A Fault Diagnosis Method Based on V-DIT Data Augmentation and CNN

Abstract

This invention relates to the field of aero-engine fault diagnosis, specifically to a fault diagnosis method based on V-DIT data augmentation and CNN. This method is based on the v-prediction time-series diffusion modeling approach, introducing an intermediate variable v to replace the noise variable to be predicted, and introducing a U-Net with a DIT structure to replace the traditional diffusion model, thus realizing a diffusion model for one-dimensional time series aero-engines based on a DIT network structure. Secondly, a CNN network is introduced for classification training of aero-engine fault data. The CNN network performs feature extraction, feature overlay, and feature classification on the input training set data, enabling it to capture local correlation features in the data, thereby achieving the identification and classification of different types of faults.

Description

Technical Field

[0001] This invention relates to the field of aero-engine fault diagnosis, and specifically to a fault diagnosis method based on V-DIT data augmentation and CNN. Background Technology

[0002] Core components of aero-engines (such as bearings, fuel regulators, and compressors) play a crucial role in ensuring overall engine performance and flight safety. Therefore, researching high-precision, real-time fault diagnosis technologies for these core components is not only essential for improving the reliability and safety of aero-engines but also a vital direction for the intelligent development of modern aviation support systems. However, in practical engineering applications, limitations imposed by objective factors often mean that only a very small amount of fault data for a specific component can be obtained for model training. This leads to problems such as poor robustness, weak generalization ability, and low diagnostic accuracy in traditional data-driven models for fault diagnosis tasks. Therefore, it is crucial to research a method that can achieve effective data augmentation and fault sample generation under small or limited sample conditions.

[0003] In recent years, diffusion models have been widely applied to data generation tasks and have achieved significant results. The Diffusion Transformer (DIT) model, as the latest form of diffusion model, introduces the Transformer architecture, replacing the traditional U-Net for modeling the diffusion process. It possesses stronger global modeling capabilities and is suitable for various data types such as images and time series. It has been widely applied in intelligent condition monitoring and predictive maintenance of critical aero-engine equipment, such as aero-engines and automotive engines, significantly improving equipment reliability and safety. However, DIT still faces many challenges in processing one-dimensional time series tasks such as aero-engine sensor data, including limited modeling capabilities, poor prediction stability, and low sampling efficiency.

[0004] To address the aforementioned issues, a fault diagnosis method for aero-engines based on V-DIT data augmentation and CNN is proposed. Summary of the Invention

[0005] To address the shortcomings and deficiencies in the aforementioned background technologies, this invention proposes a fault diagnosis method based on V-DIT data augmentation and CNN. First, this method proposes a time-series diffusion modeling approach based on v-prediction, introducing an intermediate variable v to replace the noise variable to be predicted, and introducing a DIT structure to replace the traditional diffusion model's U-Net, thus realizing a diffusion model for one-dimensional time series data of aero-engines based on the DIT network structure. Second, a CNN network is introduced for classification training of aero-engine fault data. The CNN network performs feature extraction, feature overlay, and feature classification on the input training set data, enabling it to capture local correlation features in the data, thereby achieving the identification and classification of different types of faults.

[0006] The first objective of this invention is to provide a fault diagnosis method based on V-DIT data augmentation and CNN, comprising:

[0007] Acquire aircraft engine fault data;

[0008] The processed fault data is obtained by embedding location codes into the aircraft engine fault data.

[0009] The processed fault data is processed using a diffusion DIT model and then subjected to a T-layer forward noise addition process to obtain the final noise-added feature data.

[0010] Intermediate variables are constructed based on the processed fault data and the introduced noise;

[0011] Construct a Transformer architecture network based on DIT, where the input of the network is the feature data obtained after adding noise to each layer, and the output is the predicted intermediate variables;

[0012] A loss function is established based on the constructed intermediate variables and the predicted intermediate variables;

[0013] Based on the loss function, the constructed DIT-based Transformer architecture network is trained until convergence, thus obtaining the trained DIT-based Transformer architecture network.

[0014] A V-DIT back-inference model is constructed based on the DIT model by introducing V-prediction.

[0015] Based on the final noisy feature data and the predicted intermediate variables output by the trained DIT-based Transformer architecture network, the potential fault data of layer T is obtained based on the V-DIT back-inference model.

[0016] Based on the potential fault data of layer T, the Transformer architecture network based on DIT is trained to obtain the prediction intermediate variables corresponding to the potential fault data. Then, the prediction intermediate variables are back-inferred through the V-DIT model to obtain the potential fault data of layer T-1. The prediction fault data is obtained by back-inferring in sequence.

[0017] A fault diagnosis model is constructed based on a convolutional neural network, and trained using predicted fault data and original fault data to obtain a fault diagnosis model with fault identification capabilities.

[0018] The data to be diagnosed is used to perform a diagnosis based on the diagnostic model, and the corresponding diagnostic results are obtained.

[0019] In one embodiment, the processed fault data undergoes a T-layer forward noise addition process using a diffusion DIT model, and the calculation formula is as follows:

[0020]

[0021] in, express This represents the i-th data point in the time series measured by the j-th sensor. ; It is Gaussian distributed noise. ;

[0022] The noise addition process follows the formula ;

[0023] t represents the time step; Under the conditions The conditional probability distribution;

[0024] N is normally distributed; The scaling factor for the mean; The noise variance of the time step t increases as t increases (0 < t). <……< ); ;

[0025] The initial fault data is input into the DIT Block, and noise is added using a Transformer encoder. After T forward noise addition processes, the final noise-added feature data is obtained.

[0026] In one embodiment, constructing intermediate variables based on the processed fault data and introduced noise includes:

[0027]

[0028] In the formula, Represents the actual intermediate variable; Noise representing a Gaussian distribution; Represents the original data points;

[0029] express The average value, , It is the noise variance of the time step t.

[0030] In one embodiment, a DIT-based Transformer architecture network includes:

[0031]

[0032] In the formula, Indicates the intermediate variable in the prediction; This represents the data points after noise has been added; Indicates positional encoding;

[0033] Indicates the network bias; This represents the external adjustable weight parameters of the Transformer; This represents the adjustable weight parameters within the Transformer.

[0034] In one embodiment, V-prediction is introduced to construct a V-DIT back-inference model, including:

[0035]

[0036]

[0037]

[0038] In the formula, This represents the data points predicted by the model before noise addition; This represents the noise predicted by the model; This represents the data points after noise has been added; Indicates the intermediate variable in the prediction; express The average value, , It is the noise variance of the time step t.

[0039] In one embodiment, after obtaining the predicted fault data, the method further includes: using MMD loss to determine the data generated by the inverse model. Whether it is credible, the formula is:

[0040]

[0041] In the formula, For the original fault dataset, The dataset of faults predicted by the model;

[0042] T represents the number of samples drawn from the data; k represents the selected Gaussian kernel function, whose standard form is: , This is the bandwidth function of the kernel function, controlling the kernel function's sensitivity to distance; These are data points representing faulty samples within the original dataset; The fault sample data points generated for model prediction;

[0043] when If the data is below a certain threshold, the predicted fault data will be... The distribution is reliable, meaning it serves as the final generation of fault data.

[0044] A second objective of this invention is to provide a computer program product, comprising a computer program that, when executed by a processor, implements a fault diagnosis method based on V-DIT data augmentation and CNN.

[0045] A third objective of this invention is to provide an electronic device comprising:

[0046] Processor; and

[0047] Memory for storing the executable instructions of the processor;

[0048] The processor is configured to execute a fault diagnosis method based on V-DIT data augmentation and CNN by executing the executable instructions.

[0049] The fourth objective of this invention is to provide a fault diagnosis system based on V-DIT data augmentation and CNN, comprising:

[0050] The data processing module is used to acquire aircraft engine fault data; to obtain processed fault data based on the embedded location encoding of the aircraft engine fault data; and to obtain the final noisy feature data by using a diffusion DIT model and a T-layer forward denoising process on the processed fault data.

[0051] The network construction module is used to construct intermediate variables based on the processed fault data and the introduced noise; construct a DIT-based Transformer architecture network, where the input of the network is the feature data obtained after adding noise to each layer, and the output is the predicted intermediate variables; establish a loss function based on the constructed intermediate variables and the predicted intermediate variables; and train the constructed DIT-based Transformer architecture network based on the loss function until convergence to obtain the trained DIT-based Transformer architecture network.

[0052] The fault data prediction module is used to construct a V-DIT back-inference model based on the DIT model by introducing V-prediction. Based on the final noisy feature data and the prediction intermediate variables output by the trained DIT-based Transformer architecture network, the potential fault data of layer T is obtained based on the V-DIT back-inference model. Based on the potential fault data of layer T, the corresponding prediction intermediate variables are obtained through the trained DIT-based Transformer architecture network. Then, the prediction intermediate variables are passed through the V-DIT back-inference model to obtain the potential fault data of layer T-1. This process is repeated to obtain the predicted fault data.

[0053] The fault diagnosis module is used to build a fault diagnosis model based on a convolutional neural network, and to train it using predicted fault data and original fault data to obtain a fault diagnosis model with fault recognition capabilities; the data to be diagnosed is then diagnosed according to the diagnosis model to obtain the corresponding diagnosis results.

[0054] The present invention has at least the following beneficial effects:

[0055] This invention provides a fault diagnosis method based on V-DIT data augmentation and CNN. First, it proposes a time-series diffusion modeling method based on v-prediction, introducing an intermediate variable v to replace the noise variable to be predicted, and introducing a DIT structure to replace the traditional diffusion model with U-Net, realizing a diffusion model for one-dimensional time series data of aero-engines based on the DIT network structure. Second, it introduces a CNN network for classification training of aero-engine fault data. The CNN network performs feature extraction, feature overlay, and feature classification on the input training set data, capturing local correlation features in the data, thereby achieving the identification and classification of different types of faults.

[0056] This invention applies v-prediction to a data generation and diffusion model for one-dimensional time series of aero-engines, including key steps such as forward noise addition, prediction of intermediate variable v, training through MSE loss, and reverse generation process.

[0057] This invention introduces a DIT structure to replace the traditional diffusion model U-Net, realizing a noise addition and denoising model for one-dimensional time series of aero-engines based on the DIT network structure, including VAE, time step embedding, latent space mapping, Transformer encoder stacking structure, etc., while supporting DDIM skip step back propagation to improve generation efficiency.

[0058] This invention introduces a CNN network for classification training of aero-engine fault data. The CNN network extracts, stacks, and classifies features from the input training data, capturing locally correlated features. The extracted high-dimensional features are then classified using fully connected layers and a softmax classifier, thereby enabling the identification and classification of different types of faults.

[0059] This invention can be widely applied to various small-sample or few-sample aero-engine fault vibration datasets, especially suitable for datasets with complex operating conditions and high diagnostic difficulty, such as the aero-engine bearing dataset (HIT) released by Harbin Institute of Technology. By introducing v-prediction to replace traditional noise prediction and combining DIT network and CNN for fault diagnosis, this invention can effectively improve diagnostic performance on such complex datasets and enhance the model's generalization ability and robustness. Attached Figure Description

[0060] Figure 1 A flowchart illustrating the fault diagnosis method based on V-DIT data augmentation and CNN provided by this invention;

[0061] Figure 2 This is a block diagram illustrating the principle of the V-DIT diffusion model. Detailed Implementation

[0062] In order to illustrate the technical means and effects adopted by the present invention to achieve the intended purpose, the following detailed description is provided in conjunction with the embodiments.

[0063] The purpose of this invention is to address the numerous problems that DIT (Data Augmentation Technique) still faces in processing one-dimensional time series data, such as aero-engine sensor data, including limited modeling capabilities, poor prediction stability, and low sampling efficiency. Therefore, it provides a fault diagnosis method based on V-DIT data augmentation and CNN (Network CNN).

[0064] To achieve the above objectives, a fault diagnosis method based on V-DIT data augmentation and CNN includes:

[0065] S1 acquires aircraft engine fault data;

[0066] The processed fault data is obtained by embedding location codes into the aircraft engine fault data.

[0067] The processed fault data is processed using a diffusion DIT model and then subjected to a T-layer forward noise addition process to obtain the final noise-added feature data.

[0068] The processed fault data is subjected to a T-layer forward noise addition process using the diffusion DIT model, and the calculation formula is as follows:

[0069]

[0070] in, express This represents the i-th data point in the time series measured by the j-th sensor. ; It is Gaussian distributed noise. ;

[0071] The noise addition process follows the formula ;

[0072] t represents the time step; Under the conditions The conditional probability distribution;

[0073] N is normally distributed; The scaling factor for the mean; The noise variance of the time step t increases as t increases (0 < t). <……< ); Represents a unit sequence;

[0074] The initial fault data is input into the DIT Block, and noise is added using a Transformer encoder. After T forward noise addition processes, the final noise-added feature data is obtained.

[0075] S2 constructs intermediate variables based on the processed fault data and the introduced noise;

[0076] Construct a Transformer architecture network based on DIT, where the input of the network is the feature data obtained after adding noise to each layer, and the output is the predicted intermediate variables;

[0077] A loss function is established based on the constructed intermediate variables and the predicted intermediate variables;

[0078] Based on the loss function, the constructed DIT-based Transformer architecture network is trained until convergence, thus obtaining the trained DIT-based Transformer architecture network.

[0079] Intermediate variables are constructed based on the processed fault data and the introduced noise, including:

[0080]

[0081] In the formula, Represents the actual intermediate variable; Noise representing a Gaussian distribution; Represents the original data points;

[0082] express The average value. It is a scaling factor related to the number of noise additions. , It is the noise variance of the time step t.

[0083] DIT-based Transformer architecture networks include:

[0084]

[0085] In the formula, Indicates the intermediate variable in the prediction; This represents the data points after noise has been added; Indicates positional encoding; Indicates the network bias; This represents the external adjustable weight parameters of the Transformer; This represents the adjustable weight parameters within the Transformer.

[0086] S3 constructs a V-DIT back-inference model based on the DIT model by introducing V-prediction;

[0087] Based on the final noisy feature data and the predicted intermediate variables output by the trained DIT-based Transformer architecture network, the potential fault data of layer T is obtained based on the V-DIT back-inference model.

[0088] Based on the potential fault data of layer T, the Transformer architecture network based on DIT is trained to obtain the prediction intermediate variables corresponding to the potential fault data. Then, the prediction intermediate variables are back-inferred through the V-DIT model to obtain the potential fault data of layer T-1. The prediction fault data is obtained by back-inferring in sequence.

[0089] Introducing V-prediction to construct a V-DIT back-inference model, including:

[0090]

[0091]

[0092]

[0093] In the formula, This represents the data points predicted by the model before noise addition; This represents the noise predicted by the model; This represents the data points after noise has been added; Indicates the intermediate variable in the prediction; express The average value, , It is the noise variance of the time step t; This represents the data points after t-1 rounds of noise addition.

[0094] After obtaining the predicted fault data, the process also includes: using MMD loss to determine the data generated by the inverse model. Whether it is credible, the formula is:

[0095]

[0096] In the formula, For the original fault dataset, The dataset of faults predicted by the model;

[0097] T represents the number of samples drawn from the data; k represents the selected Gaussian kernel function, whose standard form is: , This is the bandwidth function of the kernel function, controlling the kernel function's sensitivity to distance; These are data points representing faulty samples within the original dataset; The fault sample data points generated for model prediction;

[0098] when If the data is below a certain threshold, the predicted fault data is considered to be... The distribution is reliable, meaning it serves as the final generation of fault data.

[0099] S4 constructs a fault diagnosis model based on a convolutional neural network and trains it using predicted fault data and original fault data to obtain a fault diagnosis model with fault identification capabilities.

[0100] The data to be diagnosed is used to perform a diagnosis based on the diagnostic model, and the corresponding diagnostic results are obtained.

[0101] To further illustrate the fault diagnosis method based on V-DIT data augmentation and CNN provided by this invention, the following description is provided in conjunction with the accompanying drawings.

[0102] See Figure 1 As shown, a fault diagnosis method based on V-DIT data augmentation and CNN includes:

[0103] Step 1: Data Preprocessing

[0104] 1.1 Assuming the aircraft engine operation includes a time period T, input the complete operation process. Sequence data from each sensor , No. The time series of measurements from each sensor is represented as follows: Sensor data that is only slightly relevant to the fault occurrence or whose parameters remain essentially unchanged throughout the fault cycle are discarded, leaving only the following data: Time series data from individual sensors , Less than .

[0105] 1.2, Regarding the first The first sensor indivual Dimensional Data ,in The numbers 1-9 are treated as independent categorical features, with each digit representing a distinct category.

[0106] Will Convert to ;

[0107] According to the formula , generate Dimensional position encoding ,in correspond In ; Index representing the dimension; Represents the total number of dimensions.

[0108] Embed position code At the end, we get Perform the above operations on all sensor data to obtain the processed data. .

[0109] Step 2, see Figure 2 As shown, the predicted fault data obtained from the original fault data includes:

[0110] 2.1 Positive Noise Addition:

[0111] Introducing the DIT model for preprocessed data Using the forward process of the diffusion DIT model to... Noise addition: The principle of noise addition is briefly described below:

[0112] set up Let i be the i-th data point of the time series measured by the j-th sensor, or simply as ;

[0113] The noise addition process follows the formula:

[0114]

[0115] Where t represents the time step. Under the conditions The conditional probability distribution of ; N is a normal distribution; The scaling factor for the mean; The noise variance of the time step t increases as t increases (0 < t). <……< ); Represents a unit sequence;

[0116] achievable ,in, , It is Gaussian distributed noise. .

[0117] Data The input DIT block is used, and the above noise addition process is implemented using a Transformer encoder. After T forward noise addition processes, the final feature data is obtained. .

[0118] 2.2 Training V-DIT

[0119] 2.2.1 Improve the DIT model, assuming the original data points are... After the noise-adding process in step 2.1, the final data points are obtained. .

[0120] Introducing intermediate variables ;

[0121] Network inference based on DIT Transformer architecture The formula is as follows:

[0122]

[0123] In the formula, Intermediate variables in the prediction; This represents the external adjustable weight parameters of the Transformer; This represents the adjustable weight parameters within the Transformer; Indicates positional encoding; Indicates the network bias; This represents the data points after t rounds of noise addition.

[0124] During the training process of the network model based on the DIT-based Transformer architecture, the model will randomly select a data point after adding noise. (l=1,2,……,t), calculate the corresponding true intermediate variables. And obtain the predicted value. Once a data point has been selected, it will not be selected again. This step will be repeated until all T noisy data points have been selected.

[0125] 2.2.2 Defining the loss function Using MSE loss, the formula is:

[0126]

[0127] in, These are actual intermediate variables; is the intermediate variable for prediction; T is the total number of predictions.

[0128] when When it is below a certain threshold, <0.05, indicating that the prediction at this time is... Credible.

[0129] 2.3 Back-engineering prediction data

[0130] Establish a V-DIT back-inference model for the T layer. For the input value of layer T, when it is considered When reliable, according to the reverse formula:

[0131] ,

[0132] Can be reversed to predict and , which is the output value of the Tth layer;

[0133] The output value of layer T is used as the input value of layer (T-1). At this point, it is considered... and Credible

[0134] but Then imitate The inference process leads to credible conclusions. ,Depend on Predict new and During the back-reasoning process, a fixed number of K layers in the model can be randomly frozen according to the DDIM sampling mechanism. Typically, 50%-75% of the total number of layers T in the model is frozen to speed up the back-reasoning efficiency.

[0135] Thus, the output value of the last layer can be obtained, which is the generated prediction data. ;

[0136] Set up the reconstruction layer for the VAE decoder section, and... Restore to predicted fault data .

[0137] 2.4 Comparing data reliability

[0138] Using MMD loss, we can determine the data generated by the inverse model. Whether it is credible, the formula is:

[0139]

[0140] in, This is the original fault dataset;

[0141] The dataset of faults predicted by the model;

[0142] k represents the chosen Gaussian kernel function, whose standard form is: , This is the bandwidth function of the kernel function, controlling the kernel function's sensitivity to distance; These are data points representing faulty samples within the original dataset; The fault sample data points generated for model prediction.

[0143] when Below a certain threshold ( <0.05), considered as predicted fault data If the distribution is relatively reliable, it can be used as the final generation of fault data.

[0144] Step 3: Perform fault diagnosis

[0145] 3.1 The fault data generated by the V-DIT model is mixed with the original fault data, and divided into training and test sets, with the training set accounting for 80% and the test set accounting for 20%. The training set data is as follows: The test set data is .in Indicates a sample, This indicates the corresponding label.

[0146] 3.2 Input the training set samples into the CNN network, i.e., the diagnostic model, and first perform feature extraction, using the following formula:

[0147]

[0148] in, This represents the convolution operation in a CNN; This represents the parameters of the f-th convolutional layer; This represents the extracted spatial feature information.

[0149] The outputs of the feature extraction are stacked dimensionally to generate high-dimensional stacked features. These stacked features are then input into a feature classification layer, whose formula is:

[0150]

[0151] in, express Softmax Classifier, Represents the class l Softmax parameter, This represents the predicted probability distribution for different categories.

[0152] 3.3 The output of the CNN network is the class label, and the loss function is set to cross-entropy loss:

[0153]

[0154] in, This represents the total number of predictions. Labels for actual fault categories. Let be the predicted probability distribution. When If the value is below a certain threshold, the network diagnosis is considered accurate, and the fault diagnosis result is output.

[0155] Step four: Perform diagnosis on the data to be diagnosed according to the diagnostic model and obtain the corresponding diagnostic results.

[0156] This invention provides a computer program product, including a computer program that, when executed by a processor, implements a fault diagnosis method based on V-DIT data augmentation and CNN.

[0157] This invention provides an electronic device, comprising:

[0158] Processor; and

[0159] Memory for storing the executable instructions of the processor;

[0160] The processor is configured to execute a fault diagnosis method based on V-DIT data augmentation and CNN by executing the executable instructions.

[0161] This invention provides a fault diagnosis system based on V-DIT data augmentation and CNN, comprising:

[0162] The data processing module is used to acquire aircraft engine fault data; to obtain processed fault data based on the embedded location encoding of the aircraft engine fault data; and to obtain the final noisy feature data by using a diffusion DIT model and a T-layer forward denoising process on the processed fault data.

[0163] The network construction module is used to construct intermediate variables based on the processed fault data and the introduced noise; construct a DIT-based Transformer architecture network, where the input of the network is the feature data obtained after adding noise to each layer, and the output is the predicted intermediate variables; establish a loss function based on the constructed intermediate variables and the predicted intermediate variables; and train the constructed DIT-based Transformer architecture network based on the loss function until convergence to obtain the trained DIT-based Transformer architecture network.

[0164] The fault data prediction module is used to construct a V-DIT back-inference model based on the DIT model by introducing V-prediction. Based on the final noisy feature data and the prediction intermediate variables output by the trained DIT-based Transformer architecture network, the potential fault data of layer T is obtained based on the V-DIT back-inference model. Based on the potential fault data of layer T, the corresponding prediction intermediate variables are obtained through the trained DIT-based Transformer architecture network. Then, the prediction intermediate variables are passed through the V-DIT back-inference model to obtain the potential fault data of layer T-1. This process is repeated to obtain the predicted fault data.

[0165] The fault diagnosis module is used to build a fault diagnosis model based on a convolutional neural network, and to train it using predicted fault data and original fault data to obtain a fault diagnosis model with fault recognition capabilities; the data to be diagnosed is then diagnosed according to the diagnosis model to obtain the corresponding diagnosis results.

[0166] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for fault diagnosis based on V-DIT data augmentation and CNN, characterized in that, The method comprises the following steps: obtaining aero-engine fault data; the fault data comes from aero-engine bearing HIT data set; obtaining processed fault data according to aero-engine fault data embedding position code; obtaining final noise-added feature data by using diffusion DIT model on processed fault data through T-layer forward noise adding process; constructing intermediate variables according to processed fault data and introduced noise; constructing a DIT-based Transformer architecture network, wherein the input of the network is the feature data obtained after each layer of noise addition, and the output is the predicted intermediate variables; establishing a loss function according to the constructed intermediate variables and the predicted intermediate variables; training the constructed DIT-based Transformer architecture network based on the loss function until convergence, and obtaining the trained DIT-based Transformer architecture network; constructing a V-DIT backstepping model based on the DIT model by introducing V-prediction; obtaining T-layer potential fault data based on the V-DIT backstepping model according to the final noise-added feature data and the predicted intermediate variables output by the trained DIT-based Transformer architecture network; obtaining the predicted intermediate variables corresponding to the potential fault data through the trained DIT-based Transformer architecture network, and then obtaining T-1-layer potential fault data by inputting the predicted intermediate variables into the V-DIT backstepping model, and sequentially backstepping to obtain predicted fault data; constructing a fault diagnosis model based on a convolutional neural network, and training the fault diagnosis model by using the predicted fault data and the original fault data to obtain a fault diagnosis model with fault recognition capability; diagnosing the to-be-diagnosed data according to the diagnosis model to obtain corresponding diagnosis results; constructing a V-DIT backstepping model by introducing V-prediction, which comprises: wherein denotes the data point predicted by the model before adding noise; denotes the noise predicted by the model; denotes the data point after adding noise; denotes the predicted intermediate variable; denotes the average value of , is the noise variance for time step t. 2.The V-DIT data augmentation and CNN based fault diagnosis method of claim 1, wherein, using diffusion DIT model on processed fault data through T-layer forward noise adding process, the calculation formula is as follows: wherein, denotes the data point before adding noise; is the i-th data point of the time series measured by the j-th sensor; . is a Gaussian distributed noise, ; The noise addition process follows the formula ; t denotes a time step; conditional on conditional probability distribution of N is a normal distribution; is a scaling factor for the mean; is the noise variance for time step t, which increases with t; ; input the initial fault data into the DIT Block, use the Transformer encoder to add noise, and obtain the final noise-added feature data after T forward noise adding processes. 3.The V-DIT data augmentation and CNN based fault diagnosis method of claim 1, wherein, constructing intermediate variables according to processed fault data and introduced noise, which comprises: wherein represents a real intermediate variable; represents a Gaussian distributed noise; represents an original data point; denotes the average value of , is the noise variance for time step t. 4.The V-DIT data augmentation and CNN based fault diagnosis method of claim 1, wherein, the DIT-based Transformer architecture network comprises: wherein, denotes a predicted intermediate variable; denotes a data point after adding noise; denotes a position encoding; denotes a bias of the network; denotes a Transformer external tunable weight parameter; denotes a Transformer internal tunable weight parameter.

5. The V-DIT data augmentation and CNN based fault diagnosis method according to claim 1, characterized in that, After obtaining the predicted failure data, further comprising: judging whether the backstepping model generates data by using MMD loss Whether it is reliable, the formula is: wherein is the original fault dataset, is the model predicted fault dataset; T is the number of samples extracted from the data; k represents the selected Gaussian kernel function, whose standard form is , is the bandwidth function of the kernel function, which controls the sensitivity of the kernel function to distance; is the data point of the fault sample in the original data set; is the fault sample data point generated by the model prediction; When Below a certain threshold, the predicted failure data The distribution is trusted as the final generated failure data.

6. A computer program product comprising a computer program, characterized in that, the computer program is executed by the processor to realize the V-DIT data enhancement and CNN-based fault diagnosis method in any one of claims 1 to 5.

7. An electronic device, comprising: The method comprises the following steps: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to execute the V-DIT data enhancement and CNN-based fault diagnosis method in any one of claims 1 to 5 by executing the executable instructions.

8. A system for the V-DIT data augmentation and CNN based fault diagnosis method of claim 1, characterized by, The method comprises the following steps: a data processing module for obtaining aero-engine fault data and obtaining processed fault data according to aero-engine fault data embedding position code; obtaining final noise-added feature data by using diffusion DIT model on processed fault data through T-layer forward noise adding process; The network construction module is configured to construct an intermediate variable according to the processed fault data and introduced noise; construct a DIT-based Transformer architecture network, wherein an input of the network is feature data obtained after each layer is added with noise, and an output of the network is a predicted intermediate variable; establish a loss function according to the constructed intermediate variable and the predicted intermediate variable; and train the constructed DIT-based Transformer architecture network based on the loss function until convergence, to obtain a trained DIT-based Transformer architecture network; The fault data prediction module is configured to construct a V-DIT backstepping model by introducing V-prediction based on the DIT model; obtain T-layer potential fault data based on the V-DIT backstepping model according to the final feature data added with noise and the predicted intermediate variable output by the trained DIT-based Transformer architecture network; obtain a predicted intermediate variable corresponding to the potential fault data by the trained DIT-based Transformer architecture network according to the T-layer potential fault data, and obtain T-1-layer potential fault data by the V-DIT backstepping model according to the predicted intermediate variable, to sequentially backstep and obtain predicted fault data; The fault diagnosis module is configured to construct a fault diagnosis model based on a convolutional neural network, and train the fault diagnosis model by using the predicted fault data and the original fault data, to obtain a fault diagnosis model with fault recognition capability; and diagnose to-be-diagnosed data according to the diagnosis model, to obtain a corresponding diagnosis result.

Images