Intelligent detection system for mechanical parts defects based on deep learning

Abstract

The present invention belongs to the technical field of intelligent defect detection and relates to an intelligent defect detection system for mechanical parts based on deep learning. The system is configured with a parameter matrix acquisition module, a mechanical part quantum state data output module, a fractional dimension convolution feature extraction module, a mechanical part defect manifold coordinate extraction module, a manifold transposition submodule calculation module, and a mechanical part defect confidence output module. The present invention solves the problem that traditional mechanical part defect detection is generally based on identification by professional inspectors in the field, manual inspection is highly subjective, the detection results have a high false detection rate, and the detection efficiency and accuracy of manual inspection are low. By integrating multiple aspects of geometric information and decision factors, the system can more comprehensively and reliably assess the possibility of defects in mechanical parts, reduce human intervention, and improve detection efficiency and consistency.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent defect detection and relates to an intelligent defect detection system for mechanical parts based on deep learning. Background Art

[0002] Mechanical equipment is mostly used to replace human labor in high-intensity, repetitive, and highly dangerous work, liberating humans. Therefore, the working environment of mechanical equipment is often relatively harsh. Mechanical equipment is often used to replace human labor in strong alkali, strong acid, high temperature, and high humidity environments.

[0003] When operating for a long time in a harsh environment, the metal components of mechanical equipment are prone to deformation and corrosion, which may cause the mechanical equipment to fail to operate normally or reduce its operating efficiency. In order to prevent the mechanical equipment from stopping or even being scrapped due to defective mechanical parts, traditional mechanical parts defect detection is generally based on identification by professional inspectors in the field. This method has high labor costs, strong subjectivity of manual inspection, high false detection rate of inspection results, and high professional requirements for inspectors. Manual inspection has low efficiency and accuracy. Summary of the Invention

[0004] In response to the shortcomings of the existing technology, the present invention provides a deep learning-based intelligent detection system for mechanical parts defects, which can effectively solve the problem.

[0005] The present invention is achieved through the following technical solutions:

[0006] A deep learning-based intelligent detection system for mechanical part defects includes: a parameter matrix acquisition module for acquiring a material parameter matrix of a mechanical part; a mechanical part quantum state data output module for outputting the mechanical part quantum state data based on a multimodal generation model of quantum field excitation; a fractional dimension convolution feature extraction module for performing fractional dimension convolution feature extraction on the mechanical part quantum state data and performing quantum state decoding to obtain the fractal features of the mechanical part; a mechanical part defect manifold coordinate extraction module for constructing a topological manifold detection network, calculating manifold extreme points, and extracting the mechanical part defect manifold coordinates from the mechanical part fractal features; a manifold transposition subtraction module for calculating the manifold transposition subtraction based on the mechanical part defect manifold coordinates; and a mechanical part defect confidence output module for performing non-commutative geometric decision fusion and outputting the mechanical part defect fusion confidence.

[0007] As a further solution, a multimodal generation model based on quantum field excitation is used to output the quantum state data of mechanical parts. The specific analysis process includes:

[0008] (1) Parameter matrix acquisition module:

[0009] Obtain material parameters of mechanical parts, including the elastic modulus of mechanical parts , Poisson's ratio of mechanical parts , mechanical parts density , yield strength of mechanical parts , fracture toughness of mechanical parts , melting point of mechanical parts ;Construct the material parameter matrix of mechanical parts :

[0010] ; (1)

[0011] Combine the material parameters of mechanical parts with the quantum field, where each material parameter controls a component of the quantum field;

[0012] (2) Mechanical parts quantum state data output module:

[0013] Get the generated network weights stored in the database , generating quantum states through quantum field theory path integrals :

[0014] ; (2)

[0015] Where, is the quantum state function obtained by path integral, describing the quantum state of the system at space x, is the normalization constant, For quantum fields The functional integral measure of , The quantum fields stored in the database The amount of action, To generate the network, e is a natural constant; represents matrix multiplication; It refers to the combination of the material parameter matrix of mechanical parts and quantum fields;

[0016] The quantum state must satisfy the material mechanics equations:

[0017] ; (3)

[0018] Where, is the displacement field, is the Lamé constant stored in the database, is the field nonlinear coefficient stored in the database, is the displacement field About time The second derivative of , which is acceleration;

[0019] Finally, the quantum state data is obtained through projection measurement :

[0020] ; (4)

[0021] Where, is the three-dimensional complex value data obtained by quantum measurement, a, b, c are the indicators in the three-dimensional space, For quantum state The conjugate transpose of is the measurement operator of a, b, c, is the trace of the matrix, a represents the surface morphology, b represents the internal stress, and c represents the crack depth;

[0022] Output quantum state data .

[0023] As a further solution, the material parameters of mechanical parts are combined with quantum fields. The specific analysis process is as follows:

[0024] ; (5)

[0025] Where, is the quantum field at space x, The material parameters of mechanical parts and the rules of quantum field fusion, is the kth component of the quantum field at space x, is the kth component corresponding to the i-th mechanical part material parameter matrix sample, i is the mechanical part material parameter matrix sample number, and k is the quantum field component number corresponding to the mechanical part material parameter.

[0026] (3) Fractional dimension convolution feature extraction module:

[0027] As a further solution, fractional dimension convolution feature extraction is performed on the quantum state data of mechanical parts, and quantum state decoding is performed to obtain the fractal features of mechanical parts. The specific analysis process is as follows: Get the convolution kernel , initialized using the Mittag-Leffler function; obtain the fractional order stored in the database ;

[0028] Fractional differentials:

[0029] ; (6)

[0030] Where, For fractional order The result function of the differential operation, For the sum index, is the gamma function, is the space interval, is the quantum state data;

[0031] Obtain the initial fractal features, that is, the initial features of the fractal pyramid;

[0032] Fractal pyramid construction:

[0033] ; (7)

[0034] Where, It is the fractal pyramid Layer features, It is through With the convolution kernels Perform calculations and The next layer of features of the fractal pyramid obtained after transformation, N is the number of features participating in the construction of the fractal pyramid Calculated convolution kernel the number of is the Hausdorff dimension Related transformation operators, Number the convolution kernel; represents the convolution operation;

[0035] Get the last layer of features of the fractal pyramid ; The last layer of the fractal pyramid features Perform quantum state decoding, decompose the complex features into physical quantities, and obtain the fractal features of mechanical parts .

[0036] As a further solution, quantum state decoding, the specific analysis process is: decomposing the complex characteristics into physical quantities:

[0037] ; (8)

[0038] Where, Encodes geometric features for the real part, is the real part, is the imaginary quantum phase characteristic, is the imaginary part; get the fractal characteristics of mechanical parts , .

[0039] (4) Mechanical parts defect manifold coordinate extraction module:

[0040] As a further solution, a topological manifold detection network is constructed to calculate the manifold extreme points and extract the manifold coordinates of mechanical part defects from the fractal features of mechanical parts. The specific analysis process is as follows: based on X-ray, ultrasound, and infrared modalities, the multi-sensor mechanical part defect manifold coordinates are generated; the fully connected layer weights stored in the database are obtained. , output p-directional metric tensor manifold space weight , p-directional metric tensor manifold space bias ;

[0041] Get initial metrics :

[0042] ; (9)

[0043] in, is the fractal feature of mechanical parts, is the characteristic covariance matrix, p is the component number of the p-direction metric tensor, and q is the component number of the q-direction metric tensor;

[0044] Construct the differentiable manifold equations:

[0045] ; (10)

[0046] in, is the coordinate component in the p-direction metric tensor, and e is a natural constant;

[0047] Improve the manifold shape through Ricci flow:

[0048] ; (11)

[0049] in, represents the Gaussian curvature of the manifold, is the p and q direction metric tensor, t is the time index;

[0050] When Gaussian curvature The iteration is stopped when ; the manifold extreme point m is calculated, the manifold extreme point corresponds to the crack tip or the hole center, and the coordinates of the manifold extreme point are marked as the manifold coordinates of the mechanical part defect.

[0051] As a further solution, calculate the extreme points of the manifold. The specific analysis process is as follows:

[0052] ; (12)

[0053] in, Indicates that the Gaussian curvature on the manifold is calculated The set of points with zero gradient about coordinate x is called the manifold extreme point set;

[0054] Get the coordinates of the manifold extreme point m as ;in, is the x-axis coordinate of the manifold extreme point, is the y-axis coordinate of the manifold extreme point, is the z-axis coordinate of the manifold extreme point, is the defect depth at the extreme point of the manifold.

[0055] (5) Manifold transposition subcomputation module:

[0056] As a further solution, based on the manifold coordinates of mechanical parts defects, the manifold transposition is calculated. The specific analysis process is as follows: the manifold coordinates of mechanical parts defects specifically include the X-ray manifold coordinates of mechanical parts defects , ultrasonic mechanical parts defect manifold coordinates , infrared modal mechanical parts defect manifold coordinates ; Get the Dirac matrix defined by the standard Clifford algebra stored in the database;

[0057] The coordinates of the mechanical part defect manifold are encoded as Clifford numbers to generate a Dirac matrix basis that satisfies:

[0058] ; (13)

[0059] in, For the The Dirac matrix basis of dimensions, For the The Dirac matrix basis of dimensions, is the Minkowski metric, is the identity matrix, is the dimension number, is the dimension number, and the dimensions are X-ray dimension, ultrasound dimension, and infrared modality dimension;

[0060] Computational manifold transposition , determine the dominant direction of the defect, non-zero manifold transposition reveals multimodal data conflicts, the modulus length of the manifold transposition value The larger the value, the more significant the conflict: obtain the defect dominant direction-defect weight factor set mapping set stored in the database, and determine the matching defect weight factor set based on the determined defect dominant direction.

[0061] As a further solution, we calculate the manifold transposition. The specific analysis process is as follows:

[0062] ; (14)

[0063] Where, is the imaginary unit, is the three-dimensional Levi-Civita symbol, is the Dirac matrix basis of the Kth dimension, For the The coordinate component of the R direction of the dimension, For the The coordinate component in the r direction of the dimension, K is the dimension number;

[0064] like In the three-dimensional space, the directions are evenly arranged, and the three-dimensional Levi-Civita sign is equal to 1; if In the odd arrangement in three-dimensional space, the three-dimensional Levi-Civita sign is equal to -1; if If any two directions in three-dimensional space are consistent, the three-dimensional Levi-Civita sign is equal to 0.

[0065] (6) Mechanical parts defect confidence output module:

[0066] As a further solution, non-commutative geometric decision fusion is performed to output the fusion confidence of mechanical part defects. The specific analysis process is as follows:

[0067] Perform non-commutative geometric decision fusion and obtain the confidence level of mechanical part defects through feature decomposition:

[0068] ; (15)

[0069] in, is the confidence level of mechanical parts defects, To take the real part operation, For trace operation, is the chiral projection matrix, is the operation of projecting the confidence into the chiral subspace;

[0070] ; (16)

[0071] ; (17)

[0072] in, For the The Pauli matrix of dimensions, is an imaginary unit;

[0073] Based on the defect confidence of each mechanical part and combined with the matching defect weight factor set, a weighted sum is performed to obtain the matching mechanical part defect fusion confidence ; Output mechanical parts defect fusion confidence .

[0074] Compared with the prior art, the present invention has the following beneficial effects:

[0075] (1) By providing an intelligent detection system for mechanical part defects based on deep learning, the fractional dimension convolution feature extraction module uses fractional dimension convolution to extract features and perform quantum state decoding. Fractional dimension convolution can capture more subtle and complex feature relationships in the data. Compared with traditional integer dimension convolution, it can extract fractal features at a finer scale, dig deeper into features such as the microstructure of mechanical parts, and improve feature representation capabilities. The mechanical part defect manifold coordinate extraction module constructs a topological manifold detection network to calculate the manifold extreme points to extract the defect manifold coordinates. The manifold concept can characterize the intrinsic structure of the data from a geometric perspective. By finding the extreme points to determine the defect location, it can accurately locate the crack tip, hole center, and other defect locations of mechanical parts, providing accurate location information for subsequent repair and evaluation.

[0076] (2) The mechanical part defect confidence output module calculates the manifold permutation based on the defect manifold coordinates and performs non-commutative geometric decision fusion to output the confidence. Non-commutative geometric decision fusion integrates multiple aspects of geometric information and decision factors to more comprehensively and reliably assess the possibility of defects in mechanical parts, providing a more reliable basis for quality control and decision-making. Based on deep learning to realize the functions of each module, the entire system has intelligent processing capabilities. From data acquisition and feature extraction to defect location and confidence assessment, the entire process is automated, reducing manual intervention, improving detection efficiency and consistency, and adapting to the needs of large-scale rapid detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] Figure 1 This is a schematic diagram of system module connections of the present invention.

[0078] Figure 2 Flowchart of the steps for determining a matching set of defect weight factors.

[0079] Figure 3 It is a flow chart of the steps of the present invention. DETAILED DESCRIPTION

[0080] The embodiments of the present invention are described clearly and completely below with reference to the accompanying drawings.

[0081] See also Figure 1 , the embodiment of the present invention provides a technical solution for intelligent detection of mechanical parts defects based on deep learning: Figure 3 As shown in the figure, the intelligent detection system for mechanical parts defects based on deep learning includes a parameter matrix acquisition module, a mechanical parts quantum state data output module, a fractional dimension convolution feature extraction module, a mechanical parts defect manifold coordinate extraction module, a manifold transposition submodule calculation module and a mechanical parts defect confidence output module.

[0082] Parameter matrix acquisition module, obtains the material parameter matrix of mechanical parts;

[0083] The mechanical parts quantum state data output module outputs the mechanical parts quantum state data based on a multimodal generation model of quantum field excitation.

[0084] (1) Parameter matrix acquisition module:

[0085] The specific analysis process includes: obtaining the material parameters of mechanical parts, including the elastic modulus of mechanical parts , Poisson's ratio of mechanical parts , mechanical parts density , yield strength of mechanical parts , fracture toughness of mechanical parts , melting point of mechanical parts ;Construct the material parameter matrix of mechanical parts :

[0086] ; (1)

[0087] Combine the material parameters of mechanical parts with the quantum field, with each material parameter controlling a component of the quantum field;

[0088] (2) Mechanical parts quantum state data output module:

[0089] Get the generated network weights stored in the database , generating quantum states through quantum field theory path integrals :

[0090] ; (2)

[0091] Where, is the quantum state function obtained by path integral, describing the quantum state of the system at space x, is the normalization constant, For quantum fields The functional integral measure of , The quantum fields stored in the database The amount of action, To generate the network, e is a natural constant;

[0092] The quantum state must satisfy the material mechanics equations:

[0093] ; (3)

[0094] Where, is the displacement field, is the Lamé constant stored in the database, is the field nonlinear coefficient stored in the database, is the displacement field About time The second derivative of , which is acceleration;

[0095] Finally, the quantum state data is obtained through projection measurement :

[0096] ; (4)

[0097] Where, is the three-dimensional complex value data obtained by quantum measurement, a, b, c are the indicators in the three-dimensional space, For quantum state The conjugate transpose of is the measurement operator of a, b, c, is the trace of the matrix, a represents the surface morphology, b represents the internal stress, and c represents the crack depth;

[0098] Output quantum state data .

[0099] Combining the material parameters of mechanical parts with quantum fields, the specific analysis process is as follows:

[0100] ; (5)

[0101] Where, is the quantum field at space x, The material parameters of mechanical parts and the rules of quantum field fusion, is the kth component of the quantum field at space x, is the kth component corresponding to the i-th mechanical part material parameter matrix sample, i is the mechanical part material parameter matrix sample number, and k is the quantum field component number corresponding to the mechanical part material parameter.

[0102] Obtaining various material parameters of mechanical parts, such as elastic modulus and Poisson's ratio, to construct a matrix comprehensively covers the physical properties of the materials, providing a rich data foundation for subsequent analysis. This can more completely characterize the properties of mechanical parts and avoid analysis deviations due to missing information.

[0103] Combining material parameters with quantum fields, where each parameter controls a quantum field component, and leveraging quantum field characteristics to explore potential connections between material parameters, data can be integrated from a new perspective, enabling the acquisition of deeper and more unique data features and information compared to traditional methods.

[0104] The multimodal generation model based on quantum field excitation combines the quantum field theory path integral to generate quantum states, and models at the microscopic quantum level, which conforms to modern physics theory. This makes the model's description of the state of mechanical parts more scientific and accurate, and can capture physical phenomena and laws that are difficult to reach with traditional methods.

[0105] Three-dimensional complex quantum state data is obtained through projection measurement, which corresponds to surface morphology, internal stress, crack depth, etc., providing mechanical part status information from multiple dimensions. Compared with single-dimensional detection, it reflects the part condition more comprehensively and is conducive to the comprehensive evaluation of part quality.

[0106] (3) Fractional dimension convolution feature extraction module

[0107] The fractional dimension convolution feature extraction module performs fractional dimension convolution feature extraction on the quantum state data of mechanical parts and performs quantum state decoding to obtain the fractal features of mechanical parts.

[0108] The specific analysis process is: Get the convolution kernel , initialized using the Mittag-Leffler function; obtain the fractional order stored in the database ;

[0109] Fractional differentials:

[0110] ; (6)

[0111] Where, For fractional order The result function of the differential operation, For the sum index, is the gamma function, is the space interval, is the quantum state data;

[0112] Obtain the initial fractal features, that is, the initial features of the fractal pyramid;

[0113] Fractal pyramid construction:

[0114] ; (7)

[0115] Where, It is the fractal pyramid Layer features, It is through With the convolution kernels Perform calculations and The next layer of features of the fractal pyramid obtained after transformation, N is the number of features participating in the construction of the fractal pyramid Calculated convolution kernel the number of is the Hausdorff dimension Related transformation operators, Number the convolution kernel;

[0116] Get the last layer of features of the fractal pyramid ; The last layer of the fractal pyramid features Perform quantum state decoding, decompose the complex features into physical quantities, and obtain the fractal features of mechanical parts .

[0117] Quantum state decoding, the specific analysis process is: decomposing the complex characteristics into physical quantities:

[0118] ; (8)

[0119] Where, Encodes geometric features for the real part, is the real part, is the imaginary quantum phase characteristic, is the imaginary part; get the fractal characteristics of mechanical parts , .

[0120] Fractional convolution and fractional differentiation break through the traditional integer-order limitations, enabling more precise capture of complex and subtle features in quantum state data. Quantum state data contains microscopic information about mechanical components, and fractional operations can mine features at different scales, such as irregularities in the microstructure of materials, enhancing both the depth and breadth of feature extraction.

[0121] Fractal pyramid construction is based on fractal theory, which exhibits self-similar and multi-scale properties. By iteratively utilizing different convolution kernels and combining Hausdorff dimension-related transformations, features can be analyzed and integrated at different scales, comprehensively describing the characteristics of mechanical parts from the macro to the micro perspective, adapting to the analysis needs of different hierarchical structures of parts.

[0122] The convolution kernel is initialized using the Mittag-Leffler function, the concept of fractional order is introduced, and cutting-edge mathematical theories are applied to feature extraction model construction. Compared with traditional convolution methods, this gives the model new characteristics and capabilities. In view of the complex characteristics of quantum state data of mechanical parts, quantum state decoding separates the real and imaginary parts to obtain geometric features and quantum phase features respectively, which conforms to the nature of the data, enabling the model to effectively process and utilize quantum state data, thereby improving the pertinence and effectiveness of feature extraction.

[0123] The final fractal feature F contains real geometric features and imaginary quantum phase features. This clear feature representation provides intuitive and physically meaningful feature information for subsequent mechanical parts defect detection and quality assessment.

[0124] (4) Mechanical parts defect manifold coordinate extraction module:

[0125] The mechanical parts defect manifold coordinate extraction module builds a topological manifold detection network, calculates manifold extreme points, and extracts the mechanical parts defect manifold coordinates from the fractal features of the mechanical parts.

[0126] The specific analysis process is as follows: Generate multi-sensor mechanical part defect manifold coordinates based on X-ray, ultrasound, and infrared modalities; Obtain the fully connected layer weights stored in the database , output p-directional metric tensor manifold space weight , p-directional metric tensor manifold space bias ;

[0127] Get initial metrics :

[0128] ; (9)

[0129] in, is the fractal feature of mechanical parts, is the characteristic covariance matrix, p is the component number of the p-direction metric tensor, and q is the component number of the q-direction metric tensor;

[0130] Construct the differentiable manifold equations:

[0131] ; (10)

[0132] in, is the coordinate component in the p-direction metric tensor;

[0133] Improve the manifold shape through Ricci flow:

[0134] ; (11)

[0135] in, represents the Gaussian curvature of the manifold, is the p and q direction metric tensor, t is the time index;

[0136] When Gaussian curvature The iteration is stopped when ; the manifold extreme point m is calculated, the manifold extreme point corresponds to the crack tip or the hole center, and the coordinates of the manifold extreme point are marked as the manifold coordinates of the mechanical part defect.

[0137] Calculate the manifold extreme points. The specific analysis process is as follows:

[0138] ; (12)

[0139] in, Indicates that the Gaussian curvature on the manifold is calculated The set of points with zero gradient about coordinate x is called the manifold extreme point set;

[0140] Get the coordinates of the manifold extreme point m as ;in, is the x-axis coordinate of the manifold extreme point, is the y-axis coordinate of the manifold extreme point, is the z-axis coordinate of the manifold extreme point, is the defect depth at the extreme point of the manifold.

[0141] Generate manifold coordinates for mechanical part defects using multiple modalities, including X-ray, ultrasound, and infrared. Different modalities can perceive part conditions from multiple angles. X-rays can detect internal structural defects, ultrasound is sensitive to cracks, and infrared can detect defects reflected by temperature anomalies. Multimodal fusion combines the advantages of each modality, reducing the limitations and misjudgments of single-modality detection and providing comprehensive and accurate defect information.

[0142] A topological manifold detection network is constructed, leveraging the concept of manifolds to describe the characteristics and defect locations of mechanical parts from a geometric perspective. Manifolds can characterize the inherent structure of data, mapping the fractal characteristics of mechanical parts into manifold space. This allows defects to be detected from changes in geometric structure, providing a new perspective and theoretical framework for defect detection and revealing the fundamental relationship between defects and part structure.

[0143] The fully connected layer weights are used to output the manifold space weights and biases, obtaining an initial metric based on the feature covariance matrix. The fully connected layer effectively transforms features and determines the manifold space parameters; the initial metric provides the foundation for the manifold space geometry. Reasonable parameter and metric settings enable the model to more accurately describe the manifold characteristics, laying a good foundation for subsequent calculations.

[0144] A differentiable manifold equation is constructed and the manifold shape is refined using Ricci flow. The differentiable manifold equation defines the mathematical form of the manifold. Ricci flow dynamically adjusts the manifold based on Gaussian curvature, making it approximate the actual part structure and defect distribution. Iterations are terminated when Gaussian curvature meets certain conditions, ensuring convergence of the manifold shape optimization and accurate reflection of defect locations and characteristics.

[0145] Defect locations are determined by calculating manifold extreme points, which correspond to key defect locations such as crack tips or hole centers. This method accurately captures defect locations and outputs information including spatial coordinates and defect depth, providing precise location information for mechanical part repair and quality assessment, and is more accurate than traditional positioning methods.

[0146] The manifold transposition calculation module calculates the manifold transposition based on the manifold coordinates of mechanical part defects.

[0147] The mechanical part defect confidence output module performs non-commutative geometric decision fusion and outputs the mechanical part defect fusion confidence.

[0148] (5) Manifold transposition subcomputation module:

[0149] like Figure 2 As shown, the matching defect weight factor set is determined. The specific analysis process is as follows: the mechanical part defect manifold coordinates specifically include the X-ray mechanical part defect manifold coordinates , ultrasonic mechanical parts defect manifold coordinates , infrared modal mechanical parts defect manifold coordinates ; Get the Dirac matrix defined by the standard Clifford algebra stored in the database;

[0150] The coordinates of the mechanical part defect manifold are encoded as Clifford numbers to generate a Dirac matrix basis that satisfies:

[0151] ; (13)

[0152] in, For the The Dirac matrix basis of dimensions, For the The Dirac matrix basis of dimensions, is the Minkowski metric, is the identity matrix, is the dimension number, is the dimension number, and the dimensions are X-ray dimension, ultrasound dimension, and infrared modality dimension;

[0153] Computational manifold transposition , determine the dominant direction of the defect, non-zero manifold transposition reveals multimodal data conflicts, the modulus length of the manifold transposition value The larger the value, the more significant the conflict: obtain the defect dominant direction-defect weight factor set mapping set stored in the database, and determine the matching defect weight factor set based on the determined defect dominant direction.

[0154] Manifold transposition, the specific analysis process is:

[0155] ; (14)

[0156] Where, is the imaginary unit, is the three-dimensional Levi-Civita symbol, is the Dirac matrix basis of the Kth dimension, For the The coordinate component of the R direction of the dimension, For the The coordinate component in the r direction of the dimension, K is the dimension number;

[0157] like In the three-dimensional space, the directions are evenly arranged, and the three-dimensional Levi-Civita sign is equal to 1; if In the odd arrangement in three-dimensional space, the three-dimensional Levi-Civita sign is equal to -1; if If any two directions in three-dimensional space are consistent, the three-dimensional Levi-Civita sign is equal to 0.

[0158] (6) Mechanical parts defect confidence output module:

[0159] Perform non-commutative geometric decision fusion and output the fusion confidence of mechanical part defects. The specific analysis process is as follows:

[0160] Perform non-commutative geometric decision fusion and obtain the confidence level of mechanical part defects through feature decomposition:

[0161] ; (15)

[0162] in, is the confidence level of mechanical parts defects, To take the real part operation, For trace operation, is the chiral projection matrix, is the operation of projecting the confidence into the chiral subspace;

[0163] ; (16)

[0164] ; (17)

[0165] in, For the The Pauli matrix of dimensions, is an imaginary unit;

[0166] Based on the defect confidence of each mechanical part and combined with the matching defect weight factor set, a weighted sum is performed to obtain the matching mechanical part defect fusion confidence ; Output mechanical parts defect fusion confidence .

[0167] The integrated defect manifold coordinates from multiple modalities, including X-ray, ultrasound, and infrared, comprehensively consider information from different part inspection methods. Different modalities have varying sensitivities and priorities for defects. Fusion enables more complete capture of defect characteristics, reducing the incompleteness of single-modality inspection and improving defect judgment accuracy.

[0168] The calculation of the manifold transposition factor determines the dominant direction of defects and can detect conflicts in multimodal data. A non-zero transposition factor indicates data discrepancies, and its value reflects the significance of the conflict. This helps analyze contradictions between different test data, clarify the manifestation of defects in various dimensions and the main impact direction, and provide key information for accurate defect assessment.

[0169] Using the Dirac matrix defined by standard Clifford algebra, the coordinates of the defect manifold are encoded into a Clifford number-generating matrix basis. Clifford algebra has powerful descriptive capabilities in fields such as geometry and physics. It can uniformly process multimodal data from both algebraic and geometric perspectives, providing a more abstract and universal mathematical framework for defect analysis, enhancing the depth and theoretical nature of the analysis.

[0170] Confidence is obtained through non-commutative geometric decision fusion. Non-commutative geometry has unique advantages in handling complex spatial and data relationships. It comprehensively considers the non-commutative characteristics and geometric structure of multimodal data, fusing information at a more fundamental level. Compared with traditional methods, it can more rationally integrate multi-source data and make confidence calculation more scientific.

[0171] Defect confidence levels are calculated through eigenvalue decomposition and other calculations to quantify the likelihood of a part defect. This specific value allows operators to intuitively assess part quality, providing a clear basis for determining whether a part is acceptable and requires further inspection or repair.

[0172] If the mechanical part defect fusion confidence is higher than the mechanical part defect fusion confidence threshold stored in the database, and the manifold extreme point defect depth is higher than the manifold extreme point defect depth threshold stored in the database, it indicates that the mechanical part defect is serious and the detected mechanical part defect has high credibility, and further inspection and repair are required.

Claims

1. A deep learning-based intelligent detection system for mechanical parts defects, characterized by: include: Parameter matrix acquisition module: obtains the material parameter matrix of mechanical parts; Mechanical parts quantum state data output module: Based on the multimodal generation model of quantum field excitation, it outputs the quantum state data of mechanical parts; Fractional dimension convolution feature extraction module: performs fractional dimension convolution feature extraction on the quantum state data of mechanical parts, and performs quantum state decoding to obtain the fractal features of mechanical parts; Mechanical Parts Defect Manifold Coordinate Extraction Module: Builds a topological manifold detection network, calculates manifold extreme points, and extracts mechanical parts defect manifold coordinates from the fractal features of mechanical parts; Manifold transposition calculation module: calculates the manifold transposition based on the manifold coordinates of mechanical part defects; Mechanical part defect confidence output module: performs non-commutative geometric decision fusion and outputs mechanical part defect fusion confidence; The fractional dimension convolution feature extraction and quantum state decoding of the mechanical parts quantum state data are performed to obtain the fractal features of the mechanical parts. The specific analysis process is as follows: (1) Fractional dimension convolution feature extraction: Get the convolution kernel , initialized using the Mittag-Leffler function; Get the fractional order stored in the database ; Fractional differentials: ； （6）； Where, For fractional order The result function of the differential operation, For the sum index, is the gamma function, is the space interval, is the quantum state data; (2) Fractal pyramid construction: Obtain the initial fractal features, that is, the initial features of the fractal pyramid; Fractal pyramid construction: ；（7）； Where, It is the fractal pyramid Layer features, It is through With the convolution kernels Perform calculations and The next layer of features of the fractal pyramid obtained after transformation, N is the number of features participating in the construction of the fractal pyramid Calculated convolution kernel the number of is the Hausdorff dimension Related transformation operators, Number the convolution kernel; represents the convolution operation; Get the last layer of features of the fractal pyramid ; The last layer of the fractal pyramid Perform quantum state decoding, decompose complex features into physical quantities, and obtain fractal features of mechanical parts ; The manifold transposition is calculated based on the manifold coordinates of the mechanical part defect. The specific analysis process is as follows: Mechanical parts defect manifold coordinates specifically include X-ray mechanical parts defect manifold coordinates , ultrasonic mechanical parts defect manifold coordinates , infrared modal mechanical parts defect manifold coordinates ; Get the Dirac matrix defined by the standard Clifford algebra stored in the database; The coordinates of the mechanical part defect manifold are encoded as Clifford numbers, generating a Dirac matrix basis that satisfies: ；（13）； in, For the The Dirac matrix basis of dimensions, For the The Dirac matrix basis of dimensions, is the Minkowski metric, is the identity matrix, is the dimension number, is the dimension number, and the dimensions are X-ray dimension, ultrasound dimension, and infrared modality dimension; Computational manifold transposition , determine the dominant direction of the defect, non-zero manifold transposition reveals multimodal data conflicts, the modulus length of the manifold transposition value The bigger it is, the more significant the conflict is; A defect dominant direction-defect weight factor set mapping set stored in a database is obtained, and a matching defect weight factor set is determined based on the determined defect dominant direction.

2. The deep learning-based intelligent detection system for mechanical parts defects according to claim 1, characterized in that: The parameter matrix acquisition module is used to obtain the material parameter matrix of mechanical parts. The specific analysis process includes: Obtain material parameters of mechanical parts, including the elastic modulus of mechanical parts , Poisson's ratio of mechanical parts , mechanical parts density , yield strength of mechanical parts , fracture toughness of mechanical parts , melting point of mechanical parts ; Constructing material parameter matrix of mechanical parts : ；（1）； Combine the material parameters of mechanical parts with the quantum field, where each material parameter controls a component of the quantum field; The multimodal generation model based on quantum field excitation outputs quantum state data of mechanical parts. The specific analysis process includes: Get the generated network weights stored in the database , generating quantum states through quantum field theory path integrals : ；（2）； Where, is the quantum state function obtained by path integral, describing the quantum state of the system at space x, is the normalization constant, For quantum fields The functional integral measure of , The quantum fields stored in the database The amount of action, To generate the network, e is a natural constant; represents matrix multiplication; It refers to the combination of the material parameter matrix of mechanical parts and quantum fields; The quantum state must satisfy the material mechanics equations: ；（3）； Where, is the displacement field, is the Lamé constant stored in the database, is the field nonlinear coefficient stored in the database, is the displacement field About time The second derivative of , which is acceleration; Finally, the quantum state data is obtained through projection measurement : ；（4）； Where, is the three-dimensional complex value data obtained by quantum measurement, a, b, c are the indicators in the three-dimensional space, For quantum state The conjugate transpose of is the measurement operator of a, b, c, is the trace of the matrix, a represents the surface morphology, b represents the internal stress, and c represents the crack depth; Output quantum state data .

3. The deep learning-based intelligent detection system for mechanical parts defects according to claim 2, characterized in that: The specific analysis process of combining the material parameters of mechanical parts with quantum fields is as follows: ；（5）； Where, is the quantum field at space x, The material parameters of mechanical parts and the rules of quantum field fusion, is the kth component of the quantum field at space x, is the kth component corresponding to the i-th mechanical part material parameter matrix sample, i is the mechanical part material parameter matrix sample number, and k is the quantum field component number corresponding to the mechanical part material parameter.

4. The deep learning-based intelligent detection system for mechanical parts defects according to claim 1, characterized in that: The specific analysis process of the quantum state decoding is as follows: Decompose complex features into physical quantities: ；（8）； Where, Encodes geometric features for the real part, is the real part, is the imaginary quantum phase characteristic, is the imaginary part; Obtaining fractal features of mechanical parts , .

5. The deep learning-based intelligent detection system for mechanical parts defects according to claim 1, characterized in that: The topological manifold detection network is constructed, the manifold extreme points are calculated, and the manifold coordinates of the mechanical part defects are extracted from the fractal features of the mechanical part. The specific analysis process is as follows: (1) Constructing a topological manifold monitoring network: Generate multi-sensor manifold coordinates of mechanical part defects based on X-ray, ultrasound, and infrared modalities; Get the fully connected layer weights stored in the database , output p-directional metric tensor manifold space weight , p-directional metric tensor manifold space bias ; Get initial metrics : ； （9）； in, is the fractal feature of mechanical parts, is the characteristic covariance matrix, p is the component number of the p-direction metric tensor, and q is the component number of the q-direction metric tensor; Construct the differentiable manifold equations: ；（10）； in, is the coordinate component in the p-direction metric tensor, and e is a natural constant; (2) Manifold optimization and defect location: Improve the manifold shape through Ricci flow: ；（11）； in, represents the Gaussian curvature of the manifold, is the p and q direction metric tensor, t is the time index; When Gaussian curvature Stop iteration when Calculate the manifold extreme point m, which corresponds to the crack tip or the center of the hole, and mark the coordinates of the manifold extreme point as the manifold coordinates of the mechanical part defect.

6. The deep learning-based intelligent detection system for mechanical parts defects according to claim 5, characterized in that: The specific analysis process of calculating the manifold extreme point is as follows: ； （12）； in, Indicates that the Gaussian curvature on the manifold is calculated The set of points with zero gradient about coordinate x is called the manifold extreme point set; Get the coordinates of the manifold extreme point m as ; in, is the x-axis coordinate of the manifold extreme point, is the y-axis coordinate of the manifold extreme point, is the z-axis coordinate of the manifold extreme point, is the defect depth at the extreme point of the manifold.

7. The deep learning-based intelligent detection system for mechanical parts defects according to claim 1, characterized in that: The specific analysis process of the calculation manifold transposition is as follows: ； （14）； Where, is the imaginary unit, is the three-dimensional Levi-Civita symbol, is the Dirac matrix basis of the Kth dimension, For the The coordinate component in the R direction of the dimension, For the The coordinate component in the r direction of the dimension, K is the dimension number; like In three-dimensional space, the directions are evenly arranged, and the three-dimensional Levi-Civita sign is equal to 1; like In the three-dimensional space, the direction of the odd arrangement is odd, and the three-dimensional Levi-Civita sign is equal to -1; like If any two directions in three-dimensional space are consistent, the three-dimensional Levi-Civita sign is equal to 0.

8. The deep learning-based intelligent detection system for mechanical parts defects according to claim 7, characterized in that: The non-commutative geometric decision fusion is performed to output the fusion confidence of mechanical part defects. The specific analysis process is as follows: (1) Perform non-commutative geometric decision fusion: Perform non-commutative geometric decision fusion and obtain the confidence level of mechanical part defects through feature decomposition: ； （15）； in, is the confidence level of mechanical parts defects, To take the real part operation, For trace operation, is the chiral projection matrix, is the operation of projecting the confidence into the chiral subspace; ；（16）； ；（17）； in, For the The Pauli matrix of dimensions, is an imaginary unit; (2) Output mechanical parts defect fusion confidence: Based on the defect confidence of each mechanical part and combined with the matching defect weight factor set, a weighted sum is performed to obtain the matching mechanical part defect fusion confidence .

Images