Crystal grain size ultrasonic evaluation method based on phase space recursion characteristics
Through the ultrasonic evaluation method based on phase space recursive features, the losslessness and signal processing complexity of metal material grain size detection are solved, and high-precision grain size evaluation is achieved.
Patent Information
- Application Number
- CN202510160429.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-13
- Publication Date
- 2025-06-17
AI Technical Summary
The existing metal material grain size detection methods are mostly lossy metallographic methods, which cannot realize non-destructive detection. In addition, ultrasonic detection technology has complex signal processing problems in characteristic evaluation.
The grain size ultrasonic evaluation method based on phase space recursive features is adopted. By extracting the backscattered signal in the ultrasonic A-scan signal, using the mutual information method and FNN for signal optimization, the optimal embedding dimension and optimal delay time are obtained, and the phase space reconstruction is carried out, and the evaluation model is constructed in combination with the PSO-XGBoost algorithm.
Non-destructive detection of the grain size of metal materials is realized, and a high-precision nonlinear model is established through phase space characteristic characterization, which improves the accuracy of grain size evaluation.
Smart Images
Figure CN120161119A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of non-destructive characterization of the microstructure of metal materials, and particularly to an ultrasonic evaluation method for grain size based on phase space recurrence features. Background Art
[0002] Most of the metal materials commonly used in industry are polycrystalline structure materials, and their mechanical properties are affected by the grain size. At present, there are many methods for detecting and evaluating the grain size of materials. Among them, the metallographic method is the most accurate and reliable method. However, it belongs to a destructive detection method, which destroys the original structure and shape of the workpiece. Ultrasonic detection technology has the advantages of fast speed, high precision, wide application range, etc., and has gradually become the main research method for non-destructive evaluation of the characteristics of polycrystalline metal materials. This detection method has high sensitivity to the grain size of materials and has great advantages and potential in tissue evaluation.
[0003] The ultrasonic signal changes as it propagates through different material tissues over time of flight. During this process, the ultrasonic signal usually contains rich non-linear dynamic characteristics, such as periodicity, chaos, and complex non-linear behavior.
[0004] Therefore, there is an urgent need to develop an ultrasonic evaluation method for grain size based on phase space recurrence features. Summary of the Invention
[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is:
[0006] An ultrasonic evaluation method for grain size based on phase space recurrence features, comprising the following steps:
[0007] Step 1: Extract the ultrasonic A-scan signal of the metal sample to be measured;
[0008] Step 2: Extract the backscattered signal by weighting the signal between the interface wave and the bottom wave of the A-scan signal with a rectangular window;
[0009] Step 3: Use the mutual information method and FNN to perform optimization calculations on the backscattered signal to obtain the optimal embedding dimension and the optimal delay time;
[0010] Step 4: Reconstruct the phase space with the optimal embedding dimension, the optimal delay time and the backscattered signal, and draw a recurrence plot through the phase space and threshold calculation;
[0011] Step 5: Perform statistical analysis according to the recurrence plot to obtain phase space eigenvalue such as recurrence rate and capture time;
[0012] Step 6: Use the PSO-XGBoost algorithm to construct an evaluation model with the phase space features as the input and the grain size as the output.
[0013] Preferably, the surface of the metal sample in step 1 needs to be polished to at least a surface roughness Ra of 6.3 μm.
[0014] Preferably, in step 2, the rectangular window is located between the interface wave and the bottom wave, and its function expression is as follows:
[0015]
[0016] where Rect represents the rectangular window function, n represents the total length of the window function, M represents the effective length of the window function.
[0017] Preferably, the role of the mutual information method in step 3 is to obtain the optimal delay time, and the first minimum value obtained during reconstruction is the optimal delay time.
[0018] Preferably, the position of the rectangular window selected in step 2 should avoid the interface wave and the primary bottom wave pulse width region.
[0019] Preferably, the role of the FNN in step 3 is to obtain the optimal embedding dimension. When the embedding dimension increases, the number of false neighboring points will continuously decrease until the number of false neighboring points no longer changes, and then the optimal embedding dimension can be determined.
[0020] Preferably, in step 4, the optimal embedding dimension, the optimal delay time, and the backscattering signal are used for phase space reconstruction.
[0021] Preferably, the recurrence plot in step 5 is obtained from the phase space reconstruction, and statistical analysis is performed on it to obtain phase space eigenvalue such as the recurrence rate and the capture time.
[0022] Preferably, in step 6, the sampling PSO is used to optimize the hyperparameters in XGBoost, and the objective function is defined by minimizing the MSE value of the validation set. The hyperparameters to be optimized are (eta, max_depth, min_child_weight, subsample, colsample_bytree, alpha).
[0023] Preferably, for the PSO-XGBoost algorithm in step 6, its input is the phase space features and the output is the grain size.
[0024] Preferably, the ultrasonic evaluation method for grain size based on phase space recurrence features further includes step 7, and step 7 is: according to the optimal hyperparameters of the XGBoost algorithm, a grain size ultrasonic evaluation model based on phase space features is constructed.
[0025] Due to the adoption of the above technical solutions, the technical progress achieved by the present invention compared with the prior art is:
[0026] 1) The phase space eigenvalues are extracted from the backscattered signals in ultrasonic testing of metal materials for the first time in this method.
[0027] 2) The grain size is characterized by the phase space features in this method, showing another analysis angle compared with the traditional time-frequency feature characterization method.
[0028] 3) The PSO-XGBoost algorithm is used to fit the phase space eigenvalues and the grain size in this method, and a non-linear model for ultrasonic evaluation of grain size based on phase space features is established, with higher evaluation accuracy compared with the linear model. BRIEF DESCRIPTION OF THE DRAWINGS
[0029] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on the structures shown in these drawings without creative efforts.
[0030] Figure 1 Flow chart of the ultrasonic evaluation method for grain size based on phase space recursive features provided by the present invention;
[0031] Figure 2 Ultrasonic A-scan signal diagram;
[0032] Figure 3 Ultrasonic backscattered signal extracted by rectangular window weighting;
[0033] Figure 4 Optimization calculation diagram for the optimal delay time;
[0034] Figure 5 Optimization calculation diagram for the optimal embedding dimension;
[0035] Figure 6 Recurrence plot;
[0036] Figure 7 Iteration diagram for PSO to optimize the hyperparameters of XGBoost;
[0037] Figure 8 Training result of the PSO-XGBoost algorithm;
[0038] Figure 9 Test result of the PSO-XGBoost algorithm. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0039] The following embodiments can help those skilled in the art to more comprehensively understand the present invention, but cannot limit the present invention in any way.
[0040] Embodiment
[0041] An ultrasonic evaluation method for grain size based on phase space recursive features. Step 1: Perform ultrasonic testing on a metal material with its surface polished to Ra 6.3 μm, and extract the ultrasonic A-scan signal of the metal sample to be measured. This A-scan signal should at least include the interface wave and the first bottom wave of the workpiece, as Figure 2 shown;
[0042] Step 2: Extract the backscattered signal by weighting the signal between the interface wave and the bottom wave of the A-scan signal with a rectangular window, as Figure 3 shown;
[0043] Among them, the position of this rectangular window is between the interface wave and the first bottom wave, and does not include the interface wave and the pulse width range of the first bottom wave. The width of the rectangular window is [V1 + 150, V2 - 60], where V1 is the number of sampling points corresponding to the peak of the interface wave, and V2 is the number of sampling points corresponding to the peak of the first bottom wave.
[0044] Step 3: Use the mutual information method and FNN to perform optimization calculations on the backscattered signal to obtain the optimal embedding dimension and the optimal delay time;
[0045] In the mutual information method, multiple different delayed versions of copies are generated from the time series data. The delay time corresponding to the first minimum value of the mutual information entropy is the optimal delay time, as Figure 4 shown.
[0046] In FNN, when the embedding dimension increases, the number of false neighboring points will continuously decrease. Until the number of false neighboring points no longer changes, the optimal embedding dimension can be determined, that is, the first embedding dimension when the virtual zero ratio changes relatively little, as Figure 5 shown.
[0047] Step 4: Reconstruct the phase space with the optimal embedding dimension, the optimal delay time and the backscattered signal, and draw a recurrence plot through the phase space and threshold calculation, as Figure 6 shown;
[0048] Step 5: Conduct statistical analysis based on the recurrence plot to obtain phase space eigenvalue such as recurrence rate and capture time;
[0049] Step 6: Use the PSO algorithm to perform optimization calculations on the hyperparameters of the XGBoost algorithm;
[0050] The total number of samples in the experimental dataset is 33. The training set, validation set, and test set are randomly divided according to the ratio of 8:1:1. Among them, the number of samples in the training set is 26, the number of samples in the validation set is 3, and the number of samples in the test set is 3. The results of the validation set are used for parameter optimization, and the test set is used for model evaluation.
[0051] Optimize the five hyperparameters of eta, max_depth, min_child_weight, subsample, colsample_bytree, and alpha in the XGBoost algorithm. With minimizing the MSE value of the validation set as the objective function and 500 generations of iteration, calculate the optimal hyperparameters of the XGBoost algorithm. The optimization process is as Figure 7 shown;
[0052] In an embodiment of the present invention, the ultrasonic evaluation method for grain size based on phase space recursion features further includes step 7, and step 7 is: construct an ultrasonic evaluation model for grain size based on phase space features according to the optimal hyperparameters, where the training set is as Figure 8 shown, and the test set is as Figure 9 shown.
[0053] The evaluation result is as Figure 9 shown.
[0054] In an embodiment of the present invention, the surface of the metal sample in step 1 needs to be polished at least to a surface roughness Ra of 6.3 μm.
[0055] In an embodiment of the present invention, in step 2, the rectangular window is located between the interface wave and the bottom wave, and its function expression is as follows:
[0056]
[0057] Among them, Rect represents the rectangular window function, n represents the total length of the window function, and M represents the effective length of the window function.
[0058] In an embodiment of the present invention, the role of the mutual information method in step 3 is to obtain the optimal delay time, and the first minimum value obtained during reconstruction is the optimal delay time.
[0059] In an embodiment of the present invention, the position of the rectangular window selected in step 2 should avoid the interface wave and the primary bottom wave pulse width region.
[0060] In an embodiment of the present invention, the metal material in step 1 is a polycrystalline structure material.
[0061] In an embodiment of the present invention, the role of the FNN in step 3 is to obtain the optimal embedding dimension. When the embedding dimension increases, the number of false neighboring points will continuously decrease until the number of false neighboring points no longer changes, and then the optimal embedding dimension can be determined.
[0062] In an embodiment of the present invention, in step 4, perform phase space reconstruction on the optimal embedding dimension, the optimal delay time, and the backscattering signal.
[0063] In an embodiment of the present invention, in step 5, the recurrence plot is obtained by phase space reconstruction, and statistical analysis is performed on it to obtain phase space eigenvalue such as recurrence rate and capture time.
[0064] In an embodiment of the present invention, in step 6, the sampling PSO is used to optimize the hyperparameters in XGBoost, and the objective function is defined by minimizing the MSE value of the validation set, where the optimized hyperparameters are (eta, max_depth, min_child_weight, subsample, colsample_bytree, alpha).
[0065] The present invention has been generally and exhaustively described above. However, based on the present invention, some modifications or improvements can be made, which are obvious to those of ordinary skill in the technical field. Therefore, any modifications or improvements made without departing from the spirit and idea of the present invention fall within the protection scope of the present invention.
Claims
1. A method for ultrasonic evaluation of grain size based on phase space recursive characteristics, characterized in that: The following steps are involved: Step 1: Extract the ultrasonic A-scan signal of the metal sample to be tested; Step 2: Extract the backscattered signal by performing rectangular window weighting on the signal between the interface wave and the bottom wave of the A-scan signal; Step 3: Use the mutual information method and FNN to optimize the backscattered signal and obtain the optimal embedding dimension and the optimal delay time; Step 4: Reconstruct the phase space of the optimal embedding dimension, the optimal delay time and the backscattering signal, and draw a recursive graph through phase space and threshold calculation; Step 5: Perform statistical analysis based on the recurrence diagram to obtain phase space eigenvalues such as recurrence rate and capture time; Step 6: Use the PSO-XGBoost algorithm to build an evaluation model with phase space characteristics as input and grain size as output.
2. The method for ultrasonic evaluation of grain size based on phase space recursive features according to claim 1, characterized in that: In step 2, the rectangular window is located between the interface wave and the bottom wave, and its function expression is as follows: Wherein, Rect represents a rectangular window function, n represents the total length of the window function, and M represents the effective length of the window function.
3. The method for ultrasonic evaluation of grain size based on phase space recursive features according to claim 1, characterized in that: The mutual information method in step 3 is used to obtain the optimal delay time, wherein the first minimum value obtained during reconstruction is the optimal delay time.
4. The method for ultrasonic evaluation of grain size based on phase space recursive features according to claim 1, characterized in that: The role of the FNN described in step 3 is to obtain the optimal embedding dimension. When the embedding dimension increases, the number of false neighboring points will continue to decrease until the number of false neighboring points no longer changes, and the optimal embedding dimension can be determined.
5. The method for ultrasonic evaluation of grain size based on phase space recursive features according to claim 1, characterized in that: In step 4, the optimal embedding dimension, the optimal delay time and the backscattering signal are reconstructed in phase space.
6. The method for ultrasonic evaluation of grain size based on phase space recursive features according to claim 1, characterized in that: In step 5, the recursion graph is obtained by reconstructing the phase space, and statistical analysis is performed on it to obtain the phase space characteristic values such as the recursion rate and capture time.
7. The method for ultrasonic evaluation of grain size based on phase space recursive features according to claim 1, characterized in that: The PSO-XGBoost algorithm in step 6 defines the objective function by minimizing the MSE value of the validation set, and uses PSO to optimize the parameters in XGBoot (eta, max_depth, min_child_weight, subsample, colsample_bytree, alpha).
8. The method for ultrasonic evaluation of grain size based on phase space recursive features according to claim 1, characterized in that: The PSO-XGBoost algorithm in step 6 has phase space features as input and grain size as output.
9. The method for ultrasonic evaluation of grain size based on phase space recursive features according to claim 1, characterized in that: The surface of the metal sample in step 1 needs to be polished to at least a surface roughness of Ra 6.3 μm.
10. The method for ultrasonic evaluation of grain size based on phase space recursive features according to any one of claims 1 to 9, characterized in that: The method for ultrasonic evaluation of grain size based on phase space recursive features also includes step 7, which is: constructing a grain size ultrasonic evaluation model based on phase space features according to the optimal hyperparameters of the XGBoost algorithm.