Microstructure intelligent recognition and quantitative evaluation method fusing physical mechanism

Abstract

The present application relates to the technical field of physical and chemical detection of metal materials, and particularly relates to a microstructure intelligent recognition and quantitative evaluation method fusing physical mechanism, comprising: using the difference in grain boundary corrosion potential to perform adaptive contrast enhancement on a metallographic image, and constructing a high signal-to-noise ratio physical feature field; constructing a topological reconstruction model fusing martensitic phase change mechanism, flow field dynamics simulation and thermodynamic equilibrium repair, and restoring a closed grain network conforming to physical reality; based on the reconstructed grain topological network, through equivalent ellipse fitting and area weighting algorithm, multi-dimensional quantitative characterization and noise-resistant rating of grain effective size, morphological anisotropy and texture orientation are realized; through sliding window scanning to generate a local variation coefficient thermal map, the non-uniformity of the structure is quantitatively evaluated. The present application effectively solves the problems of incomplete grain boundary recognition, statistical rating distortion and local defect missed detection in the prior art, and realizes automatic early warning of local coarse grain and mixed grain risks of high-performance metal materials.

Description

Technical Field

[0001] This invention relates to the field of physical and chemical testing technology for metallic materials, and in particular to a method for intelligent identification and quantitative evaluation of microstructures that integrates physical mechanisms. Background Technology

[0002] High-performance metallic components are core parts of high-end equipment in energy, aerospace, and other fields, and their service safety directly depends on the microstructure of the materials. In the quality control of high-end equipment manufacturing, the grain size rating of metallic materials with complex phase transformation structures faces significant challenges, primarily including:

[0003] Limitations of physical feature extraction: After tempering, the microstructure of this type of material consists of interwoven martensite / bainite laths and proto-austenite grain boundaries. Due to the differences in interfacial energy between different types of grain boundaries, the contrast varies during chemical etching. Existing image detection techniques lack the constraints of prior knowledge in physical metallurgy, making it impossible to distinguish between high-energy grain boundaries and low-energy subgrain boundaries. This often results in severe grain boundary fractures or artifact interference in the detection results, failing to reconstruct the true grain topology.

[0004] The evaluation system suffers from a lack of dimensionality and statistical distortion: Current testing standards and automated equipment typically output only a single average grain size. However, for high-performance components that have undergone large deformation forging, their microstructure often exhibits significant anisotropy. A major drawback of existing technologies is the incompleteness of statistical information, losing crucial physical information such as grain morphology (e.g., aspect ratio) and preferred orientation (e.g., streamline angle), making it difficult to explain the anisotropic mechanical behavior of materials. Furthermore, traditional geometric statistical methods are susceptible to interference from non-structural noise during sample preparation (e.g., corrosion pits), often leading to inflated effective grain size ratings that fail to accurately reflect the strength and toughness potential of the material.

[0005] Missed detection of local defects: Existing testing equipment only outputs global average indicators and cannot locate mixed crystal or banded structures within the material. For such materials, local coarse-grained regions are often the source of fatigue crack initiation. This hidden danger of average values ​​being acceptable but local failures is a major blind spot in the quality control of high-end equipment.

[0006] In summary, existing evaluation methods lack the constraints of prior knowledge in physical metallurgy, resulting in severe interference from high-energy grain boundaries and low-energy subgrain boundaries in the detection of complex non-equiaxed crystalline structures. Furthermore, they suffer from limitations such as a single evaluation dimension and the inability to quantitatively characterize local thermodynamic inhomogeneities. Therefore, a new intelligent identification and evaluation method is urgently needed. Summary of the Invention

[0007] To address the aforementioned problems, the present invention aims to provide a method for intelligent identification and quantitative evaluation of microstructures that integrates physical mechanisms, thereby resolving issues such as the lack of physical authenticity and the distortion of the evaluation system in existing microstructure detection technologies for complex metallic materials.

[0008] The technical solution adopted in this invention is as follows: The present invention proposes a method for intelligent identification and quantitative evaluation of microstructures that integrates physical mechanisms, specifically including the following steps: S1. Adaptive contrast enhancement of metallographic images using grain boundary corrosion potential differences; S2. Construct a topology reconstruction model; the topology reconstruction model is based on a pre-trained dual-flow deep neural network, takes the enhanced metallographic image as input, and outputs a spatial gradient flow field and a grain boundary probability map; based on the spatial gradient flow field and the grain boundary probability map, grain boundary repair is performed on the grain boundary breaks caused by corrosion discontinuity with the grain boundary energy minimization and the mechanical balance of the three-way node as boundary conditions, and a reconstructed closed grain topology network that conforms to physical reality is obtained. S3. Based on the reconstructed closed grain topology network, the effective grain size, morphological anisotropy and texture orientation are quantitatively characterized and noise-resistantly rated in a multidimensional manner by equivalent ellipse fitting and area weighting algorithm. S4. Based on the reconstructed closed grain topology network, a local variation coefficient heatmap is generated by sliding window scanning to quantitatively evaluate the non-uniformity of the microstructure.

[0009] Furthermore, step S1 includes: acquiring metallographic microscopic data of the material to be tested; based on the principle in materials science that different types of grain boundaries have different interfacial energies, resulting in differences in chemical corrosion rates; using an adaptive contrast enhancement algorithm to nonlinearly map the gray-scale potential field of the image, thereby highlighting the deep trench corrosion characteristics of high-energy grain boundaries, while suppressing the non-structural noise of low-energy precipitates within the grains and the matrix background, and constructing a high signal-to-noise ratio physical feature field that reflects the true topological contour of the material.

[0010] Furthermore, the topology reconstruction model includes a martensitic phase transformation mechanism constraint; the martensitic phase transformation mechanism constraint includes: in the feature extraction stage, introducing a priori weights of the martensitic phase transformation shear mechanism, so that the model can automatically distinguish between the original austenitic grain boundaries and the internal subgrain boundaries based on the geometric morphology differences.

[0011] Furthermore, the topology reconstruction model includes flow field dynamics simulation; the flow field dynamics simulation includes: predicting the gradient flow field of grain space growth, and using Euler integral to simulate the dynamic process of pixel points converging towards the grain topology center to achieve instance segmentation.

[0012] Furthermore, the topology reconstruction model includes thermodynamic equilibrium repair; the thermodynamic equilibrium repair includes: for grain boundary breakpoints caused by corrosion discontinuity, based on the thermodynamic laws of interfacial tension balance and grain boundary energy minimization, constraining the grain boundary extension path so that it satisfies the angular balance condition of polycrystalline topology at the three-way node, thereby restoring a closed grain topology network that conforms to physical reality.

[0013] Furthermore, the multidimensional quantitative characterization includes: grain size distribution histogram, aspect ratio, and texture orientation rose diagram.

[0014] Furthermore, step S3 includes: Using the principle of second-order central moments of the image, each identified non-equiaxial grain is fitted to a mechanically equivalent ellipse; the major axis, minor axis and morphological orientation angle of each grain are extracted, and the aspect ratio is calculated to quantify the elongation and anisotropic characteristics of the grain. A frequency distribution histogram of the equivalent circle diameter is generated to reveal the dispersion and concentration trend of the material structure; an orientation rose diagram is generated based on the major axis angle of the fitted ellipse to intuitively characterize the preferred orientation of the microstructure caused by forging streamlines or rolling process. Using the area ratio of a single grain as the statistical weight, the effective average grain diameter of the entire field of view is calculated and converted into a standard rating.

[0015] Furthermore, step S4 includes: constructing a microstructure local difference evaluation model, locating potential defects inside the material, calculating the coefficient of variation and local average level of grain size in the local area through sliding window scanning technology, and generating a microstructure non-uniformity thermogram.

[0016] Furthermore, the dual-stream deep neural network is constructed based on a transfer learning strategy, freezing the general texture extraction layer and fine-tuning the high-level semantic layer.

[0017] Furthermore, based on the aforementioned spatial gradient flow field and grain boundary probability map, grain boundary repair is performed on grain boundary breaks caused by corrosion discontinuities, using grain boundary energy minimization and three-way node mechanical equilibrium as boundary conditions. This includes: Scan the identified grain boundary framework and locate all breakpoints; calculate the tangent direction tensor at the endpoints; Based on the physical law that grain boundaries tend to maintain the lowest interfacial energy, a curvature penalty term is introduced when searching for connection paths; candidate points with the smallest rate of change of curvature after connection are selected first to avoid forming sharp non-physical broken lines. At multiple grain boundary intersections, the three-way node balance rule of polycrystalline topology is executed; whether the angle of connection formation tends to a stable state is detected; and connection requests that deviate significantly from the mechanical equilibrium angle are rejected.

[0018] Compared with the prior art, the present invention has the following advantages: The intelligent identification and quantitative evaluation method for microstructures that integrates physical mechanisms proposed in this invention has the following significant advantages over existing technologies in terms of the accuracy of material physical property measurement and the scientific nature of the detection mechanism: 1. This invention restores the true grain topology that conforms to thermodynamic equilibrium, solving the problem of physical distortion in traditional optical detection. Existing technologies rely solely on optical contrast for geometric segmentation, often resulting in grain boundary fractures due to corrosion potential differences, leading to detection results that do not match the true material structure. This invention creatively introduces grain boundary energy minimization and the mechanical equilibrium of the three-way node as boundary conditions for topology reconstruction. This allows the detection system to automatically repair discontinuous grain boundaries according to thermodynamic laws, similar to the reverse process of grain growth. This ensures that the output grain topology network strictly follows the physical laws of polycrystalline materials, thereby achieving high-fidelity restoration of the original austenite grain boundary morphology and eliminating the interference of optical artifacts on material structure characterization.

[0019] 2. This invention establishes a multidimensional quantitative characterization and noise-resistant evaluation system based on reconstructed topology, breaking through the limitations of traditional single indicators and significantly improving the accuracy of mechanical property prediction. Addressing the deficiency of existing technologies in losing shape and orientation information, this invention introduces multidimensional descriptors such as grain size distribution histograms, aspect ratios (morphological anisotropy), and texture orientation rose diagrams, achieving a digital mapping of the entire microstructure of materials. To address the deficiency of traditional arithmetic mean methods in reflecting the true volume fraction of materials, an area-weighted stereoscopic statistical model is constructed. This model assigns higher weights to large grains occupying the dominant volume, conforming to the physical definition of the mean free path of dislocation slip in metallic materials.

[0020] The multidimensional physical parameters measured by this system can not only more accurately predict the yield strength and impact toughness of materials, but also reveal the anisotropy of mechanical properties caused by deformation texture. This makes the test results no longer just geometric statistics, but a key physical basis for directly characterizing the macroscopic mechanical property potential and process deformation history of materials.

[0021] 3. This invention achieves thermodynamic quantitative characterization of microstructure inhomogeneity, filling the gap in local failure risk detection. The fatigue life of high-performance metallic materials often depends on the weakest local region (such as a coarse-grained region). The original local variation coefficient thermographic technology of this invention is essentially a quantitative mapping of the thermodynamic inhomogeneity of the microstructure within the material. It can automatically capture and locate local coarse-grained and mixed-grained regions caused by uneven forging deformation or delayed phase transformation during heat treatment. This precise physical location of the worst field of view directly reveals the potential stress concentration points and fatigue crack initiation sources of components, significantly improving the service safety assessment level of high-end equipment in energy, aerospace, and other fields.

[0022] 4. This invention utilizes prior knowledge of physical metallurgy to suppress non-structural noise, achieving accurate analysis of complex phase transformation structures. For the complex phase transformation characteristics of martensite / bainite lath boundaries and proto-austenite grain boundaries coexisting in high-performance metallic materials, prior constraints based on phase transformation crystallography are incorporated into the model construction, suppressing signal interference from low-energy lath boundaries at the source of feature extraction. Without the need for complex chemical staining or electron microscopy, accurate separation and extraction of the target phase (proto-austenite) can be achieved solely through optical images, significantly reducing reliance on expensive testing equipment and improving the efficiency and versatility of on-site physical and chemical testing in industrial settings.

[0023] 5. This invention constructs a multi-dimensional anisotropic grain morphology descriptor, providing data feedback for optimizing the physical parameters of the forging process. By fitting the aspect ratio and orientation angle of the grains, it not only measures dimensions but also quantifies the plastic deformation history experienced by the material. These physical indicators can reflect the streamline distribution and deformation degree during the forging process, providing direct quantitative feedback for optimizing physical process parameters such as forging compression ratio and final forging temperature, thus promoting the transformation of metal material manufacturing from "experience-based trial and error" to "digital precision control". Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the technical route of the intelligent identification and quantitative evaluation method for microstructures that integrates physical mechanisms proposed in this invention. Figure 2 This is a schematic diagram of the original microstructure of the materials selected in the embodiments of the present invention; Figure 3 This is a schematic diagram illustrating the effect of grain boundary topological reconstruction in an embodiment of the present invention; Figure 4 This is a schematic diagram of the reconstructed complete grain boundary network in an embodiment of the present invention; Figure 5 This is a schematic diagram of grain size distribution in an embodiment of the present invention; Figure 6 This is a schematic diagram of the morphological orientation angle distribution in an embodiment of the present invention; Figure 7 This is a schematic diagram of the non-uniform thermodynamics of the tissue in an embodiment of the present invention. Detailed Implementation

[0025] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] See appendix Figure 1 The present invention proposes a method for intelligent identification and quantitative evaluation of microstructures that integrates physical mechanisms, which specifically includes the following steps: S1. Adaptive contrast enhancement of metallographic images is performed by utilizing the difference in grain boundary corrosion potential to construct a high signal-to-noise ratio physical feature field.

[0027] The specific process is as follows: Metallographic microscopy data of the material under test are acquired. Based on the principle in materials science that different types of grain boundaries have different interfacial energies, which leads to differences in chemical corrosion rates, an adaptive contrast enhancement algorithm is used to perform nonlinear mapping on the gray-scale potential field of the image. This highlights the deep trench corrosion characteristics of high-energy grain boundaries, while suppressing low-energy precipitates within the grains and unstructured noise from the matrix background, thus constructing a high signal-to-noise ratio physical feature field that reflects the true topological profile of the material.

[0028] S2. Construct a topological reconstruction model that integrates physical metallurgical mechanism constraints (martensitic phase transformation mechanism constraints, flow field dynamics simulation and thermodynamic equilibrium repair) to extract and repair grain boundaries in the microstructure and restore a closed grain network that conforms to physical reality.

[0029] This step abandons the traditional pure geometric threshold segmentation and instead adopts the following physical constraint strategy: Phase transformation mechanism constraint: In the feature extraction stage, a priori weights of the martensitic phase transformation shear mechanism are introduced, enabling the model to automatically distinguish between the original austenitic grain boundaries and the internal subgrain boundaries based on geometric morphological differences (such as straight laths and curved grain boundaries). Flow field dynamics simulation: Predict the gradient flow field of grain space growth, and use Euler integral to simulate the dynamic process of pixel points converging towards the grain topological center to achieve instance segmentation; Thermodynamic equilibrium repair: For grain boundary breaks caused by corrosion discontinuity, based on the thermodynamic laws of interfacial tension balance and grain boundary energy minimization, the grain boundary extension path is constrained so that it meets the angular balance condition of polycrystalline topology at the three-way node, thereby restoring a closed grain network that conforms to physical reality.

[0030] S3. Based on the reconstructed closed grain topology, a multidimensional quantitative statistical and characterization system is established.

[0031] This step aims to deeply mine the geometric morphological information of microscopic tissues and output a comprehensive dataset containing individual feature distributions and global statistical ratings: Grain equivalent fitting and anisotropy characterization: Utilizing the principle of the second-order central moment of the image, each identified non-equiaxial grain is fitted to a mechanically equivalent ellipse. The major axis, minor axis, and morphological orientation angle of each grain are extracted, and the aspect ratio is calculated to quantify the grain elongation and anisotropic characteristics.

[0032] Microscopic statistical distribution construction: Instead of outputting a single average value, it statistically analyzes the geometric parameters of all grains in the entire field of view to construct a multidimensional distribution map: it generates a frequency distribution histogram of the equivalent circle diameter (ECD) to reveal the dispersion and concentration trend of the material structure; it generates a direction rose diagram based on the major axis angle of the fitted ellipse to intuitively characterize the preferred orientation of the microstructure caused by forging streamlines or rolling processes.

[0033] Noise-resistant area-weighted rating: To address the issue that traditional arithmetic mean methods are easily lowered by minor noise in the sample preparation (such as corrosion pits and scratches), an area-weighted algorithm based on stereoscopic principles is adopted. The area proportion of a single grain is used as the statistical weight to calculate the effective average grain diameter across the entire field of view and convert it into a standard rating. This index can automatically filter out unstructured noise interference, ensuring that the rating results truly reflect the grain characteristics that occupy the dominant volume fraction of the material, and establishing a more accurate constitutive correlation with the material's macroscopic mechanical properties.

[0034] S4. Generate a local coefficient of variation heatmap by scanning a sliding window to quantitatively evaluate tissue heterogeneity: A microstructure local difference evaluation model is constructed to locate potential defects within the material. Using a sliding window scanning technique, the coefficient of variation and local average level of grain size within local regions are calculated, generating a thermographic map of microstructure inhomogeneity. This map uses color gradients to quantitatively characterize the differences in thermodynamic states within the material, visually indicating local mixed-grain and coarse-grain risk zones caused by uneven distribution of forging flow lines or delayed phase transformation during heat treatment. This enables automatic early warning of quality deficiencies in high-end equipment components.

[0035] The core principle of this invention lies in constructing a deep computational framework that integrates physical metallurgical constitutive relations, which fundamentally solves the problems of data discontinuity and missing physical features in microstructure characterization.

[0036] First, this invention establishes the physical basis for image feature enhancement and topology repair based on the electrochemical corrosion kinetics and thermodynamic equilibrium criteria of polycrystalline materials. In the microscopic imaging stage, the significant difference in corrosion potential between high-energy grain boundaries (such as proto-austenite grain boundaries) and low-energy substructures (such as lath boundaries) is utilized to map the chemical potential energy distribution of the microstructure into the grayscale gradient field of the optical image, thereby achieving signal enhancement of the target skeleton at the physical level. In the topology reconstruction stage, this invention explicitly introduces the Gibbs free energy minimization principle, treating grain boundaries as the geometric trajectories with the lowest total interface energy of the system, and forcing the three-way nodes to satisfy the interface tension vector balance condition. This strong prior constraint based on energy and mechanics enables the algorithm to automatically deduce and repair grain boundary fractures caused by incomplete corrosion according to physical laws, ensuring that the reconstructed grain network strictly follows the topological laws of polycrystalline materials.

[0037] Secondly, addressing the highly nonlinear characteristics of complex phase transformation structures, this invention innovatively designs a unique model architecture based on spatial gradient flow field and Euler dynamics integral. Unlike traditional convolutional neural networks that only statically classify pixels, this model transforms the grain segmentation problem into a particle convection evolution problem in fluid mechanics. The model learns the grain boundary morphology characteristics under the martensitic phase transformation shear mechanism to predict the normalized spatial gradient flow field pointing towards the thermodynamic center of each grain across the entire field of view. Subsequently, the Euler integral algorithm is used to simulate the inverse dynamic process of grain growth, driving pixels to converge towards the grain center along the flow field lines. This unique dynamic architecture possesses a natural ability to maintain topological continuity, enabling it to overcome local pixel gaps and achieve high-fidelity closure and instance segmentation of complex discontinuous grain boundaries. This overcomes the shortcomings of traditional watershed algorithms, such as sensitivity to noise and susceptibility to oversegmentation, from the underlying algorithmic architecture.

[0038] Finally, based on the aforementioned topological reconstruction of physical reality, this invention establishes a quantitative characterization system consistent with stereochemical principles. By introducing an area-weighted statistical mechanism, the mean free path effect of dislocation slip and grain boundary obstruction in polycrystalline materials is simulated, thereby assigning higher statistical weights to large grains occupying the dominant volume. This approach effectively filters out the interference of unstructured artifacts on the statistical results, achieving precise quantification of the effective grain size and microstructure inhomogeneity of the material, and ensuring a rigorous constitutive correlation between the microscopic characterization results and the macroscopic mechanical properties of the material.

[0039] The present invention will be further illustrated below through specific embodiments: This embodiment is applied to the material physicochemical testing process in the field of high-end heavy equipment manufacturing. The method described in this invention is integrated into an intelligent microscopic metallographic analysis system, which includes an optical microscopic imaging unit, a central processing unit (equipped with a high-performance GPU for parallel computing), and a data visualization terminal.

[0040] The metal material sample used in the experiment was a domestically produced 30Cr2Ni4MoV rotor steel forging provided by a heavy equipment manufacturing enterprise. The sample underwent quenching and tempering heat treatment, and its microstructure exhibited typical tempered martensite characteristics. Moreover, the corrosion depth at the original austenite grain boundaries was uneven, making it a complex sample that is extremely difficult to identify and grade using existing testing techniques.

[0041] The method of the present invention is used to process it, and the specific processing procedure is as follows: Step S1: Enhancement of microphysical features based on differences in grain boundary corrosion potential; During the material sample preparation stage, a specially formulated etching solution is used to etch the sample. Due to the high grain boundary energy of the original austenite grain boundaries, their corrosion rate is significantly faster than that of intragranular lath boundaries and the matrix, forming deeper corrosion trenches.

[0042] Data Acquisition: Sample images were acquired using a metallurgical microscope at a resolution of 2560×1920 pixels, with a physical resolution of 0.44. Original microscopic tissue images such as Figure 2 As shown.

[0043] Physical Contrast Enhancement: To address the issues of uneven illumination and shallow corrosion in the original image, the system employs a contrast-limited adaptive histogram equalization algorithm: setting the statistical grid to 8×8 and the shearing threshold to 2.0. This step utilizes digital image processing technology to nonlinearly stretch the grayscale gradient caused by corrosion potential differences, significantly enhancing the signal intensity of high-energy grain boundaries (dark trenches) while suppressing background noise from low-energy precipitates (light-colored spots) within the grains, thus constructing a physical feature field with a high signal-to-noise ratio.

[0044] Step S2: Grain boundary topology reconstruction incorporating physical metallurgical mechanism constraints; This step aims to reconstruct the true grain topology that conforms to the thermodynamic equilibrium state of the material.

[0045] Flow field dynamics prediction: The enhanced image is input into a pre-trained two-stream deep neural network, which is built based on a transfer learning strategy. The general texture extraction layer is frozen, and only the high-level semantic layers are fine-tuned. Specifically, a pre-trained U-net is used as the basic architecture of the two-stream network. The first three convolutional blocks are frozen (to retain the ability to extract general texture features), and only the last two convolutional blocks and the fully connected layers are fine-tuned to adapt to the grain analysis task.

[0046] Output I: Spatial gradient flow field. The system predicts a unit vector for each pixel pointing to the geometric center of its respective grain, simulating the dynamic trend of grain contraction from the boundary towards the center.

[0047] Output II: Grain boundary probability map. Based on the martensitic phase transformation shear mechanism, the network assigns lower probability weights to long, straight lath boundaries and higher probability weights to curved pre-austenite grain boundaries.

[0048] Thermodynamically Constrained Topology Repair (Core Steps): For grain boundary breaks caused by corrosion discontinuities, the system does not use simple geometric interpolation, but instead executes a repair algorithm based on minimizing grain boundary energy: Breakpoint detection and tensor calculation: Scan the identified grain boundary framework and locate all breakpoints. Calculate the tangent direction tensor at the endpoints.

[0049] Constraint I: Curvature Variational Constraint. Based on the physical law that grain boundaries tend to maintain the minimum surface area (i.e., the minimum interface energy), a curvature penalty term is introduced when the system searches for connection paths. Candidate points with the smallest rate of change of curvature after connection are preferentially selected to avoid forming non-physical sharp lines.

[0050] Constraint II: Tri-node Balance. At multiple grain boundary intersections, the system enforces the tri-node balance rule of polycrystalline topology. It checks whether the angle at which connections are formed tends towards a stable state. Connection requests that significantly deviate from the mechanical equilibrium angle are rejected.

[0051] Results: After the above physical constraint repair, the system outputs a closed, continuous grain boundary topology mask that conforms to the grain growth law. The original austenite grain boundary identification integrity rate reaches over 95%. Figure 3 and Figure 4 As shown.

[0052] Step S3: Quantitative statistics and characterization of multidimensional microstructures based on reconstructed topology; Based on the closed grain topology reconstructed in step S2, the system is no longer limited to a single average size calculation, but performs multidimensional quantitative statistics that include morphology, distribution and texture characteristics.

[0053] Grain equivalent fitting and anisotropy characterization: Using the principle of the second-order central moment of the image, each non-equiaxed proto-austenite grain identified within the field of view is fitted to a mechanically equivalent ellipse. The system automatically extracts the morphological feature parameters of each grain. Major axis length minor axis length

[0054] Morphological orientation angle : That is, the angle between the long axis of the grain and the horizontal reference axis (0°~180°).

[0055] Aspect Ratio = / ): Used to quantify the degree of grain elongation. In the rotor steel forging of this embodiment, the average aspect ratio is 1.8, which accurately reflects the plastic deformation characteristics of the material during the forging process.

[0056] Construction of micro-statistical distribution: The system statistically analyzes all effective grains (N>500) within the entire field of view and constructs a visual distribution map (e.g. Figure 5 (as shown) Size frequency distribution: The equivalent circle diameter of each grain was calculated, and a grain size histogram was generated. The results show that the grain size of the sample exhibits a positively skewed distribution, with the majority of the grains concentrated in the 30–50 μm range.

[0057] Orientation Rose Diagram: Based on Draw a rose diagram of grain orientation distribution. The diagram shows that the long axis of the grains is mainly concentrated in the 45°~90° direction, as shown below. Figure 6As shown, this is highly consistent with the forging flow line direction of the local part of the rotor forging, verifying the method's ability to analyze deformed grains.

[0058] Noise-resistant area-weighted rating: To address the issue that traditional image-based methods are susceptible to inflated ratings due to tiny corrosion pits (non-structural noise) during sample preparation, the system employs an area-weighted algorithm to calculate the effective average grain size across the entire field of view.

[0059] The calculation formula is as follows:

[0060] in, Let be the area of ​​the i-th grain. It is the diameter of its equivalent circle.

[0061] Step S4: Evaluation of the local mapping of microstructure thermodynamic inhomogeneity; To detect potential defects inside the material, the system performs microstructure uniformity analysis.

[0062] Sliding window scanning: Define a statistical kernel with a size of 256×256 pixels and a step size of 64 pixels to perform step scanning in the entire field of view.

[0063] Quantification of thermodynamic state differences: At each window location, the coefficient of variation (CV = standard deviation / mean) of the internal grain size is calculated. This index physically corresponds to the differences in the thermodynamic state of the local region.

[0064] Defect Visualization and Early Warning: The system uses a color mapping algorithm to convert the CV value matrix into a color heatmap and overlay it onto the original image. The heatmap shows several significantly bright areas (CV value > 0.6) in the image. Upon review, these areas exhibit coarse grains or mixed crystal phenomena, classifying them as typical risk areas. Figure 7 As shown.

[0065] Output report: The system automatically generates a comprehensive quality evaluation report that includes global effective grain size and local coarse grain risk warning (coordinates X, Y).

[0066] This embodiment demonstrates that by integrating physical metallurgical mechanism constraints, the challenges of identifying and rating the complex microstructure of high-performance metallic materials can be effectively overcome. The measured results not only outperform traditional image processing methods in statistical accuracy but also accurately locate potential fatigue failure sources through non-uniform thermal maps, providing a reliable physical testing method for the quality control of high-end equipment materials.

[0067] All matters not covered in this invention are common knowledge.

[0068] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for intelligent identification and quantitative evaluation of microstructures that integrates physical mechanisms, characterized in that, The method includes the following steps: S1. Adaptive contrast enhancement of metallographic images using grain boundary corrosion potential differences; S2. Construct a topology reconstruction model; the topology reconstruction model is based on a pre-trained dual-flow deep neural network, takes the enhanced metallographic image as input, and outputs a spatial gradient flow field and a grain boundary probability map; based on the spatial gradient flow field and the grain boundary probability map, grain boundary repair is performed on the grain boundary breaks caused by corrosion discontinuity with the grain boundary energy minimization and the mechanical balance of the three-way node as boundary conditions, and a reconstructed closed grain topology network that conforms to physical reality is obtained. S3. Based on the reconstructed closed grain topology network, the effective grain size, morphological anisotropy and texture orientation are quantitatively characterized and noise-resistantly rated in a multidimensional manner by equivalent ellipse fitting and area weighting algorithm. S4. Based on the reconstructed closed grain topology network, a local variation coefficient heatmap is generated by sliding window scanning to quantitatively evaluate the non-uniformity of the microstructure. Based on the aforementioned spatial gradient flow field and grain boundary probability map, grain boundary repair is performed on grain boundary breaks caused by corrosion discontinuities, using grain boundary energy minimization and three-way node mechanical equilibrium as boundary conditions. This includes: Scan the identified grain boundary framework and locate all breakpoints; calculate the tangent direction tensor at the breakpoints. Based on the physical law that grain boundaries tend to maintain the lowest interfacial energy, a curvature penalty term is introduced when searching for connection paths; candidate points with the smallest rate of change of curvature after connection are selected first to avoid forming sharp non-physical broken lines. At multiple grain boundary intersections, the three-way node balance rule of polycrystalline topology is executed; whether the angle of connection formation tends to a stable state is detected; and connection requests that deviate significantly from the mechanical equilibrium angle are rejected.

2. The method for intelligent identification and quantitative evaluation of microstructures integrating physical mechanisms according to claim 1, characterized in that, Step S1 includes: acquiring metallographic microscopic data of the material to be tested; based on the principle in materials science that different types of grain boundaries have different interfacial energies, resulting in differences in chemical corrosion rates; using an adaptive contrast enhancement algorithm to nonlinearly map the gray-scale potential field of the image, thereby highlighting the deep trench corrosion characteristics of high-energy grain boundaries, while suppressing the low-energy precipitates within the grains and the unstructured noise of the matrix background, and constructing a high signal-to-noise ratio physical feature field that reflects the true topological contour of the material.

3. The method for intelligent identification and quantitative evaluation of microstructures integrating physical mechanisms according to claim 1, characterized in that, The topology reconstruction model includes a martensitic phase transformation mechanism constraint; the martensitic phase transformation mechanism constraint includes: in the feature extraction stage, introducing a priori weights of the martensitic phase transformation shear mechanism, so that the model can automatically distinguish between the original austenitic grain boundaries and the internal subgrain boundaries based on the differences in geometric morphology.

4. The method for intelligent identification and quantitative evaluation of microstructures integrating physical mechanisms according to claim 3, characterized in that, The topology reconstruction model includes flow field dynamics simulation; the flow field dynamics simulation includes: predicting the gradient flow field of grain space growth, and using Euler integral to simulate the dynamic process of pixels converging towards the grain topology center to achieve instance segmentation.

5. The method for intelligent identification and quantitative evaluation of microstructures integrating physical mechanisms according to claim 4, characterized in that, The topology reconstruction model includes thermodynamic equilibrium repair; the thermodynamic equilibrium repair includes: for grain boundary breakpoints caused by corrosion discontinuity, based on the thermodynamic laws of interfacial tension balance and grain boundary energy minimization, constraining the grain boundary extension path so that it satisfies the angular balance condition of polycrystalline topology at the three-way node, thereby restoring a closed grain topology network that conforms to physical reality.

6. The method for intelligent identification and quantitative evaluation of microstructures integrating physical mechanisms according to claim 1, characterized in that, The multidimensional quantitative characterization includes: grain size distribution histogram, aspect ratio, and texture orientation rose diagram.

7. The method for intelligent identification and quantitative evaluation of microstructures integrating physical mechanisms according to claim 6, characterized in that, Step S3 includes: Using the principle of second-order central moments of the image, each identified non-equiaxial grain is fitted to a mechanically equivalent ellipse; the major axis, minor axis and morphological orientation angle of each grain are extracted, and the aspect ratio is calculated to quantify the elongation and anisotropic characteristics of the grain. A frequency distribution histogram of the equivalent circle diameter is generated to reveal the dispersion and concentration trend of the material structure; an orientation rose diagram is generated based on the major axis angle of the fitted ellipse to intuitively characterize the preferred orientation of the microstructure caused by forging streamlines or rolling process. Using the area ratio of a single grain as the statistical weight, the effective average grain diameter of the entire field of view is calculated and converted into a standard rating.

8. The method for intelligent identification and quantitative evaluation of microstructures integrating physical mechanisms according to claim 1, characterized in that, Step S4 includes: constructing a microstructure local difference evaluation model, locating potential defects inside the material, calculating the coefficient of variation and local average level of grain size in the local area through sliding window scanning technology, and generating a thermal map of microstructure non-uniformity.

9. The method for intelligent identification and quantitative evaluation of microstructures integrating physical mechanisms according to claim 1, characterized in that, The dual-stream deep neural network is constructed based on a transfer learning strategy, freezing the general texture extraction layer and fine-tuning the high-level semantic layer.

Images