A sliding bearing fault diagnosis model construction method, device, equipment and medium
By constructing a sliding bearing fault diagnosis model based on multidimensional features and using second-order cone programming to optimize the support vector classification algorithm, the problem of low accuracy in sliding bearing fault diagnosis is solved, and efficient fault state identification is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- TANGZHI SCI & TECH HUNAN DEV CO LTD
- Filing Date
- 2026-03-09
- Publication Date
- 2026-05-15
AI Technical Summary
The accuracy of sliding bearing fault diagnosis is low. Existing technologies cannot clearly reflect changes in its operating status. Furthermore, support vector classification algorithms have high computational complexity and slow convergence speed when dealing with large-scale data.
Based on the historical operating data of sliding bearings, a multi-dimensional target feature training set is generated, a convex quadratic programming problem of support vector machine is constructed, and it is transformed into a second-order cone programming problem. The optimal sparse weight vector and bias are obtained through the solver, and a sliding bearing fault diagnosis model is constructed.
This method improves the accuracy of sliding bearing fault diagnosis, reduces computational complexity, increases solution efficiency, and enables rapid fault diagnosis.
Smart Images
Figure CN121808991B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of bearing fault diagnosis, and in particular to a method, apparatus, equipment and medium for constructing a fault diagnosis model for sliding bearings. Background Technology
[0002] A sliding bearing is a type of bearing that operates under sliding friction. It uses a lubricating medium to separate moving parts, reducing friction and wear. It is suitable for low-speed, heavy-load applications or where lubrication is difficult. Sliding bearings are widely used in aerospace, shipbuilding, and other fields. Their failure is a major threat to the normal operation of rotating machinery. Therefore, it is necessary to diagnose sliding bearing failures to avoid sudden shutdowns and production stoppages caused by malfunctions.
[0003] In related technologies, the diagnosis of sliding bearing faults is mainly achieved by measuring its mechanical vibration signals. However, unlike rolling bearings, sliding bearings do not have a specific frequency indicating bearing faults. The collision, friction, and hydrodynamic interaction between the shaft and bearing surfaces can all generate vibrations. The superimposed vibrations are difficult to clearly reflect changes in the operating state of the sliding bearing; therefore, the accuracy of diagnosing sliding bearing friction faults is low. Summary of the Invention
[0004] In view of this, the purpose of this invention is to provide a method, apparatus, device, and storage medium for constructing a sliding bearing fault diagnosis model, which can improve the accuracy of sliding bearing fault state diagnosis and increase solution efficiency. The specific solution is as follows:
[0005] On the one hand, this application discloses a method for constructing a fault diagnosis model for sliding bearings, including:
[0006] A training set is generated based on the historical operating data of the sliding bearing and the multi-dimensional target features corresponding to the historical operating data.
[0007] According to the margin maximization strategy, the training set is used to construct a convex quadratic programming problem for support vector machines, and the convex quadratic programming problem is transformed into a second-order cone programming problem.
[0008] The optimal sparse weight vector and bias are obtained by solving the second-order cone programming problem using a solver. Based on the optimal sparse weight vector and bias, the sliding bearing fault decision function is obtained to obtain the sliding bearing fault diagnosis model, so as to use the sliding bearing fault diagnosis model for sliding bearing fault diagnosis.
[0009] Optionally, the generation of a training set based on the historical operating data of the sliding bearing and the corresponding multidimensional target features includes:
[0010] Acquire historical operating data of the sliding bearing; the historical operating data includes normal operation data and fault data; the historical operating data includes any one or more of vibration data, impact data, power data, speed data, bearing shell temperature data, and oil data; the vibration data includes any one or more of the vibration acceleration data of the bearing housing, the vibration acceleration data of the machine body, and the vibration displacement data on the shaft;
[0011] Multidimensional target features are extracted based on the historical operating data; the multidimensional target features include any one or more of vibration features, impact features, operating condition features, bearing temperature features, and oil features;
[0012] The dataset is obtained based on the historical operating data and the multidimensional target features. The dataset is then divided into a training set and a test set.
[0013] Optionally, the step of constructing a convex quadratic programming problem for a support vector machine using the training set according to the margin maximization strategy includes:
[0014] Convex quadratic programming problem with sparse constraints is constructed based on the training set; the convex quadratic programming problem includes a first objective function and a first constraint condition;
[0015] The first objective function is constructed based on half of the sum of squares of the feature weights, the sum of the slack variables of all training samples, the sum of the absolute values of the feature weights, a first hyperparameter for the sum of the slack variables of all training samples, and a second hyperparameter for the sum of the absolute values of the feature weights; the first constraint includes a classification margin constraint and a non-negativity constraint for the slack variables.
[0016] Optionally, the first objective function is:
[0017] ;
[0018] in, The feature weights of feature d are... Let C be the slack variable for the training sample n, and let C be the first hyperparameter. This is the second hyperparameter. This is a sparsity regularization term.
[0019] Optionally, transforming the convex quadratic programming problem into a second-order cone programming problem includes:
[0020] The second objective function is obtained by replacing the sum of squares of the feature weights in the first objective function with a first auxiliary variable and replacing the sum of the absolute values of the feature weights with a second auxiliary variable.
[0021] By using the first auxiliary variable as an upper bound constraint on the sum of squares of the feature weights, a second constraint condition is obtained.
[0022] By using the second auxiliary variable as an upper bound constraint on the sum of the absolute values of the feature weights, a third constraint condition is obtained.
[0023] Based on the grouping of the multidimensional target features, weight constraints are constructed for the feature groups to obtain the fourth constraint.
[0024] Based on the second objective function, the first constraint, the second constraint, the third constraint, and the fourth constraint, the second-order cone programming problem is obtained.
[0025] Optionally, the second objective function is:
[0026] ;
[0027] Where t is the first auxiliary variable, u is the second auxiliary variable, and C is the first hyperparameter. For the slack variables of training sample n, This is the second hyperparameter;
[0028] The third constraint is as follows:
[0029] ;
[0030] in, The feature weights of feature d are... All are auxiliary variables, with u being the second auxiliary variable.
[0031] Optionally, after obtaining the sliding bearing fault decision function based on the optimal sparse weight vector and bias to obtain the sliding bearing fault diagnosis model, the method further includes:
[0032] The sliding bearing fault diagnosis model is evaluated using a test set based on preset evaluation indicators.
[0033] An evaluation score is obtained based on the evaluation results of each of the preset evaluation indicators.
[0034] Optionally, after obtaining the sliding bearing fault decision function based on the optimal sparse weight vector and bias to obtain the sliding bearing fault diagnosis model, the method further includes:
[0035] A new training set is generated based on the new historical operating data of sliding bearings;
[0036] Determine the misclassification rate of the sliding bearing fault diagnosis model on the new training set;
[0037] If the misclassification rate exceeds a preset threshold, the second-order cone programming problem is resolved based on the new training set to update the optimal sparse weight vector and bias, thereby updating the sliding bearing fault diagnosis model.
[0038] On the other hand, this application discloses a method for diagnosing sliding bearing faults, including:
[0039] Obtain real-time operating data of the sliding bearing;
[0040] The real-time operating data is input into the sliding bearing fault diagnosis model, and the sliding bearing fault diagnosis result is obtained according to the model output; the sliding bearing fault diagnosis model is a model constructed using the aforementioned sliding bearing fault diagnosis model construction method.
[0041] On the other hand, this application discloses a sliding bearing fault diagnosis model construction device, including:
[0042] The training set generation module is used to generate a training set based on the historical operating data of the sliding bearing and the multi-dimensional target features corresponding to the historical operating data.
[0043] The problem transformation module is used to construct a convex quadratic programming problem for a support vector machine using the training set according to the margin maximization strategy, and to transform the convex quadratic programming problem into a second-order cone programming problem.
[0044] The problem-solving module is used to solve the second-order cone programming problem using a solver to obtain the optimal sparse weight vector and bias. Based on the optimal sparse weight vector and bias, the sliding bearing fault decision function is obtained to obtain the sliding bearing fault diagnosis model, so as to use the sliding bearing fault diagnosis model for sliding bearing fault diagnosis.
[0045] On the other hand, this application discloses an electronic device, including:
[0046] Memory, used to store computer programs;
[0047] A processor is used to execute the computer program to implement the aforementioned method for constructing a sliding bearing fault diagnosis model.
[0048] On the other hand, this application discloses a computer-readable storage medium for storing a computer program; wherein the computer program, when executed by a processor, implements the aforementioned method for constructing a sliding bearing fault diagnosis model.
[0049] In this application, a training set is generated based on the historical operating data of the sliding bearing and the corresponding multidimensional target features. Following a margin maximization strategy, a convex quadratic programming problem for a support vector machine (SVM) is constructed using the training set, and this problem is transformed into a second-order cone programming problem. The second-order cone programming problem is solved using a solver to obtain the optimal sparse weight vector and bias. Based on the optimal sparse weight vector and bias, a sliding bearing fault decision function is obtained to derive a sliding bearing fault diagnosis model, which is then used for sliding bearing fault diagnosis. It is evident that by extracting multidimensional target features from the sliding bearing's operating data and incorporating second-order cone programming into the SVM algorithm, features at different scales are effectively integrated, effectively compensating for the shortcomings of single-dimensional data analysis and improving the confidence of sliding bearing fault state diagnosis. Furthermore, introducing second-order cone programming into the convex quadratic programming solution of the sliding bearing SVM classification model transforms the complex sliding bearing classification convex quadratic programming problem into a second-order cone programming problem, reducing computational complexity and improving solution efficiency. Attached Figure Description
[0050] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.
[0051] Figure 1 A flowchart of a sliding bearing fault diagnosis model construction method provided in this application;
[0052] Figure 2 A flowchart illustrating a specific method for constructing a sliding bearing fault diagnosis model is provided in this application.
[0053] Figure 3 A specific receiver operating characteristic (ROC) curve is provided for this application.
[0054] Figure 4 A schematic diagram of a sliding bearing fault diagnosis model construction device provided in this application;
[0055] Figure 5 This application provides a structural diagram of an electronic device. Detailed Implementation
[0056] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0057] In related technologies, the diagnosis of sliding bearing faults is usually achieved by measuring its mechanical vibration signals. However, unlike rolling bearings, sliding bearings do not have a specific frequency indicating a bearing fault. Vibrations can be generated by collisions, friction, and hydrodynamic interactions between the shaft and bearing surfaces, resulting in a mixture of vibrations that are difficult to clearly reflect changes in the operating state of the sliding bearing; thus, the accuracy of diagnosing sliding bearing friction faults is low. The core of the support vector classification algorithm is a convex quadratic programming problem. Commonly used interior-point methods and sequential minimum optimization algorithms have high computational complexity and slow convergence speed when dealing with large-scale data, greatly limiting the application of support vector classification algorithms in the classification and diagnosis of sliding bearing faults. To overcome the above technical problems, this application proposes a sliding bearing fault diagnosis model construction method, which can improve the accuracy of sliding bearing fault state diagnosis and increase solution efficiency.
[0058] This application discloses a method for constructing a sliding bearing fault diagnosis model. (See also...) Figure 1 As shown, the method may include the following steps:
[0059] Step S11: Generate a training set based on the historical operating data of the sliding bearing and the multi-dimensional target features corresponding to the historical operating data.
[0060] First, a training set is constructed to prepare for subsequent model training. The historical operating data of the sliding bearing includes vibration data, impact data, power data, speed data, bearing temperature data, and oil data under normal and fault conditions; the extracted target features include vibration features, impact features, operating condition features, bearing temperature features, and oil features.
[0061] In a preferred embodiment, a training set is generated based on the historical operating data of the sliding bearing and the multi-dimensional target features corresponding to the historical operating data, for example... Figure 2 As shown, it includes the following steps:
[0062] S201: Obtain historical operating data of the sliding bearing; the historical operating data includes normal operation data and fault data; the historical operating data includes any one or more of vibration data, impact data, power data, speed data, bearing temperature data and oil data.
[0063] Data on normal and faulty sliding bearings are collected. The data mainly consists of vibration, impact, power, speed, bearing temperature, and oil data. The specific data types can be flexibly adjusted and selected according to measurement and diagnostic needs.
[0064] Among them, vibration data includes vibration acceleration data of bearing housing, vibration acceleration data of machine body, and vibration displacement data of shaft; impact data is the impact obtained by generalized resonance demodulation technology of vibration acceleration data of bearing housing; power is the power output of the equipment where the sliding bearing is located; speed is the speed of the shaft where the sliding bearing is located; bearing temperature is the temperature measured by the sliding bearing; oil data includes oil abrasive particle size count, oil temperature, pressure, moisture, contamination degree, viscosity, etc.
[0065] S202: Preprocess the historical operation data; the preprocessing includes outlier detection and processing, and data standardization.
[0066] The above-mentioned outlier detection and processing includes: performing probability density statistics on the historical operating data using a target probability density function, filtering out outliers from the historical operating data based on the statistical results, and performing removal or interpolation correction processing on the outliers.
[0067] This involves preprocessing the data before feature extraction, including outlier detection and data standardization. Specifically, probability density statistical algorithms can be used to detect outliers, marking those exceeding a set threshold as outliers, and then removing or interpolating the marked data.
[0068] The probability density statistical algorithm can employ distributions such as normal, exponential, uniform, Poisson, or gamma. Taking the probability density function of the normal distribution as an example, for... Data outside the range of (k is configurable, preferably 5) is marked as outliers:
[0069] ;
[0070] in, The average value of the data x; Let x be the variance of the data.
[0071] Data standardization eliminates differences in data units, making different features comparable. This embodiment does not limit the specific standardization method; Z-score standardization is used as an example. Standardization is achieved by transforming the original data into a distribution with a mean of 0 and a standard deviation of 1.
[0072] S203: Extract multi-dimensional target features based on the historical operating data; the multi-dimensional target features include any one or more of vibration features, impact features, operating condition features, bearing temperature features, and oil features.
[0073] The aforementioned extraction of multi-dimensional target features based on the historical operational data includes: extracting vibration features from vibration data and rotational speed data to obtain vibration features. These vibration features include equivalent vibration intensity evaluation features, waveform factors, peak values, and kurtosis factors.
[0074] (1) Equivalent vibration intensity. Equivalent vibration intensity is the composite result of vibration velocity measurements in multiple directions. The specific calculation process is as follows:
[0075] Calculate the root mean square value of the speed:
[0076] ;
[0077] in, The root mean square value of vibration velocity ( ), It is a periodic time function of vibration velocity. This is obtained by integrating the vibration acceleration data, where T is the period of the vibration velocity variation with time. (This is for the bearing housing.) Vibration acceleration data in three directions, body The vibration acceleration data in three directions are integrated to obtain the corresponding values. Vibration velocity data in three directions were collected, and the corresponding values for each direction were calculated. ( The characteristics of equivalent vibration intensity evaluation are:
[0078] ;
[0079] It is the equivalent vibration intensity ( ), The effective values of vibration velocity were measured in three mutually perpendicular directions. ), These represent the number of measurement points in the three directions.
[0080] (2) Waveform factor :
[0081] ;
[0082] Waveform factor is an effective value with absolute average The ratio of the waveform index to the bearing's surface area. When the waveform index value is too large, it indicates that the bearing may have pitting, while a smaller waveform index may indicate wear.
[0083] (3) Peak factor :
[0084] ;
[0085] Peak value With effective value The ratio of peak index to RMS value. For transient impact vibrations caused by discrete defects such as surface spalling or scratches, the peak index is more sensitive than the RMS value. Generally, the peak index for normal vibration is about 4-5. When scratches occur, the peak index can sometimes reach 10. The larger the defect, the higher the peak index. The larger the peak value, the easier it is to identify anomalies in key components of the transmission system. However, as the fault expands and the peak value gradually reaches its extreme, the root mean square value also increases, and the peak value begins to decrease. Under online monitoring, monitoring the peak value can effectively provide early warning of key components in the transmission system and reflect the development trend of the fault.
[0086] (4) Kurtosis factor :
[0087] ;
[0088] Kurtosis index can reflect the impact characteristics of vibration signals in the time domain.
[0089] The aforementioned extraction of multi-dimensional target features based on historical operating data includes: extracting impact features from impact data. Since sliding bearings generate high-frequency impacts during operation due to wear, fatigue spalling, or galling, generalized resonance demodulation technology is used to enhance the weak impact signal of the sliding bearing. The parameters obtained from generalized resonance demodulation technology, specifically the Shock Velocity (SV), reflect the high-frequency impact energy during the sliding bearing's operation. This is achieved by electronically resonating and amplifying the weak impact signal of the sliding bearing to extract high signal-to-noise ratio, high-quality data, thus enhancing the weak signal of the sliding bearing.
[0090] Specifically, the above-mentioned extraction of multi-dimensional target features based on the historical operating data includes: extracting operating condition features from power data and speed data. Regarding power data... and speed data The operating conditions can be divided into segments, such as intervals. and The range is operating condition interval 1. and The range is the operating condition interval. A total of M working conditions were obtained;
[0091] Normalize the operating conditions to obtain the operating condition characteristics. : .
[0092] The aforementioned extraction of multi-dimensional target features based on historical operating data includes: extracting bearing temperature features from bearing temperature data. For the bearing temperature data, features reflecting its changing trends and operating status are extracted using threshold settings and time-series trend analysis, including maximum values. Minimum value ,average value ,variance Temperature rise (k is configurable), temperature change rate, etc. (k can be set) (Sampling interval).
[0093] The aforementioned extraction of multi-dimensional target features based on the historical operating data includes: extracting oil features from oil data. Oil features include, but are not limited to, oil temperature, pressure, moisture content, contamination level, and viscosity, with the maximum, minimum, average, variance, increase, and rate of change of each parameter calculated. Oil features also include oil abrasive particle size, and the counting of abrasive particle size can be performed using different standards to count abrasive particles of different sizes.
[0094] It is evident that by constructing a model using multi-source data and corresponding multi-dimensional features, the problem of low accuracy in sliding bearing diagnosis can be solved.
[0095] S204: Based on the historical running data and the multidimensional target features, a dataset is obtained, and the dataset is divided to obtain a training set and a test set.
[0096] Split the dataset into training and testing sets, selecting the splitting ratio; for example, splitting it into a 7:3 ratio.
[0097] Step S12: According to the margin maximization strategy, construct a convex quadratic programming problem for the support vector machine using the training set, and transform the convex quadratic programming problem into a second-order cone programming problem.
[0098] Support Vector Machines (SVMs) are a type of generalized linear classifier that performs binary classification of data using supervised learning. Their decision boundary is the hyperplane with the maximum margin calculated from the training samples. This embodiment first constructs a convex quadratic programming problem for SVM, then transforms it into a second-order cone programming problem. The trained model is obtained by solving the second-order cone programming problem. Introducing second-order cone programming into the convex quadratic programming solution of the sliding bearing SVM classification model reduces computational complexity, improves solution efficiency, and enhances model training efficiency, supporting rapid training scenarios.
[0099] In some embodiments, constructing a convex quadratic programming problem for a support vector machine using the training set according to the margin maximization strategy includes: constructing a convex quadratic programming problem with sparsity constraints based on the training set; the convex quadratic programming problem includes a first objective function and a first constraint; the first objective function is constructed based on half the sum of squares of the feature weights, the sum of slack variables of all training samples, the sum of absolute values of the feature weights, a first hyperparameter for the sum of slack variables of all training samples, and a second hyperparameter for the sum of absolute values of the feature weights; the first constraint includes a classification margin constraint and a non-negativity constraint for slack variables.
[0100] Construct an SVM training model with the strategy of margin maximization and the learning algorithm of convex quadratic programming; its decision function is:
[0101] ;in, Let b be the weight vector, and b be the bias term. It is a nonlinear function that maps input data x to a high-dimensional feature space.
[0102] The core of SVM improves the model's generalization ability by maximizing the classification margin, which is transformed into solving the following convex quadratic programming problem:
[0103] Minimize the first objective function:
[0104] ;
[0105] in, The weight vector determines the weights of each feature. (x) represents the importance of x in the decision function; ξ is a slack variable. For sparsity regularization, L1 norm regularization can be used. C is the first hyperparameter. This is the second hyperparameter.
[0106] First constraint: ;
[0107] The first set of constraints includes the categorical margin constraint and the non-negativity constraint for slack variables. Let b be a nonlinear function that maps input data x to a high-dimensional feature space, and b be a bias term. Let i be the label of sample i.
[0108] In some embodiments, transforming the convex quadratic programming problem into a second-order cone programming problem includes:
[0109] S301: Replace the sum of squares of the feature weights in the first objective function with a first auxiliary variable, and replace the sum of the absolute values of the feature weights with a second auxiliary variable to obtain the second objective function;
[0110] The convex quadratic programming problem is equivalently transformed into a second-order cone programming (SOCP) problem. The standard form of SOCP includes a linear objective function and second-order cone constraints. First, a first auxiliary variable t and a second auxiliary variable u are introduced to constrain the norm term in the original objective function, respectively, thus linearizing the objective function and obtaining the second objective function:
[0111] .
[0112] S302: Using the first auxiliary variable as the upper bound constraint of the sum of squares of the feature weights, a second constraint condition is obtained;
[0113] By using the first auxiliary variable as an upper bound constraint on the sum of squares of the feature weights, i.e., adding an L2 norm constraint (rotated second-order cone constraint), we obtain the second constraint condition:
[0114] ;
[0115] in, Let t be the feature weight of feature d, and t be the first auxiliary variable.
[0116] S303: Using the second auxiliary variable as the upper bound constraint of the sum of the absolute values of the feature weights, a third constraint condition is obtained;
[0117] Using the second auxiliary variable as an upper bound constraint on the sum of the absolute values of the feature weights, i.e., adding an L1 norm constraint, such as with L1 norm regularization, we obtain a third constraint condition:
[0118] Specifically, by introducing auxiliary variables Add constraints:
[0119] .
[0120] S304: Based on the grouping of the multidimensional target features, construct weight constraints for the feature groups to obtain the fourth constraint.
[0121] Based on the grouping of the multidimensional target features, weight constraints are constructed for the feature groups, resulting in a fourth constraint. This means grouping the features according to the actual situation and needs, adding group constraints. For example, suppose there are three groups of features: Group 1 contains features... Group 2 contains features Group 3 contains features Introducing auxiliary variables Adding constraints yields the fourth constraint condition:
[0122] .
[0123] S305: Based on the second objective function, the first constraint, the second constraint, the third constraint, and the fourth constraint, the second-order cone programming problem is obtained.
[0124] By introducing auxiliary variables and adding constraints, a second-order cone programming problem is obtained, which includes a second objective function, a first constraint, a second constraint, a third constraint, and a fourth constraint.
[0125] By adding various constraints such as L1 norm constraints, L2 norm constraints, and group constraints, the sparse solution of the sliding bearing is automatically obtained, enabling the selection of multiple features. At the same time, it can clearly point out the features that are important for classification, thus enhancing the interpretability of the model.
[0126] Step S13: Solve the second-order cone programming problem using a solver to obtain the optimal sparse weight vector and bias. Based on the optimal sparse weight vector and bias, obtain the sliding bearing fault decision function to obtain the sliding bearing fault diagnosis model, so as to use the sliding bearing fault diagnosis model for sliding bearing fault diagnosis.
[0127] The second-order cone programming problem is input into a specialized solver to obtain the optimal sparse weight vector. and bias Based on the optimal sparse weight vector and bias, the final sliding bearing fault decision function is constructed:
[0128] ;
[0129] Due to the addition of sparsity constraints, the weight vector With a large number of elements being 0, feature selection is achieved.
[0130] In some embodiments, after obtaining the sliding bearing fault decision function based on the optimal sparse weight vector and bias to obtain the sliding bearing fault diagnosis model, the method further includes: evaluating the sliding bearing fault diagnosis model using a test set according to preset evaluation indicators; and obtaining an evaluation score based on the evaluation results of each preset evaluation indicator.
[0131] The preset evaluation metrics can be any one or more of the following: accuracy, AUC (Area Under Curve), ROC (Receiver Operating Characteristic), and F-score (F-measure).
[0132] Accuracy is a fundamental criterion for evaluating classification performance, reflecting the overall classification performance of the classifier. Its calculation formula is as follows:
[0133] ;
[0134] Among them, TN is classified as faulty even though its actual value is faulty; FN is classified as faulty even though its actual value is normal; FP is classified as normal even though its actual value is faulty; and FN is classified as normal even though its actual value is normal.
[0135] The F-score is one of the commonly used metrics for measuring the classification performance of a classifier on an imbalanced dataset, and it is defined as follows:
[0136] ;
[0137] Among them, accuracy Recall rate , Values ,like A value of 1 represents the average weight between recall and precision. For us, higher precision and recall are better, with higher precision indicating a higher accuracy in classifying the minority class. This can be achieved by adjusting... Values are a good measure of the classification performance of classifiers in imbalanced datasets.
[0138] AUC is the area under the corresponding ROC curve, commonly used to measure the performance of a classifier. It's a curve plotted with sensitivity and specificity on the axes. Each point on the curve corresponds to the classification accuracy with each sample size. The larger the area under the AUC curve, the better the classifier's performance. Different sample sizes correspond to different levels of sensitivity and specificity. A smoother curve indicates more samples. A perfectly smooth ROC curve represents the classification performance for every sampling result, thus comprehensively reflecting the classifier's performance. Therefore, it's often used to measure the classification performance of a classifier in imbalanced data. The vertical axis of the ROC curve represents specificity. The ROC curve represents the relationship between the minority class classification accuracy and the majority class classification error rate under different sampling conditions. Each time a different sample size is selected, the classifier will produce the minority class classification accuracy and the majority class classification error rate, corresponding to points on the ROC curve. By sampling multiple times and calculating multiple sensitivities and specificities, the ROC curve can be plotted. For example... Figure 3 The ROC curve diagram shows that the top-left corner A is (0,1), representing that both minority and majority class samples are correctly classified. As the classification error rate of majority class samples increases during sampling, the curve shifts to the right. Therefore, the closer the curve is to the top-left corner, the better the classification performance, and the larger the corresponding area, reflecting that a larger area is better. The ROC curve comprehensively describes the classification performance of the classifier under different samples, but it cannot intuitively show which of two similar classifiers is optimal. The area under the ROC curve, AUC, does not have this problem. It can intuitively and easily evaluate the performance of the classifier, and the type of classifier does not affect the AUC value.
[0139] The evaluation score can be:
[0140] ;
[0141] in, The sliding bearing fault diagnosis model with the best evaluation score and that meets the prediction requirements is selected for publication.
[0142] In some embodiments, after obtaining the sliding bearing fault decision function based on the optimal sparse weight vector and bias to obtain the sliding bearing fault diagnosis model, the method further includes: generating a new training set based on new historical operating data of the sliding bearing; determining the misclassification rate of the sliding bearing fault diagnosis model on the new training set; if the misclassification rate exceeds a preset threshold, resolving the second-order cone programming problem based on the new training set to update the optimal sparse weight vector and bias, thereby updating the sliding bearing fault diagnosis model.
[0143] For example Figure 2 As shown, the trained model is evaluated to assess its performance, and retraining or re-release is performed as needed. The sliding bearing fault model is updated based on subsequent new data to achieve incremental learning. The steps include: Step 401: Initialize the SVM sliding bearing fault model, including initializing parameters and kernel functions. Step 402: Collect new data and update the training dataset. Step 403: Classify the new data and calculate the misclassification rate. Step 404: Update the SVM model parameters based on the misclassification rate. Step 405: Repeat steps 402-404 until convergence or the maximum number of iterations is reached. The converged sliding bearing fault model is evaluated to analyze its performance, and training or re-release is performed as needed.
[0144] As can be seen from the above, in this embodiment, a training set is generated based on the historical operating data of the sliding bearing and the multi-dimensional target features corresponding to the historical operating data. Following the interval maximization strategy, a convex quadratic programming problem for support vector machines is constructed using the training set, and this convex quadratic programming problem is transformed into a second-order cone programming problem. The second-order cone programming problem is solved using a solver to obtain the optimal sparse weight vector and bias. Based on the optimal sparse weight vector and bias, a sliding bearing fault decision function is obtained to obtain a sliding bearing fault diagnosis model, which can then be used for sliding bearing fault diagnosis. It is evident that by extracting multi-dimensional target features from the sliding bearing's operating data, and introducing second-order cone programming into the support vector classification algorithm, features at different scales are effectively integrated, effectively compensating for the shortcomings of single-dimensional data analysis and improving the confidence of sliding bearing fault state diagnosis. Furthermore, introducing second-order cone programming into the convex quadratic programming solution of the sliding bearing support vector classification model transforms the complex sliding bearing classification convex quadratic programming problem into a second-order cone programming problem, reducing computational complexity and improving solution efficiency.
[0145] This application discloses a specific method for diagnosing sliding bearing faults, which may include the following steps:
[0146] Step S51: Obtain real-time operating data of the sliding bearing.
[0147] Step S52: Input the real-time running data into the sliding bearing fault diagnosis model, and obtain the sliding bearing fault diagnosis result according to the model output; the sliding bearing fault diagnosis model is a model constructed using the above-mentioned sliding bearing fault diagnosis model construction method.
[0148] The specific process of step S52 can be found in the relevant content disclosed in the foregoing embodiments, and will not be repeated here.
[0149] Accordingly, this application also discloses a sliding bearing fault diagnosis model construction device, see [link to relevant documentation]. Figure 4 As shown, the device includes:
[0150] Training set generation module 11 is used to generate a training set based on the historical operating data of the sliding bearing and the multi-dimensional target features corresponding to the historical operating data;
[0151] Problem transformation module 12 is used to construct a convex quadratic programming problem of support vector machine using the training set according to the interval maximization strategy, and transform the convex quadratic programming problem into a second-order cone programming problem;
[0152] The problem-solving module 13 is used to solve the second-order cone programming problem using a solver to obtain the optimal sparse weight vector and bias, and to obtain the sliding bearing fault decision function based on the optimal sparse weight vector and bias, so as to obtain the sliding bearing fault diagnosis model, so as to use the sliding bearing fault diagnosis model to diagnose sliding bearing faults.
[0153] As can be seen from the above, in this embodiment, a training set is generated based on the historical operating data of the sliding bearing and the multi-dimensional target features corresponding to the historical operating data. Following the interval maximization strategy, a convex quadratic programming problem for support vector machines is constructed using the training set, and this convex quadratic programming problem is transformed into a second-order cone programming problem. The second-order cone programming problem is solved using a solver to obtain the optimal sparse weight vector and bias. Based on the optimal sparse weight vector and bias, a sliding bearing fault decision function is obtained to obtain a sliding bearing fault diagnosis model, which can then be used for sliding bearing fault diagnosis. It is evident that by extracting multi-dimensional target features from the sliding bearing's operating data, and introducing second-order cone programming into the support vector classification algorithm, features at different scales are effectively integrated, effectively compensating for the shortcomings of single-dimensional data analysis and improving the confidence of sliding bearing fault state diagnosis. Furthermore, introducing second-order cone programming into the convex quadratic programming solution of the sliding bearing support vector classification model transforms the complex sliding bearing classification convex quadratic programming problem into a second-order cone programming problem, reducing computational complexity and improving solution efficiency.
[0154] In some specific embodiments, the training set generation module 11 may specifically include:
[0155] The data acquisition unit is used to acquire historical operating data of the sliding bearing; the historical operating data includes normal operation data and fault data; the historical operating data includes any one or more of vibration data, impact data, power data, speed data, bearing temperature data and oil data; the vibration data includes any one or more of the vibration acceleration data of the bearing housing, the vibration acceleration data of the machine body, and the vibration displacement data on the shaft.
[0156] The feature extraction unit is used to extract multi-dimensional target features based on the historical operating data; the multi-dimensional target features include any one or more of vibration features, impact features, operating condition features, bearing temperature features, and oil features;
[0157] The dataset generation unit is used to obtain a dataset based on the historical running data and the multidimensional target features, and to divide the dataset to obtain a training set and a test set.
[0158] In some specific embodiments, the problem conversion module 12 may specifically include:
[0159] A convex quadratic programming problem construction unit is used to construct a convex quadratic programming problem with sparse constraints based on the training set; the convex quadratic programming problem includes a first objective function and a first constraint condition;
[0160] The first objective function is constructed based on half of the sum of squares of the feature weights, the sum of the slack variables of all training samples, the sum of the absolute values of the feature weights, a first hyperparameter for the sum of the slack variables of all training samples, and a second hyperparameter for the sum of the absolute values of the feature weights; the first constraint includes a classification margin constraint and a non-negativity constraint for the slack variables.
[0161] In some specific embodiments, the sliding bearing fault diagnosis model construction device may specifically include:
[0162] The evaluation unit is used to evaluate the sliding bearing fault diagnosis model using a test set according to a preset evaluation index after obtaining the sliding bearing fault decision function based on the optimal sparse weight vector and bias.
[0163] The evaluation score determination unit is used to obtain an evaluation score based on the evaluation results of each of the preset evaluation indicators.
[0164] In some specific embodiments, the sliding bearing fault diagnosis model construction device may specifically include:
[0165] The new training set generation unit is used to generate a new training set based on the new historical operating data of the sliding bearing after obtaining the sliding bearing fault decision function based on the optimal sparse weight vector and bias to obtain the sliding bearing fault diagnosis model.
[0166] The misclassification rate determination unit is used to determine the misclassification rate of the sliding bearing fault diagnosis model on the new training set.
[0167] The model update unit is used to resolve the second-order cone programming problem based on the new training set if the misclassification rate exceeds a preset threshold, so as to update the optimal sparse weight vector and bias, and then update the sliding bearing fault diagnosis model.
[0168] In some specific embodiments, the problem conversion module 12 may specifically include:
[0169] The second objective function determination unit is used to replace the sum of squares of the feature weights in the first objective function with a first auxiliary variable, and replace the sum of the absolute values of the feature weights with a second auxiliary variable to obtain the second objective function;
[0170] The second constraint determination unit is used to take the first auxiliary variable as the upper bound constraint of the sum of squares of the feature weights to obtain the second constraint condition.
[0171] The third constraint determination unit is used to take the second auxiliary variable as the upper bound constraint of the sum of the absolute values of the feature weights, and obtain the third constraint condition.
[0172] The fourth constraint determination unit is used to construct weighted constraint conditions for the feature groups based on the grouping of the multidimensional target features, thereby obtaining the fourth constraint condition;
[0173] The second-order cone programming problem determination unit is used to obtain the second-order cone programming problem based on the second objective function, the first constraint, the second constraint, the third constraint, and the fourth constraint.
[0174] Furthermore, this application also discloses an electronic device, see [link to relevant documentation]. Figure 5 As shown, the content in the figure should not be considered as any limitation on the scope of use of this application.
[0175] Figure 5 This is a schematic diagram of the structure of an electronic device 20 provided in an embodiment of this application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the sliding bearing fault diagnosis model construction method disclosed in any of the foregoing embodiments.
[0176] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.
[0177] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or optical disk, etc. The resources stored on it include operating system 221, computer program 222 and data 223 including training set, etc., and the storage method can be temporary storage or permanent storage.
[0178] The operating system 221 manages and controls the various hardware devices on the electronic device 20 and the computer program 222 to enable the processor 21 to perform calculations and processing on the massive amounts of data 223 in the memory 22. The operating system 221 can be Windows Server, Netware, Unix, Linux, etc. In addition to including a computer program capable of performing the sliding bearing fault diagnosis model construction method executed by the electronic device 20 as disclosed in any of the foregoing embodiments, the computer program 222 may further include computer programs capable of performing other specific tasks.
[0179] Furthermore, this application also discloses a computer storage medium storing computer-executable instructions. When the computer-executable instructions are loaded and executed by a processor, they implement the sliding bearing fault diagnosis model construction method steps disclosed in any of the foregoing embodiments.
[0180] Furthermore, this application also discloses a computer program product, including a computer program that, when executed by a processor, implements the sliding bearing fault diagnosis model construction method steps disclosed in any of the foregoing embodiments.
[0181] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.
[0182] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.
[0183] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0184] The foregoing has provided a detailed description of the method, apparatus, equipment, and storage medium for constructing a sliding bearing fault diagnosis model. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.
Claims
1. A method for constructing a fault diagnosis model for sliding bearings, characterized in that, include: A training set is generated based on the historical operating data of the sliding bearing and the multi-dimensional target features corresponding to the historical operating data. According to the margin maximization strategy, a convex quadratic programming problem for a support vector machine is constructed using the training set, and the convex quadratic programming problem is transformed into a second-order cone programming problem; the convex quadratic programming problem includes a first objective function and a first constraint condition. The second-order cone programming problem is solved using a solver to obtain the optimal sparse weight vector and bias. Based on the optimal sparse weight vector and bias, the sliding bearing fault decision function is obtained to obtain the sliding bearing fault diagnosis model, so as to use the sliding bearing fault diagnosis model for sliding bearing fault diagnosis. The transformation of the convex quadratic programming problem into a second-order cone programming problem includes: The second objective function is obtained by replacing the sum of squares of the feature weights in the first objective function with a first auxiliary variable and replacing the sum of the absolute values of the feature weights with a second auxiliary variable. By using the first auxiliary variable as an upper bound constraint on the sum of squares of the feature weights, a second constraint condition is obtained. By using the second auxiliary variable as an upper bound constraint on the sum of the absolute values of the feature weights, a third constraint condition is obtained. Based on the grouping of the multidimensional target features, weight constraints are constructed for the feature groups to obtain the fourth constraint. Based on the second objective function, the first constraint, the second constraint, the third constraint, and the fourth constraint, the second-order cone programming problem is obtained. The second objective function is: ； Where t is the first auxiliary variable, u is the second auxiliary variable, and C is the first hyperparameter. For the slack variables of training sample n, This is the second hyperparameter; The third constraint is as follows: ； in, The feature weights of feature d are... All are auxiliary variables, with u being the second auxiliary variable.
2. The method for constructing a sliding bearing fault diagnosis model according to claim 1, characterized in that, Based on the historical operating data of the sliding bearing and the corresponding multidimensional target features, a training set is generated, including: Acquire historical operating data of the sliding bearing; the historical operating data includes normal operation data and fault data; the historical operating data includes any one or more of vibration data, impact data, power data, speed data, bearing shell temperature data, and oil data; the vibration data includes any one or more of the vibration acceleration data of the bearing housing, the vibration acceleration data of the machine body, and the vibration displacement data on the shaft; Multidimensional target features are extracted based on the historical operating data; the multidimensional target features include any one or more of vibration features, impact features, operating condition features, bearing temperature features, and oil features; The dataset is obtained based on the historical operating data and the multidimensional target features. The dataset is then divided into a training set and a test set.
3. The method for constructing a sliding bearing fault diagnosis model according to claim 2, characterized in that, Before extracting multidimensional target features based on the historical operational data, the process also includes: The historical operational data is preprocessed; the preprocessing includes outlier detection and processing, and data standardization.
4. The method for constructing a sliding bearing fault diagnosis model according to claim 1, characterized in that, Following the margin maximization strategy, the convex quadratic programming problem of constructing a support vector machine using the training set includes: Construct a convex quadratic programming problem with sparse constraints based on the training set; The first objective function is constructed based on half of the sum of squares of the feature weights, the sum of the slack variables of all training samples, the sum of the absolute values of the feature weights, a first hyperparameter for the sum of the slack variables of all training samples, and a second hyperparameter for the sum of the absolute values of the feature weights; the first constraint includes a classification margin constraint and a non-negativity constraint for the slack variables.
5. The method for constructing a sliding bearing fault diagnosis model according to claim 4, characterized in that, The first objective function is: ； in, The feature weights of feature d are... Let C be the slack variable for the training sample n, and let C be the first hyperparameter. This is the second hyperparameter. This is a sparsity regularization term.
6. The method for constructing a sliding bearing fault diagnosis model according to claim 1, characterized in that, After obtaining the sliding bearing fault decision function based on the optimal sparse weight vector and bias, and thus obtaining the sliding bearing fault diagnosis model, the following steps are also included: The sliding bearing fault diagnosis model is evaluated using a test set based on preset evaluation indicators. An evaluation score is obtained based on the evaluation results of each of the preset evaluation indicators.
7. The method for constructing a sliding bearing fault diagnosis model according to any one of claims 1 to 6, characterized in that, After obtaining the sliding bearing fault decision function based on the optimal sparse weight vector and bias, and thus obtaining the sliding bearing fault diagnosis model, the following steps are also included: A new training set is generated based on the new historical operating data of sliding bearings; Determine the misclassification rate of the sliding bearing fault diagnosis model on the new training set; If the misclassification rate exceeds a preset threshold, the second-order cone programming problem is resolved based on the new training set to update the optimal sparse weight vector and bias, thereby updating the sliding bearing fault diagnosis model.
8. A method for diagnosing sliding bearing faults, characterized in that, include: Obtain real-time operating data of the sliding bearing; The real-time operating data is input into the sliding bearing fault diagnosis model, and the sliding bearing fault diagnosis result is obtained according to the model output; the sliding bearing fault diagnosis model is a model constructed using the sliding bearing fault diagnosis model construction method according to any one of claims 1 to 7.
9. A device for constructing a fault diagnosis model for a sliding bearing, characterized in that, include: The training set generation module is used to generate a training set based on the historical operating data of the sliding bearing and the multi-dimensional target features corresponding to the historical operating data. The problem transformation module is used to construct a convex quadratic programming problem for a support vector machine using the training set according to the margin maximization strategy, and to transform the convex quadratic programming problem into a second-order cone programming problem; the convex quadratic programming problem includes a first objective function and a first constraint condition. The problem-solving module is used to solve the second-order cone programming problem using a solver to obtain the optimal sparse weight vector and bias, and to obtain the sliding bearing fault decision function based on the optimal sparse weight vector and bias, so as to obtain the sliding bearing fault diagnosis model, and to use the sliding bearing fault diagnosis model to diagnose sliding bearing faults. The problem transformation module is used to replace the sum of squares of the feature weights in the first objective function with a first auxiliary variable, and replace the sum of the absolute values of the feature weights with a second auxiliary variable to obtain a second objective function; use the first auxiliary variable as an upper bound constraint on the sum of squares of the feature weights to obtain a second constraint; use the second auxiliary variable as an upper bound constraint on the sum of the absolute values of the feature weights to obtain a third constraint; construct weight constraints for feature groups based on the grouping of the multidimensional objective features to obtain a fourth constraint; and obtain the second-order cone programming problem based on the second objective function, the first constraint, the second constraint, the third constraint, and the fourth constraint. The second objective function is: ； Where t is the first auxiliary variable, u is the second auxiliary variable, and C is the first hyperparameter. For the slack variables of training sample n, This is the second hyperparameter; The third constraint is as follows: ； in, The feature weights of feature d are... All are auxiliary variables, with u being the second auxiliary variable.
10. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor is configured to execute the computer program to implement the sliding bearing fault diagnosis model construction method as described in any one of claims 1 to 7, or the sliding bearing fault diagnosis method as described in claim 8.
11. A computer-readable storage medium, characterized in that, Used to store computer programs; wherein the computer programs, when executed by a processor, implement the sliding bearing fault diagnosis model construction method as described in any one of claims 1 to 7, or the sliding bearing fault diagnosis method as described in claim 8.