HTCC ceramic defect automatic identification method based on industrial model assistance

Abstract

The invention relates to the technical field of intelligent manufacturing, in particular to an HTCC ceramic defect automatic identification method based on industrial model assistance, which comprises the following steps: step 1, collecting sintering temperature gradient, lamination pressure and green body water content process parameters in the HTCC production process in real time; step 2, performing non-contact dielectric property scanning on the sintered HTCC substrate on a production line conveyor belt; 3, based on the coordinates of the dielectric abnormal region, controlling an annular polarization light source to inhibit the incident angle reflected by the metallization layer from irradiating the target region; 4, inputting the dielectric anomaly frequency band feature vector and the defect image texture feature in the step 3 into a cross-modal attention network to generate a fusion feature map; and 5, establishing an association mapping library of the defect type and the process deviation type. Through accurate real-time monitoring, intelligent defect identification and classification, automatic process adjustment and a continuously optimized feedback mechanism, the efficiency, quality and intelligent level of the ceramic production process are greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent manufacturing, and particularly to an automatic HTCC ceramic defect recognition method assisted by an industrial model. Background Art

[0002] High-temperature co-fired ceramics (HTCC) are widely used in high-end fields such as aerospace high-frequency devices and high-power electronic packaging due to their excellent high-temperature resistance, high mechanical strength, and multi-layer wiring capabilities. However, the HTCC production process is complex, involving key processes such as tape casting, lamination, and co-firing, and is prone to defects such as internal microcracks, interlayer holes, and metallization layer peeling due to process parameter fluctuations (such as abnormal sintering temperature gradient and uneven lamination pressure). These defects can cause serious problems such as signal distortion and thermal failure during the high-frequency operation of devices, especially posing a major threat to the reliability of spacecraft-borne electronic systems. The current technologies for HTCC ceramic defect detection have the following limitations:

[0003] Existing technologies mostly focus on off-line inspection of finished products (such as X-ray tomography and ultrasonic flaw detection). Although they can identify existing defects, they cannot trace the correlation between defects and process parameters (such as temperature fluctuations in sintering zones and excessive moisture content in green bodies) during the production process. For example, due to the out-of-control heating rate in a certain temperature zone of an aerospace HTCC substrate, batch interlayer cracks occurred. Traditional methods can only reject defective products but cannot locate the root cause of the process link, resulting in the recurrence of similar defects;

[0004] During the HTCC sintering process, the difference in thermal expansion coefficients of interlayer materials will cause the accumulation of dynamic stress, eventually forming microcracks. Existing X-ray detection requires interrupting production and cannot capture the defect evolution process in real time, while the method based on surface optical imaging is limited by the high reflectivity characteristics of the metallization layer and it is difficult to stably obtain internal defect characteristics. The accident of a satellite communication module failing during its service life due to the expansion of internal microcracks shows that the lack of on-line dynamic monitoring means will directly threaten the long-life reliability of high-value devices. Summary of the Invention

[0005] Based on the above purposes, the present invention provides an automatic HTCC ceramic defect recognition method assisted by an industrial model, including the following steps:

[0006] Step 1: Real-time collect process parameters such as sintering temperature gradient, lamination pressure, and green body moisture content during the HTCC production process, construct a process parameter spatio-temporal baseline model through a long short-term memory network, dynamically calculate the deviation degree between the current parameter and the baseline model, and generate a process deviation type label when the deviation degree exceeds a dynamic threshold adaptively based on the historical stability data of the production line;

[0007] Step 2: Conduct non-contact dielectric property scanning on the sintered HTCC substrate on the production line conveyor belt. Dynamically select the dielectric scanning frequency band according to the process deviation type label in Step 1. Obtain the dielectric constant distribution map through the resonant cavity perturbation method, calculate the local variance, and mark the coordinates of the dielectric anomaly area. Map these coordinates to the optical detection coordinate system;

[0008] Step 3: Based on the coordinates of the dielectric anomaly area, control the circularly polarized light source to irradiate the target area at an incident angle that suppresses the reflection of the metallization layer. Dynamically match the near-infrared imaging wavelength according to the dielectric scanning frequency band. Subtract the light intensity of the normal area through the background difference method to generate a high-contrast image of local defects;

[0009] Step 4: Input the dielectric anomaly frequency band feature vector and the defect image texture feature in Step 3 into the cross-modal attention network to generate a fused feature map. Activate the corresponding defect classification branch according to the process deviation type label. Use the adversarial generation network to separate the defect features and the process background texture, and output the defect type and severity level;

[0010] Step 5: Establish an association mapping library between the defect type and the process deviation type. When the same defect type appears continuously for a preset number of times, automatically adjust the temperature control parameters of the sintering furnace, and verify the change in the defect rate after parameter adjustment through the digital twin model, and update the dynamic threshold of process deviation to form a closed-loop feedback.

[0011] Preferably, the method for determining the dynamic threshold adaptively based on the production line historical stability data in Step 1 includes:

[0012] Collect the temperature gradient data, lamination pressure time series data, and green body moisture content detection data of each temperature zone during the continuous production cycle of the production line, and construct a process parameter historical database;

[0013] Divide the historical data according to the production batches, calculate the standard deviation and mean offset of the parameters within each batch, and generate a stability evaluation index;

[0014] According to the stability evaluation index corresponding to the current production batch, use the sliding window algorithm to dynamically adjust the deviation determination threshold: when the stability evaluation index is lower than the preset critical value, expand the tolerance range of the deviation determination threshold; when the index is higher than the critical value, narrow the threshold range;

[0015] The generation of the process deviation type label includes: matching the current parameter deviation mode with the predefined process anomaly mode library, and the mode library is established by clustering historical anomaly data, and each mode contains the deviation weight combination of temperature, pressure, and moisture content.

[0016] Preferably, the method for dynamically selecting the dielectric scanning frequency band according to the process deviation type label in Step 2 includes:

[0017] Construct a dielectric frequency band - defect type sensitivity mapping table: Measure the change rate of dielectric parameters of various defects at different frequency bands through experiments, and calculate the response intensity of the frequency band to the defects;

[0018] According to the process deviation type label in step 1, select the top N frequency bands with the highest response intensity from the mapping table as the scanning frequency bands, where N is dynamically adjusted according to the real - time detection requirements of the production line;

[0019] The mapping method of the coordinates of the dielectric anomaly region includes: Set a reference positioning mark on the conveyor belt, obtain the real - time position of the substrate through a laser rangefinder, and establish an affine transformation relationship between the dielectric scanning coordinate system and the optical imaging coordinate system to achieve sub - millimeter - level spatial alignment.

[0020] Preferably, the adjustment method for controlling the incident angle of the circularly polarized light source to suppress the reflection of the metallization layer in step 3 includes:

[0021] Collect the reflectivity data of the metallization layer on the surface of the HTCC substrate, train the light source angle optimization model through historical imaging samples, and output the optimal incident angle range for suppressing specular reflection;

[0022] According to the real - time detection data of the surface roughness of the current substrate, correct the incident angle in real - time. When the roughness is higher, reduce the angle between the incident angle and the normal;

[0023] The implementation of the background difference method includes: Collect the light intensity distribution template of the normal area, update the template through the dynamic weighted average algorithm, and subtract the template data from the target image to retain the defect area difference signal.

[0024] Preferably, the construction method of inputting the dielectric anomaly frequency band feature vector and the defect image texture feature into the cross - modal attention network in step 4 includes:

[0025] Normalize the dielectric frequency band feature vector according to the frequency band sensitivity weight as the Key value of the attention mechanism;

[0026] Extract the local binary pattern texture features of the defect image, and use them as the Query value after dimensionality reduction through the convolutional layer;

[0027] Calculate the cross - modal weight matrix through the multi - head attention mechanism to suppress the feature channels irrelevant to the current process deviation type;

[0028] The constraint conditions of the adversarial generative network include: Restrict the defect morphology distribution output by the generator according to the process deviation type label, and the discriminator receives real defect samples and synthetic samples generated by process parameter perturbations for adversarial training.

[0029] Preferably, the training data generation method for separating defect features and process background textures using an adversarial generative network in step 4 includes:

[0030] When a new type of defect is detected, according to the associated process deviation type label, simulate defect textures are superimposed on normal samples;

[0031] The method for generating the simulate defect textures includes: simulating the stress distribution caused by process parameter deviations through finite element analysis to generate the morphological characteristics of cracks or holes;

[0032] Spatially align the simulate defect images with the real dielectric scan data to construct an augmented training dataset.

[0033] Preferably, the method for verifying the change in defect rate after parameter adjustment in step 5 includes:

[0034] Construct a thermal-mechanical coupling simulation model for the HTCC sintering process, and input the current process parameters and the combination of parameters to be adjusted;

[0035] Simulate the difference in the coefficient of thermal expansion and the thermal stress distribution of the interlayer materials during the sintering process, and predict the probability and distribution location of defect generation;

[0036] When the simulated defect rate reduction does not meet the expectation, iteratively adjust the parameters and re-simulate until the verification conditions are met;

[0037] The method for determining the preset number of times includes: setting different thresholds according to the defect severity level, and the higher the severity, the fewer the consecutive occurrences required to trigger parameter adjustment.

[0038] Preferably, the method for establishing the dielectric frequency band-defect type sensitivity mapping table includes:

[0039] In known defect samples, measure the correlation coefficient between the change rate of dielectric constant and defect size at each frequency band;

[0040] Screen the frequency band combination with the highest contribution to defect classification through principal component analysis;

[0041] Calculate the comprehensive sensitivity score of the frequency band according to the correlation coefficient and the principal component weight, and the higher the score, the higher the scanning priority.

[0042] Preferably, the method for obtaining the surface roughness detection data includes:

[0043] Before dielectric scanning, measure the surface roughness profile of the substrate by the laser triangulation reflection method;

[0044] Calculate the surface roughness grade according to the peak-valley height difference and the root mean square value of the profile data;

[0045] The incident angle correction rule is: for each increase in the roughness grade by one level, the angle between the incident angle and the normal decreases by a preset angle step.

[0046] Preferably, the calibration method of the thermo-mechanical coupling simulation model includes:

[0047] Regularly collect the temperature gradient data and interlayer stress detection data during the actual sintering process, and compare them with the simulation results;

[0048] When the error exceeds the allowable range, update the boundary conditions of the material parameters of the simulation model through the backpropagation algorithm;

[0049] The material parameters include the thermal conductivity, thermal expansion coefficient and viscoelastic modulus of the HTCC green body.

[0050] Advantages of the present invention:

[0051] By collecting and analyzing process parameters in real time, anomalies in the production process can be detected and corrected in a timely manner, reducing the production of unqualified products. At the same time, the dynamic adjustment mechanism of process parameters ensures that the quality of ceramic products can always be maintained at the best level. With the help of dielectric property scanning and optical imaging technology, this method can accurately locate and identify defects on the HTCC ceramic substrate, and effectively classify these defects through a cross-modal attention network. This provides sufficient data support for the subsequent processing and analysis of defects. Through the correlation analysis between process parameters and defect types, this method can automatically adjust relevant production parameters when defects continuously occur, reducing defects caused by process problems. This continuous optimization process ensures the stability of the production line and the long-term reliability of product quality. At the same time, through automated data collection, analysis and process adjustment, the need for manual intervention is reduced, and the automation level of the production line is improved. The combination of the digital twin model and the closed-loop feedback mechanism makes the entire production process more intelligent and adaptive. Description of the Drawings

[0052] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only those of the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings without creative efforts.

[0053] Figure 1 It is the flowchart of the steps of the method of the present invention;

[0054] Figure 2 It is the flowchart of the steps of the method for verifying the change of the defect rate after parameter adjustment in step 5 of the method of the present invention;

[0055] Figure 3 It is the flowchart of the steps of the method for establishing the dielectric frequency band-defect type sensitivity mapping table in the method of the present invention. Detailed Embodiments

[0056] The present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments. At the same time, it should be noted here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments. For some well-known technologies, those skilled in the art can also adopt other alternative methods for implementation; moreover, the accompanying drawings are only for more specifically describing the embodiments, and are not intended to specifically limit the present invention.

[0057] Please refer to Figures 1-3 , an embodiment of the present invention provides an automatic HTCC ceramic defect recognition method assisted by an industrial model. During the production process, process parameters such as sintering temperature gradient, lamination pressure, and green body moisture content are first collected in real time. These data are analyzed by a spatio-temporal baseline model constructed by a long short-term memory network (LSTM). The LSTM model can process time series data and dynamically calculate the deviation degree of the current parameter from the baseline model. If the deviation degree exceeds the dynamic threshold adaptively based on the production line historical stability data, the system will generate a process deviation type label. This process can detect anomalies in the production process early and provide a basis for subsequent adjustments.

[0058] According to the process deviation type label generated in step 1, select an appropriate scanning frequency band for dielectric performance scanning. Use the resonant cavity perturbation method to obtain the dielectric constant distribution map of the HTCC substrate, and calculate the local variance to determine the dielectric anomaly region. By mapping the coordinates of these anomaly regions to the optical detection coordinate system, accurate spatial positioning can be provided for subsequent defect detection and image generation.

[0059] After scanning the anomaly region, control the incident angle of the circularly polarized light source to suppress the interference caused by the reflection of the metallization layer. Through the background difference method, subtract the light intensity of the normal region to generate a high-contrast image of the local defect. This process makes the defect region more prominent, facilitating subsequent image processing and defect recognition.

[0060] Input the defect image texture features generated in step 3 and the feature vectors of the dielectric anomaly frequency band into the cross-modal attention network together. This network can effectively extract important information related to defect classification by calculating the weighted weights of cross-modal features. The generative adversarial network (GAN) further separates the defect features from the process background texture, and finally outputs the defect type and severity level.

[0061] According to the association mapping library of defect types and process deviation types, when the same defect type appears continuously for a preset number of times, the system will automatically adjust the temperature control parameters of the sintering furnace. The adjusted effect is verified by a digital twin model, simulating how the new process parameters affect the defect rate, and dynamically updating the process deviation dynamic threshold according to the change of the defect rate, forming a closed-loop feedback. This mechanism can continuously optimize the process during the production process to ensure the stability of product quality.

[0062] In a possible implementation, first, the system collects temperature gradient data of each temperature zone, lamination pressure time-series data, and green body moisture content detection data during the continuous production cycle of the production line. These data will be used to construct a historical database of process parameters, and retrospective analysis and modeling of the production process will be carried out through these historical data to ensure that potential process fluctuations and abnormal trends can be captured.

[0063] Next, the system divides the historical data according to production batches and calculates the standard deviation and mean offset of each process parameter within each batch. Through these data, a stability evaluation index is generated, which can quantify the stability of each batch. The standard deviation reflects the volatility of the parameter, while the mean offset shows the long-term change trend of each process parameter. The stability evaluation index, as an important parameter to measure the current production state, directly affects the strategy of dynamically adjusting the deviation determination threshold.

[0064] According to the generated stability evaluation index, the sliding window algorithm is used to dynamically adjust the deviation determination threshold. Specifically, when the stability evaluation index is lower than the preset critical value, the system will automatically expand the tolerance range of the deviation determination threshold, which means that larger parameter deviations are allowed during the production process; when the stability evaluation index is higher than the critical value, the system will narrow the threshold range and require the process parameters to be more strictly maintained within the standard range. This dynamic adjustment mechanism can flexibly adapt to the production line state according to the different characteristics of production batches, thereby improving the stability and adaptability of the system.

[0065] After dynamically adjusting the deviation determination threshold, the system will match the deviation mode of the current process parameters with the predefined process anomaly mode library according to the deviation situation of the process parameters in the current production process. This mode library is established by clustering historical anomaly data, and each mode contains a combination of deviation weights of process parameters such as temperature, pressure, and moisture content. When the current process parameters match a certain anomaly mode, the system will generate a corresponding process deviation type label, which serves as an important basis for subsequent defect detection and classification.

[0066] In a possible implementation, first, the change rate of dielectric parameters of various defects at different frequency bands is measured through experiments. Each defect will show different response intensities at different frequency bands. Some frequency bands are more sensitive to certain defects, while other frequency bands show higher response intensities to other defects. Through these experimental data, a sensitivity mapping table between dielectric frequency bands and defect types can be established. This table records the response intensity of each frequency band to various defects, thereby providing a basis for subsequent selection of scanning frequency bands.

[0067] Based on the process deviation type label, the system can select the frequency band that is most sensitive to the current process deviation mode from the mapping table. The process deviation type label is generated in the previous step and represents the abnormal characteristics of the process in the current production process. By looking up the corresponding frequency band response intensity in the mapping table, the system will select the top N frequency bands with the highest response intensity as the scanning frequency bands. The value of N is dynamically adjusted according to the real-time detection requirements of the production line: if faster response to defects is needed, fewer frequency bands may be selected (for example, the first two frequency bands), while if more accurate identification is required, more frequency bands may be selected to obtain more comprehensive detection data.

[0068] To ensure the accurate matching of the dielectric scan results with the defect positions in the actual production process, the system needs to achieve precise spatial alignment. To do this, reference positioning marks are set on the conveyor belt to track the position of the substrate in real time. The real-time position of the substrate is obtained by a laser rangefinder, and based on the positional relationship between the reference mark and the substrate, an affine transformation relationship between the dielectric scan coordinate system and the optical imaging coordinate system is established. This step ensures the accurate docking of the scan data and the optical imaging data, thus achieving sub-millimeter-level spatial alignment and enabling the detection results to accurately indicate the specific positions of the defects.

[0069] In a possible implementation, first, the reflectivity data of the metallization layer on the surface of the HTCC substrate at different incident angles is collected. The reflection characteristics of the metallization layer change with the change of the incident angle. Especially at specific angles, the reflected light may form specular reflection on the image, affecting the detection of the defect area. Therefore, by collecting the reflectivity data at different angles, the system can construct a light source angle optimization model. Based on historical imaging samples and through training, this model can predict the range of incident angles that is most suitable for suppressing specular reflection, thereby optimizing the incident angle to reduce the specular reflection interference of the metallization layer.

[0070] The surface roughness of the HTCC substrate directly affects the reflection characteristics of light. When the surface roughness of the substrate is high, the irregularities on the surface will cause uneven light reflection. Therefore, the incident angle of the light source needs to be adjusted in real time. According to the detection data of the surface roughness of the substrate, the system can automatically correct the incident angle. Specifically, when the surface roughness increases, the system will reduce the angle between the incident angle and the normal, making the incident angle of the light source closer to the normal direction, thereby reducing the reflection interference caused by the surface roughness. This adjustment can ensure that even on a relatively rough surface, the influence of specular reflection on the image quality can be minimized to the greatest extent.

[0071] The background subtraction method is used to subtract the background data of the normal area from the target image to highlight the difference signal in the defect area. First, the system collects the light intensity distribution template of the normal area, representing the light intensity characteristics of the area without defects. Subsequently, the dynamic weighted average algorithm is used to update the template to adapt to the possible changes during the production process. By updating the background template in real time, the system can more accurately reflect the normal light intensity distribution under the current process state. Then, by performing a difference operation on the light intensity data in the target image and the updated template, the light intensity of the normal area is subtracted, and the difference signal in the defect area is retained. This process can effectively remove background interference, make the signal in the defect area more obvious, and thus improve the sensitivity and accuracy of defect detection.

[0072] In a possible implementation, first, the feature vectors in the dielectric anomaly frequency band are normalized according to the frequency band sensitivity to obtain a standardized feature representation. The frequency band sensitivity represents the response degree of different frequency bands to the defect detection of ceramic materials. By normalizing the dielectric feature vectors according to the weights of the frequency band sensitivity, it can ensure that the features in different frequency bands are reasonably weighted in the network, avoiding the excessive or insufficient influence of the features in certain frequency bands. The normalized dielectric features serve as the Key values in the attention mechanism.

[0073] The local binary pattern texture feature extraction of the defect image is to describe the texture of the local area of the image, so as to obtain visual features that can reflect defect information. The texture features of the image are processed by the convolutional layer for dimensionality reduction to generate a more compact and effective feature representation, serving as the Query value in the attention mechanism. In this way, the visual features of the image are extracted from the high-dimensional space and converted into a low-dimensional representation, thereby improving the calculation efficiency and recognition effect.

[0074] Next, the multi-head attention mechanism is used to calculate the cross-modal weight matrix between the dielectric frequency band features and the image texture features. The core idea of the attention mechanism is to calculate the importance weights of each pair of feature channels according to the relationship between the input Query and Key values. These weights represent the contributions of each feature to defect recognition in different frequency bands and image textures. Through the multi-head attention mechanism, the system can simultaneously focus on multiple feature subspaces and automatically suppress the feature channels that are irrelevant to the current process deviation type during the learning process, ensuring the accuracy and effectiveness of the recognition result.

[0075] During the training process of the generative adversarial network, process deviation type labels are used as constraints for the generator. The generator outputs the synthetic morphology of defects according to these labels. At the same time, the generated samples are also affected by process parameter perturbations, so as to simulate the defect morphology under various process deviation conditions. The discriminator receives real defect samples and synthetic samples generated by the generator, and optimizes the generator through adversarial training to enable it to generate more realistic defect samples. In this way, the generator can learn the distribution of defect morphologies under different process conditions and improve the system's ability to identify defects under complex process deviation conditions.

[0076] In a possible implementation, when the system detects a new type of defect, based on its associated process deviation type label, a simulated defect texture is first superimposed on the normal sample. At this time, the simulated defect texture is generated by simulating the stress distribution caused by process parameter deviations (such as temperature, pressure, material composition, etc.) through finite element analysis. Changes in the stress distribution may trigger cracks, holes or other types of defects, and the morphological characteristics of these defects (such as crack width, depth and shape, etc.) can be accurately predicted and simulated through finite element analysis.

[0077] Finite element analysis is a computational method used to simulate the internal stress distribution and deformation of materials. In this method, by inputting process parameters (such as temperature changes, pressure application, etc.), the microstructure of ceramic materials is simulated to predict the impact of process deviations on cracks or holes that may occur on the material surface. By simulating the stress concentration and fracture process, the geometric morphology of defects can be accurately generated, including the topological shape of cracks, the distribution of holes and other potential defect types.

[0078] After generating the simulated defect image, it needs to be spatially aligned with the real dielectric scan data. The purpose of this step is to ensure that the defect features in the simulated image can match the actual process data, so as to form a training data set with high reliability. By spatially aligning the simulated defect image with the real scan data, the model can obtain more accurate training samples, thereby improving the model's ability to identify real process deviation situations. At this time, the augmented training data set will contain defect images under various process conditions and can provide more diverse training data, thereby improving the generalization ability and adaptability of the model.

[0079] The augmented training data set generated through the above steps is used for the training of the adversarial generation network. The generator learns to generate simulated defect images from normal samples and process deviation labels, while the discriminator is responsible for judging the difference between the generated defect images and the real images. Through adversarial training, the generator is continuously optimized to generate more realistic defect images. The discriminator further improves the robustness of the system in practical applications by learning to distinguish real and generated defect images.

[0080] In a possible implementation, it is first necessary to construct a thermo-mechanical coupling simulation model for the HTCC sintering process. The thermo-mechanical coupling model can consider the effects of temperature changes and stress changes on materials. During the simulation, the current process parameters (such as temperature, pressure, etc.) and the combination of process parameters to be adjusted (such as sintering time, cooling rate, pressure distribution, etc.) are input into the model. The goal of this model is to comprehensively predict the temperature field, stress field of the material and their effects on the ceramic material under different process parameters during the sintering process.

[0081] During the sintering process of HTCC ceramic materials, due to the differences in the expansion coefficients of different layers of materials, a certain thermal stress distribution will be generated. This step simulates the effects brought about by these differences in expansion coefficients through a digital twin model, and further calculates the stress distribution between material layers. These thermal stresses may lead to the generation of defects such as cracks and pores. Therefore, through this model, it is possible to accurately predict the probability of defect generation and its distribution position in different regions.

[0082] Based on the results of the thermo-mechanical coupling simulation model, it is possible to predict which regions may have defects during the sintering process. By simulating the differences in thermal stress and expansion coefficient, the model can infer the probability of defect generation and its distribution position. This prediction is based on the calculation results of the physical properties of the material and the current process conditions, providing a scientific basis for subsequent parameter adjustment.

[0083] When the reduction rate of the simulated defect rate does not reach the expectation, the system will automatically adjust the parameters and re-perform the simulation. By continuously adjusting the process parameters and simulating the change trends of defect generation under different parameter combinations until the preset verification conditions are met. This iterative process can ensure that in actual production, the process parameters can be precisely controlled, thereby significantly reducing the occurrence of defects.

[0084] To optimize the adjustment process, the determination of the preset number of times is closely related to the defect severity level. By setting different defect severity thresholds, when the severity of the defect is higher, the corresponding number of consecutive occurrences required for parameter adjustment is less. This strategy can more finely control the response speed of the adjustment, ensuring that serious defects can be effectively handled in the shortest time and preventing high-risk defects during the production process.

[0085] In a possible implementation, it is first necessary to collect data from known defect samples, measure the rate of change of the dielectric constant at different frequency bands, and perform a correlation analysis between it and the size of the defect. Specifically, the dielectric constant reflects the response characteristics of the material at different electromagnetic frequencies, and the size and shape of the defect will directly affect the rate of change of the dielectric constant. By experimentally measuring the change in the dielectric constant at these frequency bands, the correlation coefficient between each frequency band and the defect size can be obtained. This correlation coefficient reveals the sensitivity of different frequency bands in identifying different types of defects and provides basic data for subsequent analysis.

[0086] Based on the measured correlation coefficients, the principal component analysis method is used to perform dimensionality reduction on the change in the dielectric constant of each frequency band. The goal of PCA is to extract the principal components that can retain the most information through linear transformation, so as to screen out the frequency band combinations that contribute the most to defect classification. Principal component analysis can not only reduce the influence of redundant data, but also help identify the most critical features in multi-frequency band data, thereby improving the accuracy of subsequent defect identification.

[0087] According to the correlation coefficients obtained in the previous step and the weights of the principal component analysis, the comprehensive sensitivity score of each frequency band is calculated. The comprehensive sensitivity score reflects the importance of each frequency band in the defect identification process. The higher the score of a frequency band, the more sensitive it is to defect identification. In order to ensure that the frequency bands with high sensitivity to defect types can be preferentially selected during scanning, the scanning priority of the frequency bands will also be adjusted according to this score. In this way, during the actual scanning process, the system preferentially selects those frequency bands that contribute greatly to defect classification and have high sensitivity, thereby improving the efficiency and accuracy of defect detection.

[0088] In a possible implementation, before automatically identifying HTCC ceramic defects, it is necessary to detect the surface roughness of the substrate. This step uses the laser triangulation reflection method to measure the surface profile of the substrate. The laser triangulation reflection method calculates the surface profile by emitting a laser beam onto the substrate surface and receiving the angular change of the reflected beam. This method has high precision and non-contact measurement, can accurately capture the minute undulations on the surface, and is suitable for high-precision industrial inspection.

[0089] The surface profile data obtained by the laser triangulation reflection method includes the surface peak-to-valley height difference and the root mean square value (RMS value). The peak-to-valley height difference represents the vertical distance between the highest point and the lowest point on the surface, and the root mean square value measures the overall fluctuation of the surface roughness. Based on these data, the surface roughness grade can be calculated. For example, a smaller peak-to-valley height difference and root mean square value correspond to a smoother surface, while larger values indicate a rougher surface. This calculation method usually follows specific industrial standards to generate the surface roughness grade, thus providing key information for subsequent scanning and defect identification.

[0090] Adjust the correction rule of the incident angle according to the surface roughness grade. Specifically, when the roughness grade increases by one level, the angle between the incident angle and the normal will decrease by a preset angular step. This adjustment is based on the influence of the rough surface on the dielectric constant scanning. When the surface roughness is high, the adjustment of the incident angle can help improve the quality of the scanning signal, because a rougher surface may cause uneven signal scattering or reflection, which in turn affects the recognition accuracy. By adjusting the incident angle, the scanning can more accurately reflect the true characteristics of the ceramic surface and avoid errors caused by surface roughness.

[0091] In a possible implementation, during the sintering process of HTCC ceramics, the temperature gradient and interlayer stress are important factors affecting the ceramic quality and defect formation. Therefore, during the actual sintering process, it is necessary to regularly collect these data. The temperature gradient reflects the thermal conduction difference of the ceramic at different positions, while the interlayer stress represents the stress distribution between different layers inside the ceramic. These data can effectively describe the thermo-mechanical coupling characteristics during the sintering process and serve as the basis for calibrating the simulation model.

[0092] The actually collected temperature gradient and interlayer stress data will be compared with the results output by the simulation model. Through this comparison, the accuracy of the simulation model and its agreement with the actual situation can be evaluated. If there are differences between the simulation results and the actual situation, it is necessary to further adjust the model parameters to improve the accuracy of the simulation results.

[0093] When the error between the simulation results and the actual data exceeds the preset allowable range, it is necessary to optimize the simulation model through the backpropagation algorithm. The backpropagation algorithm is a commonly used optimization method. It calculates the relationship between the error and the material parameters and continuously adjusts the material parameters in the simulation model to make the simulation results closer to the temperature gradient and interlayer stress data during the actual sintering process.

[0094] When updating the simulation model, the key material parameters to be adjusted include the thermal conductivity, thermal expansion coefficient, and viscoelastic modulus of the HTCC green body.

[0095] Specifically: Thermal conductivity: Affects the efficiency of heat conduction and is a key parameter describing the thermal conductivity characteristics of the material.

[0096] Thermal expansion coefficient: Determines the volume change of the material when heated and is an important factor in thermo-mechanical coupling.

[0097] Viscoelastic modulus: Characterizes the deformation characteristics of the material under stress and directly affects the deformation and stress distribution of the ceramic.

[0098] By precisely calibrating these material parameters, the simulation model can be made more consistent with the thermo-mechanical behavior during the actual sintering process, thereby improving the model's prediction ability.

[0099] The calibration method of the thermal-mechanical coupling simulation model collects the temperature gradient and interlayer stress data during the actual sintering process, and adjusts the material parameters in combination with the backpropagation algorithm, ensuring the consistency between the simulation model and the actual situation. This method not only improves the accuracy and prediction ability of the simulation model, but also provides more accurate theoretical support for the automatic identification of HTCC ceramic defects, thereby improving the defect identification rate and product quality control level in the industrial production process.

[0100] This invention covers any alternatives, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To enable the public to have a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments of this invention. However, those skilled in the art can fully understand this invention even without the description of these details. Additionally, well-known methods, processes, flows, components, and circuits, etc., are not described in detail to avoid unnecessary confusion to the essence of this invention.

[0101] The above are only the preferred embodiments of this invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of this invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of this invention.

Claims

1. An automatic recognition method for HTCC ceramic defects based on industrial model assistance, characterized in that, It includes the following steps: Step 1: Real-time collect the process parameters of sintering temperature gradient, lamination pressure and green body moisture content during the HTCC production process. Construct a spatio-temporal baseline model of process parameters through a long short-term memory network, and dynamically calculate the deviation degree between the current parameters and the baseline model. When the deviation degree exceeds the dynamic threshold adaptively based on the historical stability data of the production line, generate a process deviation type label. Step 2: Conduct non-contact dielectric property scanning on the sintered HTCC substrate on the production line conveyor belt. Dynamically select the dielectric scanning frequency band according to the process deviation type label in Step 1. Obtain the dielectric constant distribution map by the resonant cavity perturbation method, calculate the local variance and mark the coordinates of the dielectric anomaly area, and map the coordinates to the optical detection coordinate system. Step 3: Based on the coordinates of the dielectric anomaly area, control the circularly polarized light source to irradiate the target area at the incident angle that suppresses the reflection of the metallization layer. Dynamically match the near-infrared imaging wavelength according to the dielectric scanning frequency band, and subtract the light intensity of the normal area by the background difference method to generate a high-contrast image of local defects. Step 4: Input the dielectric anomaly frequency band feature vector and the texture feature of the defect image in Step 3 into the cross-modal attention network to generate a fused feature map. Activate the corresponding defect classification branch according to the process deviation type label, and use the generative adversarial network to separate the defect features and the process background texture, and output the defect type and severity level. Step 5: Establish an association mapping library between the defect type and the process deviation type. When the same defect type appears continuously for a preset number of times, automatically adjust the temperature control parameters of the sintering furnace, and verify the change of the defect rate after parameter adjustment through the digital twin model, and update the dynamic threshold of process deviation to form a closed-loop feedback.

2. The automatic identification method for HTCC ceramic defects based on industrial model assistance according to claim 1, wherein, The determination method of the dynamic threshold adaptively based on the historical stability data of the production line in Step 1 includes: Collect the temperature gradient data of each temperature zone, the lamination pressure time series data and the green body moisture content detection data during the continuous production cycle of the production line, and construct a historical database of process parameters. Divide the historical data according to the production batches, calculate the standard deviation and mean offset of the parameters within each batch, and generate a stability evaluation index. According to the stability evaluation index corresponding to the current production batch, use the sliding window algorithm to dynamically adjust the deviation degree judgment threshold: when the stability evaluation index is lower than the preset critical value, expand the tolerance range of the deviation degree judgment threshold; when the index is higher than the critical value, narrow the threshold range. The generation of the process deviation type label includes: matching the current parameter deviation mode with a predefined process anomaly mode library, and the mode library is established by clustering historical anomaly data, and each mode contains a combination of deviation weights of temperature, pressure and moisture content.

3. The automatic identification method for HTCC ceramic defects based on industrial model assistance according to claim 1, characterized in that The method of dynamically selecting the dielectric scanning frequency band according to the process deviation type label in Step 2 includes: Construct a dielectric frequency band - defect type sensitivity mapping table: Measure the change rate of dielectric parameters of various defects at different frequency bands through experiments, and calculate the response intensity of the frequency band to the defects. According to the process deviation type label in Step 1, select the top N frequency bands with the highest response intensity from the mapping table as the scanning frequency bands, and N is dynamically adjusted according to the real-time detection requirements of the production line. The mapping method of the coordinates of the dielectric anomaly region includes: setting a reference positioning mark on the conveyor belt, obtaining the real-time position of the substrate through a laser rangefinder, establishing an affine transformation relationship between the dielectric scanning coordinate system and the optical imaging coordinate system, and realizing sub-millimeter-level spatial alignment.

4. The automatic recognition method for HTCC ceramic defects based on industrial model assistance according to claim 1, wherein, The adjustment method for controlling the incident angle of the circularly polarized light source to suppress the reflection of the metallization layer in step 3 includes: Collecting the reflectivity data of the metallization layer on the surface of the HTCC substrate, training the light source angle optimization model through historical imaging samples, and outputting the optimal incident angle range for suppressing specular reflection; According to the surface roughness detection data of the current substrate, the incident angle is corrected in real time. When the roughness is higher, the included angle between the incident angle and the normal is reduced; The implementation of the background difference method includes: collecting the light intensity distribution template of the normal region, updating the template through the dynamic weighted average algorithm, and subtracting the template data from the target image to retain the defect region difference signal.

5. The automatic defect recognition method of HTCC ceramics based on industrial model assistance according to claim 1, wherein, The construction method of inputting the dielectric anomaly frequency band feature vector and the defect image texture feature into the cross-modal attention network in step 4 includes: Normalizing the dielectric frequency band feature vector according to the frequency band sensitivity weight as the Key value of the attention mechanism; Extracting the local binary pattern texture features of the defect image and using them as the Query value after dimensionality reduction through the convolutional layer; Calculating the cross-modal weight matrix through the multi-head attention mechanism to suppress the feature channels irrelevant to the current process deviation type; The constraint conditions of the adversarial generation network include: restricting the defect morphology distribution output by the generator according to the process deviation type label, and the discriminator receiving real defect samples and synthetic samples generated by process parameter perturbations for adversarial training.

6. The automatic recognition method for HTCC ceramic defects based on industrial model assistance according to claim 5, wherein, The training data generation method of separating defect features and process background textures using the adversarial generation network in step 4 includes: When a new type of defect is detected, according to the associated process deviation type label, simulating defect textures are superimposed on the normal samples; The generation method of the simulated defect texture includes: simulating the stress distribution caused by process parameter deviation through finite element analysis to generate the morphological features of cracks or holes; Spatially aligning the simulated defect image with the real dielectric scanning data to construct an augmented training dataset.

7. The automatic recognition method for HTCC ceramic defects based on industrial model assistance according to claim 1, characterized in that The method of verifying the change in the defect rate after parameter adjustment in step 5 through the digital twin model includes: Constructing a thermal-mechanical coupling simulation model of the HTCC sintering process and inputting the current process parameters and the combination of parameters to be adjusted; Simulating the difference in the coefficient of thermal expansion and the thermal stress distribution of the interlayer materials during the sintering process, and predicting the probability and distribution position of defect generation; When the simulated defect rate reduction does not meet the expectation, iteratively adjust the parameters and re-simulate until the verification conditions are met; The determination method of the preset number of times includes: setting different thresholds according to the defect severity level. The higher the severity, the fewer the consecutive occurrences required to trigger parameter adjustment.

8. The automatic identification method for HTCC ceramic defects based on industrial model assistance according to claim 3, wherein, The establishment method of the dielectric frequency band-defect type sensitivity mapping table includes: In known defect samples, measuring the correlation coefficient between the change rate of the dielectric constant and the defect size in each frequency band; Screening the frequency band combination with the highest contribution to defect classification through principal component analysis; Calculating the comprehensive sensitivity score of the frequency band according to the correlation coefficient and the principal component weight. The higher the score, the higher the scanning priority.

9. The automatic identification method for HTCC ceramic defects based on industrial model assistance according to claim 4, wherein The method for obtaining the surface roughness detection data includes: Before dielectric scanning, measuring the surface roughness profile of the substrate by the laser triangulation reflection method; Calculating the surface roughness grade according to the peak-to-valley height difference and the root mean square value of the profile data; The incident angle correction rule is: for each increase in the roughness grade by one level, the angle between the incident angle and the normal decreases by a preset angle step.

10. The automatic identification method for HTCC ceramic defects based on industrial model assistance according to claim 7, wherein, The calibration method for the thermal-mechanical coupling simulation model includes: Regularly collecting the temperature gradient data and the interlayer stress detection data during the actual sintering process and comparing them with the simulation results; When the error exceeds the allowable range, updating the material parameter boundary conditions of the simulation model by the backpropagation algorithm; The material parameters include the thermal conductivity, the thermal expansion coefficient, and the viscoelastic modulus of the HTCC green body.

Images