Method and system for detecting hardness of powder metallurgy gear

Abstract

The invention discloses a powder metallurgy gear hardness detection method and system, and belongs to the technical field of metallurgy gears, and the method comprises the steps: obtaining a to-be-detected gear three-dimensional model, and dividing a hardness evaluation region; a thermal response image is collected, a thermal response time gradient parameter TRTG is extracted, and a thermal diffusion characteristic matrix T is formed; establishing a hardness initial model H1 based on the TRTG and compactness; constructing a finite element stress field model, extracting a stress distribution modulation coefficient SDMC, and forming a stress characteristic matrix S; performing weighted correction on H1 based on T and S, and establishing a comprehensive hardness prediction model H2; outputting a hardness prediction result of the whole gear and the key area, and identifying a potential weak area; according to the method, thermal response and stress modulation parameters are fused, non-destructive and high-resolution hardness distribution prediction is achieved, and the method is suitable for quality evaluation and failure early warning of the powder metallurgy gear of a complex structure.

Description

Technical Field

[0001] The present invention relates to the technical field of metallurgical gears, and particularly relates to a method and system for detecting the hardness of powder metallurgy gears. Background Art

[0002] During the hardness detection process of powder metallurgy gears, traditional methods such as Rockwell hardness testers or Vickers hardness testers often require cutting, embedding, or polishing of gear specimens to meet the detection accuracy requirements. However, due to the porous structure and anisotropy of powder metallurgy materials, the traditional destructive specimen preparation process may cause local deformation of the original structure or pore collapse, resulting in hardness value deviation, seriously affecting the representativeness of the detection results and the accuracy of engineering judgment. In addition, the tooth surface of powder metallurgy gears undergoes surface treatments such as sintering and carburizing. If the general indentation method is used for hardness testing, it is difficult to reflect the gradient hardness distribution characteristics of the whole gear, especially when there is uneven hardness at the tooth root and tooth tip, it is more difficult to accurately evaluate its service life. Summary of the Invention

[0003] The purpose of the present invention is to provide a method and system for detecting the hardness of powder metallurgy gears to solve the deficiencies in the background art.

[0004] To achieve the above purpose, the present invention provides the following technical solutions: A method for detecting the hardness of powder metallurgy gears, comprising: S100, obtaining the three-dimensional geometric model and tooth surface spatial coordinate information of the powder metallurgy gear to be measured, and dividing the hardness evaluation area; S200, collecting a sequence of thermal response images of the gear under set thermal excitation conditions, extracting the thermal response time gradient parameter TRTG of the corresponding area, and forming a thermal diffusion feature matrix T; S300, establishing an initial mapping relationship model H1 between TRTG and hardness based on the material heat conduction model and the gear sintering denseness index; S400, constructing a finite element stress field simulation model of the gear under the loading state, extracting the stress distribution modulation coefficient SDMC of each evaluation area, and forming a stress action feature matrix S; S500, based on T and S, performing weighted correction on the hardness initial mapping relationship model H1 to establish a comprehensive hardness prediction model H2; S600, according to the model H2, outputting the hardness prediction results of the whole gear and the stress areas such as the tooth tip, tooth root, and tooth flank, and identifying potential weak areas.

[0005] Preferably, the S100 includes: S101, using a structured light three-dimensional reconstruction device to perform a full-circle angle scan on the powder metallurgy gear, obtaining the gear shape point cloud data set, and eliminating the interference of fine burrs on the tooth surface through an edge filtering algorithm; S102. Based on the gear point cloud model, construct a multi-resolution surface mesh, and use the topographic equal-height band weight function to automatically cluster and divide the tooth surface area to obtain refined evaluation units for structurally complex regions; S103. According to the gear meshing mechanical characteristic parameters, introduce a dynamic stress flux map, locally adjust and re-group the initial evaluation units to form a hardness evaluation area division result coupled with the stress path.

[0006] Preferably, the S200 includes: S201. Use a non-contact area array short-wave infrared thermal imager to continuously image the gear under the action of a modulated heat source pulse, record the thermal response image sequence within a preset time interval, and extract the instantaneous temperature change surface through a time-segment demodulation algorithm; S202. Perform regional normalization processing on the thermal response images, construct a local thermal image response function, and calculate the thermal response rise time and peak delay time within each evaluation area based on an improved exponential temperature decay fitting model; S203. Form a thermal diffusion feature matrix T for subsequent modeling by constructing a thermal response time gradient parameter TRTG from the thermal response rise time and delay time of each evaluation area and performing principal component dimensionality reduction on the TRTG.

[0007] Preferably, constructing the thermal response time gradient parameter TRTG includes: Based on the image partition result, divide the infrared thermal response image into multiple micro-region response units, construct a thermal response-time function for each unit, and extract its thermal response rise time trise; Use an adaptive time window algorithm to locally correct the temperature peak delay time tdelay of each response unit, and the correction basis includes the change of the first derivative inflection point of the infrared signal and the background thermal drift model; Utilize the dynamic non-linear coupling relationship between trise and tdelay to construct a local thermal time difference spectrogram, and vectorize the coupling parameters in the spectrogram to form the comprehensive thermal response time gradient parameter TRTG.

[0008] Preferably, the S300 includes: S301. Obtain the microstructure image of the powder metallurgy gear to be measured, combine the image segmentation algorithm to identify the pore distribution area, calculate the density index, and construct a denseness factor map of the tooth surface spatial distribution; S302. Based on the heat conduction coupling model, use the TRTG parameter as the input variable, and introduce the denseness factor as the spatial weighted correction parameter to construct an initial multi-factor heat diffusion hardness prediction model H10; In S303, perform a residual distribution regression analysis on H10, and optimize the model structure according to the residual space clustering characteristics to form an initial hardness mapping model H1 that distinguishes different compactness levels and thermal response paths.

[0009] Preferably, the step S400: S401, construct a finite element simulation model of the gear under typical working conditions. The model uses three-dimensional tooth surface contact elements and sets non-uniform friction boundary conditions to simulate typical load situations of tooth surface meshing and tooth root bending; S402, based on the time step control method, perform dynamic loading simulation on the model, and extract the stress distribution response curves of each evaluation area at the key loading stages; S403, use the stress response path mapping algorithm, combined with the tooth surface space coordinate system, to map the stress response curve to the thermal evaluation area in a normalized manner to form a regional response path data set; S404, perform a main frequency component spectrum analysis on the path data, extract the main frequency amplitudes of the stress in each evaluation area, and construct a stress distribution modulation coefficient SDMC for generating a stress action characteristic matrix S.

[0010] Preferably, extracting the main frequency amplitudes of the stress in each evaluation area and constructing the stress distribution modulation coefficient SDMC includes: Perform wavelet packet frequency domain decomposition on the stress response path data of each evaluation area, extract the multi-scale stress response frequency bands, and construct a frequency domain energy distribution map; Based on the frequency band energy distribution results, calculate the spectral concentration factor of each area; Extract the main frequency amplitude of the frequency band corresponding to the spectral concentration factor, and combine with the regional geometric constraint conditions to generate a normalized main frequency response intensity; Based on the main frequency response intensity and the spectral concentration factor, construct a two-factor weighted function, and define and calculate the stress distribution modulation coefficient SDMC.

[0011] Preferably, the S500 includes: S501, perform tensor fusion on the thermal diffusion characteristic matrix T and the stress action characteristic matrix S to construct a multi-source collaborative feature body, and use a mutual information matching mechanism to screen high-coupling feature channels; S502, based on the double-weight collaborative filtering model, correct the initial hardness mapping model H1. The model applies the thermal response weight and the stress modulation factor as independent convolution kernel weights to the prediction path; S503, construct a residual feedback loop, adaptively backpropagate the difference between the predicted result of the model H1 and the actual reference hardness, and dynamically update the thermal stress coupling factor to achieve iterative convergence of the model structure and output a comprehensive hardness prediction model H2.

[0012] Preferably, the S600 includes: S601, based on the output result of the comprehensive hardness prediction model H2, construct a gear surface hardness distribution map, and perform spatial matching cutting on the tooth tip, tooth root, and tooth flank regions; S602, calculate the local gradient tensor field for the hardness prediction results within the cut region, and extract the hardness sharp change path based on the gradient tensor difference to identify potential abnormal change trends; S603, in combination with the stress modulation coefficient SDMC, extract the mechanical vulnerability response factor VFI on the sharp change path, and establish an intelligent weak point marking mechanism for the abnormal aggregation area of the response factor; S604, adopt the fuzzy boundary set partitioning algorithm to identify the continuously changing area and the boundary transition area as the hardness fluctuation band, and output a potential weak area evaluation report including the coordinates and grade labels of high-risk points.

[0013] The present invention also provides a powder metallurgy gear hardness detection system, including: A region division module, which acquires the three-dimensional geometric model and tooth surface spatial coordinate information of the powder metallurgy gear to be measured, and divides the hardness evaluation region; A thermal response acquisition module, which acquires a sequence of thermal response images of the gear under set thermal excitation conditions, extracts the thermal response time gradient parameter TRTG of the corresponding region, and forms a thermal diffusion feature matrix T; A thermal hardness initial modeling module, which establishes an initial mapping relationship model H1 between TRTG and hardness based on the material heat conduction model and the gear sintering densification index; A stress simulation and modulation coefficient extraction module, which constructs a finite element stress field simulation model under the gear loading state, extracts the stress distribution modulation coefficient SDMC of each evaluation region, and forms a stress action feature matrix S; A coupled modeling and hardness prediction correction module, which performs weighted correction on the initial mapping relationship model H1 between hardness based on T and S, and establishes a comprehensive hardness prediction model H2; A weak point identification module, which, according to the output of model H2, obtains the hardness prediction results of the overall gear and the stress regions of the tooth tip, tooth root, and tooth flank, and identifies potential weak regions.

[0014] In the above technical solution, the technical effects and advantages provided by the present invention are as follows: 1. By introducing two key physical quantities, namely the thermal response time gradient parameter (TRTG) and the stress distribution modulation coefficient (SDMC), the present invention establishes a thermal-mechanical multi-field coupling modeling mechanism, significantly improving the hardness prediction accuracy of powder metallurgy gears under complex microstructures and non-uniformity conditions. Compared with traditional single-point hardness detection or average value methods, the present invention can achieve spatial distribution hardness modeling of key stress regions such as tooth tips, tooth roots, and tooth flanks, identify potential weak points with physical basis and engineering significance, and improve the representativeness of detection results and the risk discrimination ability.

[0015] 2. In the modeling process of the present invention, high-dimensional feature fusion technology, wavelet spectrum analysis, residual feedback mechanism and fuzzy set boundary recognition algorithm are integrated, which not only enhances the adaptability of the prediction model in complex structures, but also has a high degree of automation and intelligence. It is applicable to the hardness evaluation scenarios of large quantities, non-destructive, and high-resolution, providing strong technical support for powder metallurgy process optimization, quality grading and service life warning, and has broad industrial application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings.

[0017] Figure 1 It is a flowchart of the method of the present invention.

[0018] Figure 2 It is a flowchart of the system module of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0019] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0020] Embodiment 1. Please refer to Figure 1 As shown, a method for detecting the hardness of a powder metallurgy gear in this embodiment includes: S100. Obtain the three-dimensional geometric model and tooth surface space coordinate information of the powder metallurgy gear to be measured, and divide the hardness evaluation area; S200. Collect the thermal response image sequence of the gear under the set thermal excitation conditions, extract the thermal response time gradient parameter TRTG of the corresponding area, and form the thermal diffusion feature matrix T; S300. Based on the material heat conduction model and the gear sintering denseness index, establish the initial mapping relationship model H1 between TRTG and hardness; S400. Construct a finite element stress field simulation model under the gear loading state, extract the stress distribution modulation coefficient SDMC of each evaluation area, and form the stress action feature matrix S; S500. Based on T and S, the initial hardness mapping relationship model H1 is weighted and corrected to establish a comprehensive hardness prediction model H2. S600. According to the model H2, the hardness prediction results of the overall gear and the stress regions of the tooth tip, tooth root, and tooth flank are output to identify potential weak regions.

[0021] In order to solve the technical problems in the existing hardness detection of powder metallurgy gears, such as unreasonable division of the evaluation area and local distortion of the detection results due to the complex tooth surface morphology and uneven distribution of geometric features, the present invention proposes a spatial region division method combining three-dimensional modeling, morphology clustering, and stress flux analysis for a more engineering-significant evaluation of the hardness distribution of gears.

[0022] This method first quickly obtains a high-precision geometric model of the powder metallurgy gear through structured light three-dimensional reconstruction technology, and further uses the edge filtering algorithm to eliminate local interference data on the tooth surface to ensure the modeling accuracy. On this basis, combined with the contour features of the tooth surface morphology, a multi-resolution surface grid model is constructed, and spatial clustering is performed based on curvature weights to achieve refined division of structure-sensitive regions. Finally, the finite element stress flux distribution under the gear meshing state is introduced to correct and re-optimize the preliminary region division to ensure that the hardness evaluation region is consistent with the actual force characteristics.

[0023] The process of dividing the hardness evaluation region includes the following technical steps: Three-dimensional model acquisition and point cloud optimization (S101): A structured light scanning device is used to perform multi-angle dynamic scanning on the powder metallurgy gear to be measured, and a high-density point cloud data set covering the tooth surface contour is obtained. To eliminate the noise interference caused by material porosity and surface micro-irregularities, the edge gradient filtering algorithm is used to correct the point cloud, thereby generating a three-dimensional geometric model with clear boundaries and retained details.

[0024] Tooth surface grid reconstruction and partition clustering (S102): Based on the above point cloud model, the tooth surface mesh is generated through normal reconstruction, and a hierarchical weight function is constructed using morphology factors such as local curvature and surface slope. The K-means clustering algorithm based on density and structure gradient is executed to achieve differential automatic division of multiple regions such as the tooth tip, tooth root, and tooth flank, and obtain preliminary hardness evaluation units.

[0025] Stress flux-assisted region optimization (S103): By simulating the load conditions of the gear, the stress flux distribution map under the key meshing state is obtained. This distribution data is used as a secondary weighting factor to fine-tune and re-divide the boundary of the initial clustering region, so that the region division is highly consistent with the force path during the actual operation of the gear, thereby improving the accuracy and engineering reliability of subsequent hardness prediction.

[0026] Through the above steps, the present invention can accurately divide the hardness evaluation area of the powder metallurgy gear, especially having obvious advantages in the areas with high curvature and complex meshing, providing a unified spatial basis for subsequent thermal response analysis and hardness modeling.

[0027] In order to extract more representative gradient feature parameters from the thermal response data and further improve the accuracy of predicting the hardness distribution of the powder metallurgy gear, the following method is used to process the thermal response image sequence and construct the thermal response time gradient parameter (TRTG): First, a non-contact area array short-wave infrared thermal imager is used to continuously thermally image the gear. During the imaging process, a modulated pulsed heat source is adopted to apply instantaneous thermal excitation to the gear at a set frequency and pulse width. Within each excitation period, the infrared thermal imager records the thermal response image sequence at a high frame rate, and using the time-segment demodulation algorithm, the original temperature image is converted into an instantaneous temperature change rate map, thereby obtaining a temperature change surface with more dynamic characteristics.

[0028] Subsequently, the collected infrared images are subjected to spatial registration and normalization processing according to the preset regional distribution results on the tooth surface. A local thermal image response function is established for each micro-region, and an improved exponential-type temperature decay fitting model is used to curve-fit the temperature change curve of each region, and two characteristic parameters, namely the thermal response rise time trise and the temperature peak delay time tdelay, are respectively extracted from it.

[0029] In the further feature construction stage, the above two time parameters are combined to form the preliminary thermal response time pairs of each evaluation area. Based on the data information of these pairs, the following technical means are used to construct the TRTG parameter: The thermal response image is subdivided into multiple micro-region response units, and a thermal response-time function model is established for each unit to accurately depict the response dynamics of the micro-region during the heating process; To avoid peak recognition errors caused by thermal image fluctuations or surface noise, an adaptive time window algorithm is used to perform multi-region fitting correction on tdelay. The basis is the inflection point position of the first derivative of the temperature response of each unit, and at the same time, the background thermal drift trend is considered for compensation; Utilize the dynamic non-linear coupling relationship between trise and tdelay to construct a local thermal time difference spectrogram, and vectorize the coupling parameters in the spectrogram to form the comprehensive thermal response time gradient parameter TRTG.

[0030] Finally, for the convenience of subsequent modeling processing, the TRTG data is subjected to principal component dimensionality reduction processing to compress the feature dimension and remove redundant information, forming a unified format of thermal diffusion feature matrix T as the input basic data for subsequent hardness modeling.

[0031] Through the above processing method, the present invention can not only accurately reflect the differences in heat conduction and microstructure in various regions of the gear, but also significantly improve the practical application effect of non-contact thermal detection in hardness prediction analysis.

[0032] To establish an effective mapping relationship between the thermal response time gradient parameter (TRTG) and the local hardness of the gear, a multi-parameter composite modeling method considering the material densification factor is proposed to establish the initial hardness prediction model H1, which specifically includes the following steps: First, by obtaining the microstructure images of the gear samples, image processing and analysis are carried out on areas such as the tooth surface and tooth root. An image segmentation algorithm based on Otsu threshold segmentation and edge enhancement is used to automatically identify the pore areas, and then the void ratio per unit area of the material is calculated to form a quantified densification index map. This map reflects the spatial differences in the sintering quality of different parts of the gear.

[0033] Secondly, the previously calculated TRTG parameter is combined with the densification index to construct a densification thermal weight function. On this basis, a coupled model based on the Fourier heat diffusion equation and the multi-variable regression algorithm is applied to establish a preliminary heat diffusion-hardness mapping model H10. Among them, TRTG is used as the dominant prediction variable, and the densification factor participates in the expression of the characteristic function in the form of a spatial weighting coefficient, correcting the limitation of the assumption of material homogeneity in the traditional heat conduction model.

[0034] To further improve the stability and local adaptability of the mapping model, regression residual analysis is performed on H10. By superimposing the residual heat map on the three-dimensional tooth surface model, the spatial aggregation trend of the model error is identified, and it is judged whether the error is related to the local microstructure, surface state or imaging accuracy. For positions with obvious aggregation of the error distribution, the model structure is adjusted, and local response weights or piecewise mapping strategies are introduced to optimize it into the final initial hardness mapping model H1.

[0035] This method not only improves the local accuracy of the hardness prediction results, but also realizes the modeling of the thermal diffusion response differences of the heterogeneous structure of the material, which is significantly better than the traditional linear heat conduction hardness regression method, and is particularly suitable for the hardness distribution analysis of complex regions of powder metallurgy structures.

[0036] To effectively assist the hardness evaluation with the stress distribution in the key areas of powder metallurgy gears, a stress modulation parameter modeling method combining dynamic finite element simulation and spectral feature extraction is proposed, that is, the stress distribution modulation coefficient (SDMC) extraction process.

[0037] First, a finite element model of the gear structure was created using a 3D modeling tool. The tooth surfaces in the model were described using 3D contact elements, and non-uniform friction boundary conditions were introduced between contact pairs to simulate the inconsistency between the actual tooth surface lubrication state and local friction. Regarding load setting, multiple time-varying load paths were set, combining the typical meshing loads and tooth root bending conditions experienced during gear transmission, to complete the simulation modeling of the actual operating conditions.

[0038] Subsequently, a loading simulation was performed on the finite element model using a time-step-controlled transient analysis method. The simulation process divided the critical loading phases according to the gear meshing cycle, and the stress response data for each evaluation area was recorded to form a set of stress response curves that evolved over time.

[0039] In order to make the stress information correspond one-to-one with the thermal response modeling area, a stress response path mapping algorithm based on coordinate mapping relationship is adopted to project the stress time history curve obtained by simulation into the aforementioned thermal assessment area, thereby obtaining a spatially consistent stress response path dataset.

[0040] For the above path data, in order to explore its frequency structure characteristics, wavelet packet decomposition is performed on each stress path data, the energy distribution within the multi-scale frequency bandwidth is extracted, and a frequency domain energy spectrum is constructed to reflect the dominant characteristics of the stress response in different frequency bands.

[0041] Based on the spectral energy distribution results, the spectral concentration factor is calculated. This quantitative indicator measures whether the energy is concentrated in a certain dominant frequency band and represents the frequency stability of the stress response. The dominant frequency amplitude is then extracted from the frequency band corresponding to the concentration factor. This is combined with the local size or curvature constraints of the region in the geometric structure to form the normalized dominant frequency response intensity.

[0042] Finally, based on the coupling relationship between the main frequency response intensity and the factors in the spectral concentration, a two-factor weighted model was constructed to form the stress distribution modulation coefficient (SDMC), which represents the typical mechanical modulation characteristics of the evaluation area. The SDMC values of all regions constitute a complete stress action characteristic matrix (S), which is used for subsequent coupled hardness modeling analysis with thermal response parameters.

[0043] Compared with the traditional modeling method based on maximum stress or average stress, this method can better reflect the structural dynamic response characteristics of powder metallurgy materials under complex loads, and is particularly suitable for situations where the microstructure of non-uniform materials is sensitive to mechanical propagation behavior.

[0044] To overcome the problems of low accuracy and strong sensitivity to local outliers in traditional hardness prediction models when processing multi-physics field coupling information, a method is proposed to jointly modify the initial model H1 based on the thermal response matrix T and the stress characteristic matrix S to establish a more robust and accurate comprehensive hardness prediction model H2. The method includes the following steps: First, perform tensor fusion processing on the extracted thermal diffusion feature matrix T and the stress action feature matrix S. During the fusion process, use a third-order tensor to construct a multi-source collaborative feature volume, and perform feature docking on the main thermal and mechanical response channels of each evaluation region in the two physical domains. At the same time, introduce a mutual information matching mechanism to calculate the coupled channels in the feature volume, eliminate redundant channels with low mutual information, and only retain high-weight feature channels with significant collaborative relationships to provide highly correlated input data for subsequent modeling.

[0045] Second, based on the above fusion results, construct a dual-weight collaborative filtering model. This model adopts a structure similar to a multi-channel convolutional filtering structure, but uses the thermal diffusion intensity weight and the stress main frequency modulation coefficient as two independent filtering kernels, and performs weighted convolution processing on the input paths in the model respectively. This design allows the model to capture asymmetric modulation features from the two physical domains, thereby realizing directional correction of the initial hardness mapping model H1. The filtered output is injected into the prediction path as a new feature vector group to generate an intermediate corrected output.

[0046] Finally, to improve the model's fitting ability for actual hardness values, introduce a residual feedback loop mechanism in the training and regression stages. This mechanism uses the difference between the H2 prediction result and the reference hardness value as the residual index, establishes a feedback path to dynamically adjust the parameter distribution of the collaborative filter, and gives higher adjustment weights to abnormal regions in particular to achieve adaptive learning and convergence of the model in high-deviation regions. After multiple rounds of feedback correction, a converged and stable comprehensive hardness prediction model H2 is formed, which has higher prediction accuracy for regions such as local soft spots and hardness mutation bands within the entire gear range.

[0047] This method significantly improves the adaptability and accuracy of the hardness model under the conditions of thermo-mechanical non-uniform coupling characteristics of powder metallurgy materials, and is a multi-source intelligent modeling method for complex structural states.

[0048] To achieve effective analysis of the hardness distribution of the overall powder metallurgy gear and key stressed areas and intelligent identification of weak areas, a method for structural functional evaluation based on the output results of the comprehensive hardness prediction model H2 is proposed. This method not only considers the predicted value itself, but also integrates hardness gradient changes, stress modulation effects, and fuzzy boundary determination, thereby improving the accuracy and engineering usability of the identification results.

[0049] First, based on the predicted output of the H2 model, the system generates a hardness distribution map on the three-dimensional structure of the complete gear. This distribution map uses the gear tooth surface as the projection reference, performs spatial reconstruction by region, and maps the predicted values onto the model's geometric structure to form a hardness data set with spatial resolution. Subsequently, according to the functional areas marked in the three-dimensional CAD model of the gear, such as the tooth tip, tooth root, tooth flank, etc., the system performs spatial matching and cutting operations to cut out the hardness prediction results by region, providing a data basis for subsequent local analysis.

[0050] For each cut region, the system constructs a local gradient tensor field within this region to characterize the spatial variation trend of the predicted hardness. This gradient tensor not only considers the change of the first-order derivative in the main direction but also describes the "severity of hardness change" and the local concentration of the change through the introduction of a second-order derivative matrix form. For example, in the tooth root transition region or the region where the tooth flank curvature changes rapidly, if the hardness change gradient shows high-order discontinuity, it often indicates local anomalies in the material structure, density, or stress response.

[0051] Based on the gradient tensor, the system performs gradient difference analysis to identify the "drastic change paths" of local hardness, that is, the transition paths from a normal hardness region to a region with a significantly deviated hardness value. These paths characterize the potential weak point evolution trend. However, since the hardness change is not necessarily equivalent to structural weakness, it is necessary to further evaluate its actual mechanical vulnerability.

[0052] Therefore, the system further combines the stress distribution modulation coefficient SDMC in the previously calculated stress action characteristic matrix S to extract the mechanical vulnerability response factor VFI for each drastic change path region. This factor is calculated by weighted summation of the hardness mutation amplitude, change rate, and stress modulation degree, and is used to identify the regions that are most likely to undergo local failure under the action of stress coupling. Especially in powder metallurgy gears, the micropores caused by uneven sintering density are prone to cause the "fuse effect" of structural fracture. The introduction of VFI effectively avoids the misjudgment risk brought by judging defects only based on thermal response changes.

[0053] After the above VFI extraction is completed, the system fuses the VFI spatial distribution results with the gradient tensor information to establish an intelligent weak point identification mechanism. Using the outlier clustering and aggregation threshold strategy in machine learning, the regions with abnormally dense responses are marked as "primary risk weak points", the regions with drastic changes but no abnormal aggregation of VFI are regarded as "secondary warning regions", and the data in the transition zone is retained for further judgment.

[0054] Considering that the hardness gradient characteristics of powder metallurgy materials do not have strict boundaries, the system introduces a fuzzy boundary set partitioning algorithm. This algorithm takes the spatial continuity function of the predicted hardness value as the core, constructs a fuzzy membership function, fits the interval of the continuously changing region, and distinguishes the obvious mutation and the boundary of the transition zone. Using the fuzzy set recognition structure, the hardness fluctuation regions (such as the tooth tip edge, tooth root corner, etc.) are divided into three states: the boundary fuzzy zone, the stable zone, and the mutation zone, providing clear spatial information partitioning for subsequent process adjustment.

[0055] Finally, the system outputs a weak area assessment report containing the following information: The three-dimensional coordinates of each high-risk point; The identification of the tooth surface area to which it belongs (tooth tip / tooth root / tooth flank); The corresponding predicted hardness value, VFI value, and gradient tensor strength; The risk level label (primary weak point / secondary fluctuation band); The recommended marking range and suggestions for subsequent process inspection.

[0056] This method realizes an integrated closed-loop process from "numerical evaluation" to "structural risk identification" by extracting high-dimensional physical features from the model prediction values and integrating the logic of the structural functional regions. It is especially suitable for scenarios with high requirements for reliability judgment due to non-uniform hardness distribution in complex configurations and heterogeneous structure materials (such as powder metallurgy parts). Compared with the existing method of only setting thresholds based on the hardness value itself to judge defects, this method has more engineering decision-making value.

[0057] Example 2, please refer to Figure 2 As shown, a hardness detection system for a powder metallurgy gear in this embodiment includes: A region partitioning module, which obtains the three-dimensional geometric model and tooth surface spatial coordinate information of the powder metallurgy gear to be measured, and partitions the hardness evaluation region; A thermal response acquisition module, which acquires a sequence of thermal response images of the gear under set thermal excitation conditions, extracts the thermal response time gradient parameter TRTG of the corresponding region, and forms a thermal diffusion feature matrix T; A thermal hardness initial modeling module, which establishes an initial mapping relationship model H1 between TRTG and hardness based on the material heat conduction model and the gear sintering denseness index; A stress simulation and modulation coefficient extraction module, which constructs a finite element stress field simulation model under the loaded state of the gear, extracts the stress distribution modulation coefficient SDMC of each evaluation region, and forms a stress action feature matrix S; A coupled modeling and hardness prediction correction module, which based on T and S, performs weighted correction on the hardness initial mapping relationship model H1 to establish a comprehensive hardness prediction model H2; The weak point identification module identifies potential weak areas based on the hardness prediction results of the overall gear, the tooth tip, the tooth root, and the tooth flank stress areas output by model H2.

[0058] The above is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of changes or substitutions, which should all be covered within the protection scope of the present application.

Claims

1. A method for detecting the hardness of a powder metallurgy gear, characterized in that: Including: S100. Obtain the three-dimensional geometric model and tooth surface spatial coordinate information of the powder metallurgy gear to be measured, and divide the hardness evaluation area; S200. Collect the thermal response image sequence of the gear under the set thermal excitation conditions, extract the thermal response time gradient parameter TRTG of the corresponding area, and form the thermal diffusion feature matrix T; S300. Based on the material heat conduction model and the gear sintering density index, establish the initial mapping relationship model H1 between TRTG and hardness; S400. Construct a finite element stress field simulation model of the gear under the loaded state, extract the stress distribution modulation coefficient SDMC of each evaluation area, and form the stress action feature matrix S; S500. Based on T and S, perform weighted correction on the hardness initial mapping relationship model H1, and establish a comprehensive hardness prediction model H2; S600. According to the model H2, output the hardness prediction results of the overall gear and the stress areas of the tooth tip, tooth root and tooth flank, and identify potential weak areas.

2. The method for detecting the hardness of a powder metallurgy gear according to claim 1, wherein: The S100 includes: S101. Use a structured light three-dimensional reconstruction device to perform a full-circle angle scan on the powder metallurgy gear, obtain the gear shape point cloud data set, and eliminate the interference of fine burrs on the tooth surface through an edge filtering algorithm; S102. Based on the gear point cloud model, construct a multi-resolution surface mesh, and use a topography equal-height band weight function to automatically cluster and divide the tooth surface area to obtain a refined evaluation unit for the structurally complex area; S103. Introduce a dynamic stress flux map according to the gear meshing mechanical characteristic parameters, perform local adjustment and re-grouping on the initial evaluation unit, and form a hardness evaluation area division result coupled with the stress path.

3. The method for detecting the hardness of a powder metallurgy gear according to claim 1, characterized in that: The S200 includes: S201. Use a non-contact area array short-wave infrared thermal imager to continuously image the gear under the action of a modulated heat source pulse, record the thermal response image sequence within a preset time interval, and extract the instantaneous temperature change surface through a time-segment demodulation algorithm; S202. Perform regional normalization processing on the thermal response image, construct a local thermal image response function, and calculate the thermal response rise time and peak delay time in each evaluation area based on an improved exponential temperature decay fitting model; S203. The thermal response rise time and delay time of each evaluation area form the thermal response time gradient parameter TRTG, and perform principal component dimensionality reduction processing on TRTG to form the thermal diffusion feature matrix T for subsequent modeling.

4. The method for detecting the hardness of a powder metallurgy gear according to claim 3, wherein: Constructing the thermal response time gradient parameter TRTG includes: Based on the image partition result, divide the infrared thermal response image into multiple micro-area response units, construct a thermal response-time function for each unit, and extract its thermal response rise time trise; Use an adaptive time window algorithm to locally correct the temperature peak delay time tdelay of each response unit, and the correction basis includes the change of the first derivative inflection point of the infrared signal and the background thermal drift model; Utilize the dynamic non-linear coupling relationship between trise and tdelay to construct a local thermal time difference spectrogram, and vectorize the coupling parameters in the spectrogram to form the comprehensive thermal response time gradient parameter TRTG.

5. A method for detecting the hardness of a powder metallurgy gear according to claim 1, characterized in that: The S300 includes: S301. Obtain the microstructural image of the powder metallurgy gear to be measured, identify the pore distribution area by combining with the image segmentation algorithm, calculate the density index, and construct the compactness factor map of the tooth surface spatial distribution; S302. Based on the heat conduction coupling model, take the TRTG parameters as input variables, and introduce the compactness factor as the spatial weighted correction parameter to construct the initial multi-factor thermal diffusion hardness prediction model H10; S303. Conduct residual distribution regression analysis on H10, and optimize the model structure according to the residual space clustering characteristics to form the initial hardness mapping model H1 that distinguishes different compactness levels and thermal response paths.

6. A method for detecting the hardness of a powder metallurgy gear according to claim 1, characterized in that: The steps of S400: S401. Construct a finite element simulation model of the gear under typical working conditions. The model uses three-dimensional tooth surface contact elements and sets non-uniform friction boundary conditions to simulate the typical load conditions of tooth surface meshing and tooth root bending; S402. Based on the time step control method, conduct dynamic loading simulation on the model, and extract the stress distribution response curves of each evaluation area at the key loading stages; S403. Adopt the stress response path mapping algorithm, combine with the tooth surface spatial coordinate system, and map the stress response curve to the thermal evaluation area in a normalized manner to form the regional response path data set; S404. Conduct main frequency component spectrum analysis on the path data, extract the main frequency amplitudes of the stress in each evaluation area, and construct the stress distribution modulation coefficient SDMC for generating the stress action characteristic matrix S.

7. A method for detecting the hardness of a powder metallurgy gear according to claim 6, characterized in that: Extracting the main frequency amplitudes of the stress in each evaluation area and constructing the stress distribution modulation coefficient SDMC includes: Conduct wavelet packet frequency domain decomposition on the stress response path data of each evaluation area, extract the multi-scale stress response frequency bands, and construct the frequency domain energy distribution map; Based on the frequency band energy distribution results, calculate the spectral concentration factor of each area; Extract the main frequency amplitude of the frequency band corresponding to the spectral concentration factor, and combine with the regional geometric constraint conditions to generate the normalized main frequency response intensity; Based on the main frequency response intensity and the spectral concentration factor, construct a two-factor weighted function, define and calculate the stress distribution modulation coefficient SDMC.

8. A method for detecting the hardness of a powder metallurgy gear according to claim 1, characterized in that: The S500 includes: S501. Conduct tensor fusion on the thermal diffusion characteristic matrix T and the stress action characteristic matrix S to construct a multi-source collaborative feature body, and use the mutual information matching mechanism to screen the high-coupling feature channels; S502. Based on the double-weight collaborative filtering model, correct the initial hardness mapping model H1. The model applies the thermal response weight and the stress modulation factor as independent convolution kernel weights to the prediction path; S503. Construct a residual feedback loop, adaptively backpropagate the difference between the prediction result of model H1 and the actual reference hardness, and dynamically update the thermal stress coupling factor to achieve iterative convergence of the model structure and output the comprehensive hardness prediction model H2.

9. A method for detecting the hardness of a powder metallurgy gear according to claim 1, characterized in that: The S600 includes: S601. Based on the output result of the comprehensive hardness prediction model H2, construct the surface hardness distribution map of the gear, and perform spatial matching cutting on the tooth tip, tooth root, and tooth flank areas; S602. Calculate the local gradient tensor field for the hardness prediction results in the cut area, and extract the hardness sharp change path based on the gradient tensor difference to identify potential abnormal change trends; S603. In combination with the stress modulation coefficient SDMC, extract the mechanical vulnerability response factor VFI on the drastic change path, and establish an intelligent weak point marking mechanism for the abnormal aggregation area of the response factor. S604. Adopt the fuzzy boundary set partitioning algorithm to identify the continuously changing area and the boundary transition area as the hardness fluctuation band, and output a potential weak area assessment report including the coordinates and grade labels of high-risk points.

10. A powder metallurgy gear hardness detection system for implementing the powder metallurgy gear hardness detection method according to any one of claims 1-9, characterized in that: It includes: An area partitioning module, which obtains the three-dimensional geometric model and the tooth surface space coordinate information of the powder metallurgy gear to be measured, and partitions the hardness evaluation area. A thermal response acquisition module, which acquires the thermal response image sequence of the gear under the set thermal excitation conditions, extracts the thermal response time gradient parameter TRTG of the corresponding area, and forms a thermal diffusion feature matrix T. A thermal hardness initial modeling module, which based on the material heat conduction model and the gear sintering density index, establishes an initial mapping relationship model H1 between TRTG and hardness. A stress simulation and modulation coefficient extraction module, which constructs a finite element stress field simulation model under the loading state of the gear, extracts the stress distribution modulation coefficient SDMC of each evaluation area, and forms a stress action feature matrix S. A coupled modeling and hardness prediction correction module, which based on T and S, performs weighted correction on the hardness initial mapping relationship model H1, and establishes a comprehensive hardness prediction model H2. A weak point identification module, which according to the output of model H2, identifies the potential weak areas by predicting the hardness of the overall gear and the stress areas of the tooth tip, tooth root and tooth flank.

Images