Fault sample generation method and device for main bearing of turboshaft turboprop engine
By generating a standard grayscale image of the main bearing and training the target generator, the problem of insufficient fault samples in turboshaft and turboprop engines was solved, improving the accuracy and generalization ability of fault diagnosis.
Patent Information
- Application Number
- CN202511167267.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-19
- Publication Date
- 2026-01-09
AI Technical Summary
In the existing technology, the acquisition of main bearing fault samples of turboshaft and turboprop engines mainly relies on real-time acquisition, but most of them are normal samples, which makes the fault diagnosis model training insensitive and unable to be effectively applied to fault diagnosis.
By acquiring standard fault samples of the main bearing, a standard grayscale image is generated. Based on the grayscale image and fault labels, a fault sample generation model is trained. A target generator is constructed using a generator and a discriminator to generate high-quality fault samples that retain physical authenticity and feature distribution.
The training dataset was effectively expanded, improving the accuracy and generalization ability of main bearing fault diagnosis. The generated fault samples met the feature distribution and physical authenticity of the specified fault category.
Smart Images
Figure CN121301913A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of fault diagnosis technology, specifically to a method and apparatus for generating fault samples of the main bearing of a turboshaft / turboprop engine. Background Technology
[0002] As a crucial component of the rotor system of a turboshaft / turboprop engine, the main bearing's failure can lead to engine performance degradation and even trigger various serious malfunctions. Fault diagnosis methods utilizing artificial intelligence algorithms such as deep learning can automatically extract fault features from vibration signals, enabling fault model identification. However, this method requires a sufficient number of fault samples to train and model the fault diagnosis system.
[0003] In practical engineering, the acquisition of main bearing fault samples in turboshaft and turboprop engines mainly relies on real-time data acquisition. However, most of the acquired samples are normal main bearings, with few faulty samples. This results in data-driven bearing fault diagnosis being insensitive to faults, hindering effective training of diagnostic models and severely limiting the application of these models in turboshaft and turboprop engines. Therefore, how to generate fault samples is a problem that needs to be solved. Summary of the Invention
[0004] In view of this, the present invention provides a method and apparatus for generating fault samples of the main bearing of a turboshaft and turboprop engine, so as to solve the problem of generating fault samples.
[0005] In a first aspect, the present invention provides a method for generating fault samples of the main bearing of a turboshaft / turboprop engine, the method comprising:
[0006] Multiple standard fault samples of the main bearing of the turboshaft and turboprop engine are obtained, and each standard fault sample carries a standard fault label that identifies the fault category.
[0007] For each standard fault sample, generate a standard grayscale image of the standard fault sample.
[0008] Based on the standard grayscale image and standard fault label of each standard fault sample, a fault sample generation model is trained to obtain the target generator. The model structure of the fault sample generation model is determined based on the generation characteristics of the grayscale image.
[0009] Obtain the target fault label;
[0010] A target generator is used to generate multiple target fault samples corresponding to the target fault labels.
[0011] This invention obtains standard fault samples of main bearings to provide real samples for model training. Each real sample carries a label, providing a supervised basis for subsequent model generation, guiding the generation of more accurate fault samples. These standard fault samples are converted into grayscale images, preserving the temporal dependence and structural information of the original data. Simultaneously, fault features are presented in a visual grayscale mode, significantly improving the model's learning efficiency for fault features. The model is trained based on the grayscale images and fault labels, enabling it to learn the grayscale distribution characteristics of different fault categories, thus obtaining a target generator capable of generating high-quality fault samples. In practical applications, the trained target generator generates fault samples of a specified fault category, not only satisfying the feature distribution of the specified fault category but also preserving the physical authenticity of the fault through a semi-physical generation method. This effectively expands the training dataset, thereby improving the accuracy and generalization ability of main bearing fault diagnosis.
[0012] In one optional implementation, for each standard fault sample, a grayscale image of the standard fault sample is generated, including:
[0013] For each standard fault sample, the standard fault sample is standardized to obtain a standardized fault sample.
[0014] The phase angle is calculated based on the sample lengths of standardized fault samples and standard fault samples.
[0015] Based on the phase angle and sample length of the standard fault samples, an improved Gramian matrix corresponding to the standard brick fault samples is constructed.
[0016] Each element in the improved Gramian matrix is mapped to a range of gray values and arranged in the order of rows and columns of the matrix to generate a grayscale image of a standard fault sample.
[0017] This invention ensures the consistency of grayscale image generation by standardizing standard fault samples, and enhances the time series characteristics and periodicity of signals by improving the traditional Gramian matrix. The improved Gramian matrix is then used to convert matrix values into visual grayscale values, providing high-quality visual feature input for model training.
[0018] In one alternative implementation, the improved Gramian matrix is constructed using the following formula:
[0019]
[0020] In the formula, G ij θ represents the element in the i-th row and j-th column of the improved Gramian matrix; i and θ jThis represents the phase angle of the i-th sample and the phase angle of the j-th sample within the sample length of the standardized fault samples. This invention uses an improved Gramian matrix to convert matrix values into visual grayscale values, allowing fault features to be presented in image form. Furthermore, the two-dimensional grayscale image can extract richer spatial features compared to a one-dimensional signal.
[0021] In one alternative implementation, the fault sample generation model includes a generator and a discriminator;
[0022] Based on the standard grayscale image and standard fault label of each standard fault sample, a fault sample generation model is trained to obtain a target generator, including:
[0023] The random noise vector and the standard fault label of each standard fault sample are input into the generator of the fault sample generation model to obtain the training grayscale image.
[0024] The training grayscale image is input into the discriminator of the fault sample generation model to obtain the discrimination result;
[0025] Based on the random noise vector, standard grayscale image, standard fault label, training grayscale image, and discrimination result corresponding to each standard fault sample, the discriminator objective function and the generator objective function are calculated.
[0026] Update the model parameters of the discriminator based on the discriminator's objective function;
[0027] The model parameters of the discriminator are fixed, and the model parameters of the generator are updated based on the objective function of the generator.
[0028] Based on the updated generator, return to the step of inputting the random noise vector and the standard fault label of each standard fault sample into the generator of the fault sample generation model to obtain the training grayscale image, until the training stopping condition is met, and use the generator obtained from the last update as the target generator.
[0029] This invention generates training grayscale images through the generator of a fault sample generation model, providing high-quality simulated samples for subsequent discriminator training. The training grayscale images are input into the discriminator to generate discrimination results. Based on the above data, the objective functions of the generator and the discriminator are calculated to update the model parameters, thereby obtaining a target generator capable of generating high-quality samples.
[0030] In one optional implementation, based on the random noise vector corresponding to each standard fault sample, the standard grayscale image, the standard fault label, the training grayscale image, and the discrimination result, the discriminator objective function and the generator objective function are calculated, including:
[0031] Calculate the gradient penalty term and the first loss based on the standard grayscale image and training grayscale image of each standard fault sample;
[0032] The second loss is calculated based on the standard fault labels of the standard grayscale image and the discrimination results of the training grayscale image.
[0033] The sum of the first loss, the second loss, and the gradient penalty term is used as the discriminator objective function.
[0034] Calculate the difference between the first loss and the second loss, and use the sum of the difference and the gradient penalty term as the generator objective function.
[0035] This invention measures the discriminator's ability to distinguish and classify real samples by calculating a first loss and a second loss, and calculates a gradient penalty term to balance model stability and training efficiency, thereby obtaining the objective functions of the discriminator and generator to improve model training performance and increase model training efficiency.
[0036] In one alternative implementation, a gradient penalty term is calculated based on a standard grayscale image and a training grayscale image, including:
[0037] Multiple linear interpolation samples are generated based on the standard grayscale image and the training grayscale image;
[0038] Calculate the gradient of the discriminator for each linear interpolated sample;
[0039] Calculate the gradient norm of the gradient and penalize the deviation of the gradient norm from the preset value;
[0040] Based on the squared terms of the deviations corresponding to all linear interpolation samples, the expected value of all linear interpolation samples is calculated to obtain the average penalty value.
[0041] The gradient penalty term is determined based on the penalty hyperparameter and the average penalty value.
[0042] This invention provides parameters for model training by calculating gradient penalty terms, which helps to balance model stability and training efficiency.
[0043] In one optional implementation, a target generator is used to generate multiple target fault samples corresponding to the target fault labels, including:
[0044] The target fault label and random noise vector are input into the target generator to generate multiple target fault samples corresponding to the target fault label.
[0045] This invention uses a trained target generator to generate corresponding fault samples according to the target fault labels. The semi-physical generation method preserves the physical authenticity of the faults and can be used to expand the training dataset of the diagnostic model, which helps to improve the accuracy and generalization ability of fault diagnosis.
[0046] Secondly, the present invention provides a fault sample generation device for the main bearing of a turboshaft / turboprop engine, the device comprising:
[0047] The first acquisition module is used to acquire multiple standard fault samples of the main bearing of the turboshaft and turboprop engine. Each standard fault sample carries a standard fault label that identifies the fault category.
[0048] The first generation module is used to generate a standard grayscale image of each standard fault sample.
[0049] The training module is used to train the fault sample generation model based on the standard grayscale image and standard fault label of each standard fault sample to obtain the target generator. The model structure of the fault sample generation model is determined based on the generation characteristics of the grayscale image.
[0050] The second acquisition module is used to acquire the target fault label;
[0051] The second generation module is used to generate multiple target fault samples corresponding to the target fault labels using a target generator.
[0052] Thirdly, the present invention provides a computer device, comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to perform the method for generating fault samples of the main bearing of a turboshaft propeller engine as described in the first aspect or any corresponding embodiment.
[0053] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to execute the method for generating fault samples of the main bearing of a turboshaft / turboprop engine according to the first aspect or any corresponding embodiment described above. Attached Figure Description
[0054] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.
[0055] Figure 1 This is a flowchart illustrating a method for generating fault samples of the main bearing of a turboshaft / turboprop engine according to an embodiment of the present invention.
[0056] Figure 2 This is a flowchart of the training process of the target generator according to an embodiment of the present invention;
[0057] Figure 3This is a structural block diagram of a fault sample generation device for the main bearing of a turboshaft / turboprop engine according to an embodiment of the present invention.
[0058] Figure 4 This is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. Detailed Implementation
[0059] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0060] The acquisition of main bearing fault samples in turboshaft and turboprop engines primarily relies on real-time data acquisition. However, most of the acquired samples are normal main bearings, with few faulty samples. This results in insensitivity to faults when using data-driven bearing fault diagnosis, hindering effective training of diagnostic models and significantly limiting their application in turboshaft and turboprop engines. This invention obtains standard main bearing fault samples to provide real-world data for model training. Each real sample carries a label, providing a supervised basis for subsequent model generation. These standard fault samples are converted into grayscale images, and the model is trained based on these grayscale images and fault labels. This allows the model to learn the grayscale distribution characteristics of different fault categories, resulting in a target generator capable of generating high-quality fault samples. In practical applications, the trained target generator generates fault samples for specified fault categories. These samples not only satisfy the characteristic distribution of the specified fault categories but also retain the physical authenticity of the faults through a semi-physical generation method, effectively expanding the training dataset and improving the accuracy and generalization ability of main bearing fault diagnosis.
[0061] According to an embodiment of the present invention, a method for generating fault samples of the main bearing of a turboshaft propeller engine is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.
[0062] This embodiment provides a method for generating fault samples of the main bearing of a turboshaft / turboprop engine. Figure 1 This is a flowchart of a method for generating fault samples of the main bearing of a turboshaft / turboprop engine according to an embodiment of the present invention, as shown below. Figure 1 As shown, the process includes the following steps:
[0063] Step S101: Obtain multiple standard fault samples of the main bearing of the turboshaft and turboprop engine. Each standard fault sample carries a standard fault label that identifies the fault category.
[0064] Specifically, the standard fault sample is the real-time vibration data Z collected when the main bearing fails. h (t). Here, h = 1, 2, 3, 4 represent standard fault labels, corresponding to different fault categories, such as outer race, inner race, rolling elements, and cage. By obtaining standard fault samples, real samples are provided for model training. Each real sample carries a label, fully considering the influence of different fault categories, providing a supervised basis for subsequent model generation, and guiding the generation of more accurate fault samples.
[0065] Step S102: For each standard fault sample, generate a standard grayscale image of the standard fault sample.
[0066] Specifically, standard fault samples are transformed from time-series data into grayscale images, preserving the time dependence and structural information of the original data. At the same time, fault features are presented in a visual grayscale mode, significantly improving the model's learning efficiency of fault features.
[0067] Step S103: Based on the standard grayscale image and standard fault label of each standard fault sample, train the fault sample generation model to obtain the target generator. The model structure of the fault sample generation model is determined based on the generation characteristics of the grayscale image.
[0068] Specifically, since the above steps have converted the standard fault samples into grayscale images, the training process in the fault sample generation model is implemented using grayscale images as units. The model should contain a structure capable of generating and processing grayscale images. The grayscale images and labels of real standard fault samples are input into the model, allowing it to learn the grayscale image distribution characteristics of different fault categories, thereby obtaining a target generator capable of generating high-quality fault samples.
[0069] Step S104: Obtain the target fault label.
[0070] Specifically, for various fault categories of main bearings in turboshaft and turboprop engines, in practical engineering applications, the target fault type can be specified according to diagnostic needs (such as focusing on rolling element faults). By inputting the target fault label, the generation process can be controlled by category orientation, avoiding the generation of samples of irrelevant fault categories, and effectively improving the efficiency and targeting of fault diagnosis.
[0071] Step S105: Use a target generator to generate multiple target fault samples corresponding to the target fault labels.
[0072] Specifically, the target generator trained can generate fault samples of a specified fault category according to the target fault label. It not only meets the feature distribution of the specified fault category, but also retains the physical authenticity of the fault through a semi-physical generation method. It can be directly used to expand the training dataset of the diagnostic model, significantly improving the accuracy and generalization ability of main bearing fault diagnosis.
[0073] This invention obtains standard fault samples of main bearings to provide real samples for model training. Each real sample carries a label, providing a supervised basis for subsequent model generation, guiding the generation of more accurate fault samples. These standard fault samples are converted into grayscale images, preserving the temporal dependence and structural information of the original data. Simultaneously, fault features are presented in a visual grayscale mode, significantly improving the model's learning efficiency for fault features. The model is trained based on the grayscale images and fault labels, enabling it to learn the grayscale distribution characteristics of different fault categories, thus obtaining a target generator capable of generating high-quality fault samples. In practical applications, the trained target generator generates fault samples of a specified fault category, not only satisfying the feature distribution of the specified fault category but also preserving the physical authenticity of the fault through a semi-physical generation method. This effectively expands the training dataset, thereby improving the accuracy and generalization ability of main bearing fault diagnosis.
[0074] This embodiment provides a method for generating fault samples of the main bearing of a turboshaft / turboprop engine, which specifically includes the following steps:
[0075] Step S201: Obtain multiple standard fault samples of the main bearing of the turboshaft / turboprop engine. Each standard fault sample carries a standard fault label identifying the fault category. For details, please refer to [link to relevant documentation]. Figure 1 Step S101 of the illustrated embodiment will not be described again here.
[0076] Step S202: For each standard fault sample, generate a standard grayscale image of the standard fault sample.
[0077] Specifically, step S202 includes:
[0078] Step S2021: For each standard fault sample, standardize the standard fault sample to obtain a standardized fault sample.
[0079] Specifically, the standard fault samples are standardized by the following formula (1) so that the standardized samples are distributed on a uniform scale, ensuring the consistency of grayscale image generation.
[0080]
[0081] In the formula, Z represents a standardized fault sample. h(t) represents the standard fault sample; E() represents the mean; It represents the standard deviation.
[0082] Step S2022: Calculate the phase angle based on the sample lengths of the standardized fault samples and the standard fault samples.
[0083] Specifically, the phase angle of the standard fault sample is calculated by the following formula (2), and the mapping relationship between the signal time point and the polar coordinates is established to ensure that the grayscale image can retain the temporal characteristics of the original sample.
[0084]
[0085] In the formula, θ k This represents the phase angle of the k-th standardized fault sample; The k-th standardized fault sample is represented by N; the sample length is represented by N; and the m-th sample point is represented by m.
[0086] Step S2023: Based on the phase angle and sample length of the standard fault sample, construct the improved Gramian matrix corresponding to the standard brick fault sample.
[0087] Specifically, the traditional Gramian matrix is constructed based on the inner product of the sample signal amplitudes. In this embodiment of the invention, the time series characteristics and periodicity of the signal are enhanced by the trigonometric function relationship of the phase angle. The one-dimensional information is converted into angles in polar coordinates using the properties of trigonometric functions, which preserves the time dependence and structural information of the original sample and helps to generate a visualized grayscale image. More specifically, each element in the improved Gramian matrix is obtained by the following equation (3).
[0088]
[0089] In the formula, G ij θ represents the element in the i-th row and j-th column of the improved Gramian matrix; i and θ j This represents the phase angle of the i-th sample and the phase angle of the j-th sample within the sample length of the standardized fault sample.
[0090] Step S2024: Map each element in the improved Gramian matrix to a grayscale value range, arrange them in the order of the matrix rows and columns, and generate a grayscale image of the standard fault sample.
[0091] Specifically, each element in the matrix is mapped using the following formula (4) to obtain the corresponding gray value. All gray values are arranged in the order of the matrix rows and columns to form a two-dimensional array, i.e., a grayscale image. By converting the matrix values into visual gray values, the fault features are presented in the form of an image. Moreover, the two-dimensional grayscale image can extract richer spatial features than the one-dimensional signal, providing high-quality visual feature input for model training.
[0092]
[0093] In the formula, G represents the gray value of the element in the i-th row and j-th column of the improved Gramian matrix; ij G represents the element in the i-th row and j-th column of the improved Gramian matrix; min G represents the minimum gray value in the improved Gramian matrix; max This represents the maximum gray value in the improved Gramian matrix.
[0094] Step S203: Based on the standard grayscale image and standard fault label of each standard fault sample, train the fault sample generation model to obtain the target generator. The model structure of the fault sample generation model is determined based on the generation characteristics of the grayscale image. The fault sample generation model includes a generator and a discriminator.
[0095] Specifically, step S203 includes:
[0096] Step S2031: Input the random noise vector and the standard fault label of each standard fault sample into the generator of the fault sample generation model to obtain the training grayscale image.
[0097] Specifically, the fault sample generation model is an Auxiliary Classifier Generative Adversarial Network (ACGAN), comprising a generator and a discriminator. The generator consists of a 1D-CNN (1D Convolutional Neural Network), an LSTM (Long Short-Term Memory Network), and a GASF (Gramian Angular Summation Field) to generate two-dimensional grayscale images. The CNN uses three convolutional layers. Pooling layers are removed because they reduce the dimensionality of the data to a lower level and lose a significant amount of feature information, leading to decreased model accuracy. Batch normalization layers are added after each convolutional layer to enhance the diversity of generated samples and improve the stability of the trained network. Unlike traditional GAN networks, standard fault labels are introduced into the generator to guide the direction of data generation, achieving supervised learning. The random noise vector and the standard fault label for each standard fault sample are first input into the generator's CNN to capture temporal local features. Then, long-term dependencies are modeled using LSTM. Finally, the temporal features are converted into a two-dimensional grayscale image, i.e., the training grayscale image, using GASF. The generator can output a two-dimensional grayscale image similar to the distribution of real fault data, providing high-quality simulated samples for subsequent discriminator training, effectively expanding the amount of fault data. Moreover, the generation process is guided by labels, making the data highly targeted.
[0098] Step S2032: Input the training grayscale image into the discriminator of the fault sample generation model to obtain the discrimination result.
[0099] Specifically, the discriminator consists of a CNN-MLP (Multilayer Perceptron) to distinguish the generated results. The CNN structure is identical to the CNN in the generator. The CNN output is processed by a three-layer fully connected MLP, using the LeakyReLU activation function (left-hand gradient 0.2). The output layer is mapped to [0,1] using the Sigmoid function to determine the authenticity of the data, obtaining a true / false classification result. Simultaneously, an auxiliary classifier outputs a classification result indicating the fault category. The true / false classification result and the fault category classification result are combined as the discrimination result for training the grayscale image.
[0100] Step S2033: Based on the random noise vector, standard grayscale image, standard fault label, training grayscale image and discrimination result corresponding to each standard fault sample, calculate the discriminator objective function and the generator objective function.
[0101] In some optional implementations, step S2033 above includes:
[0102] Step S20331: Calculate the gradient penalty term and the first loss based on the standard grayscale image and training grayscale image of each standard fault sample.
[0103] Specifically, the standard grayscale image is the real sample, and the training grayscale image is the sample generated by the generator. The loss between the two is calculated by the following formula (5) using loss functions such as cross-entropy, so as to determine the discriminator's ability to distinguish between real and fake samples.
[0104]
[0105] In the formula, L1 represents the first loss; x represents the standard grayscale image; P r G(z) represents the actual distribution to which the standard grayscale image belongs; G(z) represents the training grayscale image, z represents the random noise vector, and P represents the actual distribution to which the standard grayscale image belongs. g This represents a random noise vector from the actual distribution P. r The generated distribution learned in the training; D(G(z)) represents the true or false classification result of the discriminator in the discrimination result of the training grayscale image; E() represents the expectation; D() represents the discriminator output; G() represents the generator output.
[0106] In some optional implementations, step S20331 above calculates a gradient penalty term based on the standard grayscale image and training grayscale image of each standard fault sample, including:
[0107] Step a1: Generate multiple linear interpolation samples based on the standard grayscale image and the training grayscale image.
[0108] Specifically, assuming the standard grayscale image is x and the training grayscale image is y, multiple linear interpolation samples l are generated, l = ε·x + (1-ε)·y, where ε is a coefficient randomly sampled from the uniform distribution U(0,1), which corresponds one-to-one with the linear interpolation samples to ensure uniform distribution along the line connecting the two grayscale images.
[0109] Step a2: Calculate the gradient of the discriminator for each linear interpolation sample.
[0110] Specifically, the gradient of the discriminator reflects its sensitivity to the input data. If the gradient norm is too large, it may lead to gradient explosion during training; if it is too small, it may lead to gradient vanishing, both of which will affect model convergence. By calculating the gradient of the interpolated samples, the behavior of the discriminator in the transition region can be monitored, providing a basis for gradient penalty.
[0111] Step a3: Calculate the gradient norm of the gradient and penalize the deviation of the gradient norm from the preset value.
[0112] Specifically, ideally, the L2 gradient norm of the discriminator on the data manifold should be close to a preset value (usually 1) to ensure the smoothness of the decision boundary. If the gradient norm deviates from the preset value, it indicates that the discriminator's mapping in that region has nonlinear distortion, which may cause the generator to learn incorrect distribution characteristics. By penalizing the bias, the gradient norm is forced to be close to the preset value, thus improving the model's stability.
[0113] Step a4: Based on the squared terms of the deviations corresponding to all linear interpolation samples, calculate the expectation of all linear interpolation samples to obtain the average penalty value.
[0114] Specifically, the penalty value of a single linear interpolation sample is random. By using the expected value of all linear interpolations as the average penalty value, the impact of random noise can be reduced, ensuring that the penalty mechanism works on the entire data distribution, rather than individual samples.
[0115] Step a5: Determine the gradient penalty term based on the penalty hyperparameter and the average penalty value.
[0116] Specifically, the penalty hyperparameter is used to adjust the strength of the gradient penalty. If it is too small, the penalty effect is not obvious; if it is too large, it may over-constrain the discriminator and affect its discrimination ability. By multiplying the average penalty value with the penalty hyperparameter as shown in Equation (6), the penalty strength can be flexibly controlled to balance model stability and training efficiency.
[0117]
[0118] In the formula, L gp λ represents the gradient penalty term; λ represents the penalty hyperparameter; x represents the standard grayscale image; P r The standard grayscale image represents the actual distribution to which it belongs; y represents the training grayscale image; ||·|| p Represents the p-norm; This represents the average penalty value. Where ε represents the coefficients randomly sampled from the uniform distribution U(0,1); G(z) represents the training grayscale image, z represents the random noise vector; and D() represents the discriminator output.
[0119] Step S20332: Calculate the second loss based on the standard fault labels of the standard grayscale image and the discrimination results of the training grayscale image.
[0120] Specifically, the standard fault label is the true category, and the discrimination result includes the fault category classification result predicted by the discriminator on the training grayscale image. The second loss is calculated by the following formula (7) using loss functions such as cross-entropy to measure the classification accuracy of the discriminator.
[0121]
[0122] In the formula, L2 represents the first loss; c represents the standard fault label of the standard grayscale image; P r The standard grayscale image represents the actual distribution to which it belongs; D(G(c)) represents the fault category results of the training grayscale image; P g This indicates that the standard fault labels are distributed from the actual distribution P. r The generated distribution learned in the process; E() represents the expectation; D() represents the discriminator output; G() represents the generator output.
[0123] Step S20333 determines the sum of the first loss, the second loss, and the gradient penalty term as the discriminator objective function.
[0124] Specifically, the sum of the two losses and the gradient penalty term mentioned above is used as the objective function of the discriminator, so that the discriminator can accurately distinguish between real data and generated data and correctly classify them.
[0125] Step S20334: Calculate the difference between the first loss and the second loss, and use the sum of the difference and the gradient penalty term as the generator objective function.
[0126] Specifically, the difference between the first loss and the second loss, along with the sum of the gradient penalty term, is used as the objective function of the generator to ensure that the generator accurately generates high-quality data that conforms to the specified fault category.
[0127] Step S2034: Update the model parameters of the discriminator based on the discriminator objective function.
[0128] Specifically, the parameters are updated through methods such as backpropagation to maximize the objective function of the discriminator, thereby improving the discriminator's discrimination ability and classification accuracy, and providing more effective feedback to the generator.
[0129] Step S2035: Fix the model parameters of the discriminator and update the model parameters of the generator based on the objective function of the generator.
[0130] Specifically, with the discriminator parameters fixed, the generator parameters are updated through backpropagation and other methods to maximize the generator's objective function. This allows the generator to focus on learning how to generate data that the discriminator misclassifies as real but is correctly classified, thus avoiding interference from changes in the discriminator parameters on the generator's optimization direction.
[0131] In some alternative implementations, the model parameters of the discriminator can be updated a preset number of times before the model parameters of the generator are updated.
[0132] Step S2036: Based on the updated generator, return to the step of inputting the random noise vector and the standard fault label of each standard fault sample into the generator of the fault sample generation model to obtain the training grayscale image, until the training stopping condition is reached, and use the generator obtained from the last update as the target generator.
[0133] Specifically, using the updated generator, random noise vectors and standard fault labels are input into the generator to generate new training grayscale images. Steps S2031-S2035 are repeated. After each generator update, it is determined whether the training stopping condition has been met, such as the number of iterations reaching a preset value, the generator's objective function stabilizing, or the discriminator's objective function stabilizing. When the training stopping condition is met, training stops, and the generator obtained from the last update is used as the target generator for generating subsequent main bearing fault diagnosis data. Through iterative training, the generator is continuously optimized in the process of competing with the discriminator. The final generated fault data can fully learn the distribution characteristics of real data, effectively improving the recognition accuracy of subsequent fault diagnosis models.
[0134] Step S204: Obtain the target fault label. For details, please refer to [link to relevant documentation]. Figure 1 Step S104 of the illustrated embodiment will not be described again here.
[0135] Step S205: Use a target generator to generate multiple target fault samples corresponding to the target fault labels.
[0136] Specifically, step S205 includes:
[0137] Step S2051: Input the target fault label and random noise vector into the target generator to generate multiple target fault samples corresponding to the target fault label.
[0138] Specifically, the target fault label and random noise vector are input into the target generator. Since the target generator is composed of CNN-LSTM-GASF, and only fault samples are needed for fault diagnosis, the output obtained by CNN-LSTM is used as the target fault sample. If a grayscale image is needed for fault diagnosis, the output of LSTM can be further processed through GASF to obtain the final grayscale image. This embodiment of the invention does not impose any limitations on this.
[0139] In some alternative implementations, Figure 2 This is a flowchart of the training process of the target generator according to an embodiment of the present invention, such as... Figure 2As shown, a random noise vector and the standard fault label for each standard fault sample are input into the generator to obtain a training grayscale image. Then, the training grayscale image is input into the discriminator to obtain the discrimination result. Next, the first loss, the second loss, and the gradient penalty term are calculated based on the above data to obtain the discriminator objective function and the generator objective function. The model parameters of the discriminator and the generator are updated according to the two objective functions respectively. It is then determined whether the training stopping condition has been met. If not, the process returns to the step of inputting data into the generator until the training stopping condition is met, thus obtaining the trained target generator.
[0140] This invention obtains standard fault samples of main bearings to provide real samples for model training. Each real sample carries a label, providing a supervised basis for subsequent model generation, guiding the generation of more accurate fault samples. These standard fault samples are converted into grayscale images, preserving the temporal dependence and structural information of the original data. Simultaneously, fault features are presented in a visual grayscale mode, significantly improving the model's learning efficiency for fault features. The model is trained based on the grayscale images and fault labels, enabling it to learn the grayscale distribution characteristics of different fault categories, thus obtaining a target generator capable of generating high-quality fault samples. In practical applications, the trained target generator generates fault samples of a specified fault category, not only satisfying the feature distribution of the specified fault category but also preserving the physical authenticity of the fault through a semi-physical generation method. This effectively expands the training dataset, thereby improving the accuracy and generalization ability of main bearing fault diagnosis.
[0141] This embodiment also provides a fault sample generation device for the main bearing of a turboshaft / turboprop engine. This device is used to implement the above embodiments and preferred embodiments, and details already described will not be repeated. As used below, the term "module" can be a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.
[0142] This embodiment provides a fault sample generation device for the main bearing of a turboshaft / turboprop engine, such as... Figure 3 As shown, it includes:
[0143] The first acquisition module 301 is used to acquire multiple standard fault samples of the main bearing of the turboshaft and turboprop engine. Each standard fault sample carries a standard fault label that identifies the fault category.
[0144] The first generation module 302 is used to generate a standard grayscale image of the standard fault sample for each standard fault sample.
[0145] Training module 303 is used to train a fault sample generation model based on the standard grayscale image and standard fault label of each standard fault sample to obtain a target generator. The model structure of the fault sample generation model is determined based on the generation characteristics of the grayscale image.
[0146] The second acquisition module 304 is used to acquire the target fault label.
[0147] The second generation module 305 is used to generate multiple target fault samples corresponding to the target fault labels using a target generator.
[0148] In some alternative implementations, the first generation module 302 includes:
[0149] The standardization unit is used to standardize each standard fault sample to obtain a standardized fault sample.
[0150] The first calculation unit is used to calculate the phase angle based on the sample length of the standardized fault sample and the standard fault sample.
[0151] The building unit is used to construct the improved Gramian matrix corresponding to the standard brick fault sample based on the phase angle and sample length of the standard fault sample.
[0152] The first generation unit maps each element in the improved Gramian matrix to a grayscale range, arranges them in the order of rows and columns of the matrix, and generates a grayscale image of a standard fault sample.
[0153] In some alternative implementations, the improved Gramian matrix is constructed using the following formula:
[0154]
[0155] In the formula, G ij θ represents the element in the i-th row and j-th column of the improved Gramian matrix; i and θ j This represents the phase angle of the i-th sample and the phase angle of the j-th sample within the sample length of the standardized fault sample.
[0156] In some alternative implementations, the fault sample generation model includes a generator and a discriminator;
[0157] Training module 303 includes:
[0158] The second generation unit is used to input the random noise vector and the standard fault label of each standard fault sample into the generator of the fault sample generation model to obtain the training grayscale image.
[0159] The third generation unit is used to input the training grayscale image into the discriminator of the fault sample generation model to obtain the discrimination result.
[0160] The second computational unit is used to calculate the discriminator objective function and the generator objective function based on the random noise vector, standard grayscale image, standard fault label, training grayscale image and discrimination result corresponding to each standard fault sample.
[0161] The first update unit is used to update the model parameters of the discriminator based on the discriminator objective function.
[0162] The second update unit is used to fix the model parameters of the discriminator and update the model parameters of the generator based on the objective function of the generator.
[0163] The second update unit is used to return to the step of inputting the random noise vector and the standard fault label of each standard fault sample into the generator of the fault sample generation model to obtain the training grayscale image based on the updated generator, until the training stopping condition is reached, and the generator obtained from the last update is used as the target generator.
[0164] In some alternative implementations, the second computing unit includes:
[0165] The first computational subunit is used to calculate the gradient penalty term and the first loss based on the standard grayscale image and the training grayscale image of each standard fault sample.
[0166] The second computational subunit is used to calculate the second loss based on the standard fault labels of the standard grayscale image and the discrimination results of the training grayscale image.
[0167] The first determining subunit is used to determine the sum of the first loss, the second loss, and the gradient penalty term as the discriminator objective function.
[0168] The third computational subunit is used to calculate the difference between the first loss and the second loss, and the sum of the difference and the gradient penalty term is used to determine the generator objective function.
[0169] In some optional implementations, the first computational subunit is specifically used to: generate multiple linear interpolation samples based on a standard grayscale image and a training grayscale image; calculate the gradient of the discriminator for each linear interpolation sample; calculate the gradient norm of the gradient and penalize the deviation of the gradient norm from a preset value; calculate the expectation of all linear interpolation samples based on the squared terms of the deviations corresponding to all linear interpolation samples to obtain an average penalty value; and determine the gradient penalty term based on the penalty hyperparameter and the average penalty value.
[0170] In some alternative implementations, the second generation module 305 includes:
[0171] The fourth generation unit is used to input the target fault label and random noise vector into the target generator to generate multiple target fault samples corresponding to the target fault label.
[0172] Further functional descriptions of the above modules and units are the same as those in the corresponding embodiments described above, and will not be repeated here.
[0173] In this embodiment, the fault sample generation device for the main bearing of the turboshaft and turboprop engine is presented in the form of a functional unit. Here, a unit refers to an ASIC (Application Specific Integrated Circuit), a processor and memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.
[0174] This invention also provides a computer device having the above-described features. Figure 3 The device shown is for generating fault samples of the main bearing of a turboshaft / turboprop engine.
[0175] Please see Figure 4 , Figure 4 This is a schematic diagram of the structure of a computer device provided in an optional embodiment of the present invention, such as... Figure 4 As shown, the computer device includes one or more processors 10, memory 20, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components communicate with each other via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on external input / output devices (such as display devices coupled to the interfaces). In some alternative implementations, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system). Figure 4 Take a processor 10 as an example.
[0176] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GDA), or any combination thereof.
[0177] The memory 20 stores instructions executable by at least one processor 10 to cause at least one processor 10 to perform the method shown in the above embodiments.
[0178] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the computer device. Furthermore, the memory 20 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.
[0179] The memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid-state drive; the memory 20 may also include a combination of the above types of memory.
[0180] The computer device also includes an input device 30 and an output device 40. The processor 10, memory 20, input device 30, and output device 40 can be connected via a bus or other means. Figure 4 Taking the example of a connection between China and Israel via a bus.
[0181] Input device 30 can receive input numerical or character information, and generate key signal inputs related to user settings and function control of the computer device, such as a touchscreen, keypad, mouse, trackpad, touchpad, joystick, one or more mouse buttons, trackball, joystick, etc. Output device 40 may include display devices, auxiliary lighting devices (e.g., LEDs), and haptic feedback devices (e.g., vibration motors). The aforementioned display devices include, but are not limited to, liquid crystal displays, light-emitting diodes, displays, and plasma displays. In some alternative embodiments, the display device may be a touchscreen.
[0182] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implements the methods shown in the above embodiments.
[0183] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.
[0184] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.
Claims
1. A method for generating fault samples of the main bearing of a turboshaft / turboprop engine, characterized in that, The method includes: Multiple standard fault samples of the main bearing of the turboshaft and turboprop engine are obtained, and each standard fault sample carries a standard fault label that identifies the fault category. For each standard fault sample, generate a standard grayscale image of the standard fault sample; Based on the standard grayscale image and standard fault label of each standard fault sample, a fault sample generation model is trained to obtain a target generator. The model structure of the fault sample generation model is determined based on the generation characteristics of the grayscale image. Obtain the target fault label; The target generator is used to generate multiple target fault samples corresponding to the target fault label.
2. The method according to claim 1, characterized in that, For each standard fault sample, generating a grayscale image of the standard fault sample includes: For each standard fault sample, the standard fault sample is standardized to obtain a standardized fault sample; The phase angle is calculated based on the standardized fault sample and the sample length of the standard fault sample; Based on the phase angle and sample length of the standard fault sample, an improved Gramian matrix corresponding to the standard fault sample is constructed. Each element in the improved Gramian matrix is mapped to a range of gray values and arranged in the order of the matrix rows and columns to generate a grayscale image of the standard fault sample.
3. The method according to claim 2, characterized in that, The improved Gramian matrix is constructed using the following formula: In the formula, G ij This represents the element in the i-th row and j-th column of the improved Gramian matrix; θ i and θ j This represents the phase angle of the i-th sample and the phase angle of the j-th sample within the sample length of the standardized fault sample.
4. The method according to claim 1, characterized in that, The fault sample generation model includes a generator and a discriminator; The fault sample generation model is trained based on the standard grayscale image and standard fault label of each standard fault sample to obtain the target generator, including: The random noise vector and the standard fault label of each standard fault sample are input into the generator of the fault sample generation model to obtain the training grayscale image. The training grayscale image is input into the discriminator of the fault sample generation model to obtain the discrimination result; Based on the random noise vector, standard grayscale image, standard fault label, training grayscale image, and discrimination result corresponding to each standard fault sample, the discriminator objective function and the generator objective function are calculated. The model parameters of the discriminator are updated based on the discriminator objective function; The model parameters of the discriminator are fixed, and the model parameters of the generator are updated based on the objective function of the generator. Based on the updated generator, return to the step of inputting the random noise vector and the standard fault label of each standard fault sample into the generator of the fault sample generation model to obtain the training grayscale image, until the training stopping condition is reached, and use the generator obtained from the last update as the target generator.
5. The method according to claim 4, characterized in that, The objective functions of the discriminator and generator are calculated based on the random noise vector, standard grayscale image, standard fault label, training grayscale image, and discrimination result corresponding to each standard fault sample, including: Calculate the gradient penalty term and the first loss based on the standard grayscale image and training grayscale image of each standard fault sample; The second loss is calculated based on the standard fault labels of the standard grayscale image and the discrimination results of the training grayscale image; The sum of the first loss, the second loss, and the gradient penalty term is determined as the discriminator objective function; The difference between the first loss and the second loss is calculated, and the sum of the difference and the gradient penalty term is determined as the generator objective function.
6. The method according to claim 5, characterized in that, The step of calculating the gradient penalty term based on the standard grayscale image and the training grayscale image includes: Based on the standard grayscale image and the training grayscale image, multiple linear interpolation samples are generated; Calculate the gradient of the discriminator for each linear interpolated sample; Calculate the gradient norm of the gradient and penalize the deviation of the gradient norm from a preset value; Based on the squared terms of the deviations corresponding to all linear interpolation samples, the expected value of all linear interpolation samples is calculated to obtain the average penalty value. The gradient penalty term is determined based on the penalty hyperparameter and the average penalty value.
7. The method according to claim 1, characterized in that, The step of using the target generator to generate multiple target fault samples corresponding to the target fault label includes: The target fault label and random noise vector are input into the target generator to generate multiple target fault samples corresponding to the target fault label.
8. A fault sample generation device for the main bearing of a turboshaft / turboprop engine, characterized in that, The device includes: The first acquisition module is used to acquire multiple standard fault samples of the main bearing of the turboshaft and turboprop engine. Each standard fault sample carries a standard fault label that identifies the fault category. The first generation module is used to generate a standard grayscale image of each standard fault sample. The training module is used to train a fault sample generation model based on the standard grayscale image and standard fault label of each standard fault sample to obtain a target generator. The model structure of the fault sample generation model is determined based on the generation characteristics of the grayscale image. The second acquisition module is used to acquire the target fault label; The second generation module is used to generate multiple target fault samples corresponding to the target fault label using the target generator.
9. A computer device, characterized in that, include: A memory and a processor are interconnected, the memory stores computer instructions, and the processor executes the computer instructions to perform the method for generating fault samples of the main bearing of a turboshaft propeller engine as described in any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to execute the method for generating fault samples of the main bearing of a turboshaft propeller engine as described in any one of claims 1 to 7.