CN114611233B - Rotating machinery fault imbalance data generation method and computer equipment - Google Patents
Rotating machinery fault imbalance data generation method and computer equipment Download PDFInfo
- Publication number
- CN114611233B CN114611233B CN202210227591.0A CN202210227591A CN114611233B CN 114611233 B CN114611233 B CN 114611233B CN 202210227591 A CN202210227591 A CN 202210227591A CN 114611233 B CN114611233 B CN 114611233B
- Authority
- CN
- China
- Prior art keywords
- data
- scale
- fault
- generating
- gearbox
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/10—Geometric CAD
- G06F30/17—Mechanical parametric or variational design
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/02—Reliability analysis or reliability optimisation; Failure analysis, e.g. worst case scenario performance, failure mode and effects analysis [FMEA]
Abstract
The invention belongs to the technical field of mechanical equipment fault diagnosis, and discloses a rotating machinery fault unbalance data generation method and computer equipment, which comprise the following steps: the method comprises the steps of preprocessing the data of the gearbox, carrying out multiscale on the data of the gearbox, constructing and initializing a multiscale incremental generation countermeasure network model, simultaneously carrying out training and optimization on the multiscale incremental generation countermeasure network model, and generating multiscale fault data by utilizing the optimized multiscale incremental generation countermeasure network model. The invention solves the problem of serious data imbalance commonly existing in the field by outputting high-quality fault data samples in a progressive generation mode. The method can supplement a few classes of samples, and achieves the effect of balancing the data set. The multi-core maximum mean distance MK-MMD measuring method for transfer learning is introduced, so that the distribution of the generated data at each level is closer to the distribution condition of real data, and the generated data can have corresponding mechanicalness.
Description
Technical Field
The invention belongs to the technical field of fault diagnosis of mechanical equipment, and particularly relates to a rotating machinery fault unbalance data generation method and computer equipment.
Background
At present, in recent decades, scientific technology and engineering technology have been developed rapidly, modern mechanical equipment is also complicated and intelligentized more and more, the usability and stability of mechanical equipment are required to be higher, and the development of a reliable and efficient intelligent operation and maintenance method for core components of the mechanical equipment is urgent. The gearbox is a core component in rotating mechanical equipment, for example, in wind power generator equipment, the planetary gearbox bears the operation of a fan transmission system, and under the high-load operation in such long-term extreme environment, a series of common faults such as gear crack, tooth breakage, abrasion and the like are relatively easy to occur. Generally, the cost penalty for post-mortem maintenance is much higher than for scheduled maintenance, while in the environment of work machine operation, predictive maintenance is much more profitable than scheduled maintenance. Effective fault diagnosis is the key of predictive maintenance, and component faults can be found in advance and timely maintained, so that huge safety problems and economic losses caused by accidents are avoided. However, unlike the conventional experimental environment, in the real engineering environment, the failure time of the mechanical component is far shorter than the normal operation time, i.e. there is a serious data set unbalance problem in the field of mechanical failure diagnosis. At present, a data-driven method becomes a mainstream method in the field of fault diagnosis, and a fault diagnosis method based on deep learning is a research hotspot and a development trend. However, these data-driven methods all need to use a proper balanced data set as a basis, and the unbalanced data set causes misleading to the classification of the model, so that the model is more expected to improve the diagnosis accuracy of most classes, and does not concern the classification accuracy of the model on a few classes, which leads to the accuracy of the fault diagnosis model trained by directly using actual engineering data, and the reliability is significantly lower than the model index in the laboratory. Generation of a countermeasure network (GAN) has been one of the mainstream methods in the field of data generation since the proposal in 2014.
The generation confrontation network generates a sub-network and judges the sub-network through construction, the two sub-networks learn to generate data distribution and judge authenticity in a confrontation mode, and after Nash balance is finally achieved, the discriminator cannot distinguish a generated sample from a real sample, and the generated sample output by the generator is considered to be fake, namely the generated sample conforms to the distribution of the real sample. This idea has an initiative, but the original method of generating the countermeasure network still has many problems, such as training is not converged, the model fails, and the mode is crashed. Many researchers in the future have improved the network structure and loss function for generating the countermeasure network, and have been gradually applied to the research field of fault diagnosis. However, the quality of data generated by the conventional method is poor, the mechanistic property of the signal to be provided is difficult to satisfy, and a model for directly generating a high-scale granularity signal is easy to fail in training.
Through the above analysis, the problems and defects of the prior art are as follows: the quality of data generated by the prior art is poor, the mechanicalness of a signal is difficult to meet, and a high-scale granularity signal model generated directly is easy to fail in training; data imbalance in the field of fault diagnosis of critical components of rotating machinery.
The difficulty in solving the above problems and defects is: although some researchers have made this generation of countermeasure structures much more reliable by improving on the original GAN network structure and the loss function. However, the existing GAN-based data generation method still has obvious instability in the data generation of the one-dimensional vibration signal, the model may not converge during training, resulting in training failure, and the quality of the generated data is difficult to guarantee. The problem inevitably increases a great deal of model training work, manual participation is increased in the process, automation of the data generation process is not facilitated, and negative effects are caused to practical application of the generation method in the field.
The significance of solving the problems and the defects is as follows: the MPGAN algorithm provided by the invention is improved aiming at the data generation of the one-dimensional vibration signal, and can ensure the stability of the model training process and the high-quality data generation. A stable and reliable model frame can improve the working efficiency, is beneficial to the automation of the data generation process and is beneficial to promoting the method to fall to the ground in practical application.
Disclosure of Invention
Aiming at the problems in the prior art, the invention provides a rotating machinery fault imbalance data generation method and computer equipment.
The invention is realized in such a way that a rotating machinery fault unbalance data generation method comprises the following steps:
the method comprises the steps of preprocessing the data of the gearbox, carrying out multi-scale processing on the data of the gearbox, constructing and initializing a multi-scale progressive generation countermeasure network model, simultaneously carrying out training and optimization of the multi-scale progressive generation countermeasure network model, and generating multi-scale fault data by utilizing the optimized multi-scale progressive generation countermeasure network model.
Further, the preprocessing the gearbox data comprises:
the method comprises the steps of collecting vibration data of a gearbox by using an acceleration sensor arranged on the gearbox, processing the extracted vibration data of the gearbox by a non-stationary signal processing method of variational modal decomposition, extracting main eigenmode functions IMFs in frequency domain signals, and filtering noise components to obtain a noise-free gearbox frequency domain vibration data set.
Further, the performing gearbox data multiscaling comprises:
the method comprises the steps of performing windowing processing on an original vibration signal according to the same step length as the window length by using sliding windows with different granularity window lengths, extracting the maximum value in each window, and obtaining frequency domain vibration data sets with different scale granularity by using the window lengths with different granularity.
Further, the building and initializing the multi-scale progressive generation countermeasure network model includes:
the method comprises the steps of constructing a model by taking a one-dimensional depth convolution generation countermeasure network 1D-DCGAN suitable for a vibration frequency domain signal data set as a basic framework, initializing parameters of the constructed model by using random numbers obeying Gaussian distribution, and introducing a multi-core maximum mean distance MK-MMD to obtain an initialized multi-scale progressive generation countermeasure network model.
Further, the 1D-DCGAN comprises a generator subnetwork and a discriminator subnetwork, and the loss function is Wasserstein distance, gradient Penalty Penalty and weighted loss of the multinuclear maximum mean distance MK-MMD.
Further, the multi-scale evolutionary generation of the loss function against the network model is as follows:
further, the training and optimizing the multi-scale progressive generation confrontation network model comprises:
(1) Generating a confrontation network model by inputting the lowest-dimensional input noise signal and the lowest-dimensional true fault sample data into a multi-scale progressive generation method, and generating a confrontation network model in a low-dimensional generator network G 1 And a low-dimensional discriminator D 1 Repeatedly iterating countertraining, and iterating and updating the weight and bias of the neural network key points;
(2) When the model reaches Nash equilibrium, G 1 Outputting the generated failure data M 1→1 Fault data M generated by a linear up-sampling method 1→1 M upscaling from a lower dimension scale to a higher dimension scale 1→2 ；
(3) Adding new noise signal, and inputting data into generator network G of higher layer 2 And arbiter network D 2 And, performing iterative training again, G 2 Outputting the generated fault data M of corresponding scale 2→2 ；
(4) And (4) repeating the steps (1) to (3), and performing up-sampling, noise signal adding and countermeasure training on the generated fault data again to obtain multi-scale progressively-increased one-dimensional fault data and multi-scale high-quality fault data.
Further, the generating multi-scale fault data comprises: and the data output by the generator networks at all levels of the trained confrontation network is the multi-scale fault data.
Further, the method for generating the rotating machinery fault imbalance data comprises the following steps:
analyzing a one-dimensional vibration time sequence signal of the planetary gearbox, and removing harmonic waves with low correlation with fault characteristics by using a time-frequency domain transformation method of variational modal decomposition for denoising;
extracting the one-dimensional vibration signals according to the fixed window length and the maximum value by using a sliding window method to obtain a plurality of sample data sets with different scale granularities;
and step three, generating fault data of high-dimensional scale granularity from data of low-order scale granularity by constructing a multi-scale incremental generation countermeasure network MPGAN method, and introducing a migration learning MK-MMD measurement mode to enable the distribution of the generated data at each level to be closer to the real data distribution.
Another object of the present invention is to provide a computer device comprising a memory and a processor, the memory storing a computer program which, when executed by the processor, causes the processor to perform the steps of the rotating machine fault imbalance data generation method.
By combining all the technical schemes, the invention has the advantages and positive effects that: the invention provides an improved method for generating a countermeasure network, which is used for outputting high-quality fault data samples in a progressive generation mode to solve the problem of serious data imbalance commonly existing in the field, supplementing a few types of samples and achieving the effect of balancing a data set. The generation method has higher applicability than a single improved diagnosis model, different diagnosis models can be combined aiming at different fault diagnosis tasks, the overfitting phenomenon of the model to specific tasks is avoided, and the usability of the method in practical application is improved.
The method comprises the steps of taking a planetary gear box as a research object, analyzing a one-dimensional vibration time sequence signal of the planetary gear box, removing harmonic waves with low correlation with fault characteristics by using a time-frequency domain transformation method of variational modal decomposition, removing noise, extracting the one-dimensional vibration signal according to the fixed window length and the maximum value by using a sliding window method, obtaining a plurality of sample data sets with different scale granularities, constructing a multi-scale gradual increase generation countermeasure network MPGAN method, generating fault data of high-dimensional scale granularity by one step for the data of low-scale granularity, and introducing a migration learning MK-MMD measurement mode to enable the distribution of the generated data of each stage to be closer to the real data distribution so as to solve the problem of unbalance of fault samples in the field of rotating machinery.
The invention adopts a variable modal decomposition non-stationary signal processing method, and the original vibration signal is subjected to denoising processing, so that the requirements on noise processing in a real engineering environment are met, the problems of modal aliasing inhibition and endpoint effect inhibition can be solved better, and more stable and reliable preprocessed data can be provided.
By adopting the multi-scale progressive generation countermeasure network MPGAN provided by the invention, a loss function with transfer learning capability is introduced, and high-fine-grained fault data with detailed characteristics can be gradually generated from coarse low-fine-grained data, so that high-quality fault data can be generated, the problem of data unbalance is relieved, the accuracy rate can be improved by 4.01% on a reference diagnosis model, and the generated sample can better accord with the mechanicalness of an original signal.
The invention introduces a migration learning MK-MMD measurement method, so that the distribution of the generated data at each level is closer to the distribution condition of real data, and the generated data can have corresponding mechanicalness.
Drawings
Fig. 1 is a schematic diagram of a method for generating imbalance data of a rotating machine according to an embodiment of the present invention.
Fig. 2 is a flowchart of a method for generating imbalance data of a rotating machine according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail with reference to the following embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and do not limit the invention.
In view of the problems in the prior art, the present invention provides a method for generating imbalance data of rotating machinery faults and a computer device, and the following describes the present invention in detail with reference to the accompanying drawings.
As shown in fig. 1, a method for generating imbalance data of a rotating machine according to an embodiment of the present invention includes:
the method comprises the steps of preprocessing the data of the gearbox, carrying out multiscale on the data of the gearbox, constructing and initializing a multiscale incremental generation countermeasure network model, simultaneously carrying out training and optimization on the multiscale incremental generation countermeasure network model, and generating multiscale fault data by utilizing the optimized multiscale incremental generation countermeasure network model.
The method for preprocessing the data of the gearbox provided by the embodiment of the invention comprises the following steps:
the method comprises the steps of collecting vibration data of a gearbox by using an acceleration sensor arranged on the gearbox, processing the extracted vibration data of the gearbox by a non-stationary signal processing method of variational modal decomposition, extracting main IMFs in frequency domain signals, and filtering noise components to obtain a noise-free gearbox frequency domain vibration data set.
The gear box data multiscale processing method provided by the embodiment of the invention comprises the following steps:
by using sliding windows with different granularity window lengths, windowing is carried out on an original vibration signal according to the step length which is the same as the window length, the maximum value in each window is extracted, and frequency domain vibration data sets with different scale granularity are obtained by using the window lengths with different granularity.
The method for constructing and initializing the multi-scale progressive generation confrontation network model provided by the embodiment of the invention comprises the following steps:
the method comprises the steps of constructing a model by taking a one-dimensional deep convolution generated countermeasure network 1D-DCGAN as a basic framework, initializing parameters of the constructed model by using random numbers obeying Gaussian distribution, introducing MK-MMD, and obtaining an initialized multi-scale progressive generation countermeasure network model.
The 1D-DCGAN provided by the embodiment of the invention comprises a generator subnetwork and a discriminator subnetwork, and the designed loss function is Wassertein distance, and the Gradient Penalty is weighted loss of Gradient Penalty and multinuclear maximum mean distance MK-MMD.
The loss function of the multi-scale progressive generation confrontation network model provided by the embodiment of the invention is as follows:
the training and optimizing method for carrying out multi-scale progressive generation of the confrontation network model provided by the embodiment of the invention comprises the following steps:
(1) Generating a confrontation network model by inputting the lowest-dimensional input noise signal and the lowest-dimensional true fault sample data into a multi-scale progressive generation method, and generating a confrontation network model in a low-dimensional generator network G 1 And a low-dimensional discriminator D 1 Performing repeated iteration confrontation training, and iteratively updating the weight and bias of the neural network key points;
(2) When the model reaches Nash equilibrium, G 1 Outputting the generated failure data M 1→1 Fault data M generated by a linear up-sampling method 1→1 M upscaling from lower dimension scale to higher dimension scale 1→2 ；
(3) Adding new noise signal, and inputting data into higher-level generator network G 2 And arbiter network D 2 And, performing iterative training again, G 2 Outputting the generated fault data M of a corresponding scale 2→2 ；
(4) And (4) repeating the steps (1) to (3), and performing up-sampling, noise signal adding and countermeasure training on the generated fault data again to obtain one-dimensional fault data with multi-scale progressive increase and multi-scale high-quality fault data.
The multi-scale fault data generation method provided by the embodiment of the invention comprises the following steps: and the data output by the generator networks at all levels of the trained confrontation network is the multi-scale fault data.
As shown in fig. 2, a method for generating imbalance data of a rotating machine according to an embodiment of the present invention includes the following steps:
s101, analyzing a one-dimensional vibration time sequence signal of the planetary gear box, and removing harmonic waves with low correlation with fault characteristics by using a time-frequency domain transformation method of variational modal decomposition for denoising;
s102, extracting the one-dimensional vibration signal according to the fixed window length and the maximum value by using a sliding window method to obtain a plurality of sample data sets with different scale granularities;
s103, by constructing a multi-scale incremental generation countermeasure network MPGAN method, the data of low-order scale granularity is used for generating fault data of high-dimensional scale granularity, and a transfer learning MK-MMD measurement mode is introduced, so that the distribution of the generated data at each level is closer to the real data distribution.
The technical solution of the present invention is further described with reference to the following specific embodiments.
Example 1:
a rotating machinery fault imbalance data generation method based on a multi-scale progressive generation countermeasure network specifically comprises the following steps:
(1) Gearbox data preprocessing
The published gearbox data set of the university of connecticut, usa, was used as experimental data, and as shown in fig. 1, vibration data of the gearbox was collected by an acceleration sensor mounted on the gearbox. However, the environment complexity of the operation of the gearbox is high, noise infection needs to be eliminated, the main IMFs in the frequency domain signals are extracted through a non-stationary signal processing method of Variational Modal Decomposition (VMD), noise components are filtered, and a noise-free gearbox frequency domain vibration data set is formed.
(2) Gearbox data multiscaling
The original vibration signal is subjected to windowing treatment according to the step length same as the window length by using sliding windows with different granularity window lengths, the maximum value in each window is extracted to better reserve spectrum peak value information, and vibration signal frequency domain data sets with different granularity can be obtained by using the window lengths with different granularity.
(3) Establishment and initialization of multi-scale progressive generation countermeasure network model
The model is constructed by taking the 1D-DCGAN as a basic framework, and comprises a generator subnetwork and a discriminator subnetwork, wherein a loss function is designed as Wassertein distance, and Gradient Penalty is given to weighted loss of Gradient Penalty and multinuclear maximum mean distance MK-MMD. In the network structure, the extraction capability of the model for multi-scale frequency domain signal features can be improved by using convolution layers, the iterative stability of the model can be improved by using the distribution of each layer of batch standardization specifications, the problems of gradient disappearance and neuron death in training can be solved by using LeakyReLU, and the data size in the generator network can be amplified by using deconvolution. In terms of a loss function, in order to solve the problem of unstable GAN training, the Wassertein distance, namely the distance between two data fields can be better measured, a reliable measure can be given even if the difference between the two distributions is large, so that the model is more stable, and the weight updating of the discriminator can be better controlled by using the Gradient Penalty Penalty, so that the problem that the discriminator cannot be converged due to overlarge weight is avoided. The established model is initialized by parameters by random numbers which obey Gaussian distribution, MK-MMD is introduced, the distance between the distribution domain of the generator and the distribution domain of a real sample can be better measured, the problem of kernel function selection of mapping the MMD to a Regeneration Kernel Hilbert Space (RKHS) can be solved by selecting multi-kernel MMD, and the robustness and the effectiveness of measurement are improved through weighting fusion. The formula for the loss function is as follows:
(4) Training and optimization of multi-scale progressive generation confrontation network model
A method for multi-scale progressive increase of one-dimensional vibration data is provided, and an input noise signal with the lowest dimension and real fault sample data with the lowest dimension are input into a modelAt low vitamin generator network G 1 And a low-dimensional discriminator D 1 After repeated iteration countertraining, the weight and bias of the neural network are iteratively updated, so that G is obtained after the model reaches Nash balance 1 Outputting the generated failure data M 1→1 After linear up-sampling method, from G 1 Generated failure data M 1→1 M upscaling from lower dimension scale to higher dimension scale 1→2 After adding new noise signal, data is input into generator network G of higher layer 2 And arbiter network D 2 After iterative training again, G 2 Outputting the generated fault data M of this scale 2→2 And repeating the steps, and after the generated fault data is subjected to up-sampling, noise signal addition and countermeasure training again, realizing multi-scale progressive increase of the one-dimensional fault data and obtaining multi-scale high-quality fault data which are more consistent with the mechanicalness of real data learned by the model.
(5) Generation of multiscale fault data
In the process of training the multi-scale progressive generation countermeasure network, different scales of generated fault data such as M can be generated at each stage 1→1 ，M 2→2 And after the training of the whole model is completed, fault data output by the generator networks at all levels can be obtained.
(6) Through the steps, high-quality generated data can be obtained, the generated fault data are mixed into the original unbalanced data set, the data set can be recovered to be balanced in category, a fault diagnosis experiment is carried out on the data generated by the gearbox by using a high-stability Support Vector Machine (SVM) as a reference model, the accuracy of fault diagnosis is increased from 90.20% to 94.21% in an unbalanced state by using the generated data balanced data set, the improvement is 4.01%, and the feasibility and the effectiveness of the generated data on the alleviation of the data unbalance problem are verified as a result. In the method, the generated model can select a coupling mode with a diagnosis model, a plurality of different models can be tried through decoupling, the usability of the whole framework is improved, and the generated model and the diagnosis model can be selected to be coupled to improve the end-to-end diagnosis performance. In practical application, an adaptive method can be selected according to fault tasks of different mechanical parts, so that development and landing of practical fault diagnosis projects are facilitated.
It should be noted that embodiments of the present invention can be realized in hardware, software, or a combination of software and hardware. The hardware portion may be implemented using dedicated logic; the software portions may be stored in a memory and executed by a suitable instruction execution system, such as a microprocessor or specially designed hardware. Those skilled in the art will appreciate that the apparatus and methods described above may be implemented using computer executable instructions and/or embodied in processor control code, such code being provided on a carrier medium such as a disk, CD-or DVD-ROM, programmable memory such as read only memory (firmware), or a data carrier such as an optical or electronic signal carrier, for example. The apparatus and its modules of the present invention may be implemented by hardware circuits such as very large scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., or by software executed by various types of processors, or by a combination of hardware circuits and software, e.g., firmware.
The above description is only for the purpose of illustrating the embodiments of the present invention, and the scope of the present invention should not be limited thereto, and any modifications, equivalents and improvements made by those skilled in the art within the technical scope of the present invention as disclosed in the present invention should be covered by the scope of the present invention.
Claims (7)
1. A method of generating rotating machine fault imbalance data, the method comprising: the method comprises the steps of preprocessing gear box data, carrying out multi-scale processing on the gear box data, constructing and initializing multi-scale progression to generate a confrontation network model; training and optimizing a multi-scale progressive generation confrontation network model, and generating multi-scale fault data by utilizing the optimized multi-scale progressive generation confrontation network model;
the performing gearbox data multiscaling includes:
performing windowing processing on an original vibration signal according to the same step length as the window length by using sliding windows with different granularity window lengths, extracting the maximum value in each window, and obtaining frequency domain vibration data sets with different scale granularities by using the window lengths with different granularity;
the constructing and initializing the multi-scale progressive generation countermeasure network model comprises the following steps:
constructing a model by taking 1D-DCGAN as a basic framework, performing parameter initialization on the constructed model by using a random number obeying Gaussian distribution, and introducing MK-MMD to obtain an initialized multi-scale progressive generation confrontation network model;
the training and optimizing for carrying out the multi-scale progressive generation of the confrontation network model comprises the following steps:
(1) Generating a confrontation network model by inputting the lowest-dimensional input noise signal and the lowest-dimensional real fault sample data into a multi-scale progression, and generating a confrontation network model in a low-dimensional generator network G 1 And a low-dimensional discriminator D 1 Performing repeated iteration confrontation training, and iteratively updating the weight and bias of the neural network key points;
(2) When the model reaches Nash equilibrium, G 1 Outputting the generated failure data M 1→1 Fault data M generated by a linear up-sampling method 1→1 M upscaling from lower dimension scale to higher dimension scale 1→2 ；
(3) Adding new noise signal, and inputting data into generator network G of higher layer 2 And arbiter network D 2 And, performing iterative training again, G 2 Outputting the generated fault data M of corresponding scale 2→2 ；
(4) And (4) repeating the steps (1) to (3), and performing up-sampling, noise signal adding and countermeasure training on the generated fault data again to obtain one-dimensional fault data with multi-scale progressive increase and multi-scale high-quality fault data.
2. The method of generating rotary machine fault imbalance data according to claim 1, wherein the preprocessing the gearbox data comprises:
the method comprises the steps of collecting vibration data of a gearbox by using an acceleration sensor arranged on the gearbox, processing the extracted vibration data of the gearbox by a non-stationary signal processing method of variational modal decomposition, extracting main IMFs in frequency domain signals, filtering noise components, and obtaining a noise-free gearbox frequency domain vibration data set.
3. The method of generating rotating machine fault imbalance data according to claim 1, wherein the 1D-DCGAN includes a generator sub-network and a discriminator sub-network, and the loss function is a Wasserstein distance, a gradient penalty gradientpennalty, and a weighted loss of a multi-kernel maximum mean distance MK-MMD.
4. A method of generating rotating machine fault imbalance data according to claim 1, wherein the multi-scale progression generates a penalty function against a network model as follows:
5. a method of generating rotary machine fault imbalance data in accordance with claim 1, wherein said generating multiscale fault data comprises: and the data output by each level of generator network of the trained countermeasure network is the multi-scale fault data.
6. A method of generating rotary machine fault imbalance data according to claim 1, wherein the rotary machine fault imbalance data generating method includes the steps of:
analyzing a one-dimensional vibration time sequence signal of the planetary gearbox, and removing harmonic waves with low correlation with fault characteristics by using a time-frequency domain transformation method of variational modal decomposition for denoising;
extracting the one-dimensional vibration signals according to the fixed window length and the maximum value by using a sliding window method to obtain a plurality of sample data sets with different scale granularities;
and step three, generating fault data of high-dimensional scale granularity by using data of low-order scale granularity through constructing a multi-scale incremental growth method for generating an antagonistic network MPGAN, and introducing a migration learning MK-MMD measurement mode to enable the distribution of the generated data at each level to be closer to the real data distribution.
7. A computer arrangement, characterized in that the computer arrangement comprises a memory and a processor, the memory storing a computer program which, when executed by the processor, causes the processor to carry out the steps of the rotating machine fault imbalance data generation method according to any one of claims 1-6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210227591.0A CN114611233B (en) | 2022-03-08 | 2022-03-08 | Rotating machinery fault imbalance data generation method and computer equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210227591.0A CN114611233B (en) | 2022-03-08 | 2022-03-08 | Rotating machinery fault imbalance data generation method and computer equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114611233A CN114611233A (en) | 2022-06-10 |
| CN114611233B true CN114611233B (en) | 2022-11-11 |
Family
ID=81861349
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210227591.0A Active CN114611233B (en) | 2022-03-08 | 2022-03-08 | Rotating machinery fault imbalance data generation method and computer equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114611233B (en) |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109918999A (en) * | 2019-01-22 | 2019-06-21 | 西安交通大学 | Based on the mechanical equipment fault intelligent diagnosing method for generating model under a kind of Small Sample Database |
| CN110428004A (en) * | 2019-07-31 | 2019-11-08 | 中南大学 | Component of machine method for diagnosing faults under data are unbalance based on deep learning |
| CN112649198A (en) * | 2021-01-05 | 2021-04-13 | 西交思创智能科技研究院(西安)有限公司 | Intelligent fault diagnosis method, system and equipment for quasi-unbalanced rolling bearing and application |
| CN113157771A (en) * | 2021-04-27 | 2021-07-23 | 广东海聊科技有限公司 | Data anomaly detection method and power grid data anomaly detection method |
| CN113376516A (en) * | 2021-06-07 | 2021-09-10 | 科润智能控制股份有限公司 | Medium-voltage vacuum circuit breaker operation fault self-diagnosis and early-warning method based on deep learning |
| CN113408578A (en) * | 2021-05-13 | 2021-09-17 | 华南理工大学 | Harmonic reducer fault diagnosis method and system based on generative countermeasure network |
| CN113887136A (en) * | 2021-10-08 | 2022-01-04 | 东北大学 | Improved GAN and ResNet based electric vehicle motor bearing fault diagnosis method |
| CN113935460A (en) * | 2021-09-27 | 2022-01-14 | 苏州大学 | Intelligent diagnosis method for mechanical fault under class imbalance data set |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111323220B (en) * | 2020-03-02 | 2021-08-10 | 武汉大学 | Fault diagnosis method and system for gearbox of wind driven generator |
-
2022
- 2022-03-08 CN CN202210227591.0A patent/CN114611233B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109918999A (en) * | 2019-01-22 | 2019-06-21 | 西安交通大学 | Based on the mechanical equipment fault intelligent diagnosing method for generating model under a kind of Small Sample Database |
| CN110428004A (en) * | 2019-07-31 | 2019-11-08 | 中南大学 | Component of machine method for diagnosing faults under data are unbalance based on deep learning |
| CN112649198A (en) * | 2021-01-05 | 2021-04-13 | 西交思创智能科技研究院(西安)有限公司 | Intelligent fault diagnosis method, system and equipment for quasi-unbalanced rolling bearing and application |
| CN113157771A (en) * | 2021-04-27 | 2021-07-23 | 广东海聊科技有限公司 | Data anomaly detection method and power grid data anomaly detection method |
| CN113408578A (en) * | 2021-05-13 | 2021-09-17 | 华南理工大学 | Harmonic reducer fault diagnosis method and system based on generative countermeasure network |
| CN113376516A (en) * | 2021-06-07 | 2021-09-10 | 科润智能控制股份有限公司 | Medium-voltage vacuum circuit breaker operation fault self-diagnosis and early-warning method based on deep learning |
| CN113935460A (en) * | 2021-09-27 | 2022-01-14 | 苏州大学 | Intelligent diagnosis method for mechanical fault under class imbalance data set |
| CN113887136A (en) * | 2021-10-08 | 2022-01-04 | 东北大学 | Improved GAN and ResNet based electric vehicle motor bearing fault diagnosis method |
Non-Patent Citations (6)
| Title |
|---|
| Imbalanced fault diagnosis of rolling bearing using improved MsR-GAN and feature enhancement-driven CapsNet;Jie Liu; Changhe Zhang;Xingxing Jiang;《Mechanical Systems and Signal Processing》;20211204;1-24 * |
| Rolling Bearing Fault Diagnosis Method Based on Parallel QPSO-BPNN Under Spark-GPU Platform;Wan, L., Li, H., Zhang, G., Li, C., Man, J., & Xiao, M.;《IEEE Access》;20210419;56786-56801 * |
| 不均衡数据集下基于生成对抗网络的改进深度模型故障识别研究;包萍等;《电子测量与仪器学报》;20190315;第33卷(第03期);176-183 * |
| 数据失衡下基于WGAN和GAPCNN的轴承故障诊断研究;薛振泽; 满君丰; 彭成; 邓河;《计算机应用研究》;20200312;3681-3685 * |
| 深度学习在故障诊断与预测中的应用;余萍等;《计算机工程与应用》;20200331(第03期);1-18 * |
| 用于轴承故障诊断的两步迁移学习法;陶启生; 彭成; 满君丰; 刘翊;《计算机工程与应用》;20210114;303-312 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114611233A (en) | 2022-06-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Wen et al. | A new reinforcement learning based learning rate scheduler for convolutional neural network in fault classification | |
| Fu et al. | Blind parameter identification of MAR model and mutation hybrid GWO-SCA optimized SVM for fault diagnosis of rotating machinery | |
| CN111914883B (en) | Spindle bearing state evaluation method and device based on deep fusion network | |
| CN112883994B (en) | Rotating machinery variable working condition fault diagnosis method with balanced distribution adaptation | |
| CN114323644A (en) | Gear box fault diagnosis and signal acquisition method and device and electronic equipment | |
| Liu et al. | LOSGAN: latent optimized stable GAN for intelligent fault diagnosis with limited data in rotating machinery | |
| CN112163640A (en) | Planetary gearbox fault diagnosis method and system based on deep convolution migration learning | |
| Saravanakumar et al. | Hierarchical symbolic analysis and particle swarm optimization based fault diagnosis model for rotating machineries with deep neural networks | |
| CN116522467B (en) | Prediction method and device for service life of spacecraft component | |
| Li et al. | IFD-MDCN: Multibranch denoising convolutional networks with improved flow direction strategy for intelligent fault diagnosis of rolling bearings under noisy conditions | |
| CN114611233B (en) | Rotating machinery fault imbalance data generation method and computer equipment | |
| Ma et al. | A health indicator construction method based on self-attention convolutional autoencoder for rotating machine performance assessment | |
| He et al. | In-situ fault detection for the spindle motor of CNC machines via multi-stage residual fusion convolution neural networks | |
| Bie et al. | An improved CEEMDAN time-domain energy entropy method for the failure mode identification of the rolling bearing | |
| Yan et al. | Digital twin-assisted imbalanced fault diagnosis framework using subdomain adaptive mechanism and margin-aware regularization | |
| Lu et al. | An imbalanced sample intelligent fault diagnosis method using data enhancement and improved broad learning system | |
| CN114235408B (en) | Bearing fault diagnosis method and system based on improved cascade variational modal decomposition | |
| CN116070126A (en) | Aviation plunger pump oil distribution disc abrasion detection method and system based on countermeasure self-supervision | |
| Wang et al. | Wavelet packet transform-assisted least squares support vector machine for gear wear degree diagnosis | |
| CN113537501B (en) | Electromagnetic crosstalk calibration and slow release method and device and electronic equipment | |
| CN111993158B (en) | Method for generating tool detection model, method, system, device and medium for detecting tool detection model | |
| CN116502516B (en) | Identification method and device for degradation stage of spacecraft component | |
| Wang et al. | A Differential Enhanced ConvNet for Rotating Machinery Diagnosis under Strong Noise | |
| Hofmeister | Vibration-based damage localisation: Impulse response identification and model updating methods | |
| CN113962254B (en) | Knowledge-driven intelligent migration fault diagnosis method and system for industrial robot |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |