GA-BP rolling bearing fault diagnosis method based on electrostatic signals

Abstract

The invention relates to the field of mechanical equipment fault diagnosis, and discloses a GA-BP rolling bearing fault diagnosis method based on electrostatic signals. The method comprises the following steps: firstly, acquiring an amplified rolling bearing electrostatic signal, and decomposing the electrostatic signal into a plurality of intrinsic mode function components (IMF) by adopting a complete ensemble empirical mode decomposition method added with adaptive white noise; then, selecting the first eight IMF components, and constructing a normalized feature vector based on the energy features of the first eight IMF components; and performing normalization processing and training set and test set division on the feature vectors, and inputting the feature vectors to a BP neural network model. And BP neural network weight and threshold parameters are optimized through a genetic algorithm, the global optimization ability of network training is improved, and finally intelligent recognition of rolling bearing fault types is realized. According to the method, the bearing operation state information in the electrostatic signals can be effectively extracted, and the accuracy and reliability of fault diagnosis are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis, and in particular to a GA-BP rolling bearing fault diagnosis method based on electrostatic signals. Background Art

[0002] Bearings are critical components in industrial equipment, and their operating status directly impacts equipment reliability and safety. According to statistics, approximately 30% of mechanical failures are related to bearings. Therefore, bearing fault diagnosis technology is crucial for improving equipment efficiency and reducing maintenance costs.

[0003] Currently, common bearing fault diagnosis methods include vibration signal analysis, temperature monitoring, and acoustic emission technology. Vibration signal analysis is the most commonly used method, but it is less sensitive to early-stage faults and is susceptible to interference from environmental noise. In recent years, electrostatic signal analysis has gained increasing attention as an emerging fault diagnosis method. Electrostatic signals can reflect the friction and wear conditions within bearings and are highly sensitive to early-stage faults. However, the processing and analysis of electrostatic signals is complex, and the diagnostic accuracy and stability of existing methods still need to be improved.

[0004] Neural network technologies (such as BP neural networks) have been widely used in fault diagnosis. BP neural networks can automatically extract fault characteristics by learning from large amounts of data, but their training process is prone to becoming stuck in local optimal solutions and suffers from low training efficiency. Genetic algorithms (GAs), as global optimization algorithms, can effectively address the local optimality issues of BP neural networks. However, existing technologies have not yet fully combined the advantages of electrostatic signal analysis and GA-BP optimization methods, leaving room for improvement in the accuracy and efficiency of bearing fault diagnosis. Summary of the Invention

[0005] To overcome the existing problems and defects, the present invention proposes a GA-BP rolling bearing fault diagnosis method based on electrostatic signals, comprising the following steps:

[0006] S1. Collecting electrostatic signals of rolling bearings after signal amplification;

[0007] S2. Decompose the electrostatic signal using an improved adaptive noise complete ensemble empirical mode decomposition method to obtain multiple intrinsic mode function components (IMFs);

[0008] S3. Extract energy features based on intrinsic mode function components and construct fault diagnosis feature vectors;

[0009] S4. Preprocess the feature vector and divide the data set;

[0010] S5. Input the preprocessed feature vector into the neural network model, and use the genetic algorithm to optimize the parameters of the neural network model, and diagnose the fault type of the rolling bearing through the optimized neural network model.

[0011] Furthermore, step S2 is specifically as follows: adding a series of adaptive white noises whose intensities are controlled by the standard deviation coefficient to the original signal, using a cubic spline function to fit the maximum and minimum points of the signal after superimposing the noise to form upper and lower envelopes, calculating the mean of the envelopes as the local mean, and extracting the intrinsic mode function components that stably meet the IMF standard in an iterative manner. The above decomposition process is performed separately under multiple groups of different noises, and the results of each corresponding component are averaged to obtain the final IMF component.

[0012] Furthermore, step S3 is specifically as follows: selecting the first 8 IMF components obtained by decomposition, square-summing all sampling points of each IMF component and taking the square root to obtain its energy value, and taking the ratio of the energy value of each component to the sum of the energy values ​​of all IMF components as the normalized fault feature vector.

[0013] Furthermore, in step S4, the ratio of the training set to the test set is 8:2, and the input data is normalized using the maximum and minimum values.

[0014] Furthermore, the neural network model is a BP neural network including an input layer, a hidden layer and an output layer, the number of input layer nodes is 8, the number of hidden layer nodes is 5, the number of output layer nodes is 3, and the activation function is a Sigmoid function.

[0015] Furthermore, during the genetic algorithm optimization process, the fitness function is the inverse of the mean square error of the neural network on the training set.

[0016] Furthermore, the selection operation of the genetic algorithm adopts a normal geometric selection method, the crossover operation adopts an arithmetic crossover method, and the mutation operation adopts a non-uniform mutation method.

[0017] Furthermore, the maximum number of iterations of the neural network model is 1000, the target error is 1e-6, and the learning rate is 0.01.

[0018] Beneficial effects of the present invention:

[0019] This method uses electrostatic signal monitoring to effectively detect friction and wear within bearings, with a particularly high sensitivity for early-stage failures. Compared to traditional vibration signal analysis methods, it can detect bearing failures earlier, reducing the risk of equipment damage.

[0020] This paper uses an improved complete ensemble empirical mode decomposition (ICEEMDAN) method with adaptive noise to decompose electrostatic signals. This method effectively extracts fault features, avoids the modal aliasing problem found in traditional empirical mode decomposition (EMD), and improves the accuracy and stability of signal decomposition. By extracting the energy values ​​of the first eight IMF components as fault feature vectors, it effectively reflects bearing fault information, reduces data dimensionality, and improves fault diagnosis efficiency.

[0021] The present invention optimizes the weights and bias parameters of the BP neural network through a genetic algorithm (GA), thereby solving the problem that traditional BP neural networks are prone to falling into local optimal solutions, improving the global search capability and training efficiency of the model, and significantly improving the accuracy of fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0023] Figure 1 This is a flow chart of the rolling bearing fault diagnosis method based on electrostatic signals and GA-BP according to the present invention;

[0024] Figure 2 This is an exploded view of the ICEEMDAN described in the present invention;

[0025] Figure 3 This is a diagram of the BP model fault diagnosis results of the test set described in the present invention;

[0026] Figure 4 This is a diagram of the fault diagnosis results of the GA-BP model of the test set described in the present invention;

[0027] Figure 5 This is a graph showing the BP model training error for the test set described in the present invention;

[0028] Figure 6 This is a graph showing the GA-BP model training error for the test set described in the present invention. DETAILED DESCRIPTION

[0029] Example 1:

[0030] like Figure 1As shown, this embodiment provides a GA-BP rolling bearing fault diagnosis method based on electrostatic signals, including the following steps: S1. Collecting the electrostatic signal of the rolling bearing after signal amplification processing; S2. Decomposing the electrostatic signal using an improved adaptive noise complete set empirical mode decomposition method to obtain multiple intrinsic mode function components (IMF); S3. Based on the intrinsic mode function components, extracting energy features and constructing fault diagnosis feature vectors; S4. Preprocessing the feature vectors and dividing the data set; S5. Inputting the preprocessed feature vectors into a neural network model, and optimizing the parameters of the neural network model using a genetic algorithm, and diagnosing the fault type of the rolling bearing through the optimized neural network model.

[0031] Each step is described below with reference to specific embodiments:

[0032] Step S1: Collect the electrostatic signal of the rolling bearing after signal amplification

[0033] In this step, a signal amplifier is used to amplify the electrostatic signal before collecting it, and the signal waveform is output through the electrostatic signal collection system.

[0034] Step S2: Decompose the electrostatic signal using an improved adaptive noise complete set empirical mode decomposition method to obtain multiple intrinsic mode function components (IMFs)

[0035] like Figure 2 As shown in Figure 2, this step decomposes the electrostatic signal through the improved adaptive noise complete ensemble empirical mode decomposition ICEEMDAN to obtain a series of modal components IMF. The specific calculation method is as follows:

[0036] S2-1 to original signal Add a series of adaptive white noise , then the new signal generated is:

[0037]

[0038] in is the coefficient of the noise standard deviation, which is used to control the intensity of the added noise. Indicates the A white noise signal.

[0039] S2-2 signal after adding noise Perform a decomposition operation similar to EMD, by finding the local extreme points of the signal, and then fitting the upper and lower envelopes with a cubic spline function, calculating the average value of the envelope and subtracting it from the original signal, iterating repeatedly, assuming that after After iterations, IMF1 is obtained. The calculation of each iteration can be expressed as:

[0040]

[0041] in It is The residual signal after iterations is is the local mean calculated from the upper and lower envelopes. When the IMF conditions are met, IMF1 is obtained.

[0042] S2-3 calculates the first residual signal:

[0043]

[0044] S2-4 adds adaptive noise to the residual signal , generate a new signal:

[0045]

[0046] in is the new noise standard deviation coefficient, It is white noise The first IMF is obtained after the first EMD-like decomposition. A decomposition operation similar to step S2-2 is performed to obtain the second modal component IMF2.

[0047] S2-5 repeats step S2-4 to calculate the subsequent residual signal , and add adaptive noise to the residual signal for decomposition until the residual signal meets the stopping criterion, and the signal diagram is as follows Figure 2 shown.

[0048] S3. Based on the intrinsic mode function components, energy features are extracted and fault diagnosis feature vectors are constructed.

[0049] The S3-1 high frequency band mainly contains the fault information of rolling bearings, so the first 8 IMF components are taken as the subsequent research objects to extract the fault feature information.

[0050] S3-2 Assume that the IMF matrix obtained after ICEEMDAN decomposition is , where each column represents an IMF component, let the matrix The size of ( is the number of sample points, and 8 is the number of IMF components).

[0051] S3-3 For the IMF components ( ), first square the elements of each column and add them up, then open the sign. The sum of all columns after this operation is:

[0052]

[0053] Rule No. The energy value of each IMF component The proportion (relative energy) can be expressed as:

[0054]

[0055] Get the fault feature vector.

[0056] S4. Preprocess the feature vector and divide the dataset.

[0057] The input feature vector data is preprocessed, including data shuffling, dividing into training and test sets, and normalization. The specific steps include the following:

[0058] S4-1 randomly shuffles the input data to generate a data set in a random order;

[0059] S4-2 divides the data set into a training set and a test set, where the training set accounts for 80% of the total data and the test set accounts for 20% of the total data;

[0060] S4-3 normalizes the input data of the training set and test set so that their value range is [0,1]. The normalization formula is as follows:

[0061]

[0062] in, and are the minimum and maximum values ​​of the data, respectively.

[0063] S5. Input the preprocessed feature vector into the neural network model, and use the genetic algorithm to optimize the parameters of the neural network model, and diagnose the fault type of the rolling bearing through the optimized neural network model.

[0064] S5: Construct a neural network model, use a genetic algorithm to optimize the weights and bias parameters of the neural network model, train the neural network model with the optimized weights and bias parameters, then predict the test set, and evaluate the model performance, specifically including the following steps:

[0065] S5-1. Neural Network Model Construction: Build a BP neural network consisting of an input layer, a hidden layer, and an output layer. Taking the energy value E of the eight IMFs, the number of input layer nodes is set to 8. The number of hidden layer nodes is set to 5. Faults are categorized as outer race faults, inner race faults, and roller faults, so the number of output layer nodes is set to 3. The Sigmoid function is used as the activation function.

[0066] S5-2, initialize the population, and set the population size (i.e. the number of individuals in the population) to , each individual consists of the weight and bias parameters of the neural network. The range of parameters is usually set to [−1,1]. For each individual, a set of weight and bias parameters are randomly generated. The formula is as follows:

[0067]

[0068] in For the weight or bias parameters, is a random number in the range [0, 1], and The minimum and maximum values ​​of the parameters, respectively

[0069] S5-3. The fitness value is used to evaluate the quality of each individual. The higher the fitness value, the better the individual. To calculate the prediction error, use the weight and bias parameters of the current individual to initialize the neural network and calculate its prediction error on the training set. The formula is as follows:

[0070]

[0071] in is the true value, is the predicted value.

[0072] The prediction error is converted into a fitness value as follows:

[0073]

[0074] in is a small constant (such as ), used to prevent division by zero errors.

[0075] S5-4, the selection operation is used to select the better individuals from the current population to enter the next generation. The present invention adopts the normal geometric selection method to calculate the selection probability of each individual according to the fitness value. The formula is as follows:

[0076]

[0077] in For the The probability of selection of an individual, For the The fitness value of an individual.

[0078] The crossover operation is used to generate new individuals. The present invention adopts the arithmetic crossover method to randomly select two parent individuals from the current population. and , generate offspring individuals through arithmetic crossover, the formula is as follows:

[0079]

[0080] in is the crossover factor (usually set to 0.5), The generated offspring individuals.

[0081] The mutation operation is used to increase the diversity of the population. The present invention adopts a non-uniform mutation method to randomly select an individual from the current population for mutation. The non-uniform mutation is performed on the selected individual. The formula is as follows:

[0082]

[0083] in is the parameter value before mutation, is the parameter value after mutation, is the randomly generated mutation amount, whose size gradually decreases with the number of iterations. The formula is as follows:

[0084]

[0085] in is the current iteration number, is the maximum number of iterations, is the mutation intensity parameter, set to 3.

[0086] Repeat the selection, crossover, and mutation operations until the maximum number of iterations is reached or the fitness value converges. Finally, the individual with the highest fitness value is selected as the optimal solution.

[0087] S5-5. Use the optimized weights and bias parameters to train the neural network. Set the maximum number of iterations to 1000, the target error to 1e-6, and the learning rate to 0.01. Update the weights and bias parameters of the neural network using the backpropagation algorithm. The weight update formula for backpropagation is as follows:

[0088]

[0089] in, The learning rate is 0.01, is the loss function.

[0090] S5-6. Calculate the prediction accuracy of the training set and test set, and draw a comparison chart between the prediction results and the true values ​​for model performance evaluation.

[0091] Combine Figure 3 , Figure 4 We can see that for Class 1 and Class 2 faults, the unoptimized BP fault diagnosis algorithm's predicted and actual values ​​are consistent, resulting in relatively high classification accuracy. However, for Class 3 faults, the unoptimized BP fault diagnosis algorithm's predicted and actual values ​​differ significantly, making Class 3 fault diagnosis inaccurate. However, for Class 1 and Class 3 faults, the GA-BP fault diagnosis algorithm achieves 100% prediction accuracy, with only Class 2 faults exhibiting deviations. Overall, fault classification is accurate.

[0092] Combine Figure 5 , Figure 6 We can see that the traditional BP neural network needs 16 steps to reach the training target, while the GA-BP neural network only needs 8 steps to reach the training target. It is not difficult to see that the genetic BP neural network has a faster convergence speed and can avoid the occurrence of local minimum values. It can also be found that the GA-BP neural network has higher training accuracy and better training effect than the traditional BP neural network.

[0093] It should be noted that the above embodiments can be freely combined as needed. The above are only preferred embodiments of the present invention. It should be pointed out that ordinary relevant personnel in this technical field can make several improvements and modifications without departing from the principles of the present invention. Such improvements and modifications should also be considered as the scope of protection of the present invention.

[0094] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

Claims

1. A GA-BP rolling bearing fault diagnosis method based on electrostatic signals, characterized in that: The steps include: S1. Collecting electrostatic signals of rolling bearings after signal amplification; S2. Decomposing the electrostatic signal using an improved adaptive noise complete ensemble empirical mode decomposition method to obtain a plurality of intrinsic mode function components (IMFs); S3. Based on the intrinsic mode function components, energy characteristics are extracted and a fault diagnosis feature vector is constructed; S4. Preprocess the feature vector and divide the data set; S5. Input the preprocessed feature vector into the neural network model, and use a genetic algorithm to optimize the parameters of the neural network model, and diagnose the fault type of the rolling bearing through the optimized neural network model.

2. The GA-BP rolling bearing fault diagnosis method based on electrostatic signals according to claim 1 is characterized in that: Step S2 is specifically as follows: adding a series of adaptive white noises whose intensities are controlled by the standard deviation coefficient to the original signal, using the cubic spline function to fit the maximum and minimum points of the signal after superimposing the noise to form upper and lower envelopes, calculating the mean of the envelopes as the local mean, and extracting the intrinsic mode function components that stably meet the IMF standard in an iterative manner. The above decomposition process is performed separately under multiple groups of different noises, and the results of each corresponding component are averaged to obtain the final IMF component.

3. The GA-BP rolling bearing fault diagnosis method based on electrostatic signals according to claim 1 is characterized in that: Step S3 specifically includes: selecting the first 8 IMF components obtained by decomposition, taking the square root of the square sum of all sampling points of each IMF component to obtain its energy value, and taking the ratio of the energy value of each component to the sum of the energy values ​​of all IMF components as the normalized fault feature vector.

4. The GA-BP rolling bearing fault diagnosis method based on electrostatic signals according to claim 1, characterized in that: In step S4, the ratio of the training set to the test set is 8:2, and the input data is normalized using the maximum and minimum values.

5. The GA-BP rolling bearing fault diagnosis method based on electrostatic signals according to claim 1 is characterized in that: The neural network model is a BP neural network comprising an input layer, a hidden layer and an output layer, wherein the number of nodes in the input layer is equal to the dimension of the extracted feature vector, the number of nodes in the output layer is equal to the number of rolling bearing fault types to be diagnosed, and the activation function is a Sigmoid function.

6. The GA-BP rolling bearing fault diagnosis method based on electrostatic signals according to claim 5, characterized in that: During the genetic algorithm optimization process, the fitness function is the inverse of the mean square error of the neural network on the training set.

7. The GA-BP rolling bearing fault diagnosis method based on electrostatic signals according to claim 6, characterized in that: The selection operation of the genetic algorithm adopts the normal geometric selection method, the crossover operation adopts the arithmetic crossover method, and the mutation operation adopts the non-uniform mutation method.

8. The GA-BP rolling bearing fault diagnosis method based on electrostatic signals according to any one of claims 5 to 7, characterized in that: The maximum number of iterations of the neural network model is 1000, the target error is 1e-6, and the learning rate is 0.01.

Images