Rapid defect positioning method based on glass detection

Abstract

The invention discloses a rapid defect positioning method based on glass detection, and relates to the technical field of industrial detection, and the method comprises the following steps: distributing quantum computing resources according to glass types and production line priorities, and driving a quantum processor to process submicron defect features; performing DNA molecular coding operation on the submicron defect features, mapping a light intensity discrete value into an adenine base sequence, mapping a polarization scattering angle into a thymine base sequence, mapping a phase gradient tensor into a guanine base sequence, and constructing a defect DNA fragment containing space-time attributes; and the defect identification model and the defect depth value are fused, the Boltzmann entropy weight is combined to calculate the defect position confidence coefficient, and when the defect position confidence coefficient exceeds a threshold value, three-dimensional coordinates and defect type labels are output. Through a DNA molecular coding technology, optical characteristics are converted into computable bioinformatics sequences, and the expression ability of defect characteristics is enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial detection, and in particular to a method for quickly locating defects based on glass detection. Background Art

[0002] As a key material in the construction, automotive, and electronics industries, glass quality directly impacts product safety and performance. As industrial manufacturing continues to demand higher quality glass, the detection of micron- and even submicron-level defects has become a critical factor limiting product quality. While methods based on optical imaging and machine learning have made progress in defect detection in recent years, they still face significant challenges in practical industrial applications.

[0003] Existing technologies primarily rely on high-resolution camera systems and computer vision algorithms for surface defect detection, but they have significant shortcomings when dealing with submicron defects. Firstly, because the optical signatures of submicron defects are extremely faint, traditional feature extraction methods struggle to accurately identify them amidst the optical interference of complex production environments. Secondly, existing detection systems employ static algorithm architectures, making it difficult to dynamically adjust detection strategies based on changes in the production environment. This makes it difficult to balance detection accuracy and real-time performance requirements under changing working conditions. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a method for rapid defect location based on glass detection to solve the problems of insufficient submicron glass defect detection accuracy and poor adaptability to dynamic production environments in the prior art.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: In a first aspect, the present invention provides a method for rapid defect location based on glass inspection, which includes allocating quantum computing resources according to glass type and production line priority, driving the quantum processor to process submicron defect features; performing DNA molecular encoding operations on submicron defect features, mapping discrete light intensity values ​​to adenine base sequences, mapping polarization scattering angles to thymine base sequences, and mapping phase gradient tensors to guanine base sequences, to construct defective DNA fragments containing spatiotemporal properties; transmitting the defective DNA fragments to a cloud-based federated aggregation server through a homomorphic encryption channel, and performing probability-driven cross-mutation and adaptive recombination with encrypted defective DNA fragments across production lines; expressing the recombined encrypted defective DNA fragments as a defect recognition model, synchronously calculating the defect state energy level offset in the photonic band structure, and inverting the defect depth value according to the phonon resonance frequency; fusing the defect recognition model and the defect depth value, and calculating the defect position confidence in combination with the Boltzmann entropy weight, and outputting the three-dimensional coordinates and defect type label when the defect position confidence exceeds the threshold value.

[0007] As a preferred solution of the defect rapid location method based on glass inspection of the present invention, wherein: the submicron defect characteristics include light intensity discrete value, polarization scattering angle and phase gradient tensor; The light intensity discrete value, polarization scattering angle and phase gradient tensor refer to two-dimensional spatial distribution parameters extracted by Fourier transform and polarization interference analysis after collecting optical signals of glass surface defects.

[0008] As a preferred solution of the glass inspection-based rapid defect location method described in the present invention, the allocation of quantum computing resources specifically includes constructing a mapping table between glass type and number of quantum bits, and dynamically adjusting the priority of the quantum processor task queue according to the real-time load of the production line.

[0009] As a preferred solution of the method for rapid defect location based on glass inspection of the present invention, the specific steps of constructing a defective DNA fragment containing spatiotemporal attributes are as follows: The discrete values ​​of light intensity are mapped to the length of adenine base sequence through quantization segmentation; The polarization scattering angle is mapped to the thymine base sequence by angle interval division; The phase gradient tensor is mapped to the guanine base combination pattern via eigenvalue decomposition; The length of adenine base sequence, the order of thymine base sequence and the combination pattern of guanine base are embedded in the time-space stamp to construct multi-dimensional defective DNA fragments.

[0010] As a preferred solution of the glass inspection-based defect rapid location method described in the present invention, the homomorphic encryption channel adopts a lattice-based cryptographic system, and the crossover mutation and adaptive recombination are optimized through Monte Carlo Markov chain sampling to perform fragment exchange probability optimization.

[0011] As a preferred solution of the method for rapid defect location based on glass inspection of the present invention, the following specific steps are taken to construct a defect recognition model: Analyze the number of consecutive adenine base repeats in the recombinant defective DNA fragment and map it to the number of layers in the convolutional neural network; According to the thymine base sequence sorting pattern, the weight initialization matrix of each convolutional layer is generated, and the guanine base combination pattern is used to determine the dimension of the neural network feature channel; The energy level offset of the photon band defect state is calculated based on the parameter-optimized tight-binding model to verify the material defect characteristics. The defect recognition model is generated through dynamic weight fusion by combining the phonon resonance frequency inversion results with the neural network output.

[0012] As a preferred solution of the glass inspection-based defect rapid positioning method described in the present invention, the Boltzmann entropy weight calculation refers to assigning an entropy weight coefficient according to the thermodynamic probability distribution of the defect depth value, and integrating the confidence output of the defect recognition model to generate a position decision factor.

[0013] As a preferred solution of the glass inspection-based defect rapid location method described in the present invention, the threshold value refers to a confidence threshold value pre-set according to the characteristics of the glass material, and when the confidence of the defect position exceeds the confidence threshold value, it is determined to be a valid defect.

[0014] In a second aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the method for rapid defect location based on glass inspection as described in the first aspect of the present invention is implemented.

[0015] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, any step of the method for rapid defect location based on glass inspection as described in the first aspect of the present invention is implemented.

[0016] The beneficial effects of the present invention are as follows: the present invention significantly improves the detection accuracy and computational efficiency of submicron glass defects through dynamic allocation technology of quantum computing resources; through DNA molecular coding technology, optical features are converted into computable bioinformatics sequences, thereby enhancing the expressive ability of defect features; through federal aggregation technology under homomorphic encryption, the sharing and optimization of defect features across production lines are realized while ensuring data security; finally, by integrating photon band structure calculation and phonon resonance inversion technology, a physically interpretable defect recognition model is constructed, which significantly improves the accuracy and reliability of defect positioning. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 This is a flow chart of a method for rapid defect location based on glass inspection.

[0019] Figure 2 Flowchart for quantum computing resource allocation and defect feature processing.

[0020] Figure 3 Flowchart of encoding operations for DNA molecules.

[0021] Figure 4 Flowchart for homomorphic encryption and federation reorganization. DETAILED DESCRIPTION

[0022] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0023] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0025] Reference Figures 1 to 4 , is an embodiment of the present invention, which provides a method for quickly locating defects based on glass inspection, comprising the following steps: S1: Allocate quantum computing resources based on glass type and production line priority, driving quantum processors to process submicron defect features.

[0026] Among them, allocating quantum computing resources specifically includes building a mapping table between glass type and number of quantum bits, and dynamically adjusting the priority of the quantum processor task queue according to the real-time load of the production line.

[0027] S1.1: Read the glass type currently being inspected. Glass types include float glass, tempered glass, or Low-E glass, which are defined in the production line configuration file.

[0028] Create a mapping table between glass type and number of qubits, where the row index is the glass type and the column is the corresponding number of qubits.

[0029] It should be noted that the quantum bit values ​​are preset based on the hardness, thickness, and defect sensitivity characteristics of the glass, ensuring the minimum computing resources required to process submicron defect features. The mapping table is stored in memory and quickly accessed through a key-value query mechanism.

[0030] Use load monitoring tools to continuously collect real-time production line load data, including CPU utilization, task queue length, and defect detection frequency. Apply a priority queue algorithm to calculate task priority weights based on real-time load data. During periods of high load (e.g., when the queue length exceeds the task queue length threshold), the weights of low-priority tasks are reduced; during periods of low load, the weights of high-priority tasks are increased. Weight calculation uses linear interpolation, and the task queue list is reordered after the task queue priority is updated.

[0031] It should be noted that the determination of the task queue length threshold needs to be combined with historical operation data and performance verification.

[0032] Specifically, the distribution characteristics of task queue lengths during normal production line operation were analyzed, and a critical value under typical high-load conditions was selected as a baseline reference. Different load scenarios were simulated in a test environment, and this baseline value was adjusted until it met the response requirements of real-time defect analysis tasks. Through actual production line operation observations, it was confirmed that the task queue length threshold ensured the processing timeliness of critical tasks under high load conditions, and the verified value was written to the configuration file. For example, on a glass production line, after multiple tests, the task queue length threshold was set to 10, ensuring that the latency of real-time inspection tasks remained within the acceptable range.

[0033] For example, during production line operation, real-time data is collected: CPU utilization is 75%, the task queue length is 15 pending tasks, and the defect detection frequency is 5 times per second. When the task queue length exceeds the preset task queue length threshold of 10 tasks, a high load state is determined. Linear interpolation is used to reduce the weight of the routine logging task from 0.6 to 0.4, while the weight of the real-time defect analysis task remains unchanged at 1.0. The task queue is reordered according to the new weights, ensuring that the real-time defect analysis task is prioritized in allocating quantum computing resources. When the queue length drops to 8 tasks, the weight of the real-time defect analysis task is increased to 1.2 to expedite processing of the backlog, while the weight of the routine logging task is restored to 0.6. This dynamic adjustment process continues in a continuous loop, ensuring that quantum processor resources are prioritized for the most time-sensitive defect detection tasks.

[0034] S1.2: Based on the mapping table query results and the task queue priority, the number of qubits is allocated to the quantum processor. This allocation process includes determining the base number of qubits based on the glass type and scaling resources based on the priority weight (e.g., increasing the number of qubits by 10% for higher weights). Resource allocation instructions are sent to the quantum processor hardware via the quantum computing API.

[0035] Execute the quantum processor instruction set and input sub-micron defect feature data.

[0036] It should be noted that the discrete values ​​of light intensity, polarization scattering angles, and phase gradient tensors are submicron-level defect characteristic data, and are derived from optical signal processing. Specifically, they refer to the two-dimensional spatial distribution parameters extracted through Fourier transform and polarization interference analysis after collecting the optical signals of defects on the glass surface.

[0037] The quantum processor runs a quantum algorithm (such as the Grover search algorithm) to process submicron-level defect signature data and outputs a processed defect signature dataset. After processing, the results are stored in shared memory for subsequent use.

[0038] S2: Perform DNA molecular encoding operations on submicron-scale defect features, mapping the discrete values ​​of light intensity to adenine base sequences, the polarization scattering angle to thymine base sequences, and the phase gradient tensor to guanine base sequences, to construct defective DNA fragments with spatiotemporal properties.

[0039] S2.1: The discrete values ​​of light intensity are mapped to the length of adenine base sequence through quantization segmentation.

[0040] Specifically, the light intensity discrete values ​​(two-dimensional matrix) are extracted from the defect feature data set output by the quantum processor, expanded into a one-dimensional sequence by row, and the sequence values ​​are normalized and scaled to the [0,1] interval; According to the preset segmentation threshold (such as 0.2, 0.4, 0.6, 0.8), the normalized value is divided into intervals (such as 5 intervals), and the adenine base sequence length is mapped according to the interval value to generate an adenine base sequence of the corresponding length; It should be noted that the preset segmentation threshold is determined through systematic statistical analysis of historical glass defect detection data combined with experimental verification. Specifically, a data-driven approach is adopted: first, defect optical detection data of large-scale glass samples are collected, discrete values ​​of light intensity are extracted and normalized, and then statistical methods such as cluster analysis (such as the K-means algorithm) and kernel density estimation are used to automatically identify significant feature points of the data distribution. The impact of different threshold combinations on defect classification performance is tested through grid search and cross-validation, and the threshold combination that achieves the optimal balance between classification accuracy and recall rate is selected as the preset value.

[0041] For example, in float glass inspection, it has been experimentally verified that the four threshold points of 0.2, 0.4, 0.6, and 0.8 can effectively distinguish five types of defect characteristics. These threshold points will be written into the system configuration file and a dynamic update mechanism will be established. They will be continuously optimized through online learning algorithms to adapt to new inspection data characteristics.

[0042] Example: If value ∈ [0, 0.2), the length of the adenine base sequence is 2; if value ∈ [0.2, 0.4), the length of the adenine base sequence is 4; if value ∈ [0.4, 0.6), the length of the adenine base sequence is 6; if value ∈ [0.6, 0.8), the length of the adenine base sequence is 8; if value ∈ [0.8, 1], the length of the adenine base sequence is 10.

[0043] S2.2: The polarization scattering angle is mapped to the thymine base sequence by angle interval division.

[0044] Specifically, the azimuth component of the polarization scattering angle tensor is extracted and the angle intervals are divided; For example, 60° is divided into 6 angle intervals. According to the interval to which the angle belongs, the thymine base sorting rule is assigned: when [0,60) → positive order ("TTT..."); when [60,120) → reverse order ("...TTT"); when [120,180) → alternating positive and reverse order ("TtTt...", t represents lowercase); when [180,240) → repeated pairs ("TTaaTTaa"); when [240,300) → triply symmetric ("TTTaaTTTaa"); when [300,360) → random order ("aTaTTa").

[0045] It should be noted that in the polarization scattering angle mapping rule: the capital letter "T" represents the standard thymine base; the lowercase letter "t" represents the modified thymine base variant; and the letter "a" represents the placeholder for the adenine base.

[0046] Generate a fixed-length (length = 6) thymine base sequence according to the sorting rules.

[0047] S2.3: The phase gradient tensor is mapped to the guanine base combination pattern via eigenvalue decomposition.

[0048] Specifically, the phase gradient tensor (2×2 matrix) is subjected to eigenvalue decomposition, the main eigenvalue λ1 is extracted, and the combination pattern is defined according to the sign and size of λ1 to generate a 3-base guanine sequence.

[0049] For example, λ1>0.5→mode G1 ("GGG"); 0<λ1≤0.5→mode G2 ("GgG"); λ1=0→mode G3 ("gGg"); λ1<0→mode G4 ("ggG").

[0050] It should be noted that in the phase gradient tensor mapping rule: the capital letter "G" represents the standard guanine base; the lowercase letter "g" represents the modified guanine base variant.

[0051] S2.4: Embed the adenine base sequence length, thymine base sequence order, and guanine base combination pattern into the time-space stamp to construct a multi-dimensional defective DNA fragment.

[0052] Specifically, time and space stamps are extracted from the output of the quantum processor, and the three types of base sequences are combined with the time and space stamps into multidimensional structured data; among them, the time stamp = the UTC millisecond time when the data processing is completed; the space stamp = the center coordinate of the defect area on the glass surface; the three types of base sequences refer to the adenine base sequence length (light intensity mapping), the thymine base sequence order (polarization angle mapping), and the guanine base combination pattern (phase gradient mapping).

[0053] S3: The defective DNA fragments are transmitted to the cloud-based federated aggregation server through a homomorphic encryption channel, and probability-driven cross-mutation and adaptive recombination are performed with the encrypted defective DNA fragments across production lines.

[0054] Among them, the homomorphic encryption channel adopts the lattice-based cryptographic system; cross-mutation and adaptive recombination optimize the fragment exchange probability through Monte Carlo Markov chain sampling.

[0055] S3.1: Generate a public-private key pair using the Kyber-1024 algorithm; read the multidimensional structural data of the defective DNA fragment (adenine sequence length, thymine sequence, guanine pattern, timestamp, and spatial coordinates); convert the multidimensional structural data into an integer array and perform byte block division.

[0056] Perform lattice encryption on each block of data, encrypt the array block with the public key to generate the corresponding ciphertext block; Specifically, the structure of each ciphertext block includes two parts: the encrypted data body and the integrity tag.

[0057] Among them, the encrypted data body is the lattice encryption result of the array block; the integrity tag is the SHA-256 hash value of the timestamp; All ciphertext blocks are combined into an encrypted block sequence according to the generation order of multi-dimensional structure data (i.e., adenine sequence length → thymine sequence → guanine pattern → timestamp → spatial coordinates) as subsequent data input.

[0058] S3.2: Deploy the TLS 1.3 protocol at the transport layer to establish a secure link and stream the encrypted block sequence to the cloud federation aggregation server via HTTP / 2; After receiving the data, the cloud federation aggregation server uses the private key to batch verify the data integrity tags; If the verification fails (the hash value does not match), a request for retransmission of the encrypted block is triggered; the encrypted block sequence that passes the verification is input into the reassembly step.

[0059] S3.3: Crossover mutation and adaptive recombination are performed by Monte Carlo Markov chain sampling to optimize the probability of segment exchange.

[0060] Specifically, the encrypted defective DNA fragments from multiple production lines are aggregated and the time and space stamps (UTC millisecond time + coordinates) in the ciphertext are extracted; The initial exchange probability is calculated based on the similarity of time and space stamps, and adaptive recombination rounds (10 rounds) are performed; For example, two encrypted defective DNA fragments are randomly selected from the fragment pool and the following operations are performed according to the current exchange probability: the time stamp is exchanged with an 80% probability; the thymine sequence and guanine pattern are exchanged with a 60% probability; the adenine sequence length variation is rejected (maintained at the original value); Calculate the Boltzmann entropy change ΔS between the recombined encrypted defective DNA fragment and the original fragment; When ΔS>0: accept the newly generated encryption defective DNA fragment and increase the probability of this type of exchange; when ΔS≤0: retain the original encryption defective DNA fragment; update the next round of exchange probability parameters.

[0061] S3.4: The recombined encryption-defective DNA fragment maintains a homomorphic encryption state; the output structure is a modified multi-dimensional ciphertext array: [encrypted adenine length, encrypted thymine sequence, encrypted guanine pattern, encrypted timestamp, and encrypted spatial coordinates], which is written to the distributed storage pool of the federated aggregation server.

[0062] The best approach is to dynamically reorganize encrypted defect data while ensuring data security. It also uses spatiotemporal correlation and entropy change evaluation mechanisms to automatically screen the optimal feature combination, enabling intelligent fusion of defect features across production lines, thereby improving detection accuracy and production line adaptability, and outputting more reliable defect identification results.

[0063] S4: Express the reconstructed encrypted defective DNA fragment as a defect recognition model, synchronously calculate the defect state energy level offset in the photonic band structure, and invert the defect depth value based on the phonon resonance frequency.

[0064] S4.1: Decrypt the recombinant encrypted defective DNA fragment using the lattice private key, analyze the number of consecutive repetitions of the adenine base, and map it to the number of layers in the convolutional neural network.

[0065] Specifically, the key management service under the federated learning architecture is called to obtain the lattice private key, homomorphic decryption is performed in the TEE, and the time-space stamp hash is verified. The number of adenine base repetitions Y is mapped to the number of CNN layers. For example, the decrypted adenine sequence is "AAAA" (Y=4) → a 4-layer CNN is constructed. Note: The actual encrypted state is stored in a homomorphic encryption format.

[0066] S4.2: Generate the weight initialization matrix for each convolutional layer based on the thymine base sequence sorting pattern, and use the guanine base combination pattern to determine the neural network feature channel dimension.

[0067] Specifically, the thymine sequence sorting patterns (such as forward, reverse, and alternating) are analyzed, and the weight matrix is ​​generated according to the following rules: forward order → use orthogonal initialization to keep the input and output variances consistent; reverse order → use truncated normal distribution to avoid gradient explosion; alternating order → use Xavier uniform distribution to adapt to nonlinear activation functions.

[0068] Extract the combination pattern of guanine (such as "GGG" and "GgG"), use the number of capital G + 1 as the number of feature channels for each convolution layer, and generate the initialization weight matrix and channel dimension parameters of each convolution layer.

[0069] Example: "GgG" (capital G = 2) → number of channels = 3.

[0070] S4.3: Calculate the energy level offset of the photon band defect state based on the parameter-optimized tight-binding model, verify the material defect characteristics, combine the phonon resonance frequency inversion results with the neural network output, and generate a defect recognition model through dynamic weight fusion.

[0071] Specifically, the photon band structure is calculated through the tight binding model to establish the Hamiltonian , the expression is: ; ; Where, and Represents the position index of the lattice point in the lattice, ( , usually limited to the nearest neighbor or next nearest neighbor), To generate the operator, we can express it at the grid point An electron is generated (increase the number of particles), is the annihilation operator, which means that at the grid point annihilates an electron (reduces the number of particles), same , but acting on the grid And with The combination represents the electrons at the lattice The number of occupants, Represents the occupancy number operator, used to describe the grid points The occupancy of the electrons, is the transition integral, which describes the electron transition from the lattice point Jump to grid point The quantum mechanical amplitude of (kinetic energy term), To represent the description grid The local potential energy (such as the nuclear potential field or the external potential), is the Hermitian conjugation operation, which means that the operator or matrix is ​​transposed and complex conjugated at the same time; It should be noted that the tight-binding model is a classic theoretical framework in condensed matter physics that describes the motion of electrons in periodic lattices. Its core concept is that electrons are primarily bound by the localized potential fields of their nearest-neighbor atoms. By constructing a special system of quantum mechanical operators, this model decomposes the collective motion of electrons in a crystal into a superposition of quantum transitions between individual atomic orbitals and the effects of localized potential energy. In practical applications, the tight-binding model transforms complex many-body problems into a computable matrix form, and the band structure characteristics of the material are obtained through mathematical eigenvalue solutions.

[0072] This theoretical approach is particularly well-suited for analyzing the perturbations of electronic states caused by crystal defects, as it can clearly distinguish between localized state changes introduced by defects and the intrinsic properties of the intact crystal. In practical applications, the tight-binding model parameters must be adjusted in accordance with the material's symmetry characteristics and boundary conditions, while numerical calculation tools must be used to perform large-scale matrix operations.

[0073] Convert the Hamiltonian into a numerically solvable matrix form, solve the eigenvalue problem, and output the angularized Hamiltonian.

[0074] Specifically, under the real space Wannier basis set, the Hamiltonian is converted into an N×N Hermitian matrix (N is the total number of supercell orbitals), and the Hermitian matrix elements satisfy: ; Where, is the Hamiltonian matrix element of the tight binding model, and the matrix it constitutes is a Hermitian matrix. Indicates the The local potential energy of the neighboring grid points reflects the influence of the nuclear potential field or the external potential. is the transition integral, which describes the electrons at the neighboring lattice points and The quantum tunneling intensity (kinetic energy term) between them.

[0075] Furthermore, the eigenvalue equation of the Hermitian matrix is ​​solved by the zheevd routine of the LAPACK library to complete the diagonalization of the Hamiltonian.

[0076] Specifically, the N×N Hermitian matrix generated by the tight-binding model is stored as a complex array in column-first order as input. The calculation parameters are configured to calculate all eigenvalues ​​and eigenvectors. The divide-and-conquer method is used to ensure numerical stability. The eigenvalue array arranged in ascending order and the eigenvector matrix with the same dimension as the input matrix are output to complete the diagonalization solution of the Hamiltonian.

[0077] Calculating defect state energy level offsets by diagonalizing the Hamiltonian , the expression is: ; Where, Indicates the ground state energy when there are defects (such as vacancies, impurities, etc.) in the crystal. represents the ground state energy of a defect-free ideal crystal (reference standard), Subscript identifiers indicating the presence of defects in the crystal, Subscript identifier representing a defect-free ideal crystal; Furthermore, to quantify the spatial distribution characteristics of defects, it is necessary to combine electronic structure analysis with lattice dynamics calculations to establish a quantitative relationship between the defect state energy level offset and the phonon vibration characteristics, thereby characterizing the defect characteristics in all dimensions. Specifically, according to the phonon resonance frequency Inversion defect depth , the expression is: ; Where, The speed of sound waves in a medium (e.g. ), phonon resonance frequency Obtained through Raman spectroscopy or first-principles phonon spectrum calculation.

[0078] It should be noted that the defect state energy level offset It reflects the intensity of the disturbance of the defect on the electronic structure, while the defect depth Describe the spatial location of the defect; when Significantly deviate from the bulk material band gap (e.g. , is the symbol of electron volts), the depth of phonon frequency inversion must be combined , confirm whether the defect is located near the surface (such as , is the abbreviation of nanometer, which means the unit of length) or bulk phase (such as ).

[0079] Example: Establishing defect state energy level offsets by combining electronic structure analysis with lattice dynamics calculations and phonon vibration characteristics The quantitative relationship between Reflects the intensity of the disturbance of the defect on the electronic structure (such as the vacancy defect in silicon causing deep energy levels), and through Inverted defect depth (like ) characterizes the spatial location characteristics of the defect. and This indicates that the defect is a near-surface defect with strong electronic perturbations (such as oxygen vacancies on the surface of silicon crystals that simultaneously exhibit and This electron-phonon joint analysis method explains the non-radiative recombination enhancement phenomenon caused by near-surface defects in experimental observations and verifies and The effectiveness of collaborative representation.

[0080] S5: The defect recognition model and defect depth value are integrated, and the Boltzmann entropy weight is used to calculate the defect location confidence. When the defect location confidence exceeds the threshold, the three-dimensional coordinates and defect type label are output.

[0081] Among them, the Boltzmann entropy weight calculation refers to assigning an entropy weight coefficient according to the thermodynamic probability distribution of the defect depth value, and integrating the confidence output of the defect recognition model to generate a position decision factor; the threshold value refers to the confidence threshold pre-set according to the characteristics of the glass material. When the confidence of the defect position exceeds the confidence threshold, it is judged as a valid defect.

[0082] It should be noted that the thermodynamic probability distribution refers to the relative frequency of defect depth values ​​appearing in the statistical ensemble, and is a probability density function obtained by normalization.

[0083] It should also be noted that when determining the pre-set confidence threshold, it is necessary to collect historical inspection data and count the depth distribution characteristics and defect recognition model output results of different glass materials and defect types. By analyzing the data distribution characteristics, the joint probability distribution of defect depth and defect recognition model confidence is established. The classification boundary is optimized based on the statistical learning method. After adjusting the parameters according to the characteristics of different glass types, the verified confidence threshold parameters are written into the configuration file and a dynamic update mechanism is established. The confidence threshold is continuously optimized through online learning to adapt to process changes. For example, differentiated confidence thresholds can be set for different defect types.

[0084] S5.1: Extract an output confidence value from the defect recognition model, wherein the confidence value represents the confidence level of the defect recognition model for a particular defect type (e.g., a value ranging from 0.0 to 1.0). Simultaneously read the defect depth value (in meters or nanometers); It should be noted that the defect depth value is calculated based on the inversion formula of the phonon resonance frequency. The operation includes directly accessing the data storage location in the memory or distributed storage pool.

[0085] S5.2: Use the statistical distribution of defect depth values ​​to determine the thermodynamic probability distribution.

[0086] Specifically, read the historical or measured defect depth value data set (the defect depth value data set is pre-accumulated during the production line operation and stored in the configuration file), construct a frequency distribution histogram of the defect depth value, and divide the depth value intervals; It should be noted that the division of depth value intervals must meet the following requirements: the interval width is consistent (such as 5 nanometer intervals); and the full data range is covered (such as 0-20 nanometers).

[0087] Calculate the probability value of each depth value interval (the value is equal to the ratio of the interval frequency to the total frequency), and calculate the Boltzmann entropy based on the probability value , the formula is: ; Where, is the Boltzmann constant, which is preset to a standard value in the configuration file. is a summation symbol, indicating the cumulative calculation of all possible defect depth intervals. is the probability value of a single defect depth interval ( ), which is calculated from the ratio of the frequency of the defect depth interval to the total frequency, Indicates the probability calculated using the natural logarithm The logarithm of , used to measure the amount of information; Take the Boltzmann entropy In the operation, the current defect depth value is mapped to the corresponding histogram interval, and the probability value of the defect depth interval is extracted. , substituted into the Boltzmann entropy weight coefficient calculation formula, and finally the entropy weight coefficient ranging from 0.0 to 1.0 is obtained.

[0088] For example, in the defect depth data set, 10% of the defects are less than 5 nanometers (p=0.1), 40% are between 5-10 nanometers (p=0.4), 30% are between 10-15 nanometers (p=0.3), and 20% are ≥15 nanometers (p=0.2). The current defect depth of 8 nanometers falls within the [5,10) nanometer interval. Calculate the Boltzmann entropy of the entire distribution, the expression is: ; In taking After simplified calculation, the Boltzmann entropy weight coefficient = 1 / (1+1.279)≈0.44.

[0089] S5.3: Define the position decision factor as the product of the defect recognition model output confidence and the Boltzmann entropy weight coefficient; where the position decision factor integrates the defect recognition model prediction and depth distribution characteristics.

[0090] During operation, the output confidence value of S5.1 and the Boltzmann entropy weight coefficient of S5.2 are read, and a multiplication operation is performed. The result is the position decision factor (the factor value range is 0.0 to 1.0).

[0091] For example: Position decision factor = output confidence × entropy weight coefficient = 0.85 × 0.44 ≈ 0.37.

[0092] S5.4: Calculate the confidence level for the defect location and determine whether it exceeds the threshold.

[0093] Specifically, the defect location confidence is directly expressed using the location decision factor value. The preset confidence threshold value is read (the confidence threshold value is defined in the configuration file based on the glass material properties such as hardness and thickness).

[0094] Compare the location decision factor with the threshold value: If the location decision factor ≥ the threshold value, the defect location confidence is determined to be exceeded and the output operation is performed; otherwise, the current process ends. During the operation, the threshold value is obtained from the configuration file, and the comparison logic is implemented using a simple conditional statement.

[0095] S5.5: Extract the spatial coordinates (derived from the spatiotemporal center coordinates of S2.4, in the format of [X, Y, Z]) from the recombinant defective DNA fragment parsed in step S4.1. Extract the defect type label (e.g., "scratch," "bubble," etc.) from the defect recognition model output.

[0096] It should be noted that the output operation includes writing the coordinates and types into a log file or sending them to the control terminal. The coordinate extraction operation involves reading the decrypted data of S4.1, and the type extraction operation reads the classification results of the defect recognition model.

[0097] For example: 3D coordinates: [120.5, 75.3, 0.2] (unit: mm). Defect type label: Scratches. Output format: "Defect location: [120.5, 75.3, 0.2], Defect type: Scratches."

[0098] Preferably, the entropy weight coefficient is used to dynamically adjust the weight balance between the defect recognition model prediction results and the physical statistical laws, which not only fully utilizes the recognition ability of the defect recognition model, but also combines the statistical prior knowledge of the defect depth distribution, effectively improving the accuracy of distinguishing near-surface defects and bulk defects, and at the same time realizes the structured recording and traceability of defect information through standardized output format.

[0099] This embodiment also provides a computer device suitable for the case of a rapid defect location method based on glass inspection, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the rapid defect location method based on glass inspection proposed in the above embodiment.

[0100] The computer device may be a terminal, comprising a processor, memory, a communication interface, a display, and an input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system and computer programs. The internal memory provides an environment for the operating system and computer programs stored in the non-volatile storage media. The communication interface of the computer device is used to communicate with external terminals via wired or wireless communication. Wireless communication may be achieved via Wi-Fi, a carrier network, NFC (near-field communication), or other technologies. The display of the computer device may be a liquid crystal display or an electronic ink display. The input device may be a touchscreen overlay on the display, buttons, a trackball, or a touchpad on the computer device housing, or an external keyboard, touchpad, or mouse.

[0101] This embodiment further provides a storage medium having a computer program stored thereon. When the program is executed by a processor, the method for rapidly locating defects based on glass inspection as proposed in the above embodiment is implemented. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0102] In summary, the present invention significantly improves the detection accuracy and computational efficiency of submicron glass defects through the dynamic allocation technology of quantum computing resources; through DNA molecular encoding technology, optical features are converted into computable bioinformatics sequences, thereby enhancing the expressive power of defect features; through the federated aggregation technology under homomorphic encryption, the sharing and optimization of defect features across production lines are achieved while ensuring data security; finally, by integrating photon band structure calculation and phonon resonance inversion technology, a physically interpretable defect recognition model is constructed, which significantly improves the accuracy and reliability of defect positioning.

[0103] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for rapid defect location based on glass inspection, characterized by: include, Allocate quantum computing resources based on glass type and production line priority, driving quantum processors to process submicron defect features; Perform DNA molecular encoding operations on submicron-scale defect features, mapping discrete light intensity values ​​to adenine base sequences, polarization scattering angles to thymine base sequences, and phase gradient tensors to guanine base sequences, thereby constructing defective DNA fragments with spatiotemporal properties. The defective DNA fragments are transmitted to the cloud federation aggregation server through a homomorphic encryption channel, and probability-driven cross-mutation and adaptive recombination are performed with the cross-production line encrypted defective DNA fragments; The recombined encrypted defective DNA fragment is expressed as a defect recognition model, and the defect state energy level offset in the photonic band structure is simultaneously calculated, and the defect depth value is inverted according to the phonon resonance frequency. The defect recognition model and defect depth value are integrated, and the Boltzmann entropy weight is used to calculate the defect location confidence. When the defect location confidence exceeds the threshold, the three-dimensional coordinates and defect type label are output.

2. The method for rapid defect location based on glass inspection according to claim 1, wherein: The submicron defect characteristics include light intensity discrete values, polarization scattering angles, and phase gradient tensors; The light intensity discrete value, polarization scattering angle and phase gradient tensor refer to two-dimensional spatial distribution parameters extracted by Fourier transform and polarization interference analysis after collecting optical signals of glass surface defects.

3. The method for rapid defect location based on glass inspection according to claim 1, wherein: The allocation of quantum computing resources specifically includes constructing a mapping table between glass type and number of quantum bits, and dynamically adjusting the priority of the quantum processor task queue based on the real-time load of the production line.

4. The method for rapid defect location based on glass inspection according to claim 1, wherein: The specific steps of constructing a defective DNA fragment containing spatiotemporal properties are as follows: The discrete values ​​of light intensity are mapped to the length of adenine base sequence through quantization segmentation; The polarization scattering angle is mapped to the thymine base sequence by angle interval division; The phase gradient tensor is mapped to the guanine base combination pattern via eigenvalue decomposition; The length of adenine base sequence, the order of thymine base sequence and the combination pattern of guanine base are embedded in the time-space stamp to construct multi-dimensional defective DNA fragments.

5. The method for rapid defect location based on glass inspection according to claim 1, wherein: The homomorphic encryption channel adopts a lattice-based cryptosystem, and the crossover mutation and adaptive recombination are performed through Monte Carlo Markov chain sampling to optimize the probability of fragment exchange.

6. The method for rapid defect location based on glass inspection according to claim 1, wherein: Construct a defect recognition model. The specific steps are as follows: Analyze the number of consecutive adenine base repeats in the recombinant defective DNA fragment and map it to the number of layers in the convolutional neural network; According to the thymine base sequence sorting pattern, the weight initialization matrix of each convolutional layer is generated, and the guanine base combination pattern is used to determine the dimension of the neural network feature channel; The energy level offset of the photon band defect state is calculated based on the parameter-optimized tight-binding model to verify the material defect characteristics. The defect recognition model is generated through dynamic weight fusion by combining the phonon resonance frequency inversion results with the neural network output.

7. The method for rapid defect location based on glass inspection according to claim 1, wherein: The Boltzmann entropy weight calculation refers to assigning an entropy weight coefficient according to the thermodynamic probability distribution of the defect depth value, and fusing the confidence output by the defect recognition model to generate a position decision factor.

8. The method for rapid defect location based on glass inspection according to claim 1, wherein: The threshold value refers to a confidence threshold value preset according to the characteristics of the glass material. When the confidence of the defect position exceeds the confidence threshold value, it is determined to be a valid defect.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for rapid defect location based on glass inspection according to any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for rapid defect location based on glass inspection according to any one of claims 1 to 8 are implemented.

Images