Aero-engine inter-shaft bearing fault diagnosis method under quasi-unbalance condition based on Cycle-GAN

Abstract

The invention discloses a Cycle-GAN-based aero-engine inter-shaft bearing fault diagnosis method under a class imbalance condition. The method specifically comprises the steps of data acquisition and preprocessing, improvement and establishment of a Cycle-GAN model, training of the improved Cycle-GAN model and completion of signal conversion, and completion of fault diagnosis of actual vibration signals through the Cycle-GAN model. The invention provides a new fault diagnosis transfer learning method based on Cycle-GAN, and the method comprises the steps: converting a signal sample under a known condition into a new signal sample under an unknown condition through an improved Cycle-GAN model; according to the method, the conversion signals with data distribution closer to real signals are provided, meanwhile, it can be ensured that the conversion signals retain fault category information in original signals, the problem that fault data are scarce in an actual scene is solved, the classifier is trained through the conversion signals, and the classifier can distinguish fault data under unknown conditions. The method has a wide application prospect in practical engineering application, especially in the fault diagnosis problem of the bearing between the shafts of the aero-engine under the quasi-unbalance condition.

Description

Technical Field

[0001] The present invention belongs to the technical field of bearing fault diagnosis, and in particular relates to a method for diagnosing inter-axle bearing faults of an aero-engine under quasi-unbalanced conditions based on Cycle-GAN. Background Art

[0002] Aircraft engines are known as the "flower of industry" and are the core machines that provide power. As a key component in the dual-rotor system of aircraft engines, inter-shaft bearings play a core role in connection and support. However, the inter-shaft bearings of aircraft engines are installed between the high-pressure rotor and the low-pressure rotor. The unique installation position makes them stay in the harsh operating environment of high speed, high-intensity impact and heavy load for a long time, which often leads to local failure of the bearings. Such failures will cause the bearings to exhibit unique and complex vibration modes, which will directly affect the safety and reliability of the engine, leading to major safety accidents and endangering people's lives and property safety. Therefore, efficient and accurate inter-shaft bearing fault diagnosis methods are of great significance to ensure the stable operation of the engine. In recent years, deep learning has made great achievements in computer vision and natural language processing. With its powerful feature extraction capabilities, it has been widely used in fault diagnosis of industrial equipment and key components. The effectiveness of deep learning-based fault diagnosis methods is based on the assumption that data should obey the same distribution. However, in actual industrial environments, changes in working conditions such as load and speed will cause large differences in data distribution, which will lead to poor generalization of intelligent diagnosis models. Domain adaptation methods, as a subfield of transfer learning, are often used to solve this problem. They can transfer knowledge from a certain field to similar fields, narrow the data distribution differences between different fields, and provide many effective methods for cross-domain fault diagnosis problems, including maximum mean difference (MMD), domain adversarial neural network (DANN), and joint distribution adaptation (JDA). However, as a high-reliability industrial product, aero-engines usually operate in normal conditions, and mechanical failures are sporadic and uncertain, which means that normal bearing samples are abundant, while fault bearing samples are scarce. In another case, the number of samples in a certain fault state is much smaller than the number of samples in other fault states. It is difficult to collect enough aero-engine inter-shaft bearing fault samples and construct an ideal data set. Sample data between different health states will have class imbalance problems. When using class imbalance data sets to train models, accurate diagnosis results cannot be obtained. This is an important factor limiting the practical application of intelligent diagnosis models in industrial scenarios. In order to realize fault diagnosis under class imbalance conditions, many studies on class imbalance fault diagnosis have been conducted, among which generative adversarial networks (GANs) have shown great application potential in solving such problems. GAN is a data generation model that can generate new sample data similar to the real data distribution through adversarial training of the generator and the discriminator. As a variant model of GAN, the Cycle Generative Adversarial Network (Cycle-GAN) can convert labeled sample data from known scenarios to unknown scenarios in the absence of paired data, which is helpful to solve the problem of no data in actual fault diagnosis scenarios.The present invention is improved based on Cycle-GAN, and uses the improved Cycle-GAN model to convert signals under known conditions into signals under unknown conditions. While retaining data features and fault category information, the data distribution difference between the converted signal and the real signal is reduced, the quality and quantity of the converted signal are improved, and the fault diagnosis of the inter-shaft bearing of the aircraft engine under quasi-unbalanced conditions is realized. The present invention proposes an unsupervised transfer learning network model based on Cycle-GAN. The improved Cycle-GAN can be applied to sample data under different conditions, generate new samples under unknown conditions through sample data under known conditions, and use the new sample data to train the model classifier to realize the fault diagnosis of samples under unknown conditions. Summary of the invention

[0003] The purpose of the invention is to solve the fault diagnosis problem of aircraft engine inter-shaft bearings under the condition of scarce fault data. The present invention proposes a fault diagnosis method for aircraft engine inter-shaft bearings under quasi-unbalanced conditions based on Cycle-GAN.

[0004] The technical solution of the present invention is: a method for diagnosing the fault of an aircraft engine inter-shaft bearing under quasi-unbalanced conditions based on Cycle-GAN, comprising the following steps: A. Obtain vibration signal samples of the engine inter-axial bearing in normal and faulty states under different working conditions, and divide them into training data sets and test data sets; B. Improve and establish the Cycle-GAN model; C. Use the improved Cycle-GAN model to complete the conversion of vibration signals under known conditions to vibration signals under unknown conditions, and train the classifier of the Cycle-GAN model through the new conversion signal; D. Use the Cycle-GAN model trained in step C to diagnose the actual bearing vibration signal and obtain the result.

[0005] Furthermore, the step A obtains vibration signal samples of the engine inter-shaft bearing in normal state and fault state under different working conditions, and divides the training data set and the test data set, which specifically includes the following sub-steps: A1. Select the normal bearing signal under different working conditions as the source domain data (in Represents signal data, represents the corresponding class label), in order to simulate the fault diagnosis scenario under class imbalance conditions, the fault state bearing signal (a signal that is not easy to obtain in reality) under one of the corresponding working conditions is randomly selected as the target domain data ,The above two types of data together constitute the training data set,as signal samples under known conditions; A2. Select the bearing signals with the same fault state under the remaining working conditions in the corresponding working conditions as the test data set and as the signal samples under unknown conditions.

[0006] Furthermore, the step B improves and establishes a Cycle-GAN model, and the improved Cycle-GAN model includes two generators and , two discriminators and And two classifiers and , specifically B1. The generator of the improved Cycle-GAN model consists of an encoder, a converter and a decoder, wherein the encoder is composed of three stacked convolutional layers, the converter is composed of three residual blocks connected in sequence, in each residual block, the first convolutional layer is replaced by an expanded convolution, and the decoder is composed of three deconvolutional layers; B2. The discriminator of the improved Cycle-GAN model consists of four convolutional blocks and one convolutional layer, and each convolutional block consists of a convolutional layer, an instance normalization layer and a LeakyReLU activation function; B3. The classifier of the improved Cycle-GAN model is composed of a convolutional layer and a fully connected layer connected in sequence.

[0007] Furthermore, the step C uses the Cycle-GAN model improved in step B to complete the conversion of the vibration signal from known conditions to unknown conditions. During the training process, the two generators and The mapping direction is opposite to that of the source domain. It is necessary to transform the signal from the source domain to the target domain through forward transformation and reverse transformation, and train the classifier with the new transformation signal, which includes the following steps: C1. First pass the generator The source domain signal Convert to target domain conversion signal , then the generator Learning from the source domain ( ) to the target domain ( ) , and then use the discriminator Distinguishing conversion signals and the target domain real signal ,pass and The adversarial training makes The generated conversion signal Data distribution and target domain signal Highly consistent, last generator Then Convert to signal in source domain , then the generator Learning a mapping from the target domain to the source domain , source domain signal The positive transition should make the positive cycle Established; C2. First pass the generator The target domain signal Convert to source domain , then the generator Learning a mapping from the target domain to the source domain , and then use the discriminator Distinguishing conversion signals and the real signal in the source domain ,pass and Adversarial training enables The generated conversion signal Data distribution and source domain signal Highly consistent, last generator Then Convert to a signal in the target domain , then the generator Learning a mapping from the source domain to the target domain , target domain signal The reverse conversion should make the reverse cycle Established; C3. Through steps C1 and C2, the signal is converted from known conditions to unknown conditions, which ensures that the samples generated by the two generators are not contradictory and improves the similarity between the converted signal and the real signal. During the training process, the target domain classifier Used to transform signals to the target domain and the target domain real signal Classification is performed to further ensure that the target domain classifier Ability to correctly classify real signals under unknown conditions.

[0008] Furthermore, the step D uses the Cycle-GAN model trained in step C to diagnose the actual bearing vibration signal to obtain the result, specifically: the test data set divided in step A is used as the target domain classifier of the trained Cycle-GAN model The input of the engine shaft bearing fault diagnosis under quasi-unbalanced conditions is realized.

[0009] The beneficial effects of the present invention are as follows: the present invention proposes a new fault diagnosis transfer learning method based on Cycle-GAN, which converts signal samples under known conditions into new signal samples under unknown conditions through an improved Cycle-GAN model, provides a conversion signal whose data distribution is closer to the real signal, and ensures that the conversion signal retains the fault category information in the original signal, solves the problem of scarce fault data in actual scenarios, and uses the conversion signal to train the classifier so that the classifier can distinguish fault data under unknown conditions. In actual engineering applications, especially in the problem of fault diagnosis of inter-shaft bearings of aircraft engines under quasi-unbalanced conditions, the present invention has broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 It is a flow chart of a method for diagnosing inter-axial bearing faults of an aero-engine under quasi-unbalanced conditions based on Cycle-GAN of the present invention.

[0011] Figure 2 It is a schematic diagram of the improved Cycle-GAN model structure in the present invention. DETAILED DESCRIPTION

[0012] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0013] like Figure 1 As shown, a flow chart of a method for diagnosing a bearing fault between an aero-engine under quasi-unbalanced conditions based on Cycle-GAN is provided. A method for diagnosing a bearing fault between an aero-engine under quasi-unbalanced conditions based on Cycle-GAN comprises the following steps: A. Obtain vibration signal samples of the engine inter-axial bearing in normal and faulty states under different working conditions, and divide them into training data sets and test data sets; A1. Obtain bearing vibration signals under four different working conditions, specifically by collecting signals through sensors installed near the engine shaft bearings. The engine is often in normal operation, and the normal bearing signal samples are abundant, that is, the bearing vibration signals under four normal working conditions are selected as source domain data. Occasional faults lead to scarce fault signal samples, that is, a bearing signal under a working fault state is randomly selected as the target domain data. The above two types of data together constitute the training data set as signal samples under known conditions; A2. Select the bearing signals under the same fault state of the remaining three working conditions as the test data set and as the signal samples under unknown conditions. Divide the data set according to steps A1 and A2 to simulate the fault diagnosis scenario under class imbalance conditions.

[0014] B. Improve and establish the Cycle-GAN model, such as Figure 2 As shown, the improved Cycle-GAN model includes two generators and , two discriminators and And two classifiers and ; B1. The generator of the improved Cycle-GAN model consists of an encoder, a converter and a decoder. The signal input to the generator first enters the encoder composed of three stacked convolutional layers to extract signal features; then enters the converter composed of three residual blocks, and uses dilated convolution to replace the first convolution layer of each residual block to further expand the receptive field of the convolution kernel without increasing network parameters. The dilation rate is set to 2, so that the generator can extract deeper features during the signal conversion process. The extracted signal features are converted from the source domain to the target domain through the converter; the converted features finally enter the decoder composed of three deconvolutional layers, and the decoder restores the learned features and reversely synthesizes them into the final required conversion signal; B2. The discriminator of the improved Cycle-GAN model consists of four convolution blocks and one convolution layer. Each convolution block consists of a convolution layer, an instance normalization layer, and a LeakyReLU activation function. This structure is conducive to the discriminant to perform discriminant analysis on each small segment of the signal, and can guide the model to pay attention to more details of the signal during training. At this time, the role of the discriminator is to distinguish between real signals and conversion signals. Through adversarial training between the discriminator and the generator, the generator generates conversion signals that are highly similar to real signals. B3. The classifier of the improved Cycle-GAN model is composed of three convolutional layers and one fully connected layer connected in sequence, and the test signal is input into the classifier to obtain the final fault diagnosis result; B4. The size of the convolution kernel of the first convolution layer of the generator, discriminator and classifier is 64×1. The large convolution kernel can help the convolution layer to better extract the global domain features of the bearing signal. Each convolution layer is connected to an instance normalization layer, which can normalize the feature channels, ensure the independence of the samples, and help align the feature distribution, thereby improving the stability of the model training. The LeakyReLU activation function is used to replace the initial rectified linear unit (ReLU) activation function to enhance the feature extraction capability and improve the robustness and computational efficiency of the model.

[0015] C. According to the data set division method in step A1, the Cycle-GAN model is trained with the source domain training data and the target domain training data. During the training process, the generator learns the mapping and , taking the respective source domain signals as input, respectively learning the data distribution of the corresponding target domain signals, realizing the conversion of signals from the source domain to the target domain, that is, completing the conversion of vibration signals under known conditions to vibration signals under unknown conditions through forward and reverse conversion of signals, and the new data after conversion (bearing signals under four working fault conditions) should correspond to the bearing signals under normal conditions, and finally using the new conversion signals to train the classifier of the Cycle-GAN model; C1. Through the generator The source domain signal Convert to target domain conversion signal , and then use the discriminator Distinguishing conversion signals and the target domain real signal ,pass and The adversarial training makes The generated conversion signal Data distribution and target domain signal Highly consistent, generator Then Convert to signal in source domain , source domain signal The positive transition should make the positive cycle The loss back-propagated from the discriminator will help the generator and the discriminator to update their parameters. and The adversarial training loss between them is expressed in the form of cross entropy as: {L}_{GAN\left ( {{G}_{st},{D}_{t},S,T} \right )}={E}_{{x}_{t}\sim {P}_{T}\left ( {{x}_{t}} \right )}\left [ {log{D}_{t}\left ( {{x}_{t}} \right )} \right ]+{E}_{{x}_{s}\sim {P}_{s}\left ( {{x}_{s}} \right )}\left [ {log\left ( {1-{D}_{t}\left ( {{x}^{\^}_{t}} \right )} \right )} \right ] (1) In the formula, and Respectively and The expectations are respectively subject to Distribution and distributed.

[0016] C2. Through the generator The target domain signal Convert to source domain , and then use the discriminator Distinguishing conversion signals and the real signal in the source domain ,pass and Adversarial training enables The generated conversion signal Data distribution and source domain signal Highly consistent, generator Then Convert to a signal in the target domain , target domain signal The reverse conversion should make the reverse cycle Established; and The adversarial training loss between them is expressed in the form of cross entropy as: {L}_{GAN\left ( {{G}_{ts},{D}_{s},S,T} \right )}={E}_{{x}_{s}\sim {P}_{S}\left ( {{x}_{s}} \right )}\left [ {log{D}_{s}\left ( {{x}_{s}} \right )} \right ]+{E}_{{x}_{t}\sim {P}_{T}\left ( {{x}_{t}} \right )}\left [ {log\left ( {1-{D}_{s}\left ( {{x}^{\^}_{s}} \right )} \right )} \right ] (2) When both the forward loop and the reverse loop are established, the model is considered to have completed the learning of the cycle-consistent mapping function. By constraining the training process through the cycle-consistency loss function, it is possible to prevent the loss of a large amount of useful information during the training process to improve the similarity between the conversion signal and the real signal and ensure that the generator and The generated samples do not contradict each other, and the cycle consistency loss function is expressed as follows: \right ]+{E}_{{x}_{t}\sim {P}_{T}\left ( {{x}_{t}} \right )}\left [ {{||{x}^{\~}_{t}-{x}_{t}||}_{1}} \right ] (3) In the formula, express Norm.

[0017] C3. Steps C1 and C2 realize the conversion of the signal from known conditions to unknown conditions. In addition, in order to make the converted signal consistent with the real signal of the same category, the source domain classifier The training is done to ensure that the signal retains the category information in the original signal during the conversion process, using the standard cross entropy To train: (4) In the formula, and Represents the number of samples of source domain and target domain signals respectively, and denote the source domain and target domain class label spaces respectively.

[0018] During the training process, the target domain classifier Used to transform signals to the target domain and the target domain real signal Classification is performed to further ensure that the target domain classifier It can correctly classify the real signal under unknown conditions, through the standard cross entropy To train: (5) In the formula, represents the class label space of the outlier source domain (classes belonging to the source domain but not to the target domain), Represents the number of outlier source domain signal samples.

[0019] D. Input the test data (real fault signal) obtained according to the data set division method in step A2 into the trained Cycle-GAN model target domain classifier In this paper, the fault diagnosis of engine inter-shaft bearings under quasi-unbalanced conditions is realized.

[0020] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.

Claims

1. A method for diagnosing inter-axle bearing faults of aircraft engines under quasi-unbalanced conditions based on Cycle-GAN, characterized in that: The following steps are involved: A. Obtain vibration signal samples of the engine inter-shaft bearing in normal state and different fault states under different working conditions, and divide them into training data set and test data set; B. Improve and establish the Cycle-GAN model; C. Use the improved Cycle-GAN model to complete the conversion of vibration signals under known conditions to vibration signals under unknown conditions, and train the classifier of the Cycle-GAN model through the new conversion signal; D. Use the Cycle-GAN model trained in step C to diagnose the actual bearing vibration signal and obtain the result.

2. The method for diagnosing the inter-shaft bearing fault of an aircraft engine under quasi-unbalanced conditions based on Cycle-GAN as claimed in claim 1 is characterized in that: The step A obtains vibration signal samples of the engine inter-axial bearing in normal state and fault state under different working conditions, and divides the training data set and the test data set, which specifically includes the following sub-steps: A1. Select the normal bearing signal under different working conditions as the source domain data (in Represents signal data, represents the corresponding class label), in order to simulate the fault diagnosis scenario under class imbalance conditions, the fault state bearing signal (a signal that is not easy to obtain in reality) under one of the corresponding working conditions is randomly selected as the target domain data ,The above two types of data together constitute the training data set,as signal samples under known conditions; A2. Select the bearing signals with the same fault state under the remaining working conditions in the corresponding working conditions as the test data set and as the signal samples under unknown conditions.

3. The method for diagnosing the inter-shaft bearing fault of an aircraft engine under quasi-unbalanced conditions based on Cycle-GAN as claimed in claim 1, characterized in that: The step B improves and establishes the Cycle-GAN model. The improved Cycle-GAN model includes two generators: and , two discriminators and And two classifiers and , specifically: B1. The generator of the improved Cycle-GAN model consists of an encoder, a converter and a decoder, wherein the encoder is composed of three stacked convolutional layers, the converter is composed of three residual blocks connected in sequence, in each residual block, the first convolutional layer is replaced by an expanded convolution, and the decoder is composed of three deconvolutional layers; B2. The discriminator of the improved Cycle-GAN model consists of four convolutional blocks and one convolutional layer, and each convolutional block consists of a convolutional layer, an instance normalization layer, and a leaky rectified linear unit (LeakyReLU) activation function; B3. The classifier of the improved Cycle-GAN model is composed of a convolutional layer and a fully connected layer connected in sequence.

4. The method for diagnosing the inter-shaft bearing fault of an aircraft engine under quasi-unbalanced conditions based on Cycle-GAN as claimed in claim 1, characterized in that: The step C completes the conversion of the vibration signal from known conditions to unknown conditions through the Cycle-GAN model constructed in step B. During the training process, the two generators and The mapping direction is opposite to that of the source domain. It is necessary to transform the signal from the source domain to the target domain through forward transformation and reverse transformation, and train the classifier with the new transformation signal, which includes the following steps: C1. First pass the generator The source domain signal Convert to target domain conversion signal , then the generator Learning from the source domain ( ) to the target domain ( ) , and then use the discriminator Distinguishing conversion signals and the target domain real signal ,pass and The adversarial training makes The generated conversion signal Data distribution and target domain signal Highly consistent, last generator Then Convert to signal in source domain , then the generator Learning a mapping from the target domain to the source domain , source domain signal The positive transition should make the positive cycle Established; C2. First pass the generator The target domain signal Convert to source domain , then the generator Learning a mapping from the target domain to the source domain , and then use the discriminator Distinguishing conversion signals and the real signal in the source domain ,pass and Adversarial training enables The generated conversion signal Data distribution and source domain signal Highly consistent, last generator Then Convert to a signal in the target domain , then the generator Learning a mapping from the source domain to the target domain , target domain signal The reverse conversion should make the reverse cycle Established; C3. Through steps C1 and C2, the signal is converted from known conditions to unknown conditions, which ensures that the samples generated by the two generators are not contradictory and improves the similarity between the converted signal and the real signal. During the training process, the target domain classifier Used to transform signals to the target domain and the target domain real signal Classification is performed to further ensure that the target domain classifier Ability to correctly classify real signals under unknown conditions.

5. The method for diagnosing the inter-shaft bearing fault of an aircraft engine under quasi-unbalanced conditions based on Cycle-GAN as claimed in claim 1, characterized in that: The step D uses the Cycle-GAN model trained in step C to diagnose the actual bearing vibration signal to obtain the result, specifically: the test data set divided in step A is used as the target domain classifier of the trained Cycle-GAN model The input of the engine shaft bearing fault diagnosis under quasi-unbalanced conditions is realized.

Images