinary meaning as understood by one of ordinary skill in the art to which this application pertains. The terms “first,” “second,” “third,” and similar terms used in this patent application specification and claims do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, the terms “an” or “a” and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms “comprising” or “including” and similar terms mean that the elements or objects preceding “comprising” or “including” encompass the elements or objects listed following “comprising” or “including” and their equivalents, and do not exclude other elements or objects. The “multiple” mentioned in the embodiments of this application refers to two or more. A and / or B indicate three possibilities: A; B; and A and B.

[0189] The above description is merely an exemplary embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and such modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A multi-point pressure compensation lamination method for large-size, multilayer boards to ensure thickness uniformity, characterized in that: Includes the following steps: S1: Embed micro thermo-mechanical coupling marker units composed of low-melting-point phase change materials and oriented micron-sized glass fiber bundles in key areas of large-size high-multilayer board lamination to construct a physical information carrier that can generate controllable micro-deformation and record local flow state during the heating and pressurization process. S2: Based on the thermal and pressure behavior of the micro thermo-mechanical coupling marker unit in the initial stage of heating and pressurization, the infrared thermal radiation offset and surface micro-deformation optical interferogram of each marker unit are collected simultaneously to obtain the original spatiotemporal dataset characterizing the micro-response trajectory of the material. S3: Analyze the original spatiotemporal dataset to identify characteristic relaxation time windows that characterize the inertia of material pressure response at each location, so as to generate time-series characteristic parameters that reflect the coupling relationship between local stress relaxation rate and thermal history. S4: Map the feature relaxation time window corresponding to all marked units of the entire plate into a dynamic weight matrix of the pressure field; S5: Based on the dynamic weight matrix of the pressure field, the preset multi-point pressure compensation command sequence is subjected to spatiotemporal decoupling processing, and a pressure pulse sequence with asynchronous start-up and gradient propagation characteristics is generated by combining the spatial neighborhood pressure diffusion constraint condition, so as to eliminate the response lag caused by the unified timing command.

2. The multi-point pressure compensation lamination method for large-size, multilayer board thickness uniformity according to claim 1, characterized in that, The S3 specifically includes: Multidimensional tensor reconstruction processing is performed on the original spatiotemporal dataset to form a four-dimensional spatiotemporal response tensor with a unified spatial index and time axis alignment; Based on the four-dimensional spatiotemporal response tensor, a sliding window cross-correlation operation is performed to calculate the dynamic coupling system between the thermal radiation intensity gradient vector and the surface micro-variable gradient vector at each spatial coordinate point on a continuous time slice. The nonlinear hysteresis loop fitting algorithm is used to process the dynamic coupling coefficient matrix to extract the time span between the stress relaxation inflection point and the saturation equilibrium point of each marked unit under the heating and pressurization path.

3. The multi-point pressure compensation lamination method for large-size, multilayer board thickness uniformity according to claim 1, characterized in that, S3 specifically includes: performing spatial neighborhood weighted smoothing filtering based on the initial relaxation time sequence to fuse the initial relaxation time values ​​of adjacent marker units and remove abnormal jump points caused by sensor noise or local defects; and performing threshold segmentation and clustering mapping based on the smoothed relaxation time distribution map to divide continuous relaxation time values ​​into discrete time window intervals representing different pressure response inertia levels.

4. The multi-point pressure compensation lamination method for large-size, multilayer board thickness uniformity according to claim 1, characterized in that, The S4 specifically includes: Spatial coordinate index reconstruction processing is performed on the set of time-series feature parameters to map the discrete marker unit relaxation time data to the physical grid coordinate system of the large-size high-multilayer board laminated stack. An anisotropic diffusion filtering algorithm is executed based on the initial relaxation time distribution matrix to fuse the relaxation time values ​​of adjacent grid nodes and suppress spatiotemporal jumps caused by local measurement noise, thereby generating modified relaxation time field data with spatial continuity and physical smoothness. The modified relaxation time field data is used to perform nonlinear normalization mapping operations to convert the absolute time dimension into a dimensionless weighting coefficient that characterizes the material response hysteresis under unit pressure disturbance. The neighborhood pressure diffusion constraint coupling process is performed based on the original pressure field dynamic weight tensor to introduce the lateral stress transfer effect in the resin rheological process and correct the weight deviation of isolated points. The dynamic weight matrix of the pressure field is subjected to spatiotemporal topology encapsulation processing to format it into a standard control data structure that can directly drive the asynchronous scheduling of multi-point pressure compensation instruction sequences.

5. The multi-point pressure compensation lamination method for large-size, multilayer board thickness uniformity according to claim 1, characterized in that, S5 specifically includes: performing time axis reference alignment processing on a preset multi-point pressure compensation instruction sequence to obtain a standard timing instruction set containing the initial synchronization trigger moment.

6. The multi-point pressure compensation lamination method for large-size, multilayer board thickness uniformity according to claim 5, characterized in that, Specifically, S5 further includes: performing nonlinear time delay mapping operations based on the standard timing instruction set and the dynamic weight matrix of the pressure field to calculate the independent start-up delay corresponding to each spatial coordinate point and generate an asynchronous instruction fragment set carrying local response hysteresis characteristics.

7. The multi-point pressure compensation lamination method for large-size, multilayer board thickness uniformity according to claim 6, characterized in that, Specifically, S5 also includes: using the asynchronous instruction fragment set to construct a spatial neighborhood pressure diffusion constraint model to simulate the lateral stress transmission effect in the resin rheological process and calculate the pressure interference coefficient between adjacent actuation units.

8. The multi-point pressure compensation lamination method for large-size, multilayer board thickness uniformity according to claim 7, characterized in that, S5 specifically includes: performing gradient propagation characteristic superposition processing on the asynchronous instruction fragment set according to the spatial coupling correction parameter set, so as to integrate the asynchronous startup logic in the time dimension and the pressure diffusion law in the spatial dimension.

9. The multi-point pressure compensation lamination method for large-size high-multilayer board thickness uniformity according to claim 8, characterized in that, S5 specifically includes: performing amplitude normalization and pulse width truncation optimization processing on the original pressure pulse waveform sequence to limit the duration and intensity range of a single micro-pressure pulse of each actuation unit and output the final pressure pulse sequence.

10. The multi-point pressure compensation lamination method for large-size high-multilayer board thickness uniformity according to claim 2, characterized in that, The multidimensional tensor reconstruction process unifies the discrete labeled unit data into a four-dimensional spatiotemporal response tensor. It uses sliding window cross-correlation analysis and nonlinear hysteresis loop fitting algorithm to automatically identify the pressure response hysteresis intervals of each region of the material and extract the characteristic relaxation time windows.

Images