An interface bonding method between a calcium silicate panel and an EPS core material

Abstract

The application provides an interface sticking method between a calcium silicate panel and an EPS core material, belongs to the technical field of interface sticking between a calcium silicate panel and an EPS core material, and through thickness sanding treatment on the EPS plate and application of a silane coupling agent primer, gradient polyurethane composite glue is coated on the surface of the EPS plate by using a follow-up scraper system, time sequence data of the glue coating amount is input into a thermal-mechanical coupling glue layer quality prediction model, the model embeds Arrhenius curing kinetics into a gated recurrent unit memory updating mechanism, glue repairing is triggered when the debonding area growth rate exceeds a warning threshold, after temperature cycle aging verification, measured data is fed back to the model training set to complete iterative updating, and the technical problems that the curing quality of the glue layer between the calcium silicate panel and the EPS core material is difficult to be predicted in real time due to the discreteness of process parameters, and the debonding defect cannot be quantitatively tracked in the subcritical stage are solved.

Description

Technical Field

[0001] This invention belongs to the field of interface bonding technology between calcium silicate panels and EPS core materials. Specifically, it relates to an interface bonding method between calcium silicate panels and EPS core materials. Background Technology

[0002] Calcium silicate panel and EPS core composite panels are widely used in building envelopes and decoration. The quality of the interface bonding directly determines the long-term reliability of the composite panel. In existing technologies, interface bonding typically involves roller-applying polyurethane adhesive or cement-based interface agent, combined with pressing fixtures to complete the composite process. The curing effect is controlled by adjusting parameters such as pressing pressure, temperature, and time. Building upon this, some production lines have introduced laser thickness gauges to monitor the adhesive layer thickness or used ultrasonic scanning for offline defect detection of the cured composite panel, aiming to improve the stability of the interface quality.

[0003] However, under actual production conditions, the combined effects of variations in the thickness of incoming EPS boards, fluctuations in ambient temperature and humidity, and batch-to-batch differences in adhesive viscosity cause a continuous shift in the curing kinetics during the pressing process. Due to the chemical-mechanical coupling characteristics of the curing process, adjusting a single process parameter cannot accurately reflect the true level of current curing, and offline ultrasonic testing can only detect defects after complete debonding.

[0004] In current composite panel production, the lack of a predictive mechanism that can correlate process parameter timing data with interface curing kinetics in real time, and the lack of a method to continuously and quantitatively track the geometric evolution of the debonding region under temperature cycling loads, means that the setting of the pressing time relies on experience. Debonding defects are only discovered after reaching a critical state, resulting in delayed repair opportunities and insufficient interface reliability. In other words, existing technologies suffer from technical problems such as the difficulty in predicting the real-time curing quality of the adhesive layer at the interface between calcium silicate panel and EPS core material due to the discreteness of process parameters, and the inability to quantitatively track debonding defects in the subcritical stage. Summary of the Invention

[0005] In view of this, the present invention provides an interface bonding method between a calcium silicate panel and an EPS core material, which can solve the technical problems in the prior art where the curing quality of the adhesive layer at the interface between the calcium silicate panel and the EPS core material is difficult to predict in real time due to the discreteness of process parameters, and the debonding defects cannot be quantitatively tracked in the subcritical stage.

[0006] This invention is implemented as follows: This invention provides a method for bonding a calcium silicate panel to an EPS core material, comprising the following steps:

[0007] The EPS board is sanded to a fixed thickness, and the calcium silicate panel surface is mechanically polished and activated. After activation, a silane coupling agent base coating is applied to the calcium silicate panel surface and the EPS board surface respectively.

[0008] A follow-up doctor blade system is used to roll-coat gradient polyurethane composite adhesive. The follow-up doctor blade system collects the height deviation signal of the EPS board surface in real time and compensates for the surface undulation of the EPS board. The time series data of adhesive application amount is collected by the displacement sensor of the follow-up doctor blade system and then input into the interface thermo-mechanical coupling adhesive layer quality prediction model.

[0009] The EPS board that has been rolled and coated is pressed and laminated with the calcium silicate panel in a pressing fixture. The pressing time is dynamically determined by the current degree of curing value output by the interface thermo-mechanical coupling adhesive layer quality prediction model. The pressing pressure curve, curing temperature-time curve and rotational rheometer viscosity curve during the pressing process are collected in real time and input into the interface thermo-mechanical coupling adhesive layer quality prediction model.

[0010] After lamination, the interface area of ​​the composite board is tracked by variational level set interface defect evolution based on interface thermo-mechanical coupling energy functional. The tracking results output the debonding area growth rate. The debonding area growth rate is compared with the debonding warning threshold. When the debonding area growth rate exceeds the debonding warning threshold, the glue repair process is triggered.

[0011] The composite board after inspection was subjected to temperature cycling aging verification. After aging, the interface peel strength was measured. The standard for acceptance was that the interface peel strength was not lower than the qualified threshold.

[0012] The time-series data of adhesive application amount, pressing pressure curve, curing temperature-time curve, rotational rheometer viscosity curve, measured value of interfacial peel strength, measured value of adhesive shrinkage rate and measured value of thermal cycle life corresponding to qualified composite boards are fed back to the training set of the interfacial thermo-mechanical coupling adhesive layer quality prediction model to complete one iterative update of the interfacial thermo-mechanical coupling adhesive layer quality prediction model.

[0013] Specifically, the fixed-thickness sanding process involves using a wide-width sander to simultaneously grind both the top and bottom surfaces of the EPS board. The grinding amount is determined based on the thickness deviation of the incoming EPS board material, and the surface flatness deviation of the EPS board after grinding is controlled within the flatness deviation threshold.

[0014] The silane coupling agent undercoating layer refers to the layer in which the silane coupling agent is applied. -aminopropyltriethoxysilane or - Glycidyl etheroxypropyltrimethoxysilane is uniformly sprayed in aqueous solution onto the surface of calcium silicate panel and EPS board. After static hydrolysis, it forms an interface transition layer mainly composed of covalent bonds.

[0015] The aforementioned follow-up scraper system refers to an adhesive application device consisting of a servo motor-driven scraper lifting mechanism, a displacement sensor, and a closed-loop controller. The closed-loop controller drives the scraper to follow the undulations of the EPS board surface in the vertical direction based on the height deviation signal.

[0016] The gradient polyurethane composite adhesive refers to a composite adhesive prepared by mixing flexible silicone-modified polyurethane and rigid cement-based interface agent in a continuous gradient ratio. The adhesive layer density on the side closer to the EPS board is lower than that on the side closer to the calcium silicate panel, and the density transitions continuously in the thickness direction of the adhesive layer.

[0017] The interface thermo-mechanical coupling adhesive layer quality prediction model refers to a time-series recursive residual prediction model based on the mapping relationship between process parameters and interface performance. The input layer receives adhesive application time-series data, pressing pressure curve, curing temperature-time curve and rotational rheometer viscosity curve, and is connected to a dual-layer gated loop unit encoder. A residual jump connection is set between the dual-layer gated loop units.

[0018] The interface thermo-mechanical coupling adhesive layer quality prediction model inserts a physical constraint layer after each gated cycle unit time step. The physical constraint layer explicitly calculates the current degree of cure based on the Arrhenius equation. The current degree of curing The hidden state of the next time step is injected by modulating the forget gate weight matrix.

[0019] The interface thermo-mechanical coupling adhesive layer quality prediction model includes an interface quality comprehensive evaluation function. The interface quality comprehensive evaluation function calculates the interface quality comprehensive evaluation value based on the normalized value of the predicted interface peel strength, the normalized value of the predicted adhesive layer shrinkage rate, and the normalized value of the predicted thermal cycle life. The learning rate parameter is adjusted according to the interval in which the interface quality comprehensive evaluation value is located.

[0020] The variational level set interface defect evolution tracking based on the interface thermo-mechanical coupling energy functional refers to modeling the geometric boundary of the interface debonding region as an implicit level set function. The energy functional of the debonding evolution consists of three terms: elastic strain energy, interface fracture energy, and penalty regularization term. At each time step, the level set function is updated along the negative gradient direction of the energy functional with respect to the Gâteaux derivative of the level set function.

[0021] The debonding warning threshold refers to the upper limit of the debonding area growth rate after a single temperature cycle. The debonding warning ratio that leads to the average debonding area growth rate of the overall peeling is taken as the debonding warning threshold.

[0022] The elastic strain energy is calculated based on Eshelby inclusion theory to determine the equivalent stress concentration in the debonding zone. The interface fracture energy is of the Griffith type, and the fracture energy per unit area is... The sign distance property of the level set function constrained by the penalty regularization term was determined by fracture toughness test of double cantilever beam specimens.

[0023] Wherein, the fracture energy per unit area The values ​​are calibrated at multiple temperatures and interpolated based on the current temperature field in the variational level set interface defect evolution tracking based on the interface thermo-mechanical coupling energy functional.

[0024] The temperature cycling aging verification shall be performed on a number of cycles that are not less than a threshold number of cycles, and the interface peel strength shall be measured after the aging is completed.

[0025] The loss function of the interface thermo-mechanical coupling adhesive layer quality prediction model is composed of a weighted sum of a data error term and a physical consistency penalty term. The physical consistency penalty term is the L2 norm of the difference between the integral of the predicted curing degree curve and the integral of the Kamal kinetic equation, and the weight coefficient increases with the training rounds.

[0026] The flatness deviation threshold is ±0.3mm; the coating amount of the silane coupling agent primer is 20-40g. The static hydrolysis time is 10–20 min; the displacement accuracy of the follow-up scraper system is ±0.05 mm, and the response frequency is not less than 50 Hz; the adhesive layer density of the gradient polyurethane composite adhesive on the side closest to the EPS board is 60. The adhesive layer density on the side closest to the calcium silicate panel is 300. The following parameters are required: elongation at break greater than 300%; adhesive layer thickness controlled between 100 and 500 μm; pressing pressure controlled between 0.05 and 0.20 MPa; pressing temperature controlled between 15 and 45℃; interfacial peel strength qualification threshold of 1.5 MPa; cycle count threshold of 200 cycles; debonding warning ratio of 30%; and unit area fracture energy. The calibration was performed at three temperatures: -20℃, 23℃, and 80℃.

[0027] This invention employs a method that combines an interface thermo-mechanical coupling adhesive layer quality prediction model with variational level set interface defect evolution tracking, thereby solving the technical problems of difficulty in real-time prediction of adhesive layer curing quality and inability to quantitatively track debonding defects in the subcritical stage.

[0028] This invention embeds the Arrhenius curing kinetics equation as a physical constraint layer into the memory update mechanism of a gated loop unit. This allows the model to infer the current degree of curing based on chemical reaction laws even when process parameters fluctuate, thus dynamically determining the pressing time and avoiding problems of insufficient curing or over-pressing caused by empirical settings. Traditional offline detection can only provide defect judgment after debonding has fully formed, while variational level set interface defect evolution tracking implicitly embeds the debonding boundary into a level set function and uses elastic strain energy to drive boundary evolution, achieving continuous tracking of the geometry and expansion path of the debonding region. This allows the timing of adhesive repair to be triggered when debonding expansion is still in the subcritical stage.

[0029] In summary, this invention solves the technical problems mentioned in the background art, such as the difficulty in predicting the curing quality of the adhesive layer at the interface between the calcium silicate panel and the EPS core material due to the discreteness of process parameters, and the inability to quantitatively track debonding defects in the subcritical stage. Attached Figure Description

[0030] Figure 1 This is a flowchart of the method of the present invention.

[0031] Figure 2 This is a real-time tracking curve for compensating for the surface height deviation and adhesive layer thickness of EPS boards in a follow-up scraper system.

[0032] Figure 3 This is a comparison chart of the curing degree prediction curve of the interfacial thermo-mechanical coupling adhesive layer quality prediction model and the theoretical curve of the Kamal kinetic equation.

[0033] Figure 4 This is a graph showing the evolution of the interface debonding area growth rate with the number of cycles during the temperature cycling aging process of the variational level set tracking algorithm. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below.

[0035] like Figure 1 The diagram shows a flowchart of an interface bonding method between a calcium silicate panel and an EPS core material provided by the present invention. This method includes the following steps:

[0036] S01. The EPS board is sanded to a fixed thickness to control the surface flatness deviation within ±0.3mm. Then, the calcium silicate panel surface is mechanically polished and activated. After activation, a silane coupling agent base coating is applied to the calcium silicate panel surface and the EPS board surface respectively.

[0037] S02. A follow-up doctor blade system is used to roll-coat the gradient polyurethane composite adhesive. The follow-up doctor blade system collects the height deviation signal of the EPS board surface in real time and compensates for the surface undulation of the EPS board. The adhesive layer thickness is controlled within the range of 100 to 500 μm. The adhesive application amount time series data is collected by the displacement sensor of the follow-up doctor blade system and then input into the interface thermo-mechanical coupling adhesive layer quality prediction model.

[0038] S03. The EPS board with roller coating and the calcium silicate panel are pressurized and laminated in a pressing fixture. The pressing pressure is controlled at 0.05-0.20 MPa, the pressing temperature is controlled at 15-45℃, and the pressing time is dynamically determined by the current degree of curing value output by the interface thermo-mechanical coupling adhesive layer quality prediction model. The pressing pressure curve, curing temperature-time curve and rotational rheometer viscosity curve during the pressing process are collected in real time and input into the interface thermo-mechanical coupling adhesive layer quality prediction model.

[0039] S04. After pressing, perform variational level set interface defect evolution tracking based on interface thermo-mechanical coupling energy functional on the interface area of ​​the composite board. The tracking result outputs the debonding area growth rate. Compare the debonding area growth rate with the debonding warning threshold. When the debonding area growth rate exceeds the debonding warning threshold, trigger the glue repair process. After the glue repair is completed, re-execute S04.

[0040] S05. After passing the S04 test, the composite board is subjected to temperature cycling aging verification. The temperature cycling range is -20 to 80℃, and the number of cycles is not less than 200. After aging, the interfacial peel strength is measured. The standard for acceptance is that the interfacial peel strength is not less than 1.5MPa.

[0041] S06. The time-series data of adhesive application amount, pressing pressure curve, curing temperature-time curve, rotational rheometer viscosity curve, measured value of interfacial peel strength, measured value of adhesive shrinkage rate and measured value of thermal cycle life corresponding to the qualified composite board are transmitted back to the training set of the interfacial thermo-mechanical coupling adhesive layer quality prediction model to complete one iterative update of the interfacial thermo-mechanical coupling adhesive layer quality prediction model.

[0042] The thickness sanding process refers to simultaneously grinding both sides of the EPS board using a wide-width sander. The grinding amount is determined based on the thickness deviation of the incoming EPS board material. After grinding, the surface flatness deviation of the EPS board is controlled within ±0.3mm. The ±0.3mm threshold is obtained as follows: In the roller coating process, using EPS boards with different surface flatness deviations as variables, the uniformity of the adhesive layer thickness and the interfacial peel strength are measured respectively. The relationship between the surface flatness deviation and the standard deviation of the interfacial peel strength is statistically analyzed. The upper limit of the surface flatness deviation corresponding to the standard deviation of the interfacial peel strength being lower than 10% of the average interfacial peel strength is set at ±0.3mm, and this is verified after at least three batches of repeated experiments.

[0043] Among them, the silane coupling agent primer coating refers to the coating layer containing silane coupling agents. -aminopropyltriethoxysilane or Glycidyl etheroxypropyltrimethoxysilane was uniformly sprayed in aqueous solution onto the surface of calcium silicate panels and EPS boards, with a coating amount of 20-40 g / L. The static hydrolysis time is 10-20 minutes to form an interface transition layer mainly composed of covalent bonds, which is used to improve the chemical bonding strength between the gradient polyurethane composite adhesive and the calcium silicate panel and EPS board.

[0044] The follow-up scraper system refers to a glue application device consisting of a servo motor-driven scraper lifting mechanism, a displacement sensor, and a closed-loop controller. The displacement sensor collects the surface height of the EPS board in real time, and the closed-loop controller drives the scraper to follow the undulations of the EPS board surface in the vertical direction based on the height deviation signal. The displacement accuracy is ±0.05mm, and the response frequency is not less than 50Hz, so that the glue layer thickness remains stable in the undulating area of ​​the EPS board surface.

[0045] Among them, gradient polyurethane composite adhesive refers to a composite adhesive prepared by mixing flexible silicone-modified polyurethane and rigid cement-based interface agent in a continuously gradient ratio, with the adhesive layer density on the side closest to the EPS board being 60. The adhesive layer density on the side closest to the calcium silicate panel is 300. The density transitions continuously along the thickness of the adhesive layer, and the elongation at break is greater than 300%. The purpose of this gradient design is that the coefficient of thermal expansion of the EPS board is approximately... The coefficient of thermal expansion of calcium silicate panels is approximately The difference between the two is as high as Under temperature cycling, alternating shear stress is generated at the interface. The gradient polyurethane composite adhesive concentrates and transfers strain to the flexible region for dissipation through continuous modulus changes, suppressing the propagation of interface fatigue cracks. The density gradient range and elongation at break threshold of the gradient polyurethane composite adhesive are obtained by the following method: The formulation density starts from 50... Up to 350 A series of polyurethane foam samples were tested under temperature cycling conditions (-20 to 80℃, 200 cycles) to determine the interfacial peel strength retention rate of each sample. The density gradient range was determined by the density start and end values ​​corresponding to the highest interfacial peel strength retention rate. The elongation at break threshold was determined by a combination of tensile tests and fatigue crack propagation rate calculations of the same batch of samples, and was verified by repeated experiments in no less than 3 batches.

[0046] Among them, the interface thermo-mechanical coupling adhesive layer quality prediction model (THCB-Net) is a time-series recursive residual prediction model based on the mapping relationship between process parameters and interface performance. The specific structure of the interface thermo-mechanical coupling adhesive layer quality prediction model is as follows: The input layer receives time-series data of adhesive application amount, pressing pressure curve, curing temperature-time curve, and rotational rheometer viscosity curve at 30-second sampling intervals, totaling four types of time-series features, with a feature dimension of 4 per time step; A dual-layer gated loop unit encoder follows the input layer, with a hidden state dimension of 256 for each layer; Residual jump connections are set between the dual-layer gated loop units, with one level of residual jump connection each from the input layer to the first gated loop unit, from the first gated loop unit to the second gated loop unit, and from the second gated loop unit to the output layer, for a total of three levels of residual jump connections; A physical constraint layer is inserted after each time step of the gated loop unit, explicitly calculating the current degree of curing according to the Arrhenius equation. The current degree of curing As a bypass signal injected into the hidden state of the next time step, the specific implementation method is based on the current solidification degree. A forget gate weight matrix is ​​modulated to form a memory update mechanism driven by a solidified dynamic equation. The network output head has three fully connected branches in parallel, outputting the predicted interface peeling strength, predicted gel layer shrinkage rate, and predicted thermal cycling lifetime, respectively. Weight allocation between neurons uses L1 norm sparsity regularization constraints, activation allocation between layers uses batch normalized layer gradient stabilization, data loop sequence allocation uses truncated backpropagation with a truncation step size of 32 steps, memory allocation uses a gradient checkpoint mechanism to retain only key intermediate activations during forward propagation, and memory allocation uses a prefetch data loader to preload the next batch of data into memory. CUDA stream allocation separates the forward computation stream and the data transmission stream into two independent CUDA streams for parallel execution. Parallel computation of different convolutional kernels is achieved using CUDA warp-level parallel allocation. In terms of hierarchical allocation, the computation of the physical constraint layer is offloaded to the CPU to free up GPU computing power for gated recurrent unit matrix operations. The loss function consists of a weighted sum of a data error term and a physical consistency penalty term. The data error term is the mean squared error between the predicted and measured interface peel strength, and the physical consistency penalty term is the L2 norm of the difference between the integral of the predicted curing degree curve and the integral of the Kamal kinetic equation. The weighting coefficients for both are... The initial value is 0.1, increasing to 0.5 with each training round. The principle of the interface thermo-mechanical coupling adhesive layer quality prediction model is as follows: the adhesive layer curing process is regarded as a chemical-mechanical coupling time series driven by the adhesive amount time series data, the pressing pressure curve, the curing temperature-time curve, and the rotational rheometer viscosity curve. The gated loop unit captures the time dependence of process parameters. The physical constraint layer explicitly embeds the Arrhenius curing kinetics into the gating mechanism, internalizing the chemical reaction law in the network. The three-level residual jump connection prevents gradient vanishing and retains the direct transmission path of process characteristics of each layer. The technical effect of the interface thermo-mechanical coupling adhesive layer quality prediction model is that by embedding the physical law of curing kinetics into the network gating structure, the model can still give a reasonable interface performance prediction based on the chemical reaction law when the process parameters deviate from the historical training distribution. Under small sample conditions, it improves the guidance accuracy of parameters such as pressing time and curing temperature, and reduces the risk of insufficient interface peel strength or adhesive layer cohesion failure caused by improper process parameter settings.

[0047] The steps for establishing the training dataset for the interface thermo-mechanical coupling adhesive layer quality prediction model specifically include: collecting no fewer than 500 sets of actual production process records, each set of records covering adhesive application amount time-series data, pressing pressure curve, curing temperature-time curve, rotational rheometer viscosity curve, and corresponding measured values ​​of interface peel strength, adhesive layer shrinkage rate, and thermal cycle life; and removing outliers from the collected data, with outlier removal based on box curves. Figure 3 The interquartile range criterion is used to fill in missing time steps with linear interpolation; the dataset is divided into training, validation and test sets in a 7:1:2 ratio; and zero-mean unit variance normalization is performed on each input feature.

[0048] The specific steps for training the interface thermo-mechanical coupling adhesive layer quality prediction model include: using the Adam optimizer with an initial learning rate of... The training batch size is 32; the loss is calculated on the validation set every 5 training epochs; if the loss does not decrease for 10 consecutive validation epochs, the learning rate is multiplied by 0.5; the reaction rate constant of the Kamal kinetic equation in the physical constraint layer. , and activation energy After being calibrated by differential scanning calorimetry experiments, the model weights were fixed as constants; the total number of training rounds was no less than 200 rounds, and the model weights with the lowest root mean square error in predicting the interface peel strength on the test set were used as the final weights.

[0049] The interface thermo-mechanical coupling adhesive layer quality prediction model includes an interface quality comprehensive evaluation function, which is used to adjust the learning rate parameter of the model. The interface quality comprehensive evaluation function is based on the normalized value of the predicted interface peel strength at the current time step. Normalized value of predicted adhesive layer shrinkage and the predicted normalized value of thermal cycling lifetime Calculate the overall evaluation value of interface quality The formula is expressed as follows: ;in , , These are the standard reference values ​​for predicting interfacial peel strength, predicting adhesive layer shrinkage, and predicting thermal cycling life, respectively. , , The weighting coefficients were determined through orthogonal experimental design to statistically analyze the prediction errors of the model under different weight combinations. When the comprehensive evaluation value of the interface quality... When the learning rate remains unchanged, the overall interface quality evaluation value remains constant. At that time, the learning rate is multiplied by 0.8 for decay adjustment; when the overall interface quality evaluation value... At that time, the learning rate is multiplied by 0.5 for decay adjustment, while the weight coefficient of the physical consistency penalty term is increased. When the overall interface quality evaluation value When the learning rate is reset to its initial value, the learning rate is reset to its initial value. This triggers a rollback of the model weights to the previous optimal point of the validation loss. The aforementioned standard reference value... , , The following methods were used to obtain the corresponding standard reference values: In the training dataset, the 90th percentile of the measured values ​​of interface peel strength, adhesive shrinkage rate, and thermal cycle life were taken as the corresponding standard reference values, and the values ​​were verified by no less than 3 batches of repeated experiments.

[0050] Among them, variational level set interface defect evolution tracking based on interface thermo-mechanical coupling energy functional refers to modeling the geometric boundary of the interface debonding region as an implicit level set function. Energy functional of decohesion evolution It consists of three terms: elastic strain energy, interfacial fracture energy, and a penalty regularization term; the elastic strain energy is calculated based on Eshelby inclusion theory to determine the equivalent stress concentration in the debonding zone; the interfacial fracture energy is Griffith type, and the fracture energy per unit area is... The sign distance property of the level set function constrained by the penalty regularization term is determined by fracture toughness test of double cantilever beam specimens; the algorithm solves the thermoelastic coupled finite element equations sequentially at each time step to obtain the full-field stress and calculate the energy functional. Regarding level set functions The Gâteaux derivative is used to update the level set function along the negative gradient direction. The algorithm performs narrowband reinitialization and extracts the zero level set as the current debonding boundary; it outputs the debonding area growth rate and predicts the critical moment of peeling failure. The principle of variational level set interface defect evolution tracking based on the interface thermo-mechanical coupling energy functional lies in implicitly embedding the debonding region boundary into the level set function. Topological changes such as the splitting and merging of decohesion regions are naturally achieved during function evolution without the need for mesh remapping. The Gâteaux derivative provides the energy functional. Lowering the fastest boundary update direction physically corresponds to the mechanical mechanism of debonding propagation driven by the release of elastic strain energy. The thermoelastic coupled finite element equation explicitly incorporates the contribution of the temperature field to the interface stress into the calculation. The effect of the variational level set interface defect evolution tracking technology based on the interface thermo-mechanical coupled energy functional is that it realizes continuous tracking of the geometry and propagation path of the debonding region of the composite plate under temperature cyclic loading, providing quantitative geometric evolution information for online determination of whether debonding has reached the critical state, and shifting the timing of glue repair to when the debonding propagation is still in the subcritical stage, thereby improving the overall interface reliability.

[0051] The debonding warning threshold refers to the upper limit of the debonding area growth rate after a single temperature cycle. The debonding warning threshold is obtained by the following method: accelerated aging test is carried out on composite plate samples with different initial debonding sizes, and the average debonding area growth rate after each cycle and the average debonding area growth rate when the whole is finally debonded are recorded. 30% of the average debonding area growth rate that leads to the whole debonding is taken as the debonding warning threshold, and it is determined after being verified by no less than 3 batches of repeated experiments.

[0052] Among them, fracture energy per unit area The following methods were used to obtain the fracture toughness of the interface between the calcium silicate panel and the gradient polyurethane composite adhesive using double cantilever beam specimens. Load-displacement curves were measured on no fewer than 20 specimens, and the fracture energy per unit area of ​​each specimen was calculated using the linear elastic fracture mechanics formula. The average value was taken as the calibration parameter for the variational level set interface defect evolution tracking based on the interface thermo-mechanical coupling energy functional, and the fracture energy per unit area was calculated at three temperatures: -20℃, 23℃, and 80℃. Each value is calibrated separately, and interpolation is performed based on the current temperature field in the variational level set interface defect evolution tracking based on the interface thermo-mechanical coupling energy functional.

[0053] The specific implementation method of step S01 is as follows. When performing fixed-thickness sanding on EPS boards, a wide-width sander is used to simultaneously grind both the upper and lower surfaces of the EPS board. The grinding amount is determined based on the measured value of the thickness deviation of the incoming material. After grinding, a laser rangefinder is used to scan the board surface point by point to confirm that the surface flatness deviation is controlled within ±0.3mm. The ±0.3mm threshold is derived from statistical experiments: using EPS boards with different flatness deviations as variables, the uniformity of adhesive layer thickness and interfacial peel strength are measured respectively. The upper limit of the flatness deviation corresponding to the standard deviation of interfacial peel strength being lower than 10% of the mean is set as this threshold, and it is verified by at least 3 batches of repeated experiments. Subsequently, the surface of the calcium silicate panel is mechanically sanded and activated to remove the surface contamination layer and increase the surface roughness, thereby improving the wetting and spreading ability of the subsequent adhesive. After the activation treatment is completed, -aminopropyltriethoxysilane or Glycidyl etheroxypropyltrimethoxysilane was uniformly sprayed in aqueous solution onto the surface of calcium silicate panels and EPS boards, with a coating amount of 20-40 g / L. The hydrolysis time is 10-20 minutes. After hydrolysis, silane molecules form an interfacial transition layer mainly composed of covalent bonds at the interface, which is used to improve the chemical bonding strength between the gradient polyurethane composite adhesive and the substrates on both sides.

[0054] The specific implementation of step S02 is as follows. The follow-up scraper system consists of a scraper lifting mechanism driven by a servo motor, a displacement sensor, and a closed-loop controller. The displacement sensor collects the surface height of the EPS board in real time with a response frequency of not less than 50Hz. The closed-loop controller drives the scraper to follow the undulations of the EPS board surface in the vertical direction based on the height deviation signal, with a displacement accuracy of ±0.05mm, thereby controlling the adhesive layer thickness within the range of 100-500μm. The gradient polyurethane composite adhesive is prepared by mixing flexible silicone-modified polyurethane and rigid cement-based interface agent in a continuous gradient ratio. The adhesive layer density on the side closest to the EPS board is 60. The adhesive layer density on the side closest to the calcium silicate panel is 300. The density transitions continuously along the thickness of the adhesive layer, and the elongation at break is greater than 300%. The purpose of the gradient design is that the coefficient of thermal expansion of the EPS board is approximately... At / ℃, the coefficient of thermal expansion of calcium silicate panels is approximately At a temperature of / ℃, alternating shear stress is generated at the interface under temperature cycling. The gradient modulus concentrates and transfers strain to the flexible region for dissipation, inhibiting the propagation of fatigue cracks at the interface. The time-series data of the adhesive application amount is collected by a displacement sensor at a sampling interval of 30 seconds and then input into the interface thermo-mechanical coupling adhesive layer quality prediction model.

[0055] The specific implementation of step S03 is as follows: The rolled EPS board and calcium silicate panel are aligned and stacked in a pressing fixture, and a pressing pressure of 0.05–0.20 MPa is applied. The pressing temperature is controlled between 15–45°C. The interface thermo-mechanical coupling adhesive layer quality prediction model (THCB-Net) is a time-series recursive residual prediction model. The input layer receives four types of time-series features, with a feature dimension of 4 at each time step. It is connected to a dual-layer gated loop unit encoder, with a hidden state dimension of 256 at each layer. Residual jump connections are set between the dual-layer gated loop units and between the input layer and the output layer, for a total of three levels of residual jump connections. A physical constraint layer is inserted after each time step of the gated loop unit. The physical constraint layer explicitly calculates the current degree of cure based on the Arrhenius equation. ,by The forget gate weight matrix is ​​modulated to form a memory update mechanism driven by the solidification kinetic equation. Three fully connected branches are connected in parallel to the network output head, outputting the predicted interfacial peel strength, predicted adhesive shrinkage rate, and predicted thermal cycling lifetime, respectively. The comprehensive interface quality evaluation function is based on... , , The comprehensive evaluation value is calculated using the three normalized values. ,in accordance with Learning rate adjustment for the current interval: Learning rate remains constant Multiply the learning rate by 0.8 Multiply the learning rate by 0.5 and increase the weight coefficient of the physical consistency penalty term. The learning rate is reset to its initial value and weight rollback is triggered. The pressing time is dynamically determined by the current degree of curing value output by the model in real time. During the pressing process, the pressing pressure curve, curing temperature-time curve, and rotational rheometer viscosity curve are collected in real time and continuously input into the model.

[0056] The specific implementation of step S04 is as follows: After lamination, variational level set interface defect evolution tracking based on interface thermo-mechanical coupling energy functional is performed on the interface region of the composite plate. The geometric boundary of the debonding region is modeled as an implicit level set function. Deadhesion evolution energy functional It consists of three terms: elastic strain energy, interfacial fracture energy, and a penalty regularization term. The elastic strain energy is calculated based on Eshelby inclusion theory to determine the equivalent stress concentration in the debonding zone. The interfacial fracture energy is Griffith type, and the fracture energy per unit area is... The fracture toughness of the double cantilever beam specimens was calibrated at -20℃, 23℃, and 80℃, with values ​​interpolated based on the current temperature field during tracking. At each time step, the thermoelastic coupled finite element equations were solved sequentially to obtain the total stress, and the energy functional was calculated. The Gâteaux derivative is updated along the negative gradient direction. The algorithm performs narrowband reinitialization and extracts the zero-level set as the current debonding boundary. It outputs the debonding area growth rate and compares it with a debonding warning threshold, which is set to 30% of the average debonding area growth rate leading to overall debonding. When the debonding area growth rate exceeds the debonding warning threshold, a glue replenishment repair process is triggered. After repair, S04 is re-executed.

[0057] The specific implementation method of step S05 is as follows. The composite board that passed the S04 test is subjected to temperature cycling aging verification, with a cycling range of -20 to 80℃ and a cycle count of no less than 200 times. After aging, the interfacial peel strength is measured according to the adhesive strength test standard. The pass / fail criterion is an interfacial peel strength of not less than 1.5MPa. Unqualified boards are returned to the S04 process or scrapped.

[0058] The specific implementation of step S06 is as follows. The time-series data of adhesive application amount, pressing pressure curve, curing temperature-time curve, rotational rheometer viscosity curve, measured values ​​of interfacial peel strength, measured values ​​of adhesive layer shrinkage, and measured values ​​of thermal cycle life corresponding to the qualified composite board are fed back to the training set of the interfacial thermo-mechanical coupling adhesive layer quality prediction model. Outliers in the newly added data are removed (based on box curves). Figure 3 The model is then updated using the interquartile range criterion and linear interpolation, followed by an Adam optimizer that starts with the current weights to continuously accumulate and correct the model's predictive capabilities based on production data.

[0059] It should be noted that the key technologies of this invention include: a physical constraint structure for curing kinetics embedded in a neural network gating structure, enabling the model to still predict behavior based on the Arrhenius equation even when process parameters deviate from historical distributions, thus avoiding prediction distortion of purely data-driven models under small sample conditions; a continuous density gradient design for the gradient polyurethane composite adhesive, which disperses and dissipates the interfacial alternating shear stress caused by the difference in thermal expansion coefficients in the flexible region through continuous modulus variation, suppressing fatigue crack propagation to the brittle interface; and a variational level set method that implicitly embeds the debonding boundary into a function, allowing topological changes such as splitting and merging of the debonding region to be realized naturally, with the Gâteaux derivative providing the boundary update direction with the fastest energy decrease, achieving continuous tracking in the subcritical stage. These three key technologies work synergistically: the gradient adhesive layer suppresses initial debonding during the manufacturing stage, the physical constraint neural network ensures curing quality during the pressing stage, and the variational level set tracking achieves subcritical early warning of defects during service, forming a complete technology chain from material design and process control to online monitoring, resulting in significantly better overall interface reliability than traditional discrete control methods.

[0060] It should be noted that during the temperature cycling aging test, after the composite board undergoes a large number of cycles, the interface debonding area may evolve in the form of multiple dispersed small areas expanding simultaneously, rather than linearly propagating from a single crack source. The reason for this technical problem is the significant difference in the coefficients of thermal expansion between the EPS board and the calcium silicate panel. At / ℃, under conditions of multiple coexisting defects, the stress fields around each debonding region superimpose, causing a sudden acceleration in the strain energy release rate in some areas. This leads to bifurcation and merging of the debonding propagation path, making it difficult for traditional methods to track the debonding geometry of multiple simultaneous topological changes within the same time step. A common solution to this problem is to use a finite element crack propagation model (such as the cohesive element method) and re-mesh the crack front at each topological change to adapt to the new geometry. However, mesh re-meshing is computationally expensive, and when multiple crack fronts split or merge simultaneously, the mesh re-meshing algorithm struggles to guarantee topological consistency, easily leading to numerical divergence. In practical engineering, multi-crack problems often have to be simplified to single-crack problems, losing crucial information about the interactions of multiple debonding regions. This invention effectively solves this problem. The variational level set method implicitly embeds the debonding boundary into the level set function. The function is defined across the entire computational domain, and topological changes such as the splitting and merging of debonded regions are naturally achieved through positive and negative changes in the function value, without any mesh re-division operation. Each time step only requires solving the thermoelastic coupling finite element equations, calculating the Gâteaux derivative, and updating the function value on a fixed mesh. The simultaneous evolution of multiple debonded regions is handled self-consistently within the same computational framework. The narrowband re-initialization technique concentrates numerical computation in a narrowband region near the zero level set, controlling the computational load while ensuring accuracy. Therefore, regardless of the number and topological complexity of the debonded regions, the tracking algorithm of this invention can provide an accurate geometric evolution description within a unified framework, providing a reliable quantitative basis for precise positioning of adhesive repair.

[0061] Specifically, the principle of this invention is:

[0062] The present invention is able to solve the above-mentioned technical problems, and the fundamental reason lies in the synergistic effect of the two core mechanisms.

[0063] The first core mechanism is to embed the physical laws of curing kinetics into the neural network gating structure. The curing process of the adhesive layer is essentially a chemical-mechanical coupled time-series process driven by multiple parameters such as temperature, pressure, and viscosity, and its curing rate satisfies the temperature dependence described by the Arrhenius equation. Traditional data-driven models, when the training samples are limited or the process parameters deviate from historical distributions, produce unreasonable predictions due to a lack of physical constraints. This invention inserts a physical constraint layer after each time step of the gating loop unit, explicitly calculating the current degree of curing according to the Arrhenius equation, and injecting it as a bypass signal into the forget gate weight matrix of the next time step, thereby transforming the laws of chemical reaction kinetics into an intrinsic constraint for network memory updates. This mechanism ensures that the model's predictive behavior is explicitly constrained by the curing kinetic equation when process parameters fluctuate, rather than relying solely on statistical fitting, thus maintaining reasonable guidance for parameters such as pressing time and curing temperature under small sample conditions. The three-level residual skip connections further prevent gradient vanishing in multi-layer networks, preserving the direct transmission path of process characteristics from each layer, allowing process information at different time scales to effectively participate in degree of curing prediction.

[0064] The second core mechanism is the variational level set interface defect evolution tracking based on energy functionals. The propagation of debonding defects is essentially a mechanical process driven by the release of elastic strain energy, propelling the crack front. Its propagation path dynamically evolves with temperature load changes and may undergo topological changes such as splitting and merging. Traditional ultrasonic testing struggles to capture the continuous geometric evolution of debonding propagation because it is essentially a static sampling at discrete moments. This invention implicitly embeds the debonding boundary into the level set function. By solving the thermoelastic coupling finite element equations to obtain the full-field stress, the Gâteaux derivative of the energy functional with respect to the level set function is calculated. The level set function is then updated along the negative gradient direction, ensuring that the evolution direction of the debonding boundary physically corresponds to the path with the fastest release of elastic strain energy. Topological changes such as splitting and merging in the debonding region are naturally realized during function evolution, without the need for mesh re-division. Narrow-band reinitialization ensures the numerical stability of the level set function, and zero-level set extraction provides a precise geometric description of the current debonding boundary. By outputting the debonding area growth rate and comparing it with the warning threshold, the timing of adhesive repair is shifted to the subcritical stage, fundamentally changing the logic of passive defect detection in traditional methods.

[0065] The logic behind the combined effect of the two mechanisms is that the former ensures that the curing quality during the pressing stage is under control, while the latter ensures that the interface degradation during the service stage is traceable, forming a complete closed loop from manufacturing to testing.

[0066] The following provides a specific embodiment 1 of the present invention, and the specific implementation of each step in this embodiment 1 is described in detail below.

[0067] The specific implementation of step S01 is as follows: A wide-width sander is used to simultaneously grind both sides of the EPS board. The grinding amount is determined based on the thickness deviation of the incoming material, and the surface flatness deviation after grinding is controlled within ±0.3mm. This threshold is determined as follows: Using EPS boards with different surface flatness deviations as variables, the uniformity of the adhesive layer thickness and the interfacial peel strength are measured respectively. The relationship between the standard deviation of the interfacial peel strength and the surface flatness deviation is statistically analyzed. The upper limit of the deviation corresponding to the standard deviation of the interfacial peel strength being less than 10% of the mean is set at ±0.3mm. This is confirmed after verification through at least three batches of repeated experiments. Subsequently, the surface of the calcium silicate panel is mechanically polished and activated, and then... -aminopropyltriethoxysilane or Glycidyl etheroxypropyltrimethoxysilane was uniformly sprayed in aqueous solution onto the surface of calcium silicate panels and EPS boards, with a coating amount of 20-40 g / L. The static hydrolysis time is 10-20 minutes to form an interface transition layer mainly composed of covalent bonds, which is used to improve the chemical bonding strength between the gradient polyurethane composite adhesive and the substrates on both sides.

[0068] The specific implementation of step S02 is as follows: The follow-up scraper system consists of a servo motor-driven scraper lifting mechanism, a displacement sensor, and a closed-loop controller. The displacement sensor collects the surface height of the EPS board in real time. The closed-loop controller drives the scraper to follow the undulations of the EPS board surface in the vertical direction based on the height deviation signal. The displacement accuracy is ±0.05mm, the response frequency is not less than 50Hz, and the adhesive layer thickness is controlled between 100 and 500 mm. Within the specified range. The gradient polyurethane composite adhesive used is prepared by mixing flexible silicone-modified polyurethane and rigid cement-based interface agent in a continuously gradient ratio. The adhesive layer density on the side closest to the EPS board is 60. The adhesive layer density on the side closest to the calcium silicate panel is 300. The density transitions continuously along the thickness of the adhesive layer, and the elongation at break is greater than 300%. This gradient design is based on the fact that the coefficient of thermal expansion of EPS boards is approximately... The coefficient of thermal expansion of calcium silicate panels is approximately The difference between the two is as high as Under temperature cycling, alternating shear stress is generated at the interface. The gradient polyurethane composite adhesive concentrates and transfers strain to the flexible region for dissipation through continuous modulus changes, suppressing the propagation of interface fatigue cracks. The density gradient range and elongation at break threshold are determined as follows: the formulation density starts from 50... Up to 350 A series of polyurethane foam samples were tested under temperature cycling conditions (-20~80℃, 200 cycles) to determine the interfacial peel strength retention rate of each sample. The density gradient range was determined by the density start and end values ​​corresponding to the highest interfacial peel strength retention rate. The elongation at break threshold was determined by combining tensile tests and fatigue crack propagation rate calculations of the same batch of samples, and was verified after at least three batches of repeated experiments. The adhesive application time series data was collected by the displacement sensor of the follow-up scraper system and then input into the interfacial thermo-mechanical coupling adhesive layer quality prediction model.

[0069] The specific implementation of step S03 is as follows: the EPS board that has undergone roller coating and the calcium silicate panel are pressure-bonded together in a pressing fixture, and the pressing pressure is controlled between 0.05 and 0.20. The pressing temperature is controlled between 15 and 45°C, and the pressing time is dynamically determined by the current degree of cure value output by the interface thermo-mechanical coupling adhesive layer quality prediction model. During the pressing process, the pressing pressure curve, curing temperature-time curve, and rotational rheometer viscosity curve are collected in real time and input into the interface thermo-mechanical coupling adhesive layer quality prediction model. The input layer of the interface thermo-mechanical coupling adhesive layer quality prediction model receives four types of time-series features: adhesive application amount time-series data with a sampling interval of 30 seconds, pressing pressure curve, curing temperature-time curve, and rotational rheometer viscosity curve. Each time step has a feature dimension of 4, forming the input vector. ,in For discrete time step index, dimensionless. In a two-layer gated cyclic encoder, the first... layer( The hidden state update process of ) is as follows, forget gate Input gate Candidate hidden states and hidden state All are dimensionless vectors, namely:

[0070] ;

[0071] ;

[0072] ;

[0073] .

[0074] In the formula, the input layer performs zero-mean, unit-variance normalization on each feature before inputting it into the network, so that... A dimensionless vector , , For the first The weight matrix of each gate in the layer is dimensionless. , , The corresponding bias vector is dimensionless. For the first Layer input dimension, layer 1 , second floor Logistic activation function Hyperbolic tangent activation function For element-wise multiplication For the first Layer time step The input, the first layer input is The input of the second layer is The three-stage residual jump connection will respectively... Add directly to ,Will Add directly to ,Will The residuals are added directly to the output input, and all residual additions are element-wise sums. Dimensional mismatches are aligned via linear projection. The physical constraint layer explicitly calculates the current degree of solidification based on the Arrhenius equation. The formula is expressed as follows:

[0075] ;

[0076] In the formula, For time step The degree of curing is dimensionless and ranges from 0 to 1. The sampling interval is 30 seconds. , The reaction rate constant is expressed in units of 1000 m / s. After being calibrated by differential scanning calorimetry experiments, it was fixed as a constant. , The corresponding activation energy is expressed in units of 1. After being calibrated by differential scanning calorimetry experiments, it was fixed as a constant. is the molar gas constant, with a value of 8.314. For time step The curing temperature, in units of Read directly from the curing temperature-time curve , , The reaction order parameter is dimensionless and determined through differential scanning calorimetry (DSC) experimental fitting. The physical constraint layer will... As a bypass signal injected into the hidden state of the next time step, the forget gate weight matrix is ​​modulated in the following manner:

[0077] ;

[0078] In the formula, The modulated forget gate weight matrix is ​​dimensionless. The modulation coefficient is dimensionless and has an empirical value of 0.1. It is used to control the intensity of the effect of curing degree on memory update. Dimensionless, therefore Dimensionless, and The units are consistent. The network output header has three fully connected branches in parallel, each outputting the predicted interface peeling strength. Predicting the shrinkage rate of the adhesive layer and predicting thermal cycle life The output layer formula is:

[0079] ;

[0080] In the formula, The output weight matrix is ​​dimensionless. The output bias vector is dimensionless. This is the index of the final discrete-time step of the sequence, dimensionless, and its value is the maximum number of steps in the current input sequence; each component of the output vector is processed... , , After normalization, it becomes a dimensionless quantity, consistent with the dimensions of the product of the dimensionless matrices on the right. This is a standard reference value for predicting interfacial peel strength, in units of... Dimensionless. This serves as a standard reference value for predicting adhesive layer shrinkage. The standard reference values ​​for predicting thermal cycling lifetime are given in cycles; all three are taken as the 90th percentile of the corresponding measured values ​​in the training dataset and determined after verification through at least three batches of repeated experiments. The loss function is a weighted sum of the data error term and the physical consistency penalty term, expressed in the following formula:

[0081] ;

[0082] In the formula, Number of samples in a batch For the first The predicted interfacial peel strength for each sample, in units of To correspond to the measured interfacial peel strength, the unit is... For predicting the curing degree curve of the network, dimensionless The curing degree curve is calculated using the Kamal kinetic equation and is dimensionless. This represents the total time for lamination and curing, in units of... Used to normalize the integral difference in degree of cure. The weight coefficient for the physical consistency penalty term is dimensionless, initially set to 0.1, and linearly increases to 0.5 with each training epoch; both terms are dimensionless and have consistent dimensions. The total loss after adding L1 sparsity regularization constraints is:

[0083] ;

[0084] In the formula, The total loss after adding sparsity regularization is dimensionless. Let L1 be the regularization coefficient, dimensionless, with an empirical value of [value missing]. For the first Layer weight matrix Let L1 be the L1 norm of the matrix, which is the sum of the absolute values ​​of all its elements and is dimensionless. The truncation step size for backpropagation is 32 steps, and the gradient flow is blocked at the beginning time step of each truncation window, as expressed in the following formula:

[0085] ;

[0086] In the formula, Index of the start time step for each truncated window, dimensionless. This indicates that gradient flow is blocked during backpropagation, and backpropagation is performed only within a window of length 32 steps. The interface quality comprehensive evaluation function is used to dynamically adjust the learning rate, and its formula is as follows:

[0087] ;

[0088] In the formula, This is a dimensionless comprehensive evaluation value for interface quality. The predicted interface peeling intensity at the current time step, in units of The predicted shrinkage rate of the adhesive layer at the current time step is dimensionless. The predicted thermal cycle life at the current time step, in cycles. , , The weighting coefficients are dimensionless and were determined through statistical analysis of the model prediction errors under different weight combinations using orthogonal experimental methods. All terms are normalized to corresponding standard reference values ​​to be dimensionless, with consistent dimensions between the numerator and denominator. When the learning rate remains constant; when When, the learning rate is multiplied by 0.8; when At that time, the learning rate is multiplied by 0.5, and simultaneously increased. ;when At that time, the learning rate is reset to This triggers a rollback of the model weights to the previous validation loss optimum. Training uses the Adam optimizer with an initial learning rate of... The batch size is 32. The loss is calculated on the validation set every 5 training rounds. If the loss does not decrease for 10 consecutive validation rounds, the learning rate is multiplied by 0.5. The total number of training rounds is no less than 200 rounds. The model weights that have the lowest root mean square error in the predicted interface peeling strength on the test set are used as the final weights.

[0089] The specific implementation of step S04 is as follows: After lamination, variational level set interface defect evolution tracking based on interface thermo-mechanical coupling energy functional is performed on the interface region of the composite plate. The geometric boundary of the interface debonding region is modeled as an implicit level set function. ,in Spatial coordinates, in units of This is the signed distance function, with units of . Positive values ​​indicate that the point is outside the decoupling region, while negative values ​​indicate that the point is inside the decoupling region. The energy functional of decoupling evolution. It consists of three terms: elastic strain energy, interfacial fracture energy, and penalty regularization term, and the formula is expressed as follows:

[0090] ;

[0091] In the formula, The computational domain for the composite plate is expressed in units of 1. These are the stress tensor components, in units of . For the strain tensor components, dimensionless , These are the tensor component subscripts, with values ​​of 1, 2, and 3, corresponding to the directions of the three-dimensional spatial coordinates, respectively. level set function The corresponding debonding boundary, in units of The fracture energy per unit area is expressed in units of 1000 kJ / m². The fracture toughness of the specimens was calibrated using a double cantilever beam test at three temperatures: -20℃, 23℃, and 80℃. The values ​​were interpolated based on the current temperature field during calculation. The coefficient for the penalty regularization term, in units of The empirical value corresponds to making the penalty term comparable in magnitude to the fracture energy term, used to constrain the sign distance property of the level set function; the first term Units are The second item Units are The third item middle Units are , Units are Therefore, the unit of the third item is The three quantities have consistent dimensions. The algorithm flow for each time step is as follows: First, solve the thermoelastic coupling finite element equation to obtain the full-field stress. The thermoelastic coupling finite element equation is:

[0092] ;

[0093] ;

[0094] In the formula, This is the stiffness matrix, in units of 1. This is the thermo-mechanical coupling matrix, in units of... This is the force-thermal coupling matrix, in units of This is the heat conduction matrix, in units of... Here is the nodal displacement vector, in units of This is the temperature vector at the finite element nodes, in units of... Obtained from temperature field calculations This is a mechanical load vector, in units of... This is the thermal load vector, in units of The dimensions of each matrix are determined by the number of nodes in the finite element mesh. The energy functional is then calculated. Regarding level set functions The Gâteaux derivative (i.e., variational derivative) is used to update the level set function along the negative gradient direction, as expressed in the following formula:

[0095] ;

[0096] In the formula, The virtual time of the level set evolution, in units of Units are The Dirac function is given by units of 1000 ppm. Concentrated at the horizontal set zero interface The gradient magnitude of the level set function, in units of (The ideal value of the signed distance function after processing is 1, a dimensionless approximation) The curvature of the interface is expressed in units of 1. ,Depend on Calculate; first term Units are The unit of the corresponding level set update after multiplying by the virtual time step is... ,and Units consistent; Item 3 Units are The dimensions correspond to those of the first two items. After narrowband reinitialization, the zero-level set is extracted. As the current debonding boundary, calculate the debonding area growth rate. The formula is expressed as follows:

[0097] ;

[0098] In the formula, The debonding area at the current time step, in units of This represents the debonding area in the previous cycle, in units of... The time step for a single temperature cycle, in units of This is a dimensionless ratio. With de-adhesion warning threshold Comparison, The following method was used to determine the debonding area growth rate: accelerated aging tests were conducted on composite plate samples with different initial debonding sizes, and the average debonding area growth rate at the time of final overall peeling was recorded after each cycle. ,Pick It was determined after verification through no fewer than three batches of repeated experiments. When the repair process is completed, step S04 is re-executed.

[0099] The specific implementation of step S05 is as follows: The composite plate inspected in step S04 undergoes temperature cycling aging verification. The temperature cycling range is -20 to 80°C, and the number of cycles is no less than 200. After aging, the interfacial peel strength is measured, with a requirement that the interfacial peel strength is not less than 1.5. This is the standard for determining whether a product is qualified.

[0100] The specific implementation of step S06 is as follows: The time series data of adhesive application amount, pressing pressure curve, curing temperature-time curve, rotational rheometer viscosity curve, measured value of interfacial peel strength, measured value of adhesive layer shrinkage rate and measured value of thermal cycle life corresponding to the qualified composite board are transmitted back to the training set of the interfacial thermo-mechanical coupling adhesive layer quality prediction model. The model is iteratively updated once according to the training process described in step S03, so as to realize the continuous adaptive optimization of the model to the new production conditions.

[0101] To better understand and implement this invention, the following is a specific application scenario of the invention, Example 2: To verify the effect of the invention, technicians set up a composite board interface bonding test environment, using a batch of calcium silicate panels and EPS boards as test objects, and performed interface bonding and quality verification according to the entire process of the method of the invention.

[0102] The measured thickness deviation of the incoming EPS board ranged from -0.8 to +0.6 mm. After simultaneous grinding of both sides using a wide-width sander, a laser rangefinder scanned point by point to confirm that the surface flatness deviation had narrowed to ±0.25 mm, meeting the ±0.3 mm threshold requirement. Subsequently, the calcium silicate panel surface was mechanically polished and activated, and then sprayed with... -Aqueous aminopropyltriethoxysilane, spraying amount 30 The EPS board surface is treated simultaneously with static hydrolysis for 15 minutes.

[0103] A follow-up scraper system acquires the surface height of the EPS board in real time with a response frequency of 55Hz. A closed-loop controller drives the scraper to follow the surface undulations, and the thickness of the gradient polyurethane composite adhesive roller coating is controlled between 180 and 220 μm. The adhesive application time-series data is input into the interface thermo-mechanical coupling adhesive layer quality prediction model at 30-second sampling intervals. For example... Figure 2As shown in the figure, the height deviation distribution and corresponding adhesive layer thickness compensation curve of the EPS board surface collected by the follow-up scraper system in this test can be seen. The figure shows the real-time following relationship between the height deviation and the scraper displacement compensation.

[0104] The rolled-coated EPS boards and calcium silicate panels are stacked in a laminating fixture, with a lamination pressure of 0.12 MPa and a lamination temperature of 25°C. The interface thermo-mechanical coupling adhesive layer quality prediction model receives four types of temporal characteristics in real time, and the physical constraint layer calculates the current degree of cure based on calibrated Kamal kinetic parameters. The model outputs a signal indicating that the curing degree has reached the target value 38 minutes after the start of pressing. The pressing fixture then releases pressure, and the pressing time is dynamically fixed at 38 minutes. Compared to the traditional method using a fixed pressing time of 45 minutes for the same batch, the interface thermo-mechanical coupling adhesive layer quality prediction model saves unnecessary waiting time and avoids curing defects caused by insufficient pressing. The interface quality comprehensive evaluation function outputs a comprehensive evaluation value at the end of pressing. The learning rate remains unchanged, and the model state is stable.

[0105] After lamination, variational level set interface defect evolution tracking based on interface thermo-mechanical coupling energy functionals is performed on the interface region of the composite plate. Fracture energy per unit area... Double cantilever beam samples were calibrated at three temperatures: -20℃, 23℃, and 80℃. Twenty samples were measured at each temperature, and the average value was used as the calibration parameter. The tracking algorithm updated the level set function step-by-step on a fixed grid, and extracted the zero level set to obtain the geometry of the debonding boundary. Initial detection revealed three scattered debonding areas at the interface, with a debonding area growth rate of 0.8% / cycle, which is lower than the debonding warning threshold (the calibration value for this batch is 2.1% / cycle). No re-adhesion repair was required, and the composite board passed the S04 inspection. The main process parameters and tracking results are shown in Table 1.

[0106] Table 1 Summary of Interface Adhesion Process Parameters and Tracking Results

[0107]

[0108] After passing the S04 test, the composite board underwent temperature cycling aging verification, with a cycling range of -20 to 80℃ and 200 cycles. After aging, the interfacial peel strength was measured, and the peel strength distribution at each measuring point is shown in Table 2.

[0109] Table 2 Distribution of interfacial peel strength after temperature cycling aging

[0110]

[0111] It can be seen from Table 2 that the interfacial peel strength of each measuring point is not lower than the qualified threshold of 1.5 MPa, so the composite board is judged as qualified. As Figure 3 shows, it is the comparison between the degree of curing output by the interface thermal-mechanical coupling adhesive layer quality prediction model in this test evolution curve over time and the theoretical curve of the Kamal kinetic equation. It can be seen from the figure that the model prediction curve is highly consistent with the theoretical curve throughout the pressing stage, which verifies that the physical constraint layer effectively embeds the curing kinetic law. The full set of process data of the qualified composite board is transmitted back to the model training set, completing one model iterative update, and the sample size of the training set increases to 503 groups. As Figure 4 shows, it is the tracking result of the variational level set tracking algorithm on the geometric boundary evolution of the interfacial debonding region during the temperature cycle aging process in this test. The figure shows the area growth trend of 3 small initial debonding regions in 200 cycles and the change curve of the debonding area growth rate with the number of cycles. It can be seen that the debonding area growth rate is always lower than the early warning threshold, and the debonding expansion is effectively suppressed.

[0112] Compared with traditional methods, the present invention achieves the following technical progress. The traditional fixed empirical pressing time scheme cannot perceive the fluctuations of rubber viscosity and curing temperature between batches, and the deviation of curing kinetic state can only be found through peel strength testing after the event. In contrast, the present invention explicitly embeds the Arrhenius equation into the memory update mechanism of the gated recurrent unit, so that the dynamic determination of pressing time is supported by the constraints of physical equations, which fundamentally changes the logic of experience dependence. Traditional ultrasonic offline inspection can only provide a static judgment that debonding has been formed, while variational level set tracking converts the subcritical dynamic process of debonding expansion into quantifiable geometric information through function evolution driven by energy functional, which moves the repair decision forward to the stage where the defect has not reached the critical state and eliminates the root cause of delayed discovery from the technical mechanism. The synergistic effect of the two technologies changes the interface quality control from post-inspection to an active control mode of process prediction and subcritical intervention.

[0113] It should be noted that the detailed explanation of the variables involved in the present invention is shown in Tables 3 and 4.

[0114] Table 3 Variable Explanation Table (Part I)

[0115]

[0116] Table 4 Variable Explanation Table (Part II)

[0117]

[0118] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for bonding a calcium silicate panel to an EPS core material, characterized in that, Includes the following steps: The EPS board is sanded to a fixed thickness, and the calcium silicate panel surface is mechanically polished and activated. After activation, a silane coupling agent base coating is applied to the calcium silicate panel surface and the EPS board surface respectively. A follow-up doctor blade system is used to roll-coat gradient polyurethane composite adhesive. The follow-up doctor blade system collects the height deviation signal of the EPS board surface in real time and compensates for the surface undulation of the EPS board. The time series data of adhesive application amount is collected by the displacement sensor of the follow-up doctor blade system and then input into the interface thermo-mechanical coupling adhesive layer quality prediction model. The EPS board that has been rolled and coated is pressed and laminated with the calcium silicate panel in a pressing fixture. The pressing time is dynamically determined by the current degree of curing value output by the interface thermo-mechanical coupling adhesive layer quality prediction model. The pressing pressure curve, curing temperature-time curve and rotational rheometer viscosity curve during the pressing process are collected in real time and input into the interface thermo-mechanical coupling adhesive layer quality prediction model. After lamination, the interface area of ​​the composite board is tracked by variational level set interface defect evolution based on interface thermo-mechanical coupling energy functional. The tracking results output the debonding area growth rate. The debonding area growth rate is compared with the debonding warning threshold. When the debonding area growth rate exceeds the debonding warning threshold, the glue repair process is triggered. The composite board after inspection was subjected to temperature cycling aging verification. After aging, the interface peel strength was measured. The standard for acceptance was that the interface peel strength was not lower than the qualified threshold. The time-series data of adhesive application amount, pressing pressure curve, curing temperature-time curve, rotational rheometer viscosity curve, measured value of interfacial peel strength, measured value of adhesive shrinkage rate, and measured value of thermal cycle life corresponding to the qualified composite board are fed back to the training set of the interfacial thermo-mechanical coupling adhesive layer quality prediction model to complete one iterative update of the interfacial thermo-mechanical coupling adhesive layer quality prediction model.

2. The interface bonding method according to claim 1, characterized in that, The specified thickness sanding process involves using a wide-width sander to simultaneously grind both the top and bottom surfaces of the EPS board. The amount of grinding is determined based on the thickness deviation of the incoming EPS board material. After grinding, the surface flatness deviation of the EPS board is controlled within the flatness deviation threshold.

3. The interface bonding method according to claim 2, characterized in that, The silane coupling agent undercoat refers to the coating layer that... -aminopropyltriethoxysilane or - Glycidyl etheroxypropyltrimethoxysilane is uniformly sprayed in aqueous solution onto the surface of calcium silicate panel and EPS board. After static hydrolysis, it forms an interface transition layer mainly composed of covalent bonds.

4. The interface bonding method according to claim 3, characterized in that, The aforementioned follow-up scraper system refers to an adhesive application device consisting of a servo motor-driven scraper lifting mechanism, a displacement sensor, and a closed-loop controller. The closed-loop controller drives the scraper to follow the undulations of the EPS board surface in the vertical direction based on the height deviation signal.

5. The interface bonding method according to claim 4, characterized in that, The gradient polyurethane composite adhesive refers to a composite adhesive prepared by mixing flexible silicone-modified polyurethane and rigid cement-based interface agent in a continuous gradient ratio. The adhesive layer density on the side closer to the EPS board is lower than that on the side closer to the calcium silicate panel, and the density transitions continuously in the thickness direction of the adhesive layer.

6. The interface bonding method according to claim 5, characterized in that, The interface thermo-mechanical coupling adhesive layer quality prediction model refers to a time-series recursive residual prediction model based on the mapping relationship between process parameters and interface performance. The input layer receives adhesive application time-series data, pressing pressure curve, curing temperature-time curve and rotational rheometer viscosity curve, and is connected to a dual-layer gated loop unit encoder. A residual jump connection is set between the dual-layer gated loop units.

7. The interface bonding method according to claim 6, characterized in that, The interface thermo-mechanical coupling adhesive layer quality prediction model inserts a physical constraint layer after each gated loop unit time step. The physical constraint layer explicitly calculates the current degree of curing according to the Arrhenius equation and injects the current degree of curing into the hidden state of the next time step by modulating the forget gate weight matrix.

8. The interface bonding method according to claim 7, characterized in that, The interface thermo-mechanical coupling adhesive layer quality prediction model has an interface quality comprehensive evaluation function. The interface quality comprehensive evaluation function calculates the interface quality comprehensive evaluation value based on the normalized value of the predicted interface peel strength, the normalized value of the predicted adhesive layer shrinkage rate, and the normalized value of the predicted thermal cycle life. The learning rate parameter is adjusted according to the interval in which the interface quality comprehensive evaluation value is located.

9. The interface bonding method according to claim 8, characterized in that, The variational level set interface defect evolution tracking based on the interface thermo-mechanical coupling energy functional refers to modeling the geometric boundary of the interface debonding region as an implicit level set function. The energy functional of the debonding evolution consists of three terms: elastic strain energy, interface fracture energy, and penalty regularization term. At each time step, the level set function is updated along the negative gradient direction of the energy functional with respect to the Gâteaux derivative of the level set function.

10. The interface bonding method according to claim 9, characterized in that, The debonding warning threshold refers to the upper limit of the debonding area growth rate after a single temperature cycle. The debonding warning ratio that leads to the average debonding area growth rate of the overall peeling is taken as the debonding warning threshold.

Images