A bearing fault classification method based on snake optimization support vector machine with fault feature selection

Abstract

The application relates to a bearing fault classification method, in particular to a bearing fault classification method of a snake optimization support vector machine with fault feature selection. The method can quickly and accurately and efficiently complete bearing fault classification under the input of a small amount of fault features. The method comprises a data processing module, a feature selection module, a model parameter optimization module and a fault classification module. The data processing module divides the collected original data, and then converts each group of data into a time domain signal index through calculation, serving as input data of fault features of the model. The feature selection module adopts a decision tree to select optimal fault features, and the decision tree divides the root node based on a division attribute, taking the division attribute as the optimal fault feature of the model. The model parameter optimization module combines a snake optimization algorithm and a support vector machine, improves the classification accuracy and classification time of the model. The fault classification module divides bearing faults into four forms and accurately locates the fault state.

Description

Technical Field

[0001] This invention belongs to the field of fault diagnosis in the machinery manufacturing industry, and more specifically, relates to a bearing fault classification method using a snake-optimized support vector machine for fault feature selection. Background Technology

[0002] Bearings are important components in mechanical equipment, with a high failure rate and easy damage. If fault characteristic information can be effectively extracted in the early stage of failure, the bearing's operating status can be accurately identified, and damaged bearings can be replaced or repaired in a timely manner, which can effectively avoid the occurrence of chain failures and is of great significance for reducing economic losses.

[0003] Bearing fault diagnosis technology is mainly divided into quantitative analysis-based methods and qualitative analysis-based methods. Notably, data-driven machine learning methods within the latter are widely used in fault diagnosis. These mainly include decision trees, neural networks, and random forests. However, in real-world scenarios, data acquisition is difficult, and equipment often operates under normal conditions, resulting in scarce effective data, thus limiting the application of these methods. Support vector machines (SVMs) can effectively solve fault classification problems with small sample data. However, the classification performance of SVMs is largely dependent on the fault features and the selection of its own model parameters. The input fault features for SVM methods are generally subjectively selected by fault diagnostic personnel; therefore, the subjectively selected features largely determine the performance ceiling of the SVM method. Regarding parameter selection, some methods employ swarm intelligence optimization algorithms such as particle swarm optimization, genetic algorithms, and whale optimization algorithms to optimize SVMs. While these algorithms have certain advantages, they also suffer from drawbacks such as slow optimization speed and susceptibility to getting trapped in local optima.

[0004] In summary, identifying appropriate fault characteristics and model parameters is crucial for classifying bearing faults in an SVM classification model. Summary of the Invention

[0005] To address the classification problem of bearing faults, this invention proposes a bearing fault classification method based on snake-optimized support vector machines for fault feature selection. This method uses the partitioning attributes selected by decision trees through the Gini index as fault features, avoiding the subjective will of fault personnel. At the same time, the selected optimization method has a fast optimization speed and is not prone to getting trapped in local optima.

[0006] The objective of this invention is achieved through the following technical solution: This method employs a modular approach to classify bearing faults, primarily comprising a data processing module, a feature selection module, a model parameter optimization module, and a fault classification module. The data processing module divides the collected raw data into units of 1024, then converts each group of data into time-domain signal indicators, which serve as input data for the model's fault features. The feature selection module uses a decision tree to select the optimal fault features. The decision tree partitions the root node based on the partitioning attribute, and this partitioning attribute is used as the model's optimal fault features. The model parameter optimization module combines the snake optimization algorithm and support vector machine to improve the model's classification accuracy and time. The fault classification module classifies bearing faults into four types: normal state, inner race fault, rolling element fault, and outer race fault, and displays the classification results using visual graphics.

[0007] A bearing fault classification method using a snake-optimized support vector machine for fault feature selection includes the following steps:

[0008] S1: Set the evaluation index for the decision tree splitting attribute; perform decision tree classification;

[0009] S2: Use the classification attribute index as the optimal fault feature;

[0010] S3: Initialize the environment model, initialize the number of snakes in the snake population, the number of iterations, and the initial population solution;

[0011] S4: Set up separate female and male snake groups to divide the population into female and male groups;

[0012] S5: Set initial fitness values, optimal female individuals, and optimal male individuals;

[0013] S6: Set the temperature to 0.6 degrees Celsius and the food mass (Q) to 0.25 kg.

[0014] S7: Determine if the food mass (Q) is less than 0.25. If yes, proceed to step S8; otherwise, proceed to step S9.

[0015] S8: Finding Food

[0016] S9: Determine if the temperature T is greater than 0.6. If yes, proceed to S10; otherwise, proceed to S11.

[0017] S10: Move closer to the food and update the position according to the food's location; if it is...

[0018] S11: Determine if the temperature change range is less than 0.6. If yes, proceed to S12; otherwise, proceed to S13.

[0019] S12: Battle Mode;

[0020] S13: Mating pattern:

[0021] S14: Determine if the maximum number of iterations has been reached. If the maximum number of iterations has been reached, proceed to S15; otherwise, proceed to the next iteration and jump to S5.

[0022] S15: Output the optimal penalty parameter (c) and kernel parameter (g).

[0023] S16: Training the Support Vector Machine Classification Model

[0024] S17: Determine if the classification accuracy has reached its maximum. If it has, proceed to S18; otherwise, jump to S3.

[0025] S18: Output the classification results.

[0026] Due to the adoption of the above technical solution, the present invention has the following beneficial effects:

[0027] This invention proposes a bearing fault classification method using a snake-optimized support vector machine (SVM) for fault feature selection. Representative fault features are selected as input data through a decision tree, while the SVM parameters are optimized using a snake optimization algorithm. The model can quickly and accurately classify bearing faults. The proposed method has good reference and application value for bearing fault detection and can improve the work efficiency of maintenance personnel. Attached Figure Description

[0028] Figure 1 This is a schematic diagram of the overall method of the present invention;

[0029] Figure 2 This is the decision tree classification process of the present invention;

[0030] Figure 3 The process of selecting evaluation and partitioning attribute indicators for decision trees;

[0031] Figure 4 To select the optimal fault feature results;

[0032] Figure 5 Flowchart of the optimization parameters of this invention;

[0033] Figure 6 Comparison chart of the invented DT-SO-SVM, WOA-SVM, MFO-SVM, PSO-SVM, GA-SVM and GridSearch-svm algorithms. Detailed Implementation

[0034] The invention will now be described in further detail with reference to the accompanying drawings. The invention proposes a bearing fault classification method using a snake-optimized support vector machine for fault feature selection, comprising four parts: a data processing module, a feature selection module, a model parameter optimization module, and a fault classification module. Figure 1 As shown.

[0035] The data processing module divides the raw data collected by the sensors installed on the bearing into four states: normal state, inner ring fault, outer ring fault, and rolling element fault. These four states are numbered 1, 2, 3, and 4 respectively. The data is truncated using 1024 as a unit dimension, and the truncated data is converted and calculated into 6-dimensional time-domain signal indices according to formulas (1)-(6), namely waveform index, peak index, pulse index, margin index, skewness index, and kurtosis index.

[0036] Waveform Indicators:

[0037] Peak metrics:

[0038] Pulse Indicators:

[0039] Margin indicators:

[0040] Skewness index;

[0041] Kurtosis index:

[0042] The feature selection module uses the splitting attributes of the decision tree as the selected fault features. Figure 2 This is a visualization of a decision tree classification system. Three metrics are used to determine the splitting attribute: information gain, gain ratio, and Gini index. After ten experiments, we selected the Gini index as the splitting attribute metric. Figure 3 The purity of sample set D can be measured using the Gini coefficient.

[0043]

[0044] The Gini index of attribute α is defined as:

[0045] Therefore, among the candidate attributes, the attribute that minimizes the Gini index after partitioning is selected as the optimal attribute. Simultaneously, ten experiments are conducted using a decision tree based on the Gini index, such as... Figure 4 The most frequently used indicators are identified as the optimal fault characteristics, and the selection results are waveform indicators, peak indicators, and skewness indicators.

[0046] The model parameter optimization module, such as Figure 5This paper presents the Snake Optimization Algorithm (SO), an optimization algorithm based on the foraging and reproductive behaviors of snakes. The mating of male and female snakes is influenced by several factors. Snakes mate in late spring and early summer when temperatures are low. However, the mating process depends not only on temperature but also on food availability. If temperatures are low and food is plentiful, male snakes will fight each other to attract the attention of females. Females have the right to decide whether to mate. If mating occurs, the female begins to lay eggs in her nest or burrow, and once the eggs are laid, she leaves. The Snake Optimization Algorithm uses the classification accuracy of the Support Vector Machine as the objective function, and the penalty parameter (c) and kernel parameter (g) as variables.

[0047] When food is unavailable, snakes search for food by choosing arbitrary, random locations and updating their positions as follows:

[0048] Male update location: X i,m (t+1)=X rand,m (t)±c2×A m ×((X max -X min )×rand+X min (9)

[0049] Where X i,m Let X be the position of the i-th male. rand,m The position of the random male, rand is a random number between 0 and 1, A m The formula for calculating the male's ability to find food is as follows:

[0050]

[0051] Where f rand,m For males in the X group rand,m Individual fitness, f i,m Let c2 be the fitness of the i-th male individual, and c2 = 0.05 be a constant.

[0052] Female update location: X i,f =X rand,f (t+1)±c2×A f ×((X max -X min )×rand+X min (11)

[0053] Where X i,f Let X be the position of the i-th female. rand,f The position of the random female, rand is a random number between 0 and 1, A f The ability to find food for females is calculated using the following formula:

[0054] In the formula, f rand,f For X rand,f fitness, f i,f Let be the fitness of the i-th individual in the female population.

[0055] If there is food available, update the location based on the location of the food.

[0056] If food > 0.25

[0057] If the temperature is greater than 0.6

[0058] Snakes will approach their prey and update their location:

[0059] X i,j (t+1)=X food ±c3×Temp×rand×(X food -X i,j (t)) (13)

[0060] Where X i,j It refers to the location of an individual (female or male), X food It is the position of the best individual (the best individual in the initial population), c3 = 2.

[0061] The system will switch between combat mode and mating mode based on temperature changes.

[0062] If the temperature variation range is <0.6

[0063] Battle Mode:

[0064] X i,m (t+1)=X i,m (t)±c3×FM×rand×(X best,f -X i,m (t)) (14)

[0065] Where X i,m X represents the location of the male individual. best,f The position of the best individual in the female group; FM represents the combat ability of the male individual.

[0066] X i,f (t+1)=X i,f (t+1)±c3×FF×rand×(X best,m -X i,F (t+1)) (15)

[0067] Where X i,f X represents the position of the i-th female. best,m This indicates the position of the best individual in the female sex group, and FF represents the female's combat ability.

[0068] FM and FF can be calculated using the following formula:

[0069]

[0070]

[0071] Where f best,f For optimal fitness in the female population, f best,m For optimal fitness in male populations, f i For individual fitness.

[0072] Otherwise, enter mating mode:

[0073] X i,m (t+1)=X i.m (t)±c3×M m ×rand×(Q×X i,f (t)-X i,m (t)) (18)

[0074] X i,f (t+1)=X i,f (t)±c3×M f ×rand×(Q×X i,m (t)-X i,f (t)) (19)

[0075] Where X i,f Let X be the position of the i-th individual in the female group. i,m Let M be the position of the i-th individual in the male population. m &M f The mating ability of males and females are calculated using the following formulas:

[0076]

[0077]

[0078] If the eggs hatch, select the worst male and female and replace them.

[0079] X worst,m =X min +rand×(X max -X min ) (twenty two)

[0080] X worst,f =X min +rand×(X max -X min ) (twenty three)

[0081] Among them, X worst,mIt is the worst individual in the male population, X worst,f It is the worst individual in the female population.

[0082] The fault classification module employs a Support Vector Machine (SVM) model. The SVM takes fault features selected using a decision tree as input, with parameters as variables for the classification function (so). The model is trained to obtain classification accuracy, which is then used as the so target function. Iterative training continues until the target function value is met, outputting model parameters. Sample training is then performed to output classification results. Simultaneously, comparisons are made with DT-SO-SVM, WOA-SVM, MFO-SVM, PSO-SVM, GA-SVM, and GridSearch-SVM. The results are as follows: Figure 6 As shown.

[0083] It should be emphasized that the embodiments described in this invention are illustrative rather than limiting. Therefore, this invention includes, but is not limited to, the embodiments described in the specific implementation. Any other implementations derived by those skilled in the art based on the technical solutions of this invention are also within the scope of protection of this invention.

Claims

1. A bearing fault classification method using a snake-optimized support vector machine for fault feature selection, characterized in that... It consists of four modules: data processing module, feature selection module, model parameter optimization module, and fault classification module. First, the data processing module divides the collected raw data into groups of 1024. Then, each group of data is converted into a time-domain signal index through calculation, which serves as the fault feature input data for the model. Secondly, the feature selection module uses a decision tree to select the optimal fault features. The decision tree divides the root node based on the partitioning attribute, and the partitioning attribute is used as the optimal fault feature of the model. Then, the model parameter optimization module combines the snake optimization algorithm and support vector machine to improve the classification accuracy and classification time of the model. Finally, the fault classification module classifies bearing faults into four types: normal state, inner ring fault, rolling element fault, and outer ring fault, and uses visual graphics to display the classification results. The feature selection module uses the splitting attributes of the decision tree as the selected fault features. Three metrics are used to select the splitting attributes: information gain, gain ratio, and Gini index. After ten experiments, the Gini index was selected as the splitting attribute metric. The purity of the sample set D can be measured using the Gini value. The Gini index of attribute α is defined as: Therefore, among the candidate attributes, the attribute that minimizes the Gini index after partitioning is selected as the optimal attribute. Ten experiments are conducted using a decision tree based on the Gini index to partition the most frequently used index as the optimal fault feature.

2. The bearing fault classification method using snake-optimized support vector machine for fault feature selection according to claim 1, characterized in that... The data processing module divides the raw data collected by the sensor installed on the bearing into four states: normal state, inner ring fault, outer ring fault and rolling element fault. The four states are numbered 1, 2, 3 and 4 respectively. The data is truncated with 1024 as a unit dimension. The truncated data is converted and calculated into 6-dimensional time domain signal indicators according to formulas (1)-(6), namely waveform indicator, peak indicator, pulse indicator, margin indicator, skewness indicator and kurtosis indicator. Waveform Indicators: Peak metrics: Pulse Indicators: Margin indicators: Skewness index: Kurtosis index:

3. The bearing fault classification method using snake-optimized support vector machine for fault feature selection according to claim 1, characterized in that... The model parameter optimization module employs the Snake Optimization Algorithm (SO), an optimization algorithm proposed based on the foraging and reproductive behaviors of snakes. Mating between male and female snakes is influenced by temperature and food availability. Snakes mate in late spring and early summer when temperatures are low, but the mating process depends not only on temperature but also on food availability. If the temperature is sufficient, male snakes will fight to attract the attention of females, who have the right to decide whether to mate. If mating occurs, the female begins to lay eggs in her nest or burrow, and once the eggs are laid, she leaves. The Snake Optimization Algorithm uses the classification accuracy of the Support Vector Machine as the objective function, and the penalty parameter c and kernel parameter g as variables. When food is unavailable, snakes search for food by choosing arbitrary, random locations and updating their positions as follows: Male update location: X i,m (t+1)=X rand,m (t)±c2×A m ×((X max -X min )×rand+X min (9); Where X i,m Let X be the position of the i-th male. rand,m The position of the random male, rand is a random number between 0 and 1, A m The formula for calculating the male's ability to find food is as follows: Among them, f rand,m For males in the X group rand,m Individual fitness, f i,m Let c2 be the fitness of the i-th male individual, and c2 = 0.05 be a constant. Female update location: X i,f =X rand,f (t+1)±c2×A f ×((X max -X min )×rand+X min (11); Among them, X i,f Let X be the position of the i-th female. rand,f The position of the random female, rand is a random number between 0 and 1, A f The ability to find food for females is calculated using the following formula: In the formula, f rand,f For X rand,f fitness, f i,f Let be the fitness of the i-th individual in the female population; If food is available, update the location based on the location of the food. If food > 0.25; If the temperature is greater than 0.6; Snakes will approach their prey and update their location: X i,j (t+1)=X food ±c3×Temp×rand×(X food -X i,j (t)) (13); Among them, X i,j It is the location of an individual, X food The optimal position for the individual is c3 = 2; The mode will be either combat mode or mating mode depending on the temperature. If the temperature variation range is <0.6; Battle Mode: X i,m (t+1)=X i,m (t)±c3×FM×rand×(X best,f -X i,m (t)) (14); Among them, X i,m X represents the location of the male individual. best,f The position of the best individual in the female group; FM represents the combat ability of the male individual. X i,f (t+1)=X i,f (t+1)±c3×FF×rand×(X best,m -X i,F (t+1)) (15); Among them, X i,f X represents the position of the i-th female. best,m This indicates the position of the best individual in the male group, while FF indicates the female's combat ability. FM and FF are calculated using the following formula: Among them, f best,f For optimal fitness in the female population, f best,m For optimal fitness in male populations, f i For individual fitness; Otherwise, enter mating mode: X i,m (t+1)=X i.m (t)±c3×M m ×rand×(Q×X i,f (t)-X i,m (t)) (18); X i,f (t+1)=X i,f (t)±c3×M f ×rand×(Q×X i,m (t)-X i,f (t)) (19); Among them, X i,f Let X be the position of the i-th individual in the female group. i,m Let M be the position of the i-th individual in the male population. m &M f The mating ability of males and females are calculated using the following formulas: If the eggs hatch, select the worst males and females and replace them; X worst,m =X min +rand×(X max -X min ) (22); X worst,f =X min +rand×(X max -X min ) (23); Among them, X worst,m It is the worst individual in the male population, X worst,f It is the worst individual in the female population.

4. The bearing fault classification method using snake-optimized support vector machine for fault feature selection according to claim 1, characterized in that... The fault classification module adopts a support vector machine classification model. The support vector machine takes the fault features selected by the decision tree as input, the parameters as variables of so, trains the model to obtain the classification accuracy, and uses it as the objective function of so. Iterative training is carried out until the objective function value is met, outputs the model parameters, performs sample training, and outputs the classification results, which are four results: normal state, inner ring fault, outer ring fault and rolling element fault.

Images