Adaptive sampling method and related equipment for fault diagnosis under multi-class imbalanced data

Abstract

This application provides an adaptive sampling method and related equipment for fault diagnosis under multi-class imbalanced data, which effectively solves the problems of low sample generation quality, strong parameter dependence, high computational complexity, and poor adaptability to complex working conditions. The method of this application includes: using a parameter adaptive calculation mechanism to respond to different data feature scenarios, then an evolutionary mechanism to explore the global optimal solution in the multi-classification modeling accuracy model solution optimization by employing the Newton-Raphson optimizer idea, and finally using a feature recombination mechanism to convert the optimal solution set into various types of fault samples, thereby achieving high-quality sample balance. By combining the Multi-mechanism driven Adaptive Evolutionary Sampling Technique (MAESTE) with multi-class LS-SVM, small sample multi-class imbalanced fault diagnosis is performed to increase the interpretability of the model.

Description

Technical Field

[0001] The present application relates to the technical field of mechanical fault diagnosis, and in particular to an adaptive sampling method and related equipment for fault diagnosis under multi-class imbalanced data. Background Art

[0002] With the rapid development of Industry 4.0 and the era of big data, the importance of mechanical equipment health monitoring and fault diagnosis technology in high-end manufacturing is becoming increasingly prominent. By analyzing large amounts of equipment operating data, potential faults can be identified in a timely manner, unexpected equipment downtime can be prevented, production efficiency can be improved, and operational safety can be guaranteed. However, in actual industrial applications, due to the diversity and complexity of fault types, they are often accompanied by simultaneous failures of multiple components or multiple failure modes of the same component. At the same time, the normal state data of the equipment far exceeds the fault state data, and the fault samples are often unevenly distributed and scarce in number, forming a typical small sample multi-class imbalanced data feature, which poses a huge challenge to the intelligent fault diagnosis of high-end equipment.

[0003] Currently, four main approaches are commonly used for fault diagnosis in multi-class imbalanced data: data sampling, classifier adaptation, ensemble learning, and cost-sensitive methods. Traditional sampling techniques interpolate or replicate minority class samples to balance the class distribution. While these methods are easy to implement, the quality of the synthesized samples is difficult to control, and they may introduce noise or lead to overfitting, making them particularly unstable in complex multi-class scenarios. Classifier adaptation techniques modify the classifier structure or loss function (such as Focal Loss and multi-task learning) to enhance the model's ability to identify the minority class. While these methods improve model performance to some extent, they are sensitive to hyperparameters, complex to debug, and have limited effectiveness when data is highly imbalanced. Ensemble learning methods improve model generalization by combining multiple base learners. While these methods can alleviate data imbalance to some extent, they are computationally expensive, require long model training times, and offer limited performance improvements when samples are extremely scarce. Cost-sensitive methods introduce class weights or misclassification costs during training to increase the model's focus on minority class samples. This type of method does not rely on sample generation, but requires accurate setting of cost weights, and has poor adaptability to different data sets, and faces greater uncertainty in practical applications.

[0004] In summary, in the field of high-end equipment fault diagnosis research under actual working conditions, both data-level processing methods and technical-level technologies have alleviated the problem of decreased diagnostic performance caused by small sample multi-class imbalanced data to a certain extent. However, there are generally problems such as low sample generation quality, strong parameter dependence, high computational complexity, and poor adaptability to complex working conditions. Therefore, there is an urgent need for more efficient, stable, and adaptive diagnostic methods to solve the above problems. Summary of the Invention

[0005] This application provides an adaptive sampling method and related equipment for fault diagnosis under multi-class imbalance of data, which are used to solve the above technical problems.

[0006] In a first aspect, the present application provides an adaptive sampling method for fault diagnosis under multi-class imbalanced data, the method comprising at least the following steps:

[0007] According to the original data set, each category of samples is divided, and the Euclidean distance between each category of samples and their k nearest neighbors is calculated to identify noise points, and the k nearest neighbor method is used for denoising preprocessing;

[0008] The denoised data set is subjected to feature processing. The ICEEMDAN modal decomposition method is used to perform multi-level decomposition of the vibration signal data to obtain a series of intrinsic mode functions. The Shannon entropy of each modal function is calculated, and its eigenvalue is extracted to obtain the reconstructed feature data set.

[0009] Based on the analysis of data distribution characteristics, the sample imbalance ratio of different data sets and the sample difference information within and between classes are calculated, and the key parameters in the sampling process are dynamically adjusted;

[0010] By defining an objective function that minimizes classification error, a Newton-Raphson evolutionary mechanism is used to obtain a global optimal solution in the multi-classification modeling accuracy optimization. The Newton-Raphson evolutionary mechanism includes a Newton-Raphson search and a trap avoidance mechanism.

[0011] The generated optimal solution samples are converted into new synthetic samples of the minority class using a feature recombination mechanism, and the synthetic new samples are evaluated using a fitness evaluation mechanism to ensure that the quality of the newly generated samples meets the requirements of the diagnostic model;

[0012] The MAESTE (Multi-mechanism driven Adaptive Evolutionary Sampling Technique) method is combined with multi-class LS-SVM to perform small sample multi-class imbalanced data fault diagnosis.

[0013] Optionally, divide each class of samples according to the original data set, calculate the Euclidean distance between each class of samples and their k nearest neighbors to identify noise points, and use the k-nearest neighbor method for denoising preprocessing, including:

[0014] The original data set is divided into majority class samples and minority class samples according to the number of samples in each category. The Euclidean distance between each class of samples and its k nearest neighbors is calculated to identify noise points, and the k-nearest neighbor method is used to perform denoising preprocessing on each class of samples.

[0015] Optionally, Shannon entropy is calculated for each modal function, and its eigenvalues are extracted to obtain a reconstructed feature data set, including:

[0016] The ICEEMDAN-Shannon energy entropy method is used to extract features from the denoising preprocessed data and reconstruct the dataset.

[0017] Optionally, based on data distribution characteristics, calculate the sample imbalance ratio and intra-class and inter-class sample differences of different data sets, and dynamically adjust key parameters in the sampling process, including:

[0018] According to different data characteristics, the number of fault samples that need to be generated is dynamically and adaptively calculated, and then the total number of features that need to be solved in the solution process is calculated based on the total number of samples to be generated and the number of features.

[0019] Optionally, by defining an objective function that minimizes the classification error, a Newton-Raphson evolution mechanism is used to obtain a global optimal solution in the multi-classification modeling accuracy model solution optimization. The Newton-Raphson evolution mechanism includes a Newton-Raphson search and a trap avoidance mechanism, including:

[0020] The objective function in the solution process is defined to minimize the classification error, and the Newton-Raphson search criterion is used to explore and solve the global optimal solution.

[0021] Optionally, a feature recombination mechanism is used to convert the generated optimal solution samples into newly synthesized samples of the minority class. The newly synthesized samples are evaluated using a fitness evaluation mechanism to ensure that the quality of the newly generated samples meets the requirements of the diagnostic model, including:

[0022] Use the feature recombination mechanism to reorganize the optimal solution and generate minority class fault samples;

[0023] Perform fitness evaluation on the generated samples to ensure that the quality of the newly generated samples meets the requirements;

[0024] The fitness evaluation mechanism is used to evaluate the performance of new samples in the classification model. If the fitness of the newly generated sample is better, the sample is updated. The expression of the fitness function is as follows:

[0025]

[0026] Among them, fit represents the fitness function, S i Represents the newly generated sample, fit(S i ) represents the fitness of the newly generated sample, ACC is the classification error, G represents the error of the G-Mean indicator, F represents the F-Measure error, A represents the AUC error, and 𝛼1 and 𝛼2 are two weight parameters.

[0027] Optionally, MAESTE is combined with multi-class LS-SVM to perform small sample multi-class imbalanced data fault diagnosis, including:

[0028] The newly generated high-quality samples are merged with the original sample set and divided into a training sample set and a test sample set. Multiple LS-SVM models are trained using a one-to-one strategy and classified through a voting mechanism.

[0029] The test sample set is used as the input of the multi-class LS-SVM, and the four evaluation indicators of accuracy, G-mean, F-measure and AUC are calculated. The MAESTE method is evaluated by the four evaluation indicators;

[0030] The MAESTE method is combined with multi-class LS-SVM to perform small sample multi-class imbalanced data fault diagnosis.

[0031] A second aspect of the present application provides an adaptive sampling system for fault diagnosis under multi-class imbalanced data, the system comprising:

[0032] The division unit is used to divide each category of samples according to the original data set, calculate the Euclidean distance between each category of samples and their k nearest neighbors to identify noise points, and use the k nearest neighbor method for denoising preprocessing;

[0033] The processing unit is used to perform feature processing on the denoised data set, perform multi-level decomposition of the vibration signal data using the ICEEMDAN modal decomposition method, obtain a series of intrinsic mode functions, calculate the Shannon entropy of each mode function, and extract its eigenvalue to obtain the reconstructed feature data set;

[0034] The adjustment unit is used to analyze the data distribution characteristics, calculate the sample imbalance ratio of different data sets, and the sample difference between classes, and dynamically adjust the key parameters in the sampling process;

[0035] An acquisition unit, configured to obtain a global optimal solution in the multi-classification modeling accuracy optimization solution by defining an objective function that minimizes classification error and adopting a Newton-Raphson evolution mechanism, wherein the Newton-Raphson evolution mechanism includes a Newton-Raphson search and a trap avoidance mechanism;

[0036] An evaluation unit is used to convert the generated optimal solution samples into new synthesized samples of the minority class using a feature recombination mechanism, and to evaluate the synthesized new samples using a fitness evaluation mechanism so that the quality of the newly generated samples meets the requirements of the diagnostic model;

[0037] Diagnostic unit for small sample multi-class imbalanced data fault diagnosis using MAESTE method combined with multi-class LS-SVM

[0038] A third aspect of the present application provides an adaptive sampling device for fault diagnosis under multi-class imbalance of data, the device comprising:

[0039] processor, memory, input and output units, and buses;

[0040] The processor is connected to the memory, the input and output unit, and the bus;

[0041] The storage stores a program, and the processor calls the program to execute the first aspect and any optional method in the first aspect.

[0042] In a fourth aspect, the present application provides a computer-readable storage medium, on which a program is stored. When the program is executed on a computer, the program executes the first aspect and any optional method in the first aspect.

[0043] It can be seen from the above technical solutions that this application has the following advantages:

[0044] This application uses a parameter adaptive calculation mechanism to cope with different data feature scenarios, and then uses an evolutionary mechanism to explore the global optimal solution in the multi-classification modeling accuracy model solution optimization by employing the Newton-Raphson optimizer idea. Then, the feature recombination mechanism is used to convert the optimal solution set into various types of fault samples, thereby achieving high-quality sample balance. Finally, multi-class LS-SVM is combined for classification modeling to improve the model interpretability, in order to effectively improve the classification model's ability to identify various types of fault samples in data with complex characteristics.

[0045] Therefore, this application effectively addresses issues such as low sample generation quality, strong parameter dependence, high computational complexity, and poor adaptability to complex operating conditions. MAESTE cleverly addresses the challenges of fault diagnosis modeling for high-end equipment such as cutting tools and bearing systems using small sample sizes and multiple types of unbalanced data. This addresses the limitations of existing sampling technologies, such as strong parameter dependence, uncontrollable sample quality, and an inability to adapt to complex and unknown scenarios, resulting in a wider range of applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions in this application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of this application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0047] Figure 1 This is a flow chart of an embodiment of an adaptive sampling method for fault diagnosis under multi-class imbalanced data of this application;

[0048] Figure 2A schematic diagram of an embodiment of an adaptive sampling system for fault diagnosis under multi-class imbalanced data of this application;

[0049] Figure 3 A schematic diagram of an embodiment of an adaptive sampling device for fault diagnosis under multi-class imbalance of data in this application;

[0050] Figure 4 Schematic diagram of an embodiment of the cooperation relationship between the MAESTE technology and the processing platform of this application;

[0051] Figure 5 This is the structural diagram of the MAESTE sampling algorithm in this application;

[0052] Figure 6 This is the framework diagram for combining MAESTE with multi-class LS-SVM for small sample multi-class imbalance fault diagnosis in this application. DETAILED DESCRIPTION

[0053] This application provides an adaptive sampling method and related equipment for fault diagnosis under multi-class imbalanced data, which effectively solves problems such as low sample generation quality, strong parameter dependence, high computational complexity and poor adaptability to complex working conditions.

[0054] See also Figure 1 In a first aspect, the present application provides an embodiment of an adaptive sampling method for fault diagnosis under multi-class imbalanced data, the embodiment comprising:

[0055] 101. Divide each category of samples according to the original data set, calculate the Euclidean distance between each category of samples and their k nearest neighbors to identify noise points, and use the k nearest neighbor method for denoising preprocessing;

[0056] 102. Perform feature processing on the denoised data set. Use the ICEEMDAN modal decomposition method to perform multi-level decomposition on the vibration signal data to obtain a series of intrinsic mode functions. Perform Shannon entropy calculation on each mode function and extract its eigenvalue to obtain the reconstructed feature data set.

[0057] 103. Based on the data distribution characteristics analysis, calculate the sample imbalance ratio of different data sets, the sample difference information within and between classes, and dynamically adjust the key parameters in the sampling process;

[0058] 104. By defining an objective function that minimizes classification error, a Newton-Raphson evolution mechanism is used to obtain a global optimal solution in solving and optimizing the multi-classification model accuracy model, wherein the Newton-Raphson evolution mechanism includes a Newton-Raphson search and a trap avoidance mechanism;

[0059] 105. Use the feature recombination mechanism to convert the generated optimal solution samples into new synthesized samples of the minority class, and use the fitness evaluation mechanism to evaluate the synthesized new samples so that the quality of the newly generated samples meets the requirements of the diagnostic model;

[0060] 106. The MAESTE method is combined with multi-class LS-SVM to perform small sample multi-class imbalanced data fault diagnosis.

[0061] In the embodiments of this application, the Newton-Raphson Optimizer (NRBO) is a novel meta-heuristic optimization algorithm that combines the classic Newton-Raphson method with a swarm intelligence evolutionary strategy. This algorithm leverages the local search capabilities of the gradient and Hessian matrix in the Newton-Raphson method (NRSR) to achieve rapid approximation of the objective function. It also combines population search with a trap avoidance mechanism (TAO) to enhance global search capabilities and the ability to escape local optima. Through this combined exploration and exploitation optimization approach, the algorithm demonstrates excellent convergence speed and solution accuracy for continuous optimization problems, making it particularly suitable for optimization tasks in high-dimensional, nonlinear, and complex search spaces.

[0062] k-NN preprocessing denoising (k-Nearest Neighbors) technology is a method used to cleanse and optimize data quality before modeling. It builds on the fundamental idea of the k-NN algorithm: by analyzing the categories or values of each data point and its k nearest neighbors, the data point is determined to be significantly different from its neighbors. If a point's label is inconsistent with that of most of its neighbors, it may be considered noise and subsequently corrected, smoothed, or removed. This method effectively reduces the interference of outliers and incorrect labels on model training, thereby improving the classification or regression performance of k-NN and other subsequent algorithms. It is particularly suitable for datasets that are significantly affected by noise.

[0063] The LS-SVM algorithm, or Least Squares Support Vector Machine (LS-SVM), is a machine learning algorithm used to solve problems such as pattern classification and function estimation. The core of LS-SVM regression lies in selecting an appropriate kernel function. Common kernel functions include linear, polynomial, and Gaussian kernels. The choice of kernel function depends on the specific characteristics of the problem. By selecting an appropriate kernel function, LS-SVM can handle nonlinear problems in high-dimensional space.

[0064] ICEEMDAN-Shannon feature extraction is a feature extraction method that combines signal decomposition with information entropy analysis. ICEEMDAN (Improved Fully Adaptive Ensemble Empirical Mode Decomposition with Noise) is a signal decomposition technique that decomposes non-stationary signals into a set of physically meaningful intrinsic mode functions (IMFs), effectively avoiding modal aliasing. Subsequently, by calculating the Shannon entropy of each IMF component, the information complexity and uncertainty in the time series are extracted, forming a feature vector that reflects the dynamic characteristics of the signal. This method is widely used in fields such as fault diagnosis and medical signal processing, and has strong time-frequency localization and information extraction capabilities.

[0065] Furthermore, in step 101, samples of each category are divided according to the original data. Under actual working conditions, most of the data belong to normal operation samples, which are divided into majority class samples, and samples of different faults are divided into minority class samples respectively.

[0066] Furthermore, in step 101, k-NN data preprocessing is included, that is, using the k-NN method to identify the noise samples in each category and delete the noise. x i , its neighboring sample set is N k ( x i ), the local density of the sample can be expressed as:

[0067]

[0068] in, dist ( x i , x j ) represents a sample x i With neighboring samples x j For points whose density is significantly lower than that of their neighboring samples (i.e., noise points), they are removed from the data set.

[0069] Furthermore, in step 102, the denoised data set is subjected to feature processing, and the vibration signal data is subjected to multi-level decomposition using the ICEEMDAN modal decomposition method to obtain a series of intrinsic mode functions (IMFs). Shannon entropy is calculated for each modal function (IMF), and its feature value is extracted to obtain a reconstructed feature data set.

[0070] Furthermore, in step 103, the number of generated fault samples is calculated, and the imbalance problem of small samples and the continuity of the sampling process are considered, and the following formula is used for dynamic adjustment:

[0071]

[0072] in, Num i Indicates the i The difference in the number of samples between the minority class and the majority class, Num Indicates the total number of samples that need to be generated, N len and P ilen The problem dimension dim is the total number of features that need to be solved during the solution process. It is determined by the total number of samples to be generated and the number of features. The calculation formula is as follows:

[0073]

[0074] Where fen_num represents the number of features after feature extraction. Therefore, each candidate solution represents a potential sampling strategy, whose length is equal to the product of the total number of samples and the number of features.

[0075] Furthermore, in step 104, an objective function is defined, the goal of which is to find the minimum classification error, and its calculation formula is as follows:

[0076]

[0077] in, S represents the generated samples used for training, R prd represents the classification accuracy on the test set, is the class label predicted by the model, is the true category label. n represents the total number of test samples, 1(⋅) is the indicator function, when = When , its value is 1, otherwise it is 0. Then, during the population initialization process, initial values are generated for the candidate solutions of each dimension based on the minority class samples. These values are uniformly and randomly distributed between the predetermined upper and lower boundaries to ensure that all candidate solutions are within the defined search space. The population initialization formula is as follows:

[0078]

[0079] Among them, represents the coordinate of the nth individual in the jth dimension, and rand is a random number between 0 and 1. The entire population matrix is obtained as follows:

[0080]

[0081] Then the current position vector is searched by the NRSR criterion. Perform a location update as follows:

[0082]

[0083] in, ,

[0084] , , r1, r2 represent random numbers between 0 and 1 but they are not equal, randn represents a normally distributed random number with mean 0 and variance 1, X w is the position vector of the individual with the worst fitness value in the current iteration, X b is the position vector of the current optimal solution, and They are the corresponding position updates and parameters in the NRSR search process. Guiding the population in the right direction is the adaptive coefficient that controls the balance between exploration and exploitation, ensuring that the algorithm strikes a balance between diversity and strength.

[0085] Next, we use the TAO mechanism to avoid traps and prevent falling into local optimality. The formula is as follows:

[0086]

[0087] in, , , θ1 and θ2 are random values between 0 and 1, -1 to 1, and −0.5 to 0.5, respectively. β represents a binary number, either 1 or 0. This allows the algorithm to dynamically adjust its search strategy based on the current search state, effectively guiding the population toward promising areas while maintaining sufficient exploration capacity. When the maximum number of iterations is reached, the algorithm terminates and outputs the optimal solution set.

[0088] Furthermore, in step 105, a feature recombination mechanism is used to drive sample generation. Each optimal solution corresponds to a portion of the feature dimensions, and new samples can be generated by combining these portions. Specifically, each new sample generated is composed of a combination of fen_num optimal solutions. This method ensures that the generated samples can effectively capture the diversity of minority class features while increasing the number of minority class samples. The formula for synthesizing samples can be expressed as:

[0089]

[0090] in, is the generated new sample of the i-th minority class, Represents feature dimensions fen _ num The corresponding optimal solution is then evaluated for the fitness of the newly generated samples to ensure that the quality of these samples meets the requirements of the model. The fitness evaluation is performed by measuring the performance of the new samples in the classification model. The fitness function fit The expression:

[0091]

[0092] Among them, fit represents the fitness function, S i Represents the newly generated sample, fit(S i ) represents the fitness of the newly generated sample, ACC is the classification error, G represents the error of the G-Mean indicator, F represents the F-Measure error, A represents the AUC error, 𝛼1 and 𝛼2 are two important weight parameters, 𝛼1 ranges from 0 to 1, representing the degree of emphasis on classification accuracy, and 𝛼2=(1−𝛼1) / 3, representing the weights of the other three classification evaluation indicators. Considering that classification accuracy is the main goal of the classifier, 𝛼1 is set to 0.7 and 𝛼2 is 0.1. If the fitness of the new sample is better, the sample position is updated: , by minimizing the fitness function fit , which can ensure the quality of the generated samples and improve the model's ability to recognize minority class samples.

[0093] Furthermore, in step 106, the generated samples and the sample set are merged and divided into a training sample set and a test sample set in a ratio of 7:3. For the multi-class data classification problem, a one-to-one strategy is adopted to train multiple LS-SVM models, and the final classification is performed through a voting mechanism. The test sample set is used as the input of the multi-class LS-SVM, and the label value of each test sample belonging to each class is output. According to the size of the output label values of each class, the class to which the corresponding test sample belongs is determined. Finally, the accuracy, G-mean, F-measure and AUC four evaluation indicators are calculated to evaluate the MAESTE method. For the structure of the MAESTE sampling algorithm, MAESTE is combined with multi-class LS-SVM for small sample multi-class imbalance fault diagnosis framework diagram, please refer to Figure 5 and Figure 6 .

[0094] To further illustrate the technical solution of this application, experimental tests were conducted on three publicly available datasets and two self-collected data sets based on actual operating conditions. The following description uses one of the bearing datasets as an example; the operations on the other datasets are the same. The specific implementation plan includes the following contents and steps:

[0095] Step 1: Divide the samples into each category based on the original data. Under actual working conditions, most of the data belongs to normal operation samples, which are divided into majority class samples, and samples with different faults are divided into minority class samples. Then use k-NN to preprocess the data and remove noise. The specific steps are as follows:

[0096] Step 1.1: Divide each category of samples according to the original data, divide the normal operation samples into the majority class samples N1, and divide the samples with different faults into the minority class samples P2, P3, and P4, and attach corresponding labels to them respectively.

[0097] Step 1.2: Use the k-NN method to identify the noise samples in each category and remove the noise. The local density of each category of samples is calculated using the following formula:

[0098]

[0099] Among them, dist(xi,xj) represents the Euclidean distance between sample xi and its neighboring sample xj. For points whose density is significantly lower than that of their neighboring samples (i.e., noise points), they are removed from the dataset to obtain a denoised dataset.

[0100] Step 2: Perform feature processing on the denoised dataset. Use the ICEEMDAN modal decomposition method to perform multi-level decomposition on the vibration signal data to obtain a series of intrinsic mode functions (IMFs). Perform Shannon entropy calculation on each modal function (IMF) and extract its feature value to obtain the reconstructed feature dataset.

[0101] Step 3: Adaptively calculate the key parameters of the sampling process based on data characteristics. The specific steps are as follows:

[0102] Step 3.1: Dynamically adjust key parameters based on information such as the sample imbalance ratio and intra-class and inter-class sample differences of different data sets. The difference in the number of majority and minority class samples directly affects parameter settings. First, calculate the number of faulty samples generated, that is, the difference in the number of minority and majority class samples. Since the sampling process is continuous, the following formula is used for dynamic adjustment:

[0103]

[0104] Among them, Numi represents the difference in the number of samples between the minority class and the majority class of the i-th class, that is, the number of samples that need to be generated, Nlen represents the number of normal multi-class samples, and Pilen represents the number of fault samples of the i-th class.

[0105] Step 3.1. Next, we need to define two core parameters of the sampling algorithm: population size (pop_size) and problem dimension (dim). Population size (pop_size) represents the number of candidate solutions generated in the population, while problem dimension is determined by the total number of samples to be generated and the number of features. The specific formula is:

[0106]

[0107] Here, fen_num represents the number of features. Therefore, each candidate solution represents a potential sampling strategy, whose length is equal to the product of the total number of samples and the number of features.

[0108] Step 4: Use evolutionary mechanisms to explore the global optimal solution in the optimization of multi-classification model accuracy. The NRBO algorithm combines the advantages of gradient and swarm optimization methods. By introducing the Newton-Raphson search rule (NRSR) and the trap avoidance operator (TAO), it achieves a balance between exploration and exploitation, thereby effectively improving the convergence speed and avoiding local optimal traps. The specific steps are as follows:

[0109] Step 4.1. Define the objective function to be solved. The goal is to find the function that minimizes the classification error:

[0110]

[0111] Step 4.2: Population initialization process. Initial values are generated for candidate solutions of each dimension based on minority class samples. These values are uniformly and randomly distributed between predetermined upper and lower boundaries to ensure that all candidate solutions are within the defined search space. The initialization formula is as follows:

[0112]

[0113] The entire population matrix is obtained as follows:

[0114]

[0115] Step 4.3: Use NRSR search criteria to guide the search direction. The corresponding position update is performed in NRSR, and its expression is as follows:

[0116]

[0117] Step 4.4: Using TAO can significantly change Position, to avoid falling into the local optimum, by combining the optimal position Xb and the current vector position Combined, a quality improvement The solution to Tao's formula is:

[0118]

[0119] When the maximum number of iterations is reached, the algorithm is terminated and the global optimal solution set is output.

[0120] Step 5: The optimal solution of the objective function is converted into minority class fault samples through the feature recombination mechanism, and the fitness function is evaluated on the samples. The specific steps are as follows:

[0121] Step 5.1: The optimal candidate solution is converted into a new minority class sample through the feature recombination mechanism. Each optimal solution corresponds to a part of the feature dimension. By combining these parts, a new sample can be generated. Specifically, fen_num is the feature dimension, and each generated new sample is composed of fen_num optimal solutions. This method ensures that the generated samples can effectively capture the diversity characteristics of the minority class while increasing the number of minority class samples. The formula for synthesizing samples can be expressed as:

[0122]

[0123] Step 5.2: After generating new samples, their fitness needs to be evaluated to ensure that the quality of these samples meets the requirements of the model. The fitness evaluation is performed by measuring the performance of the new samples in the classification model. The expression of the fitness function fit is:

[0124]

[0125] Step 6: Combined with multi-class LS-SVM to perform small sample multi-class imbalance fault diagnosis, increase the interpretability of the model. For multi-class data classification problems, a one-to-one strategy is used to train multiple LS-SVM models, and the final classification is performed through a voting mechanism. The specific steps are as follows:

[0126] Step 6.1: Merge the generated samples and sample sets and divide them into training sample set and test sample set, where the training sample set accounts for 70% and the test sample set accounts for 30%;

[0127] Step 6.2: Train an LS-SVM classifier for each pair of categories, and train C(C−1) / 2 classifiers in total, where C is the total number of categories.

[0128] Step 6.3: Use the test sample set as the input of LS-SVM. For each test sample, all classifiers vote and select the category with the most votes as the final classification result.

[0129] In step 6.4, the predicted values output by the classification model voting are compared with the actual categories to obtain the confusion matrix, and then the relevant evaluation indicators of the classification model are calculated to comprehensively evaluate the samples generated by evolutionary sampling and the trained classification model;

[0130] In step 6.5, the four evaluation indicators of accuracy, G-mean, F-measure and AUC are calculated to evaluate the MAESTE method.

[0131] To verify the effectiveness and superiority of the proposed MAESTE algorithm, we evaluate it on three publicly available datasets and two self-collected datasets, and compare it with eight other popular oversampling methods:

[0132] 1)No-sampling, 2) SMOTE, 3) BSMOTE, 4) IA-SUWO, 5) Cluster-SMOTE, 6) SCOTE, 7) NI-MWMOTE, 8) MWMOTE

[0133] The experimental results are shown in Table 1. The experimental results show that the resampling performance of the present invention is better than that of other compared methods.

[0134] Table 1 Comparison of resampling performance results:

[0135]

[0136] See also Figure 2 The second aspect of the present application provides an adaptive sampling system for fault diagnosis under multi-class imbalanced data, including:

[0137] The division unit 201 is used to divide each category of samples according to the original data set, calculate the Euclidean distance between each category of samples and their k nearest neighbors to identify noise points, and use the k-nearest neighbor method for denoising preprocessing;

[0138] The processing unit 202 is used to perform feature processing on the denoised data set, perform multi-level decomposition of the vibration signal data using the ICEEMDAN modal decomposition method to obtain a series of intrinsic mode functions, calculate the Shannon entropy of each mode function, and extract its eigenvalue to obtain a reconstructed feature data set;

[0139] An adjustment unit 203 is configured to calculate the sample imbalance ratio and intra-class and inter-class sample difference information of different data sets based on data distribution characteristics analysis, and dynamically adjust key parameters in the sampling process;

[0140] An acquisition unit 204 is configured to acquire a global optimal solution in the multi-classification modeling accuracy optimization by defining an objective function that minimizes classification error and adopting a Newton-Raphson evolution mechanism, wherein the Newton-Raphson evolution mechanism includes a Newton-Raphson search and a trap avoidance mechanism;

[0141] An evaluation unit 205 is configured to convert the generated optimal solution samples into newly synthesized minority class samples using a feature recombination mechanism, and to evaluate the newly synthesized samples using a fitness evaluation mechanism so that the quality of the newly generated samples meets the requirements of the diagnostic model;

[0142] The diagnosis unit 206 is used to perform small sample multi-class imbalanced data fault diagnosis by combining the MAESTE method with multi-class LS-SVM.

[0143] See also Figure 3 The present application also provides an adaptive sampling device for fault diagnosis under multi-class imbalanced data, including:

[0144] Processor 301, memory 302, input and output unit 303, bus 304;

[0145] The processor 301 is connected to the memory 302, the input and output unit 303 and the bus 304;

[0146] The memory 302 stores a program, and the processor 301 calls the program to execute any of the above methods.

[0147] The present application also relates to a computer-readable storage medium on which a program is stored, wherein when the program is run on a computer, the computer is caused to execute any of the above methods.

[0148] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0149] It should be noted that the technology proposed in the present invention can be executed in a PC, mobile terminal or similar computing device, please refer to Figure 4 , Figure 4 This is a structural block diagram of the hardware platform for classified diagnosis of health data of key components of high-end equipment involved in the present invention.

[0150] The platform includes a processor 102 and a data memory 104 , wherein the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA.

[0151] The terminal may further include a transmission device 106 and an input / output device 108 for communication functions.

[0152] Professional operation and maintenance technicians or managers in this field can understand that Figure 4 The structure shown is for illustration only and does not limit the structure of the above-mentioned terminal.

[0153] Memory 104 can be used to store computer programs, such as application software programs and modules, such as the computer program corresponding to the steam turbine data classification and prediction in this embodiment. Processor 102 executes the computer programs stored in memory 104 to perform various functional applications and data processing, thereby implementing the above-described method. Memory 104 can include high-speed random access memory (RAM) and non-volatile memory, such as a magnetic storage device, flash memory, or other non-volatile solid-state memory.

[0154] In some examples, the memory 104 may further include a memory remotely located relative to the processor 102, and these remote memories may be connected to the terminal via a network. Examples of the aforementioned network include, but are not limited to, the Internet, a power plant intranet, a local area network, a mobile communication network, and combinations thereof.

[0155] The transmission device 106 is used to receive or send data via a network, which may include a wireless network provided by the terminal's communication provider.

[0156] In one example, the transmission device 106 includes a network interface controller (NIC), which can be connected to other network devices through a base station so as to communicate with the Internet.

[0157] In one example, the transmission device 106 may be a wifi module, which is used to communicate with the Internet wirelessly.

[0158] The present application also relates to a computer-readable storage medium having a program stored thereon. When the program is run on a computer, the computer executes the method as described in the first aspect and any one of the first aspects.

[0159] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0160] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of units is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interface, device or unit, which can be electrical, mechanical or other forms.

[0161] Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0162] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0163] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the existing technology, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, read-only memory), random access memory (RAM, random access memory), disk or optical disk, and other media that can store program code.

Claims

1. An adaptive sampling method for fault diagnosis under multi-class imbalanced data, characterized by: At least the following steps are included: According to the original data set, each category of samples is divided, and the Euclidean distance between each category of samples and their k nearest neighbors is calculated to identify noise points, and the k nearest neighbor method is used for denoising preprocessing; The denoised data set is subjected to feature processing. The ICEEMDAN modal decomposition method is used to perform multi-level decomposition of the vibration signal data to obtain a series of intrinsic mode functions. The Shannon entropy of each modal function is calculated, and its eigenvalue is extracted to obtain the reconstructed feature data set. Based on the analysis of data distribution characteristics, the sample imbalance ratio of different data sets and the sample difference information within and between classes are calculated, and the key parameters in the sampling process are dynamically adjusted; By defining an objective function that minimizes classification error, a Newton-Raphson evolutionary mechanism is used to obtain a global optimal solution in the multi-classification modeling accuracy optimization. The Newton-Raphson evolutionary mechanism includes a Newton-Raphson search and a trap avoidance mechanism. The generated optimal solution samples are converted into new synthetic samples of the minority class using a feature recombination mechanism, and the synthetic new samples are evaluated using a fitness evaluation mechanism to ensure that the quality of the newly generated samples meets the requirements of the diagnostic model; The MAESTE method, a multi-mechanism driven adaptive evolutionary sampling technology, is combined with multi-class LS-SVM to perform small sample multi-class imbalanced data fault diagnosis.

2. The adaptive sampling method for fault diagnosis under multi-class imbalanced data according to claim 1, characterized in that: Divide each category of samples according to the original data set, calculate the Euclidean distance between each category of samples and their k nearest neighbors to identify noise points, and use the k-nearest neighbor method for denoising preprocessing, including: The original data set is divided into majority class samples and minority class samples according to the number of samples in each category. The Euclidean distance between each class of samples and its k nearest neighbors is calculated to identify noise points, and the k-nearest neighbor method is used to perform denoising preprocessing on each class of samples.

3. The adaptive sampling method for fault diagnosis under multi-class imbalanced data according to claim 1, characterized in that: Shannon entropy is calculated for each modal function, and its eigenvalues are extracted to obtain the reconstructed feature data set including: The ICEEMDAN-Shannon energy entropy method is used to extract features from the denoising preprocessed data and reconstruct the dataset.

4. The adaptive sampling method for fault diagnosis under multi-class imbalanced data according to claim 1, characterized in that: Based on data distribution characteristics analysis, we calculate the sample imbalance ratio of different data sets and the sample difference between classes, and dynamically adjust key parameters in the sampling process, including: According to different data characteristics, the number of fault samples that need to be generated is dynamically and adaptively calculated, and then the total number of features that need to be solved in the solution process is calculated based on the total number of samples to be generated and the number of features.

5. The adaptive sampling method for fault diagnosis under multi-class imbalanced data according to claim 2, characterized in that: By defining an objective function that minimizes classification error, a Newton-Raphson evolutionary mechanism is used to obtain a global optimal solution in the multi-classification modeling accuracy optimization. The Newton-Raphson evolutionary mechanism includes Newton-Raphson search and trap avoidance mechanisms, including: The objective function in the solution process is defined to minimize the classification error, and the Newton-Raphson search criterion is used to explore and solve the global optimal solution.

6. The adaptive sampling method for fault diagnosis under multi-class imbalanced data according to claim 1, characterized in that: The generated optimal solution samples are converted into new synthetic samples of the minority class using a feature recombination mechanism. The newly synthesized samples are evaluated using a fitness evaluation mechanism to ensure that the quality of the newly generated samples meets the requirements of the diagnostic model, including: Use the feature recombination mechanism to reorganize the optimal solution and generate minority class fault samples; Perform fitness evaluation on the generated samples to ensure that the quality of the newly generated samples meets the requirements; The fitness evaluation mechanism is used to evaluate the performance of new samples in the classification model. If the fitness of the newly generated sample is better, the sample is updated. The expression of the fitness function is as follows: Among them, fit represents the fitness function, S i Represents the newly generated sample, fit(S i ) represents the fitness of the newly generated sample, ACC is the classification error, G represents the error of the G-Mean indicator, F represents the F-Measure error, A represents the AUC error, and 𝛼1 and 𝛼2 are two weight parameters.

7. The adaptive sampling method for fault diagnosis under multi-class imbalanced data according to claim 1, characterized in that: The MAESTE method is combined with multi-class LS-SVM to perform small sample multi-class imbalanced data fault diagnosis, including: The newly generated high-quality samples are merged with the original sample set and divided into a training sample set and a test sample set. Multiple LS-SVM models are trained using a one-to-one strategy and classified through a voting mechanism. The test sample set is used as the input of the multi-class LS-SVM, and the four evaluation indicators of accuracy, G-mean, F-measure and AUC are calculated. The MAESTE method is evaluated by the four evaluation indicators; The MAESTE method is combined with multi-class LS-SVM to perform small sample multi-class imbalanced data fault diagnosis.

8. An adaptive sampling system for fault diagnosis under multi-class imbalanced data, characterized by: The system comprises: The division unit is used to divide each category of samples according to the original data set, calculate the Euclidean distance between each category of samples and their k nearest neighbors to identify noise points, and use the k nearest neighbor method for denoising preprocessing; The processing unit is used to perform feature processing on the denoised data set, perform multi-level decomposition of the vibration signal data using the ICEEMDAN modal decomposition method, obtain a series of intrinsic mode functions, calculate the Shannon entropy of each mode function, and extract its eigenvalue to obtain the reconstructed feature data set; The adjustment unit is used to analyze the data distribution characteristics, calculate the sample imbalance ratio of different data sets, and the sample difference between classes, and dynamically adjust the key parameters in the sampling process; An acquisition unit, configured to obtain a global optimal solution in the multi-classification modeling accuracy optimization solution by defining an objective function that minimizes classification error and adopting a Newton-Raphson evolution mechanism, wherein the Newton-Raphson evolution mechanism includes a Newton-Raphson search and a trap avoidance mechanism; An evaluation unit is used to convert the generated optimal solution samples into new synthesized samples of the minority class using a feature recombination mechanism, and to evaluate the synthesized new samples using a fitness evaluation mechanism so that the quality of the newly generated samples meets the requirements of the diagnostic model; The diagnostic unit is used to perform small sample multi-class imbalanced data fault diagnosis by combining the multi-mechanism driven adaptive evolutionary sampling technology MAESTE method with multi-class LS-SVM.

9. An adaptive sampling device for fault diagnosis under multi-class imbalanced data, characterized in that: The device comprises: processor, memory, input and output units, and buses; The processor is connected to the memory, the input and output unit, and the bus; The storage stores a program, and the processor calls the program to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a program, and when the program is executed on a computer, the method according to any one of claims 1 to 7 is executed.

Images