A component strength judgment system based on big data

Abstract

The present invention relates to the technical field of component strength detection, and discloses a component strength judgment system based on big data, which includes a data acquisition module, a data preprocessing module, a single-point strength analysis module, and a continuous strength evaluation module. The data acquisition module acquires multi-dimensional parameters of the component and performs standardization processing, separating the main features from the abnormal fluctuation area; the preprocessing module identifies the boundary and calibrates the parameter data space coordinate system; the single-point strength analysis module generates a topological network through key nodes, and judges the strength of the detection unit by combining the density and span of the structural features; the continuous strength evaluation module generates a continuous fluctuation curve based on the parameter sequence, and evaluates the strength continuity through the curvature change rate and the recurrent neural network model. The system realizes intelligent analysis of multi-dimensional data, improves the comprehensiveness and accuracy of component strength detection, is suitable for complex structure strength evaluation, and provides technical support for engineering safety.

Description

Technical Field

[0001] The present invention relates to the technical field of component strength detection, and in particular to a component strength judgment system based on big data. Background Art

[0002] In modern engineering, component strength is a core indicator for ensuring the safety and reliability of various structures, including buildings, machinery, and aerospace. With the rapid development of industrial technology, component design is becoming increasingly complex, and application scenarios are constantly expanding. Traditional component strength testing methods are gradually exposing their limitations.

[0003] Traditional detection methods are usually based on single-point detection or local sampling, which makes it difficult to fully capture the multi-dimensional parameter characteristics of components. For example, measuring local stress only through strain gauges cannot reflect the overall performance changes under the coupling of multiple physical fields such as temperature field and vibration frequency. At the same time, traditional methods rely on manually set thresholds and empirical judgments, and lack the ability to intelligently analyze complex structural features. Especially when faced with complex forms such as nonlinear distributions and mesh structures, misjudgments or omissions are prone to occur. In addition, for the continuity assessment of component strength, traditional technologies have difficulty in realizing data linkage analysis of multiple detection units, and cannot effectively identify subtle anomalies in continuous fluctuation curves, resulting in the inability to timely discover structural hazards.

[0004] With the rise of big data and intelligent algorithms, the demand for intelligent and precise component strength testing in the engineering field is becoming increasingly urgent. How to integrate multi-dimensional parameter data and build an intelligent analysis model covering the entire life cycle of components has become a technical problem that needs to be solved urgently. In the existing technology, although there are some detection systems based on data processing, there are still technical bottlenecks in key links such as data standardization processing, spatial mapping conversion, topological network construction, and continuous strength assessment. For example, the lack of effective separation of the main characteristic area of ​​the component and the abnormal fluctuation area in the data preprocessing process has limited the accuracy of subsequent analysis; the topological network generation algorithm fails to fully consider the spatial correlation and path optimization of the reference points, affecting the accuracy of strength judgment; the lack of dynamic modeling of the curvature change rate in the continuous strength assessment makes it difficult to achieve accurate prediction of the probability of structural failure.

[0005] Therefore, developing a component strength assessment system that integrates big data technologies to achieve standardized processing of multidimensional parameters, intelligent identification of complex structural features, and continuous dynamic strength assessment is of great practical significance for improving the safety and reliability of engineering structures. This system must address the shortcomings of traditional methods in terms of data comprehensiveness, intelligent analysis, and continuous assessment, providing a more efficient and accurate technical solution for component strength testing. Summary of the Invention

[0006] The purpose of the present invention is to provide a component strength judgment system based on big data to solve the problems raised in the above background technology.

[0007] To achieve the above-mentioned object, the present invention provides the following technical solution: a component strength judgment system based on big data, the system comprising:

[0008] The data acquisition module is used to obtain multi-dimensional parameter data of the component to be tested, and to perform standardization on the parameter data to divide the component main feature area and abnormal fluctuation area;

[0009] A data preprocessing module is used to identify the boundaries of the main feature area of ​​the component and perform spatial mapping conversion on the parameter data according to the physical form of the component;

[0010] The single-point strength analysis module is used to select any detection unit area of ​​the component, collect the starting and ending values ​​of each stress node based on the distributed sensor, mark them as reference points, obtain several reference points closest to the geometric center point of the detection unit, define them as key nodes, and sequentially associate the adjacent key nodes to generate a topological network. The density of the topological network covering the preset standard strength interval is obtained. When the density is less than a preset threshold, it is identified whether it is a linear distribution or a mesh distribution structural feature. If not, it is determined that the bearing pressure of the detection unit is concentrated; if so, the span of the longitudinal or transverse distribution of the structural feature is detected. If the span is greater than the preset span value, it is determined that the structural strength of the unit is abnormal.

[0011] The continuous strength assessment module is used to select any one-dimensional parameter data sequence, expand the data area of ​​each detection unit longitudinally based on the axis of the unit center point, until the data areas in adjacent detection units form a continuous fluctuation curve, generate a continuous data energy distribution band, and calculate the curvature change rate of the distribution band fitting curve. When the deviation of the change rate from the standard value exceeds a preset range, it is judged that the strength continuity of the component in this dimension does not meet the standard.

[0012] Preferably, the process of the standardization process is:

[0013] The multidimensional parameter data is processed to eliminate outliers to generate cleaned data, the cleaned data is subjected to dynamic range compression, and the cleaned data is converted into normalized data by setting a data threshold. The normalized data is smoothed based on sliding window filtering, and trend decomposition is performed on the smoothed normalized data to separate the component main feature area from the abnormal fluctuation area.

[0014] Preferably, the preset standard intensity interval is composed of 8 reference anchor points, of which 3 anchor points are set in the upstream area of ​​the detection unit, and the other 5 anchor points are set in the downstream area of ​​the detection unit. The three points in the upstream area are positioned at the energy peak point of the axial distribution; the two anchor points in the downstream area are positioned at the intersection of the unit boundary line and the stress conduction path, that is, the energy release point, and the remaining three points in the downstream area are positioned at the three uniformly distributed nodes of the energy release point connection line.

[0015] Preferably, the process of performing spatial mapping conversion on parameter data is:

[0016] The boundary contour line of the main characteristic area of ​​the component is obtained, and the azimuth deviation angles between the four main axes of the boundary and the parameter data coordinate system are calculated respectively. The parameter data is calibrated in the spatial coordinate system according to the weighted average of the azimuth deviation angles.

[0017] Preferably, the process of detecting the longitudinal or transverse distribution span of the structural feature is:

[0018] The data area is traversed according to the detection direction, which is a longitudinal section or a transverse section. The energy intensity value of each group of data detected is counted, the energy value of the abnormal fluctuation area is 255, and the energy value of the main feature area is 0. The first detection group whose energy value exceeds the main feature threshold is selected and marked as the abnormal starting group. The data area is continued to be traversed. When a detection group whose energy value returns to the main feature threshold is detected, it is marked as a pending group, and the spatial interval L1 between the pending group and the starting group is obtained. If the spatial interval L1 is greater than or equal to the preset span N, and no energy abnormality is detected along the detection direction, the pending group is marked as the termination group, and the structure has a uniform distribution feature; if the spatial interval is less than l1, the data area is continued to be traversed. When a detection group whose energy exceeds the main feature threshold is detected again, it is marked as the second starting group, and the interval L2 between the second starting group and the pending group is obtained. If the interval L2 is greater than the preset span M, it is determined that the structure distribution is abnormal.

[0019] Preferably, the specific process of judging whether the component strength continuity does not meet the standard is as follows:

[0020] For any selected detection unit, select the reference axis where the center point of the unit is located, obtain the offset of all energy points in the data area relative to the axis, adjust the offset of each energy point by a preset amplification factor, and continuously increase the value of the amplification factor until adjacent detection units form a data superposition area, obtain a continuous fluctuation band of energy distribution, obtain the corresponding fitting curve equation by mean filtering the fluctuation band, and obtain the curvature change rate of the curve equation.

[0021] Preferably, the process of generating a topological network is:

[0022] First, the four reference points closest to the center point of the detection unit are selected, and the adjacent reference points are associated to generate an initial topological structure. Then, the nearest points among the remaining reference points are selected one by one and added to the association of the topological structure. When the length of the connection path corresponding to the original reference point in the association path formed by the newly added reference point increases, the newly added point is removed, and the association path formed by the original reference point is the topological network of the detection unit.

[0023] Preferably, the data acquisition module includes a temperature sensor array, a vibration frequency collector and a strain gauge group, wherein the temperature sensors are arranged on the surface of the component in a honeycomb structure, the vibration frequency collectors are axially distributed at equal intervals, and the strain gauge group covers all stress concentration areas of the component.

[0024] Preferably, the dynamic range compression adopts a nonlinear transformation method to compress the range of the original data to a preset range while retaining the data distribution characteristics, and the window length of the sliding window filter is dynamically adjusted according to the elastic modulus of the component material.

[0025] Preferably, the continuous strength assessment module has a built-in recurrent neural network model, which is trained using historical component strength data, with the curvature change rate of the energy distribution zone as the input feature and the structural failure probability as the output result. A dynamic weight adjustment strategy is used to optimize the loss function during model training.

[0026] Compared with the prior art, the present invention has the following beneficial effects:

[0027] The data acquisition module, through a combination of a temperature sensor array, a vibration frequency collector, and a strain gauge array, comprehensively collects multidimensional parameters such as component temperature, vibration frequency, stress, and strain. The temperature sensors are arranged in a honeycomb pattern on the component surface, ensuring high-density coverage of temperature field data. The vibration frequency collectors are evenly spaced axially to capture vibration characteristics along the component axis. The strain gauge array covers all stress concentration areas, accurately acquiring stress data at key locations. This integration of multidimensional parameters provides a rich data foundation for subsequent analysis, avoiding the one-sidedness of traditional single-point detection.

[0028] The data preprocessing module effectively improves data quality through standardized processing procedures. Outlier rejection removes noise interference, dynamic range compression compresses the original data range to a preset range while preserving data distribution characteristics, and sliding window filtering dynamically adjusts the window length based on the elastic modulus of the component material, achieving adaptive smoothing of data from different materials. Trend decomposition technology separates the main characteristic areas of the component from areas of abnormal fluctuations, laying the foundation for subsequent feature analysis and ensuring that subsequent modules focus on valid data areas, improving analysis efficiency and accuracy.

[0029] The single-point strength analysis module achieves refined judgment of the strength of the detection unit through topological network construction and structural feature identification. By selecting the key nodes closest to the geometric center point of the detection unit to generate a topological network and calculating its density covering the preset standard strength range, it can effectively identify pressure concentration phenomena. For linear or mesh-distributed structural features, by detecting the longitudinal or lateral distribution span, it can accurately determine whether the structural strength is abnormal. This module not only considers the spatial correlation of nodes, but also incorporates the distribution patterns of structural features, avoiding the mechanical nature of traditional threshold judgments and improving the analysis capabilities of complex structures.

[0030] The continuous strength assessment module generates continuous fluctuation curves and energy distribution bands by vertically expanding the data region. Combined with a recurrent neural network model, it analyzes the curvature change rate, enabling dynamic assessment of component strength continuity. The recurrent neural network model, trained with historical data, uses the curvature change rate as an input feature to predict the probability of structural failure. A dynamic weight adjustment strategy optimizes model training, enabling it to capture temporal characteristics and nonlinear relationships within the data sequence. This module transcends the limitations of traditional discrete testing, enabling data linkage analysis across multiple testing units. This module can promptly identify substandard strength continuity and provide strong support for early warning of structural hazards.

[0031] The nonlinear transformation dynamic range compression and adaptive sliding window filtering used in the standardization process ensure compatibility and comparability of data across components of different types and materials, enhancing the system's versatility. The topological network generation algorithm gradually optimizes the paths associated with reference points, avoiding interference from redundant nodes and improving the rationality of the network structure and analysis efficiency. The amplification factor adjustment and mean filtering used in the continuous strength assessment effectively enhance the recognizability of data fluctuations, ensuring that the fitted curve better reflects the true strength trend. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 This is a working principle diagram of the component strength judgment system based on big data according to the present invention;

[0033] Figure 2 Design diagram for anchor point configuration for standard strength intervals;

[0034] Figure 3 Logic diagram for structural feature span detection;

[0035] Figure 4 Step diagram generated for topological network;

[0036] Figure 5 This is the design diagram of the data acquisition module. DETAILED DESCRIPTION

[0037] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0038] See also Figure 1-Figure 5 The present invention relates to a component strength judgment system based on big data, which includes: a data acquisition module, a data preprocessing module, a single-point strength analysis module, and a continuous strength assessment module. The specific implementation steps are as follows:

[0039] The data acquisition module obtains the multi-dimensional parameter data of the component to be tested, and standardizes the parameter data to divide the component main feature area and abnormal fluctuation area; the data preprocessing module identifies the boundary of the component main feature area, and performs spatial mapping conversion on the parameter data according to the physical form of the component; the single point strength analysis module selects any detection unit area of ​​the component, collects the starting value and ending value of each stress node based on the distributed sensor, marks them as reference points, obtains several reference points closest to the geometric center point of the detection unit, defines them as key nodes, and associates the adjacent key nodes in turn to generate a topological network, and obtains the density of the topological network covering the preset standard strength range. When the density is less than the preset When setting a threshold, it identifies whether the structural feature is linearly distributed or mesh-distributed. If not, it is judged that the bearing pressure of the detection unit is concentrated; if so, the span of the longitudinal or transverse distribution of the structural feature is detected. If the span is greater than the preset span value, the structural strength of the unit is judged to be abnormal; the continuous strength assessment module selects any one-dimensional parameter data sequence, and longitudinally expands the data area of ​​each detection unit based on the axis where the unit center point is located, until the data area in the adjacent detection units forms a continuous fluctuation curve, generates a continuous data energy distribution band, and calculates the curvature change rate of the distribution band fitting curve. When the deviation of the change rate from the standard value exceeds the preset range, it is judged that the strength continuity of the component in this dimension does not meet the standard.

[0040] The technical solution of the present invention is further described in detail below with reference to specific embodiments.

[0041] Example 1:

[0042] The system's data acquisition module consists of a temperature sensor array, a vibration frequency collector, and a strain gauge assembly. The temperature sensors are arranged on the component surface in a honeycomb-like structure. This layout creates a regular honeycomb array across the component surface, evenly distributing each sensor unit and maintaining a specific spacing between adjacent sensors. This arrangement covers the majority of the component surface, ensuring temperature parameter acquisition from diverse locations and avoiding data blind spots caused by uneven sensor layout. The vibration frequency collectors are evenly spaced axially, with acquisition points set at fixed, equal intervals along the component's axial direction. This distribution allows for a continuous sampling sequence of vibration frequency data along the component's axial dimension, facilitating subsequent analysis of the vibration frequency's variation and distribution characteristics along the component's length. The strain gauge assembly covers all stress concentration areas of the component. Preliminary mechanical analysis of the component identifies key stress concentration locations, such as corners, interfaces, and cross-sectional variations. Strain gauges are then tightly affixed to the surface of these areas, ensuring real-time and accurate capture of strain data in these stress concentration areas, providing key mechanical parameters for component strength analysis.

[0043] After the data acquisition module acquires multidimensional parameter data, it needs to be standardized. The specific process is as follows: First, the multidimensional parameter data is processed to eliminate outliers. By setting reasonable outlier judgment rules (such as the Z-score method or the IQR method based on statistical methods), outliers in the data that significantly deviate from the overall distribution are identified and removed to generate cleaned data. This eliminates the interference of accidental errors or noise data on subsequent analysis and improves the purity and reliability of the data. Next, the cleaned data is subjected to dynamic range compression. Nonlinear transformation methods such as logarithmic transformation, exponential transformation, or customized nonlinear function mapping are used. The appropriate transformation method is selected based on the distribution characteristics of the data to compress the range of the original data to a preset interval (such as [0,1] or [-1,1]). At the same time, the relative distribution characteristics of the data are retained to avoid distortion of the data distribution form due to linear compression. The compressed data can not only meet the numerical range requirements of subsequent processing but also truly reflect the changing trend of the original data. The cleaned data is then converted into normalized data by setting a data threshold. Reasonable upper and lower thresholds are set based on the compressed data range, and the data is uniformly mapped to a specific normalized interval, achieving dimensional unification of parameter data of different dimensions, facilitating subsequent data analysis and model processing. The normalized data is then smoothed using a sliding window filter. The window length of the sliding window filter is dynamically adjusted based on the elastic modulus of the component material. Specifically, by establishing a mapping relationship between the material elastic modulus and the window length (e.g., materials with a larger elastic modulus correspond to a smaller window length to retain more high-frequency details; materials with a smaller elastic modulus correspond to a larger window length to enhance the smoothing of low-frequency trends), the window size is automatically adjusted based on the material properties of the actual detected component, and the data sequence is smoothed point by point, reducing random fluctuations in the data and making the data curve smoother and more stable. Finally, the smoothed normalized data is trend decomposed by using trend decomposition methods in time series analysis (such as moving average method, exponential smoothing method or empirical mode decomposition method) to decompose the data into the main characteristic part reflecting the long-term change trend and the detail part reflecting short-term fluctuations or anomalies, thereby separating the main characteristic area of ​​the component from the abnormal fluctuation area, providing a clear data partitioning basis for subsequent data preprocessing and strength analysis.

[0044] During the actual layout of the temperature sensor, the side length of each hexagonal grid of the honeycomb structure can be adjusted according to the surface size of the component and the detection accuracy requirements to ensure that the sensor density can meet the requirements of the fine characterization of the surface temperature field of the component. The equally spaced axial distribution spacing of the vibration frequency collector needs to be determined according to the length of the component and the requirements of the vibration characteristic analysis. It is usually set to 1 / 10 to 1 / 20 of the characteristic length of the component to ensure that the collected vibration frequency data has sufficient spatial resolution. When the strain gauge group covers the stress concentration area, it is necessary to ensure that the pasting direction of the strain gauge is consistent with the main stress direction to accurately measure the strain value in this direction. At the same time, a moisture-proof and corrosion-resistant pasting process is used to ensure the stability and reliability of the strain gauge during long-term detection.

[0045] In the outlier removal process, outlier detection is performed separately for each dimension in the multidimensional parameter data to avoid the data integrity of other dimensions being affected by the misjudgment of outliers in a single dimension. The nonlinear transformation function of dynamic range compression needs to be pre-verified to ensure that while compressing the data extremes, the data distribution will not be overly distorted or the peak shifted. The dynamic adjustment mechanism of the sliding window filter needs to be based on a large amount of historical analysis of the data characteristics of components of different materials, and through the pre-established material-window length database, the window length can be quickly matched and adjusted. In the trend decomposition process, it is necessary to select an appropriate decomposition method based on the characteristics of the data. For data with obvious linear trends, the moving average method can be used, and for data with nonlinear trends, the empirical mode decomposition method can be used to ensure the accuracy of separation of the main feature area and the abnormal fluctuation area.

[0046] The layout design and standardized processing flow of the aforementioned data acquisition module enable comprehensive and accurate collection and standardized processing of multi-dimensional parameter data for the component under test. This provides high-quality data input for the subsequent data preprocessing module, single-point strength analysis module, and continuous strength assessment module, ensuring the reliability and accuracy of the analysis results of the entire component strength judgment system. The various components of the data acquisition module work together, and each step of the standardized processing is executed sequentially, forming a complete data processing chain, laying a solid data foundation for comprehensive judgment of component strength.

[0047] Example 2:

[0048] When the system's data preprocessing module performs spatial mapping on the parameter data, it first processes the raw data of the component's characteristic regions using edge detection algorithms (such as the Canny or Sobel operators) to identify the boundary contours of the data distribution. These contours outline the approximate shape and extent of the component in the parameter data space, reflecting the component's geometric characteristics and physical boundaries. Next, four principal axes are extracted from the boundary contours: a horizontal principal axis, a vertical principal axis, and two diagonal principal axes. These four principal axes are orthogonal or at specific angles to each other, comprehensively characterizing the boundary's primary extension direction and spatial orientation. The system then calculates the azimuth deviation angle (ADO) of each principal axis relative to the parameter data coordinate system (typically a Cartesian coordinate system). This deviation angle is the angle between the principal axis and the corresponding coordinate axis in the coordinate system (e.g., the horizontal principal axis and the x-axis, the vertical principal axis and the y-axis). The deviation angle is determined using trigonometric functions or vector dot products. This angle reflects the difference between the actual and ideal orientation of the component in the parameter data coordinate system.

[0049] Based on the azimuth deviation angles of the four main axes, their weighted average is calculated. The weights can be determined based on the importance of the boundary features represented by the main axes (e.g., the vertical and horizontal main axes can be given higher weights because they are more in line with conventional geometric analysis practices, while the diagonal main axes are given lower weights). A comprehensive azimuth deviation angle is obtained through weighted averaging, which serves as the basis for calibrating the parameter data spatial coordinate system. When calibrating the spatial coordinate system, the coordinate transformation matrix is ​​used to rotate the original parameter data. The rotation angle is the azimuth deviation angle obtained by weighted average, so that the adjusted parameter data coordinate system is consistent with the actual orientation of the component body. For example, if the deviation angle between the horizontal main axis and the x-axis is calculated to be θ, the entire parameter data coordinate system is rotated counterclockwise by θ so that the adjusted horizontal main axis coincides with the x-axis, thereby achieving spatial orientation calibration of the parameter data.

[0050] In the single-point strength analysis module, the preset standard strength interval consists of eight benchmark anchor points. In the upstream region of the test cell, data analysis algorithms (such as peak detection) identify peak points in the axial energy distribution curve. The top three peak points with the highest energy values ​​are selected as the benchmark anchor points for the upstream region. These three anchor points represent key locations in the upstream energy distribution, reflecting the degree of energy concentration and locational characteristics in that region. In the downstream region, the intersection of the cell boundary and the stress conduction path is first determined. The stress conduction path can be predetermined through finite element analysis or physical modeling. The intersection locations are obtained by solving the boundary line equation and the stress conduction path equation simultaneously. These two intersection points serve as energy release points and are key locations for stress transmission from the test cell outward. Changes in their energy values ​​directly affect the strength assessment of the test cell. The remaining three anchor points in the downstream region are evenly distributed three times along the line connecting the two energy release points. This means that the line is divided into four equal segments, with the three middle points serving as anchor points. This uniform distribution ensures comprehensive coverage of the downstream energy distribution range, ensuring the integrity and representativeness of the preset standard strength interval.

[0051] The spatial distribution of the eight reference anchor points in the detection unit forms a specific geometric structure. The three anchor points in the upstream area are arranged axially, and two of the five anchor points in the downstream area are located at the boundary intersection and three are located at equally divided nodes, together forming a three-dimensional reference frame that includes key energy positions upstream and downstream. In practical applications, the peak point detection in the upstream area needs to be combined with the material properties and stress mode of the component. The energy peak of different materials may manifest differently (for example, the peak of elastic materials is sharper, and the peak of plastic materials is relatively flat), and an adaptive peak detection algorithm needs to be adopted. The determination of the stress conduction path in the downstream area needs to take into account the actual force direction and boundary conditions of the component, and improve the accuracy of intersection positioning through mechanical modeling or historical data accumulation. The calculation of three evenly distributed nodes needs to be based on precise geometric coordinate operations to ensure the accuracy of the anchor point position.

[0052] The spatial mapping transformation process in the data preprocessing module aligns parameter data with the physical form of the component, avoiding analysis errors caused by coordinate system deviations. For example, when the component body is tilted in real space, uncalibrated parameter data may exhibit a distorted distribution, causing subsequent strength analysis results to deviate from the actual situation. Through principal axis analysis of the boundary contours and coordinate system rotation, the parameter data truly reflects the spatial orientation and geometric characteristics of the component body, providing an accurate spatial reference for the selection of key nodes and the generation of the topological network in the single-point strength analysis module.

[0053] Eight benchmark anchor points within a pre-set standard strength range provide a quantitative reference for single-point strength analysis. When assessing the strength of a test cell, the density of the strength ranges formed by the topological network covering these anchor points allows for a rapid assessment of the uniformity of stress distribution within the test cell. The energy peak anchor point in the upstream region is used to assess energy input upstream of the test cell, while the energy release point and uniform distribution anchor points in the downstream region are used to determine the efficiency of energy conduction and diffusion, thereby comprehensively evaluating the strength performance of the test cell during stress transfer.

[0054] In practice, boundary contour extraction requires selecting an appropriate edge detection algorithm based on the type of parameter data (e.g., temperature field, vibration field, strain field). Data with high noise levels must undergo smoothing preprocessing. Principal axis calculation can be performed using principal component analysis (PCA) by performing dimensionality reduction on the coordinate data of the boundary contour line. The two principal component directions with the largest variance are extracted as the horizontal and vertical principal axes to ensure that the principal axes truly reflect the primary extension direction of the boundary. The weighted average calculation of the azimuth deviation angle requires the establishment of a reasonable weight distribution model. Weight coefficients can be determined through expert experience or historical data training to improve the accuracy of the comprehensive deviation angle.

[0055] When locating anchor points within a preset standard strength range, the energy peaks in the upstream region must be free of interference from abnormal fluctuations, and valid peaks must be screened by setting peak width and height thresholds. The calculation of intersection points for stress conduction paths in the downstream region must consider the continuity and uniqueness of the paths to avoid ambiguity in intersection location due to the presence of multiple conduction paths. The coordinates of the three uniformly distributed nodes must be calculated using high-precision numerical methods to ensure equal division of the anchor points along the connecting line and avoid coordinate errors that affect the accuracy of the strength range.

[0056] Through the spatial mapping conversion of the data preprocessing module and the preset standard strength interval construction of the single-point strength analysis module, the conversion from original parameter data to structured analysis data is realized, providing a spatial benchmark and quantitative basis for subsequent key node selection, topological network generation and strength judgment, so that the entire component strength judgment system can more accurately reflect the actual strength characteristics of the component, and improve the reliability and scientificity of the judgment results.

[0057] Example 3:

[0058] When the single-point strength analysis module generates a topological network, it first determines the initial key nodes within the detection unit based on spatial positional relationships. By calculating the Euclidean distance between each reference point and the center point of the detection unit, the four closest reference points are selected as the initial key nodes. These four nodes are located in the core area of ​​the detection unit and can effectively characterize the basic mechanical properties and stress distribution state of this area. Subsequently, the Delaunay triangulation algorithm is used to connect these four initial key nodes to form the initial topological structure. This algorithm ensures that the generated triangular mesh has good geometric quality, avoiding narrow or misshapen triangles, and more accurately reflects the relationships between nodes.

[0059] After the initial topology is formed, the network is gradually expanded. Nodes from the remaining reference points are selected and added to the topology, in descending order of distance from the detection unit center. For each node to be added, its spatial distance to all existing nodes is calculated, and the three closest nodes are selected for connection. When establishing new connections, it is necessary to ensure that the newly formed triangle meets the empty circle property, that is, the circumcircle of the triangle does not contain other nodes, to ensure the stability and rationality of the topology.

[0060] As nodes are added, changes in the topology are continuously monitored. If the addition of a new node causes the length of the connection path between existing nodes to increase by more than a preset threshold (e.g., 10% of the original path length), the new node is considered to have disrupted the rationality of the original topology and is removed. This ensures that the topology network maintains an accurate representation of the stress transfer paths of the detection units during expansion.

[0061] When all benchmarks are evaluated, the resulting topological network includes all key nodes and their relationships within the detection unit. The weight of each edge in the network is determined by the distance between nodes and the stress transfer strength. The specific calculation formula is:

[0062]

[0063] in, Representation node With node The weight of the edge between and Represents nodes respectively and nodes The stress transfer coefficient is determined by the material properties and geometry of the node location; Representation node With node The formula shows that the stress transfer intensity between nodes is proportional to the characteristics of the nodes themselves and inversely proportional to the square of the distance.

[0064] When detecting the vertical or horizontal span of structural features, the data region is scanned row by row or column by column, depending on the detection direction (vertical or horizontal). During the scanning process, the data is segmented using a fixed-size sliding window, which is set based on the required detection accuracy. For the data within each window, the mean and standard deviation of the energy intensity are calculated and used as the characteristic parameters of the region.

[0065] When the energy intensity mean detected in a window exceeds a preset threshold (e.g., 1.5 times the global mean) and the standard deviation exceeds a specific value (e.g., 0.8 times the global standard deviation), the window is marked as a suspected abnormal region. The window is moved continuously until the energy intensity of multiple consecutive windows returns to the normal range. The positions of the first and last abnormal windows are recorded, and the spatial separation L between them is calculated.

[0066] The calculated spatial interval L is compared with the preset span threshold N. If L is greater than or equal to N, the abnormal region is considered to have sufficient span and may represent a structural feature. If L is less than N, the abnormal region is considered to be a local fluctuation and has no structural significance. For abnormal regions confirmed as structural features, their distribution pattern (such as linear distribution, network distribution, etc.) is further analyzed to determine their impact on the strength of the detection unit.

[0067] In practical applications, node selection and connection during topological network generation must be tailored to the component's actual stress conditions. For components subject to complex loads, the number of initial key nodes may need to be increased to more comprehensively capture stress transfer paths. The implementation of the Delaunay triangulation algorithm must consider computational efficiency. Optimized data structures (such as the Bowyer-Watson algorithm) can be used to accelerate node insertion and triangulation updates.

[0068] The window size and threshold setting for structural feature span detection must be optimized based on the component type and detection objective. Larger window sizes can detect larger-scale structural features but may miss smaller-scale anomalies; smaller window sizes have the opposite effect. Threshold setting must strike a balance between sensitivity and specificity to avoid false positives and missed detections.

[0069] By constructing a topological network and detecting structural feature spans, we can comprehensively assess the stress distribution and structural integrity within the test unit. Density analysis of the topological network reveals areas of stress concentration, while detection of structural feature spans identifies potential structural defects. The combination of these two provides strong support for accurate assessment of component strength.

[0070] Example 4:

[0071] The Continuous Strength Assessment Module determines if a component's strength continuity fails to meet the standard. Starting with an arbitrarily selected inspection unit, the module first determines the reference axis of the unit's center point. This reference axis is typically the unit's geometric center axis in the parameter data space. For example, in two-dimensional data, this could be a horizontal or vertical axis passing through the unit's center point, or in three-dimensional data, a coordinate axis passing through the center point. Using this axis as a reference for data expansion and analysis ensures that data from adjacent inspection units are linked within a unified spatial reference.

[0072] Obtain the offset of all energy points within the data region relative to the reference axis. An energy point is a discrete data point in the parametric data that represents a component's physical properties (such as stress, strain, and temperature). The offset is the absolute difference between the coordinate value of each energy point and the corresponding coordinate value of the reference axis. For example, in a two-dimensional plane, if the reference axis is the horizontal x-axis, the offset of the energy point is the absolute value of its y-coordinate value; if it is the vertical y-axis, the offset is the absolute value of its x-coordinate value. By calculating the offset, the degree of dispersion of the energy points on either side of the reference axis can be quantified.

[0073] The offset of each energy point is adjusted by a preset amplification factor. The amplification factor is a proportional factor greater than 1, and its initial value can be set to a smaller multiple such as 1.1 or 1.2. The purpose is to gradually amplify the offset so that the data areas of adjacent detection units can gradually overlap. During the adjustment process, the value of the amplification factor is continuously increased, for example, increasing it in steps of 0.1. After each adjustment, it is checked whether the data areas of adjacent detection units are superimposed. When the amplification factor increases to a certain value, the data areas of adjacent detection units begin to overlap. At this time, the amplification is stopped to obtain a continuous fluctuation band of energy distribution. This fluctuation band is a continuous curve formed by connecting the data areas of multiple detection units after amplifying the offset, which can reflect the overall change trend of the parameter data in the selected dimension.

[0074] The fluctuation band is processed by mean filtering to obtain the fitting curve equation. Mean filtering is a linear smoothing filtering method that slides a window of fixed length on the fluctuation band and calculates the average offset of all energy points in the window as the filtered value of the center point of the window. For example, if the window length is 5, then for the i-th point on the fluctuation band, its filtered value is the average offset of the i-2, i-1, i, i+1, and i+2 points (assuming i is in the middle position). Mean filtering can effectively reduce the random noise in the data, making the curve of the fluctuation band smoother and facilitating subsequent curvature analysis. The filtered curve can generate corresponding mathematical equations through polynomial fitting, spline fitting and other methods, such as the quadratic polynomial equation Or cubic spline equation, etc., this equation is the fitting curve equation, which is used to describe the overall shape of the energy distribution fluctuation band.

[0075] Get the curvature change rate of the fitted curve equation. Curvature is a geometric quantity that describes the degree of curvature of a curve. For any point in the fitted curve equation, its curvature can be calculated by the first-order derivative and second-order derivative of that point. The curvature change rate is the ratio of the difference between the curvature values ​​of adjacent points to the length of the corresponding interval, which is used to reflect the speed of change of the curvature of the curve. For example, for a point on the curve and , whose curvatures are and , if the length of the interval between two points is , then the curvature change rate is When the deviation between the rate of change and the preset standard value exceeds the preset range, it indicates that there is an abnormality in the continuity of the energy distribution of the component in this dimension, and it is judged that the strength continuity of the component in this dimension does not meet the standard.

[0076] The core structure of the recurrent neural network (RNN) model built into the continuous strength assessment module consists of an input layer, a hidden layer, and an output layer. The input layer receives data on the rate of change of curvature of the energy distribution band. This data is normalized to a range of [0, 1] or [-1, 1] to improve model training efficiency and stability. The hidden layer uses RNN units (such as LSTM or GRU units) to capture temporal dependencies in sequential data and is suitable for analyzing the contextual relevance of data in continuous strength assessment. The output layer consists of a single neuron that outputs a structural failure probability value, which ranges from 0 to 1. A larger value indicates a higher probability of structural failure.

[0077] The model is trained using historical component strength data. The training data must include parameter data for components of different types, materials, and operating conditions, as well as corresponding structural failure annotations. During training, a dynamic weight adjustment strategy is used to optimize the loss function. This loss function, typically a cross-entropy loss function, measures the difference between the model's predicted probability of structural failure and the actual annotated value. This dynamic weight adjustment strategy automatically adjusts the weight parameters of each layer in the model using an adaptive algorithm (such as the Adam optimization algorithm). Based on changes in the loss value during training, the learning rate and momentum parameters are dynamically adjusted to accelerate model convergence and avoid falling into local optimal solutions.

[0078] In practical applications, the choice of reference axis should be determined based on the geometry of the detection unit and the distribution characteristics of the parameter data. For example, for rectangular detection units, the horizontal and vertical axes passing through the center point can be selected as the reference; for circular detection units, the axis in any diameter direction can be selected as the reference. The calculation of energy point offsets must ensure the consistency of the coordinate system to avoid deviations in the analysis results due to coordinate conversion errors.

[0079] The step size and termination criteria for increasing the amplification factor must be determined through preliminary experiments to balance data stacking adequacy with computational efficiency. A step size that is too small may result in excessive computational effort, while a step size that is too large may miss the optimal stacking state. The window length of the mean filter should be adjusted based on the degree of data fluctuation. For data with high noise, the window length can be increased to enhance the smoothing effect; for data with gentle fluctuations, a smaller window length can be used to preserve more detail.

[0080] When training recurrent neural network models, it's important to avoid overfitting. The model's generalization can be improved by adding dropout layers, setting early stopping mechanisms, or using regularization methods (such as L2 regularization). The diversity and representativeness of historical training data directly impact the model's predictive accuracy. It's important to collect component data covering a wide range of possible failure modes to ensure the model's reliability in real-world applications.

[0081] Through this process, the continuous strength assessment module evaluates component strength from the perspective of parameter data continuity. Combining topological network analysis with single-point strength assessment, it forms a multi-dimensional, comprehensive assessment system for component strength. This module not only identifies strength anomalies in individual inspection units but also, through analysis of continuous fluctuation bands, identifies strength continuity defects across the entire component or in localized regions, providing a scientific basis for component design optimization, service monitoring, and fault warning.

[0082] Example 5:

[0083] In the standardized processing link of the data acquisition module, dynamic range compression is one of the key steps. This processing adopts a nonlinear transformation method. Its core idea is to compress the original data range while retaining the relative distribution characteristics of the data through nonlinear function mapping. For example, for temperature data showing an exponential growth trend, a logarithmic transformation function can be used for compression to map a large range of temperature values ​​to a smaller numerical interval, avoiding numerical overflow or analysis deviation in subsequent processing due to the large data span. The specific function form of the nonlinear transformation needs to be selected according to the distribution characteristics of the data type (such as stress, strain, vibration frequency, etc.). For strain data with obvious peak characteristics, an S-type function can be used for compression to retain the data distribution details near the peak. At the same time, the long-tail data at both ends are compressed to a preset interval (such as [0, 1]) to ensure that data of different dimensions are analyzed at the same scale.

[0084] The dynamic adjustment mechanism of the sliding window filter's window length is closely related to the elastic modulus of the component material. The elastic modulus is a physical quantity that measures a material's ability to resist elastic deformation. Components of different materials (such as steel, aluminum alloy, and composite materials) have varying elastic modulus values, and their parameter data fluctuations also vary significantly. In implementation, a mapping table is first established between material elastic modulus and window length. For example, for steel with a high elastic modulus (e.g., 200 GPa), data fluctuations are typically more severe, so a smaller window length (e.g., 5 data points) is required to preserve high-frequency fluctuation details and avoid oversmoothing that obscures data features in areas of stress concentration. For aluminum alloys with a lower elastic modulus (e.g., 70 GPa), data fluctuations are relatively gradual, so a larger window length (e.g., 15 data points) can be used to effectively suppress low-frequency noise and highlight overall trends. Window length adjustment is automatically performed by the system's built-in algorithm. Upon detecting component material information, the system retrieves the corresponding window length parameter from the mapping table and applies it to the sliding window filtering process, achieving adaptive smoothing of data for components of different materials.

[0085] In the single-point strength analysis module, density analysis and structural feature identification of the topological network are key components in determining the strength of the test unit. When determining the density of the topological network covering a preset standard strength interval, the spatial extent of the preset standard strength interval must first be determined. This interval is defined by a geometric region formed by eight reference anchor points (e.g., three upstream energy peak points, five downstream energy release points, and evenly distributed nodes). Density is calculated by calculating the ratio of the number of edges and nodes in the topological network that fall within this interval to the total number of edges and nodes. If the density is less than a preset threshold (e.g., 60%), it indicates that the stress distribution within the test unit fails to effectively cover the critical strength region, and further structural feature analysis is required.

[0086] Structural feature identification first determines whether the distribution is linear or mesh-like. A linear distribution typically manifests as a chain-like connection of stress nodes along a single direction (such as the axial or radial direction), commonly seen in linear load-bearing components such as simply supported beams. A mesh-like distribution, on the other hand, features complex, multi-sided connections between nodes, resembling a grid structure, and is commonly seen in plate-like or shell-like components. If the topological network's structural features fall outside of these two categories (e.g., a scattered distribution or a locally clustered distribution), the detected unit is considered to be bearing concentrated pressure and may harbor a stress concentration defect. If the distribution is linear or mesh-like, the longitudinal or transverse span of the distribution is further examined. For linear distribution, for example, the longitudinal span refers to the distance the structural feature extends along the length of the component, while the transverse span refers to the extent of its extension along the cross-sectional direction. By traversing the data region and counting the energy intensity value variations, the starting and ending positions of the structural feature are determined, and the spatial span between them is calculated. If this span exceeds a preset span value (e.g., 10% of the component length), it indicates an abnormal distribution of the structural feature, potentially leading to localized strength deficiencies.

[0087] In the continuous strength assessment module, the longitudinal expansion of the data area is based on the axis where the center point of the unit is located. In specific operations, the data area of ​​each detection unit is extended to both sides along the axis direction, and the extension length is determined according to the spacing between adjacent units and the data superposition requirements. For example, if the center spacing between adjacent detection units is L, in order to ensure the overlap of the data area, the data area of ​​each unit can be extended to a length of 0.6L on both sides, so that the overlapping part of the data area of ​​adjacent units reaches 0.2L. Through this expansion method, the originally discrete detection unit data forms a continuous fluctuation curve, and the shape of the curve reflects the continuous change of the parameter data in the axis direction, such as the increase, decrease or periodic fluctuation of the stress value.

[0088] After generating a continuous data energy distribution band, the curvature rate of the distribution band fitting curve is calculated. The curvature rate reflects the rate of change in the curve's curvature. For an ideal continuous-strength component, its energy distribution curve should be smooth and continuous, with a low curvature rate of change. However, if there are strength discontinuities (such as material inhomogeneity or internal cracks), the curve will exhibit localized abrupt changes, resulting in a significant increase in the curvature rate of change. The calculated curvature rate of change is compared with a standard value (such as the average curvature rate of change for similar components). If the deviation exceeds a preset range (e.g., ±30%), the component strength continuity for that dimension is considered substandard.

[0089] In practical applications, nonlinear transformations for dynamic range compression require pre-verification. By comparing the data histograms before and after the transformation, the peak position, distribution pattern, and quantile relationship of the data are ensured to remain consistent, thus avoiding data distortion caused by the transformation. The mapping table for sliding window filtering must be established based on a large amount of historical data. Spectral analysis of sample data from components of different materials can be performed to determine the optimal functional relationship between window length and elastic modulus, thereby improving the universality of the filtering effect.

[0090] The density calculation and structural feature identification of the topological network need to be combined with the stress pattern of the component. For example, for components bearing axial tensile loads, a linearly distributed topological structure is normal, while a mesh-like distribution may indicate a complex stress state. For components bearing planar loads, a mesh-like distribution may be a reasonable structural feature, while a scattered distribution requires vigilance against stress concentration risks. The setting of the preset span value should refer to the component's design specifications and historical failure cases. For critical load-bearing components, the preset span value can be set to a smaller value (such as 50mm) to improve detection sensitivity. For non-critical components, the preset span value can be appropriately relaxed to a larger value (such as 100mm) to reduce misjudgment.

[0091] When expanding data for continuous strength assessment, it's important to ensure the axis direction aligns with the primary force direction of the component. For example, for beam components, longitudinal expansion along the length of the axis can more sensitively capture strength continuity issues caused by bending deformation. The curve fitting method (such as polynomial fitting or spline fitting) should be selected based on the data's fluctuation characteristics. For data with significant nonlinear trends, cubic spline fitting can more accurately describe the curve shape and avoid endpoint divergence issues associated with polynomial fitting.

[0092] Through the synergistic effect of various links in this embodiment, the system achieves full process optimization, from data acquisition and standardization to single-point strength analysis and continuous strength assessment. Dynamic range compression and sliding window filtering ensure data quality, topological network analysis and structural feature detection enable accurate single-point strength assessment, and continuous fluctuation band analysis and curvature change rate calculation assess the continuity of component strength from a holistic perspective. Together, they form a component strength assessment system based on big data, providing a scientific and reliable technical means for component safety assessment in the engineering field.

[0093] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "includes," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0094] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A component strength judgment system based on big data, characterized in that: include: The data acquisition module is used to obtain multi-dimensional parameter data of the component to be tested, and to perform standardization on the parameter data to divide the component main feature area and abnormal fluctuation area; A data preprocessing module is used to identify the boundaries of the main feature area of ​​the component and perform spatial mapping conversion on the parameter data according to the physical form of the component; The single-point strength analysis module is used to select any detection unit area of ​​the component, collect the starting and ending values ​​of each stress node based on the distributed sensor, mark them as reference points, obtain several reference points closest to the geometric center point of the detection unit, define them as key nodes, and sequentially associate the adjacent key nodes to generate a topological network. The density of the topological network covering the preset standard strength interval is obtained. When the density is less than a preset threshold, it is identified whether it is a linear distribution or a mesh distribution structural feature. If not, it is determined that the bearing pressure of the detection unit is concentrated; if so, the span of the longitudinal or transverse distribution of the structural feature is detected. If the span is greater than the preset span value, it is determined that the structural strength of the unit is abnormal. The continuous strength assessment module is used to select any one-dimensional parameter data sequence, longitudinally expand the data area of ​​each detection unit based on the axis of the unit center point, until the data areas in adjacent detection units form a continuous fluctuation curve, generate a continuous data energy distribution band, and calculate the curvature change rate of the distribution band fitting curve. When the deviation of the change rate from the standard value exceeds a preset range, it is judged that the strength continuity of the component in this dimension does not meet the standard; The preset standard strength interval consists of 8 reference anchor points, of which 3 anchor points are set in the upstream area of ​​the detection unit, and the other 5 anchor points are set in the downstream area of ​​the detection unit. The three points in the upstream area are located at the energy peak point of the axial distribution; the two anchor points in the downstream area are located at the intersection of the unit boundary line and the stress conduction path, that is, the energy release point, and the remaining three points in the downstream area are located at the three uniformly distributed nodes of the energy release point connection line; The specific process of judging whether the component strength continuity does not meet the standards is as follows: For any selected detection unit, select the reference axis where the center point of the unit is located, obtain the offset of all energy points in the data area relative to the axis, adjust the offset of each energy point by a preset amplification factor, and continuously increase the value of the amplification factor until adjacent detection units form a data superposition area, obtain a continuous fluctuation band of energy distribution, obtain the corresponding fitting curve equation by mean filtering the fluctuation band, and obtain the curvature change rate of the curve equation.

2. The component strength judgment system based on big data according to claim 1, characterized in that: The process of the standardization process is: The multidimensional parameter data is processed to eliminate outliers to generate cleaned data, the cleaned data is subjected to dynamic range compression, and the cleaned data is converted into normalized data by setting a data threshold. The normalized data is smoothed based on sliding window filtering, and trend decomposition is performed on the smoothed normalized data to separate the component main feature area from the abnormal fluctuation area.

3. The component strength judgment system based on big data according to claim 1, characterized in that: The process of spatial mapping conversion of parameter data is: The boundary contour line of the main characteristic area of ​​the component is obtained, and the azimuth deviation angles between the four main axes of the boundary and the parameter data coordinate system are calculated respectively. The parameter data is calibrated in the spatial coordinate system according to the weighted average of the azimuth deviation angles.

4. The component strength judgment system based on big data according to claim 1, characterized in that: The process of detecting the vertical or horizontal distribution span of structural features is: The data area is traversed according to the detection direction, which is a longitudinal section or a transverse section. The energy intensity value of each group of data detected is counted, the energy value of the abnormal fluctuation area is 255, and the energy value of the main feature area is 0. The first detection group whose energy value exceeds the main feature threshold is selected and marked as the abnormal starting group. The data area is continued to be traversed. When a detection group whose energy value returns to the main feature threshold is detected, it is marked as a pending group, and the spatial interval L1 between the pending group and the starting group is obtained. If the spatial interval L1 is greater than or equal to the preset span N, and no energy abnormality is detected along the detection direction, the pending group is marked as the termination group, and the structure has a uniform distribution feature; if the spatial interval is less than l1, the data area is continued to be traversed. When a detection group whose energy exceeds the main feature threshold is detected again, it is marked as the second starting group, and the interval L2 between the second starting group and the pending group is obtained. If the interval L2 is greater than the preset span M, it is determined that the structure distribution is abnormal.

5. The component strength judgment system based on big data according to claim 1, characterized in that: The process of generating a topology network is as follows: First, the four reference points closest to the center point of the detection unit are selected, and the adjacent reference points are associated to generate an initial topological structure. Then, the nearest points among the remaining reference points are selected one by one and added to the association of the topological structure. When the length of the connection path corresponding to the original reference point in the association path formed by the newly added reference point increases, the newly added point is removed, and the association path formed by the original reference point is the topological network of the detection unit.

6. The component strength judgment system based on big data according to claim 1, characterized in that: The data acquisition module includes a temperature sensor array, a vibration frequency collector and a strain gauge group, wherein the temperature sensors are arranged on the surface of the component in a honeycomb structure, the vibration frequency collectors are axially distributed at equal intervals, and the strain gauge group covers all stress concentration areas of the component.

7. The component strength judgment system based on big data according to claim 2, characterized in that: The dynamic range compression adopts a nonlinear transformation method to compress the range of the original data to a preset range while retaining the data distribution characteristics. The window length of the sliding window filter is dynamically adjusted according to the elastic modulus of the component material.

8. The component strength judgment system based on big data according to claim 1, characterized in that: The continuous strength assessment module has a built-in recurrent neural network model, which is trained using historical component strength data, using the curvature change rate of the energy distribution zone as an input feature and the probability of structural failure as an output result. During model training, a dynamic weight adjustment strategy is used to optimize the loss function.

Images