Steel structure net rack welding defect magnetic memory detection method and system

Abstract

This invention provides a magnetic memory detection method and system for welding defects in steel structure space frames. It combines differential geometry theory with magnetic memory detection technology, using a first finite element model to predict welding defects and a second finite element model to monitor the actual welding state. The weld area is represented as a differential manifold, and the magnetic field distribution is represented as a tensor field. Gradient tensors are calculated and multi-scale analysis is performed. This invention establishes a correlation table between defect types and severity, uses comprehensive indicators to quantitatively evaluate defects, and employs an integrated system of a three-dimensional robotic arm and magnetic sensors for automated detection. This invention achieves improved detection accuracy, defect prevention, full-spectrum detection, and quantitative evaluation, providing a novel solution for the quality control of steel structure space frame welding.

Description

Technical Field

[0001] This invention relates to the field of welding quality inspection technology, specifically to a magnetic memory detection method and system for welding defects in steel structure space frames, and particularly to a technology for accurate identification and evaluation of welding defects based on differential geometry theory. Background Technology

[0002] As a key load-bearing structure in modern large-scale buildings and infrastructure, the welding quality of steel space frames directly affects the safety and service life of the overall structure. Defects such as porosity, slag inclusions, and lack of fusion that may occur during the welding process, if not detected and addressed in a timely manner, will become potential hazards to structural safety.

[0003] Traditional methods for detecting welding defects mainly include ultrasonic testing, X-ray testing, and penetrant testing. Although these methods are widely used in engineering practice, they still have some limitations: ultrasonic testing requires high surface roughness and the test results are easily affected by human factors; X-ray testing has strict environmental requirements and poses radiation safety issues; penetrant testing is only suitable for surface-opening defects and cannot detect internal defects.

[0004] In recent years, magnetic memory testing technology, as a non-destructive testing method, has gradually gained attention due to its advantages such as being non-contact, requiring no removal of the protective layer, and being able to detect stress concentration areas. Traditional magnetic memory testing technology is based on the magnetization characteristics of metallic materials under an applied magnetic field, determining the location of stress concentration areas and defects within the material by detecting changes in the magnetic field. However, traditional magnetic memory testing technology has the following problems:

[0005] 1. Limited detection accuracy and weak ability to identify minute defects;

[0006] 2. It lacks predictive capabilities; it can only detect defects that have already formed, but cannot prevent defects from occurring.

[0007] 3. The data processing method is simplistic, mainly relying on scalar analysis of magnetic field strength, and does not fully utilize the vector characteristics of the magnetic field;

[0008] 4. Defect evaluation lacks quantitative standards and mainly relies on experience-based judgment, which is highly subjective.

[0009] Therefore, there is an urgent need for a magnetic memory detection method that can improve detection accuracy, achieve defect prediction, make full use of magnetic field vector characteristics, and provide quantitative evaluation, so as to meet the actual needs of steel structure space frame welding quality inspection. Summary of the Invention

[0010] The purpose of this invention is to provide a magnetic memory detection method and system for welding defects in steel structure space frames. By combining differential geometry theory with magnetic memory detection technology, it achieves high-precision identification, prediction and quantitative evaluation of welding defects.

[0011] This invention proposes a magnetic memory detection method for welding defects in steel structure space frames, including:

[0012] Obtain structural information, material information, and welding parameters of the welded steel structure space frame component to be tested;

[0013] Based on the structural information, the material information, and the welding parameters, a first finite element model is constructed, and finite element excitation is applied to the first finite element model. By analyzing the stress distribution law in the weld zone, the possible defect types of the weld are predicted.

[0014] During the welding process, weld feature scanning is performed to obtain welding position signals and weld feature information, and a second finite element model is constructed.

[0015] In the second finite element model, based on the weld feature information, a differential manifold characterizing the weld region is constructed, the magnetic field distribution is characterized as a tensor field on the differential manifold, and the gradient tensor of the tensor field is calculated.

[0016] Multi-scale analysis is performed on the gradient tensor to construct a gradient tensor image, and the defect location of the test piece is identified from the gradient tensor image.

[0017] Establish a data table of defect types and defect severity, evaluate the identified defect locations based on the data table, and generate defect evaluation results;

[0018] The defect evaluation results are processed and displayed to determine the degree of defect in the weld under preset conditions.

[0019] Preferably, after constructing the first finite element model, the method further includes:

[0020] Based on the prediction results of the first finite element model, it is determined whether the welding operation is normal. If there are weld defects, the welding parameters are adjusted according to the defect type.

[0021] The adjusted welding parameters are bound to the structural information, the material information, and the defect type, and the prefabricated files stored in the database are updated.

[0022] Preferably, the construction of the differential manifold characterizing the weld region includes:

[0023] The weld area is converted into a parameter space through a coordinate mapping function to form a parameterized representation of the area.

[0024] Establish a tangent space at each point in the weld region, define a local coordinate basis, and form a metric tensor.

[0025] Establish metric-compatible connections to ensure that the inner product remains unchanged during the parallel transmission of the tensor field on the differential manifold;

[0026] Calculate the fundamental geometric quantities of the differential manifold, including the first and second fundamental forms.

[0027] Preferably, characterizing the magnetic field distribution as a tensor field on the differential manifold includes:

[0028] Discrete magnetic field data are obtained by sampling the magnetic field on the differential manifold using a magnetic sensor.

[0029] A continuous tensor field is constructed based on the discrete magnetic field data, and an interpolation method that preserves the differential structure is adopted.

[0030] Calculate the covariant derivative of the tensor field on the differential manifold to form the gradient tensor;

[0031] The gradient tensor is subjected to eigenvalue decomposition to obtain the principal direction and eigenvalues.

[0032] Preferably, the multi-scale analysis of the gradient tensor includes:

[0033] A multi-scale representation is generated by applying a scale transformation group to the original tensor field.

[0034] Based on the feature response intensity, the optimal feature scale is selected for each region;

[0035] Construct a representation method that can adaptively adjust the scaling parameters;

[0036] Features at different scales are integrated to form a multi-resolution feature description.

[0037] Preferably, the construction of the gradient tensor image includes:

[0038] Discretize the continuous tensor field into tensor values ​​at grid points;

[0039] Apply a smoothing filter to the tensor values ​​to reduce the impact of noise;

[0040] Valid tensor data points are filtered according to a preset gradient range;

[0041] The tensor data points are sorted to form a gradient tensor set;

[0042] Based on preset conditions, a gradient tensor set is formed;

[0043] A tensor image is formed based on the gradient tensor set;

[0044] Obtain the attribute matrix of the tensor image, where the elements of the attribute matrix are the feature parameters of the gradient tensors in the gradient tensor group;

[0045] The feature parameters in the attribute matrix are binarized.

[0046] Preferably, the establishment of the data table for defect types and defect severity includes:

[0047] Multiple defect records are generated by labeling each defect type, with each record including the degree of defect corresponding to a particular type.

[0048] The defect severity corresponding to each defect record is normalized.

[0049] A correlation coefficient is set, which is formed by the weighted sum of the product of the defect type and the corresponding defect severity and the coefficient.

[0050] The defect records are sorted according to the correlation coefficient;

[0051] Based on preset conditions, the gradient tensors corresponding to the degree of defects are selected to form a set of gradient tensors for each defect record.

[0052] Based on the number of gradient tensors in the set, a baseline gradient tensor is set to form a defect severity feature model for the defect type.

[0053] Preferably, the defect evaluation of the identified defective parts includes:

[0054] The defect records in the gradient tensor image are matched with the data table to obtain the corresponding defect type and defect severity.

[0055] Based on the curvature flow characteristics of the defect region, singularities and critical points in the flow field are analyzed.

[0056] A critical graph is constructed to represent the topology of the flow field, dividing the flow field into different homotopy categories, corresponding to different types of defects;

[0057] Calculate the persistence of topological features to distinguish between real defects and noise;

[0058] Construct an evaluation model for the degree of defects, define a comprehensive index T, and determine a defective area when T is greater than a preset threshold;

[0059] The defective part is evaluated according to the evaluation model, and the defect evaluation result is generated.

[0060] Preferably, the process of processing and displaying the defect evaluation results includes:

[0061] Cluster analysis was performed on the defect evaluation results;

[0062] The clustering results are then normalized.

[0063] Based on the processed clustering results, data compression is performed;

[0064] The clustering results after data compression are visualized.

[0065] Based on the clustering results, a defect evaluation region map is generated;

[0066] The gradient tensor image and the defect evaluation result are fused and superimposed on the defect evaluation region map.

[0067] A magnetic memory detection system for welding defects in steel space frames includes:

[0068] Magnetic sensor, data acquisition system, 3D robotic arm and host computer;

[0069] The magnetic sensor is fixed on the three-dimensional robotic arm and is used to collect magnetic field data in the weld area;

[0070] The data acquisition system includes an FPGA data acquisition system and a computer data acquisition system. The FPGA data acquisition system is responsible for data acquisition, transmission and control, while the computer data acquisition system is responsible for recording and processing the pose information of the three-dimensional robotic arm and the weld seam.

[0071] The three-dimensional robotic arm communicates wirelessly with the host computer, and the host computer controls the acquisition position of the magnetic sensor and the storage of the acquired data.

[0072] The host computer is equipped with a feature acquisition module, a defect identification module, a defect prediction and analysis module, a calculation module, a signal processing module, and a signal display module;

[0073] The feature acquisition module is used to acquire the structural information, material information, and welding parameters of the steel structure space frame welded test piece;

[0074] The defect identification module is used to construct a first finite element model based on the structural information, the material information and the welding parameters, and to apply finite element excitation to the first finite element model to predict the types of defects that may occur in the weld.

[0075] The defect prediction and analysis module is used to construct a second finite element model during the welding process, characterize the weld area as a differential manifold, and calculate the gradient tensor of the magnetic field distribution.

[0076] The calculation module is used to perform multi-scale analysis on the gradient tensor, construct a gradient tensor image, and identify the defect location of the test piece from the gradient tensor image;

[0077] The signal processing module is used to establish a data table of defect types and defect severity, and to evaluate the identified defect locations based on the data table.

[0078] The signal display module is used to process and display the defect evaluation results, and to determine the degree of defect of the weld under preset conditions.

[0079] The present invention has the following beneficial effects:

[0080] 1. Improved detection accuracy: By introducing differential geometry theory, the weld area is regarded as a differential manifold and the magnetic field distribution is regarded as a tensor field, which fully captures the spatial variation characteristics of the magnetic field, improving the detection accuracy by more than 80% compared with traditional methods, and can identify micro-cracks with a width as small as 0.1mm.

[0081] 2. Predictive and preventive capabilities: A first finite element model is established to conduct pre-welding predictive analysis, and welding parameters are adjusted based on the prediction results to achieve a shift from post-detection to predictive prevention, thereby reducing the incidence of welding defects by 60%.

[0082] 3. Multi-scale analysis capability: By introducing multi-scale analysis technology, defect features are captured at different scales, solving the problem of inconsistent sensitivity of traditional methods to defects of different scales, and realizing full-spectrum detection from micro-cracks to large-area defects.

[0083] 4. Quantitative evaluation mechanism: Establish a data table showing the correlation between defect types and severity, define comprehensive indicator evaluation standards, provide objective and quantitative defect evaluation results, and reduce subjective judgment factors.

[0084] 5. System integration optimization: The integrated system, which uses a three-dimensional robotic arm and a magnetic sensor to work together, realizes the automation and high efficiency of the detection process, and the detection speed is 3 times faster than the traditional method. Attached Figure Description

[0085] Figure 1 This is a flowchart of the magnetic memory detection method for welding defects in steel structure space frames provided by the present invention;

[0086] Figure 2 This is a structural block diagram of the magnetic memory detection system for welding defects in steel structure space frames provided by the present invention. Detailed Implementation

[0087] Please refer to Figure 1 - Figure 2 The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and should not be construed as limiting the scope of protection of the present invention.

[0088] Please refer to Figure 1This invention provides a magnetic memory detection method for welding defects in steel structure space frames, comprising the following steps:

[0089] First, the structural information, material information, and welding parameters of the steel structure space frame to be welded are obtained. In one embodiment of the present invention, this information can be obtained in the following ways: structural information is imported through CAD design drawings, including weld geometry, location distribution, and welding direction; material information includes the steel grade, chemical composition, mechanical properties, and magnetic properties, which can be obtained through material certificates or material testing; welding parameters include welding current, voltage, speed, welding wire diameter, and welding method, which can be obtained from welding process cards or welding equipment control systems.

[0090] Next, based on the acquired structural information, material information, and welding parameters, a first finite element model is constructed, and finite element excitation is applied to this model. By analyzing the stress distribution in the weld zone, the types of defects that may occur in the weld are predicted. Preferably, the construction process of the first finite element model includes: discretizing the structural geometric model into a finite element mesh; assigning material properties to the mesh elements; setting boundary conditions and load conditions; and applying a finite element solver to calculate the stress distribution. By analyzing the stress distribution, potential defect types can be predicted. For example, when the stress concentration factor exceeds 3.0, cracks are predicted to occur; when the temperature gradient exceeds 400°C / mm, porosity is predicted to occur; and when the cooling rate exceeds 50°C / s, hardened structures are predicted to form.

[0091] In embodiments of the present invention, if the first finite element model predicts the risk of weld defects, the welding parameters are adjusted according to the defect type. For example, for predicted potential porosity defects, the welding speed can be reduced by 10%–20%, and the preheating temperature increased to 150–200°C; for predicted potential cracks, the welding current can be reduced by 20–30A, and the post-heat treatment time increased to 1–2 hours. The adjusted welding parameters are bound to structural information, material information, and defect type, updating the prefabricated files stored in the database, forming a closed-loop system for welding parameter optimization.

[0092] During the welding process, weld feature scanning is performed to acquire welding position signals and weld feature information, thus constructing a second finite element model. Here, weld feature scanning is achieved by a three-dimensional robotic arm controlling a magnetic sensor to move along the weld trajectory, with a sampling frequency of 50–100 Hz, capable of capturing minute changes in the magnetic field. The acquired welding position signals include three-dimensional coordinate data with an accuracy better than ±0.1 mm; the weld feature information includes data such as magnetic field strength, direction, and gradient.

[0093] In the second finite element model, based on the acquired weld feature information, a differential manifold characterizing the weld region is constructed. The magnetic field distribution is represented as a tensor field on this differential manifold, and the gradient tensor of the tensor field is calculated. The construction of the differential manifold is one of the key innovations of this invention, and its specific implementation is as follows:

[0094] First, the weld region is transformed into a parameter space using a coordinate mapping function, forming a parameterized representation of the region. This mapping function can be expressed as:

[0095] ,

[0096] in: For the mapping function, from the two-dimensional parameter space Mapped to three-dimensional physical space ; and The coordinates are parameters, and their values ​​typically range from 1 to 2. ; The physical coordinates of the weld area are in millimeters. .

[0097] In practical applications, a parametric method suitable for the weld geometry can be selected. For example, for straight welds, a simple linear mapping can be used; for curved welds, spline functions or NURBS surfaces can be used.

[0098] Then, a tangent space is established at each point in the weld region, and a local coordinate basis is defined to form a metric tensor. The basis vectors of the tangent space can be represented as:

[0099] , ,

[0100] in: For parameters Tangent vector in direction; For parameters Tangent vector in direction; Represents mapping function For parameters The partial derivatives; Represents mapping function For parameters The partial derivatives of .

[0101] The metric tensor is represented as:

[0102] ,

[0103] in: To measure the components of a tensor, it is a Matrix; and Parameters and Tangent vector in direction; This represents the dot product operation. The metric tensor describes the local deformation from the parameter space to the physical space and is the foundation for subsequent differential geometric analysis.

[0104] Next, a metric-compatible connection is established to ensure that the inner product remains invariant during parallel propagation of the tensor field on the differentiable manifold. The connection is represented by Christoffel notation:

[0105] ,

[0106] in: The Christoffel notation indicates the direction of the parameter. and In parallel transmission, in direction Rate of change on; To measure the inverse of a tensor, satisfying ,in For the Kroneckerdelta function (when (1 if it is 1, otherwise 0). Represents the metric tensor components coordinates The partial derivatives of .

[0107] The summation convention in the formula refers to the summation of repeated indicators. Perform summation, with the summation range being... The introduction of connections allows us to define covariant derivatives on surfaces and correctly calculate the rate of change of tensor fields.

[0108] Finally, the fundamental geometric quantities of the differentiable manifold are calculated, including the first and second fundamental forms.

[0109] The first basic form is represented as:

[0110] ,

[0111] in: This is the first fundamental form, describing the metric properties of a surface; , and To measure the components of a tensor; and This refers to a small change in the parameter.

[0112] The second basic form is expressed as:

[0113] ,

[0114] in: This is the second basic form, describing the degree of curvature of the surface; , and The coefficients for the second basic form can be calculated using the following formula:

[0115] ,

[0116] in: The unit normal vector of the surface. ; , and They represent mapping functions respectively. The second-order partial derivative; This represents the cross product operation; It represents the magnitude of the vector.

[0117] These basic forms describe the intrinsic and extrinsic geometric properties of differentiable manifolds, providing a mathematical foundation for subsequent analysis.

[0118] On the constructed differential manifold, the magnetic field distribution is characterized as a tensor field, processed through the following steps:

[0119] First, discrete magnetic field data is acquired by sampling the magnetic field on the differential manifold using a magnetic sensor. In practical applications, the sampling point density is 4–9 points per square centimeter to ensure sufficient spatial resolution. The magnetic sensor has a sensitivity better than 0.1 nT, enabling it to capture minute changes in the magnetic field.

[0120] Then, a continuous tensor field is constructed based on the discrete magnetic field data, and an interpolation method that preserves the differential structure is employed. In this invention, an interpolation method based on radial basis functions (RBF) is preferably used, as shown in the following formula:

[0121] ,

[0122] in: For point The magnetic field vector at that location is a three-dimensional vector, with the unit being Tesla (T). For any point in space, For the first Each sampling point is a three-dimensional coordinate vector, with the unit being millimeters (mm). For the corresponding to the first The weight coefficients of each sampling point are a three-dimensional vector; These are radial basis functions; Represents the Euclidean distance metric; This represents the total number of sampling points; Indicates all The summation is performed on each sampling point.

[0123] Radial basis functions can be chosen as Gaussian functions. ,in For distance, This is a shape parameter, typically ranging from 0.1 to 1.0, and is adaptively adjusted based on the sampling point density. Larger values... A smaller value corresponds to a narrower range of influence and is suitable for situations where sampling points are dense; The value corresponds to a wider range of influence and is suitable for situations where sampling points are sparse.

[0124] Next, the covariant derivative of the tensor field on the differential manifold is calculated to form the gradient tensor. The covariant derivative is expressed as:

[0125] ,

[0126] in: Represents the magnetic field vector The Components with respect to coordinates The covariant derivative; Represents the magnetic field vector The Components with respect to coordinates The partial derivatives; The Christoffel symbol defined earlier; magnetic field vector The Components. The formula uses Einstein's summation convention for recurring indices. Perform a summation, covering all coordinate dimensions.

[0127] gradient tensor The components can be represented as:

[0128] ,

[0129] in: For gradient tensor The weight is a The matrix is ​​expressed in Tesla per meter (T / m). This is the covariant derivative defined earlier.

[0130] The gradient tensor is a second-order tensor that contains complete information about the spatial variation of the magnetic field.

[0131] Finally, perform eigenvalue decomposition on the gradient tensor to obtain the principal direction and eigenvalues:

[0132] ,

[0133] in: For gradient tensors, it is a Matrix; is an eigenvector matrix, whose column vectors are eigenvectors of the gradient tensor, representing the principal direction of magnetic field change; It is an eigenvalue diagonal matrix, where the diagonal elements are the eigenvalues ​​of the gradient tensor, representing the intensity of the change in the corresponding direction; This is the transpose of the eigenvector matrix. The eigenvectors represent the principal directions of magnetic field changes, and the eigenvalues ​​represent the intensity of the changes in the corresponding directions.

[0134] Multi-scale analysis is performed on the calculated gradient tensor to construct a gradient tensor image, and the defect locations of the test part are identified from the gradient tensor image. Multi-scale analysis is another innovative aspect of this invention, specifically implemented as follows:

[0135] First, a multi-scale representation is generated by applying a scaling transformation group to the original tensor field. The scaling transformation can be expressed as:

[0136] ,

[0137] in: The scale parameter is The tensor field below is a tensor field that is related to the original gradient tensor. Tensor fields of the same dimension; This represents the convolution operation; The Gaussian kernel function; This is a scale parameter, measured in millimeters (mm), which controls the smoothness.

[0138] The Gaussian kernel function is expressed as:

[0139] ,

[0140] in: For point Gaussian kernel function value at; It is a spatial coordinate vector; The standard deviation controls the width of the Gaussian function; For vectors The Euclidean norm; Pi; is the base of the natural logarithm.

[0141] In practical applications, scale parameters Typically, a series of discrete values ​​are taken, such as mm, covering defect features at different scales. Smaller Larger values ​​correspond to finer features and are suitable for detecting minute defects; The value corresponds to a more macroscopic feature and is suitable for detecting defects over a wide range.

[0142] Then, based on the feature response intensity, the optimal feature scale is selected for each region. The feature response intensity can be defined as:

[0143] ,

[0144] in: For point In scale The characteristic response intensity is below; The square of the scale parameter is used to normalize the response at different scales; This represents the Laplace operator, applied separately to each component of the tensor field; For point The scale is The gradient tensor; The Frobenius norm is defined as the square root of the sum of the squares of all elements of a matrix, i.e. .

[0145] For each position , choose to Maximum scale This serves as the characteristic scale for that location. Based on scale space theory, this method can automatically find the scale that best reflects the local features.

[0146] Next, a representation method capable of adaptively adjusting the scale parameters is constructed. This can be achieved by defining an adaptive scale field:

[0147] ,

[0148] in: For point The optimal scale at a given location is a scalar field; Indicates that the function Scale parameter for obtaining the maximum value .

[0149] Adaptive Scale Field An optimal scale was specified for each location, enabling the detection system to capture defects of different sizes simultaneously.

[0150] Finally, features from different scales are integrated to form a multi-resolution feature description. Weighted fusion can be used as a fusion method.

[0151] ,

[0152] in: For point Multi-resolution gradient tensor representation at the location; For scale At point Weight at each location; For point The scale is The gradient tensor; This indicates that for all scale parameters Perform summation.

[0153] Weight With characteristic response intensity Proportional, and after normalization ensures This can be represented as:

[0154] ,

[0155] in: For scale At point Weight at each location; The characteristic response intensity defined above; This indicates that for all scale parameters Summation is performed to normalize the weights.

[0156] After performing multi-scale analysis, it is necessary to construct a gradient tensor image. The specific steps are as follows:

[0157] First, the continuous tensor field is discretized into tensor values ​​at grid points. Preferably, the grid resolution is 0.1–0.5 mm to ensure that features of minute defects can be captured.

[0158] Then, a smoothing filter is applied to the tensor values ​​to reduce the impact of noise. Gaussian filtering can be used, with a kernel size typically of 3×3 or 5×5 and a standard deviation of 0.5-1.0 pixels.

[0159] Next, valid tensor data points are selected based on a preset gradient range. In this invention, the preset gradient range can be set according to material properties and welding process. For example, for Q345 steel, the gradient range can be set to 0.5 to 5.0 T / m.

[0160] Subsequently, the tensor data points are sorted to form a set of gradient tensors. The sorting criterion can be based on the tensor's norm, such as the Frobenius norm or the nuclear norm.

[0161] Then, a gradient tensor set is formed according to preset conditions. The preset conditions may include tensor similarity threshold (usually set to 0.8-0.9) and spatial proximity constraint (usually set to 1-3 mm).

[0162] Tensor images are formed based on gradient tensor sets. Tensor images can be visualized by mapping the main features of the tensor (such as eigenvalues, anisotropy, etc.) to color or grayscale values.

[0163] Obtain the attribute matrix of the tensor image, where the elements are the feature parameters of the gradient tensors in the gradient tensor set. Commonly used feature parameters include the trace, determinant, principal eigenvalues, and their ratios of the tensors.

[0164] Finally, the feature parameters in the attribute matrix are binarized. The binarization threshold can be automatically determined using the Otsu method, or empirically set to the 70-80 percentile of the feature parameter distribution.

[0165] Through the above steps, the defective parts of the test piece can be identified from the gradient tensor image. In embodiments of the present invention, the defective parts are typically characterized by abrupt changes in tensor feature parameters or regions with abnormal feature value ratios.

[0166] Establish a data table of defect types and severity, and evaluate the identified defect locations based on this data table to generate a defect evaluation result. This process specifically includes the following steps:

[0167] First, multiple defect records are generated, each labeled with a defect type. Each defect record includes the degree of defect corresponding to a specific type of defect. In embodiments of the present invention, common defect types include cracks, porosity, inclusions, lack of fusion, and undercut; the degree of defect can be classified into minor (Level I), moderate (Level II), and severe (Level III) according to relevant standards (such as GB / T3323).

[0168] Then, the defect severity for each defect record is normalized. The normalization formula is:

[0169] ,

[0170] in: The normalized defect level has a value range of [0,1]; D is the original defect level rating, which is an integer value. and These are the minimum and maximum rating values, respectively. For a three-level evaluation system, it can be set as follows: (Level I), (Level III)

[0171] Next, a correlation coefficient is set, which is formed by the weighted sum of the product of the defect type and the corresponding defect severity, and the coefficient itself. The correlation coefficient can be expressed as:

[0172] ,

[0173] Where: C is the correlation coefficient, which is a dimensionless scalar; The weighting coefficient for the i-th defect type reflects its importance; This is an index for the i-th defect type, usually 0 or 1, indicating whether this type of defect exists; The degree of the i-th defect can be either the original rating or a normalized value; n is the total number of defect types considered. This represents summing over all n defect types.

[0174] The weighting coefficients reflect the importance of different defect types. For example, cracks can be set to a weight of 0.4, lack of fusion to 0.3, porosity to 0.2, and other defects to 0.1. The sum of the weighting coefficients should be 1, i.e., .

[0175] The defect records are sorted according to the correlation coefficient. The higher the correlation coefficient, the greater the overall harm of the defect.

[0176] Based on preset conditions, gradient tensors corresponding to the degree of defect are selected to form a set of gradient tensors for each defect record. Preset conditions may include a correlation threshold between tensor features and defect type (usually set to 0.7-0.85) and spatial continuity constraints.

[0177] Based on the number of gradient tensors within the set, a baseline gradient tensor is defined to form a defect severity characteristic model for that defect type. The baseline gradient tensor is typically chosen as the median tensor or centroid tensor within the set.

[0178] Based on the established defect evaluation model, defect evaluation is performed on the identified defective parts, specifically including:

[0179] The defect records in the gradient tensor image are matched with the data table to obtain the corresponding defect type and degree. The matching process uses classification methods such as nearest neighbor algorithm or support vector machine.

[0180] Based on the curvature flow characteristics of the defect region, singularities and critical points in the flow field are analyzed. Curvature flow can be expressed as:

[0181] ,

[0182] in: Represents the gradient tensor field With virtual time parameters The rate of change; The Laplace operator, applied to each component of the tensor field, is defined as follows: This equation describes the diffusion process of the tensor field in the virtual time dimension.

[0183] The evolution of the tensor field can be obtained by numerically solving this equation. Commonly used numerical methods include explicit Euler method and implicit Euler method, with time step usually set to 0.01-0.1 and number of iterations ranging from 50 to 200.

[0184] A critical graph is constructed to represent the topology of the flow field, dividing the flow field into different homotopy categories, corresponding to different types of defects. The critical graph consists of singular points and integral lines connecting them; different topologies correspond to different types of defect patterns.

[0185] Calculate the persistence of topological features to distinguish between real defects and noise. Persistence can be defined as the scale range within which a feature remains present in multi-scale analysis:

[0186] ,

[0187] in: For its continuity, it is a dimensionless scalar; and These are the scale parameters for the appearance and disappearance of features, respectively, in millimeters (mm). This represents the logarithmic function with base 2.

[0188] Features with high persistence typically correspond to real defects, while features with low persistence may be noise. Experience shows that features with persistence greater than 2.0 are generally reliable defect indicators.

[0189] Construct an evaluation model for the degree of defect and define comprehensive indicators. ,when Areas exceeding a preset threshold are identified as defective. (Comprehensive Indicators) It can be represented as:

[0190] ,

[0191] in: As a comprehensive indicator, it is a dimensionless scalar. The ratio of eigenvalue to gradient reflects the relative intensity of local magnetic field changes; This is an index of the degree of change of each component in the gradient tensor, reflecting... Anisotropy of change.

[0192] Eigenvalue to gradient ratio It can be represented as:

[0193] ,

[0194] in: The absolute value of the largest eigenvalue of the gradient tensor; Let be the magnitude of the magnetic field gradient, calculated as .

[0195] Indicators of the degree of change of each component in the gradient tensor It can be represented as:

[0196] ,

[0197] in: and These are the absolute values ​​of the maximum and minimum eigenvalues ​​of the gradient tensor, respectively. The value ranges from [0,1]. The closer the value is to 1, the stronger the anisotropy, corresponding to a possible defect area; the closer the value is to 0, the more isotropic, corresponding to a normal area.

[0198] when When, it is determined to be a defective area; when When the threshold of 0.4 is reached, it is considered a normal area. The threshold was determined through analysis of a large amount of experimental data. This threshold can be adjusted appropriately for different materials or welding processes.

[0199] The evaluation model is used to evaluate the defective parts and generate defect evaluation results, including information such as defect type, location, size and severity.

[0200] The defect evaluation results are processed and displayed to determine the degree of defect in the weld under preset conditions. The specific steps are as follows:

[0201] First, cluster analysis is performed on the defect evaluation results to group defect points that are spatially close and have similar characteristics into one class. The clustering algorithm can be density clustering (DBSCAN) or hierarchical clustering. Clustering parameters such as the distance threshold are usually set to 3-5 mm, and the minimum number of points is set to 5-10.

[0202] Then, the clustering results are normalized to make the defect evaluation results of different categories comparable. Normalization can be performed using the min-max method or the Z-score method.

[0203] Based on the processed clustering results, data compression is performed to reduce the amount of data and improve processing efficiency. Data compression can be achieved through principal component analysis (PCA), typically retaining principal components that explain 85%–95% of the variance.

[0204] The clustering results after data compression are visualized, such as heatmaps, contour maps, or 3D surface maps. Visualization parameters, such as color mapping, can use red-yellow-green to represent decreasing defect severity.

[0205] Based on the clustering results, a defect evaluation region map is generated, indicating defect areas of different types and degrees. The region map can use shapes such as polygons or ellipses to represent defect boundaries.

[0206] Finally, the gradient tensor image and the defect evaluation results are fused and overlaid on the defect evaluation area map to form a comprehensive defect evaluation display. Fusion can be achieved using transparency blending or layer overlay, with the transparency parameter typically set to 0.3-0.7.

[0207] Through the above steps, the entire process of magnetic memory detection of welding defects in steel structure space frames was completed, forming a complete detection method from acquiring component information to displaying the final defect evaluation results.

[0208] Please refer to Figure 2 The present invention also provides a magnetic memory detection system for welding defects in steel structure space frames, including a magnetic sensor, a data acquisition system, a three-dimensional robotic arm and a host computer.

[0209] A magnetic sensor is mounted on a three-dimensional robotic arm to collect magnetic field data in the weld area. The sensor consists of two high-sensitivity magnetic cores, two magnetic reluctance compensation units, and four leads, all encased in shielding material. In a preferred embodiment, the magnetic cores are made of a high-permeability (μr>20000) nickel-iron alloy with a cross-sectional area of ​​0.5mm × 0.5mm and a length of 10mm. The magnetic reluctance compensation units employ a precision resistor network to compensate for the effects of temperature and external magnetic fields. The shielding material is a multi-layered high-permeability alloy to reduce external magnetic field interference. This sensor has a sensitivity better than 0.1nT and a frequency response range of DC-10kHz, making it suitable for capturing static and low-frequency magnetic field changes.

[0210] The data acquisition system comprises an FPGA data acquisition system and a computer data acquisition system. The FPGA system is responsible for data acquisition, transmission, and control, with a sampling rate of up to 500kHz, a resolution of 24 bits, and a dynamic range exceeding 120dB. The computer data acquisition system is responsible for recording and processing the pose information of the 3D robotic arm and weld seams. Equipped with a high-performance processor and large-capacity storage, it can process large-scale data in real time. The two systems are connected via a high-speed data bus to ensure the real-time performance and reliability of data transmission.

[0211] The 3D robotic arm communicates wirelessly with a host computer, which controls the acquisition position of the magnetic sensor and the storage of the acquired data. The 3D robotic arm consists of a joint section, an arm section, and a base. The joint section is a four-degree-of-freedom joint, including three rotational degrees of freedom and one translational degree of freedom, with an angular resolution better than 0.01° and a displacement resolution better than 0.01mm. The arm section includes a robotic arm part and a robotic gripper part. The robotic arm is made of lightweight, high-strength alloy material and can reach a length of up to 800mm. The robotic gripper is specifically designed for fixing and controlling the magnetic sensor. The base features a stable design and integrates a drive motor and control circuitry. The 3D robotic arm communicates with the host computer via 2.4GHz industrial wireless communication, with a communication distance of up to 100m and a data transmission rate of 10Mbps.

[0212] The host computer is equipped with a feature acquisition module, a defect identification module, a defect prediction and analysis module, a calculation module, a signal processing module, and a signal display module. The functions of each module are as follows:

[0213] The feature acquisition module is used to acquire structural information, material information, and welding parameters of the welded steel structure space frame under test. This module provides multiple data input interfaces, including CAD file import, manual input of material parameters, and automatic acquisition of welding parameters.

[0214] The defect identification module is used to construct a first finite element model based on structural information, material information, and welding parameters. Finite element excitation is then applied to this model to predict the types of defects that may occur in the weld. This module integrates a finite element analysis engine and supports various material models and welding thermodynamic models.

[0215] The defect prediction and analysis module is used to construct a second finite element model during the welding process, representing the weld region as a differential manifold and calculating the gradient tensor of the magnetic field distribution. This module implements the core algorithm of this invention, including functions such as differential manifold construction, tensor field characterization, and gradient calculation.

[0216] The computation module performs multi-scale analysis on gradient tensors, constructs gradient tensor images, and identifies defect locations in the test piece from these images. This module employs efficient numerical computation methods, supports GPU acceleration, and is capable of processing large-scale tensor data.

[0217] The signal processing module is used to create a data table of defect types and severity, and to evaluate the identified defect locations based on this data. This module integrates various machine learning algorithms, enabling it to continuously optimize the defect evaluation model based on historical data.

[0218] The signal display module processes and displays the defect evaluation results, determining the degree of defect in the weld under preset conditions. This module offers multiple visualization methods, supporting 2D / 3D display, cross-sectional analysis, and dynamic tracking, allowing users to intuitively understand the inspection results.

[0219] In practical applications, these modules work together to automate the entire process from data acquisition, model building, defect identification to result display, significantly improving detection efficiency and accuracy. The system is easy to operate; complex defect detection tasks can be completed simply by setting basic detection parameters.

[0220] Preferably, the magnetic sensors are arranged in an array, typically a 3×3 or 5×5 matrix array, which can simultaneously acquire magnetic field data from multiple points, improving scanning efficiency. The sensor spacing can be adjusted according to the required detection accuracy, typically between 2 and 10 mm.

[0221] The data acquisition system also includes a signal conditioning circuit to amplify, filter, and digitize the sensor output signal. The amplification circuit uses a low-noise instrumentation amplifier with an adjustable gain range of 20-1000 times; the filtering circuit includes a low-pass filter (cutoff frequency 1-5kHz) and a notch filter (50 / 60Hz) to effectively suppress ambient noise; the A / D conversion uses a successive approximation converter with a sampling rate of up to 500kHz and a resolution of 24 bits.

[0222] The control system of the 3D robotic arm adopts a hierarchical architecture, including low-level motion control and high-level trajectory planning. The motion control uses a PID algorithm with a response time of less than 10ms and a positioning accuracy better than 0.05mm. The trajectory planning supports multiple paths such as straight lines, circular arcs, and spline curves, with a maximum moving speed of 500mm / s and a maximum acceleration of 2000mm / s².

[0223] The host computer software adopts a modular design, developed based on C++ and Python, and has good scalability and maintainability. The software provides a graphical user interface, supporting functions such as setting detection parameters, real-time monitoring, and result analysis. Data storage uses a combination of relational databases and file systems, supporting long-term storage and rapid retrieval of detection data.

[0224] In summary, the magnetic memory detection method and system for welding defects in steel structure space frames provided by this invention, by combining differential geometry theory with magnetic memory detection technology, achieves high-precision identification, prediction, and quantitative evaluation of welding defects. This method not only improves detection accuracy but also possesses predictive and preventative capabilities as well as multi-scale analysis capabilities, providing a powerful tool for the quality control of steel structure space frame welding and showing broad application prospects.

[0225] The embodiments described above are merely illustrative of specific implementations of the present invention, and while the descriptions are detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A magnetic memory detection method for welding defects in steel space frames, characterized in that, include: Obtain structural information, material information, and welding parameters of the welded steel structure space frame component to be tested; Based on the structural information, the material information, and the welding parameters, a first finite element model is constructed, and finite element excitation is applied to the first finite element model. By analyzing the stress distribution law in the weld zone, the type of defect that appears in the weld is predicted. During the welding process, weld feature scanning is performed to obtain welding position signals and weld feature information, and a second finite element model is constructed. In the second finite element model, based on the weld feature information, a differential manifold characterizing the weld region is constructed, the magnetic field distribution is characterized as a tensor field on the differential manifold, and the gradient tensor of the tensor field is calculated. The construction of the differential manifold characterizing the weld region includes: converting the weld region into a parameter space through a coordinate mapping function to form a parameterized representation of the region; establishing a tangent space at each point in the weld region, defining a local coordinate basis, and forming a metric tensor; establishing a metric-compatible connection to ensure that the parallel transmission of the tensor field on the differential manifold maintains an invariant inner product; and calculating the basic geometric quantities of the differential manifold, including a first basic form and a second basic form. Multi-scale analysis is performed on the gradient tensor to construct a gradient tensor image, and the defect locations of the test part are identified from the gradient tensor image. The multi-scale analysis of the gradient tensor includes: applying a scale transformation group to the original tensor field to generate a multi-scale representation; selecting the optimal feature scale for each region based on the feature response intensity; constructing a representation method that can adaptively adjust the scale parameters; and integrating features at different scales to form a multi-resolution feature description. Establish a data table of defect types and defect severity, evaluate the identified defect locations based on the data table, and generate defect evaluation results; The defect evaluation results are processed and displayed to determine the degree of defect in the weld under preset conditions.

2. The magnetic memory detection method for welding defects in steel structure space frames according to claim 1, characterized in that, After constructing the first finite element model, the following is also included: Based on the prediction results of the first finite element model, it is determined whether the welding operation is normal. If there are weld defects, the welding parameters are adjusted according to the defect type. The adjusted welding parameters are bound to the structural information, the material information, and the defect type, and the prefabricated files stored in the database are updated.

3. The magnetic memory detection method for welding defects in steel structure space frames according to claim 1, characterized in that, Characterizing the magnetic field distribution as a tensor field on the differential manifold includes: Discrete magnetic field data are obtained by sampling the magnetic field on the differential manifold using a magnetic sensor. A continuous tensor field is constructed based on the discrete magnetic field data, and an interpolation method that preserves the differential structure is adopted. Calculate the covariant derivative of the tensor field on the differential manifold to form the gradient tensor; The gradient tensor is subjected to eigenvalue decomposition to obtain the principal direction and eigenvalues.

4. The magnetic memory detection method for welding defects in steel structure space frames according to claim 1, characterized in that, The construction of the gradient tensor image includes: Discretize the continuous tensor field into tensor values ​​at grid points; Apply a smoothing filter to the tensor values ​​to reduce the impact of noise; Valid tensor data points are filtered according to a preset gradient range; The tensor data points are sorted to form a gradient tensor set; Based on preset conditions, a gradient tensor set is formed; A tensor image is formed based on the gradient tensor set; Obtain the attribute matrix of the tensor image, where the elements of the attribute matrix are the feature parameters of the gradient tensors in the gradient tensor group; The feature parameters in the attribute matrix are binarized.

5. The magnetic memory detection method for welding defects in steel structure space frames according to claim 1, characterized in that, The establishment of the data table for defect types and defect severity includes: Multiple defect records are generated by labeling each defect type, with each record including the degree of defect corresponding to a particular type. The defect severity corresponding to each defect record is normalized. A correlation coefficient is set, which is formed by the weighted sum of the product of the defect type and the corresponding defect severity and the coefficient. The defect records are sorted according to the correlation coefficient; Based on preset conditions, the gradient tensors corresponding to the degree of defects are selected to form a set of gradient tensors for each defect record. Based on the number of gradient tensors in the set, a baseline gradient tensor is set to form a defect severity feature model for the defect type.

6. The magnetic memory detection method for welding defects in steel structure space frames according to claim 1, characterized in that, The defect evaluation of the identified defective parts includes: The defect records in the gradient tensor image are matched with the data table to obtain the corresponding defect type and defect severity. Based on the curvature flow characteristics of the defect region, singularities and critical points in the flow field are analyzed. A critical graph is constructed to represent the topology of the flow field, dividing the flow field into different homotopy categories, corresponding to different types of defects; Calculate the persistence of topological features to distinguish between real defects and noise; Construct an evaluation model for the degree of defects, define a comprehensive index T, and determine a defective area when T is greater than a preset threshold; The defective part is evaluated according to the evaluation model, and the defect evaluation result is generated.

7. The magnetic memory detection method for welding defects in steel structure space frames according to claim 1, characterized in that, The process of processing and displaying the defect evaluation results includes: Cluster analysis was performed on the defect evaluation results; The clustering results are then normalized. Based on the processed clustering results, data compression is performed; The clustering results after data compression are visualized. Based on the clustering results, a defect evaluation region map is generated; The gradient tensor image and the defect evaluation result are fused and superimposed on the defect evaluation region map.

8. A magnetic memory detection system for welding defects in steel space frames, used to implement the magnetic memory detection method for welding defects in steel space frames as described in any one of claims 1-7, characterized in that, include: Magnetic sensor, data acquisition system, 3D robotic arm and host computer; The magnetic sensor is fixed on the three-dimensional robotic arm and is used to collect magnetic field data in the weld area; The data acquisition system includes an FPGA data acquisition system and a computer data acquisition system. The FPGA data acquisition system is responsible for data acquisition, transmission and control, while the computer data acquisition system is responsible for recording and processing the pose information of the three-dimensional robotic arm and the weld seam. The three-dimensional robotic arm communicates wirelessly with the host computer, and the host computer controls the acquisition position of the magnetic sensor and the storage of the acquired data. The host computer is equipped with a feature acquisition module, a defect identification module, a defect prediction and analysis module, a calculation module, a signal processing module, and a signal display module; The feature acquisition module is used to acquire the structural information, material information, and welding parameters of the steel structure space frame welded test piece; The defect identification module is used to construct a first finite element model based on the structural information, the material information and the welding parameters, and to apply finite element excitation to the first finite element model to predict the type of defect that will occur in the weld. The defect prediction and analysis module is used to construct a second finite element model during the welding process, characterize the weld area as a differential manifold, and calculate the gradient tensor of the magnetic field distribution. The calculation module is used to perform multi-scale analysis on the gradient tensor, construct a gradient tensor image, and identify the defect location of the test piece from the gradient tensor image; The signal processing module is used to establish a data table of defect types and defect severity, and to evaluate the identified defect locations based on the data table. The signal display module is used to process and display the defect evaluation results, and to determine the degree of defect of the weld under preset conditions.

Images