Fault diagnosis method and system based on time sequence data enhancement of analytic wavelet

Abstract

The invention relates to the field of fault diagnosis, and particularly discloses a fault diagnosis method and system based on time sequence data enhancement of analytic wavelets. The method comprises the following steps: acquiring an original vibration signal during the operation of mechanical equipment, and dividing the original vibration signal into a training data set and a test data set; performing generalized Morse analysis wavelet transform on the original vibration signals in the training data set to generate an original scale map; determining parameters to be optimized of generalized Morse wavelet transform; using the optimized parameters to perform generalized Morse analysis wavelet transform on the original vibration signals in the training data set again to generate an enhanced scale map; combining the original scale image with the enhanced scale image to form an expanded training data set; inputting the expanded training data set into a convolutional neural network for training, and obtaining a fault diagnosis model after training convergence; and performing generalized Morse analysis wavelet transform on the original vibration signals in the test data set, inputting the original vibration signals into the fault diagnosis model, and outputting a fault diagnosis result.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis, and in particular to a fault diagnosis method and system based on time series data augmentation using analytic wavelets. Background Technology

[0002] In the field of machinery, rolling bearings are core components of rotating machinery systems, and their operating condition directly affects the overall performance and safety of the equipment. Because rolling bearings are susceptible to wear, fatigue, and other factors, a failure can lead to serious economic losses or even safety accidents. Therefore, early and accurate fault detection is crucial for ensuring the safe operation of machinery.

[0003] Traditional mechanical fault diagnosis methods primarily rely on signal processing techniques, including blind source separation and wavelet analysis. With the development of artificial intelligence, the application of deep neural networks in fault diagnosis has significantly improved diagnostic accuracy. However, the performance of deep learning models is highly dependent on the support of a large amount of labeled data, while in real-world industrial scenarios, fault data is often scarce. This is because mechanical equipment is designed to avoid faults, rather than intentionally creating faults to collect data, making it difficult to obtain the key fault data needed to train robust diagnostic models. Furthermore, the diversity of fault types further increases the complexity of data collection, making it difficult to meet the training requirements of deep neural networks using only limited measured fault data.

[0004] To address the problem of data scarcity, data augmentation techniques have emerged as a promising solution. Data augmentation generates additional training samples by performing specific transformations on the original data, while preserving the semantic validity of the data. Common data augmentation methods include geometric transformations, noise addition, and generative models based on generative adversarial networks (GANs). However, there are relatively few augmentation methods for time series data, especially those based on GANs, which suffer from low generation efficiency and insufficient sample reliability, making it difficult to effectively generate augmented data that reflects the characteristics of the original vibration signals.

[0005] Therefore, there is an urgent need for an efficient and reliable time series data augmentation method to improve the training effect and diagnostic accuracy of mechanical fault diagnosis models. Summary of the Invention

[0006] The purpose of this invention is to address the shortcomings of existing technologies by proposing a fault diagnosis method and system based on time series data enhancement using analytic wavelets. To achieve the above objective, the embodiments of this invention employ the following technical solutions:

[0007] In a first aspect, embodiments of the present invention propose a fault diagnosis method based on time series data augmentation using analytic wavelets, comprising the following steps:

[0008] S1: Collect the raw vibration signals of the mechanical equipment during operation, and divide the raw vibration signals into training dataset and test dataset;

[0009] S2: Perform a generalized Morse analytical wavelet transform on the original vibration signals in the training dataset to generate the original scale map; the formula for the generalized Morse analytical wavelet transform is:

[0010]

[0011] In the formula, ψ γ，β (t) is the generalized Morse wavelet function, A γ，β ω0 is the normalization constant, t is the time variable, ω0 is the center frequency, γ is the initial set value, and β is the parameter to be optimized.

[0012] S3: Determine the parameter β to be optimized in the generalized Morse wavelet transform based on the Heisenberg uncertainty principle; where the Heisenberg area is defined as: A = a × b

[0013] In the formula, a is the time-domain variance, b is the frequency-domain variance, and β affects the concentration of the generalized Morse wavelet in the time and frequency domains. By adjusting the parameter β to minimize the Heisenberg area A, the parameter to be optimized and the optimized parameter β are determined. opt ;

[0014] S4: Use the optimized parameter β opt The generalized Morse analytical wavelet transform is performed again on the original vibration signals in the training dataset to generate an enhanced scaling map.

[0015] S5: Merge the original scale map generated in step S2 with the enhanced scale map generated in step S4 to form the expanded training dataset;

[0016] S6: Input the expanded training dataset into the convolutional neural network for training. After the training converges, the fault diagnosis model is obtained.

[0017] S7: Perform generalized Morse analytical wavelet transform on the original vibration signals in the test dataset and input them into the fault diagnosis model to output the fault diagnosis results of the mechanical equipment.

[0018] Preferably, the initial value of the parameter β to be optimized is set to 20, and the initial value of γ is set to 3; wherein, the parameter β is adjusted stepwise with a step size Δβ, the Heisenberg area A corresponding to β after each adjustment is calculated, and the β corresponding to the minimum Heisenberg area is determined. opt .

[0019] Preferably, the parameter selection step for the enhanced scale map includes: selecting the optimized parameter β opt The difference between the initial value of 20 and the parameter to be optimized β is defined as the deviation δ, and the enhanced scale map is generated using β = 20 ± δ.

[0020] Preferably, the convolutional neural network includes: three cascaded convolutional layers, each convolutional layer being followed by a ReLU activation function layer and a pooling layer; and a fully connected layer and a Softmax classifier connected after the last pooling layer.

[0021] Preferably, the fault diagnosis results include one of the following categories: normal condition, inner ring fault, outer ring fault, or rolling element fault.

[0022] Preferably, the criterion for determining training convergence is: using a validation set to evaluate the classification accuracy of the convolutional neural network, and determining that training has converged when the classification accuracy no longer significantly improves.

[0023] Preferably, the loss function during training is the cross-entropy loss, and its calculation formula is as follows:

[0024]

[0025] In the formula, N is the number of samples participating in the training, and y i p is the true label of sample i. i Let be the predicted probability of sample i.

[0026] Secondly, embodiments of the present invention propose a fault diagnosis system based on time series data augmentation using analytic wavelets, comprising:

[0027] The data acquisition module is used to collect raw vibration signals during the operation of mechanical equipment;

[0028] The data processing module is used to divide the original vibration signal into a training dataset and a test dataset, and to perform a generalized Morse analytical wavelet transform on the training dataset to generate the original scale map.

[0029] The parameter optimization module is used to optimize the parameter β of the generalized Morse wavelet transform based on the Heisenberg uncertainty principle, so that the energy of the original scale map is concentrated in the time-frequency domain.

[0030] The data augmentation module is used to generate an augmented scale map using the optimized parameter β, and then merge it with the original scale map to form an expanded training dataset.

[0031] The model training module is used to train a convolutional neural network on the expanded training dataset to obtain a fault diagnosis model.

[0032] The diagnostic module is used to input the original vibration signals of the test dataset into the fault diagnosis model after generalized Morse analytical wavelet transform, and output the fault diagnosis results of mechanical equipment.

[0033] Preferably, the parameter optimization module sets the initial value of the parameter β to be optimized to 20, calculates the corresponding Heisenberg area by gradually adjusting the parameter δ, and selects the parameter β that minimizes the Heisenberg area. opt .

[0034] Preferably, the data augmentation module uses the optimized parameter β opt The difference between the initial value 20 and the parameter to be optimized β is defined as the deviation δ, and an enhanced scale map is generated with β = 20 ± δ.

[0035] Beneficial Effects: This invention proposes a fault diagnosis method based on analytic wavelet-based time-series data augmentation. It generates original and augmented scale maps using generalized Morse analytic wavelet transform, significantly expanding the training dataset. Optimizing the parameters of the generalized Morse wavelet based on the Heisenberg uncertainty principle concentrates signal energy in the time-frequency domain, resulting in augmented scale maps that better preserve the time-frequency characteristics of the original vibration signal, thus improving the accuracy of fault feature extraction. Compared to traditional data augmentation methods based on generative adversarial networks, directly generating augmented scale maps by adjusting wavelet transform parameters is highly efficient and reliable, closely reflecting the characteristics of the original signal and suitable for augmenting time-series data. By merging the original and augmented scale maps to form a comprehensive training dataset and training it with a convolutional neural network, the model's classification accuracy for various faults, including normal states, inner race faults, outer race faults, and rolling element faults, is effectively improved, providing higher reliability and practicality for fault diagnosis of mechanical equipment. The generalized Morse wavelet transform, through parameter adjustment, can adapt to different types of vibration signals, exhibiting high flexibility and wide applicability to various mechanical equipment fault diagnosis scenarios. Attached Figure Description

[0036] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0037] Figure 1 This is a flowchart illustrating a fault diagnosis and evaluation method based on time series data augmentation using analytic wavelets, provided in an embodiment of the present invention.

[0038] Figure 2 This is a schematic diagram of the structure of a fault diagnosis system based on analytic wavelet time series data enhancement provided in an embodiment of the present invention. Detailed Implementation

[0039] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention. It should be noted that similar reference numerals and letters in the following drawings denote similar items; therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0040] For the first aspect, please refer to... Figure 1 This invention proposes a fault diagnosis method based on analytic wavelet-based time series data augmentation, which is applied to a fault diagnosis system based on analytic wavelet-based time series data augmentation. The fault diagnosis system based on analytic wavelet-based time series data augmentation can be executed by, but is not limited to, computer devices with certain computing resources, such as personal computers (PCs, which are multi-purpose computers of a size, price, and performance suitable for personal use; desktops, laptops, mini-laptops, tablets, and ultrabooks are all considered personal computers), smartphones, personal digital assistants (PADs), or platform servers.

[0041] This embodiment details a data augmentation method based on analytic wavelet transform for fault diagnosis of mechanical equipment, particularly suitable for diagnosing critical mechanical components such as rolling bearings. Rolling bearings, as a crucial component of rotating machinery, are highly susceptible to damage leading to equipment failure, which can result in significant economic losses. Therefore, timely and accurate fault detection of bearings is essential. However, in practical applications, traditional diagnostic methods struggle to train robust deep neural network models due to the limited number and difficulty in collecting fault samples. To address this issue, this embodiment proposes a data augmentation method based on analytic wavelet transform to increase the amount of training data and improve the model's diagnostic accuracy. The specific method includes the following steps:

[0042] Step S1: Collect the raw vibration signals of the mechanical equipment during operation, and divide the raw vibration signals into training dataset and test dataset;

[0043] First, vibration sensors are used to collect vibration signals from mechanical equipment (such as rolling bearings). The raw signals are then preprocessed to remove noise and interference. Next, the processed signals are divided into training and testing datasets. The training dataset is used to train the model, while the testing dataset is used to verify the model's diagnostic performance.

[0044] Step S2: Perform a generalized Morse analytical wavelet transform on the original vibration signals in the training dataset to generate the original scale map; the formula for the generalized Morse analytical wavelet transform is:

[0045]

[0046] In the formula, ψ γ，β (t) is the generalized Morse wavelet function, A γ，β ω0 is the normalization constant, t is the time variable, ω0 is the center frequency, γ is the initial set value, and β is the parameter to be optimized.

[0047] The vibration signals in the training dataset are processed using the generalized Morse analytic wavelet transform. The definition of the generalized Morse wavelet is as follows:

[0048]

[0049] Among them, A γ，β The parameter γ is a normalization constant to ensure that the wavelet meets certain energy normalization conditions; the parameter γ is an initial setting value, which can be fixed at 3; the initial value of the parameter β to be optimized is 20, so as to obtain the initial original scale map sample.

[0050] Step S3: Optimize the generalized Morse wavelet parameter β. Determine the parameter β to be optimized in the generalized Morse wavelet transform according to the Heisenberg uncertainty principle; whereby the Heisenberg area is defined as: A = a × b

[0051] In the formula, a is the time-domain variance, b is the frequency-domain variance, and β affects the concentration of the generalized Morse wavelet in the time and frequency domains. By adjusting the parameter β to minimize the Heisenberg area A, the parameter to be optimized and the optimized parameter β are determined. opt ;

[0052] Based on the Heisenberg uncertainty principle, the Heisenberg area A is defined as the product of the signal's time-domain variance a and frequency-domain variance b:

[0053] A = a × b

[0054] By adjusting the parameter β, the energy concentration of the scale map in the time-frequency domain is increased, thus minimizing the Heisenberg area A. Specifically, starting with an initial value β = 20, the parameter β is gradually adjusted, and the Heisenberg area corresponding to each adjustment is calculated. Finally, the parameter value β that minimizes the Heisenberg area is selected. opt .

[0055] Step S4, use the optimized parameter β opt The generalized Morse analytical wavelet transform is performed again on the original vibration signals in the training dataset to generate an enhanced scaling map.

[0056] Using the optimization parameter β determined aboveopt Then, a generalized Morse analytical wavelet transform was performed again on the vibration signals in the training dataset to obtain enhanced scale map samples. These enhanced scale maps effectively enrich the data.

[0057] Step S5: Merge the original scale map generated in step S2 with the enhanced scale map generated in step S4 to form the expanded training dataset.

[0058] The original scale map and the enhanced scale map are merged to form an expanded, comprehensive training dataset. The expanded dataset contains more diverse features, effectively improving the robustness of subsequent model training.

[0059] S6: Input the expanded training dataset into the convolutional neural network for training. After the training converges, the fault diagnosis model is obtained.

[0060] This embodiment uses a convolutional neural network for model training. The network architecture consists of three cascaded convolutional layers, each followed by a ReLU activation function and a pooling layer, then a fully connected layer for feature integration, and finally a softmax classifier for classification and diagnosis. During training, an expanded dataset is used, and training is considered converged when the model's classification accuracy on the validation set no longer significantly improves. The specific loss function used is cross-entropy loss.

[0061]

[0062] In the formula, N is the number of samples participating in the training, and y i p is the true label of sample i. i Let be the predicted probability of sample i.

[0063] S7: Perform generalized Morse analytical wavelet transform on the original vibration signals in the test dataset and input them into the fault diagnosis model to output the fault diagnosis results of the mechanical equipment.

[0064] After the model is trained, the vibration signals from the test dataset are processed using the same generalized Morse analytical wavelet transform and then input into the model. The model outputs mechanical equipment fault diagnosis results. The diagnosis results include four categories: normal state, inner ring fault, outer ring fault, and rolling element fault.

[0065] By using the above method, this embodiment effectively solves the bottleneck of insufficient data in traditional mechanical fault diagnosis and significantly improves the accuracy and generalization ability of the fault diagnosis model.

[0066] For the second aspect, please refer to... Figure 2This embodiment provides a schematic diagram of a system structure for implementing the above-described mechanical fault diagnosis method. The system includes a data acquisition module, a data processing module, a parameter optimization module, a data augmentation module, a model training module, and a diagnosis module, as detailed below:

[0067] The data acquisition module is used to collect raw vibration signals from the mechanical equipment during operation. The signals are acquired by a high-sensitivity sensor, and after preliminary noise reduction processing, they are transmitted to the data processing module.

[0068] The data processing module is used to divide the original vibration signal into training and testing datasets. Then, the generalized Morse analytical wavelet transform is used to convert the training data into the original scale map, providing basic data for subsequent data augmentation steps.

[0069] The parameter optimization module, based on the Heisenberg uncertainty principle, automatically optimizes the parameter β of the generalized Morse wavelet transform by defining and calculating the Heisenberg area, selecting the optimal parameter that maximizes the concentration of time-frequency energy in the original scale map. In the specific implementation, the initial value of the parameter β is set to 20. Subsequently, the parameter is adjusted with a fixed step size, and the corresponding Heisenberg area is calculated to finally determine the optimal parameter β. opt .

[0070] The data augmentation module utilizes the optimal parameter β determined by the parameter optimization module. opt The training dataset is subjected to analytical wavelet transform again to generate an enhanced scale map. Subsequently, the generated enhanced scale map is fused with the original scale map to form an expanded training dataset, effectively increasing the diversity and richness of the data samples and improving the model training effect. Preferably, the data augmentation module uses the optimized parameter β... opt The difference between the initial value 20 and the parameter to be optimized β is defined as the deviation δ, and an enhanced scale map is generated with β = 20 ± δ.

[0071] The model training module is responsible for training the convolutional neural network on the expanded training dataset. The network structure includes three cascaded convolutional layers followed by ReLU activation function layers, pooling layers, fully connected layers, and a Softmax classifier. The training process continues until the classification accuracy on the validation set stabilizes, i.e., the training converges.

[0072] The diagnostic module uses a convolutional neural network trained by the model training module to diagnose data in the test dataset after generalized Morse analytical wavelet transform. The module outputs real-time fault diagnosis results for the mechanical equipment and performs classification and identification, including normal status, inner race fault, outer race fault, or rolling element fault, helping users monitor equipment status and perform timely maintenance.

[0073] The mechanical fault diagnosis system constructed in this embodiment realizes a complete automated process of data acquisition, processing, enhancement and diagnosis, solves the problem of insufficient data in traditional methods, improves the accuracy and practicality of fault diagnosis, and meets the actual needs of fault diagnosis of complex mechanical systems.

[0074] Thirdly, embodiments of the present invention also provide a computer-readable storage medium storing instructions comprising the fault diagnosis method based on analytic wavelet time-series data augmentation of the first embodiment. That is, the computer-readable storage medium stores instructions that, when executed on a computer, perform the fault diagnosis method based on analytic wavelet time-series data augmentation as described in the first aspect. The computer-readable storage medium refers to a data storage medium, which may include, but is not limited to, floppy disks, optical disks, hard disks, flash memory, USB flash drives, and / or Memory Sticks. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0075] Fourthly, embodiments of the present invention provide a computer program product including a non-transitory computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, cause the processor to perform the method proposed in the above embodiments.

[0076] The present invention also discloses a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the fault diagnosis method based on time series data enhancement of analytical wavelet provided by the present invention. The computer program product should be understood as a software product that mainly implements its solution through a computer program, such as a program product integrated in the cloud or software library.

[0077] In the description of this specification, the references to terms such as "an embodiment," "some embodiments," "example," "specific example," "a implementation," "a preferred implementation," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0078] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A fault diagnosis method based on time series data augmentation using analytic wavelets, characterized in that, Includes the following steps: S1: Collect the raw vibration signals of the mechanical equipment during operation, and divide the raw vibration signals into a training dataset and a test dataset; S2: Perform a generalized Morse analytical wavelet transform on the original vibration signals in the training dataset to generate the original scale map; the formula for the generalized Morse analytical wavelet transform is: In the formula, ψ γ，β (t) is the generalized Morse wavelet function, A γ，β ω0 is the normalization constant, t is the time variable, ω0 is the center frequency, γ is the initial set value, and β is the parameter to be optimized. S3: Determine the optimization parameter β of the generalized Morse wavelet transform based on the Heisenberg uncertainty principle; where the Heisenberg area is defined as: A = a × b In the formula, a is the time-domain variance, b is the frequency-domain variance, and β affects the concentration of the generalized Morse wavelet in the time and frequency domains. By adjusting the parameter β to minimize the Heisenberg area A, the parameter to be optimized and the optimized parameter β are determined. opt ; S4: Use the optimized parameter β opt The generalized Morse analytical wavelet transform is performed again on the original vibration signals in the training dataset to generate an enhanced scaling map. S5: Merge the original scale map generated in step S2 with the enhanced scale map generated in step S4 to form the expanded training dataset; S6: Input the expanded training dataset into the convolutional neural network for training. After the training converges, the fault diagnosis model is obtained. S7: Perform a generalized Morse analytical wavelet transform on the original vibration signal in the test dataset and input it into the fault diagnosis model to output the fault diagnosis result of the mechanical equipment.

2. The method according to claim 1, characterized in that, The initial value of the parameter β to be optimized is set to 20, and the initial value of γ is set to 3. The parameter β is adjusted stepwise with a step size Δβ. The Heisenberg area A corresponding to β after each adjustment is calculated, and the β corresponding to the minimum Heisenberg area is determined. opt .

3. The method according to claim 2, characterized in that, The parameter selection step for the enhanced scale map includes: selecting the optimized parameter β. opt The difference between the initial value of 20 and the parameter to be optimized β is defined as the deviation δ, and the enhanced scale map is generated using β = 20 ± δ.

4. The method according to claim 1, characterized in that, The convolutional neural network includes: three cascaded convolutional layers, each followed by a ReLU activation function layer and a pooling layer; and a fully connected layer and a Softmax classifier connected after the last pooling layer.

5. The method according to claim 1, characterized in that, The fault diagnosis results include one of the following categories: normal condition, inner ring fault, outer ring fault, or rolling element fault.

6. The method according to claim 1, characterized in that, The criteria for determining training convergence are as follows: the classification accuracy of the convolutional neural network is evaluated using a validation set, and training convergence is determined when the classification accuracy no longer significantly improves.

7. The method according to claim 6, characterized in that, The loss function used in the training process is cross-entropy loss, and its calculation formula is as follows: In the formula, N is the number of samples participating in the training, and y i p is the true label of sample i. i Let be the predicted probability of sample i.

8. A fault diagnosis system based on analytic wavelet-based time series data augmentation, characterized in that, include: The data acquisition module is used to collect raw vibration signals during the operation of mechanical equipment; The data processing module is used to divide the original vibration signal into a training dataset and a test dataset, and to perform a generalized Morse analytical wavelet transform on the training dataset to generate the original scale map. The parameter optimization module is used to optimize the parameter β of the generalized Morse wavelet transform based on the Heisenberg uncertainty principle, so that the energy of the original scale map is concentrated in the time-frequency domain. The data augmentation module is used to generate an augmented scale map using the optimized parameter β, and then merge it with the original scale map to form an expanded training dataset. The model training module is used to train a convolutional neural network on the expanded training dataset to obtain a fault diagnosis model. The diagnostic module is used to input the original vibration signal of the test dataset into the fault diagnosis model after passing it through a generalized Morse analytical wavelet transform, and output the fault diagnosis results of the mechanical equipment.

9. The system according to claim 8, characterized in that, The parameter optimization module sets the initial value of the parameter β to be optimized to 20, calculates the corresponding Heisenberg area by gradually adjusting the parameter δ, and selects the parameter β that minimizes the Heisenberg area. opt .

10. The system according to claim 9, characterized in that, The data enhancement module uses the optimized parameter β opt The difference between the initial value 20 and the parameter to be optimized β is defined as the deviation δ, and an enhanced scale map is generated with β = 20 ± δ.

Images