A method for accurate identification of wear state based on wear particle feature optimization

Abstract

The present invention discloses a method for accurately identifying wear states based on abrasive feature optimization. The method uses a motion abrasive image analyzer to obtain abrasive features during equipment operation to construct an initial sample, including multi-dimensional features such as abrasive quantity, size, concentration, and type. The method also uses the abrasive feature attention and correlation coefficient to eliminate features with high collinearity. A method for distinguishing abrasive feature affiliation is constructed using clustering and binning to divide the original abrasive features into several abrasive particle features. The method uses the average sample impurity to evaluate the importance of abrasive features to achieve abrasive feature optimization, and uses the optimized abrasive features to assess the wear state of mechanical equipment. Based on the random forest model architecture, the present invention establishes a wear state identification model for mechanical equipment by embedding an abrasive feature preselection module, a densification module, and a classification module. This solves the current problem of low identification accuracy in wear state assessment due to irrational abrasive feature selection.

Description

Technical Field

[0001] The present invention belongs to the technical field of wear state monitoring of mechanical equipment, and specifically relates to a method for accurately identifying wear states based on the optimization of abrasive particle characteristics. Background Art

[0002] During the operation of mechanical equipment, contact surface wear caused by the relative motion between friction pairs is inevitable. Wear is the basic form of component failure and the main cause of mechanical equipment failure. Abrasive particles, as a direct product of the wear process, carry a large amount of wear information. Classifying and identifying the wear status of mechanical equipment based on abrasive particle characteristics can not only provide decision-making support for the operation and maintenance of mechanical equipment, but also analyze the internal connection of abrasive particle characteristics, effectively promoting the intelligent operation and maintenance of mechanical equipment. However, the wear status of mechanical equipment is complex, and the numerous abrasive particle characteristics increase the difficulty of identifying the wear status of mechanical equipment. Therefore, selecting appropriate features and constructing effective classification models are research hotspots in the field of wear status monitoring of mechanical equipment.

[0003] Among the numerous wear state identification methods based on abrasive particle features, mechanical equipment learning algorithms in the field of artificial intelligence have been favored by many scholars due to their powerful nonlinear fitting capabilities. Traditional mechanical equipment learning algorithms mainly classify the wear state of mechanical equipment by constructing complex high-dimensional data features using models such as neural networks, logistic regression, and support vector machines. Such methods can integrate the characteristics of multiple abrasive particle features to achieve wear state identification. However, when there is a significant correlation between the selected features, feature redundancy can easily occur, resulting in reduced model classification accuracy. Therefore, the present invention aims to propose a wear state identification method to address the impact of feature redundancy on wear state identification accuracy.

[0004] The random forest model based on decision trees can evaluate the importance of input indicators while completing classification, thereby screening the best features. However, when applied to the task of identifying the wear status of mechanical equipment, this model still has the following problems:

[0005] 1) The random forest model evaluates the quality of feature subsets based on the training results. When there is strong collinearity in the input indicators, the model will overfit, thereby reducing the reliability of the model.

[0006] 2) The traditional random forest model ignores the impact of excessive volatility of wear particle characteristics on the wear state of mechanical equipment, resulting in low model classification accuracy and poor generalization ability.

[0007] In summary, wear particle analysis technology can characterize wear states by acquiring wear particle characteristics. However, the large fluctuation of wear particle characteristics and the existence of redundant information between characteristics restrict the accuracy of wear state identification. Summary of the Invention

[0008] The purpose of the present invention is to provide a method for accurately identifying the wear state based on the optimization of abrasive particle features, which solves the problem of information redundancy between abrasive particle features and insufficient data mining during wear state identification, and improves the accuracy of wear state identification of mechanical equipment.

[0009] The technical solution adopted by the present invention is a method for accurately identifying the wear state based on the optimization of abrasive particle characteristics.

[0010] Step 1: Collect the wear particle characteristics of mechanical equipment under different wear conditions to create a database;

[0011] Step 2: To solve the problem of over-fitting of the wear state characterization model due to the strong collinearity of wear particle features, a wear particle feature pre-selection module guided by tribological knowledge was established. The features with high collinearity were eliminated through a feature screening strategy that coordinated the wear particle feature attention and correlation coefficient.

[0012] Step 3: To further explore the inherent laws of wear particle characteristics and avoid the problem of low model accuracy caused by excessive data fluctuations, a wear particle feature densification module based on a binning strategy is established. A wear particle feature membership discrimination method is constructed in a clustering and binning manner to divide the original wear particle characteristics into several wear particle characteristics.

[0013] Step 4: To achieve wear state identification of mechanical equipment based on adaptive optimization of abrasive particle features, a high-precision wear state classification module integrating multi-level decision-making is established. The importance of abrasive particle features is first evaluated by evaluating the average impurity of the samples. The importance of the original abrasive particle features is then calculated based on the importance of the abrasive particle features. Finally, the original abrasive particle features are optimized based on the importance evaluation results, and the wear state of the mechanical equipment is then determined using a scoring mechanism.

[0014] Step 5: Based on the random forest architecture, the wear particle feature initial screening module, densification module and classification module are embedded to establish a wear state identification model based on the optimization of wear particle features to realize the wear state identification of mechanical equipment.

[0015] The present invention is also characterized in that:

[0016] Step 1 is as follows:

[0017] Step 1.1: Collect the wear characteristics of the mechanical equipment under different wear conditions, and mark each set of sampled data according to the actual wear state level of the mechanical equipment;

[0018] Step 1.2: Preprocess the monitoring data and correct or remove the label samples with missing features and obvious abnormalities in the data through mean filling and 3σ rule;

[0019] Step 1.3: Create a wear state identification dataset and divide it into training and test sets with a ratio of 5:1;

[0020] The training set is used as input to build the random forest model and complete the wear particle feature screening, and the test set is used to test the effect of the random forest model;

[0021] Step 2 is as follows:

[0022] Step 2.1: Based on tribological knowledge, a sample data selection mechanism based on the attention of wear particle characteristics is established to provide data support for the pre-selection of wear particle characteristics;

[0023] Step 2.2: Establish a feature preselection strategy based on the wear particle feature correlation coefficient matrix;

[0024] Furthermore, the feature pre-selection strategy in step 2.2 is as follows:

[0025] Step 2.2.1: Calculate the wear particle feature correlation coefficient matrix and select the features M and N with the largest correlation coefficient;

[0026] Step 2.2.2: Calculate the average values ​​m and n of the correlation coefficients between the wear particle characteristics M and N and other wear particle characteristics respectively;

[0027] Step 2.2.3: If m>n, then eliminate the wear particle feature M, otherwise eliminate the wear particle feature N;

[0028] Step 2.2.4: Repeat the above steps until the correlation coefficients between all wear particle features are less than the threshold value of 0.8.

[0029] Step 3 is as follows:

[0030] Step 3.1: Normalize the wear particle characteristics after preliminary screening;

[0031]

[0032] Where M(A) represents the Ath wear particle feature, and X(A) represents the normalized data;

[0033] Step 3.2: Establish a wear particle feature binning module based on clustering algorithm to divide the wear particle features into several sub-feature intervals according to their numerical values;

[0034] Step 3.3: Using the center and standard deviation of the sub-characteristic intervals in step 3.2, construct a mill particle characteristic membership function based on a mixed Gaussian function;

[0035]

[0036] Among them, σ k (A) and μ k(A) is the standard deviation and mean of the kth sub-feature of the original abrasive feature A.

[0037] Step 3.4: According to the result of the sample data belonging to the sub-feature interval, the original wear particle feature is divided into several wear particle features based on the maximum membership principle.

[0038] Step 4 is as follows:

[0039] Step 4.1: Use the wear particle features obtained in step 3.4 as the input of the wear state classification module of the multi-level decision-making, and use the average impurity reduction method to evaluate the importance of sub-features;

[0040] The feature importance calculation formula is as follows:

[0041]

[0042]

[0043]

[0044] Among them, VIM Ak Representative sub-feature A k The importance of; Z represents the number of decision trees in the constructed random forest model; Gini z (X) represents the Gini index of the data set X before the z-th decision tree is divided; Gini z (X, A) represents the Gini index of the dataset after the z-th decision tree is divided by feature A; H represents the number of categories of wear status in the sample dataset; p h Represents the proportion of randomly selected samples in the dataset X belonging to category h; dataset sub-feature A k A certain value of divides the data set X into two subsets X1 and X2; |X|, |X1| and |X2| represent the number of samples in the data set X and subsets X1 and X2 respectively;

[0045] Step 4.2: Perform arithmetical weighting on the abrasive particle features belonging to the same abrasive particle feature to obtain a descending set of the importance of the original abrasive particle features; when the difference in importance between adjacent abrasive particle features exceeds the threshold T, all features with the lowest order are eliminated, and the remaining abrasive particle features are retained as the optimal abrasive particle features.

[0046] Step 4.3: Use the optimal wear particle characteristics to establish several decision trees to make first-level decisions on the wear status of mechanical equipment;

[0047] Step 4.4: Establish a secondary decision-making system for wear status based on a voting scoring mechanism, and obtain the final result of the wear status assessment of mechanical equipment by statistically analyzing the results of the primary decision-making.

[0048] Step 5 is as follows:

[0049] Step 5.1: Based on the random forest architecture, embed the wear particle feature screening module, densification module, and classification module to establish a wear state identification model for mechanical equipment based on wear particle feature optimization. The specific model parameters involved include:

[0050] n_estimators = 500

[0051] max_features=6

[0052] max_depth=None (6)

[0053] min_samples_leaf=1

[0054] max_leaf_nodes=None

[0055] Among them, n_estimators represents the number of decision trees; max_features represents the number of features considered when building the optimal decision tree model; max_depth represents the maximum depth of the decision tree, and None represents the depth without limiting the number of subtrees; min_samples_leaf represents the minimum number of samples contained in a leaf node; max_leaf_nodes represents the maximum number of leaf nodes, and None represents that the model has no limit on leaf nodes.

[0056] Step S5.2: Input the test set data into the random forest model based on the optimization of wear particle characteristic parameters to complete the wear state identification of the mechanical equipment.

[0057] The beneficial effects of the present invention are:

[0058] 1. The present invention improves the original random forest model and proposes a random forest wear state identification method based on abrasive feature optimization. A data set is constructed by collecting abrasive features under different wear states of mechanical equipment; an abrasive feature preselection module guided by tribological knowledge is established, and high collinearity features are eliminated by coordinating the abrasive feature attention and correlation coefficient matrix; an abrasive feature densification module based on a binning strategy is established, and the original abrasive feature is divided into several abrasive particle features by the abrasive feature membership; a wear state classification module integrating multi-level decision-making is established, and the importance of abrasive features is evaluated by evaluating the average impurity of samples, and then the optimized abrasive features are used to distinguish the wear state of mechanical equipment.

[0059] 2. The present invention is improved based on the random forest architecture, which solves the problem of information redundancy between wear particle features and insufficient data mining when identifying the wear state, and improves the accuracy of wear state identification of mechanical equipment.

[0060] 3. The present invention eliminates wear particle features with strong collinearity based on the correlation coefficient matrix; proposes a binning strategy based on clustering algorithm and membership function to further explore the intrinsic characteristics of the data; uses a random forest model based on the original wear particle features for feature optimization; and establishes a random forest model based on the optimized wear particle features to realize the wear status identification of mechanical equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 Schematic diagram of the wear state identification model based on random forest architecture;

[0062] Figure 2 Schematic diagram of the membership function curve used in the present invention;

[0063] Figure 3 It is the 12-dimensional abrasive feature importance ranking diagram used in the present invention;

[0064] Figure 4 It is a schematic diagram of the model identification accuracy with different numbers of wear particle features;

[0065] Figure 5 3. It is a schematic diagram comparing the generalization ability of the random forest model of the present invention before and after feature optimization;

[0066] Figure 6 It is a schematic diagram of the confusion matrix of the random forest wear state identification method based on the optimization of abrasive particle features of the present invention. DETAILED DESCRIPTION

[0067] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0068] Example 1

[0069] To address the low accuracy and efficiency of wear state identification for mechanical equipment in existing technologies, the present invention proposes a random forest wear state identification method based on wear particle feature optimization. This method first constructs a random forest model using 12 features, then uses the average impurity reduction method to evaluate the importance of the 12 wear particle features and select the wear particle features most suitable for the wear state characterization task. Based on this, the optimal wear particle features are used to identify the wear state of the mechanical equipment, thereby efficiently utilizing the rich information in the wear particle features, reducing model complexity, and improving model identification efficiency.

[0070] Example 2

[0071] Reference Figure 1 The random forest wear state identification method based on wear particle feature optimization includes the following steps:

[0072] Step 1: Establish a wear state identification dataset. The specific steps are as follows:

[0073] S1: Collect the characteristics of abrasive particles under different wear conditions of mechanical equipment, including the total number of abrasive particles, the proportion of large abrasive particles, the number of 1-5μm abrasive particles, the number of 5-15μm abrasive particles, the number of 15-25μm abrasive particles, the number of 25-50μm abrasive particles, the number of 50-100μm abrasive particles, the number of abrasive particles larger than 100μm, the number of normal abrasive particles, the number of cutting abrasive particles, the number of fatigue abrasive particles and the number of severe sliding abrasive particles;

[0074] S2: Preprocess the acquired wear particle features, supplement missing samples and eliminate obviously abnormal samples through mean filling and 3σ rule;

[0075] S3: Use the remaining samples to establish a wear state identification dataset and divide it into training set and test set according to a 5:1 ratio;

[0076] Step 2: The wear particle feature pre-selection module guided by tribological knowledge eliminates features with high collinearity through a feature screening strategy that combines wear particle feature attention and correlation coefficient;

[0077] S1: Define the initial 12-dimensional wear particle feature attention levels as A to L and select sample data;

[0078] S2: Calculate the correlation coefficient matrix of all wear particle features and select a set of wear particle features M and N with the highest correlation coefficient;

[0079] S3: Calculate the average values ​​m and n of the correlation coefficients between the wear particle characteristics M and N and other wear particle characteristics respectively;

[0080] S4: If m>n, then eliminate the wear particle feature M, otherwise eliminate the wear particle feature N;

[0081] S5: Repeat the above steps until the correlation coefficient between all wear particle features is less than the threshold value of 0.8. At this time, the number of remaining wear particle features is recorded as Q.

[0082] Step 3: Establish a wear particle feature density module based on the binning strategy, construct a wear particle feature membership discrimination method in a clustering and binning manner, and divide the original wear particle features into several wear particle features. The specific steps are as follows;

[0083] S1: normalize the remaining wear particle characteristics;

[0084] S2: Use K-means clustering algorithm to divide the wear particle characteristics into several sub-feature intervals according to the numerical value;

[0085] S3: If Figure 2 As shown, the center and standard deviation of the sub-feature interval in step S2 are used to construct the abrasive particle feature membership function based on the Gaussian function, and the abrasive particle features are divided into several levels in the form of probabilistic membership, and each level is a abrasive particle feature;

[0086] Step 4: Establish a high-precision wear state classification module that integrates multi-level decision-making. Optimize the wear particle features by evaluating the average impurity of the samples. Then use the optimized wear particle features to identify the wear state of the mechanical equipment. The specific steps are as follows:

[0087] S1: Use the average impurity reduction method to evaluate the importance of the abrasive particle features and select the best features. The feature importance calculation formula is as follows:

[0088]

[0089]

[0090]

[0091] Among them VIM Ak Representative sub-feature A k The importance of; Z represents the number of decision trees in the constructed random forest model; Gini z (X) represents the Gini index of the data set X before the z-th decision tree is divided; Gini z (X, A) represents the Gini index of the dataset after the z-th decision tree is divided by feature A; H represents the number of categories of wear status in the sample dataset; p h Represents the proportion of randomly selected samples in the dataset X that belong to category h; dataset sub-feature A k A certain value of divides the data set X into two subsets X1 and X2; |X|, |X1| and |X2| represent the number of samples in the data set X and subsets X1 and X2 respectively;

[0092] S2: Weight the importance of the abrasive particle features belonging to the same abrasive particle feature to obtain a descending set of the importance of the original abrasive particle features; and finally select the 6 abrasive particle features with the greatest feature importance; Figure 3 The results of ranking the importance of abrasive features evaluated by the abrasive feature optimization method established by the present invention are shown.

[0093] Figure 4 The results of the evaluation of the importance of abrasive features are shown. The changes in the wear state identification accuracy are shown by gradually increasing the number of features in descending order of importance. It can be found that as the number of features increases, the model accuracy first increases and then decreases. When the number of features is 6, the model accuracy reaches the highest.

[0094] S3: Using the selected 6 wear particle characteristics, several decision trees are established to achieve the first-level decision of wear status;

[0095] S4: Count the results of the first-level decision and use the voting score mechanism to evaluate the final result of the wear status assessment of the mechanical equipment.

[0096] Step 5: Based on the random forest architecture, the wear particle feature screening module, densification strategy, and classification module are embedded to establish a wear state identification model. The optimized features are then used to identify the wear state of mechanical equipment. The specific steps are as follows;

[0097] S1: Establish a wear state identification model that includes a wear particle feature initial screening module, a densification strategy, and a classification module. The model parameters involved in the model training process include:

[0098] n_estimators = 500

[0099] max_features=6

[0100] max_depth=None (6)

[0101] min_samples_leaf=1

[0102] max_leaf_nodes=None

[0103] Among them, n_estimators represents the number of decision trees; max_features represents the number of features considered when building the optimal decision tree model; max_depth represents the maximum depth of the decision tree, and None represents the depth without limiting the number of subtrees; min_samples_leaf represents the minimum number of samples contained in a leaf node; max_leaf_nodes represents the maximum number of leaf nodes, and None represents that the model has no limit on leaf nodes.

[0104] S2: Use the optimal wear particle characteristic parameters to identify the wear state, and present the prediction results of the test set in the form of a confusion matrix, as shown in the following example: Figure 6 shown.

[0105] Example 3

[0106] The present invention provides a method for accurately identifying the wear state based on the optimization of abrasive particle characteristics, which is specifically implemented according to the following steps: Figure 3 Among all the abrasive characteristics, fatigue abrasive particles are the most important, reaching 11.4%. The importance of sliding abrasive particles, 50-100μm abrasive particles and the total number of abrasive particles is relatively high, at 10.87%, 10.82% and 10.57% respectively; the importance of normal abrasive particles and 1-5μm abrasive particles is relatively low, at 3.87% and 5.41% respectively.

[0107] Reference Figure 4When the number of features increases from 1 to 6, the recognition accuracy of the model increases rapidly due to the high importance of the features and the low correlation between the features. When the number of features increases from 6 to 9, the model accuracy increases slowly due to the presence of some duplicate information between the features. When the number of features increases from 9 to 12, the model not only fails to learn knowledge that helps with recognition from the dataset, but also increases the complexity of the model, resulting in a decrease in the model's recognition ability. Therefore, considering the recognition accuracy and complexity of the model, the first 6 features in the wear particle feature importance ranking are selected as the best wear particle features.

[0108] Reference Figure 5 The generalization error of the random forest model based on the original features is as high as 40%, and the model has overfitting during the training process; the generalization error of the random forest model based on the optimized features is only 4.5%, and the model has good convergence.

[0109] Reference Figure 6 The prediction results of the random forest model based on the optimized wear particle features are presented in the form of a confusion matrix, with each column representing the actual category and each row representing the predicted category. As can be seen, the prediction results for each category are generally distributed along the diagonal of the confusion matrix, indicating that the random forest model based on the optimized features can accurately identify the wear state of mechanical equipment. The model's identification accuracy for each wear state of mechanical equipment is above 90%, and the overall identification accuracy reaches 96.3%.

Claims

1. A method for accurately identifying wear states based on abrasive particle feature optimization, characterized in that: Please follow the steps below to implement: Step 1: Collect the wear particle characteristics of mechanical equipment under different wear conditions to create a database; Step 2: Establish a wear particle feature pre-selection module guided by tribological knowledge, and eliminate features with high collinearity through a feature screening strategy that combines wear particle feature attention and correlation coefficient; Step 3: Establish a wear particle feature density module based on the binning strategy, construct a wear particle feature membership discrimination method in a clustering and binning manner, and divide the original wear particle features into several wear particle features; Step 4: Establish a high-precision wear state classification module that integrates multi-level decision-making. First, evaluate the importance of the wear particle characteristics by evaluating the average impurity of the samples. Then, use the importance of the wear particle characteristics to calculate the importance of the original wear particle characteristics. Finally, optimize the original wear particle characteristics based on the importance evaluation results, and then use the scoring mechanism to determine the wear state of the mechanical equipment. Step 5: Based on the random forest architecture, the wear particle feature initial screening module, densification module and classification module are embedded to establish a wear state identification model based on wear particle feature selection to realize automatic identification of the wear state of mechanical equipment.

2. The method for accurately identifying wear states based on abrasive particle feature selection according to claim 1, characterized in that: Step 1 is implemented as follows: Step 1.1: Use a motion wear particle image analyzer to test the lubricating oil sample to obtain the multi-dimensional characteristics of the wear particles under different wear conditions of the mechanical equipment. Combined with the actual wear state of the mechanical equipment, the collected wear particle characteristics are marked with the wear state; Step 1.2: Preprocess the monitoring data. The specific operations include: using the mean filling method to fill the missing values ​​in the sample and using the 3𝛔 method to eliminate abnormal samples; Step 1.3: Establish a database for mechanical equipment wear status identification and divide it into training set and test set in a ratio of 5:

1.

3. The method for accurately identifying wear states based on abrasive particle feature selection according to claim 1, characterized in that: The specific steps of step 2 are as follows: Step 2.1: Guided by tribological knowledge, establish a sample data selection mechanism based on the attention of wear particle characteristics to provide data support for wear particle feature pre-selection; Step 2.2: Calculate the correlation coefficients between wear particle features and establish a wear particle feature preselection strategy based on feature correlation ranking.

4. The method for accurately identifying wear states based on abrasive particle feature selection according to claim 1, characterized in that: Step 3 is as follows: Step 3.1: Normalize the characteristics of the initially screened wear particles; Step 3.2: Establish a wear particle feature binning module based on the K-means clustering algorithm, in which the wear particle features are divided into several sub-feature intervals according to their numerical values; Step 3.3: Construct the abrasive particle characteristic membership function based on the Gaussian function using the center and standard deviation of the sub-characteristic interval in step 3.2, and divide the abrasive particle characteristics into several abrasive particle characteristics based on the maximum membership principle: The membership function is: (2) represents the normalized data, and Original wear particle characteristics A No. k The standard deviation and mean of each sub-characteristic.

5. The method for accurately identifying wear states based on abrasive particle feature selection according to claim 1, characterized in that: Step 4 is as follows: Step 4.1: Use the wear particle features obtained in step 3.3 as the input of the wear state classification module of the multi-level decision-making, and use the average impurity reduction method to evaluate the importance of sub-features; Step 4.2: Perform arithmetic weighting on the abrasive particle features belonging to the same abrasive particle feature to obtain a set of original abrasive particle features in descending order of importance; when the difference in importance between adjacent abrasive particle features exceeds a threshold T, all features with the lowest order are eliminated, and the remaining abrasive particle features are retained as the best abrasive particle features; Step 4.3: Use the optimal wear particle characteristics to establish several decision trees to make first-level decisions on the wear status of mechanical equipment; Step 4.4: Establish a secondary decision-making method for wear status based on the scoring mechanism, and obtain the final result of the wear status assessment of mechanical equipment by statistically analyzing the results of the primary decision.

6. The method for accurately identifying wear states based on abrasive particle feature selection according to claim 1, characterized in that: The specific steps of step 5 are as follows: Step 5.1: Based on the random forest architecture, the wear feature pre-selection module, densification module, and classification module are embedded to establish a wear state identification model based on wear feature optimization; Step 5.2: Use the wear particle features obtained in step 1 through the motion wear particle image analyzer as input to the wear state identification model based on wear particle feature optimization to achieve accurate identification of the wear state of mechanical equipment.

7. The method for accurately identifying wear states based on abrasive particle feature selection according to claim 3, characterized in that: The abrasive particle feature preselection strategy described in step 2.2 is as follows: Step 2.2.1: Calculate the wear particle feature correlation coefficient matrix and select the features M and N with the largest correlation coefficient; Step 2.2.2: Calculate the average values ​​m and n of the correlation coefficients between the wear particle characteristics M and N and other wear particle characteristics respectively; Step 2.2.3: If m>n, then eliminate the wear particle feature M, otherwise eliminate the wear particle feature N; Step 2.2.4: Repeat the above steps S2.2.2 to S2.2.3 until the correlation coefficients between all wear particle features are less than the threshold value of 0.

8.

8. The method for accurately identifying wear states based on abrasive particle feature selection according to claim 6, characterized in that: The parameters of the random forest model in step 5 are as follows: the number of decision trees is 500, the maximum number of features is 6, the maximum depth of the decision tree is unlimited, and the minimum number of samples contained in a leaf node is 1.

9. The method for accurately identifying wear states based on abrasive particle feature selection according to claim 4, characterized in that: The method for evaluating the original abrasive grain feature importance through abrasive grain feature importance is arithmetic summation.

Images