Friction deposition forming defect detection method based on three-dimensional force time-frequency domain characteristics

Abstract

The invention discloses a friction deposition forming defect detection method based on three-dimensional force time-frequency domain characteristics, which comprises the following steps: carrying out defect detection on a sample by adopting a trained defect detection model to obtain a prediction result, and training the defect detection model by adopting samples in a data set; the method for acquiring the data set comprises the following steps: acquiring an original three-dimensional force signal; obtaining a decomposed three-dimensional force signal and preprocessing the decomposed three-dimensional force signal; dividing the three-dimensional force signal in the stable deposition stage to obtain a plurality of sample units; calculating the characteristics of each signal curve in each sample unit; screening strong correlation features and weak correlation features; screening the strong correlation features to obtain N3 strong correlation features; obtaining a final feature set H according to the information gains of the N3 strong correlation features and the weak correlation features; and constructing a data set. According to the friction deposition forming defect detection method, dimension reduction is carried out on the feature number of the force signals, redundant information is removed, and the relation between the features and the defects and the accuracy of the prediction result are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of additive manufacturing quality monitoring, and specifically relates to a friction deposition forming defect detection method based on three-dimensional force time-frequency domain characteristics. Background Art

[0002] In recent years, the Friction Deposition Forming (FDF) process has attracted much attention as a new solid-state additive manufacturing method. During the Friction Deposition Forming (FDF) process, the deposited material undergoes plastic deformation under the action of extrusion friction, forming a metallurgical connection with the substrate or the previous deposited layer. Since the deposition process does not involve the melting and solidification of the metal, metallurgical defects such as pores and cracks can be effectively avoided. However, if the parameters of the Friction Deposition Forming process are not properly selected, it will still lead to internal defects such as weak bonding and grooves, as well as surface defects, thereby affecting the quality of the deposited parts. At present, the commonly used non-destructive testing methods mainly include: Ultrasonic Testing (UT), X-ray Computed Tomography (X-CT) methods and real-time monitoring methods based on sensor signals (such as temperature signal and mechanical signal monitoring).

[0003] Among them, ultrasonic testing can effectively identify holes and cracks inside deposited parts, and has the advantages of being non-destructive and convenient. However, due to the high requirements of ultrasonic testing on the material surface and the difficulty in detecting deposited parts with complex geometric shapes, it is difficult to use for friction deposition defect identification.

[0004] X-ray computed tomography (CT) can accurately display the three-dimensional defect distribution inside deposited parts and provide high spatial resolution. However, the X-ray CT method is expensive and the scanning and data analysis are time-consuming, making it difficult to achieve online real-time detection.

[0005] Among real-time process monitoring methods, mechanical signal monitoring is the most widely used. Force signals contain more abundant information about material flow and forming, and can be measured and acquired in real time online. In the field of friction stir welding (FSW), researchers have applied force signals to defect type detection. FSW is a solid-state joining technology that achieves material connection by generating plastic flow through frictional heat between a high-speed rotating stir head and the workpiece. Compared with the FSW process, FDF not only involves solid-state joining of the material interface, but also includes the plasticization of the bar, material flow, and layer-by-layer deposition and forming processes. This makes the force signal in the FDF process more complex, dynamic, and has obvious stage characteristics. To date, the dimension of the force signal measured by FDF is relatively single and cannot fully reflect the information contained in the deposition process. In the FDF process, the number of force signal features is relatively large. Therefore, during the analysis process, the number of features needs to be reduced to eliminate redundant features and improve analysis efficiency.

[0006] Currently existing dimensionality reduction methods, such as principal component analysis (PCA), are unsupervised linear dimensionality reduction methods that fail to fully consider the relationship between each feature and the classification target, and will retain some information irrelevant to the discrimination task. The principal components obtained after dimensionality reduction are linear combinations of the original features, lack intuitive physical explanations, and have relatively vague physical meanings. Summary of the Invention

[0007] In view of the deficiencies in the prior art, an object of the present invention is to provide a defect detection model.

[0008] Another object of the present invention is to provide a method for detecting defects in friction deposition forming based on three-dimensional force time-frequency domain characteristics. The method can predict whether a deposited part has defects and classify defects in the deposited part.

[0009] Another object of the present invention is to provide a use of the above-mentioned defect detection model for real-time defect detection in friction deposition forming.

[0010] The present invention is achieved through the following technical solutions.

[0011] A defect detection model, wherein each input sample includes: a value of a feature in a final feature set H corresponding to a sample unit; a method for obtaining the final feature set H includes:

[0012] S1, prepare the original three-dimensional force signal, which is the signal of three mutually perpendicular forces between the workpiece and the spindle in the same time period during the friction deposition forming process;

[0013] S2, performing variational modal decomposition on the original three-dimensional force signal to obtain three intrinsic mode function sets, each of which includes K intrinsic mode functions; in each intrinsic mode function set, selecting the eigenmode function with the lowest center frequency band as the selected eigenmode function, combining the three selected eigenmode functions and using them as the decomposed three-dimensional force signal, and performing Gaussian smoothing to obtain three signal curves as the pre-processed three-dimensional force signal;

[0014] S3, dividing the three-dimensional force signal according to a fixed window size to obtain multiple sample units, and the multiple sample units constitute a sample set D; calculating the features of each signal curve in each sample unit in the sample set D, and obtaining M features for each sample unit;

[0015] S4, taking each type of feature of all sample units in the sample set D as a variable, and calculating the correlation coefficient between any two variables; setting a correlation threshold, if the correlation coefficient of a certain variable with the remaining M-1 variables is less than or equal to the correlation threshold, then the type of feature corresponding to the variable is regarded as a weakly correlated feature; otherwise, it is regarded as a strongly correlated feature, and N strongly correlated features and N2 weakly correlated features are obtained;

[0016] Hierarchical clustering is used to divide N1 strongly correlated features into multiple clusters; a greedy algorithm is used to screen the strongly correlated features in each cluster, and after screening, a total of N3 strongly correlated features are obtained from all clusters; the N3 strongly correlated features obtained after screening and the N2 weakly correlated features form a feature set;

[0017] The information gain of each feature in the feature set is calculated, and the features with information gain values ​​greater than or equal to 0.01 in the feature set are selected to form the final feature set H.

[0018] In the above technical solution, the characteristics of each signal curve include: maximum value (max), minimum value (min), mean value (mean), median value (med), peak-to-peak value (ran), average absolute value (arv), standard deviation (std), kurtosis (Kur), skewness (Ske), root mean square (Rms), root mean square amplitude (Srma), form factor (Wf), peak factor (Pf), pulse factor (If), margin factor (Cf), center of gravity frequency (FC), root mean square frequency (Rmsf) and frequency standard deviation (Rvf) One or more.

[0019] In the above technical solution, in S4, the input of the hierarchical clustering is the correlation coefficients of N1 strongly correlated features that are greater than the correlation threshold.

[0020] In the above technical solution, in S4, the input of the greedy algorithm is the strongly correlated features in the screened cluster and the correlation coefficient corresponding to each strongly correlated feature in the cluster that is greater than the correlation threshold, and the output of the greedy algorithm is at least one strongly correlated feature that is representative of the cluster.

[0021] In the above technical solution, the greedy algorithm is specifically as follows: traverse each cluster, and each cluster is iterated according to the following loop steps until the correlation coefficient in the coefficient set of the cluster is 0, and all the selected strongly correlated features in the iterative process are combined into a strongly correlated feature representative set, and the strongly correlated feature representative set obtained by all clusters is the N3 strongly correlated features; wherein, the loop step: the correlation coefficients of the strongly correlated features in the cluster are combined into a coefficient set, and each time the strongly correlated feature with the largest number of correlation coefficients in the coefficient set is selected as the selected strongly correlated feature, and the correlation coefficient corresponding to the selected strongly correlated feature is deleted from the coefficient set and the selected strongly correlated feature is deleted from the cluster.

[0022] In the above technical solution, the defect detection model is constructed using a random forest algorithm.

[0023] In the above technical solution, a five-fold cross validation method is used to train the defect detection model, and a grid search method is used to optimize the parameters of the defect detection model.

[0024] In the above technical solution, the output of the defect detection model is a prediction result. When the defect detection model is used for binary classification, the prediction result is non-defective or defective, and the true value of the sample is that the deposited part corresponding to the sample unit is non-defective or defective.

[0025] When the defect detection model is used for three-classification, the prediction result is good forming, slight defect or severe defect, and the true value of the sample is that the deposited part corresponding to the sample unit has no defect, the deposited part has slight defect or the deposited part has severe defect.

[0026] A method for detecting defects in friction deposition forming based on three-dimensional force time-frequency domain characteristics uses the above-mentioned trained defect detection model to perform defect detection on a sample to be tested and obtains a prediction result corresponding to the sample to be tested.

[0027] The above-mentioned defect detection model is used for real-time defect detection in friction deposition forming.

[0028] The present invention has the following advantages due to the adoption of the above technical solution:

[0029] 1. The defect detection model constructed in the present invention is based on the original three-dimensional force signal of the deposited part during the friction deposition forming process, which realizes the real-time monitoring and evaluation of the forming quality of the deposited part during the friction deposition process. It not only improves the intelligence level of the friction deposition process, but also provides solid theoretical and technical support for online parameter adjustment and defect control.

[0030] 2. The friction deposition forming defect detection method of the present invention reduces the number of features of the force signal to ensure that the 21 types of features after dimensionality reduction have clear physical meanings and effectively remove redundant information, thereby ensuring the full retention of discriminant information, further improving the connection between features and defects, providing a basis for the prediction of the defect detection model, and improving the accuracy of the prediction results of the defect detection model. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 This is a flow chart of Example 1;

[0032] Figure 2 A schematic diagram of friction deposition forming of a deposited part;

[0033] Figure 3 The result diagram of the original three-dimensional force signal and the decomposed three-dimensional force signal;

[0034] Figure 4 The clustering tree result diagram output by hierarchical clustering;

[0035] Figure 5 The information gain value result graph of the feature;

[0036] Figure 6 is the prediction result of the present invention;

[0037] Figure 7 The confusion matrix and ROC curve corresponding to the prediction results of Example 6. DETAILED DESCRIPTION

[0038] The friction deposition forming defect detection method based on three-dimensional force time-frequency domain characteristics of the present invention is described in detail below with reference to the accompanying drawings.

[0039] In the following embodiments, the definition of a slight defect in a deposited part is: in a cross section of the deposited layer along a direction perpendicular to the deposited part, the number of pixels in the defective portion of the cross section of the deposited layer accounts for less than 20% of the total number of pixels in the cross section of the deposited layer; the definition of a serious defect in a deposited part is: in a cross section of the deposited layer along a direction perpendicular to the deposited part, the number of pixels in the defective portion of the cross section of the deposited layer accounts for more than 20% of the total number of pixels in the cross section of the deposited layer.

[0040] Among the four three-way force sensors, each three-way force sensor collects a three-directional force signal between the workpiece and the spindle, and the three-directional force signals collected by the four three-way force sensors are superimposed to obtain an original three-dimensional force signal.

[0041] Example 1

[0042] like Figure 1 As shown, a method for obtaining a data set includes:

[0043] S1, four three-force sensors are respectively installed at the four corners of the bottom of the friction deposition forming tooling table (the top surface of the tooling table is fixed with a workpiece, and friction deposition forming is used to form a deposition layer on the workpiece), so that the spindle contacts the workpiece, and the spindle performs friction deposition forming on the workpiece to form a deposition layer to obtain a deposition part, such as Figure 2 As shown in the figure, during the friction deposition forming process, four three-dimensional force sensors were used to collect the original three-dimensional force signals between the workpiece and the spindle (the acquisition frequency was set to 4000 Hz) within the same time period. The original three-dimensional force signals represent the three mutually perpendicular forces between the workpiece and the spindle. The original three-dimensional force signals include an axial original force signal, a horizontal original force signal, and a lateral original force signal. That is, the forces in the axial original force signal, the horizontal original force signal, and the lateral original force signal are mutually perpendicular. The axial original force signal represents the force along the spindle.

[0044] The original three-dimensional force signal is Figure 3 As shown in (a), Figure 3 In (a), Z-Force is the axial original force signal, X-Force is the horizontal original force signal, and Y-Force is the lateral original force signal. Figure 3 In (a), the horizontal axis is the acquisition time, the vertical axis is the force value, and the unit of the vertical axis is kN. The force values ​​of the horizontal original force signal and the lateral original force signal at different acquisition times correspond to Figure 3 The left vertical axis of (a) shows the force values ​​of the axial original force signal at different acquisition times. Figure 3 The right vertical axis of (a); the axial original force signal, the horizontal original force signal and the lateral original force signal show periodic sinusoidal wave fluctuations, and there are obvious high-frequency mutation noise signals and clutter.

[0045] S2, including S2-1, S2-2 and S2-3:

[0046] S2-1, perform variational modal decomposition (VMD) on the original three-dimensional force signal (Zhang Ping, Li Yongqiang, Xing Hualiang. Optimal configuration of hybrid energy storage capacity for smoothing wind power fluctuations based on variational modal decomposition [J]. Power Generation Technology, 1-10 [2025-03-19].), and obtain three intrinsic mode function sets, among which the axial original force signal, the horizontal original force signal and the lateral original force signal each correspond to an intrinsic mode function set, and each intrinsic mode function set includes K intrinsic mode functions, and each intrinsic mode function is a characteristic component at a different center frequency.

[0047] In this embodiment, K is set to 3, and each set of intrinsic mode functions includes: a first eigenmode function, a second eigenmode function, and a third eigenmode function. The frequency bands of the center frequencies corresponding to the first eigenmode function, the second eigenmode function, and the third eigenmode function are from high to low. The first eigenmode function and the second eigenmode function fluctuate frequently and contain random disturbances or noise components. The third eigenmode function has a high degree of smoothness, a small overall variation amplitude, and is more consistent with the low-frequency component of the original three-dimensional force signal. The regularization penalty coefficient of the variational mode decomposition is set to 2500 to optimize the smoothness of the eigenmode function and improve the accuracy of the center frequency.

[0048] S2-2, in each eigenmode function set, an eigenmode function with the lowest center frequency band is selected as the selected eigenmode function (i.e., in this embodiment, the third eigenmode function in each eigenmode function set is selected as the selected eigenmode function), and the selected eigenmode functions of the three eigenmode function sets are combined and used as the decomposed three-dimensional force signal. The decomposed three-dimensional force signal is as follows: Figure 3 As shown in (b).

[0049] S2-3, performing Gaussian smoothing on the decomposed three-dimensional force signal to obtain a preprocessed three-dimensional force signal. The preprocessed three-dimensional force signal includes: three signal curves, each signal curve represents a force in one direction, wherein Gaussian smoothing is used to remove noise and clutter in the decomposed three-dimensional force signal.

[0050] S3, including S3-1 and S3-2;

[0051] S3-1, since the friction deposition process is divided into three stages, including the initial in-situ deposition stage, the stable deposition stage, and the terminated deposition stage, the 3D force signal corresponding to the stable deposition stage is screened from the pre-processed 3D force signal. The 3D force signal of the stable deposition stage is divided according to a fixed window size to obtain multiple sample units. Each sample unit is three signal curves corresponding to the same time period. Multiple sample units constitute the sample set D;

[0052] In this embodiment, the fixed window size is 800 points, each sample unit includes 800 sample points, and the sample set D includes 3600 sample units, of which 1200 sample units correspond to deposited layers with no defects, 1200 sample units correspond to deposited layers with slight defects, and 1200 sample units correspond to deposited layers with severe defects.

[0053] S3-2, calculate the characteristics of each signal curve in each sample unit in the sample set D. The number of characteristics of each signal curve is M / 3 categories. Each sample unit obtains M categories of characteristics. The characteristics of each signal curve include: time domain characteristics and frequency domain characteristics. The time domain characteristics are: maximum value (max), minimum value (min), mean (mean), median (med), peak-to-peak value (ran), average absolute value (arv), standard deviation (std), kurtosis (Kur), skewness (Ske), root mean square (Rms), root mean square amplitude (Srma), waveform factor (Wf), peak factor (Pf), impulse factor (If) and margin factor (Cf). The frequency domain characteristics are: center of gravity frequency (FC), root mean square frequency (Rmsf) and frequency standard deviation (Rvf).

[0054] In this embodiment, since each signal curve has 18 features, the total number of features of the three signal curves in each sample unit is M=3*18=54.

[0055] Step S3 is used to reflect the response characteristics of the three-dimensional force signal to the defect within a fixed time (fixed window).

[0056] S4, including S4-1, S4-2 and S4-3, has the following steps:

[0057] S4-1, taking each type of feature of all sample units in the sample set D as a variable, using the Pearson correlation coefficient method (PCC) to calculate the correlation coefficient between any two variables, and obtaining a correlation coefficient matrix; setting a correlation threshold, if the correlation coefficient of a certain variable with the remaining M-1 variables is less than or equal to the correlation threshold, then the type of feature corresponding to the variable is regarded as a weakly correlated feature; otherwise, it is regarded as a strongly correlated feature (that is, if the correlation coefficient of a certain variable with any of the remaining variables is greater than the correlation threshold, then the feature corresponding to the variable is regarded as a strongly correlated feature), and obtaining N1 strongly correlated features and N2 weakly correlated features;

[0058] The correlation coefficient matrix is ​​a symmetric matrix, and its diagonal elements represent the correlation coefficients of the variables themselves. The values ​​of the diagonal elements are 1, and each element outside the diagonal in the correlation coefficient matrix represents the correlation coefficient between two variables.

[0059] In this embodiment, the dimension of the correlation coefficient matrix is ​​54×54; the correlation threshold is 0.9, N1=36, N2=18, and the correlation coefficients of 45 pairs of strongly correlated features corresponding to the 36 strongly correlated features are shown in Table 1. In Table 1, "(x)" represents the horizontal original force signal corresponding to the feature, "(y)" represents the lateral original force signal corresponding to the feature, and "(z)" represents the axial original force signal corresponding to the feature.

[0060] Table 1

[0061]

[0062]

[0063] S4-2, hierarchical clustering (Cai Fapeng, Feng Ji, Yang Degang, et al. Hierarchical clustering algorithm based on natural neighborhood graph partitioning [J]. Computer Engineering and Science, 2025, 47(02): 370-380.) is used to divide N1 strongly correlated features into multiple clusters; a greedy algorithm (Zhang Qian, Nie Yibing, Dong Cunjun. Research on route optimization based on the combination of greedy algorithm and dynamic programming [J]. Modern Commerce and Industry, 2025, (07): 236-238.) is used to filter the strongly correlated features in each cluster. After filtering, a total of N3 strongly correlated features are obtained from all clusters; the N3 strongly correlated features obtained after screening and the N2 weakly correlated features are made into a feature set. The specific steps are as follows:

[0064] S4-2-1, the input of hierarchical clustering is the correlation coefficient of N1 strongly correlated features that is greater than the correlation threshold. Hierarchical clustering divides N1 strongly correlated features into multiple clusters. The formula for calculating the distance between two strongly correlated features when dividing into multiple clusters is: distance = 1-correlation coefficient. The clustering method in the hierarchical clustering algorithm (when calculating the distance between two clusters) adopts the average linkage method;

[0065] In this embodiment, the clipping threshold of hierarchical clustering is set to 0.1, and the multiple clusters output by hierarchical clustering are as follows: Figure 4 The clustering tree shown in the figure intuitively shows the clustering relationship and similarity between the strongly correlated features.

[0066] S4-2-2: Use a greedy algorithm to screen the strongly correlated features within each cluster. The input of the greedy algorithm is the strongly correlated features within the cluster being screened and the correlation coefficient of each strongly correlated feature within the cluster that is greater than the correlation threshold. The output of the greedy algorithm is at least one strongly correlated feature that is representative of the cluster, forming a representative set of strongly correlated features. The greedy algorithm is specifically as follows:

[0067] Traverse each cluster, and iterate each cluster according to the following loop steps until the correlation coefficient in the coefficient set of the cluster is 0, and form all the selected strong correlation features in the iterative process into a strong correlation feature representative set, wherein the loop steps are: the correlation coefficients of the strong correlation features in the cluster are formed into a coefficient set, each time the strong correlation feature with the largest number of correlation coefficients in the coefficient set is selected as the selected strong correlation feature, the correlation coefficient corresponding to the selected strong correlation feature is deleted from the coefficient set, and the selected strong correlation feature is deleted from the cluster;

[0068] The representative set of strongly correlated features obtained from all clusters contains N3 strongly correlated features.

[0069] In this embodiment, N3=12, and the strongly correlated features represent N3 strongly correlated features in the set. Figure 4 Strongly correlated features in bold are shown in the figure, including mean (x), impulse factor (x), RMS frequency (x), mean absolute value (x), mean (y), mean absolute value (y), impulse factor (y), RMS frequency (y), mean (z), standard deviation (z), impulse factor (z), and RMS frequency (z).

[0070] S4-2-3, N3 strongly correlated features and N2 weakly correlated features in the strongly correlated feature representative set are combined into a feature set.

[0071] S4-3, calculate the information gain (IG) of each type of feature in the feature set (Chen Xi. Research on volleyball training effect evaluation model based on information gain and random forest algorithm [J]. Journal of Kashgar University, 2024, 45(06): 79-84.), select the features with information gain values ​​greater than or equal to 0.01 in the feature set and form the final feature set H, which includes N4 types of features.

[0072] In this embodiment, if Figure 5 Shown is the information gain value of each type of feature in the final feature set H ( Figure 5 The feature abbreviations are preceded by x, y, or z, where x represents the feature corresponding to the horizontal raw force signal, y represents the feature corresponding to the lateral raw force signal, and z represents the feature corresponding to the axial raw force signal). N4 = 21, and the 21 features are: pulse factor (z), average absolute value (x), maximum value (x), peak-to-peak value (x), peak-to-peak value (y), standard deviation (x), pulse factor (x), waveform factor (x), maximum value (y), standard deviation (y), center of gravity frequency (x), standard deviation (z), mean (z), kurtosis (x), mean (x), average absolute value (y), kurtosis (y), kurtosis (z), minimum value (y), center of gravity frequency (y), and root mean square frequency (y).

[0073] Step S4-3 ranks the importance of each type of features in the feature set according to the information gain value, and selects the features in the feature set that are more important for distinguishing defects.

[0074] S5, constructing a data set. The data set consists of multiple samples, each of which includes: the value of the feature in the final feature set H corresponding to a sample unit;

[0075] In this embodiment, the data set includes 3600 samples.

[0076] Example 2

[0077] A defect detection model training method includes: training the defect detection model using a five-fold cross-validation method based on samples in a data set to obtain a trained defect detection model, wherein the true value of each sample is that the deposited part corresponding to the sample unit is defect-free or defective, and when the deposited part has a slight defect or a serious defect, the deposited part is considered defective;

[0078] Among them, the defect detection model is constructed using the random forest algorithm. The input of the defect detection model is the sample in Example 1, and the output is the prediction result. The prediction result is defect-free or defective. During the training process, the grid search method (Zhao Yucheng, Li Yingjian, Shen Shimin, et al. Research on injector fault diagnosis based on grid search and voting classification model [J]. Machine Tools and Hydraulics, 2024, 52(05): 213-220.) is used to optimize the parameters of the defect detection model.

[0079] In this embodiment, 30 decision trees are set in the random forest algorithm, each decision tree has a maximum of 19 branches, the maximum number of features allowed for each decision tree is set to 7, and at least 10 samples are required on the leaf nodes of each decision tree.

[0080] Example 3

[0081] A method for detecting defects in friction deposition forming based on three-dimensional force time-frequency domain characteristics comprises: using the defect detection model trained in Example 2 to perform defect detection on a sample to be tested, and obtaining a prediction result corresponding to the sample to be tested.

[0082] The method for obtaining the sample to be tested is consistent with the method for obtaining the sample in Example 1.

[0083] Example 4

[0084] A defect detection model training method is substantially the same as that of Example 2, except that the defect detection model outputs a prediction result of good forming, minor defects, or severe defects. The true value of each sample is that the deposited part corresponding to the sample unit has no defects, has a minor defect, or has a severe defect.

[0085] Example 5

[0086] A method for detecting defects in friction deposition forming based on three-dimensional force time-frequency domain characteristics comprises: using the trained defect detection model of Example 4 to perform defect detection on a sample to be tested, and obtaining a prediction result corresponding to the sample to be tested.

[0087] The method for obtaining the sample to be tested is consistent with the method for obtaining the sample in Example 1.

[0088] Example 6

[0089] A method for detecting defects in friction deposition forming based on three-dimensional force time-frequency domain features is basically the same as that of Example 5, with the only difference being that: each sample in this embodiment includes: the value of a feature in the final feature set H′ corresponding to a sample unit;

[0090] The method for obtaining the final feature set H′ is basically the same as the “A method for obtaining a data set” in Example 1. The only difference is that step S4 in Example 1 is replaced by the principal component analysis (PCA) algorithm, that is, the PCA algorithm is used to reduce the dimension of the number of M types of features, and the obtained multiple features are used as the final feature set H′.

[0091] like Figure 6 (a)-(c) show the accuracy of the defect detection model of Example 4 under different parameters, and the boxed values ​​in the figure are the optimal parameters.

[0092] The confusion matrix corresponding to the prediction results of the samples in the data set in Example 4 is as follows: Figure 6 As shown in (e), the corresponding visualization decision tree diagram of any decision tree is as follows Figure 6 As shown in (f), the accuracy of the prediction results of Example 4 is 98.8%; the confusion matrix corresponding to the prediction results of the samples in the data set in Example 6 is as follows Figure 7 As shown in (a), Figure 7 (b) is the ROC curve under each classification, and the accuracy of the prediction result of Example 6 is 95.7%;

[0093] In summary, Example 4 effectively distinguished well-formed samples and achieved significantly improved overall accuracy in distinguishing between minor and major defects. Example 4 was able to accurately determine the presence of forming defects in a three-dimensional force signal within a short period of time, providing a theoretical basis and technical support for real-time parameter adjustment and quality control in continuous tribo-deposition manufacturing processes based on force signals. However, the tribo-deposition forming defect detection method of Example 6 was particularly inadequate for classifying minor defects, with 120 samples misclassified.

[0094] Example 7-1

[0095] A method for detecting defects in friction deposition forming based on three-dimensional force time-frequency domain characteristics is basically the same as Example 3, with the only difference being that the defect detection model is constructed using a decision tree.

[0096] Example 7-2

[0097] A method for detecting defects in friction deposition forming based on three-dimensional force time-frequency domain characteristics is basically the same as Example 5, with the only difference being that the defect detection model is constructed using a decision tree.

[0098] Example 8-1

[0099] A method for detecting defects in friction deposition forming based on three-dimensional force time-frequency domain characteristics is basically the same as Example 3, with the only difference being that the defect detection model is constructed using a support vector machine (SVM).

[0100] Example 8-2

[0101] A method for detecting defects in friction deposition forming based on three-dimensional force time-frequency domain characteristics is basically the same as that in Example 5, with the only difference being that a defect detection model is constructed using a support vector machine (SVM).

[0102] Example 9-1

[0103] A method for detecting defects in friction deposition forming based on three-dimensional force time-frequency domain characteristics is basically the same as Example 3, with the only difference being that the defect detection model is constructed using the k-nearest neighbor algorithm.

[0104] Example 9-2

[0105] A method for detecting defects in friction deposition forming based on three-dimensional force time-frequency domain characteristics is basically the same as Example 5, with the only difference being that the defect detection model is constructed using the k-nearest neighbor algorithm.

[0106] The accuracy of the prediction results of Example 7-1, Example 7-2, Example 8-1, Example 8-2, Example 9-1, Example 9-2, Example 3 and Example 5 are shown in Tables 4 and Figure 6 (d) Figure 6 In (d), DD represents "the predicted result is no defect or defective", and DC represents "the predicted result is good molding, slight defect or severe defect"). Among them, the defect detection model of Example 3 has an accuracy rate of 98.5%, a recall rate of 98.8%, an F1 value of 98.65%, and an AUC (Area Under the Curve) of 0.9998, which is close to 1, fully demonstrating that the defect detection model performs excellently in distinguishing positive and negative samples.

[0107] Table 4

[0108]

[0109] In summary, the feature screening and dimensionality reduction method in the friction deposition forming defect detection method based on three-dimensional force time-frequency domain characteristics of the present invention has a high response to defects and has a significant impact on the model accuracy.

[0110] The above is an exemplary description of the present invention. It should be noted that, without departing from the core of the present invention, any simple deformation, modification or other equivalent replacement that can be made by other skilled in the art without expending creative labor falls within the scope of protection of the present invention.

Claims

1. A defect detection model, wherein each input sample includes: The value of the feature in the final feature set H corresponding to a sample unit; characterized in that the method for obtaining the final feature set H includes: S1, prepare the original three-dimensional force signal, which is the signal of three mutually perpendicular forces between the workpiece and the spindle in the same time period during the friction deposition forming process; S2, performing variational modal decomposition on the original three-dimensional force signal to obtain three intrinsic mode function sets, each of which includes K intrinsic mode functions; in each intrinsic mode function set, selecting the eigenmode function with the lowest center frequency band as the selected eigenmode function, combining the three selected eigenmode functions and using them as the decomposed three-dimensional force signal, and performing Gaussian smoothing to obtain three signal curves as the pre-processed three-dimensional force signal; S3, dividing the three-dimensional force signal according to a fixed window size to obtain multiple sample units, and the multiple sample units constitute a sample set D; calculating the features of each signal curve in each sample unit in the sample set D, and obtaining M features for each sample unit; S4, taking each type of feature of all sample units in the sample set D as a variable, and calculating the correlation coefficient between any two variables; setting a correlation threshold, if the correlation coefficient of a certain variable with the remaining M-1 variables is less than or equal to the correlation threshold, then the type of feature corresponding to the variable is regarded as a weakly correlated feature; otherwise, it is regarded as a strongly correlated feature, and N strongly correlated features and N2 weakly correlated features are obtained; Hierarchical clustering is used to divide N1 strongly correlated features into multiple clusters; a greedy algorithm is used to screen the strongly correlated features in each cluster, and after screening, a total of N3 strongly correlated features are obtained from all clusters; the N3 strongly correlated features obtained after screening and the N2 weakly correlated features form a feature set; The information gain of each feature in the feature set is calculated, and the features with information gain values ​​greater than or equal to 0.01 in the feature set are selected to form the final feature set H.

2. The defect detection model according to claim 1, characterized in that The characteristics of each signal curve include: one or more of: maximum value, minimum value, mean value, median value, peak-to-peak value, mean absolute value, standard deviation, kurtosis, skewness, root mean square, root mean square amplitude, shape factor, peak factor, pulse factor, margin factor, center of gravity frequency, root mean square frequency and frequency standard deviation.

3. The defect detection model according to claim 1, characterized in that In S4, the input of the hierarchical clustering is the correlation coefficients of N1 strongly correlated features that are greater than a correlation threshold.

4. The defect detection model according to claim 1, characterized in that In S4, the input of the greedy algorithm is the strongly correlated features in the screened cluster and the correlation coefficient corresponding to each strongly correlated feature in the cluster that is greater than the correlation threshold, and the output of the greedy algorithm is at least one strongly correlated feature that is representative of the cluster.

5. The defect detection model according to claim 4, characterized in that: The greedy algorithm is specifically as follows: traverse each cluster, and each cluster is iterated according to the following loop steps until the correlation coefficient in the coefficient set of the cluster is 0, and all the strongly correlated features selected in the iterative process are combined into a strongly correlated feature representative set, and the strongly correlated feature representative set obtained from all clusters is the N3 strongly correlated features; wherein, the loop step: the correlation coefficients of the strongly correlated features in the cluster are combined into a coefficient set, and each time the strongly correlated feature with the largest number of correlation coefficients in the coefficient set is selected as the selected strongly correlated feature, and the correlation coefficient corresponding to the selected strongly correlated feature is deleted from the coefficient set and the selected strongly correlated feature is deleted from the cluster.

6. The defect detection model according to claim 1, characterized in that: The defect detection model is constructed using a random forest algorithm.

7. The defect detection model according to claim 1, characterized in that: The five-fold cross validation method is used to train the defect detection model, and the grid search method is used to optimize the parameters of the defect detection model.

8. The defect detection model according to claim 1, characterized in that: The output of the defect detection model is a prediction result. When the defect detection model is used for binary classification, the prediction result is either defect-free or defective, and the true value of the sample is that the deposited part corresponding to the sample unit is defect-free or defective. When the defect detection model is used for three-classification, the prediction result is good forming, slight defect or severe defect, and the true value of the sample is that the deposited part corresponding to the sample unit has no defect, the deposited part has slight defect or the deposited part has severe defect.

9. A method for detecting defects in friction deposition forming based on three-dimensional force time-frequency domain characteristics, characterized in that: The trained defect detection model according to any one of claims 1 to 8 is used to perform defect detection on a sample to be tested, and a prediction result corresponding to the sample to be tested is obtained.

10. Use of the defect detection model according to any one of claims 1 to 8 for real-time defect detection in friction deposition forming.

Images