Rolling bearing fault diagnosis method based on WOA-VMD-YOLOv8

Abstract

The invention discloses a WOA-VMD-YOLOv8-based rolling bearing fault diagnosis method. The method comprises the steps of data acquisition, data preprocessing, and dividing a fault sample data set into a training set, a test set and a verification set in proportion; a WOA-VMD-YOLOv8 model is built, and the The model comprises a whale optimization algorithm WOA, VMD decomposition and a YOLOv8 model. A WOA-VMD-YOLOv8 model is trained, and a rolling bearing fault diagnosis model based on the WOA-VMD-YOLOv8 is obtained; a sound signal of a to-be-diagnosed rolling bearing is input into the WOA-VMD-YOLOv8-based rolling bearing fault diagnosis model, and then a fault type can be output. According to the rolling bearing fault diagnosis method, the WOA is used for optimizing VMD decomposition, the IMF component is selected by introducing the related kurtosis value and the weighted energy entropy, and the YOLOv8 model is combined, so that the efficiency and accuracy of rolling bearing fault diagnosis are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of rolling bearing fault diagnosis, and specifically relates to a rolling bearing fault diagnosis method based on WOA-VMD-YOLOv8. Background Art

[0002] Rolling bearings are very important components in mechanical equipment. Their main function is to support the mechanical rotating body, reduce the friction coefficient during the movement process, and ensure its rotational accuracy. The health state of rolling bearings has a huge impact on the performance, stability, and service life of the entire mechanical equipment. With the continuous improvement of China's industrialization level, mechanical equipment is more and more widely used in various industries. Due to the harshness and complexity of the working environment, rolling bearings are one of the most easily damaged components in mechanical parts; at the same time, since it is difficult to directly observe the faults of rolling bearings, it is also very difficult to judge whether their operating state is normal or faulty through simple means.

[0003] Currently, the industrial and academic circles generally use vibration signal-based methods to indirectly evaluate the operating state of rolling bearings. Common signal analysis methods include envelope demodulation, wavelet transform, empirical mode decomposition, etc. However, wavelet transform depends on the selection of wavelet bases, EMD lacks a mathematical basis, and there are mode mixing and end effects. When the characteristic signal is relatively weak, the fault diagnosis effect is not obvious. In 2014, K. Dragomiretskiy et al. proposed the variational mode decomposition algorithm. Once VMD was proposed, it was widely used in the field of mechanical fault diagnosis. VMD has a strict mathematical theory basis and has strong advantages in the extraction of nonlinear and non-stationary signal characteristics. However, two parameters of VMD need to be manually selected, and inappropriate selection will seriously affect the effect of the algorithm.

[0004] Currently, a large number of studies have been carried out on the parameter optimization problem of VMD. In 2018, Jiang et al. used the energy fluctuation spectrum to quickly determine the initial center frequency of potential modes, and then constructed an optimization strategy guided by the center frequency to adaptively optimize the penalty factor to extract the optimal signal components. In 2019, Jiang et al. proposed a coarse-to-fine VMD decomposition strategy. First, the decomposition layer number is initialized to 1, and by iteratively using VMD on the signal, the components with characteristic signals are selected. Then, by adjusting the size of the penalty factor, the frequency components of the characteristic signal become more obvious. However, neither of these two methods can determine the best parameter combination of VMD.

[0005] In order to directly determine the optimal parameters of VMD and automatically detect bearing faults. In 2022, He et al. used an improved sparrow search algorithm to optimize the VMD parameters and used a reverse residual convolutional neural network for fault diagnosis, with an identification accuracy rate reaching 97.5%

[15] . Lin et al. used the cuckoo search algorithm to optimize the VMD parameters and used a probabilistic neural network for fault identification, with an identification accuracy rate that could reach 98.5%

[16] . However, both the sparrow search algorithm and the cuckoo search algorithm do not have good global search capabilities and are prone to falling into local optima. In addition, the inference time of the neural network model is relatively long and cannot meet the requirements of real-time fault diagnosis. The whale optimization algorithm (WOA) has a simple model and good global optimization performance. Mao et al. used WOA to optimize the VMD parameters and used the Hausdorff distance to select appropriate components, and simulations and experiments have proven the effectiveness of the proposed method. In 2024, Liu et al. used the YOLOv8 model to detect bearing defects, significantly improving the detection efficiency and accuracy. However, how to combine VMD with the YOLOv8 model to further improve the accuracy has not been made public. Summary of the Invention

[0006] 1. Technical problems to be solved:

[0007] In view of the above technical problems, the present invention provides a rolling bearing fault diagnosis method based on WOA-VMD-YOLOv8, which uses WOA to optimize VMD decomposition, selects IMF components by introducing relevant kurtosis values and weighted energy entropy, and combines with the YOLOv8 model to improve the detection efficiency and accuracy.

[0008] 2. Technical solution:

[0009] A rolling bearing fault diagnosis method based on WOA-VMD-YOLOv8, characterized by comprising the following steps:

[0010] Step 1: Data acquisition; collect the original acoustic signals during the operation of rolling bearings under different fault conditions.

[0011] Step 2: Data preprocessing; preprocess the collected original acoustic signals using a band-pass filter to remove part of the noise and obtain a fault sample data set.

[0012] Step 3: Divide the fault sample data set into a training set, a test set and a validation set according to a certain proportion.

[0013] Step 4: Build a WOA-VMD-YOLOv8 model; this model includes the whale optimization algorithm WOA, VMD decomposition and the YOLOv8 model.

[0014] Among them, the VMD decomposition decomposes the acoustic signal into K IMF components, calculates the correlation kurtosis value and weighted energy entropy of each IMF component, sets the thresholds of the correlation kurtosis value and weighted energy entropy. When the results of the correlation kurtosis value and weighted energy entropy of the IMF component are both greater than the set thresholds, it is determined that the IMF component contains fault feature information; add all the IMF components that meet the requirements of fault feature information to obtain a reconstructed signal, and draw the time-frequency image of the reconstructed signal envelope spectrum; input the time-frequency image into the YOLOv8 model for feature processing and fault diagnosis fitting to obtain fault information; the whale optimization algorithm WOA optimizes the parameters in the VMD model to improve the accuracy of VMD decomposition;

[0015] Step 5: Train the WOA-VMD-YOLOv8 model to obtain a trained rolling bearing fault diagnosis model based on WOA-VMD-YOLOv8;

[0016] Step 6: Input the acoustic signal of the rolling bearing to be diagnosed into the rolling bearing fault diagnosis model based on WOA-VMD-YOLOv8, and the fault type can be output.

[0017] Further, in Step 1, different fault conditions include inner ring fault, outer ring fault, ball fault, and the corresponding normal working state of the bearing.

[0018] Further, the whale optimization algorithm WOA optimizes the parameters of VMD by simulating the foraging behavior of whales until the optimal parameter combination is found; the parameters of VMD include the decomposition layer number and the penalty factor.

[0019] Further, in the WOA-VMD-YOLOv8 model, the VMD optimized by WOA decomposes the fault sample data into K IMF components, and calculates the correlation kurtosis value and weighted energy entropy of each IMF respectively; the specific process includes:

[0020] Calculate the correlation kurtosis value CK of each IMF as follows:

[0021]

[0022] In the above formula, N is the data length of the sample, that is, the sample consists of N data; τ is the preset time delay; IMF i (k) represents the i-th IMF component;

[0023] Calculate the sum of the energy entropy E i of all IMF components obtained by WOA-VMD decomposition:

[0024]

[0025] Calculate the proportion of the i-th IMF component in the total energy entropy as follows:

[0026]

[0027] Among them, K is the total number of IMF components;

[0028] Introduce the weighted energy entropy H we Enhance the fault sensitivity, the weighted energy entropy H we As follows:

[0029]

[0030] In the above formula, f i is the center frequency of the IMF component; f c is the fault characteristic frequency; P i is the energy proportion of the i-th IMF component.

[0031] Furthermore, the obtaining of the reconstructed signal specifically includes:

[0032] Obtain the mean value μ Hwe and the standard deviation σ Hwe of the weighted energy entropy when the sample has no fault;

[0033] For the weighted energy entropy calculated by formula (4), select the IMF components with |H we -μ Hwe |>3σ Hwe and the relevant kurtosis value greater than 5, and add them up to obtain the reconstructed signal; draw the time-frequency image of the reconstructed signal envelope spectrum.

[0034] Furthermore, the YOLOv8 model inputs the time-frequency image and outputs the bearing fault detection result.

[0035] 3. Beneficial effects:

[0036] (1) A rolling bearing fault diagnosis method based on WOA-VMD-YOLOv8 disclosed by the present invention can adaptively decompose the fault sample data into K IMF components by using VMD optimized by WOA, solves the problem of VMD parameter selection, and effectively avoids the generation of false components and mode mixing.

[0037] (2) A rolling bearing fault diagnosis method based on WOA-VMD-YOLOv8 disclosed by the present invention. Based on the fact that different faults of rolling bearings will cause energy concentration in specific frequency bands, the energy proportion of fault characteristic signals will be relatively large. At the same time, rolling bearing faults will generate periodic impacts, and the kurtosis value will have a sudden change in the data change characteristics. By calculating the relevant kurtosis values and weighted energy entropies of each IMF component, setting the thresholds of the relevant kurtosis values and weighted energy entropies, when the results of the relevant kurtosis values and weighted energy entropies of the IMF component are both greater than the set thresholds, it is determined that the IMF component contains fault characteristic information; add all the IMF components that meet the requirements of fault characteristic information to obtain a reconstructed signal, and draw the time-frequency image of the reconstructed signal envelope spectrum; input the time-frequency image into the YOLOv8 model for feature processing and fault diagnosis fitting to obtain fault information; the whale optimization algorithm WOA optimizes the parameters in the VMD model to improve the accuracy of VMD decomposition.

[0038] (3) A rolling bearing fault diagnosis method based on WOA-VMD-YOLOv8 disclosed by the present invention. By comparing with the WT-YOLOv8 model, the fault diagnosis effect of WOA-VMD-YOLOv8 for rolling bearings is better. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 is the flow chart of the present invention;

[0040] Figure 2 is the envelope spectrum diagram of the signals output by the WT-YOLOv8 model (a) and the method adopted in the verification example (b);

[0041] Figure 3 is the time-frequency image of the envelope spectrum of the reconstructed signal drawn by the method in the verification example. DETAILED DESCRIPTION OF THE INVENTION

[0042] The present invention will be specifically described below with reference to the accompanying drawings.

[0043] As shown in the Figure 1 accompanying drawings, a rolling bearing fault diagnosis method based on WOA-VMD-YOLOv8 is characterized by including the following steps:

[0044] Step 1: Data acquisition; collect the original acoustic signals when the rolling bearing is working under different fault conditions;

[0045] Step 2: Data preprocessing; use a band-pass filter to preprocess the collected original acoustic signals to remove part of the noise and obtain a fault sample data set;

[0046] Step 3: Divide the fault sample data set into a training set, a test set and a validation set according to a ratio;

[0047] Step 4: Build the WOA-VMD-YOLOv8 model; this model includes the Whale Optimization Algorithm (WOA), VMD decomposition, and the YOLOv8 model;

[0048] Among them, VMD decomposition decomposes the acoustic signal into K IMF components, calculates the correlation kurtosis value and weighted energy entropy of each IMF component, sets the thresholds of the correlation kurtosis value and weighted energy entropy. When the results of the correlation kurtosis value and weighted energy entropy of the IMF component are both greater than the set thresholds, it is determined that the IMF component contains fault feature information; add all the IMF components that meet the requirements of fault feature information to obtain a reconstructed signal, and draw the time-frequency image of the reconstructed signal envelope spectrum; input the time-frequency image into the YOLOv8 model for feature processing and fault diagnosis fitting to obtain fault information; the Whale Optimization Algorithm (WOA) optimizes the parameters in the VMD model to improve the accuracy of VMD decomposition;

[0049] Step 5: Train the WOA-VMD-YOLOv8 model to obtain a trained rolling bearing fault diagnosis model based on WOA-VMD-YOLOv8;

[0050] Step 6: Input the acoustic signal of the rolling bearing to be diagnosed into the rolling bearing fault diagnosis model based on WOA-VMD-YOLOv8, and the fault type can be output.

[0051] Further, in Step 1, different fault conditions include inner race fault, outer race fault, ball fault, and the corresponding normal working state of the bearing.

[0052] Further, the Whale Optimization Algorithm (WOA) optimizes the parameters of VMD by simulating the predation behavior of whales until the optimal parameter combination is found; the parameters of VMD include the decomposition layer number and the penalty factor.

[0053] Further, in the WOA-VMD-YOLOv8 model, the VMD optimized by WOA decomposes the fault sample data into K IMF components, and calculates the correlation kurtosis value and weighted energy entropy of each IMF respectively; the specific process includes:

[0054] Calculate the correlation kurtosis value CK of each IMF as follows:

[0055]

[0056] In the above formula, N is the data length of the sample, that is, the sample consists of N data; τ is the preset time delay; IMF i (k) represents the i-th IMF component;

[0057] Calculate the sum of the energy entropy E i of all IMF components obtained by WOA-VMD decomposition:

[0058]

[0059] The proportion of the i-th IMF component in the total energy entropy is calculated as follows:

[0060]

[0061] where K is the total number of IMF components;

[0062] Introduce the weighted energy entropy H we Enhance the fault sensitivity, the weighted energy entropy H we is as follows:

[0063]

[0064] In the above formula, f i is the center frequency of the IMF component; f c is the fault characteristic frequency; P i is the energy proportion of the i-th IMF component.

[0065] Furthermore, the obtaining of the reconstructed signal specifically includes:

[0066] Obtain the mean value μ Hwe and the standard deviation σ Hwe ;

[0067] For the weighted energy entropy calculated by formula (4), select the IMF components with |H we -μ Hwe |>3σ Hwe and the relevant kurtosis value greater than 5, and add them to obtain the reconstructed signal; draw the time-frequency image of the envelope spectrum of the reconstructed signal.

[0068] Furthermore, the YOLOv8 model inputs the time-frequency image and outputs the bearing fault detection result.

[0069] Verification example:

[0070] In this verification example, the WT-YOLOv8 model (hereinafter referred to as WT) composed of wavelet transform WT and YOLOv8 is compared with the model constructed by this method.

[0071] Input the same fault signal to be detected into the WT model and the WOA-VMD of this method respectively, and the two respectively output the signals of the envelope spectrum shown in the appendix Figure 2 ; It can be clearly seen from the figure that the output signals of WT and WOA-VMD both have the characteristic frequency 162Hz and its second harmonic. In addition, compared with WT, there is less noise in the output signal of WOA-VMD. In addition, compared with WT, there is less noise in the output signal of WOA-VMD.

[0072] Using this method, the obtained envelope spectrum is used to generate a time-frequency image, as shown in the appendix Figure 3 As shown, it can be clearly seen from the figure that there are periodic characteristic signals in the time-frequency image. According to the characteristics of the bearing fault characteristic frequency, Figure 3 it can be judged that the bearing has an inner ring fault. The generated time-frequency image is input into the YOLOv8 model, and the corresponding fault label can be output by inputting the time-frequency image.

[0073] The following table shows the comparison of the performance indicators for the fault judgment of rolling bearings using the WT-YOLOv8 model and the WOA-VMD-YOLOv8 model of this method:

[0074] Table 1. Comparison of performance indicators of WOA-VMD-YOLOv8 and WT-YOLOv8

[0075]

[0076] Table 1 shows the comparison of the performance indicators of WOA-VMD-YOLOv8 and WT-YOLOv8. After training, the recall rate of the WOA-VMD-YOLOv8 model is 99.7%, mAP@0.5 is 99.5%, and mAP@0.5-0.95 is 99.4%. The performance indicators of the diagnostic results of this method are all better than those of WT-YOLOv8.

[0077] Although the present invention has been disclosed above with preferred embodiments, they are not used to limit the present invention. Any person skilled in this art can make various changes or modifications without departing from the spirit and scope of the present invention. Therefore, the protection scope of the present invention should be defined by the protection scope of the claims of this application.

Claims

1. A rolling bearing fault diagnosis method based on WOA-VMD-YOLOv8, characterized in that: It includes the following steps: Step 1: Data acquisition; Collect the original acoustic signals during the operation of rolling bearings under different fault conditions; Step 2: Data preprocessing; Preprocess the collected original acoustic signals using a band-pass filter to remove some noise and obtain a fault sample data set; Step 3: Divide the fault sample data set into a training set, a test set, and a validation set according to a certain proportion; Step 4: Build a WOA-VMD-YOLOv8 model; this model includes the Whale Optimization Algorithm (WOA), VMD decomposition, and the YOLOv8 model; Among them, VMD decomposition decomposes the acoustic signal into K IMF components, calculates the correlation kurtosis value and weighted energy entropy of each IMF component, sets the thresholds of the correlation kurtosis value and weighted energy entropy. When the results of the correlation kurtosis value and weighted energy entropy of the IMF component are both greater than the set thresholds, it is determined that this IMF component contains fault feature information; add all the IMF components that meet the requirements of fault feature information to obtain a reconstructed signal, and draw the time-frequency image of the reconstructed signal envelope spectrum; input the time-frequency image into the YOLOv8 model for feature processing and fault diagnosis fitting to obtain fault information; the Whale Optimization Algorithm (WOA) optimizes the parameters in the VMD model to improve the accuracy of VMD decomposition; Step 5: Train the WOA-VMD-YOLOv8 model to obtain a trained rolling bearing fault diagnosis model based on WOA-VMD-YOLOv8; Step 6: Input the acoustic signal of the rolling bearing to be diagnosed into the rolling bearing fault diagnosis model based on WOA-VMD-YOLOv8, and the fault type can be output.

2. A rolling bearing fault diagnosis method based on WOA-VMD-YOLOv8 according to claim 1, characterized in that: In Step 1, different fault conditions include inner race fault, outer race fault, ball fault, and the corresponding normal working state of the bearing.

3. A rolling bearing fault diagnosis method based on WOA-VMD-YOLOv8 according to claim 1, characterized in that: The Whale Optimization Algorithm (WOA) optimizes the parameters of VMD by simulating the foraging behavior of whales until the optimal parameter combination is found; the parameters of VMD include the decomposition layer number and the penalty factor.

4. A rolling bearing fault diagnosis method based on WOA-VMD-YOLOv8 according to claim 1, characterized in that: In the WOA-VMD-YOLOv8 model, the VMD optimized by WOA decomposes the fault sample data into K IMF components, and calculates the correlation kurtosis value and weighted energy entropy of each IMF respectively; the specific process includes: Calculate the correlation kurtosis value CK of each IMF as follows: In the above formula, N is the data length of the sample, that is, the sample consists of N data; τ is the preset time delay; IMF i (k) represents the i-th IMF component; Calculate the sum of the energy entropy E of all IMF components obtained by WOA-VMD decomposition i and: Calculate the proportion of the i-th IMF component in the total energy entropy as follows: where K is the total number of IMF components; Introduce the weighted energy entropy H we Enhance the fault sensitivity, the weighted energy entropy H we As shown in the following formula: In the above formula, f i is the central frequency of the IMF component; f c is the fault characteristic frequency; P i is the energy proportion of the i-th IMF component.

5. A rolling bearing fault diagnosis method based on WOA-VMD-YOLOv8 according to claim 4, characterized in that: The specific process of obtaining the reconstructed signal includes: Obtain the mean value μ of the weighted energy entropy when the sample has no failure Hwe and the standard deviation σ Hwe ; Select the weighted energy entropy calculated by formula (4) where |H we - μ Hwe | > 3σ Hwe and add the IMF components with a relevant kurtosis value greater than 5 to obtain a reconstructed signal; plot the time-frequency image of the envelope spectrum of the reconstructed signal.

6. A rolling bearing fault diagnosis method based on WOA-VMD-YOLOv8 according to claim 1, characterized in that: The YOLOv8 model inputs the time-frequency image and outputs the bearing fault detection result.

Images