PCB circuit microdefect real-time detection algorithm optimization method based on artificial intelligence

Abstract

The invention discloses a PCB circuit microdefect real-time detection algorithm optimization method based on artificial intelligence, and relates to the field of printed circuit board monitoring, and the method comprises the steps: collecting multi-physical field data including a temperature field, a stress field, current density distribution and corrosion medium concentration in real time; generating time-space synchronous multi-modal fusion data; quantifying a thermal-mechanical-electric-chemical field interaction relationship through an inter-field coupling coefficient matrix; generating a four-dimensional synergistic effect matrix through tensor fusion; and based on the four-dimensional synergistic effect matrix, in combination with a spatio-temporal evolution diagram output by the carbonization path prediction model, establishing a three-dimensional mapping relationship among the crack size, the carbonization path and the material residual strength, and predicting a residual life prediction value. The method has the advantages that through real-time synchronous fusion of multi-physical field data and tensor coupling modeling, accurate tracking of a crack propagation path and dynamic evaluation of the residual life are achieved, the detection sensitivity is improved, and the life prediction error is reduced.

Description

Technical Field

[0001] The present invention relates to the field of printed circuit board monitoring, and in particular to an artificial intelligence-based PCB circuit micro-defect real-time detection algorithm optimization method. Background Art

[0002] With the rapid development of high-density interconnection technology, the line width / line spacing of PCB circuits has entered the micron level. During their actual use, traditional manual visual inspection and offline sampling inspection modes are unable to capture early microcracks (<3μm) and sub-surface delamination defects due to the limitation of 5-10μm resolution of conventional optical inspection equipment, which leads to insufficient defect identification sensitivity. At the same time, existing technologies do not consider the synergistic effect of thermal, mechanical, electro-chemical fields, and the multi-physical field coupling effect is not quantified, resulting in large errors in the prediction of defect expansion rate. Finally, offline inspection cannot capture the material performance degradation process caused by Joule heat accumulation under power-on conditions, resulting in a lack of dynamic monitoring capabilities. Summary of the Invention

[0003] To solve the above technical problems, an artificial intelligence-based real-time detection algorithm optimization method for PCB circuit micro-defects is provided. This technical solution solves the problem raised in the above background technology that the multi-physical field coupling effect is not quantified, resulting in an increase in the prediction error of the defect expansion rate, and the lack of dynamic monitoring capabilities.

[0004] In order to achieve the above objects, the technical solution adopted by the present invention is:

[0005] The AI-based real-time detection algorithm optimization method for PCB circuit micro-defects includes:

[0006] An infrared thermal imaging sensor array is deployed on the PCB circuit board, a stress and strain sensor group is mounted on the surface of the copper wire, a current density detection device is connected across the line node, and a corrosive medium concentration monitoring device is attached to the substrate surface to collect multi-physics field data including temperature field, stress field, current density distribution and corrosive medium concentration in real time;

[0007] Establish a 3D coordinate system for the PCB circuit, align the infrared thermal image, stress distribution map, and visible light image with the copper conductor direction, and align the spatial resolution of the current density data and stress distribution map through the copper foil grid nodes to generate multimodal fusion data synchronized in time and space.

[0008] Based on the thermal expansion coefficient of the standard PCB substrate, the yield strength parameters of the copper conductor, and the chemical corrosion sensitivity of the solder mask layer, the physical fields of thermal softening effect, mechanical fatigue, chemical corrosion, and Joule heat carbonization were established respectively. The interaction relationship between the thermal, mechanical, electro-chemical fields was quantified through the field coupling coefficient matrix.

[0009] A polar coordinate system is established in the crack tip area of ​​the copper conductor. The stress intensity factors in all directions are calculated based on the fracture mechanics formula. The equivalent softening coefficient, mechanical expansion coefficient, material embrittlement coefficient, and conduction enhancement coefficient are calculated by combining real-time multi-physics field data. A four-dimensional synergistic effect matrix is ​​generated through tensor fusion.

[0010] Based on the four-dimensional synergistic effect matrix and combined with the spatiotemporal evolution diagram output by the carbonization path prediction model, a three-dimensional mapping relationship among crack size, carbonization path and residual strength of the material is established to predict the remaining life prediction value.

[0011] Preferably, the method of aligning the spatial resolution of the current density data with the stress distribution map through the copper foil grid nodes to generate spatiotemporally synchronized multimodal fusion data specifically includes:

[0012] Use hardware to trigger the synchronous pulse signal to lock the sampling clock synchronization of the infrared thermal imager, stress sensor and current probe;

[0013] Identify the copper conductor direction in the visible light image and generate a reference coordinate system for the copper foil grid topology structure;

[0014] The stress distribution map is mapped to the reference coordinate system according to the grid nodes, and the spatial resolution of the current density data is improved to the stress distribution map level through the bicubic interpolation algorithm;

[0015] The temperature field, stress field, and current density field data streams are aligned along the time axis to generate a multimodal data cube with consistent timestamps. The multimodal data cube includes coordinates, time, temperature, stress, current density, and corrosion concentration.

[0016] Preferably, the quantification of the interaction relationship between the thermal-mechanical-electrical-chemical fields by the field coupling coefficient matrix specifically includes:

[0017] The linear thermal expansion coefficient is obtained by measuring the expansion rate of the PCB substrate length with temperature changes at a standard temperature change rate. A three-point bending test device is used to apply a constant strain to the PCB laminate structure. The stress decay curve over time is recorded by a stress sensor. The absolute value of the stress decay slope is calculated as the stress relaxation rate. The product of the thermal expansion coefficient and the stress relaxation rate is used as the thermal-mechanical coupling coefficient.

[0018] Based on the current density distribution and copper conductor resistivity collected in real time by the current density detection device, the local Joule heating power is calculated and mapped to three-dimensional space using the copper conductor grid node coordinates. The thermal diffusivity of the PCB substrate is measured using the laser flash method. The thermal conductivity is calculated based on the specific heat capacity test data. The ratio of the Joule heating power to the substrate thermal conductivity is used to generate the electro-thermal coupling coefficient.

[0019] Based on the electrochemical impedance spectroscopy test, the migration resistance of the corrosive medium solution is fitted, the ion mobility is calculated, and a convolution operation is performed on the spatial distribution matrix output by the current density detection device to obtain the current density gradient amplitude of each node. Based on the spatial correlation model between the ion mobility of the corrosive medium and the current density gradient, the chemical-electrical coupling coefficient is output;

[0020] The thermal-mechanical coupling coefficients, electro-thermal coupling coefficients, and chemical-electric coupling coefficients are summarized to construct a four-row, four-column coupling kernel matrix. The off-diagonal elements of the coupling kernel matrix are the above coupling coefficients, and the diagonal elements are the autocorrelation coefficients of each physical field.

[0021] Preferably, the calculation of the isotropic stress intensity factor based on the fracture mechanics formula specifically includes:

[0022] The copper foil grain orientation distribution data was obtained by electron backscatter diffraction technology, and a three-row and three-column material anisotropy parameter matrix was established. The main diagonal elements of the material anisotropy parameter matrix are the elastic modulus of the grain major axis, the elastic modulus of the grain minor axis, and the shear modulus. The first row and second column and the second row and first column of the material anisotropy parameter matrix are both looseness ratios, and the values ​​of all other positions are 0.

[0023] A fan-shaped grid is created in the crack tip area, multiple layers of transition units are set radially, and multiple wedge-shaped units are created circumferentially. Displacement boundary conditions consistent with the PCB laminate structure are applied to constrain the Z-direction degrees of freedom of the dielectric nodes between the substrate layers.

[0024] Nano-alumina speckle patterns were prepared on the surface of the copper wire. Dual high-speed microscope cameras were used to synchronously capture dynamic images of crack propagation, perform sub-pixel displacement calculations, obtain sub-pixel displacement field distribution data blocks, and generate a stress intensity factor spatial distribution cloud map adapted to the laminated structure.

[0025] Preferably, performing sub-pixel displacement calculation, obtaining sub-pixel displacement field distribution data blocks, and generating a stress intensity factor spatial distribution cloud map adapted to the laminated structure specifically includes:

[0026] Using the spatial pose parameters of dual high-speed microscope cameras and the principle of stereo vision triangulation, the three-dimensional coordinates and dynamic displacement images of nano-alumina speckles were analyzed. The crack tip area of ​​the copper wire was projected into the three-dimensional coordinates, and the radial displacement signal within the crack tip was subjected to empirical mode decomposition to extract the intrinsic mode function.

[0027] A circular integral path is set around the crack tip, and the displacement gradient tensor components are modified in combination with the material anisotropy parameter matrix. The stress intensity factor is inversely solved through the three-dimensional fracture mechanics relationship, and a three-dimensional vector field with the crack tip as the origin is output;

[0028] The vector field is converted to the material principal axis coordinate system, and the elastic modulus gradient constraint is introduced at the interlayer interface to generate a stress intensity factor cloud map. The X-axis of the material principal axis coordinate system is the copper foil rolling direction, the Y-axis is the transverse direction, and the Z-axis is the interlayer direction.

[0029] Preferably, generating a four-dimensional synergy matrix by tensor fusion specifically includes:

[0030] The equivalent softening coefficient, mechanical expansion coefficient, material embrittlement coefficient, and conduction enhancement coefficient are mapped into independent components of a four-dimensional tensor, each component corresponding to one dimension of the thermal, mechanical, electrical, and chemical physical fields, forming a four-dimensional tensor component set T = [equivalent softening coefficient, mechanical expansion coefficient, material embrittlement coefficient, conduction enhancement coefficient];

[0031] Perform outer product operation on the four-dimensional tensor components to generate a four-row and four-column four-dimensional coupling tensor C = T*T T , the tensor element represents the interaction strength between the physical field corresponding to the row vector and the physical field corresponding to the column vector;

[0032] Obtain the material thermal expansion coefficient and stress relaxation rate, calculate the product of the material thermal expansion coefficient and the stress relaxation rate, and obtain the thermal-mechanical coupling coefficient. Generate the mechanical-chemical coupling coefficient based on the exponential function of the historical stress corrosion sensitivity index and the environmental humidity parameter. Obtain the electro-thermal coupling coefficient through the ratio of Joule heat power to the material thermal conductivity coefficient. Output the chemical-electrical coupling coefficient based on the spatial correlation model of the corrosion medium conductivity and the current density gradient. Summarize the thermal-mechanical coupling coefficient, mechanical-chemical coupling coefficient, electrical-thermal coupling coefficient, and chemical-electrical coupling coefficient to construct a four-row and four-column inter-field coupling kernel matrix K, where the matrix elements represent the coupling coefficients between the physical field corresponding to the row and the physical field corresponding to the column.

[0033] The inter-field coupling kernel matrix K and the four-dimensional coupling tensor C are tensor-contracted and then added to the trace of the four-dimensional coupling tensor C to obtain the energy transfer weight matrix. The diagonal elements of the energy transfer weight matrix E represent the energy contribution of a single physical field, and the off-diagonal elements represent the cross-field coupling energy.

[0034] The four-dimensional coupling tensor is orthogonally decomposed based on the energy transfer weight, and the principal component eigenvectors are extracted and reconstructed into a four-dimensional synergy matrix. The diagonal elements of the matrix represent the independent contribution of each physical field to crack propagation, and the off-diagonal elements represent the coupling effect across physical fields.

[0035] Preferably, the establishing of the three-dimensional mapping relationship among the crack size, carbonization path and residual strength of the material specifically includes:

[0036] Perform spectral decomposition on the four-dimensional synergy matrix, calculate its eigenvalues ​​and corresponding orthogonal eigenvectors, sort them in descending order of their absolute values, and select the first three to form a four-row and three-column eigensubspace matrix. The rows of the eigensubspace matrix correspond to the contribution weights of the four physical fields in the principal components after dimensionality reduction, and the columns of the eigensubspace matrix correspond to the first three data retained after eigenvalue decomposition;

[0037] The eigenvectors are Schmidt orthogonalized in the material principal axis coordinate system to generate a main effect eigenvector group with physical field coupling direction characterization, and the standard stress intensity factor of the PCB circuit material is obtained by

[0038] The stress intensity factor gradient vector at the crack tip is calculated using the spatial difference method. The orthogonal eigenvector is multiplied by the stress intensity factor gradient vector at the crack tip to obtain the crack propagation direction discriminant function.

[0039] The spatiotemporal evolution diagram output by the carbonization path prediction model is processed into a three-dimensional grid. Within each grid cell, the crack size measurement value of the corresponding coordinate point, the material residual strength detection value, and the coupling weight of the four-dimensional synergy matrix are associated. A nonlinear regression model of crack length, carbonization path curvature radius, and residual strength value is established through a deep convolutional neural network. The residual strength prediction value is obtained through a feature fusion layer.

[0040] Based on the residual strength prediction value, a material strength failure threshold surface is set in the three-dimensional mapping space. The confidence interval of the residual strength value is dynamically adjusted based on the crack tip stress intensity factor to generate a three-dimensional mapping relationship map containing the probability density distribution.

[0041] Preferably, the predicted remaining life prediction value specifically includes:

[0042] Based on the stress intensity factor cloud map, the stress intensity factor gradient vector in the crack tip direction is extracted, and combined with the anisotropy parameter matrix, the rate vector is decomposed into components along the main axis of the material grain.

[0043] Perform tensor dot product operation on the eigenvector group of the four-dimensional synergy matrix and the energy transfer weight matrix to generate the equivalent crack growth rate correction coefficient under the multi-physics field coupling effect;

[0044] According to the concentration gradient of the corrosive medium in the grid unit of the carbonization path spatiotemporal evolution diagram, the corrosion front advancement rate is calculated and vector synthesized with the crack growth rate vector;

[0045] Based on the probability density distribution of the material's residual strength test values, the crack propagation paths are randomly sampled. Based on the random sampling results of the crack propagation paths, a joint probability density function of crack length, carbonization curvature radius, and residual strength is established in a three-dimensional mapping relationship map to obtain the material strength failure threshold.

[0046] A time convolution module is embedded in the deep convolutional neural network, which inputs the current multimodal data cube, four-dimensional synergy matrix and equivalent crack growth rate into the network and outputs the residual strength prediction value of the future time series.

[0047] When the remaining strength prediction value falls below the material strength failure threshold surface for the first time, the time difference from the current moment to the failure moment is calculated as the remaining life prediction value.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] The present invention proposes an artificial intelligence-based PCB circuit micro-defect real-time detection algorithm optimization method, which integrates infrared thermal imaging sensors, stress and strain sensors, current density detection devices and corrosion medium concentration monitoring devices to achieve synchronous acquisition of temperature field, stress field, current field and chemical field, forming a multi-modal sensor array, using hardware-triggered synchronous pulse signals and copper foil grid topology registration, combined with bicubic interpolation algorithm to align multi-source data resolution, and construct a six-dimensional data cube containing coordinates, time, temperature, stress, current density and corrosion concentration for time-space alignment technology, and then based on thermal expansion coefficient, Joule heat power, ion mobility The system combines parameters such as thermal, mechanical, electro-chemical fields to quantify their interactions, constructs a four-dimensional coupling kernel matrix, generates an energy transfer weight matrix by combining tensor outer product and contraction operations, and extracts principal component eigenvectors by combining orthogonal decomposition to characterize the independent contributions of physical fields and cross-field coupling effects. This allows for multi-physics field data fusion, capturing hidden defects (such as microcracks and delamination) caused by the synergistic effects of thermal, mechanical, electro-chemical fields with sub-micron sensitivity. Furthermore, the inter-field coupling kernel matrix quantifies cross-interactions, reducing the crack growth rate prediction error compared to traditional methods. The spatiotemporal data cube and tensor hardware acceleration operations support microsecond-level responses, meeting the real-time monitoring needs of production lines.

[0050] The present invention proposes an artificial intelligence-based real-time detection algorithm optimization method for PCB circuit micro-defects. It uses dual high-speed microscope cameras combined with nano-alumina speckle and stereo visual triangulation to analyze the three-dimensional displacement field, filter out noise through empirical mode decomposition, and build a sub-pixel displacement solution system. Then, it constructs the material anisotropy parameter matrix based on the grain orientation distribution, corrects the displacement gradient tensor, combines the annular integral path to invert the three-dimensional stress intensity factor, and embeds the residual network of the time convolution module. It inputs the multimodal data cube and the four-dimensional synergistic matrix, and outputs the residual intensity sequence prediction value for the next 120 seconds. Then, it can generate 10 5The crack path is organized and the joint probability density function of crack length, carbonization curvature and residual strength is established. During its use, the intrinsic displacement signal of the crack tip is extracted through dual-camera stereo vision and sub-pixel matching algorithm, combined with empirical mode decomposition, and the three-dimensional stress intensity factor cloud map is inverted to improve the accuracy of crack propagation direction discrimination, breaking through the limit of traditional optical detection. The anisotropic parameter matrix is ​​introduced into the integral calculation, the interlayer elastic modulus gradient constraint is corrected, and the stress field distribution adapted to the PCB laminate structure is generated. The joint probability density model is rasterized (0.1mm voxel) through the spatiotemporal evolution graph, and the remaining life prediction error is reduced. Finally, the spatiotemporal convolution and attention mechanism are integrated in the deep network to associate the crack size, carbonization path curvature and residual strength. The temporal dependency is captured through the time convolution module, which can achieve the prediction of nanoscale speckle displacement to macroscopic life, realize the through-type analysis of multi-scale defect dynamics, and support implicit failure warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is a schematic diagram of the process of the present invention;

[0052] Figure 2 A schematic diagram of the process of aligning the spatial resolution of current density data and stress distribution diagrams through copper foil grid nodes to generate spatiotemporally synchronized multimodal fusion data in the present invention;

[0053] Figure 3 Schematic diagram of the process of quantifying the interaction relationship between thermal, mechanical, electrochemical fields by means of the field coupling coefficient matrix in the present invention;

[0054] Figure 4 Schematic diagram of the process of calculating the stress intensity factor in all directions based on the fracture mechanics formula in the present invention;

[0055] Figure 5 Schematic diagram of the process of performing sub-pixel displacement calculation, obtaining sub-pixel displacement field distribution data blocks, and generating a stress intensity factor spatial distribution cloud map adapted to the laminated structure in the present invention;

[0056] Figure 6 Schematic diagram of the process of generating a four-dimensional synergy matrix through tensor fusion in the present invention;

[0057] Figure 7 A schematic diagram of the process of establishing a three-dimensional mapping relationship between crack size, carbonization path and material residual strength in the present invention;

[0058] Figure 8 This is a schematic diagram of the process of predicting the remaining life prediction value in the present invention. DETAILED DESCRIPTION

[0059] The following description is intended to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are merely examples, and those skilled in the art may conceive of other obvious variations.

[0060] Reference Figure 1 As shown in FIG, the artificial intelligence-based PCB circuit micro-defect real-time detection algorithm optimization method includes:

[0061] An infrared thermal imaging sensor array is deployed on the PCB circuit board, a stress and strain sensor group is mounted on the surface of the copper wire, a current density detection device is connected across the line node, and a corrosive medium concentration monitoring device is attached to the substrate surface to collect multi-physics field data including temperature field, stress field, current density distribution and corrosive medium concentration in real time;

[0062] Establish a 3D coordinate system for the PCB circuit, align the infrared thermal image, stress distribution map, and visible light image with the copper conductor direction, and align the spatial resolution of the current density data and stress distribution map through the copper foil grid nodes to generate multimodal fusion data synchronized in time and space.

[0063] Based on the thermal expansion coefficient of the standard PCB substrate, the yield strength parameters of the copper conductor, and the chemical corrosion sensitivity of the solder mask layer, the physical fields of thermal softening effect, mechanical fatigue, chemical corrosion, and Joule heat carbonization were established respectively. The interaction relationship between the thermal, mechanical, electro-chemical fields was quantified through the field coupling coefficient matrix.

[0064] A polar coordinate system is established in the crack tip area of ​​the copper conductor. The stress intensity factors in all directions are calculated based on the fracture mechanics formula. The equivalent softening coefficient, mechanical expansion coefficient, material embrittlement coefficient, and conduction enhancement coefficient are calculated by combining real-time multi-physics field data. A four-dimensional synergistic effect matrix is ​​generated through tensor fusion.

[0065] Based on the four-dimensional synergistic effect matrix and combined with the spatiotemporal evolution diagram output by the carbonization path prediction model, a three-dimensional mapping relationship among crack size, carbonization path and residual strength of the material is established to predict the remaining life prediction value.

[0066] First, an infrared thermal imaging sensor array was deployed on the PCB surface using an embedded mounting process. MEMS stress and strain sensors were mounted with a 0.5mm pitch. Hall-effect current density detection chips were integrated at the circuit nodes. An electrochemical sensitive film was applied to the substrate surface to monitor the concentration of the corrosive medium. Synchronous pulse signals were sent via an FPGA hardware trigger module, ensuring that the sampling clock error of each sensor was less than 1μs. This enabled the spatiotemporal synchronization of temperature, stress, current, and chemical fields. Furthermore, copper conductor topology was identified from the visible light image. After generating a grid reference coordinate system, the bicubic interpolation algorithm was used to increase the resolution of the current density data, constructing a spatiotemporal data cube containing six-dimensional feature vectors.

[0067] A constant strain of 0.5% / min was applied to PCB specimens using a three-point bending test rig. The thermal expansion coefficient α = 18 × 10^-6 / °C and the stress relaxation rate β, recorded by an infrared thermal imager, were used to calculate the thermal-mechanical coupling. The substrate thermal diffusivity κ was measured using a laser flash method, and the electro-thermal coupling coefficient was calculated using the current density J. Ultimately, a four-dimensional kernel matrix containing 16 coupling coefficients was constructed, enabling improved microcrack identification accuracy through multi-physics data fusion.

[0068] Based on electron backscatter diffraction, the grain orientation data of copper foil is obtained, and an anisotropic elastic matrix is ​​established. The fan-shaped meshing method is used to establish a finite element model with an accuracy of 50μm at the crack tip. The nano-speckle displacement is captured by dual high-speed cameras, and the sub-pixel correlation algorithm is used to achieve pixel displacement resolution. The three-dimensional stress intensity factor distribution is inverted. Then, the tensor fusion algorithm is used to improve the efficiency of extracting the eigenvalues ​​of the four-dimensional synergy matrix. Combined with the time series prediction module of the deep time convolutional network, the remaining life prediction error is reduced, so that the overall hidden defects caused by thermal-mechanical-electrochemical synergy can be accurately identified by quantifying the inter-field coupling effect.

[0069] Reference Figure 2 As shown, the method of aligning the spatial resolution of current density data with the stress distribution map through the copper foil grid nodes and generating spatiotemporally synchronized multimodal fusion data specifically includes:

[0070] Use hardware to trigger the synchronous pulse signal to lock the sampling clock synchronization of the infrared thermal imager, stress sensor and current probe;

[0071] Identify the copper conductor direction in the visible light image and generate a reference coordinate system for the copper foil grid topology structure;

[0072] The stress distribution map is mapped to the reference coordinate system according to the grid nodes, and the spatial resolution of the current density data is improved to the stress distribution map level through the bicubic interpolation algorithm;

[0073] The temperature field, stress field, and current density field data streams are aligned along the time axis to generate a multimodal data cube with consistent timestamps. The multimodal data cube includes coordinates, time, temperature, stress, current density, and corrosion concentration.

[0074] Using a synchronous trigger pulse signal generated by an FPGA chip, the infrared thermal imager's sampling frequency is locked at 120Hz, the stress sensor's sampling rate is set to 1kHz, and the current probe's sampling interval is synchronized to 0.83ms. The GPS clock module is used to calibrate the time base of each sensor to ensure that the acquisition error of the temperature field, stress field, and current density field is less than 0.1ms. This allows for copper wire edge detection in visible light images. The Hough transform is used to extract the characteristic lines of the wire's trajectory. A reference coordinate system is constructed using the intersections of adjacent wires as grid nodes. The pixel coordinates of the stress distribution map are mapped to this coordinate system using a bilinear transformation. To address the insufficient spatial resolution of the current density data, a super-resolution reconstruction algorithm based on an interpolation kernel function is employed. The weighted values ​​of the 16 surrounding current detection nodes are calculated using a weight function: Here, a is set to -0.5, increasing the current density resolution to 0.15 mm / pixel, matching the stress distribution map's 0.15 mm / pixel resolution. Finally, a sliding window matching is performed on the three physical field data streams using a timestamp hash table to construct a spatiotemporal data cube containing six-dimensional feature vectors. This data cube is structured as a 4D tensor in HDF5 format. This reduces the temporal alignment error of the multi-physics data through hardware-level clock synchronization, and improves spatial positioning accuracy by combining copper wire mesh topology registration technology. Furthermore, the bicubic interpolation algorithm reduces the mean square error of the current density reconstruction compared to bilinear interpolation.

[0075] Reference Figure 3 As shown, the interaction relationship between the thermal-mechanical-electrical-chemical fields quantified by the field coupling coefficient matrix specifically includes:

[0076] The linear thermal expansion coefficient is obtained by measuring the expansion rate of the PCB substrate length with temperature changes at a standard temperature change rate. A three-point bending test device is used to apply a constant strain to the PCB laminate structure. The stress decay curve over time is recorded by a stress sensor. The absolute value of the stress decay slope is calculated as the stress relaxation rate. The product of the thermal expansion coefficient and the stress relaxation rate is used as the thermal-mechanical coupling coefficient.

[0077] Based on the current density distribution and copper conductor resistivity collected in real time by the current density detection device, the local Joule heating power is calculated and mapped to three-dimensional space using the copper conductor grid node coordinates. The thermal diffusivity of the PCB substrate is measured using the laser flash method. The thermal conductivity is calculated based on the specific heat capacity test data. The ratio of the Joule heating power to the substrate thermal conductivity is used to generate the electro-thermal coupling coefficient.

[0078] Based on the electrochemical impedance spectroscopy test, the migration resistance of the corrosive medium solution is fitted, the ion mobility is calculated, and a convolution operation is performed on the spatial distribution matrix output by the current density detection device to obtain the current density gradient amplitude of each node. Based on the spatial correlation model between the ion mobility of the corrosive medium and the current density gradient, the chemical-electrical coupling coefficient is output;

[0079] The thermal-mechanical coupling coefficients, electro-thermal coupling coefficients, and chemical-electric coupling coefficients are summarized to construct a four-row, four-column coupling kernel matrix. The off-diagonal elements of the coupling kernel matrix are the above coupling coefficients, and the diagonal elements are the autocorrelation coefficients of each physical field.

[0080] First, a three-point bending test bench was built to apply a constant strain load of 0.5% / min to the PCB laminate structure. The stress decay curve was recorded in real time by a high-precision fiber Bragg grating stress sensor (accuracy ±0.05MPa), and the stress decay slope within 300 seconds was calculated. At the same time, a thermomechanical analyzer was used to measure the linear thermal expansion coefficient α of the substrate at a certain temperature change rate. The two were multiplied to obtain the thermal-mechanical coupling coefficient. The measurement accuracy of the thermal-mechanical coupling coefficient can be improved by combining the three-point bending test with the thermomechanical analysis. In the calculation of the electro-thermal coupling coefficient, based on the current density distribution data collected by the Hall array sensor and combined with the resistivity ρ of the copper wire, the formula P=J was used. 2 ρ is used to calculate the local Joule heat power. The thermal diffusivity D of the substrate is measured by the laser flash method (ASTM E1461 standard). Combined with the specific heat capacity c measured by differential scanning calorimetry, the thermal conductivity λ is calculated as D·c p ·ρ m =0.96W / (m·K)(ρ m is the substrate density), and the final electro-thermal coupling coefficient K = P / λ, so the laser flash method is combined with in-situ Joule heat measurement to realize the real-time dynamic calculation of electro-thermal coupling. Finally, a 10mV AC disturbance is applied through the electrochemical workstation, and the corrosion medium map is obtained by scanning in the 1Hz-1MHz frequency range. The migration resistance R is fitted and the ion mobility μ = σ / (zFc) (σ is the solution conductivity, z is the ion valence, F is the Faraday constant, and c is the concentration) is calculated. Then, through the electrochemical impedance spectroscopy and the current density gradient spatial correlation analysis, a quantitative model of chemical-electric coupling is established for the first time. The current density gradient matrix is ​​convolved with a 3×3 operator to calculate the spatial gradient amplitude. A Pearson correlation coefficient model with ion mobility is established, outputting the chemical-electric coupling coefficient K. The diagonal elements of the final coupling kernel matrix are the thermal field autocorrelation, force field, electric field, and chemical field, respectively. The off-diagonal elements are filled with the above coefficients, forming a complete quantitative model of the inter-field interaction relationship. The resulting four-dimensional coupling kernel matrix can accurately reflect the interactions between thermal softening, mechanical fatigue, Joule heating effect, and chemical corrosion, reducing the quantification error of multi-physics field coupling and laying a precise physical foundation for the subsequent construction of the synergistic interaction matrix.

[0081] Reference Figure 4 As shown, the calculation of the stress intensity factor in each direction based on the fracture mechanics formula specifically includes:

[0082] The copper foil grain orientation distribution data was obtained by electron backscatter diffraction technology, and a three-row and three-column material anisotropy parameter matrix was established. The main diagonal elements of the material anisotropy parameter matrix are the elastic modulus of the grain major axis, the elastic modulus of the grain minor axis, and the shear modulus. The first row and second column and the second row and first column of the material anisotropy parameter matrix are both looseness ratios, and the values ​​of all other positions are 0.

[0083] A fan-shaped grid is created in the crack tip area, multiple layers of transition units are set radially, and multiple wedge-shaped units are created circumferentially. Displacement boundary conditions consistent with the PCB laminate structure are applied to constrain the Z-direction degrees of freedom of the dielectric nodes between the substrate layers.

[0084] Nano-alumina speckle patterns were prepared on the surface of the copper wire. Dual high-speed microscope cameras were used to synchronously capture dynamic images of crack propagation, perform sub-pixel displacement calculations, obtain sub-pixel displacement field distribution data blocks, and generate a stress intensity factor spatial distribution cloud map adapted to the laminated structure.

[0085] First, an electron backscatter diffraction system was used to scan multiple areas of the copper foil surface. The Kikuchi pattern was used to analyze the grain orientation distribution, extracting the elastic modulus along the major and minor axes of the grains, as well as the shear modulus. Combined with the Poisson's ratio measured by X-ray diffraction, a third-order anisotropy parameter matrix was constructed. Finite element pre-processing software was used to create six layers of radial transition elements (each 50 μm thick) in the crack tip region, and 36 wedge-shaped elements spaced 15° apart were circumferentially divided. Displacement boundary conditions consistent with the PCB laminate structure were applied, constraining the Z-direction displacement of the nodes in the substrate FR4 dielectric layer to less than 0.1 μm, simulating the actual interlayer constraint effect. To achieve submicron displacement resolution, a 200nm aluminum oxide speckle layer was formed on the copper conductor surface using aerosol spraying technology. A stereoscopic vision system was constructed using dual high-speed microscope cameras (100,000 fps, 2048×2048 resolution). Subpixel matching of the speckle patterns in consecutive frames was performed using a digital image correlation algorithm. A gradient-based iterative method was used to achieve a displacement resolution of 0.02 pixels. The 2D image displacements were mapped to the material principal axis coordinate system using a 3D coordinate transformation matrix. Finally, a 3D stress intensity factor distribution cloud map containing the KI, KII, and KIII components was inverted and calculated using the J-integral method. Grain-level anisotropic parameter modeling reduced the error in stress intensity factor calculation. The design of sector-shaped grid transition elements reduced the error in stress singularities at the crack tip. Subpixel displacement resolution improved the spatial resolution of the displacement field. The dual-camera stereoscopic vision system broke through the limitations of traditional 2D DIC and accurately captured the Z-axis displacement component of the laminated structure, improving the match between the 3D stress intensity factor cloud map and the actual mechanical state of the PCB.

[0086] Reference Figure 5As shown, performing sub-pixel displacement calculation, obtaining sub-pixel displacement field distribution data blocks, and generating a stress intensity factor spatial distribution cloud map adapted to the laminated structure specifically include:

[0087] Using the spatial pose parameters of dual high-speed microscope cameras and the principle of stereo vision triangulation, the three-dimensional coordinates and dynamic displacement images of nano-alumina speckles were analyzed. The crack tip area of ​​the copper wire was projected into the three-dimensional coordinates, and the radial displacement signal within the crack tip was subjected to empirical mode decomposition to extract the intrinsic mode function.

[0088] A circular integral path is set around the crack tip, and the displacement gradient tensor components are modified in combination with the material anisotropy parameter matrix. The stress intensity factor is inversely solved through the three-dimensional fracture mechanics relationship, and a three-dimensional vector field with the crack tip as the origin is output;

[0089] The vector field is converted to the material principal axis coordinate system, and the elastic modulus gradient constraint is introduced at the interlayer interface to generate a stress intensity factor cloud map. The X-axis of the material principal axis coordinate system is the copper foil rolling direction, the Y-axis is the transverse direction, and the Z-axis is the interlayer direction.

[0090] A stereo vision system was constructed using two high-speed microscope cameras (frame rate 100,000 fps, pixel size 10 μm). The two cameras were fixed to a six-degree-of-freedom gimbal at a 60° angle. The spatial pose parameters were obtained using a checkerboard calibration method (translation matrix accuracy ±5 μm, rotation angle error <0.1°). A 200 nm aluminum oxide speckle layer (surface density 1200 particles / mm) was sprayed on the surface of the copper wire. 2 ), a random speckle pattern is formed through an aerosol deposition process. During dynamic acquisition, two cameras are synchronized by hardware triggering to capture crack growth sequence images with a microsecond time difference. A three-dimensional coordinate mapping model is established based on the principle of stereo vision triangulation: Where d is the parallax and f is the focal length), the submicron-level reconstruction of the three-dimensional coordinates of the nanospeckle is achieved (axial resolution 0.3μm). In the displacement solution process, a sub-pixel matching algorithm based on grayscale gradient is used to perform a local window search on the crack tip area (window size 32×32 pixels), and a displacement resolution of 0.02 pixels is achieved through bicubic spline interpolation. The radial displacement signal is subjected to empirical mode decomposition (EMD), and the screening stop criterion is set to SD=0.3. The first three-order intrinsic mode functions are extracted to filter out high-frequency noise and retain the low-frequency components reflecting the true deformation of the material. A circular integration path with a radius of 50μm is set around the crack tip, and the displacement gradient tensor components are corrected in combination with the material anisotropic parameter matrix (including elastic modulus E1=110GPa, E2=100GPa, Poisson's ratio ν=0.33). The three-dimensional J-integral formula J k =∫ I (Wδ 1k -σ ij ui,k )n j d I , inversely solve the stress intensity factor KⅠ, KⅡ, KⅢ components, and generate a three-dimensional fracture parameter field containing space vectors. During the coordinate system conversion, the calculation results can be converted to the material principal axis coordinate system (X axis along the copper foil rolling direction, Y axis horizontal direction, Z axis vertical interlayer direction), by introducing the elastic modulus gradient constraint function ( h is the laminate thickness) to correct the stress intensity factor amplitude at the interlayer interface, and finally generate a three-dimensional stress intensity factor cloud map with a grid accuracy of 0.1mm. The Z-axis displacement detection accuracy can be improved through the stereo vision system. The anisotropic parameter correction reduces the KⅠ calculation error to . The elastic modulus gradient constraint model effectively solves the stress singularity problem caused by interlayer delamination, thereby improving the reliability of crack propagation prediction in PCB laminate structures.

[0091] Reference Figure 6 As shown, generating a four-dimensional synergy matrix by tensor fusion specifically includes:

[0092] The equivalent softening coefficient, mechanical expansion coefficient, material embrittlement coefficient, and conduction enhancement coefficient are mapped into independent components of a four-dimensional tensor, each component corresponding to one dimension of the thermal, mechanical, electrical, and chemical physical fields, forming a four-dimensional tensor component set T = [equivalent softening coefficient, mechanical expansion coefficient, material embrittlement coefficient, conduction enhancement coefficient];

[0093] Perform outer product operation on the four-dimensional tensor components to generate a four-row and four-column four-dimensional coupling tensor C = T*T T , the tensor element represents the interaction strength between the physical field corresponding to the row vector and the physical field corresponding to the column vector;

[0094] Obtain the material thermal expansion coefficient and stress relaxation rate, calculate the product of the material thermal expansion coefficient and the stress relaxation rate, and obtain the thermal-mechanical coupling coefficient. Generate the mechanical-chemical coupling coefficient based on the exponential function of the historical stress corrosion sensitivity index and the environmental humidity parameter. Obtain the electro-thermal coupling coefficient through the ratio of Joule heat power to the material thermal conductivity coefficient. Output the chemical-electrical coupling coefficient based on the spatial correlation model of the corrosion medium conductivity and the current density gradient. Summarize the thermal-mechanical coupling coefficient, mechanical-chemical coupling coefficient, electrical-thermal coupling coefficient, and chemical-electrical coupling coefficient to construct a four-row and four-column inter-field coupling kernel matrix K, where the matrix elements represent the coupling coefficients between the physical field corresponding to the row and the physical field corresponding to the column.

[0095] The inter-field coupling kernel matrix K and the four-dimensional coupling tensor C are tensor-contracted and then added to the trace of the four-dimensional coupling tensor C to obtain the energy transfer weight matrix. The diagonal elements of the energy transfer weight matrix E represent the energy contribution of a single physical field, and the off-diagonal elements represent the cross-field coupling energy.

[0096] The four-dimensional coupling tensor is orthogonally decomposed based on the energy transfer weight, and the principal component eigenvectors are extracted and reconstructed into a four-dimensional synergy matrix. The diagonal elements of the matrix represent the independent contribution of each physical field to crack propagation, and the off-diagonal elements represent the coupling effect across physical fields.

[0097] During its use, the embedded data processing unit is used to vectorize the four-dimensional tensor components, and the outer product results are calculated in parallel by the matrix multiplier to generate a four-dimensional tensor structure containing 16 coupling strengths. The construction of the field coupling core matrix K is based on the product of the thermal expansion coefficient α (18×10^-6 / ℃) and the stress relaxation rate β (0.03MPa / s) to obtain the thermal-mechanical coupling coefficient, which is obtained by the Joule heat power P=J 2 The ratio of ρ to the substrate's thermal conductivity λ is used to calculate the electro-thermal coupling coefficient (P / λ), and the chemical-electrical coupling coefficient (σ·▽J) is calculated by combining the spatial correlation model of the corrosion medium's electrical conductivity σ and the current density gradient ▽J. The final constructed 4×4 inter-field coupling kernel matrix K populates the 16 coupling coefficients in the order of the physical fields. The off-diagonal elements are cross-coupling terms, and the diagonal elements calculate the energy contribution of each field through the autocorrelation function. The tensor fusion architecture enables dynamic quantification of multi-physics field coupling effects, minimizing modeling errors in the interaction between thermal, mechanical, electro-chemical fields.

[0098] During the generation of the energy transfer weight matrix, the tensor contraction operation is used to perform bilinear mapping between the kernel matrix K and the coupling tensor C, and its mathematical expression is E ij =Σ{k l}K ik C kj +tr(C)·δ ij , where δ ij is the Kronecker function. The tensor operation acceleration module implemented via FPGA can complete a trillion floating-point operations in 1μs, significantly improving computational efficiency. The resulting energy transfer weight matrix E quantifies the independent energy contributions of each physical field (diagonal elements) and the cross-field coupling energy (off-diagonal elements). Based on the orthogonal decomposition of the energy weights, the principal component eigenvectors are extracted. The error rate of the diagonal elements of the reconstructed four-dimensional synergy matrix is ​​reduced, and the quantification accuracy of the coupling effects of the off-diagonal elements is improved compared to traditional multivariate regression models. By introducing the inter-field coupling kernel matrix, direct physical effects (such as material softening caused by Joule heating) and indirect coupling effects (such as current density redistribution caused by chemical corrosion) are effectively distinguished, reducing the mean absolute error of the crack growth prediction model. In addition, the synergy matrix after orthogonal decomposition has a clear physical meaning, and its principal component eigenvectors reflect the physical field combination pattern that dominates crack growth, providing highly interpretable feature inputs for the artificial intelligence model. Real-time tensor operations achieved through hardware acceleration meet the timeliness requirements of real-time detection on the production line.

[0099] Reference Figure 7 As shown, the establishment of the three-dimensional mapping relationship between crack size, carbonization path and material residual strength specifically includes:

[0100] Perform spectral decomposition on the four-dimensional synergy matrix, calculate its eigenvalues ​​and corresponding orthogonal eigenvectors, sort them in descending order of their absolute values, and select the first three to form a four-row and three-column eigensubspace matrix. The rows of the eigensubspace matrix correspond to the contribution weights of the four physical fields in the principal components after dimensionality reduction, and the columns of the eigensubspace matrix correspond to the first three data retained after eigenvalue decomposition;

[0101] The eigenvectors are Schmidt orthogonalized in the material principal axis coordinate system to generate a main effect eigenvector group with physical field coupling direction characterization, and the standard stress intensity factor of the PCB circuit material is obtained by

[0102] The stress intensity factor gradient vector at the crack tip is calculated using the spatial difference method. The orthogonal eigenvector is multiplied by the stress intensity factor gradient vector at the crack tip to obtain the crack propagation direction discriminant function.

[0103] The spatiotemporal evolution diagram output by the carbonization path prediction model is processed into a three-dimensional grid. Within each grid cell, the crack size measurement value of the corresponding coordinate point, the material residual strength detection value, and the coupling weight of the four-dimensional synergy matrix are associated. A nonlinear regression model of crack length, carbonization path curvature radius, and residual strength value is established through a deep convolutional neural network. The residual strength prediction value is obtained through a feature fusion layer.

[0104] Based on the residual strength prediction value, a material strength failure threshold surface is set in the three-dimensional mapping space. The confidence interval of the residual strength value is dynamically adjusted based on the crack tip stress intensity factor to generate a three-dimensional mapping relationship map containing the probability density distribution.

[0105] The parallel matrix decomposition algorithm performs spectral decomposition on the four-dimensional synergy matrix and calculates its eigenvalue λ by the QR iteration method. i and orthogonal eigenvectors v i , after sorting by |λ1|≥|λ2|≥|λ3|≥|λ4|, the first three order eigenvectors are selected to form a four-row and three-column eigensubspace matrix V4×3. The standard stress intensity factor is measured by scanning the crack tip area with a 50μm step size using a high-precision six-axis robotic arm equipped with an X-ray stress analysis probe. Calculation of stress intensity factor gradient using spatial difference method In the material principal axis coordinate system, taking the rolling direction (X axis) as the reference vector, the projection components of the transverse direction (Y axis) and the interlayer direction (Z axis) are eliminated in turn to generate the main effect characteristic vector group v with direction sensitivity i, perform tensor dot product operation on the orthogonalized eigenvector and the stress intensity factor gradient The discriminant function of crack propagation direction is generated. When φ>0.85, it is judged as the propagation mode along the grain boundary, and when φ<0.3, it is judged as the transgranular propagation mode.

[0106] In the three-dimensional rasterization processing stage, the octree spatial segmentation algorithm is used to divide the spatiotemporal evolution diagram of the carbonization path into voxel units with an accuracy of 0.1mm. Each voxel integrates the crack length measurement value (measured by scanning electron microscopy SEM), the material residual strength (measured by a micro-tensile test bench) and the coupling weight of the four-dimensional synergy matrix. The constructed deep convolutional neural network contains 5 residual blocks, each of which consists of a 3×3×3 convolution kernel, a batch normalization layer and an activation function. The input layer receives a 64×64×64 voxel data block and extracts the spatial correlation features of the carbonization path curvature radius ρc and the crack length a through 3D convolution. The attention mechanism is introduced in the feature fusion layer to calculate the attention score of the coupling weight of each physical field. Where Q, K, and V are query and key-value matrices, and d = 256 is the dimension parameter. The coupling weights are weightedly fused with the geometric features. The residual strength prediction value is output through the fully connected layer, and the loss function is used. in is the predicted value of residual strength, δ=5MPa, and Training optimization is performed, and the sensitivity to outliers is reduced compared to the traditional MSE loss function.

[0107] Based on the prediction results, the moving least squares method is used to construct the material strength failure threshold surface f(a,ρ c ,S r )=0, when the confidence interval α=0.95, 10 5 The crack propagation path is calculated and the probability density distribution of the residual strength p(Sr|a,ρ c ). Introducing the dynamic correction module of the stress intensity factor at the crack tip, when the real-time monitored KI value exceeds 0.8K IC When the confidence interval is narrowed, the half-width of the confidence interval is reduced to improve failure warning sensitivity. The resulting three-dimensional mapping relationship map can achieve accurate positioning of crack length and carbonization curvature radius, shorten the residual strength prediction error, and thus reduce the prediction error. This simultaneous orthogonal decomposition of multi-physics field coupling characteristics and attention-weighted fusion effectively solves the problem of associative modeling of hidden defects under complex working conditions, ensuring the reliability and real-time performance of PCB circuit remaining life prediction meet production-line application requirements.

[0108] Reference Figure 8 As shown, the predicted remaining life prediction value specifically includes:

[0109] Based on the stress intensity factor cloud map, the stress intensity factor gradient vector in the crack tip direction is extracted, and combined with the anisotropy parameter matrix, the rate vector is decomposed into components along the main axis of the material grain.

[0110] Perform tensor dot product operation on the eigenvector group of the four-dimensional synergy matrix and the energy transfer weight matrix to generate the equivalent crack growth rate correction coefficient under the multi-physics field coupling effect;

[0111] According to the concentration gradient of the corrosive medium in the grid unit of the carbonization path spatiotemporal evolution diagram, the corrosion front advancement rate is calculated and vector synthesized with the crack growth rate vector;

[0112] Based on the probability density distribution of the material's residual strength test values, the crack propagation paths are randomly sampled. Based on the random sampling results of the crack propagation paths, a joint probability density function of crack length, carbonization curvature radius, and residual strength is established in a three-dimensional mapping relationship map to obtain the material strength failure threshold.

[0113] A time convolution module is embedded in the deep convolutional neural network, which inputs the current multimodal data cube, four-dimensional synergy matrix and equivalent crack growth rate into the network and outputs the residual strength prediction value of the future time series.

[0114] When the remaining strength prediction value falls below the material strength failure threshold surface for the first time, the time difference from the current moment to the failure moment is calculated as the remaining life prediction value.

[0115] At the hardware level, a six-axis robotic arm equipped with an X-ray stress analysis probe is used to scan the crack tip area with a step size of 50μm, and the gradient vector in the crack propagation direction is extracted based on the stress intensity factor cloud map. KI, combined with the material anisotropic parameter matrix [Eij] (the main diagonal elements are the grain major axis elastic modulus 110GPa, the minor axis modulus 100GPa, and the shear modulus 40GPa), is calculated by vector decomposition algorithm Decompose the expansion rate into the direction of the main axis of the grain. Build a tensor operation acceleration module in the data processing unit and transform the eigenvector group V of the four-dimensional synergy matrix into 4×3 and the energy transfer weight matrix E 4×4 Perform double dot product operation Generate a correction coefficient matrix containing the thermal-mechanical-electrical-chemical field coupling strength, and obtain the equivalent crack growth rate correction coefficient through normalization. For corrosion path prediction, an electrochemical microprobe array is used to measure the concentration gradient of the corrosive medium within the grid unit. Calculation of corrosion front advancement rate based on equation (where D z =2.3×10 -9 m2 / s is the ion mobility), and the crack growth rate vector V ext Perform space vector synthesis V = ηV + βV, (β = 0.15 is the environmental humidity correction factor). The Weibull distribution of the material residual strength test value is analyzed by Monte Carlo method for 10 5 Random sampling is performed to establish the crack length a and carbonization curvature radius ρ in the three-dimensional mapping map. c With the residual strength S r The joint probability density function p(a,ρ c ,S r ), when the probability density value exceeds the threshold p th =0.92 is determined as the material strength failure threshold. A temporal convolutional module (TCN) is embedded in the neural network architecture, whose input layer receives a multimodal data cube containing 256×256×256 voxels, and the dilated causal convolution (dilation factor d = 2 k , k=1-5) extract time series features, combine the principal component eigenvectors of the four-dimensional synergy matrix to perform cross-channel attention weighting, and output the remaining intensity prediction value S in the next 120 seconds r(t) When S r(t) The first time it falls below the failure threshold surface f(a,ρ c ,S r )=0, cubic spline interpolation is used to accurately calculate the failure time point t f , remaining life t rem =t f -t now The prediction error is reduced.

[0116] In summary, the advantages of the present invention are: through real-time synchronous fusion of multi-physics field data and tensor coupling modeling, combined with sub-pixel three-dimensional defect dynamics analysis and deep learning timing prediction, accurate tracking of PCB micron-level crack propagation paths and dynamic evaluation of remaining life are achieved, the sensitivity of hidden defect detection is improved, and the life prediction error is reduced.

[0117] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions merely illustrate the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. The artificial intelligence-based PCB circuit micro-defect real-time detection algorithm optimization method is characterized by: include: An infrared thermal imaging sensor array is deployed on the PCB circuit board, a stress and strain sensor group is mounted on the surface of the copper wire, a current density detection device is connected across the line node, and a corrosive medium concentration monitoring device is attached to the substrate surface to collect multi-physics field data including temperature field, stress field, current density distribution and corrosive medium concentration in real time; Establish a 3D coordinate system for the PCB circuit, align the infrared thermal image, stress distribution map, and visible light image with the copper conductor direction, and align the spatial resolution of the current density data and stress distribution map through the copper foil grid nodes to generate multimodal fusion data synchronized in time and space. Based on the thermal expansion coefficient of the standard PCB substrate, the yield strength parameters of the copper conductor, and the chemical corrosion sensitivity of the solder mask layer, the physical fields of thermal softening effect, mechanical fatigue, chemical corrosion, and Joule heat carbonization were established respectively. The interaction relationship between the thermal, mechanical, electro-chemical fields was quantified through the field coupling coefficient matrix. A polar coordinate system is established in the crack tip area of ​​the copper conductor. The stress intensity factors in all directions are calculated based on the fracture mechanics formula. The equivalent softening coefficient, mechanical expansion coefficient, material embrittlement coefficient, and conduction enhancement coefficient are calculated by combining real-time multi-physics field data. A four-dimensional synergistic effect matrix is ​​generated through tensor fusion. Based on the four-dimensional synergistic effect matrix and combined with the spatiotemporal evolution diagram output by the carbonization path prediction model, a three-dimensional mapping relationship among crack size, carbonization path and residual strength of the material is established to predict the remaining life prediction value.

2. The method for optimizing the PCB circuit micro-defect real-time detection algorithm based on artificial intelligence according to claim 1, characterized in that: The method of aligning the spatial resolution of the current density data with the stress distribution map through the copper foil grid nodes and generating spatiotemporally synchronized multimodal fusion data specifically includes: Use hardware to trigger the synchronous pulse signal to lock the sampling clock synchronization of the infrared thermal imager, stress sensor and current probe; Identify the copper conductor direction in the visible light image and generate a reference coordinate system for the copper foil grid topology structure; The stress distribution map is mapped to the reference coordinate system according to the grid nodes, and the spatial resolution of the current density data is improved to the stress distribution map level through the bicubic interpolation algorithm; The temperature field, stress field, and current density field data streams are aligned along the time axis to generate a multimodal data cube with consistent timestamps. The multimodal data cube includes coordinates, time, temperature, stress, current density, and corrosion concentration.

3. The method for optimizing the PCB circuit micro-defect real-time detection algorithm based on artificial intelligence according to claim 2, characterized in that: The quantification of the interaction relationship between the thermal, mechanical, electrochemical and electrical fields by the field coupling coefficient matrix specifically includes: The linear thermal expansion coefficient is obtained by measuring the expansion rate of the PCB substrate length with temperature changes at a standard temperature change rate. A three-point bending test device is used to apply a constant strain to the PCB laminate structure. The stress decay curve over time is recorded by a stress sensor. The absolute value of the stress decay slope is calculated as the stress relaxation rate. The product of the thermal expansion coefficient and the stress relaxation rate is used as the thermal-mechanical coupling coefficient. Based on the current density distribution and copper conductor resistivity collected in real time by the current density detection device, the local Joule heating power is calculated and mapped to three-dimensional space using the copper conductor grid node coordinates. The thermal diffusivity of the PCB substrate is measured using the laser flash method. The thermal conductivity is calculated based on the specific heat capacity test data. The ratio of the Joule heating power to the substrate thermal conductivity is used to generate the electro-thermal coupling coefficient. Based on the electrochemical impedance spectroscopy test, the migration resistance of the corrosive medium solution is fitted, the ion mobility is calculated, and a convolution operation is performed on the spatial distribution matrix output by the current density detection device to obtain the current density gradient amplitude of each node. Based on the spatial correlation model between the ion mobility of the corrosive medium and the current density gradient, the chemical-electrical coupling coefficient is output; The thermal-mechanical coupling coefficients, electro-thermal coupling coefficients, and chemical-electric coupling coefficients are summarized to construct a four-row, four-column coupling kernel matrix. The off-diagonal elements of the coupling kernel matrix are the above coupling coefficients, and the diagonal elements are the autocorrelation coefficients of each physical field.

4. The method for optimizing the PCB circuit micro-defect real-time detection algorithm based on artificial intelligence according to claim 3, characterized in that: The calculation of the isotropic stress intensity factor based on the fracture mechanics formula specifically includes: The copper foil grain orientation distribution data was obtained by electron backscatter diffraction technology, and a three-row and three-column material anisotropy parameter matrix was established. The main diagonal elements of the material anisotropy parameter matrix are the elastic modulus of the grain major axis, the elastic modulus of the grain minor axis, and the shear modulus. The first row and second column and the second row and first column of the material anisotropy parameter matrix are both looseness ratios, and the values ​​of all other positions are 0. A fan-shaped grid is created in the crack tip area, multiple layers of transition units are set radially, and multiple wedge-shaped units are created circumferentially. Displacement boundary conditions consistent with the PCB laminate structure are applied to constrain the Z-direction degrees of freedom of the dielectric nodes between the substrate layers. Nano-alumina speckle patterns were prepared on the surface of the copper wire. Dual high-speed microscope cameras were used to synchronously capture dynamic images of crack propagation, perform sub-pixel displacement calculations, obtain sub-pixel displacement field distribution data blocks, and generate a stress intensity factor spatial distribution cloud map adapted to the laminated structure.

5. The method for optimizing the PCB circuit micro-defect real-time detection algorithm based on artificial intelligence according to claim 4, characterized in that: The performing of sub-pixel displacement calculation, obtaining sub-pixel displacement field distribution data blocks, and generating a stress intensity factor spatial distribution cloud map adapted to the laminated structure specifically includes: Using the spatial pose parameters of dual high-speed microscope cameras and the principle of stereo vision triangulation, the three-dimensional coordinates and dynamic displacement images of nano-alumina speckles were analyzed. The crack tip area of ​​the copper wire was projected into the three-dimensional coordinates, and the radial displacement signal within the crack tip was subjected to empirical mode decomposition to extract the intrinsic mode function. A circular integral path is set around the crack tip, and the displacement gradient tensor components are modified in combination with the material anisotropy parameter matrix. The stress intensity factor is inversely solved through the three-dimensional fracture mechanics relationship, and a three-dimensional vector field with the crack tip as the origin is output; The vector field is converted to the material principal axis coordinate system, and the elastic modulus gradient constraint is introduced at the interlayer interface to generate a stress intensity factor cloud map. The X-axis of the material principal axis coordinate system is the copper foil rolling direction, the Y-axis is the transverse direction, and the Z-axis is the interlayer direction.

6. The method for optimizing the PCB circuit micro-defect real-time detection algorithm based on artificial intelligence according to claim 5, characterized in that: Generating a four-dimensional synergy matrix by tensor fusion specifically includes: The equivalent softening coefficient, mechanical expansion coefficient, material embrittlement coefficient, and conduction enhancement coefficient are mapped into independent components of a four-dimensional tensor, each component corresponding to one dimension of the thermal, mechanical, electrical, and chemical physical fields, forming a four-dimensional tensor component set T = [equivalent softening coefficient, mechanical expansion coefficient, material embrittlement coefficient, conduction enhancement coefficient]; Perform outer product operation on the four-dimensional tensor components to generate a four-row and four-column four-dimensional coupling tensor C = T*T T , the tensor element represents the interaction strength between the physical field corresponding to the row vector and the physical field corresponding to the column vector; Obtain the material thermal expansion coefficient and stress relaxation rate, calculate the product of the material thermal expansion coefficient and the stress relaxation rate, and obtain the thermal-mechanical coupling coefficient. Generate the mechanical-chemical coupling coefficient based on the exponential function of the historical stress corrosion sensitivity index and the environmental humidity parameter. Obtain the electro-thermal coupling coefficient through the ratio of Joule heat power to the material thermal conductivity coefficient. Output the chemical-electrical coupling coefficient based on the spatial correlation model of the corrosion medium conductivity and the current density gradient. Summarize the thermal-mechanical coupling coefficient, mechanical-chemical coupling coefficient, electrical-thermal coupling coefficient, and chemical-electrical coupling coefficient to construct a four-row and four-column inter-field coupling kernel matrix K, where the matrix elements represent the coupling coefficients between the physical field corresponding to the row and the physical field corresponding to the column. The inter-field coupling kernel matrix K and the four-dimensional coupling tensor C are tensor-contracted and then added to the trace of the four-dimensional coupling tensor C to obtain the energy transfer weight matrix. The diagonal elements of the energy transfer weight matrix E represent the energy contribution of a single physical field, and the off-diagonal elements represent the cross-field coupling energy. The four-dimensional coupling tensor is orthogonally decomposed based on the energy transfer weight, and the principal component eigenvectors are extracted and reconstructed into a four-dimensional synergy matrix. The diagonal elements of the matrix represent the independent contribution of each physical field to crack propagation, and the off-diagonal elements represent the coupling effect across physical fields.

7. The method for optimizing the PCB circuit micro-defect real-time detection algorithm based on artificial intelligence according to claim 6, characterized in that: The establishment of the three-dimensional mapping relationship between crack size, carbonization path and material residual strength specifically includes: Perform spectral decomposition on the four-dimensional synergy matrix, calculate its eigenvalues ​​and corresponding orthogonal eigenvectors, sort them in descending order of their absolute values, and select the first three to form a four-row and three-column eigensubspace matrix. The rows of the eigensubspace matrix correspond to the contribution weights of the four physical fields in the principal components after dimensionality reduction, and the columns of the eigensubspace matrix correspond to the first three data retained after eigenvalue decomposition; Perform Schmidt orthogonalization on the eigenvectors in the material principal axis coordinate system to generate a group of main effect eigenvectors representing the physical field coupling direction. Obtain the standard stress intensity factor of the PCB circuit material and calculate the crack tip stress intensity factor gradient vector using the spatial difference method. Multiply the orthogonal eigenvectors by the crack tip stress intensity factor gradient vector to obtain the crack propagation direction discriminant function. The spatiotemporal evolution diagram output by the carbonization path prediction model is processed into a three-dimensional grid. Within each grid cell, the crack size measurement value of the corresponding coordinate point, the material residual strength detection value, and the coupling weight of the four-dimensional synergy matrix are associated. A nonlinear regression model of crack length, carbonization path curvature radius, and residual strength value is established through a deep convolutional neural network. The residual strength prediction value is obtained through a feature fusion layer. Based on the residual strength prediction value, a material strength failure threshold surface is set in the three-dimensional mapping space. The confidence interval of the residual strength value is dynamically adjusted based on the crack tip stress intensity factor to generate a three-dimensional mapping relationship map containing the probability density distribution.

8. The method for optimizing the PCB circuit micro-defect real-time detection algorithm based on artificial intelligence according to claim 7, characterized in that: The predicted remaining life prediction value specifically includes: Based on the stress intensity factor cloud map, the stress intensity factor gradient vector in the crack tip direction is extracted, and combined with the anisotropy parameter matrix, the rate vector is decomposed into components along the main axis of the material grain. Perform tensor dot product operation on the eigenvector group of the four-dimensional synergy matrix and the energy transfer weight matrix to generate the equivalent crack growth rate correction coefficient under the multi-physics field coupling effect; According to the concentration gradient of the corrosive medium in the grid unit of the carbonization path spatiotemporal evolution diagram, the corrosion front advancement rate is calculated and vector synthesized with the crack growth rate vector; Based on the probability density distribution of the material's residual strength test values, the crack propagation paths are randomly sampled. Based on the random sampling results of the crack propagation paths, a joint probability density function of crack length, carbonization curvature radius, and residual strength is established in a three-dimensional mapping relationship map to obtain the material strength failure threshold. A time convolution module is embedded in the deep convolutional neural network, which inputs the current multimodal data cube, four-dimensional synergy matrix and equivalent crack growth rate into the network and outputs the residual strength prediction value of the future time series. When the remaining strength prediction value falls below the material strength failure threshold surface for the first time, the time difference from the current moment to the failure moment is calculated as the remaining life prediction value.

Images