CN-121997200-A - Fault sample generation and diagnosis method, system, equipment and program product

CN121997200ACN 121997200 ACN121997200 ACN 121997200ACN-121997200-A

Abstract

The present invention relates to the field of mechanical fault diagnosis, and in particular, to a fault sample generation and diagnosis method, system, device, and program product. The method comprises the steps of S1, collecting vibration signals of different fault types of a bearing to form a data set, S2, constructing and training a condition generation countermeasure network model, wherein the condition generation countermeasure network model comprises a generator model and a discriminator model, the generator model is used for generating a fault sample with a pseudo tag, the discriminator model is used for receiving the sample and judging and classifying the sample, the data set is used for training the model, and S3, the trained generator model is used for generating the fault sample with the pseudo tag. According to the invention, a UNet architecture is introduced into a generator model, and a CBN is combined to code class labels, and by combining jump connection of the UNet with self-adaptive condition learning capability of the CBN, dual information enhancement is realized, so that information loss in a forward propagation process is relieved, and synthesis of high-fidelity condition data is ensured.

Inventors

CHEN ZAIGANG
CHEN XIN
LIU YUQING
XIN JUNSHENG
CHEN JIANGTAO

Assignees

西南交通大学

Dates

Publication Date: 20260508
Application Date: 20251231

Claims (10)

1. A fault sample generation method, characterized by comprising the steps of: s1, collecting and processing sample data, namely collecting vibration signals of different fault types of a bearing, and converting the vibration signals of each fault type into a time-frequency diagram to form a data set for training; S2, constructing and training a condition generation countermeasure network model, wherein the condition generation countermeasure network model comprises a generator model and a discriminator model, the generator model is used for generating a fault sample with a pseudo tag according to a real fault sample, the discriminator model is used for receiving the sample and discriminating and classifying the sample, and the condition generation countermeasure network model is trained by utilizing the data set; And S3, generating a fault sample, namely generating the fault sample with the pseudo tag by utilizing a generator model in the training-completed generation countermeasure network model.
2. A failure sample generation method according to claim 1, wherein, The generator model includes an encoder and a decoder; the encoder comprises a first layer, a second layer, a third layer and a fourth layer, wherein the first layer is a double convolution condition normalization module DoubleConvCBN, the second layer is a first downsampling condition normalization module Down1_CBN, and the third layer is a second downsampling condition normalization module Down2_CBN; the decoder comprises a first layer, a second layer, a third layer and a third layer, wherein the first layer is a first Up-sampling condition normalization module Up1_CBN, the second layer is a second Up-sampling condition normalization module Up2_CBN, and the third layer is a convolution output module Conv; A jumper connection is adopted between the second downsampling condition normalization module Down2 CBN of the encoder and the first upsampling condition normalization module Up1 CBN of the decoder, And a jumper connection is adopted between the first downsampling condition normalization module Down1 CBN of the encoder and the second upsampling condition normalization module Up2 CBN of the decoder.
3. The method for generating a fault sample according to claim 2, wherein the network structure of the discriminator model comprises a discriminating input layer, four full convolution layers, a flattening layer, and a first full connection layer and a second full connection layer which are arranged in parallel and are connected with the output of the flattening layer; the judging input layer is used for inputting sample data; the four full convolution layers are used for performing four convolution downsampling operations on sample data; the flattening layer is used for changing the shape of the tensor; The first full connection layer is a discrimination output for discriminating the true or false of an input sample; The second full connection layer is an auxiliary judging classifier for judging the true or false of the input sample and classifying the input sample.
4. A failure sample generation method according to claim 3, wherein in S2, a counterdamage of a generator model and a discriminator model is calculated using a finger damage, and a total damage of the generator model and the discriminator model is constructed in combination with an auxiliary discrimination classification damage of the pair of counterdamage and the auxiliary discrimination classifier.
5. The method of claim 4, wherein in S2, the generator model is trained using a class conditional feature moment matching loss function; the class conditional feature moment matching loss function can simultaneously align the first moment and the second moment of the generated feature distribution and the real feature distribution in each class.
6. A fault sample generation system, comprising: an input unit for inputting random noise and a given class label; A processing unit for obtaining a synthesized fault sample according to the fault sample generation method as set forth in any one of claims 1 to 5; and the output unit is used for outputting the synthesized fault samples.
7. An electronic device comprising at least one processor and a memory communicatively coupled to the at least one processor, wherein the memory stores a computer program executable by the at least one processor to enable the at least one processor to perform the method of generating a faulty sample of any one of claims 1-5.
8. A computer program product, characterized in that the computer program product comprises a computer program which, when executed by a processor, implements the fault sample generation method of any of claims 1-5.
9. A fault diagnosis method, characterized by comprising: generating a fault sample by using the fault sample generation method according to any one of claims 1 to 5; training a fault diagnosis model constructed based on a deep learning algorithm by utilizing the fault sample; And performing fault diagnosis by using the fault diagnosis model built based on the deep learning algorithm after training.
10. A fault diagnosis system, comprising A sample input unit for inputting a sample to be diagnosed; The fault diagnosis unit is used for carrying out fault diagnosis by using a fault diagnosis model constructed based on a deep learning algorithm, wherein the fault diagnosis model constructed based on the deep learning algorithm is trained by using a fault sample generated by the fault sample generation method according to any one of claims 1-5; and the result output unit is used for outputting a fault diagnosis result.

Description

Fault sample generation and diagnosis method, system, equipment and program product Technical Field The present invention relates to the field of mechanical fault diagnosis, and in particular, to a fault sample generation and diagnosis method, system, device, and program product. Background In industrial rotary machine systems, bearings are used as key support members, and serve the important functions of supporting a rotating shaft and reducing friction, and are often called "industrial joints". However, statistics data show that bearings have a higher failure rate and more frequent failure phenomena than other rotating components, so that it is important to ensure safe and stable operation of industrial equipment for implementation of efficient and accurate intelligent failure diagnosis. In recent years, with the rapid development of artificial intelligence technology, deep learning methods, including convolutional neural networks, cyclic neural networks, self-encoders, transformers, and representative architectures for generating countermeasure networks, have been widely used in the field of industrial fault diagnosis, and significantly promote the evolution of equipment maintenance modes from traditional preventive maintenance to data-driven predictive maintenance, and effectively improve the accuracy and efficiency of fault identification. However, in a practical industrial scenario, real fault data is often difficult to obtain and very limited in number due to stringent requirements for data privacy and operational safety. On one hand, once abnormality is detected, the equipment is usually stopped immediately to avoid accident expansion, and on the other hand, the equipment stably operates for a long time under normal working conditions, so that a fault sample is naturally scarce. The data scarcity severely restricts the training effect of the data-driven intelligent diagnosis model, and restricts the improvement of the diagnosis precision and generalization capability of the data-driven intelligent diagnosis model. To alleviate this problem, researchers have begun to explore the technological paths of data-based generation, augmenting training sets by synthesizing high quality, high fidelity failure samples, thereby enhancing the learning ability of the model. Although generating a generator model such as an antagonism network has a certain potential in the task of fault diagnosis under the condition of limited data, obvious defects still exist in the prior art. Especially in the synthesis process, the existing model has the common serious problem of network forgetting, namely when a deep generation structure is trained, initially input category condition information (such as a fault type label) is gradually attenuated and even lost along with the deepening of the network layer number, so that the generated sample is close to real data in overall distribution, but key distinguishing characteristics of a specific fault category are difficult to accurately keep. The problem makes the composite data limited in effectiveness when being used for downstream diagnosis tasks, and is difficult to truly improve the recognition capability of the model to rare fault types, so that the reliable deployment and application of the intelligent diagnosis system in small sample industrial scenes are hindered. Disclosure of Invention The invention provides a fault sample generation and diagnosis method, a system, equipment and a program product, which aim to solve the problem that a fault sample generated in the prior art has network forgetting. In a first aspect, the present invention provides a fault sample generation method, including the steps of: s1, collecting and processing sample data, namely collecting vibration signals of different fault types of a bearing, and converting the vibration signals of each fault type into a time-frequency diagram to form a data set for training; S2, constructing and training a condition generation countermeasure network model, wherein the condition generation countermeasure network model comprises a generator model and a discriminator model, the generator model is used for generating a fault sample with a pseudo tag according to a real fault sample, the discriminator model is used for receiving the sample and discriminating and classifying the sample, and the condition generation countermeasure network model is trained by utilizing the data set; And S3, generating a fault sample, namely generating the fault sample with the pseudo tag by using a generator model in the trained network model. According to a preferred embodiment, the generator model comprises one encoder and one decoder. The encoder comprises a first layer, a second layer, a third layer and a fourth layer, wherein the first layer is a double convolution condition normalization module DoubleConvCBN, the second layer is a first downsampling condition normalization module down1_CBN, and the third layer is a second downsam