Gearbox composite fault diagnosis method based on optimal Legendre multi-wavelet frequency band
By using the optimal Legendre multi-wavelet band and improved adaptive denoising algorithm in gearbox fault diagnosis, combined with genetic algorithm and statistical feature extraction, the strong noise and small sample problems in gearbox fault diagnosis are solved, achieving high-precision Compound fault diagnosis.
Patent Information
- Application Number
- CN202410734297.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-07
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2044-06-07
AI Technical Summary
There are strong noise interference, label noise and small sample problems in gearbox fault diagnosis, resulting in low diagnostic accuracy and it is difficult for existing methods to effectively identify the composite fault type.
Using the composite fault diagnosis method based on the optimal Legendre multi-wavelet frequency band, the network structure parameters are optimized by constructing the Max min objective function, genetic algorithm and gradient descent learning algorithm, and combining the improved adaptive denoising algorithm and statistical feature extraction to achieve high-precision diagnosis .
High-precision gearbox composite fault diagnosis is achieved under conditions of strong noise, label noise and small sample size, which is better than traditional methods and shows high diagnostic accuracy especially in limited fault samples.
Smart Images

Figure CN119003978A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the cross-technical field of rotating machinery fault diagnosis and artificial intelligence technology, and in particular to a gearbox composite fault diagnosis method based on an optimal Legendre multi-wavelet frequency band. Background Art
[0002] The health of the gearbox directly affects whether the rotating mechanical equipment can work properly. If the type of fault can be accurately identified and predicted, the huge losses caused by the fault can be effectively avoided. However, when a gearbox fails, different types of faults can easily couple together, and the fault signal is easily contaminated by strong noise. In addition, in actual environments, incorrect labels are inevitable due to operator errors. Mechanical equipment is usually in normal working condition, which means that only a limited number of fault samples can be collected. Therefore, the lack of typical fault samples, that is, the small sample problem, remains a major challenge in mechanical equipment fault diagnosis.
[0003] There are generally three methods for diagnosing gearbox compound faults: analytical model-based methods, knowledge-driven methods, and data-driven methods. The analytical model-based method mainly relies on the mathematical model of the mechanical operation process. Therefore, it requires sufficient knowledge of the mathematical model knowledge of the mechanical equipment system. However, it is usually difficult to achieve mechanical fault diagnosis by developing an effective mathematical model. The knowledge-driven method basically relies on domain knowledge. However, the knowledge-driven method requires a deep understanding of the causal relationship between faults and symptoms in mechanical equipment, which may be quite challenging for many technicians, thus hindering them from accurately identifying the fault type in mechanical equipment. In contrast, the rapid development of artificial intelligence has enabled data-driven methods to provide more significant benefits for mechanical equipment fault diagnosis in practical applications. This method does not require too much mathematical model and domain knowledge, and can find corresponding good classification schemes through raw data. Traditional machine learning-based fault diagnosis methods require complex manual selection of fault features, which is difficult to meet the analysis of various fault data in engineering practice. Training a deep learning model in practical applications usually requires a large number of labeled samples. Generative adversarial networks, diffusion models, transfer learning, etc. are used to amplify existing fault sample data sets, but deep models are difficult to train.
[0004] De-noising algorithms based on wavelet transform are widely used in signal processing to improve the signal-to-noise ratio for fault detection and diagnosis by estimating thresholds. However, simple denoising algorithms usually cannot adequately remove noise because its frequency overlaps with the fault characteristic signal.
[0005] Therefore, a new solution to the above problems needs to be proposed. Summary of the invention
[0006] The purpose of the present invention is to provide a gearbox composite fault diagnosis method based on an optimal Legendre multi-wavelet frequency band. Aiming at the problem of high-precision diagnosis of gearbox composite faults under conditions of strong noise interference, label noise and limited fault samples, the present invention constructs a Max min objective function, uses a genetic algorithm and a gradient descent learning algorithm to learn the optimal value between high diagnostic accuracy and low signal energy loss, and obtains optimized network structure parameters (the number of Legendre multi-wavelets, the level of wavelet decomposition, and the number of neurons in the hidden layer of the BP neural network); an improved adaptive denoising algorithm is proposed on each optimal Legendre wavelet frequency band to effectively remove the interference of strong noise; based on the statistical feature combination standard deviation + kurtosis, entropy + skewness, the gearbox composite fault features are effectively extracted, avoiding the complex feature selection algorithm of the traditional machine learning method and the difficulties of the deep network in extracting features and learning, thereby effectively solving the problem of high-precision diagnosis of gearbox composite faults under small sample conditions.
[0007] To achieve the above object, the present invention provides the following technical solution: a gearbox composite fault diagnosis method based on optimal Legendre multi-wavelet frequency band, comprising at least the following steps:
[0008] Step 1: Data collection, using the PHM 2009 composite gearbox fault dataset;
[0009] Step 2: Initialize the algorithm;
[0010] Step 3: Sample processing, 75% of the total samples of each gearbox fault category are allocated to training samples and 25% are allocated to test samples;
[0011] Step 4: Legendre multi-wavelet decomposition process, including taking the initial values of network parameters and implementing Legendre multi-wavelet decomposition;
[0012] Step 5: Improved adaptive threshold denoising algorithm;
[0013] Step 6: Feature extraction: on each optimal Legendre multi-wavelet high-frequency band after denoising and the Legendre multi-wavelet low-frequency band without denoising, calculate the statistical feature combination standard deviation + kurtosis, entropy + skewness as the feature input of the BP neural network;
[0014] Step 7: Train the BP neural network, normalize the optimal features obtained to between 0 and 1, and train the BP neural network through the gradient descent learning algorithm. After that, pass the normalized results of the optimal features obtained on the test sample to the BP neural network, and calculate the test diagnosis accuracy to complete the performance evaluation of the proposed method.
[0015] Step 8: Fusion of genetic algorithm and improved adaptive threshold denoising algorithm. If the fault diagnosis accuracy does not reach the expected diagnostic accuracy, the decomposition level of multiwavelet, the number of Legendre multiwavelet, and the number of neurons of BP neural network are updated by genetic algorithm, and steps 4 to 7 are repeated to obtain higher fault diagnosis accuracy.
[0016] Step 9: Output the results. If the fault diagnosis accuracy reaches the expected diagnostic accuracy, then output the highest diagnostic accuracy and the corresponding multi-wavelet decomposition level, the number of Legendre multi-wavelets, and the number of neurons in the BP neural network.
[0017] Further, the gearbox model used in the PHM 2009 composite gearbox fault data set is ER-10K;
[0018] The gearbox fault signals in the PHM 2009 composite gearbox fault data set have 6 different health conditions. Each signal is collected under 5 conditions of low load and speed of 30, 35, 40, 45, and 50 Hz, respectively. The sampling frequency of each signal is 66.7 kHz and the sampling time is 8 s.
[0019] Furthermore, the initial algorithm at least includes the number of iterations of the initialization genetic algorithm, the number of iterations of the gradient descent algorithm, the decomposition level, the number of Legendre multi-wavelets, the number of hidden layer neurons of the BP neural network, the number of chromosomes, the crossover probability, the mutation probability, the training samples and the test samples, the iterative termination training error of the BP neural network and the expected fault diagnosis accuracy.
[0020] Furthermore, the step 5 at least includes the following steps:
[0021] S1: Noise estimation in the optimal high-frequency band of Legendre multiwavelets:
[0022] Let W = [w 1 ,w 2 ,…,w N ] is a vector, which is composed of the squares of the Legendre multiwavelet coefficients of each optimal high-frequency band arranged in ascending order;
[0023] First, let the vector W be the minimum value r of the risk vector R calculated according to formula (1) b (b th r) is used as an unbiased risk estimate, as shown in formula (1):
[0024]
[0025] Then, using the subscript b corresponding to the risk value, select the square w of the Legendre multiwavelet coefficient at the corresponding position in the vector W b As the minimum risk factor,
[0026] Therefore, the improved threshold estimation is formula (2):
[0027]
[0028] S2: Threshold parameter selection based on experiments:
[0029] According to formula (2), the Legendre multiwavelet coefficients of the optimal decomposition level are used to estimate the optimal threshold, where the parameter σ is the standard deviation of the optimal high-frequency band signal, and η is a parameter determined by numerical simulation experiments according to different application scenarios;
[0030] S3: Improved soft threshold denoising algorithm:
[0031] Based on the threshold estimation of formula (2), the improved soft threshold denoising algorithm is obtained as follows:
[0032]
[0033] in, is the constructed adaptive weight coefficient model. It should be noted that through this unbiased risk estimation, different thresholds λ can be obtained in different optimal Legendre multi-wavelet high-frequency bands, so that the improved algorithm is more helpful to remove complex noise interference;
[0034] S4: Evaluation of improved denoising algorithm:
[0035] The performance of the improved adaptive threshold denoising algorithm is evaluated by the root mean square error (RMSE). The performance of the improved adaptive threshold denoising algorithm is calculated by the root mean square error as shown in formula (4):
[0036]
[0037] Among them, N 0 is the number of sample points, x i is the i-th element of the original signal, y i is the i-th element of the denoised signal reconstructed using the inverse Legendre multiwavelet transform;
[0038] The lower the RMSE value, the better the denoising performance of the denoising algorithm.
[0039] Furthermore, the fusion of the genetic algorithm and the improved adaptive threshold denoising algorithm comprises at least the following steps:
[0040] S1: The fitness function of the genetic algorithm adopts the expected accuracy of fault diagnosis, which is specifically formula (5):
[0041]
[0042] Where TP, TN, FP and FN represent true positive, true negative, false positive and false negative respectively;
[0043] S2: The weight coefficient of the BP neural network is learned through the gradient descent algorithm. When the loss function reaches the predetermined lower limit or the number of iterations reaches the predetermined maximum value, the adjustment process of the network weight coefficient stops. At this time, if the fault diagnosis accuracy does not reach the target, the genetic algorithm is used to further select the better value of the network structure parameter;
[0044] S3: Let GA(·) represent the optimization process of the genetic algorithm for the network structure parameters, GD(·) represent the BP neural network weight coefficient learning process, and Fitness(·) represent the fitness function of the genetic algorithm. The designed Max min objective function is formula (6):
[0045]
[0046] Where n, p, ξ represent the decomposition level, the number of Legendre multiwavelets and the number of hidden layer neurons of the BP neural network, respectively, and λ represents the optimal unbiased risk estimation threshold in each optimal Legendre multiwavelet high frequency band.
[0047] Compared with the prior art, the present invention has the following beneficial effects:
[0048] 1. The present invention improves the Rigrsure threshold selection method, establishes a weight model for the soft threshold, adds experimental parameters to improve the threshold estimation, and proposes an adaptive threshold denoising algorithm improved in each optimized Legendre multi-wavelet frequency band. This improved adaptive threshold denoising algorithm in the optimized frequency band can effectively eliminate strong noise, label noise and interference from unknown environments under complex working conditions, and retain important fault feature components in each optimal frequency band.
[0049] 2. The Legendre multi-wavelet frequency band optimization in the present invention is achieved by constructing a Max min objective function, using a genetic algorithm and a gradient descent learning algorithm. Such optimization avoids the traditional method of obtaining network structure parameters (the number of Legendre wavelet bases, the decomposition level, and the number of neurons in the hidden layer of the BP neural network) through a large number of experiments.
[0050] 3. The present invention selects the traditional and effective statistical feature combination standard deviation + kurtosis, entropy + skewness to effectively extract the gearbox compound fault characteristics. The fundamental reason for the effectiveness of this method is that: the interference of environmental noise is removed in the optimized Legendre multi-wavelet frequency band; the frequency domain of the Legendre multi-wavelet optimized decomposition layer obtained can fully characterize the complex characteristics of the gearbox compound fault, that is, by entering a simple classifier through simple statistical features, high-precision diagnosis of gearbox compound faults under complex conditions can be achieved. The present invention avoids the complex feature selection of traditional machine learning methods and avoids the use of deep networks to extract features, thereby effectively solving the problem of high-precision diagnosis of gearbox compound faults under small sample conditions.
[0051] 4. The present invention verifies the effectiveness and robustness of the proposed method through a large number of comparative experiments on the PHM 2009 gearbox fault dataset (Prognostics and Health Management, from the IEEE International Prognostics and Health Management Conference 2009). The proposed method shows 100% diagnostic accuracy on the gearbox dataset without noise interference; under the three conditions of 6dB noise, label-to-noise ratio of 0.05 and limited fault samples (10), the diagnostic accuracy of the proposed method reaches 99.83%, 95.33% and 90.80% respectively, which is better than the method based on Daubechies wavelet. BRIEF DESCRIPTION OF THE DRAWINGS
[0052] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for describing the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.
[0053] Figure 1 It is a schematic diagram of the overall process of the present invention. DETAILED DESCRIPTION
[0054] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.
[0055] See also Figure 1 , a gearbox composite fault diagnosis method based on optimal Legendre multi-wavelet frequency band, comprising at least the following steps:
[0056] Step 1: Data collection, using the PHM 2009 composite gearbox fault dataset;
[0057] Step 2: Initialize the algorithm;
[0058] Step 3: Sample processing: 75% of the total samples of each gearbox fault category are allocated to training samples, and 25% are allocated to test samples; in order to avoid particularity and contingency, training samples and test samples are randomly selected, and such random experiments are repeated 10 times;
[0059] Step 4: Legendre multi-wavelet decomposition process, including taking the initial values of the network parameters and implementing Legendre multi-wavelet decomposition. That is, three Legendre multi-wavelets are used, and the level of Legendre multi-wavelet decomposition is 2;
[0060] Step 5: Improved adaptive threshold denoising algorithm;
[0061] Step 6: Feature extraction: on each optimal Legendre multi-wavelet high-frequency band after denoising and the Legendre multi-wavelet low-frequency band without denoising, calculate the statistical feature combination standard deviation + kurtosis, entropy + skewness as the feature input of the BP neural network;
[0062] Step 7: Train the BP neural network, normalize the optimal features obtained to between 0 and 1, and train the BP neural network through the gradient descent learning algorithm. After that, pass the normalized results of the optimal features obtained on the test sample to the BP neural network, and calculate the test diagnosis accuracy to complete the performance evaluation of the proposed method.
[0063] Step 8: Fusion of genetic algorithm and improved adaptive threshold denoising algorithm. If the fault diagnosis accuracy does not reach the expected diagnostic accuracy, the decomposition level of multiwavelet, the number of Legendre multiwavelet, and the number of neurons of BP neural network are updated by genetic algorithm, and steps 4 to 7 are repeated to obtain higher fault diagnosis accuracy.
[0064] Step 9: Output the results. If the fault diagnosis accuracy reaches the expected diagnostic accuracy, then output the highest diagnostic accuracy and the corresponding multi-wavelet decomposition level, the number of Legendre multi-wavelets, and the number of neurons in the BP neural network.
[0065] The gearbox model used in the PHM 2009 composite gearbox fault dataset is ER-10K;
[0066] The gearbox fault signals in the PHM 2009 composite gearbox fault data set have 6 different health conditions. Each signal is collected under low load and 5 conditions of speed 30, 35, 40, 45, and 50 Hz. The sampling frequency of each signal is 66.7 kHz and the sampling time is 8 s. That is, there are 533,312 sampling points for the signal under one working condition and one running speed, so there are 2,666,560 sample points in total under five speed conditions. For each composite fault signal, 4096 data points are taken as a sample, and a total of 600 samples are collected. In order to meet the actual diagnosis needs as much as possible, 450 samples are randomly selected from each health condition as training samples, and the remaining 150 samples are used to test the diagnostic accuracy of the proposed method. In addition, in order to match the composite fault mode of composite fault diagnosis, the fault label is set to a multi-label form, so that a composite fault label can represent the fault information of multiple parts.
[0067] The detailed description of the six composite fault types of the gearbox and the corresponding label arrays are shown in Table 1.
[0068] Table 1. Six types of composite faults in gearboxes
[0069]
[0070]
[0071] The position of the fault type in Table 1 represents the position of the fault in the multi-label array.
[0072] The initial algorithm at least includes the number of iterations of the initial genetic algorithm, the number of iterations of the gradient descent algorithm, the decomposition level, the number of Legendre multiwavelets, the number of hidden layer neurons of the BP neural network, the number of chromosomes, the crossover probability, the mutation probability, the training samples and the test samples, the iterative termination training error of the BP neural network and the expected fault diagnosis accuracy.
[0073] In actual engineering environments, the collected gearbox vibration signals usually encounter a lot of noise interference, employee mislabeling, and limited fault samples, which poses a huge challenge to high-precision fault diagnosis. In order to reduce the impact of noise on fault diagnosis, the present invention designs an improved adaptive threshold denoising algorithm on the optimized Legendre multi-wavelet high frequency band, as follows:
[0074] S1: Noise estimation in the optimal high-frequency band of Legendre multiwavelets:
[0075] Let W = [w 1 ,w 2 ,…,w N] is a vector, which is composed of the squares of the Legendre multiwavelet coefficients of each optimal high-frequency band arranged in ascending order;
[0076] First, let the vector W be the minimum value r of the risk vector R calculated according to formula (1) b (b th r) is used as an unbiased risk estimate, as shown in formula (1):
[0077]
[0078] Then, using the subscript b corresponding to the risk value, select the square w of the Legendre multiwavelet coefficient at the corresponding position in the vector W b As the minimum risk factor,
[0079] Therefore, the improved threshold estimation is formula (2):
[0080]
[0081] S2: Threshold parameter selection based on experiments:
[0082] According to formula (2), the Legendre multiwavelet coefficients of the optimal decomposition level are used to estimate the optimal threshold, where the parameter σ is the standard deviation of the optimal high-frequency band signal, and η is a parameter determined by numerical simulation experiments according to different application scenarios;
[0083] S3: Improved soft threshold denoising algorithm:
[0084] Based on the threshold estimation of formula (2), the improved soft threshold denoising algorithm is obtained as follows:
[0085]
[0086] in, is the constructed adaptive weight coefficient model. It should be noted that through this unbiased risk estimation, different thresholds λ can be obtained in different optimal Legendre multi-wavelet high-frequency bands, so that the improved algorithm is more helpful to remove complex noise interference;
[0087] S4: Evaluation of improved denoising algorithm:
[0088] The performance of the improved adaptive threshold denoising algorithm is evaluated by the root mean square error (RMSE). The specific calculation formula of the improved adaptive threshold denoising algorithm is formula (4):
[0089]
[0090] Among them, N0 is the number of sample points, x i is the i-th element of the original signal, y i is the i-th element of the denoised signal reconstructed using the inverse Legendre multiwavelet transform;
[0091] The lower the RMSE value, the better the denoising performance of the denoising algorithm.
[0092] The fusion of genetic algorithm and improved adaptive threshold denoising algorithm includes at least the following steps:
[0093] The method proposed in the present invention adopts a combination of a genetic algorithm and an improved adaptive threshold denoising algorithm, specifically: using a genetic algorithm to optimize three network structure parameters, namely the number of Legendre multi-wavelets, the decomposition level, and the number of hidden layer neurons of a BP neural network, so as to obtain the optimal Legendre multi-wavelet high-frequency band information under complex conditions; and then using the improved adaptive threshold denoising algorithm for each optimal Legendre multi-wavelet high-frequency band to effectively remove complex noise interference.
[0094] S1: The fitness function of the genetic algorithm adopts the expected accuracy of fault diagnosis, which is specifically formula (5):
[0095]
[0096] Among them, TP, TN, FP and FN represent true positive, true negative, false positive and false negative respectively;
[0097] S2: The weight coefficient of the BP neural network is learned through the gradient descent algorithm. When the loss function reaches the predetermined lower limit or the number of iterations reaches the predetermined maximum value, the adjustment process of the network weight coefficient stops. At this time, if the fault diagnosis accuracy does not reach the target, the genetic algorithm is used to further select the better value of the network structure parameter;
[0098] S3: Let GA(·) represent the optimization process of the genetic algorithm for the network structure parameters, GD(·) represent the BP neural network weight coefficient learning process, and Fitness(·) represent the fitness function of the genetic algorithm. The designed Max min objective function is formula (6):
[0099]
[0100] Where n, p, ξ represent the decomposition level, the number of Legendre multiwavelets and the number of hidden layer neurons of the BP neural network, respectively, and λ represents the optimal unbiased risk estimation threshold in each optimal Legendre multiwavelet high frequency band.
[0101] On the PHM 2009 dataset, the optimal network structure parameters obtained by the method proposed in this invention are shown in Table 2:
[0102] Table 2. Optimal network structure parameters obtained by the method proposed in this invention
[0103]
[0104] As shown in Table 2, the optimal structure of the BP neural network consists of 1 input layer formed by 12 input neurons, 3 hidden layers formed by 12, 56, and 26 neurons, and 1 output layer formed by 12 neurons. The threshold denoising method proposed in the present invention has a multi-wavelet decomposition level of 1 in the optimal case, a multi-wavelet number of 3, and an experimental parameter η of 1.2.
[0105] This embodiment uses a genetic algorithm to optimize the structural parameters of the proposed model: the number of Legendre multi-wavelets, the level of wavelet decomposition, and the number of hidden layer neurons of the classifier BP neural network, and proposes an adaptive threshold denoising algorithm based on Legendre multi-wavelet optimization band improvement, which effectively removes the strong noise interference in complex environments, and effectively solves the problem of high-precision diagnosis of gearbox composite faults with label noise and small samples by extracting fault statistical features and combining standard deviation + kurtosis, entropy + skewness and inputting them into the BP neural network, avoiding the problem that deep neural networks are difficult to design and train due to their complex structure, and making it easier to realize online fault diagnosis of rotating machinery.
[0106] A verification of the method of the present invention under complex conditions is proposed for the above-mentioned embodiment;
[0107] In order to verify the effectiveness of the fault diagnosis method proposed in the present invention, it is tested under the conditions of strong noise interference, label noise, and limited fault samples. Table 3 lists in detail the genetic algorithm configuration parameters involved in the experiments under three different conditions.
[0108] Table 3. Configuration parameters of the genetic algorithm
[0109]
[0110] The fault diagnosis method proposed in the present invention is compared with the Daubechies wavelet transform method to show that the proposed method has higher fault diagnosis accuracy than the traditional wavelet transform method. For the sake of clarity, the fault diagnosis method involved in the experiment is abbreviated as:
[0111] Method 1: The method proposed by the present invention, statistical characteristics (standard deviation + kurtosis);
[0112] Method 2: The method proposed by the present invention, statistical characteristics (entropy + skewness);
[0113] Method 3: Based on Daubechies wavelet transform method, statistical characteristics (standard deviation + kurtosis);
[0114] Method 4: Based on Daubechies wavelet transform method, statistical characteristics (entropy + skewness).
[0115] Experimental results under different strong noise interference
[0116] -2, 0, 2, 4, 6 dB signal-to-noise ratios were added to each gearbox fault signal to verify the anti-noise capability of the fault diagnosis method proposed in the present invention. The accuracy of the above four fault diagnosis methods under various signal-to-noise ratio conditions is shown in Table 4.
[0117] Table 4. Average test accuracy of different methods under different signal-to-noise ratio conditions
[0118]
[0119] It can be seen from the experimental results in Table 4 that when the white noise added to the vibration signal ranges from -2dB to 6dB, the method 2 proposed in the present invention achieves the highest diagnostic accuracy. In particular, when the signal-to-noise ratio decreases, the diagnostic accuracy of the traditional wavelet methods 3 and 4 decreases significantly. When the signal-to-noise ratio value is 0dB, both methods 3 and 4 have an average test accuracy of less than 60%. In contrast, the average test accuracy of the proposed method 1 is 95.17%, and the minimum standard deviation is 2.63. In addition, the average test accuracy of method 2 using the optimal feature combination entropy + skewness is higher than that of method 1 using the optimal feature combination standard deviation + kurtosis.
[0120] Experimental results under different label noise
[0121] In the actual mechanical fault diagnosis scenario, a small number of labels have noise. In order to evaluate the performance of the proposed method in this label noise environment, data sets with 0%, 2%, 5%, and 10% label noise ratios were prepared. Under each label noise ratio, 10 verification experiments were conducted, and then the average test accuracy of the experiment was used as the evaluation metric for the performance of different fault diagnosis methods. The experimental results are described in detail in Table 5.
[0122] Table 5. Average test accuracy of the proposed method under different label-to-noise ratio conditions
[0123]
[0124] As shown in Table 5, when the label-to-noise ratio changes from none to 0.1, the proposed method 2 and method 1 present higher average test accuracy than method 4 and method 3. When the label-to-noise ratio is 0.1, the average test accuracy of method 2 is above 91%. In contrast, the average diagnostic accuracy of methods 4 and 3 is only around 90%. This demonstrates that the proposed method has excellent fault identification and feature extraction capabilities in an environment with noisy labels.
[0125] Experimental results under limited fault sample conditions
[0126] In order to evaluate the performance of the method proposed in this invention under limited fault sample conditions, a small sample data set with sample numbers ranging from 10 to 100 is set, and the evaluation results are shown in Table 6.
[0127] Table 6. Average test accuracy of different methods under different limited fault samples
[0128]
[0129]
[0130] As shown in Table 6, the average test accuracy of the methods 1 and 2 proposed in the present invention is higher than that of the methods 3 and 4 when the number of fault samples is limited from 10 to 70. In particular, when the number of fault samples is 10, the average diagnostic accuracy of the proposed method is as high as more than 90%. This shows that the proposed method successfully overcomes the difficulty of the deep learning model being heavily dependent on sufficient labeled training samples, and exhibits good robustness in the case of limited fault samples.
[0131] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.
Claims
1. A gearbox composite fault diagnosis method based on optimal Legendre multi-wavelet frequency band, characterized by: At least the following steps are included: Step 1: Data collection, using the PHM 2009 composite gearbox fault dataset; Step 2: Initialize the algorithm; Step 3: Sample processing, 75% of the total samples of each gearbox fault category are allocated to training samples and 25% are allocated to test samples; Step 4: Legendre multi-wavelet decomposition process, including taking the initial values of network parameters and implementing Legendre multi-wavelet decomposition; Step 5: Improved adaptive threshold denoising algorithm; Step 6: Feature extraction: on each optimal Legendre multi-wavelet high-frequency band after denoising and the Legendre multi-wavelet low-frequency band without denoising, calculate the statistical feature combination standard deviation + kurtosis, entropy + skewness as the feature input of the BP neural network; Step 7: Train the BP neural network, normalize the optimal features obtained to between 0 and 1, and train the BP neural network through the gradient descent learning algorithm. After that, pass the normalized results of the optimal features obtained on the test sample to the BP neural network, and calculate the test diagnosis accuracy to complete the performance evaluation of the proposed method. Step 8: Fusion of genetic algorithm and improved adaptive threshold denoising algorithm. If the fault diagnosis accuracy does not reach the expected diagnostic accuracy, the decomposition level of multiwavelet, the number of Legendre multiwavelet, and the number of neurons of BP neural network are updated by genetic algorithm, and steps 4 to 7 are repeated to obtain higher fault diagnosis accuracy. Step 9: Output the results. If the fault diagnosis accuracy reaches the expected diagnostic accuracy, then output the highest diagnostic accuracy and the corresponding multi-wavelet decomposition level, the number of Legendre multi-wavelets, and the number of neurons in the BP neural network.
2. The gearbox composite fault diagnosis method based on optimal Legendre multi-wavelet frequency band according to claim 1 is characterized by: The gearbox model in the PHM 2009 composite gearbox fault data set is ER-10K; The gearbox fault signals in the PHM 2009 composite gearbox fault data set have 6 different health conditions. Each signal is collected under 5 conditions of low load and speed of 30, 35, 40, 45, and 50 Hz, respectively. The sampling frequency of each signal is 66.7 kHz and the sampling time is 8 s.
3. The gearbox composite fault diagnosis method based on optimal Legendre multi-wavelet frequency band according to claim 1 is characterized by: The initial algorithm at least includes the number of iterations of the initialization genetic algorithm, the number of iterations of the gradient descent algorithm, the decomposition level, the number of Legendre multi-wavelets, the number of hidden layer neurons of the BP neural network, the number of chromosomes, the crossover probability, the mutation probability, the training samples and the test samples, the iteration termination training error of the BP neural network and the expected fault diagnosis accuracy.
4. The gearbox composite fault diagnosis method based on optimal Legendre multi-wavelet frequency band according to claim 1 is characterized by: The step 5 at least comprises the following steps: S1: Noise estimation in the optimal high-frequency band of Legendre multiwavelets: Let W = [w1,w2,…,w N ] is a vector, which is composed of the squares of the Legendre multiwavelet coefficients of each optimal high-frequency band arranged in ascending order; First, let the vector W be the minimum value r of the risk vector R calculated according to formula (1) b (b th r) is used as an unbiased risk estimate, as shown in formula (1): Then, using the subscript b corresponding to the risk value, select the square w of the Legendre multiwavelet coefficient at the corresponding position in the vector W b As the minimum risk factor, Therefore, the improved threshold estimation is formula (2): S2: Threshold parameter selection based on experiments: According to formula (2), the Legendre multiwavelet coefficients of the optimal decomposition level are used to estimate the optimal threshold, where the parameter σ is the standard deviation of the optimal high-frequency band signal, and η is a parameter determined by numerical simulation experiments according to different application scenarios; S3: Improved soft threshold denoising algorithm: Based on the threshold estimation of formula (2), the improved soft threshold denoising algorithm is obtained as follows: in, is the constructed adaptive weight coefficient model. It should be noted that through this unbiased risk estimation, different thresholds λ can be obtained in different optimal Legendre multi-wavelet high-frequency bands, so that the improved algorithm is more helpful to remove complex noise interference; S4: Evaluation of improved denoising algorithm: The performance of the improved adaptive threshold denoising algorithm is evaluated by the root mean square error. The specific calculation formula of the improved adaptive threshold denoising algorithm by the root mean square error is formula (4): Among them, N0 is the number of sample points, x i is the i-th element of the original signal, y i is the i-th element of the denoised signal reconstructed using the inverse Legendre multiwavelet transform; The lower the RMSE value, the better the denoising performance of the denoising algorithm.
5. The gearbox composite fault diagnosis method based on optimal Legendre multi-wavelet frequency band according to claim 1 is characterized by: The fusion of the genetic algorithm and the improved adaptive threshold denoising algorithm is at least The following steps are involved: S1: The fitness function of the genetic algorithm adopts the expected accuracy of fault diagnosis, which is specifically formula (5): Where TP, TN, FP and FN represent true positive, true negative, false positive and false negative respectively; S2: The weight coefficient of the BP neural network is learned through the gradient descent algorithm. When the loss function reaches the predetermined lower limit or the number of iterations reaches the predetermined maximum value, the adjustment process of the network weight coefficient stops. At this time, if the fault diagnosis accuracy does not reach the target, the genetic algorithm is used to further select the better value of the network structure parameter; S3: Let GA(·) represent the optimization process of the genetic algorithm for the network structure parameters, GD(·) represent the BP neural network weight coefficient learning process, and Fitness(·) represent the fitness function of the genetic algorithm. The designed Max min objective function is formula (6): Where n, p, ξ represent the decomposition level, the number of Legendre multiwavelets and the number of hidden layer neurons of the BP neural network, respectively, and λ represents the optimal unbiased risk estimation threshold in each optimal Legendre multiwavelet high frequency band.
Citation Information
Patent Citations
Gearbox fault diagnosis method based on wavelet packet and depth auto-encoder
CN114818806A
Gearbox fault diagnosis method
CN115452362A
Fault diagnosis method for series hybrid electric vehicle ac / DC converter
US20190242936A1
Cited By
Steel surface defect classification method based on Legendre multi-wavelet transformation
CN116109865A
Frequency refinement diffusion model fault diagnosis method and system for data imbalance
CN120781232A
