Copper-clad plate pressing implicit abnormality prediction method based on contour pattern recognition, server and storage medium

Abstract

The application provides a copper-clad plate pressing hidden abnormality prediction method based on contour pattern recognition, a server and a storage medium. A cross-scale contour mapping state field is generated through curvature and fractal two-dimensional coding by obtaining a copper-clad plate edge contour sampling sequence and a reference contour sequence; the mapping field is gradient flow decomposed to separate geometric constraint components and instability components to construct an invariant contour pattern distribution manifold; the trend mutual induction degree of the disturbance components corresponding to interlayer slip and resin uneven infiltration hidden abnormalities in the multiple batch manifolds is measured to generate an evolution trend atlas; when the two types of disturbance components in the atlas are synchronously monotonically increasing in the abnormal response time window and the slope convergence index exceeds a threshold value, a prediction signal is generated. The application realizes early coupling prediction of copper-clad plate pressing hidden abnormalities, and improves the quality monitoring capability and the response timeliness of process control.

Description

Technical Field

[0001] This application relates to the field of anomaly detection, and in particular to a method for predicting latent anomalies in copper-clad laminate lamination based on contour pattern recognition, a server, and a storage medium. Background Technology

[0002] The lamination process of copper clad laminates is a critical step in determining the quality of interlayer bonding. During lamination, fluctuations in process conditions can lead to latent anomalies such as interlayer slippage or uneven resin impregnation. These anomalies cannot be visually detected as defects after lamination, but they can potentially impact the long-term reliability of the board. Current copper clad laminate lamination quality inspection typically relies on offline measurement of the edge contour of the laminated board, comparing the measured contour with a standard template to determine if the dimensions exceed allowable tolerances. However, this method, based on overall deviation comparison, only focuses on the macroscopic compliance of the contour and struggles to capture subtle anomaly patterns hidden in the local morphology of the contour. It lacks the ability to identify latent anomalies that have not yet formed significant dimensional deviations, often requiring the accumulation of latent anomalies and their evolution into explicit defects before they are discovered. This approach fails to meet the need for early-stage prediction of lamination process anomalies. Summary of the Invention

[0003] This invention provides a method for predicting latent anomalies in copper clad laminate lamination based on contour pattern recognition, as well as a server and storage medium.

[0004] In a first aspect, the present invention provides a method for predicting latent anomalies in copper-clad laminate lamination based on contour pattern recognition, comprising: Obtain the discrete sampling point sequence of the edge contour formed by the copper clad laminate after the lamination process, and obtain the lamination process reference contour sequence corresponding to the copper clad laminate; Cross-dimensional contour morphology mapping is performed on the discrete sampling point sequence of the copper clad laminate edge profile and the lamination process reference profile sequence. In the contour pattern observation space jointly defined by the curvature response scale and the fractal feature scale, a cross-scale contour mapping state field of the copper clad laminate edge profile relative to the lamination process reference profile is generated. The cross-scale contour mapping state field characterizes the constraint deviation distribution between the geometric shape of the copper clad laminate edge profile and the geometric shape of the lamination process reference profile in different observation dimensions. Based on the distribution of contour morphology constraint deviation in the cross-scale contour mapping state field, contour pattern decomposition at the gradient flow level is performed on the cross-scale contour mapping state field. The geometric instability component of the copper-clad laminate edge contour is separated from the geometric constraint component of the cross-scale contour mapping state field, generating an intrinsic contour pattern distribution manifold. The intrinsic contour pattern distribution manifold contains the contour pattern tensor in the region where the contour morphology constraint deviation is not significant and the contour pattern tensor in the region where the contour morphology constraint deviation changes abruptly. The intrinsic profile pattern distribution manifold is subjected to profile pattern perturbation evolution analysis along the time series. Based on the spatial adjacency relationship and temporal propagation trajectory of the profile pattern tensor, the trend mutual induction effect of the profile pattern perturbation component corresponding to the latent anomaly of interlayer slip and the profile pattern perturbation component corresponding to the latent anomaly of resin wetting unevenness in the intrinsic profile pattern distribution manifold is measured between consecutive pressing batches, and a profile anomaly perturbation mutual induction evolution trend map is generated. When the contour pattern perturbation component corresponding to the latent anomaly of interlayer slip and the contour pattern perturbation component corresponding to the latent anomaly of uneven resin wetting in the contour anomaly perturbation mutual induced evolution trend map show a synchronous monotonically increasing trend within a preset anomaly response time window and the slope convergence index of the trend increasing curve exceeds the preset trend coordination threshold, a copper clad laminate lamination latent anomaly prediction signal is generated.

[0005] Secondly, embodiments of the present invention provide a server, including a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the program to implement the steps in the above method.

[0006] Thirdly, embodiments of this application provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method.

[0007] The method for predicting latent anomalies in copper-clad laminate (CCL) lamination provided by this invention acquires a discrete sampling point sequence of the CCL edge contour and a baseline contour sequence of the lamination process, and performs cross-dimensional contour morphology mapping. A cross-scale contour mapping state field is generated within the observation space jointly defined by the curvature response scale and the fractal feature scale, allowing contour geometric deviations to be characterized in both curvature morphology and fractal roughness dimensions, thus obtaining richer constraint deviation distribution information than a single contour metric. Based on the distribution morphology of constraint deviations in this mapped state field, contour pattern decomposition at the gradient flow level is performed, separating geometric instability components from geometric constraint components to form an intrinsic contour pattern distribution manifold containing contour pattern tensors of constraint regions and abrupt change regions. This allows contour deformation patterns related to latent anomalies to be presented independently without being masked by the overall deformation trend. Contour pattern perturbation evolution analysis is performed on this manifold along a time series. Based on the trend mutual induction effect of perturbation components corresponding to interlayer slip and uneven resin wetting between consecutive lamination batches, a contour anomaly perturbation mutual induction evolution trend map is generated, which can capture the temporal mutual excitation and co-evolution relationship between the two types of latent anomalies. When the two disturbance components in the spectrum show a synchronous monotonically increasing trend within the abnormal response time window and the slope convergence index meets the preset conditions, a prediction signal is generated, which realizes the early detection of latent anomalies in the pressing process. This helps to intervene in the process before irreversible interlayer defects are formed, thereby improving the consistency of pressing quality and production stability. Attached Figure Description

[0008] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present invention and, together with the specification, serve to explain the technical solutions of the present invention.

[0009] Figure 1 This is a schematic diagram illustrating the implementation process of a method for predicting latent anomalies in copper clad laminate lamination based on contour pattern recognition, provided in an embodiment of the present invention.

[0010] Figure 2 This is a schematic diagram of a server hardware entity provided in an embodiment of the present invention. Detailed Implementation

[0011] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. The described embodiments should not be regarded as limitations on the present invention. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0012] In the following description, references to "some embodiments" describe a subset of all possible embodiments; however, it is understood that "some embodiments" may be the same or different subsets of all possible embodiments and may be combined with each other without conflict. The terms "first / second / third" are used merely to distinguish similar objects and do not represent a specific ordering of objects. It is understood that "first / second / third" may be interchanged in a specific order or sequence where permissible, so that the embodiments of the invention described herein can be implemented in an order other than that illustrated or described herein.

[0013] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein is for descriptive purposes only and is not intended to limit the scope of the invention.

[0014] This invention provides a method for predicting latent anomalies in copper clad laminate lamination based on contour pattern recognition, which can be executed by a server processor. Figure 1 This is a schematic diagram illustrating the implementation process of a method for predicting latent anomalies in copper-clad laminate lamination based on contour pattern recognition, as provided in an embodiment of the present invention. Figure 1 As shown, the method includes: Step S100: Obtain the discrete sampling point sequence of the edge contour formed by the copper clad laminate after the lamination process, and obtain the lamination process reference contour sequence corresponding to the copper clad laminate.

[0015] The edge profile formed by the lamination process of copper-clad laminate (CCL) is the geometric boundary that is finally solidified and formed by the material flow, mold constraints, and thermal stress release of the substrate edge after the hot lamination process. The discrete sampling point sequence is a set of coordinate points collected sequentially at certain spatial intervals along this edge profile. Each sampling point contains at least its abscissa and ordinate values ​​in a two-dimensional plane, used to discretize and characterize the geometric shape of the continuous profile curve. The lamination process reference profile sequence is the theoretical geometric trajectory or standard template trajectory that the edge profile of the CCL should present under ideal lamination process conditions. It serves as a reference benchmark for subsequent profile shape comparison. Its data organization is also a sequentially arranged coordinate point sequence, and its sampling density maintains a spatial correspondence with the discrete sampling point sequence of the edge profile, in order to establish a point-to-point or arc-length-to-arc-length mapping basis.

[0016] To obtain a discrete sampling point sequence of the edge contour formed by the copper-clad laminate after lamination, a high-resolution industrial line scan camera can be used to scan the complete edge of the laminated copper-clad laminate line by line to obtain a high-contrast grayscale image of the edge region. This grayscale image is then input into a vision processing unit based on a sub-pixel edge detection algorithm to perform contour extraction. This sub-pixel edge detection algorithm first uses the Canny operator to calculate the gradient magnitude and suppress non-maximum values ​​in the grayscale image, generating a pixel-level edge point set. Then, in the gradient direction of each pixel-level edge point, a one-dimensional grayscale moment or quadratic curve fitting method is used to analyze the neighborhood grayscale distribution, calculating a more refined sub-pixel-level coordinate position of the edge point within the pixel, thus forming a discrete sampling point sequence of the copper-clad laminate edge contour. To obtain the lamination process reference contour sequence corresponding to the copper clad laminate, in this embodiment, the reference coordinate point sequence with the same number of sampling points and consistent arc length distribution as the discrete sampling point sequence can be generated by using the equal arc length resampling algorithm based on the edge contour geometric data defined in the computer-aided design drawings of the copper clad laminate product; or, copper clad laminate products that have been manually inspected and confirmed to have no appearance defects and excellent dimensional accuracy can be selected from historical production batches, and their edge contours can be extracted using the same sub-pixel edge detection algorithm as described above, and the edge contours can be solidified into the lamination process reference contour sequence after smoothing and denoising.

[0017] Step S200: Perform cross-dimensional contour morphology mapping on the discrete sampling point sequence of the copper clad laminate edge contour and the lamination process reference contour sequence. In the contour pattern observation space jointly defined by the curvature response scale and the fractal feature scale, generate a cross-scale contour mapping state field of the copper clad laminate edge contour relative to the lamination process reference contour. The cross-scale contour mapping state field characterizes the constraint deviation distribution between the geometric shape of the copper clad laminate edge contour and the geometric shape of the lamination process reference contour in different observation dimensions.

[0018] The curvature response scale is a metric that defines the accuracy of contour morphology observation based on the degree of local curvature. It focuses on the degree of change in tangent direction at each sampling point on the contour and uses multi-level curvature response encoding layers to characterize the contribution of curvature amplitudes of different magnitudes to the overall contour morphology differences. The fractal feature scale is a metric that defines the accuracy of contour morphology observation based on the self-similarity of multi-granular roughness. It focuses on the variation of the apparent length of the contour under different observation step sizes and uses multi-level observation granularity layers to reveal the cross-scale correlation between the microscopic irregularities and macroscopic geometry of the contour. The contour pattern observation space is an abstract multi-dimensional observation grid spanned by the curvature response scale dimension and the fractal feature scale dimension. Each observation position in this space corresponds to a specific combination of curvature response encoding layers and fractal observation granularity layers, providing a unified quantitative coordinate system for fine comparison of copper-clad laminate edge contour morphology. The cross-scale contour mapping state field is a multidimensional state distribution field defined in the above contour pattern observation space. It uses the contour position index and observation dimension combination as independent variables, and the corresponding curvature deviation tensor component and fractal deviation encoding component as field values. In this way, it systematically describes the constraint deviation distribution of the geometric shape of the copper clad laminate edge contour relative to the geometric shape of the lamination process reference contour in different observation dimensions.

[0019] In an optional embodiment, step S200 may specifically include the following steps S210 to S260: Step S210: Perform curvature response contour encoding on the discrete sampling point sequence of the copper clad laminate edge contour to generate an edge curvature encoding sequence. Each encoding element of the edge curvature encoding sequence records the local curvature direction mark and local curvature amplitude description at the corresponding sampling point.

[0020] Curvature response contour coding is a quantization process that transforms the local bending characteristics of a contour curve into a structured coded sequence. Its core lies in the joint analysis of the change in tangential direction and the arc length step at each sampling point, extracting a curvature direction marker and a curvature amplitude description that uniquely identify the bending morphology at that point, and then fusing these two into a compact combined codeword. The local curvature direction marker is a sign bit used to identify the bending direction of the contour at that sampling point. Typically, it uses clockwise and counterclockwise bending as two basic states to capture the concavity / convexity direction information of the contour. The local curvature amplitude description is a discretized hierarchical representation of the degree of bending at that sampling point. It is obtained by non-uniform quantization coding of continuous curvature amplitudes, allowing different levels of bending intensity to map to different coding levels, thus compressing data volume while retaining the ability to distinguish morphological differences. The edge curvature coding sequence is the serialized data formed by sequentially arranging the combined codewords corresponding to each sampling point along the discrete sampling point sequence of the copper-clad laminate edge contour.

[0021] In an optional embodiment, step S210 may specifically include the following steps S211 to S216: Step S211: Extract the tangent direction angle recursively from the discrete sampling point sequence of the copper clad laminate edge contour to generate a sequence of tangent direction angle changes between adjacent sampling points.

[0022] The tangent direction angle is the angle between the tangent line to the edge contour curve at a sampling point and the positive direction of a preset reference coordinate axis. It is usually expressed in radians or degrees, and its value reflects the orientation of the contour at that point. Tangent direction angle recursive extraction is a method that calculates the tangent direction angle at each sampling point sequentially along a discrete sampling point sequence. First, a local tangent direction vector is formed by selecting the current sampling point and its next adjacent sampling point along the sampling sequence direction. Then, the angle between this vector and the positive direction of the reference coordinate axis is solved using the arctangent function, thus obtaining the tangent direction angle of the current sampling point. The sequence of tangent direction angle changes between adjacent sampling points is generated by performing a backward difference operation on the recursively obtained tangent direction angle sequence. Each element corresponds to the change in the rotation angle of the contour orientation between two adjacent sampling points.

[0023] For example, assume that the discrete sampling point sequence of the copper-clad laminate edge contour contains N sampling points, with coordinates Pi(xi, yi), where i ranges from 0 to N-1. For each sampling point Pi, calculate the direction vector Vi from Pi to Pi+1, whose components are (Δxi=x... i+1 -x i , Δyi=y i+1 -y i Furthermore, the angle between this direction vector and the positive X-axis is calculated using the four-quadrant arctangent function atan2(Δyi, Δxi), yielding the tangent direction angle θi at point Pi. For the final point P of the sequence... N-1 Its tangent direction angle is directly taken as θ. N-2 Extend and assign values. Then, perform a difference operation on adjacent elements of the tangent direction angle sequence {θi} to generate the tangent direction angle change sequence {Δθi}, where... The value of i ranges from 0 to N-2. During the calculation, the angle difference needs to be normalized and wrapped. That is, when the absolute value of Δθi exceeds π, it is adjusted to the interval [-π, π] by adding or subtracting an integer multiple of 2π to ensure the physical consistency of the angle change.

[0024] Step S212: Based on the sequence of changes in tangent direction angle and the arc length step between adjacent sampling points, generate the local curvature amplitude at each sampling point. The local curvature amplitude is directly proportional to the change in tangent direction angle and inversely proportional to the arc length step.

[0025] Arc length step size is the actual arc length distance along the edge profile curve between two adjacent sampling points. Within the framework of a discrete sampling point sequence, this arc length can be approximated as the Euclidean distance between two adjacent sampling points. Local curvature amplitude is a scalar physical quantity that describes the degree of curvature of the profile curve at a certain sampling point. Its value directly corresponds to the rate of change of the tangent direction angle along the arc length at that point; the larger the value, the more significant the curvature at that point.

[0026] Specifically, for sampling point P i The corresponding arc length step s i P can be calculated i With P i+1 The Euclidean distance between them is obtained, i.e., s i =sqrt((Δx i ) 2 +(Δy i ) 2 Furthermore, the change in tangent direction angle Δθ obtained in step S211 is... i With arc length step s i Substitute into the formula for calculating the local curvature amplitude κ i =|Δθ i | / s i P can then be obtained. i The local curvature amplitude at point s. In this embodiment, for s i The value of is protected against zero when s i When the arc length is less than the preset minimum arc length threshold, s will be... i This is forcibly set to the minimum arc length threshold to avoid numerical overflow caused by overly dense sampling points. The local curvature amplitude κ... i The unit is the reciprocal of radians, and its magnitude directly reflects P. i The bending strength of the contour curve.

[0027] Step S213: Determine the sign and direction of the tangent direction angle change sequence. Based on the clockwise or counterclockwise attribute of the tangent direction angle change trend, assign a curvature direction label to each sampling point. The curvature direction label and the local curvature amplitude together constitute a curvature description pair.

[0028] The sign direction determination is based on each element Δθ in the sequence of changes in the tangent direction angle. i The operation uses algebraic symbols to determine the direction of rotation of the contour bending at the sampling point. When Δθ i When the value is positive, it indicates that the tangential direction angle increases counterclockwise along the sampling advance direction, and the contour at that location exhibits a counterclockwise bending characteristic; when Δθ iWhen the value is negative, it indicates that the tangential direction angle decreases clockwise along the sampling advance direction, and the contour at that point exhibits a clockwise curvature characteristic. The curvature direction marker is a binary or binary identifier that records the direction of this curvature rotation. For example, the counterclockwise curvature marker can be set to the code value "1", and the clockwise curvature marker can be set to the code value "0". The curvature description pair is a binary data structure composed of the curvature direction marker and the corresponding local curvature amplitude, used to completely characterize the curvature response features at a single sampling point.

[0029] In practical implementation, the sequence of changes in tangent direction angle {Δθ} is traversed. i}, for each Δθ i If its value is greater than zero, then it is a sampling point P. i Assign a curvature direction marker "1" to indicate counterclockwise curvature; if its value is less than 0, assign a curvature direction marker "0" to indicate clockwise curvature; if Δθ i If the value is exactly 0, then there is no directional curvature at that point, and it can be classified into one of the aforementioned two categories according to preset rules, or assigned a neutral label alone. The curvature direction label generated therefrom is then compared with the local curvature amplitude κ generated in step S212. i Combined into an ordered binary tuple (direction marker, κ) i ), that is, constituting sampling point P i The curvature description is correct.

[0030] Step S214: Arrange the curvature description pairs of each sampling point into a curvature description sequence along the sampling order of the discrete sampling point sequence, and perform spatial index encoding on the curvature description sequence so that each curvature description pair is associated with its normalized arc length position label on the edge contour.

[0031] The curvature description sequence is a one-dimensional sequence formed by arranging the curvature description pairs corresponding to each sampling point in ascending order of sampling point index. This sequence faithfully preserves the spatial sequence of the contour curves. Spatial index encoding is an encoding operation that assigns a unique label to each description pair in the curvature description sequence, identifying its spatial position on the contour curve. This allows subsequent morphological analysis to accurately correlate the encoded features with the physical contour position. The normalized arc length position label is a dimensionless position parameter calculated with the total arc length of the contour as the denominator and the cumulative arc length of the current sampling point relative to the starting sampling point as the numerator. It ranges from 0 to 1 and eliminates the influence of the absolute size of the contour, unifying the position reference system for the edge contours of copper-clad laminates of different sizes.

[0032] In this embodiment, starting from sampling point P0, the arc length step s between adjacent sampling points is accumulated point by point along the sampling sequence. i To obtain each sampling point P i The cumulative arc length L at the location i=Σ j=0 i-1 s j Where L0 is specified as 0. The total arc length L of the profile. total For L N-1 With s N-1 The sum of the samples. Then the sampling point P. i Normalized arc length position label l i =L i / L total Subsequently, the curvature description of each sampling point is paired with the corresponding normalized arc length position label l. i Bind to generate a triple (l i , Direction marker, κ i The indexed curvature description terms are formed by arranging all the indexed curvature description terms in the sampling order, which constitutes the curvature description sequence after spatial index encoding.

[0033] Step S215: Based on the distribution range of local curvature amplitude in the curvature description sequence, perform non-uniform quantization encoding on the local curvature amplitude, and map the continuous amplitude to the curvature response encoding level to obtain the local curvature amplitude description.

[0034] The distribution range of the local curvature amplitude is the local curvature amplitude κ of all sampling points in the curvature description sequence. i The non-uniform quantization coding level is a technique that maps continuous numerical values ​​to discrete coding levels based on the statistical characteristics of data distribution. It abandons the uniform quantization approach of using the same quantization step size across all intervals, instead allocating narrower quantization intervals in regions with dense curvature amplitude distribution to preserve subtle differences, and wider quantization intervals in regions with sparse curvature amplitude distribution to reduce redundant coding. The curvature response coding level is a set of discrete levels defined during the non-uniform quantization coding process. Each level corresponds to a specific local curvature amplitude description, which serves as the identifier or sequence number of that level.

[0035] For example, a non-uniform quantization method based on the cumulative distribution function is employed. First, the local curvature amplitude κ of all sampling points in the curvature description sequence is collected. i A set of amplitude values ​​is formed. After removing the minimum values ​​that are zero or close to noise, an amplitude histogram is constructed, and its probability density function and cumulative distribution function are calculated. The number of curvature response coding levels is preset to K. K quantiles are selected at equal intervals from the 0 to 1 interval of the cumulative distribution function. These quantiles are mapped back to the amplitude axis, resulting in K threshold values. This threshold sequence exhibits a non-uniform distribution characteristic on the amplitude axis, with smaller intervals where amplitudes are dense and larger intervals where amplitudes are sparse. For each sampling point, the local curvature amplitude κ... iThe hierarchical index or hierarchical code corresponding to the threshold interval into which the sample falls is used as the description of the local curvature amplitude of that sampling point. Thus, continuous local curvature amplitudes are transformed into discrete curvature response coding hierarchical identifiers, enabling subsequent coding fusion and bias analysis to be performed efficiently within the discrete coding domain.

[0036] Step S216: Perform joint encoding and fusion of curvature direction marker and local curvature amplitude description to generate edge curvature encoding sequence. Each encoding element of the edge curvature encoding sequence is a combined encoding word, which contains direction marker bit field and amplitude description bit field.

[0037] Joint coding fusion is the operation of combining curvature direction markers and local curvature amplitude descriptions belonging to the same sampling point into a single binary or decimal combined codeword according to a preset bit allocation rule. A combined codeword is a structured coding unit, internally divided into direction marker bit fields and amplitude description bit fields. The direction marker bit field occupies the highest or a specific bit of the codeword and is used to store the binary value of the curvature direction marker. The amplitude description bit field occupies the remaining lower bits of the codeword and is used to store the hierarchical coded value of the local curvature amplitude description. The edge curvature coding sequence is the sequence formed by arranging the combined codewords corresponding to all sampling points in the sampling order.

[0038] In this embodiment, assuming the curvature direction marker requires 1 bit and the local curvature magnitude description requires Q bits to meet the encoding requirements of K curvature response encoding levels, the total bit width of the combined coded word is 1 + Q bits. For sampling point P... i The binary value of the curvature direction marker is filled into the direction marker bit field, and the hierarchical encoded binary value of the local curvature amplitude description is filled into the amplitude description bit field. After concatenation, the combined code word of that point is obtained. The above concatenation operation is performed on all sampling points to obtain the edge curvature encoding sequence. This sequence uniformly expresses the curvature direction and curvature degree at each sampling point on the contour in a compact structured encoding form.

[0039] Step S220: Perform curvature response contour encoding on the pressing process reference contour sequence to generate a reference curvature encoding sequence. The encoding element structure of the reference curvature encoding sequence is dimensionally aligned with the encoding element structure of the edge curvature encoding sequence.

[0040] The reference curvature coding sequence is serialized data with the exact same coding structure and coding parameter configuration as the edge curvature coding sequence. Its generation process is completely consistent with the operation performed on the discrete sampling point sequence of the copper-clad laminate edge contour in step S210. Specifically, the following steps are performed sequentially: Step S211: Tangent direction angle recursive extraction is performed on the pressing process reference contour sequence to obtain the tangent direction angle change sequence between adjacent sampling points; Step S212: Local curvature amplitude at each sampling point is generated based on the tangent direction angle change sequence and the arc length step between adjacent sampling points; Step S213: Sign direction discrimination is performed to assign curvature direction markers and form curvature description pairs; Step S214: Curvature description pairs are arranged along the sequence and spatial index encoding is performed based on normalized arc length position labels; Step S215: Non-uniform quantization encoding of amplitudes is performed based on the local curvature amplitude distribution range to obtain local curvature amplitude descriptions; and Step S216: Joint encoding and fusion of curvature direction markers and local curvature amplitude descriptions is performed to generate a reference curvature encoding sequence composed of combined encoding words. Dimensional alignment means that the number of orientation marker bits, amplitude description bits, number of curvature response coding levels, and thresholding strategy of non-uniform quantization of the combined codes in the baseline curvature coding sequence are completely consistent with those in the edge curvature coding sequence. This ensures that the combined codes in the two sequences can be compared and the deviation calculated element by element under the same curvature response scale system.

[0041] Step S230: Based on the edge curvature coding sequence and the reference curvature coding sequence, construct the contour deviation tensor field under the curvature response scale. The contour deviation tensor field takes the contour position index as the independent variable and the curvature direction offset and curvature amplitude offset as tensor components.

[0042] The profile deviation tensor field is a two-dimensional or higher-order tensor distribution field defined in the index space of the copper-clad laminate edge profile position. Each field point corresponds to a sampling point position on the edge profile. The tensor at the field point consists of two components: curvature direction offset and curvature amplitude offset. The curvature direction offset measures the consistency of the bending direction between the actual copper-clad laminate edge profile and the lamination process reference profile at that position. When the bending directions are consistent, the offset is zero or takes its minimum value; when the directions are opposite, the offset takes its maximum value. The curvature amplitude offset measures the difference in bending intensity between the actual copper-clad laminate edge profile and the lamination process reference profile at that position. Its value is the layer difference described by the local curvature amplitude of the two, or a function transformation of the layer difference.

[0043] In this embodiment, the direction marker segment value of the i-th combined codeword in the edge curvature coding sequence is set to D. i_actual The amplitude description segment value is A. i_actualThe direction marker segment value of the i-th combined codeword in the reference curvature coding sequence is D. i_ref The amplitude description segment value is A. i_ref Curvature direction offset ΔD i The calculation rule is: if D i_actual Equals D i_ref Then ΔD i Take 0; if D i_actual Not equal to D i_ref Then ΔD i Set to 1 to indicate direction mismatch. Curvature amplitude offset ΔA i Take A directly i_actual With A i_ref The absolute difference, i.e. ΔA i =|A i_actual -A i_ref For each contour position index i, construct a two-component tensor T. i =[ΔD i , ΔA i The tensors at all locations, arranged along the contour position index, constitute the contour deviation tensor field at the curvature response scale. This tensor field, in a compact coded offset form, characterizes the curvature shape deviation of the copper clad laminate edge contour relative to the lamination process reference contour on the same scale dimension.

[0044] Step S240: Perform fractal-scale contour encoding on the discrete sampling point sequence of the copper-clad laminate edge contour to generate an edge fractal-scale contour encoding sequence. The edge fractal-scale contour encoding sequence describes the roughness distribution structure of the edge contour at a multi-level observation granularity.

[0045] Fractal-scale contour coding is a process of quantitatively and coded descriptions of the geometric irregularities of contour curves across granularities, starting from multi-scale roughness characteristics. Multi-level observation granularity involves changing the step size or resolution when observing and measuring the contour curve, thereby obtaining several levels of apparent length at different levels of refinement. Coarse-grained observation ignores micro-level fluctuations and retains only macro-level trends, while fine-grained observation captures more detailed contour fluctuations. The roughness distribution structure is a set of length variation patterns exhibited by the contour curve at different observation granularities, revealing the strength of self-similarity or self-affineness of the contour curve at different scales. The edge fractal-scale contour coding sequence is a serialized coding structure that arranges the fractal-scale descriptive components extracted from different observation granularity levels in order from fine to coarse observation granularity and binds them to the corresponding observation granularity level index.

[0046] In an optional embodiment, step S240 may specifically include the following steps S241 to S246: Step S241: Perform multi-scale step size resampling on the discrete sampling point sequence of the copper clad laminate edge contour to generate several sets of resampled contour sequences with different observation granularities. The sampling step size of each set of resampled contour sequences is greater than the sampling step size of the previous set of resampled contour sequences.

[0047] Multi-scale step resampling is an operation that uses a discrete sequence of sampling points on the edge contour of a copper-clad laminate as the original data, and resamples it according to a set of incremental sampling step sizes to obtain contour representation sequences at different spatial resolutions. The sampling step size is defined as the cumulative arc length distance traversed each time along the arc length direction when selecting resampling points on the original discrete sampling point sequence. The observation granularity is positively correlated with the sampling step size; the larger the sampling step size, the coarser the observation granularity, and the higher the degree to which contour details are filtered out.

[0048] In this embodiment, a set of observation granularity levels is predefined, containing M levels, and the corresponding sampling step size set is {t1, t2, ..., t}. M}, satisfying t1 <t2<...<t M Where t1 is taken as the average sampling interval of the original discrete sampling point sequence as the finest observation granularity step size, t M A preset proportion of the total arc length of the contour is taken as the coarsest observation granularity step size. For the m-th observation granularity level, an equal arc length resampling algorithm is used: starting from the initial sampling point P0 of the original sequence, the arc length step s between the original sampling points is accumulated along the sampling order. i Whenever the accumulated arc length reaches or exceeds the current sampling step size t m When the value is an integer multiple of the arc length, a resampling point coordinate is determined at the current arc length position using linear interpolation or nearest neighbor selection. This resampling point is added to the m-th resampling contour sequence, and the accumulated arc length is reset to continue until the entire original sequence has been traversed. After M levels of resampling, M resampling contour sequences are generated with the number of sampling points decreasing sequentially and the spatial details becoming increasingly simplified.

[0049] Step S242: Measure the total length of the contour segments for each group of resampled contour sequences to obtain the apparent length of the contour at the corresponding observation granularity. The apparent length of the contour decreases as the observation granularity increases.

[0050] The total length measurement of the profile segment is the process of calculating the total length of the piecewise linear polyline formed by connecting adjacent resampled points in a set of resampled profile sequences. The apparent length of the profile is the length value of the copper-clad laminate edge profile measured by approximating it with the aforementioned piecewise linear polyline at a specific observation granularity.

[0051] Specifically, for the m-th resampled contour sequence, which contains Q m The resampling points are sequentially denoted as R0, R1, ..., R2.Qm-1 Calculate R for adjacent resampling points sequentially. j With R j+1 The Euclidean distance d between them j Then the apparent length L of the profile at this observation granularity m =Σ j=0 Qm-2 d j Because the resampling sequence corresponding to a coarser observation granularity ignores the high-frequency micro-ripples on the profile curve, its piecewise linear polygonal line is a smooth approximation of the true curve. Therefore, the measured apparent length L of the profile is... m Typically smaller than the apparent length L at a finer observation granularity. m-1 The apparent length of the profile shows a monotonically decreasing trend as the observation granularity increases.

[0052] Step S243: Based on the ratio of the apparent length of the profile at adjacent observation granularities to the ratio of the corresponding observation granularities, generate the profile length change rate in double logarithmic coordinates, which serves as the fractal scale description component within the observation granularity interval.

[0053] The profile apparent length ratio is the ratio of the profile apparent length of the coarser level to the profile apparent length of the finer level in two adjacent observation granularity levels. The observation granularity ratio is the ratio of the sampling step size of the corresponding two levels. The profile length change rate in double logarithmic coordinates is a metric obtained by taking the slope of the two ratios respectively. It characterizes the sensitivity of the profile length to changes in the observation scale, that is, the fractal scaling characteristics of the profile within that scale range.

[0054] For example, for the observation granularity interval formed by the (m-1)th observation granularity level and the mth observation granularity level, let the sampling step size of the finer level be t. m-1 The apparent length of the profile is L m-1 The sampling step size for coarser levels is t. m The apparent length of the profile is L m Calculate the length ratio r L =L m-1 / L m Calculate the step size ratio r t =t m / t m-1 In double logarithmic coordinates, the fractal scale description component F m From formula F m =log(r L ) / log(r t Determined. This component F... m F represents the fractal dimension correlation characteristics exhibited by the contour curve within this specific scale range. m The fluctuations in the values ​​reflect the variation patterns of profile roughness at different scales.

[0055] Step S244: Arrange the fractal scale description components within all observation granularity intervals in order from fine to coarse observation granularity to form a fractal scale description component sequence.

[0056] The fractal scale descriptor component sequence is all the fractal scale descriptor components F1, F2, ..., F2 calculated sequentially from adjacent observation granularity levels in step S243. M-1 The sequence is a one-dimensional sequence obtained by arranging the observation granularity intervals from fine to coarse in spatial order. The components with earlier indices in this sequence correspond to the rate of change of profile length in the fine-scale interval, reflecting the micro-roughness characteristics; the components with later indices correspond to the rate of change of profile length in the coarse-scale interval, reflecting the macro-profile irregularity. The sequence as a whole constitutes a discretized representation of the cross-scale roughness structure of the copper-clad laminate edge profile.

[0057] Step S245: Apply inter-scale continuity constraints to the fractal scale description component sequence, use a sliding window to smooth the trend consistency of adjacent fractal scale description components, and generate a smooth fractal scale description sequence. The smooth fractal scale description sequence suppresses inter-scale jumps caused by sampling noise.

[0058] The inter-scale continuity constraint means that the fractal scale descriptive components corresponding to adjacent observation granularity intervals should theoretically have a certain degree of consistency in change, rather than exhibiting high-frequency, drastic jumps. This is because the edge profile of a copper-clad laminate, as a physically continuous entity, typically exhibits gradual rather than abrupt changes in its multi-scale roughness characteristics. A sliding window is a fixed-width data selection box that moves across sequential data. By aligning the window's center with each element in the sequence, it calculates the statistical representative value of all elements within the window and replaces the current element's value, thereby achieving a smooth and consistent denoising operation.

[0059] In this embodiment, a mean-smoothing or median-smoothing sliding window with a window width of W is used, where W is a preset odd number, such as 3 or 5. The center of the sliding window is sequentially aligned with the 2nd to M-2nd components in the fractal scale description component sequence (when W=3). An arithmetic mean is calculated on the W fractal scale description components currently covered by the window, and the resulting average value is used as the smoothed fractal scale description component at the corresponding position in the smoothed sequence. For the boundary components at both ends of the sequence, the boundary values ​​can be copied by symmetric extension to ensure complete window coverage. After sliding window processing, isolated jump values ​​in the original fractal scale description component sequence caused by sampling noise or local small-scale defects are effectively suppressed, and the generated smoothed fractal scale description sequence has higher inter-scale continuity while maintaining the overall trend.

[0060] Step S246: Bind and encode the smooth fractal scale description sequence with the corresponding observation granularity level index to generate the edge fractal scale contour coding sequence. Each coding element in the edge fractal scale contour coding sequence corresponds to a local contour roughness code of an observation granularity level.

[0061] The observation granularity level index is a label used to uniquely identify each observation granularity level; it can be a level number or the sampling step size value corresponding to the level. Binding encoding is the operation of combining each smooth component in the smooth fractal scale description sequence with its corresponding observation granularity level index into a joint description unit. Local contour roughness encoding is the encoded form of this joint description unit, which contains observation scale information and fractal roughness feature information at that scale.

[0062] Specifically, for the m-th observation granularity interval (corresponding to the smooth fractal scale description component at the m-th level), its observation granularity level index value m and the corresponding smooth fractal scale description component value SF are... m These are combined into an encoding pair, and both are binary-encoded using a preset bit width to generate the local contour roughness code for that observation granularity interval. The local contour roughness codes corresponding to all M-1 observation granularity intervals are arranged in order from fine to coarse observation granularity, thus generating an edge fractal scale contour encoding sequence. This sequence, in a multi-level joint encoding form, fully characterizes the roughness distribution structure of the copper-clad laminate edge contour in the fractal feature scale dimension.

[0063] Step S250: Perform fractal scale contour coding on the reference contour sequence of the pressing process to generate a reference fractal scale contour coding sequence. The observation granularity level of the reference fractal scale contour coding sequence corresponds one-to-one with the observation granularity level of the edge fractal scale contour coding sequence.

[0064] The reference fractal scale contour encoding sequence is generated by taking the lamination process reference contour sequence as input and performing the same fractal scale contour encoding operation as the discrete sampling point sequence of the copper clad laminate edge contour in step S240. The operation specifically includes: step S241, which involves multi-scale step-size resampling of the pressing process baseline contour sequence to generate several sets of resampled contour sequences with different observation granularities; step S242, which involves measuring the total length of contour segments in each set of resampled contour sequences to obtain the apparent length of the contour at the corresponding observation granularity; step S243, which involves generating the contour length change rate in double logarithmic coordinates as a fractal scale description component based on the ratio of the apparent length of the contour at adjacent observation granularities to the ratio of the corresponding observation granularity; step S244, which involves arranging all fractal scale description components in order of observation granularity from fine to coarse to form a fractal scale description component sequence; step S245, which involves using a sliding window to perform trend consistency smoothing to obtain a smoothed fractal scale description sequence; and step S246, which involves binding the smoothed fractal scale description sequence with the observation granularity level index to generate a baseline fractal scale contour encoding sequence composed of local contour roughness encoding. A one-to-one correspondence means that the two encoding sequences use the exact same set of observation granularity levels, i.e., the sampling step set {t1, t2, ...,t...}. M The number of observation granularity levels M and the resampling interpolation algorithm are kept strictly consistent to ensure that the encoded elements of the two sequences can be compared layer by layer within the same fractal feature scale framework.

[0065] Step S260: Input the contour deviation tensor field, the edge fractal scale contour coding sequence and the reference fractal scale contour coding sequence into the contour pattern observation space construction stage. Generate a cross-scale contour mapping state field through cross-dimensional mapping fusion. The cross-scale contour mapping state field describes the constraint deviation distribution on the observation grid jointly formed by the curvature dimension and the fractal dimension.

[0066] The contour pattern observation space construction step is a signal integration and processing module. Its core operational logic lies in registering and fusing deviation information from the curvature response scale and the fractal feature scale, respectively, on a two-dimensional observation grid jointly defined by the two scale dimensions. The curvature dimension uses the curvature response encoding level or curvature deviation tensor components as the observation coordinate axes, while the fractal dimension uses the observation granularity level index as the observation coordinate axes. The intersection of the two dimensions forms a grid covering all observation combinations. Cross-dimensional mapping and fusion involves retrieving the corresponding curvature direction offset and curvature amplitude offset from the contour deviation tensor field for each sampling point on the copper-clad laminate edge contour, based on its normalized arc length position label. Simultaneously, it extracts the differences in the fractal scale descriptive components at the corresponding observation granularity level from the edge fractal scale contour encoding sequence and the reference fractal scale contour encoding sequence, thereby associating and integrating these two sets of deviation description information from different dimensions at the corresponding nodes of the observation grid.

[0067] In this embodiment, the input to the contour pattern observation space construction step includes three parts: first, the contour deviation tensor field generated in step S230, which provides curvature dimension deviation using normalized arc length position labels as indices; second, the edge fractal scale contour encoding sequence generated in step S246; and third, the reference fractal scale contour encoding sequence generated in step S250, the latter two providing fractal dimension deviation using observation granularity level indices as dimensions. The construction process is as follows: First, a two-dimensional observation grid is established with curvature dimension coordinate k and fractal dimension coordinate m as row and column indices, where k traverses all possible curvature amplitude offset value ranges, and m traverses all observation granularity levels. For each sampling point on the copper-clad laminate edge contour, it is assumed that its normalized arc length position is l. i Extract the corresponding tensor T from the contour deviation tensor field. i =[ΔD i , ΔA i ], ΔA i Mapped to curvature dimension coordinate k i Simultaneously, smooth fractal scale description components for all observation granularity levels corresponding to the point are extracted from the edge fractal scale profile encoding sequence, and the difference is calculated with the corresponding components in the baseline fractal scale profile encoding sequence to obtain the fractal deviation code ΔF for each granularity level. i m. The curvature dimension coordinates k above. iThe grid cell corresponding to the fractal dimension coordinate m is assigned a comprehensive state vector containing curvature deviation tensor and fractal deviation encoding. After all sampling points have been traversed, multiple state vectors within the same grid cell are statistically aggregated to generate the final state field value of that grid cell. The resulting cross-scale contour mapping state field systematically describes the overall distribution of constraint deviations between the geometry of the copper clad laminate edge contour and the geometry of the lamination process reference contour in two independent yet complementary observation dimensions: curvature bending characteristics and fractal roughness characteristics, on the observation grid jointly constructed by the curvature and fractal dimensions.

[0068] Step S300: Based on the distribution pattern of the contour morphology constraint deviation in the cross-scale contour mapping state field, perform contour pattern decomposition at the gradient flow level on the cross-scale contour mapping state field, separate the geometric instability component of the copper-clad laminate edge contour from the geometric constraint component of the cross-scale contour mapping state field, and generate the intrinsic contour pattern distribution manifold. The intrinsic contour pattern distribution manifold includes the contour pattern tensor in the region where the contour morphology constraint deviation is not significant and the contour pattern tensor in the region where the contour morphology constraint deviation changes abruptly.

[0069] The distribution pattern of contour morphology constraint deviations is a spatial distribution pattern of field values, such as high and low fluctuations, continuity or discontinuity, smoothness or abrupt changes, exhibited by the cross-scale contour mapping state field on its observation grid. Contour pattern decomposition at the gradient flow level is a field source separation technique based on gradient flow field analysis. It calculates the spatial gradient flow direction field of the scalar function of the state field on the observation grid and decomposes the entire state field into two main parts based on the divergence properties of the gradient flow: a geometric constraint component and a geometric instability component. The geometric constraint component corresponds to a stable region with low gradient flow divergence. The contour morphology constraint deviations within this region maintain a stable distribution along the tangential direction of the gradient flow, indicating that the copper-clad laminate edge contour morphology in this region is effectively controlled by the lamination mold, and the deviations are in a constrained and stable state. The geometric instability component corresponds to a divergent or convergent region with high gradient flow divergence. The contour morphology constraint deviations within this region exhibit obvious local divergent or convergent distributions along the normal direction of the gradient flow, indicating that the copper-clad laminate edge contour morphology in this region exhibits free deformation or abnormal flow trends unconstrained by the reference mold during the lamination process. The intrinsic contour pattern distribution manifold is a manifold data structure that decomposes the above two components and further refines the regions where each component is located into a manifold data structure with contour position as index and contour pattern tensor as expression form. It integrates the constrained deformation pattern description of constrained regions and the free deformation pattern description of abrupt regions at the global contour position.

[0070] In an optional embodiment, step S300 may specifically include the following steps S310 to S360: Step S310: On the observation grid composed of the curvature response scale and the fractal feature scale, calculate the spatial gradient flow direction field of the constraint deviation scalar function of the cross-scale contour mapping state field. The deviation change direction along the curvature dimension and the fractal dimension is defined at each grid point of the spatial gradient flow direction field.

[0071] The constraint deviation scalar function is a scalar function S(k, m) extracted from the comprehensive state vector of the cross-scale profile mapping state field. It measures the comprehensive constraint deviation strength of the copper-clad laminate edge profile relative to the lamination process reference profile under the observation conditions identified by the curvature dimension coordinate k and the fractal dimension coordinate m. One way to construct this scalar function is to take the weighted fusion value of the curvature deviation tensor component and the fractal deviation encoded component in the state vector. The spatial gradient flow direction field is a vector field obtained at each discrete grid point of the observation grid by calculating the partial derivatives of the constraint deviation scalar function in the curvature dimension and the fractal dimension. Its vector direction points to the direction in which the scalar function value increases the fastest at that point, and the vector magnitude reflects the intensity of change in that direction.

[0072] For example, suppose the constraint deviation scalar function value at coordinate (k, m) on the observation grid is S(k, m). The step size in the curvature dimension is Δk, and the step size in the fractal dimension is Δm. At grid point (k, m), the average of the forward and backward differences is first calculated along the curvature dimension to obtain the approximate partial derivative of the central difference in the curvature dimension. Similarly, the central difference approximate partial derivatives are calculated along the fractal dimension. Therefore, the spatial gradient flow direction vector at this grid point is obtained. The direction of this vector indicates the path of the most drastic change in local constraint deviation at that point. Its components along the curvature dimension and along the fractal dimension describe the driving forces of the deviation changes at the curvature response level and the fractal roughness level, respectively. The same calculation is performed on all grid points across the observation grid to generate the spatial gradient flow direction field.

[0073] Step S320: Extract the geometric constraint components of the cross-scale contour mapping state field based on the spatial gradient flow direction field. The geometric constraint components are composed of regions where the gradient flow divergence is lower than the preset divergence limit, and the contour shape constraint deviation within these regions remains stably distributed along the tangential direction of the gradient flow.

[0074] Gradient flow divergence is the divergence of the scalar field gradient, corresponding to the divergence of the spatial gradient flow direction field G on a discrete grid. Its physical meaning is to measure the degree of sourcelessness at which the gradient flow at that point is divergent or convergent. Within the stable region corresponding to the geometric constraint components, the gradient flow with constraint deviations typically exhibits nearly parallel or weakly divergent characteristics, with relatively small divergence values. The preset divergence limit is a pre-defined numerical threshold used to divide the gradient flow divergence into high-divergence and low-divergence regions.

[0075] Specifically, at each grid point of the observation grid, the divergence value Div(k,m) is calculated based on the spatial gradient flow direction field G. The discrete calculation method for the divergence is: Div(k, m)≈[G k (k+Δk, m)-G k (k-Δk, m)] / (2Δk)+[G m (k, m+Δm)-G m (k, m-Δm)] / (2Δm), where G k and G m Let G represent the components of G in the curvature and fractal dimensions, respectively. The divergence value Div(k, m) is compared with the preset divergence bound θ. div For comparison, those with divergence values ​​lower than θ div The grid points are marked as candidate points for geometric constraint components. Based on this, the connectivity of these candidate points is checked, requiring that the contour morphology constraint deviation within the region they form maintains a stable distribution along the gradient flow tangential direction; that is, when moving along a tangential direction perpendicular to the gradient direction within the region, the fluctuation amplitude of the scalar function S does not exceed a preset stable fluctuation threshold. The contiguous regions that satisfy the above two conditions are finally extracted as geometric constraint components.

[0076] Step S330: Extract the geometric instability component of the cross-scale contour mapping state field based on the spatial gradient flow direction field. The geometric instability component is composed of regions where the gradient flow divergence is higher than the preset divergence limit, and the contour shape constraint deviation inside the region shows a local divergent or convergent distribution along the normal direction of the gradient flow.

[0077] Corresponding to step S320, the geometric instability component is another type of component separated from the cross-scale profile mapping state field. The gradient flow divergence Div(k, m) in the observation grid is set above a preset divergence limit θ. div The grid points were identified as candidate points for geometrically unstable components. Further, the field value distribution in these grid point regions was analyzed along the gradient flow normal direction, observing the cross-sectional distribution of the deviation scalar function S along the normal direction. If the cross-sectional distribution exhibited obvious local peaks decreasing towards both sides (divergent) or local valleys increasing towards both sides (convergent), then the region was confirmed to correspond to a geometrically unstable component. The spatial location of geometrically unstable components is concentrated in anomalous regions where constraint deviations change drastically and the profile deviates from the bonding process reference constraint framework.

[0078] Step S340: Perform topological connectivity analysis on the geometric constraint components, extract the connected component skeleton of the geometric constraint components, and map the connected component skeleton to the spatial position of the edge contour to generate the contour pattern tensor of the constraint region. The contour pattern tensor of the constraint region describes the constrained deformation mode in the region where the contour shape constraint deviation is not significant.

[0079] Topological connectivity analysis involves segmenting all grid points occupied by geometrically constrained components on the observation grid according to their adjacency relationships, identifying the geometric extension range and shape of each connected region. The connected region skeleton is a thin, linear structure obtained after morphological refinement of each connected region, preserving its topological connectivity properties and geometric extension direction. The contour pattern tensor is a multidimensional data tensor stored using the spatial location of the edge contour as an index. In this embodiment, the contour pattern tensor of the constrained region stores the joint cross-dimensional morphological description of the skeleton branches in the curvature and fractal dimensions, as well as the local morphological transition tensor at the nodes, used to characterize the constrained deformation mode of the copper-clad laminate edge contour at that location, influenced by the lamination process reference constraints.

[0080] In an optional embodiment, step S340 may specifically include the following steps S341 to S346: Step S341: Binary labeling is performed on the observation grid area where the geometric constraint components are located, and a binary distribution map of the constraint area is generated. Grid points belonging to the geometric constraint components in the binary distribution map of the constraint area are assigned activation labels.

[0081] Binary labeling is an operation that transforms different field properties on an observation grid into a logical image containing only two states. In this step, all grid points of the entire observation grid are traversed. For each grid point, if it is determined to belong to a geometric constraint component in step S320, its binary label value is set to "1" (activated); if it does not belong to a geometric constraint component, its binary label value is set to "0" (deactivated). This generates a binary distribution map of the constraint region, consistent with the spatial resolution of the observation grid and composed of activated and deactivated elements. This distribution map visually highlights the stable constraint regions in the continuous field in a set form.

[0082] Step S342: Perform morphological refinement on the binary distribution map of the constrained region, peeling away the boundary grid points layer by layer until a connected domain skeleton segment with a single grid width is obtained. The connected domain skeleton segment maintains the topological connectivity and geometric extension direction of the original constrained region.

[0083] Morphological thinning is based on hit-and-miss transformation or lookup table methods. In each iteration, the local neighborhood pattern of each active pixel in the binary image is judged. Pixels located at the region boundary that will not disrupt the region's connectivity are changed from an active state to an inactive state. This process is repeated layer by layer until the region is reduced to a skeleton line of single-pixel width. In this embodiment, the Zhang-Suen fast parallel thinning algorithm is used to process the binary distribution map of the constrained region. This algorithm divides each iteration cycle into two sub-stages. In each sub-stage, a set of conditions is checked for the 8-neighborhood configuration of the active pixel. Boundary pixels that meet the conditions are marked as to be deleted, and are deleted uniformly after the sub-stage ends. Iteration continues until no more pixels can be deleted. The remaining active pixels constitute the connected domain skeleton line segment. This skeleton line segment maintains the branch topology of the original constrained region and the curvature-fractal extension path within the observation grid.

[0084] Step S343: Perform node-branch topology analysis on the skeleton line segments of the connected domain, extract the branch node positions and end node positions in the skeleton line segments, and record the curvature dimension coordinate sequence and fractal dimension coordinate sequence of the skeleton branches between adjacent nodes in the observation grid.

[0085] Node-branch topology analysis transforms skeleton segments from sets of pixels into graph-theoretic topological structures. A branch node is a pixel in a skeleton segment that is adjacent to three or more skeleton pixels, representing a fork in the skeleton. An end node is a pixel in a skeleton segment that is adjacent to only one skeleton pixel, representing the end point of the skeleton line. A skeleton branch is an unbranched skeleton path connecting two nodes (between branch nodes or between a branch node and an end node). The curvature dimension coordinate sequence is the sequence of curvature dimension k-coordinates of each skeleton pixel traversed sequentially along the skeleton branch in the observation grid, and the fractal dimension coordinate sequence is the sequence of fractal dimension m-coordinates of the corresponding skeleton pixels.

[0086] For example, a connected component labeling algorithm is used to scan the skeleton segments, locating the grid coordinates of all branch nodes and end nodes. Then, starting from each end node, the skeleton path is traced until the first branch node is encountered, recording all grid points along the path; these grid points constitute a skeleton branch. For each skeleton branch, the curvature dimension coordinate values ​​of the grid points along the path are extracted and arranged according to the tracing order to obtain a curvature dimension coordinate sequence; similarly, a fractal dimension coordinate sequence is obtained. These two coordinate sequences together describe the precise geometric path of the skeleton branch in the contour pattern observation space.

[0087] Step S344: Convert the curvature dimension coordinate sequence of the skeleton branch into curvature description code, and convert the fractal dimension coordinate sequence of the skeleton branch into fractal scale description code to form a joint cross-dimensional morphological description of the skeleton branch.

[0088] The curvature description code is a set of descriptive codes generated based on the curvature dimension coordinate sequence of the skeleton branch. Specifically, it is generated by mapping each coordinate value in the curvature dimension coordinate sequence back to the curvature response coding level identifier defined in step S215, thus obtaining the curvature description code corresponding to that coordinate value. The fractal scale description code is generated by mapping each coordinate value in the fractal dimension coordinate sequence of the skeleton branch back to the joint representation of the observation granularity level index and the local contour roughness code defined in step S246. The curvature description code corresponding to each point on the same skeleton branch is paired with the fractal scale description code to form a joint cross-dimensional morphological description of that skeleton branch. This description accurately records the morphological feature changes of the constrained region along the skeleton path within the contour pattern observation space.

[0089] Step S345: Propagate and aggregate the joint cross-dimensional morphological description of the skeleton branch along the connected path of the node-branch topology, and generate a local morphological transition tensor at each branch node based on the difference in morphological description between the in-degree branch and the out-degree branch.

[0090] Propagation aggregation is the process of transmitting morphological description information branch by branch from the terminal node along the skeleton branches toward the branch nodes. At each branch node, the joint cross-dimensional morphological description of all in-degree branches pointing to that node is aggregated, as well as the joint cross-dimensional morphological description of all out-degree branches branching outward from that node. The local morphological transition tensor is a tensor defined at the branch node location that quantifies the transformation characteristics of the controlled deformation mode of the contour morphology at that branch node by comparing the degree of difference between the in-degree and out-degree branches in curvature description encoding and fractal scale description encoding.

[0091] In this embodiment, for a given branch node, the average code of the joint cross-dimensional morphological description of all its in-degree branches is taken as the in-degree representative, and the average code of the joint cross-dimensional morphological description of all its out-degree branches is taken as the out-degree representative. The difference vectors between the in-degree and out-degree representatives in the curvature description coding component and the fractal scale description coding component are calculated. These two difference vectors, along with the grid coordinates of the branch node, are packaged to form the local morphological transition tensor at that branch node.

[0092] Step S346: Bind the local shape transition tensor to the corresponding edge contour position index to generate the contour pattern tensor of the constrained region. The contour pattern tensor of the constrained region stores the description of the constrained deformation mode at the contour position as the index.

[0093] The local morphological transition tensor generated in step S345 is reverse-mapped to the specific copper-clad laminate edge contour spatial position index using the curvature dimension coordinate sequence and fractal dimension coordinate sequence recorded in step S343 for the skeleton branch where the node is located, combined with the correspondence between the observed grid and the normalized arc length position label in step S260. For local morphological transition tensors that can be mapped to a unique contour position index, a binding relationship is directly established; for those mapped to a contour position interval, the center position of the interval is selected for binding. The resulting contour pattern tensor of the constraint region is a data structure with the edge contour position as the key and the local morphological transition tensor as the value. It systematically stores the spatial distribution information of the constrained deformation modes in all spatial regions identified as geometric constraint components.

[0094] Step S350: Detect local extrema of the geometrically unstable component, extract the spatial distribution of local scalar extrema in the geometrically unstable component, and map the distribution of local scalar extrema to the spatial position of the edge contour to generate the contour pattern tensor of the abrupt region. The contour pattern tensor of the abrupt region describes the free deformation mode in the abrupt region of the contour shape constraint deviation.

[0095] Local extremum detection is the process of finding grid points in the observation grid region where the geometrically unstable component is located where the constraint deviation scalar function S(k, m) achieves a local maximum or minimum value. These scalar extremum points represent the locations where constraint deviation mutations are most concentrated and are the core sites where free deformation modes occur. The profile mode tensor of the mutation region is structurally isomorphic to the profile mode tensor of the constraint region, but its stored content is a description of the free deformation trend extracted from the scalar extremum points, used to characterize the free deformation mode generated when the copper-clad laminate edge profile deviates from the reference constraint during the lamination process.

[0096] In an optional embodiment, step S350 may specifically include the following steps S351 to S356: Step S351: Within the observation grid region where the geometric instability component is located, calculate the Hessian response of the constraint deviation scalar function at each grid point. The Hessian response is used to measure the principal curvature intensity of the local surface at that grid point.

[0097] The Hessian response is a scalar index constructed based on the second-order partial derivative matrix (i.e., the Hessian matrix) of the constraint deviation scalar function S(k, m). The Hessian matrix H is a 2×2 symmetric matrix whose elements contain the second-order partial derivatives of S with respect to k. The second-order partial derivative of S with respect to m and mixed partial derivatives One way to construct the Hessian response is to take the absolute value of the sum of the eigenvalues ​​of the Hessian matrix or the absolute value of the determinant of the Hessian matrix, which reflects the total measure of the bending intensity of the local surface.

[0098] In this embodiment, for each grid point (k, m) within the geometrically unstable component region, the second-order partial derivatives and mixed partial derivatives are calculated using the central difference scheme: , , After constructing the Hessian matrix H, its two eigenvalues ​​λ1 and λ2 are calculated, and then the Hessian response R is obtained. H =|λ1|+|λ2|. The larger the Hessian response of a grid point, the more severe the curvature of its local constraint deviation surface, and the greater the probability that it is a potential extreme point.

[0099] Step S352: Based on the local peak position of the Hessian response, determine the coordinates of the scalar extreme point of the profile morphology in the geometric instability component. The coordinates of the scalar extreme point include curvature dimension coordinates and fractal dimension coordinates.

[0100] Local peak location identification involves searching for grid points in the Hessian response field whose Hessian response values ​​are greater than those of all their neighboring grid points. In this embodiment, a 3×3 window local nonmaximum suppression method is used: each grid point in the geometric instability component is sequentially traversed, and its Hessian response value is compared with the Hessian response values ​​of all grid points in its 8-neighborhood. If the value is the maximum within the window, the grid point is identified as a scalar extremum point of the profile morphology. The coordinates of this scalar extremum point are determined by the curvature dimension coordinate value k of the grid point in the observation grid. ext and fractal coordinates m ext Together they form a whole.

[0101] Step S353: For each scalar extreme point, extract the constraint deviation scalar distribution within the local neighborhood window centered on it, and construct the deformation driving direction vector of the extreme point based on the principal direction change rate of the scalar values ​​within the local neighborhood window.

[0102] A local neighborhood window is a rectangular or circular region within the observation grid, centered on a scalar extremum point, used to capture details of the deviation distribution around that extremum point. The principal direction rate of change is the rate of change of the constraint deviation scalar function S along two principal orthogonal directions within this local neighborhood window, typically obtained by performing principal component analysis or calculating the principal directions of the gradient direction histogram within the window. The deformation-driving direction vector is a two-dimensional vector pointing in the direction of the most dramatic deviation change at the extremum point; its magnitude reflects the intensity of the change.

[0103] For example, for the scalar extremum point (k) ext , m exFor example, extract the S-values ​​within a 5×5 grid neighborhood window centered at the point to form a sample matrix. After weighted centering of the grid point coordinates in this sample matrix, calculate the sample covariance matrix and solve for its eigenvectors. Take the eigenvector corresponding to the largest eigenvalue as the principal direction and calculate the rate of change of the S-value along this principal direction, thus obtaining the principal direction rate of change. Multiply the unit vector of the principal direction by the principal direction rate of change to obtain the deformation driving direction vector at the extreme point.

[0104] Step S354: Combine the deformation driving direction vector with the corresponding extreme point coordinates to obtain the free deformation trend description of the extreme point. The free deformation trend description indicates the tendency of the geometric instability component at this location to expand outward or contract inward.

[0105] The free deformation trend description is a structured descriptive unit containing spatial location and directional intensity information. It associates the signs of the curvature dimension and fractal dimension components of the deformation-driving direction vector with two trends: "expansion" and "contraction," respectively. If the curvature dimension component of the deformation-driving direction vector is positive, it indicates that the constraint deviation at that point exhibits an expansion trend in the direction of increasing curvature; if it is negative, it indicates a contraction trend in the direction of decreasing curvature. The fractal dimension component is interpreted similarly. This free deformation trend description fully records the free deformation motion tendency at a single abrupt change center point.

[0106] Step S355: Sort the free deformation trend descriptions of all extreme points according to the curvature dimension coordinates of the extreme points, and aggregate the distribution histograms along the fractal dimension to obtain the free deformation trend distribution sequence based on the curvature dimension coordinates.

[0107] All free deformation trend descriptions are sorted in ascending order of curvature coordinate values ​​at extreme points, ensuring an orderly arrangement of descriptions along the curvature dimension. Histogram aggregation along the fractal dimension involves grouping and statistically analyzing the free deformation trend descriptions along the fractal dimension using a sliding window or fixed statistical bin, calculating the number of free deformation trend descriptions and the average trend direction within each fractal dimension interval. The free deformation trend distribution sequence sorted by curvature coordinates is a multidimensional sequence that integrates the sorting and aggregation results, revealing the overall distribution pattern of free deformation trends in the geometric instability components along both the curvature and fractal dimensions.

[0108] Step S356: Map each free deformation trend description in the free deformation trend distribution sequence to the edge contour sampling point location index corresponding to the extreme point, generate the contour pattern tensor of the abrupt region, and store the free deformation pattern description at the edge contour location as the index of the abrupt region contour pattern tensor.

[0109] Similar to the mapping mechanism in step S346, the observed grid coordinates of each extreme point are mapped inversely to the normalized arc length position label of the copper-clad laminate edge contour through the correlation established during the cross-scale contour mapping state field establishment process, thereby determining the corresponding edge contour sampling point position index. The free deformation trend description of each extreme point is bound to this position index. For multiple extreme points mapped to the same position index, the average vector of their free deformation trend descriptions is taken as the comprehensive free deformation mode description for that position. The final contour mode tensor of the abrupt change region is a set of free deformation mode descriptions stored with the edge contour position as the index in all regions where geometric instability components are significant.

[0110] Step S360: The contour pattern tensor of the constrained region and the contour pattern tensor of the abrupt region are joined along the edge contour position index to obtain the intrinsic contour pattern distribution manifold. The intrinsic contour pattern distribution manifold integrates the distribution of constrained deformation modes and free deformation modes at the global contour position.

[0111] Manifold splicing is an operation that arranges and merges two sets of contour pattern tensors, derived from geometric constraints and geometric instabilities respectively, into a complete manifold data structure according to a unified position indexing system. For each position index on the copper clad laminate edge contour, the existence of a valid record in the contour pattern tensor of the abrupt region is checked first. If it exists, the pattern at that position is marked as a free deformation mode, and the corresponding free deformation mode description is filled in. If it does not exist, the corresponding constrained deformation mode description is retrieved from the contour pattern tensor of the constrained region and filled in. For positions where neither tensor has a record, the default stationary mode description is filled in. The intrinsic contour pattern distribution manifold formed by this splicing is a one-dimensional manifold structure that is continuously distributed along the edge contour position index, storing the corresponding contour pattern tensor (either a constrained mode or an abrupt mode) at each position node. It provides a native, intrinsic pattern decomposition expression for the deviation between the geometric shape of the copper clad laminate edge contour and the lamination process reference contour.

[0112] Step S400: Perform contour pattern perturbation evolution analysis on the intrinsic contour pattern distribution manifold along the time series. Based on the spatial adjacency relationship and temporal propagation trajectory of the contour pattern tensor, measure the trend mutual induction effect between continuous pressing batches of the contour pattern perturbation component corresponding to the latent anomaly of interlayer slip and the latent anomaly of resin wetting unevenness in the intrinsic contour pattern distribution manifold, and generate a trend map of mutual induction evolution of contour anomaly perturbation.

[0113] The time series here is a set of intrinsic profile pattern distribution manifolds arranged in chronological order of lamination batches. Spatial adjacency is the topological association between two adjacent profile pattern tensor nodes along the profile position index within the intrinsic profile pattern distribution manifold of the same lamination batch, reflecting the spatial coupling structure of perturbations on the edge profile of a single product. The temporal propagation trajectory is the evolution path of the profile pattern tensor at the same profile position index across different lamination batches, reflecting the cumulative change of perturbations over time or batches. The interlayer slip latent anomaly is an invisible, minute relative displacement trend between copper clad laminate layers during lamination due to uneven material flow or insufficient interfacial bonding. Its corresponding profile pattern perturbation component exhibits a profile tangential deformation mode with specific directional characteristics. The resin impregnation uneven latent anomaly is a local enrichment or depletion defect caused by the resin failing to uniformly impregnate every area of ​​the reinforcing material during lamination. Its corresponding profile pattern perturbation component exhibits a profile normal expansion and contraction deformation mode with specific directional characteristics. The trend-mutually induced effect is a phenomenon in which changes in the trend of one disturbance component guide or amplify changes in the trend of the other component in a directional manner during continuous batch evolution. The profile anomaly disturbance mutual-induced evolution trend map is a comprehensive visualization data structure that integrates the above-mentioned temporal, spatial, and disturbance type information. It presents the evolution of the induced correlation strength between latent anomalies of interlayer slip and latent anomalies of resin wetting unevenness in the form of spatiotemporal thermal distribution and directional propagation vectors.

[0114] In an optional embodiment, step S400 may specifically include the following steps S410 to S460: Step S410: Obtain the intrinsic profile pattern distribution manifold sequence corresponding to multiple consecutive pressing batches. Each manifold element in the intrinsic profile pattern distribution manifold sequence corresponds to the profile pattern tensor set of a pressing batch under the unified edge profile position index.

[0115] A continuous lamination batch is a sequence of copper-clad laminated products manufactured sequentially according to time. Each batch corresponds to one or a group of copper-clad laminates laminated under the same process parameter settings. The intrinsic profile pattern distribution manifold sequence is a sequence formed by arranging the intrinsic profile pattern distribution manifolds obtained after processing each batch in step S300 according to the batch sequence. To achieve comparability between batches, the intrinsic profile pattern distribution manifolds of all batches adopt a unified aligned edge profile position indexing system. That is, for all batches, the same position index value points to the same geometrically corresponding position area on the edge profile of the copper-clad laminate. This can be completed in the preprocessing stage through a profile registration algorithm (such as linear mapping registration based on arc length ratio).

[0116] Step S420: Perform spatial adjacency modeling on the intrinsic contour pattern distribution manifold sequence. Construct a spatial adjacency graph of the contour pattern tensor within each compression batch. Adjacent nodes in the spatial adjacency graph are defined by tensor pairs that are adjacent along the contour position index and whose tensor similarity exceeds a preset similarity threshold.

[0117] Spatial adjacency modeling is the process of establishing a graph structure describing the coupling relationships between different contour locations within the same batch, based on the physical proximity of contour location indices and the feature similarity of tensor content. A spatial adjacency graph is an undirected or directed graph where nodes are contour pattern tensors at each contour location index, and edges connect node pairs that satisfy adjacency conditions. Tensor similarity is a measure of the similarity between two contour pattern tensors in constrained or free deformation modes, and can be calculated on the flattened tensor vector using distance metrics such as Pearson correlation coefficient and cosine similarity. A preset similarity threshold is an empirical threshold used to filter out weak connections with excessively low similarity.

[0118] In this embodiment, for the intrinsic profile pattern distribution manifold of the b-th pressing batch, all adjacent profile position index pairs (i, i+1) are traversed. The profile pattern tensor T at position i is... i and the contour pattern tensor T at position i+1 i+1 Flattened into vectors v i and v i+1 Calculate the cosine similarity cos(v) i , v i+1 If the cosine similarity is higher than the preset similarity threshold θ sim Then, an edge is added between node i and node i+1 in the spatial adjacency graph, and the edge weight is recorded as the similarity value. Thus, each batch generates an irregular graph reflecting its spatial association structure of contour patterns.

[0119] Step S430: Project the perturbation type onto the contour pattern tensor corresponding to each node in the spatial adjacency graph. Project the tensor components associated with the interlayer slip deformation mode onto the interlayer slip perturbation channel, and project the tensor components associated with the resin impregnation uneven deformation mode onto the resin impregnation uneven perturbation channel to obtain the perturbation components of each node in the interlayer slip perturbation channel and the perturbation components in the resin impregnation uneven perturbation channel.

[0120] Perturbation type projection is an operation that decomposes a high-dimensional profile pattern tensor and assigns it to multiple perturbation channels with explicit physical meaning. The interlayer slip perturbation channel is specifically designed to carry deformation information related to interlayer slip deformation modes, while the resin wetting unevenness perturbation channel is specifically designed to carry deformation information related to resin wetting unevenness deformation modes. The interlayer slip deformation direction template is a predefined set of unit vectors or orthogonal bases characterizing typical interlayer slip deformation directions, with its direction typically dominated along the profile tangent. The resin wetting unevenness deformation direction template is an orthogonal direction template, with its direction typically dominated along the profile normal.

[0121] In an optional embodiment, step S430 may specifically include the following steps S431 to S436: Step S431: Extract the principal direction tensor component and deformation amplitude tensor component of the contour deformation from the contour pattern tensor. The principal direction tensor component indicates the distribution of deformation in the contour tangent and contour normal, and the deformation amplitude tensor component indicates the deformation intensity.

[0122] The contour pattern tensor can be resolved into two interrelated components: the principal direction tensor component and the deformation amplitude tensor component. The principal direction tensor component is a unit vector or direction matrix that indicates the orientation of the dominant deformation direction at the contour location within the local coordinate system formed by the local tangent and normal directions of the contour. The deformation amplitude tensor component is a scalar or vector that quantifies the energy or intensity of the deformation along the principal direction. In this embodiment, for a given contour pattern tensor, if it belongs to a constrained region, its principal direction tensor component can be determined by the direction of the difference vector of the local morphological transition tensor, while the deformation amplitude tensor component is determined by the magnitude of the difference vector. If it belongs to abruptly changed regions, its principal direction tensor component is determined by the direction of the deformation driving direction vector, while the deformation amplitude tensor component is determined by the magnitude of the deformation driving direction vector.

[0123] Step S432: Match the principal direction tensor component with the preset interlayer slip deformation direction template, calculate the projection amplitude of the principal direction tensor component in the interlayer slip deformation direction, and generate the projection intensity of the interlayer slip deformation direction.

[0124] The preset interlayer slip deformation direction template is one or more fixed direction reference vectors defined in the local coordinate system of the profile. In actual implementation, this template can be predetermined based on the mechanical theory of interlayer slip in copper-clad laminates, and is typically a unit vector along the tangent direction of the profile. Direction matching is the process of performing a dot product operation between the actual deformation direction indicated by the principal direction tensor components and the template vector. The projection intensity of the interlayer slip deformation direction is the absolute or square value of this dot product, reflecting the degree of fit between the actual deformation direction and the ideal interlayer slip direction. The higher the projection amplitude, the more the deformation mode at that location conforms to the characteristics of interlayer slip.

[0125] Step S433: Jointly encode the deformation amplitude tensor component and the projection intensity of the interlayer slip deformation direction as the perturbation component of the profile mode tensor in the interlayer slip perturbation channel. The perturbation component of the interlayer slip perturbation channel carries both deformation direction conformity and deformation intensity information.

[0126] Joint encoding is the operation of combining the scalar values ​​of the deformation amplitude tensor components with the scalar values ​​of the projection intensity of the interlayer slip deformation direction into a two-dimensional vector or composite scalar. For example, a simple joint encoding method is to directly take the product of the two as the perturbation component value of the interlayer slip perturbation channel, i.e., P. slip =Deformation amplitude × Projected intensity of interlayer slip deformation direction. This value is high when the deformation direction is highly consistent with the interlayer slip direction and the deformation intensity is large, and low when the direction is inconsistent or the deformation intensity is small, thus faithfully condensing the tendency information of interlayer slip anomaly in the profile pattern tensor into the single-channel perturbation component.

[0127] Step S434: Match the principal direction tensor component with the preset resin impregnation uneven deformation direction template, calculate the projection amplitude of the principal direction tensor component in the resin impregnation uneven deformation direction, generate the resin impregnation uneven deformation direction projection intensity, and the resin impregnation uneven deformation direction template and the interlayer slip deformation direction template have an orthogonal constraint relationship.

[0128] The preset resin wetting uneven deformation direction template is another unit vector reference template orthogonal to the interlayer slip deformation direction template, typically set along the contour normal. Since the copper clad laminate edge contour usually exhibits outward bulging or inward contraction normal deformation during resin wetting unevenness, this direction template maintains a strict orthogonal constraint relationship with the interlayer slip tangential deformation template. Similar to step S432, the dot product of the principal direction tensor component vector and the resin wetting uneven deformation direction template vector is calculated, and its absolute value or square value is taken as the resin wetting uneven deformation direction projection intensity. This value reflects the magnitude of the actual deformation component in the resin wetting uneven deformation direction.

[0129] Step S435: Jointly encode the deformation amplitude tensor component and the projection intensity of the resin impregnation uneven deformation direction as the perturbation component of the contour mode tensor in the resin impregnation uneven perturbation channel. The perturbation component of the resin impregnation uneven perturbation channel carries both deformation direction conformity and deformation intensity information.

[0130] Similar to step S433, the deformation amplitude tensor component is multiplied or weighted and summed with the projected intensity of the resin impregnation uneven deformation direction to obtain the disturbance component value P of the resin impregnation uneven disturbance channel. resin= Deformation amplitude × Projected intensity of resin impregnation uneven deformation direction. The physical meaning of this component is a comprehensive measure of the degree and intensity to which the edge profile of the copper clad laminate at this location matches the abnormal deformation characteristics of resin impregnation unevenness.

[0131] Step S436: Perform orthogonal decomposition verification on the disturbance components of the interlayer slip disturbance channel and the resin impregnation uneven disturbance channel at the same contour position node, adjust the ratio of the disturbance components of the two channels so that there is no redundant directional projection overlap, and obtain the verified disturbance components.

[0132] Orthogonal decomposition verification is an operation that checks the orthogonality and eliminates redundancy of the perturbation components of the two channels obtained in steps S433 and S435. Since the direction of the principal direction tensor component may not be completely parallel to the interlayer slip direction template or the resin wetting unevenness direction template, but rather lies in between, the projections of the two channels may contain overlapping components due to incomplete orthogonal decomposition of directions. In this embodiment, the two projection intensity values ​​are calculated, and their sum of squares is checked to see if it significantly exceeds 1 (when the principal direction is a unit vector). If redundancy exists, the two projection coefficients are adjusted according to the Gram-Schmidt orthogonalization method, forcing the two components to be orthogonal and the sum of the squares of their magnitudes equal to the square of the original principal direction magnitude. The adjusted P... slip and P resin This refers to the verified perturbation component, ensuring that the information from the two perturbation channels is strictly complementary and without redundant overlap in subsequent analysis.

[0133] Step S440: Along the time dimension of the distribution manifold sequence of the intrinsic profile pattern, track the change trajectory of the disturbance component of the same profile position node between different pressing batches to obtain the temporal propagation trajectory of the interlayer slip disturbance component and the temporal propagation trajectory of the resin impregnation unevenness disturbance component.

[0134] The time dimension refers to the sequence of lamination batches. For a given contour location index i, from batch 1 to batch B, the verified disturbance components of the interlayer slip disturbance channel at that location node are extracted sequentially. These component values ​​are arranged in batch order, thus forming the temporal propagation trajectory of the interlayer slip disturbance components. Similarly, the verified disturbance components of the resin impregnation unevenness disturbance channel are extracted sequentially and arranged in batch order, forming the temporal propagation trajectory of the resin impregnation unevenness disturbance components. These two one-dimensional time series describe the evolution history of two latent anomalous disturbance modes at the same spatial location with respect to process batches.

[0135] Step S450: Perform trend mutual induction detection on the temporal propagation trajectory of the interlayer slip disturbance component and the temporal propagation trajectory of the resin wetting unevenness disturbance component. When the change direction of the interlayer slip disturbance component in one batch causes the resin wetting unevenness disturbance component to change in the same direction in subsequent batches, record the induced correlation strength of the batch and its contour position.

[0136] Trend mutual induction detection is a time-delay-based causal relationship screening algorithm that aims to capture the guiding role of interlayer slip perturbation trends on resin wetting unevenness perturbation trends. The direction of change refers to the increase or decrease of the perturbation component between adjacent batches, categorized as positive trends (increase) and negative trends (decrease). Same-direction change occurs when, within a defined time delay window, the direction of change of the resin wetting unevenness perturbation component is consistent with the direction of change of the interlayer slip perturbation component. The strength of the induced correlation is a comprehensive indicator quantifying the significance and strength of this induced relationship.

[0137] In an optional embodiment, step S450 may specifically include the following steps S451 to S456: Step S451: Differentiate the temporal propagation trajectories of interlayer slip disturbance components at the same contour location to generate a batch-to-batch change direction sequence of interlayer slip disturbance components. The change direction sequence includes positive trend markers for increasing disturbance components and negative trend markers for decreasing disturbance components.

[0138] Differential operation is the process of processing the interlayer slip perturbation component sequence. Perform a first-order backward difference. b ranges from 1 to B-1. For the difference result, if its value is positive and its absolute value exceeds the preset minimum change threshold, the change direction of this batch is marked as a positive trend, such as coded as "+1"; if its value is negative and its absolute value exceeds the threshold, it is marked as a negative trend, such as coded as "-1"; if the change amplitude is less than the threshold, it is marked as no change trend, coded as "0". This generates a change direction sequence of length B-1.

[0139] Step S452: Differentiate the temporal propagation trajectory of the resin impregnation unevenness disturbance component at the same contour position to generate a batch-to-batch variation direction sequence of the resin impregnation unevenness disturbance component.

[0140] Using the exact same difference rule and threshold parameters as in step S451, the resin wetting unevenness perturbation component sequence is processed. Perform a first-order backward difference and transform the difference result into a sequence of change directions consisting of "+1", "-1", and "0". The two sets of change direction sequences are perfectly aligned on the time base.

[0141] Step S453: Within a preset time delay window, detect whether there is a direction consistency event between the batch-to-batch change direction sequence of the interlayer slip disturbance component and the batch-to-batch change direction sequence of the resin wetting unevenness disturbance component. When the positive trend marker of the interlayer slip disturbance component and the positive trend marker of the resin wetting unevenness disturbance component appear correspondingly within the delay window, it is determined that a positive induction event has occurred.

[0142] The preset time delay window is a positive integer time offset in units of batch size. Its physical meaning is that it allows the trend change of interlayer slip disturbance to pass through Only after several batches did a response occur regarding the tendency of uneven resin impregnation. For a specific batch b as the inducing batch, the detection in batch b... At this point, it is determined whether the direction markers for interlayer slip change and resin wetting unevenness change are both positive trend markers. If both are positive trend markers, then it is determined that within the delay window... Below, from batch b to batch A positive induction event has occurred. Similarly, a negative induction event can be defined.

[0143] Step S454: Count the cumulative number of positive induced events under the same contour position and the same delay window configuration, and combine the positive trend amplitude of the interlayer slip disturbance component and the response amplitude of the resin wetting unevenness disturbance component to generate the initial induced correlation intensity under the delay condition.

[0144] The positive trend amplitude is the absolute difference of the interlayer slip disturbance component at the inducing source batch, and the response amplitude is the absolute difference of the resin wetting unevenness disturbance component at the response batch. One way to construct the initial induced correlation intensity is to use the same contour position i and the same delay window. The sum of the products of the trend magnitude and the response magnitude of each positively induced event is multiplied by the total number of occurrences, and then divided by a normalization factor. The summation is performed only on batches b that are determined to be positive induction events. This value reflects the total energy of the induction effect of interlayer slip on resin wetting unevenness under a specific delay assumption.

[0145] Step S455: Perform initial induced association strength calculation on all possible delay windows in multiple batches of sequences, select the delay window that maximizes the initial induced association strength as the effective induced delay window at the contour position, and use the initial induced association strength corresponding to the effective induced delay window as the induced association strength at the contour position.

[0146] All possible delay windows are from To the preset maximum delay A set of candidate time offsets. For contour position i, iterate through each candidate delay. Calculate the initial induced correlation strength according to steps S453 and S454. In all In the middle, select such that Take the maximum value As an effective induced delay window The corresponding maximum initial induced correlation strength is the induced correlation strength C at that contour location. opt(i) This optimization mechanism adaptively captures the optimal response time of the mutual induced effect at different spatial locations.

[0147] Step S456: Arrange all contour locations and their corresponding induced correlation intensities along the contour location index, and store them together with the effective induced delay window and the batch identifier of the interlayer slip disturbance source to obtain the induced correlation intensity dataset used to generate the mutual induced evolution trend map of contour anomaly disturbances.

[0148] Record the induced association strength C of each contour position i in ascending order of edge contour position index. opt(i) Effectively induce delay window And the set of source batch identifiers that generate this induced association strength. This induced association strength dataset is the complete data source for constructing the visualization map, which contains three layers of information: spatial dimension (outline location), temporal dimension (source batch), and delay dimension (effective induced delay window), as well as the core strength measure.

[0149] Step S460: The induced correlation intensity between multiple batches at all contour locations is fused and plotted along both time and space dimensions to generate a contour anomaly perturbation mutual induced evolution trend map. The contour anomaly perturbation mutual induced evolution trend map contains the spatiotemporal thermal distribution and directional propagation vector of the induced correlation intensity.

[0150] Fusion mapping is the operation of transforming the induced correlation intensity dataset into an intuitive graphical representation. The spatiotemporal thermal distribution is a two-dimensional grid constructed with the contour location index as the horizontal axis and the lamination batch as the vertical axis. Color saturation or brightness is used to map the induced correlation intensity values ​​of each batch at each location, forming a heatmap characterizing the overall evolution. The directional propagation vector, on the contour geometric unfolding diagram, uses arrows to indicate the spatial direction of the induced propagation of interlayer slip disturbances to resin wetting unevenness disturbances at that location, and arrow length or thickness to represent the relative magnitude of the induced correlation intensity. Together, these elements constitute a graph of the mutual induced evolution trend of contour anomaly disturbances, allowing process engineers to clearly observe the multi-batch evolution trend and spatial hotspot distribution of latent anomalies in copper clad laminate lamination.

[0151] Step S500: When the contour pattern perturbation component corresponding to the interlayer slip latent anomaly and the contour pattern perturbation component corresponding to the resin impregnation uneven latent anomaly in the contour anomaly perturbation mutual induced evolution trend map show a synchronous monotonically increasing trend within a preset anomaly response time window and the slope convergence index of the trend increasing curve exceeds the preset trend coordination threshold, a copper clad laminate lamination latent anomaly prediction signal is generated.

[0152] The anomaly response time window is a pre-defined lower limit of the time span measured by the number of consecutive lamination batches, used to ensure that the monitored trend has sufficient persistence rather than random fluctuations. A synchronous monotonically increasing trend is defined within the batch intervals defined by the anomaly response time window, where the time evolution curves of the interlayer slip disturbance component and the resin wetting unevenness disturbance component simultaneously exhibit a non-decreasing characteristic with at least one strictly increasing characteristic. The increasing slope sequence is a sequence describing the relative magnitude of the increase in each batch interval. The slope convergence index is a comprehensive indicator quantifying the consistency in the rhythm of change between the two increasing slope sequences; a higher index indicates a more consistent evolutionary pace between the two latent anomalies and a more significant coupling and excitation relationship between the latent anomalies. The trend coordination threshold is a pre-defined threshold for judging the slope convergence index. When all the above conditions are simultaneously met, the system generates a latent anomaly prediction signal for copper clad laminate lamination, indicating that the production process contains latent defects that are about to appear or have already appeared, resulting in the amplification of interlayer slip and resin wetting unevenness coupled together.

[0153] In an optional embodiment, step S500 may specifically include the following steps S510 to S560: Step S510: Extract the time evolution curves of the interlayer slip disturbance component and the resin impregnation unevenness disturbance component from the contour anomaly disturbance mutual induced evolution trend map, and align the two time evolution curves to a unified pressing batch time axis.

[0154] The time evolution curve is a curve showing the change in the intensity of perturbation components at a certain contour position or region of interest, extracted from the batch-level data. In this embodiment, from the perspective of the whole or hot spot areas, the batch average of the interlayer slip perturbation components at all contour positions in the contour anomaly perturbation mutual induced evolution trend map can be taken to obtain the time evolution curve of the global interlayer slip perturbation components, with the batch number as the horizontal axis and the perturbation component average as the vertical axis; similarly, the time evolution curve of the global resin impregnation unevenness perturbation components can be obtained. The pressing batch time axis is a batch sequence axis represented by consecutive natural numbers, ensuring that the data points of the two curves correspond one-to-one by batch.

[0155] Step S520: Perform monotonicity analysis on the time evolution curves of interlayer slip disturbance component and resin impregnation unevenness disturbance component within the abnormal response time window, and detect the segmented monotonically increasing intervals of each time curve within the window and its continuous batch span.

[0156] Monotonicity analysis is the process of determining whether a curve is increasing, decreasing, or flat within a local interval by comparing curve values ​​in batches. First, a fixed width W is defined. anomaly The abnormal response time window, where W anomaly The number of batches must be no less than a preset threshold. The window is slid along the time axis from the starting batch to the ending batch, and at each window position, subsequences of interlayer slip perturbation and resin impregnation unevenness perturbation are extracted. For each extracted subsequence, a monotonically increasing judgment algorithm is used: the perturbation component values ​​of adjacent batches are compared sequentially; if the value of the subsequent batch is greater than or equal to the value of the previous batch, then the subsequence is determined to be monotonically increasing within that window. The starting and ending batches of all windows showing monotonically increasing values ​​are recorded, forming a set of segmented monotonically increasing intervals for the two curves and their respective continuous batch spans.

[0157] Step S530: Merge the segmented monotonically increasing intervals with time overlap in the two time evolution curves to generate a set of synchronously monotonically increasing time periods. Each element in the set of synchronously monotonically increasing time periods is a time period marker, in which both disturbance components show a monotonically increasing trend.

[0158] Temporal overlap occurs when a monotonically increasing interval of the interlayer slip perturbation component and a monotonically increasing interval of the resin wetting unevenness perturbation component intersect on the batch axis. The merging operation identifies all intersecting interval pairs and treats the intersection interval as a synchronously monotonically increasing period. The period label must include at least the start and end batch indices. All synchronously monotonically increasing periods constitute a set of synchronously monotonically increasing periods. Each period in this set satisfies the basic condition of synchronous increase of both perturbation components, providing the temporal basis for further determining trend synergy.

[0159] Step S540: Extract the slope sequence of the increasing trend of the two time evolution curves within each synchronous monotonically increasing time period to obtain the increasing slope sequence of the interlayer slip disturbance component and the increasing slope sequence of the resin impregnation unevenness disturbance component.

[0160] An increasing slope sequence is a detailed description of the growth rhythm within an increasing curve segment. It consists of the ratio of the increment between adjacent batches to the increment between the previous adjacent batches, and can reflect whether the increasing trend is accelerating, constant, or decelerating.

[0161] In an optional embodiment, step S540 may specifically include the following steps S541 to S546: Step S541: During the synchronous monotonically increasing period, perform batch-by-batch difference operation on the time evolution curve of the interlayer slip disturbance component to obtain the batch increment sequence of the interlayer slip disturbance component. Each element of the batch increment sequence represents the increase of the interlayer slip disturbance component between two adjacent batches.

[0162] For the batch range covered by the synchronously monotonically increasing time period, from b start to b end Extract a portion of the interlayer slip perturbation component sequence within this range, and calculate the first-order difference Δslip. b =P slip(b+1) -P slip(b) , b from b start Take to b end-1 The resulting increments Δslip b This constitutes a batch increment sequence.

[0163] Step S542: Perform adjacent increment ratio transformation on the batch increment sequence of interlayer slip disturbance components to generate an increasing slope sequence of interlayer slip disturbance components. Each slope value in the increasing slope sequence is the ratio of the current batch increment to the previous batch increment.

[0164] The adjacent increment ratio is calculated for the j-th increment in the batch increment sequence, comparing it to the previous increment. To prevent the denominator from being zero, if the numerator is also zero when the previous increment is zero, the ratio is set to 1 (indicating no change); if the numerator is positive, the ratio is preset to a large constant to indicate a sharp acceleration. All effective ratios are arranged in chronological order, forming an increasing slope sequence of the interlayer slip perturbation components.

[0165] Step S543: During the synchronous monotonically increasing time period, perform batch-by-batch differential operation on the time evolution curve of the resin impregnation unevenness disturbance component to obtain the batch increment sequence of the resin impregnation unevenness disturbance component.

[0166] The operation is the same as step S541, performing a first-order difference on the resin wetting unevenness perturbation component sequence within the synchronously monotonically increasing time period to generate the corresponding batch's incremental sequence Δresin. b .

[0167] Step S544: Perform adjacent increment ratio transformation on the batch increment sequence of the resin impregnation unevenness disturbance component to generate an increasing slope sequence of the resin impregnation unevenness disturbance component. Each slope value in the increasing slope sequence is the ratio of the current batch increment to the previous batch increment.

[0168] The operation is the same as step S542, using the same rules for Δresin. b By performing adjacent increment ratio transformation, an increasing slope sequence of the resin impregnation unevenness disturbance component is obtained.

[0169] Step S545: Align the increasing slope sequence of the interlayer slip disturbance component and the increasing slope sequence of the resin wetting unevenness disturbance component with length, and extract the slope subsequence within the time alignment interval from the start batch to the end batch of the synchronous monotonically increasing period.

[0170] Since difference and ratio operations may cause slight differences in sequence length, the length alignment step ensures that the two slope sequences have a completely consistent index correspondence. The approach taken is to use the starting and ending batches of the synchronous monotonically increasing period as boundaries, and to extract the maximum common index range where both sequences have valid slope values, resulting in slope subsequences of equal length for the interlayer slip perturbation component and the resin wetting unevenness perturbation component.

[0171] Step S546: Perform outlier batch detection on the two truncated slope subsequences, remove the slope values ​​corresponding to batches whose slopes change abnormally due to local perturbations, and obtain the incremental slope sequence of the corrected interlayer slip perturbation component and the incremental slope sequence of the corrected resin impregnation unevenness perturbation component.

[0172] Outlier batch detection can be performed using either the standard deviation method or the box plot method. For example, the mean μ and standard deviation σ of the slope subsequence are calculated, and batches with slope values ​​exceeding the range of μ ± 3σ are marked as outliers. The slope values ​​corresponding to these outliers in both slope subsequences are simultaneously removed, and the remaining sequence is the corrected increasing slope sequence. This operation eliminates the interference of a few instantaneous process fluctuations on the determination of overall trend consistency.

[0173] Step S550: Based on the increasing slope sequence of the interlayer slip disturbance component and the increasing slope sequence of the resin impregnation unevenness disturbance component, calculate the slope convergence index of the two during the synchronous monotonically increasing period. The slope convergence index is used to measure the consistency of the change pace of the two slope sequences.

[0174] The slope convergence index is a scalar indicator obtained by comprehensively calculating the degree of deviation and the consistency of fluctuation of two slope sequences. The larger the value, the more coordinated the increasing rhythm of the two sequences.

[0175] In an optional embodiment, step S550 may specifically include the following steps S551 to S556: Step S551: Perform batch-by-batch slope difference calculation on the incremental slope sequence of the corrected interlayer slip disturbance component and the incremental slope sequence of the corrected resin wetting unevenness disturbance component to generate a slope difference sequence, which represents the degree of deviation of the incremental slope of the two disturbance components in each batch.

[0176] Let the two corrected slope sequences be S slip_k and S resin_kk ranges from 1 to K. Calculate the difference d at each corresponding position k. k =|S slip_k -S resin_k These differences form a slope difference sequence. The smaller the difference, the closer the two slopes are on that batch.

[0177] Step S552: Squaring and summing the slope difference sequence to obtain the cumulative slope deviation. The cumulative slope deviation increases monotonically with the increase of the number of batches during the synchronous monotonically increasing period.

[0178] For the cumulative amount of slope deviation The squaring operation amplifies the effect of larger deviations, making the cumulative amount more sensitive to significant deviations.

[0179] Step S553: ​​Normalize the cumulative slope deviation by batch quantity to obtain the normalized slope deviation index. The normalized slope deviation index eliminates the influence of batch quantity differences on the cumulative deviation.

[0180] Normalized slope deviation index D norm =D sum / K, by dividing by the sequence length K, eliminates the problem of incomparable cumulative deviation caused by different batch sizes in different synchronous monotonically increasing time periods.

[0181] Step S554: Calculate the batch-to-batch variation rate of the increasing slope sequence of the corrected interlayer slip disturbance component to generate a first variation rate sequence, and calculate the batch-to-batch variation rate of the increasing slope sequence of the corrected resin wetting unevenness disturbance component to generate a second variation rate sequence.

[0182] The inter-batch rate of change is a measure of the volatility of an increasing slope sequence. The rate of change Δ² is defined as follows. slip_k =S slip_{k+1} -S slip_k This value is the first difference of the slope sequence, reflecting the amplitude and direction of the slope's fluctuations. All K-1 Δ² values... slip_k Arrange the sequences to obtain the first rate of change sequence, and similarly obtain the second rate of change sequence.

[0183] Step S555: Perform correlation analysis on the first rate of change sequence and the second rate of change sequence to obtain the rate of change linkage coefficient. The rate of change linkage coefficient describes the synergy of the rhythm of the increasing slope of the two disturbance components.

[0184] Correlation analysis was performed using the Pearson linear correlation coefficient method. Let the first rate of change sequence be X, and the second rate of change sequence be Y, both with a length of K⁻¹. The covariance Cov(X, Y) and standard deviation σ of X and Y were calculated. X and σ Y Then the rate of change linkage coefficient ρ = Cov(X, Y) / (σ X ×σ Y The value of ρ ranges from -1 to 1. A positive value close to 1 indicates that the two slope fluctuations are highly synchronized and resonant, while a negative value indicates that the two fluctuation rhythms are opposite.

[0185] Step S556: Input the normalized slope deviation index and the rate of change linkage coefficient into the convergence index synthesis function to generate the slope convergence index. The slope convergence index is inversely proportional to the normalized slope deviation index and directly proportional to the rate of change linkage coefficient.

[0186] One example of a convergence index composition function is constructed as the slope convergence index Ψ = (1 + ρ)² / (ε+D norm ), where ε is a very small positive constant to prevent the denominator from being zero, ρ is the rate of change linkage coefficient, and D norm This is a normalized slope bias index. When the bias between two slope sequences is extremely small (D... norm When Ψ approaches 0 and the degree of fluctuation coordination is extremely high (ρ approaches 1), Ψ reaches a maximum value, indicating that the two increase at a highly consistent pace; otherwise, Ψ is smaller.

[0187] Step S560: When the time span of at least one synchronous monotonically increasing period exceeds the abnormal response time window and the slope convergence index corresponding to the period exceeds the preset trend coordination threshold, a copper clad laminate lamination latent anomaly prediction signal is generated.

[0188] The system examines each time period in the set of synchronously monotonically increasing time periods generated in step S530, and filters out time periods whose time span (end batch minus start batch) is greater than a preset abnormal response time window. If, within any of the filtered time periods, the slope convergence index calculated in step S550 exceeds a preset trend coordination threshold, the system concludes that the copper clad laminate lamination process exhibits a latent abnormality due to the enhanced coupling between interlayer slippage and uneven resin wetting, and immediately generates a latent abnormality prediction signal for copper clad laminate lamination, triggering an alarm or process feedback adjustment mechanism.

[0189] Figure 2 A hardware entity diagram of a server provided in an embodiment of the present invention, such as... Figure 2 As shown, the hardware entity of the server 1000 includes a processor 1001 and a memory 1002, wherein the memory 1002 stores a computer program that can run on the processor 1001, and the processor 1001 executes the program to implement the steps in the method of any of the above embodiments.

[0190] The memory 1002 stores computer programs that can run on the processor. The memory 1002 is configured to store instructions and applications that can be executed by the processor 1001. It can also cache data to be processed or already processed (e.g., image data, audio data, voice communication data, and video communication data) in the processor 1001 and various modules in the server 1000. It can be implemented by flash memory or random access memory (RAM).

[0191] When the processor 1001 executes the program, it implements any of the steps of the above-mentioned method for predicting latent anomalies in copper-clad laminate lamination based on contour pattern recognition. The processor 1001 typically controls the overall operation of the server 1000.

[0192] This invention provides a computer storage medium storing one or more programs that can be executed by one or more processors to implement the steps of the copper clad laminate lamination latent anomaly prediction method based on contour pattern recognition as described in any of the above embodiments.

[0193] The above description is merely an embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting latent anomalies in copper-clad laminate lamination based on contour pattern recognition, characterized in that, The method includes: Obtain the discrete sampling point sequence of the edge contour formed by the copper clad laminate after the lamination process, and obtain the lamination process reference contour sequence corresponding to the copper clad laminate; A cross-dimensional contour morphology mapping is performed on the discrete sampling point sequence of the copper clad laminate edge contour and the lamination process reference contour sequence. In the contour pattern observation space jointly defined by the curvature response scale and the fractal feature scale, a cross-scale contour mapping state field of the copper clad laminate edge contour relative to the lamination process reference contour is generated. The cross-scale contour mapping state field characterizes the constraint deviation distribution between the geometric shape of the copper clad laminate edge contour and the geometric shape of the lamination process reference contour in different observation dimensions. Based on the distribution of contour shape constraint deviation in the cross-scale contour mapping state field, contour pattern decomposition at the gradient flow level is performed on the cross-scale contour mapping state field. The geometric instability component of the copper-clad laminate edge contour is separated from the geometric constraint component of the cross-scale contour mapping state field, generating an intrinsic contour pattern distribution manifold. The intrinsic contour pattern distribution manifold includes the contour pattern tensor in the region where the contour shape constraint deviation is not significant and the contour pattern tensor in the region where the contour shape constraint deviation changes abruptly. The intrinsic profile pattern distribution manifold is subjected to profile pattern perturbation evolution analysis along the time series. Based on the spatial adjacency relationship and temporal propagation trajectory of the profile pattern tensor, the trend mutual induction effect of the profile pattern perturbation component corresponding to the latent anomaly of interlayer slip and the profile pattern perturbation component corresponding to the latent anomaly of resin wetting unevenness in the intrinsic profile pattern distribution manifold is measured between consecutive pressing batches, and a profile anomaly perturbation mutual induction evolution trend map is generated. When the contour pattern perturbation component corresponding to the interlayer slip latent anomaly and the contour pattern perturbation component corresponding to the resin impregnation uneven latent anomaly in the contour anomaly perturbation mutual induced evolution trend map show a synchronous monotonically increasing trend within a preset anomaly response time window and the slope convergence index of the trend increasing curve exceeds a preset trend coordination threshold, a copper clad laminate lamination latent anomaly prediction signal is generated.

2. The method for predicting latent anomalies in copper-clad laminate lamination based on contour pattern recognition according to claim 1, characterized in that, The process of performing cross-dimensional contour morphology mapping between the discrete sampling point sequence of the copper-clad laminate edge contour and the lamination process reference contour sequence, and generating a cross-scale contour mapping state field of the copper-clad laminate edge contour relative to the lamination process reference contour within the contour pattern observation space jointly defined by the curvature response scale and the fractal feature scale, includes: Curvature response contour encoding is performed on the discrete sampling point sequence of the copper clad laminate edge contour to generate an edge curvature encoding sequence. Each encoding element of the edge curvature encoding sequence records the local curvature direction mark and local curvature amplitude description at the corresponding sampling point. The curvature response contour sequence of the pressing process is encoded to generate a reference curvature encoding sequence. The encoding element structure of the reference curvature encoding sequence is dimensionally aligned with the encoding element structure of the edge curvature encoding sequence. Based on the edge curvature coding sequence and the reference curvature coding sequence, a contour deviation tensor field is constructed under the curvature response scale. The contour deviation tensor field takes the contour position index as the independent variable and the curvature direction offset and curvature magnitude offset as tensor components. The discrete sampling point sequence of the copper-clad laminate edge contour is subjected to fractal-scale contour encoding to generate an edge fractal-scale contour encoding sequence. The edge fractal-scale contour encoding sequence describes the roughness distribution structure of the edge contour at a multi-level observation granularity. The reference profile sequence of the pressing process is coded using fractal scale to generate a reference fractal scale profile coding sequence. The observation granularity level of the reference fractal scale profile coding sequence corresponds one-to-one with the observation granularity level of the edge fractal scale profile coding sequence. The contour deviation tensor field, the edge fractal scale contour encoding sequence, and the reference fractal scale contour encoding sequence are input into the contour pattern observation space construction stage. The cross-scale contour mapping state field is generated by cross-dimensional mapping fusion. The cross-scale contour mapping state field describes the constraint deviation distribution on the observation grid jointly formed by the curvature dimension and the fractal dimension.

3. The method for predicting latent anomalies in copper-clad laminate lamination based on contour pattern recognition according to claim 2, characterized in that, The step of performing curvature response contour encoding on the discrete sampling point sequence of the copper-clad laminate edge contour to generate an edge curvature encoding sequence includes: Tangent direction angle is recursively extracted from the discrete sampling point sequence of the copper clad laminate edge contour to generate a sequence of tangent direction angle changes between adjacent sampling points; Based on the sequence of changes in the tangent direction angle and the arc length step between adjacent sampling points, a local curvature amplitude is generated at each sampling point. The local curvature amplitude is directly proportional to the change in the tangent direction angle and inversely proportional to the arc length step. The sign direction of the tangent direction angle change sequence is determined, and a curvature direction label is assigned to each sampling point according to the clockwise or counterclockwise attribute of the tangent direction angle change trend. The curvature direction label and the local curvature amplitude together constitute a curvature description pair. Following the sampling order of the discrete sampling point sequence, the curvature description pairs of each sampling point are arranged into a curvature description sequence, and the curvature description sequence is spatially indexed and encoded so that each curvature description pair is associated with its normalized arc length position label on the edge contour. Based on the distribution range of local curvature amplitudes in the curvature description sequence, non-uniform quantization encoding is performed on the local curvature amplitudes, and continuous amplitudes are mapped to the curvature response encoding level to obtain the local curvature amplitude description. The curvature direction marker and the local curvature amplitude description are jointly encoded and fused to generate the edge curvature coding sequence. Each coding element of the edge curvature coding sequence is a combined coding word, which includes a direction marker bit field and an amplitude description bit field.

4. The method for predicting latent anomalies in copper-clad laminate lamination based on contour pattern recognition according to claim 3, characterized in that, The step of performing fractal-scale contour encoding on the discrete sampling point sequence of the copper-clad laminate edge contour to generate an edge fractal-scale contour encoding sequence includes: Multi-scale step size resampling is performed on the discrete sampling point sequence of the copper clad laminate edge contour to generate several sets of resampled contour sequences with different observation granularities. The sampling step size of each set of resampled contour sequences is larger than the sampling step size of the previous set of resampled contour sequences. For each group of resampled contour sequences, the total length of the contour segments is measured to obtain the apparent length of the contour at the corresponding observation granularity. The apparent length of the contour decreases as the observation granularity increases. Based on the ratio of the apparent length of the profile at adjacent observation granularities to the ratio of the corresponding observation granularities, the profile length change rate in double logarithmic coordinates is generated as the fractal scale description component within the observation granularity interval; wherein, the observation granularity interval is the interval formed by the observation granularities corresponding to two adjacent sets of resampled profile sequences. The fractal scale description components within all observation granularity ranges are arranged in order from fine to coarse observation granularity to form a fractal scale description component sequence. The fractal scale description component sequence is subjected to inter-scale continuity constraints, and adjacent fractal scale description components are smoothed for trend consistency using a sliding window to generate a smooth fractal scale description sequence. The smooth fractal scale description sequence suppresses inter-scale jumps caused by sampling noise. The smooth fractal scale description sequence is bound and encoded with the corresponding observation granularity level index to generate the edge fractal scale contour encoding sequence. Each encoding element in the edge fractal scale contour encoding sequence corresponds to a local contour roughness encoding of an observation granularity level.

5. The method for predicting latent anomalies in copper-clad laminate lamination based on contour pattern recognition according to claim 1, characterized in that, Based on the distribution pattern of the contour morphology constraint deviation in the cross-scale contour mapping state field, the contour pattern decomposition at the gradient flow level is performed on the cross-scale contour mapping state field to separate the geometric instability component of the copper-clad laminate edge contour from the geometric constraint component of the cross-scale contour mapping state field, generating an eigencontact contour pattern distribution manifold, including: On the observation grid jointly composed of the curvature response scale and the fractal feature scale, the spatial gradient flow direction field of the constraint deviation scalar function of the cross-scale contour mapping state field is calculated, and the deviation change direction along the curvature dimension and the fractal dimension is defined at each grid point of the spatial gradient flow direction field. The geometric constraint components of the cross-scale contour mapping state field are extracted based on the spatial gradient flow direction field. The geometric constraint components are composed of regions where the gradient flow divergence is lower than a preset divergence limit, and the contour shape constraint deviation within these regions remains stably distributed along the tangential direction of the gradient flow. The geometric instability component of the cross-scale contour mapping state field is extracted based on the spatial gradient flow direction field. The geometric instability component is composed of regions where the gradient flow divergence is higher than a preset divergence limit, and the contour shape constraint deviation within these regions exhibits a local divergent or convergent distribution along the normal direction of the gradient flow. Topological connectivity analysis is performed on the geometric constraint components to extract the connected component skeleton, and the connected component skeleton is mapped to the spatial position of the edge contour to generate the contour pattern tensor of the constraint region. The contour pattern tensor of the constraint region describes the constrained deformation mode in the region where the contour shape constraint deviation is not significant. Local extremum point detection is performed on the geometrically unstable component, the spatial distribution of local scalar value extremum points in the geometrically unstable component is extracted, and the distribution of the local scalar value extremum points is mapped to the spatial position of the edge contour to generate the contour mode tensor of the abrupt region. The contour mode tensor of the abrupt region describes the free deformation mode in the abrupt region of the contour shape constraint deviation. The contour pattern tensor of the constrained region and the contour pattern tensor of the abrupt region are joined along the edge contour position index to obtain the intrinsic contour pattern distribution manifold.

6. The method for predicting latent anomalies in copper-clad laminate lamination based on contour pattern recognition according to claim 5, characterized in that, The step of performing topological connectivity analysis on the geometric constraint components, extracting the connected component skeleton, and mapping the connected component skeleton to the spatial location of the edge contour to generate the contour pattern tensor of the constraint region includes: The observation grid region where the geometric constraint component is located is binary-labeled to generate a binary distribution map of the constraint region. Grid points belonging to the geometric constraint component in the binary distribution map of the constraint region are assigned activation labels. Morphological thinning is performed on the binary distribution map of the constrained region, and the boundary grid points are peeled off layer by layer until a connected domain skeleton segment with a single grid width is obtained. The connected domain skeleton segment maintains the topological connectivity and geometric extension direction of the original constrained region. Node-branch topology analysis is performed on the connected domain skeleton segments to extract the branch node positions and terminal node positions in the skeleton segments, and the curvature dimension coordinate sequence and fractal dimension coordinate sequence of the skeleton branches between adjacent nodes in the observation grid are recorded. The curvature dimension coordinate sequence of the skeleton branch is converted into curvature description code, and the fractal dimension coordinate sequence of the skeleton branch is converted into fractal scale description code to form a joint cross-dimensional morphological description of the skeleton branch. The joint cross-dimensional morphological description of the skeleton branch is propagated and aggregated along the connectivity path of the node-branch topology, and a local morphological transition tensor is generated at each branch node based on the difference in morphological description between the in-degree branch and the out-degree branch. The local morphological transition tensor is bound to the corresponding edge contour position index to generate the contour pattern tensor of the constrained region. The contour pattern tensor of the constrained region stores the description of the constrained deformation pattern at the contour position as the index.

7. The method for predicting latent anomalies in copper-clad laminate lamination based on contour pattern recognition according to claim 6, characterized in that, The step of detecting local extrema of the geometrically unstable components, extracting the spatial distribution of local scalar extrema in the geometrically unstable components, and mapping the distribution of the local scalar extrema to the spatial location of the edge contour to generate the contour pattern tensor of the abrupt region includes: Within the observation grid region where the geometrically unstable component is located, the Hessian response of the constraint deviation scalar function at each grid point is calculated. The Hessian response is used to measure the principal curvature intensity of the local surface at that grid point. Based on the local peak position of the Hessian response, the coordinates of the scalar extreme point of the profile shape in the geometric instability component are determined, and the coordinates of the scalar extreme point include curvature dimension coordinates and fractal dimension coordinates. For each scalar extreme point, extract the constraint deviation scalar distribution within a local neighborhood window centered on it, and construct the deformation driving direction vector of the extreme point based on the principal direction change rate of the scalar values ​​within the local neighborhood window. The deformation driving direction vector is combined with the corresponding extreme point coordinates to obtain a free deformation trend description of the extreme point. The free deformation trend description indicates the tendency of the geometric instability component at the extreme point to expand outward or contract inward. The free deformation trend descriptions of all extreme points are sorted according to the curvature dimension coordinates of the extreme points, and the distribution histograms are aggregated along the fractal dimension direction to obtain the free deformation trend distribution sequence based on the curvature dimension coordinates. Each free deformation trend description in the free deformation trend distribution sequence is mapped to the edge contour sampling point position index corresponding to the extreme point to generate the contour pattern tensor of the abrupt region. The contour pattern tensor of the abrupt region stores the free deformation pattern description at the edge contour position as the index.

8. The method for predicting latent anomalies in copper-clad laminate lamination based on contour pattern recognition according to claim 1, characterized in that, The analysis of the evolution of contour pattern perturbations along the time series of the intrinsic contour pattern distribution manifold, based on the spatial adjacency relationship and temporal propagation trajectory of the contour pattern tensor, measures the mutual inductive effect of the contour pattern perturbation components corresponding to interlayer slip latent anomalies and the contour pattern perturbation components corresponding to resin wetting unevenness latent anomalies in the intrinsic contour pattern distribution manifold between consecutive pressing batches, and generates a mutual inductive evolution trend map of contour anomaly perturbations, including: Obtain the intrinsic profile pattern distribution manifold sequence corresponding to multiple consecutive pressing batches, wherein each manifold element in the intrinsic profile pattern distribution manifold sequence corresponds to a set of profile pattern tensors of a pressing batch under a unified edge profile position index. Spatial adjacency modeling is performed on the intrinsic contour pattern distribution manifold sequence. A spatial adjacency graph of the contour pattern tensor is constructed within each pressing batch. Adjacent nodes in the spatial adjacency graph are defined by tensor pairs that are adjacent along the contour position index and whose tensor similarity between contour pattern tensors exceeds a preset similarity threshold. Perturbation type projection is performed on the contour pattern tensor corresponding to each node in the spatial adjacency graph. The tensor components associated with the interlayer slip deformation mode are projected to the interlayer slip perturbation channel, and the tensor components associated with the resin impregnation uneven deformation mode are projected to the resin impregnation uneven perturbation channel, so as to obtain the perturbation components of each node in the interlayer slip perturbation channel and the perturbation components in the resin impregnation uneven perturbation channel. Along the time dimension of the distribution manifold sequence of the intrinsic profile pattern, the change trajectory of the disturbance component of the same profile position node between different pressing batches is tracked to obtain the temporal propagation trajectory of the interlayer slip disturbance component and the temporal propagation trajectory of the resin impregnation unevenness disturbance component. The temporal propagation trajectories of interlayer slip disturbance components and resin wetting unevenness disturbance components are subjected to trend mutual induction detection. When the change direction of interlayer slip disturbance components in one batch causes the resin wetting unevenness disturbance components to change in the same direction in subsequent batches, the induced correlation strength of the batch and its contour position is recorded. The induced correlation strength between multiple batches at all contour locations is fused and plotted along both time and space dimensions to generate the contour anomaly perturbation mutual induced evolution trend map. The contour anomaly perturbation mutual induced evolution trend map contains the spatiotemporal thermal distribution and directional propagation vector of the induced correlation strength.

9. The method for predicting latent anomalies in copper-clad laminate lamination based on contour pattern recognition according to claim 8, characterized in that, The step of perturbation-type projection of the contour pattern tensor corresponding to each node in the spatial adjacency graph, projecting the tensor components associated with interlayer slip deformation modes to the interlayer slip perturbation channel, and projecting the tensor components associated with resin impregnation uneven deformation modes to the resin impregnation uneven perturbation channel includes: Extract the principal direction tensor component and deformation amplitude tensor component of the contour deformation from the contour pattern tensor. The principal direction tensor component indicates the distribution of deformation in the contour tangent and contour normal, and the deformation amplitude tensor component indicates the deformation intensity. The principal direction tensor component is matched with the preset interlayer slip deformation direction template, and the projection amplitude of the principal direction tensor component in the interlayer slip deformation direction is calculated to generate the projection intensity of the interlayer slip deformation direction. The deformation amplitude tensor component and the projection intensity of the interlayer slip deformation direction are jointly encoded as the perturbation component of the profile pattern tensor in the interlayer slip perturbation channel. The perturbation component of the interlayer slip perturbation channel carries both deformation direction conformity and deformation intensity information. The principal direction tensor component is matched with the preset resin impregnation uneven deformation direction template, and the projection amplitude of the principal direction tensor component in the resin impregnation uneven deformation direction is calculated to generate the resin impregnation uneven deformation direction projection intensity. The resin impregnation uneven deformation direction template and the interlayer slip deformation direction template have an orthogonal constraint relationship. The deformation amplitude tensor component and the projection intensity of the resin impregnation uneven deformation direction are jointly encoded as the perturbation component of the contour mode tensor in the resin impregnation uneven perturbation channel. The perturbation component of the resin impregnation uneven perturbation channel carries both deformation direction conformity and deformation intensity information. The perturbation components of the interlayer slip perturbation channel and the perturbation components of the resin impregnation uneven perturbation channel at the same contour position node are orthogonally decomposed and verified. The ratio of the perturbation components of the two channels is adjusted so that there is no redundant directional projection overlap, and the verified perturbation components are obtained.

10. A server comprising a memory and a processor, the memory storing a computer program capable of running on the processor, characterized in that, When the processor executes the program, it implements the steps of the method according to any one of claims 1 to 9.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the method according to any one of claims 1 to 9.

Images