Method and system for diagnosing cross-working-condition fault of planetary gearbox driven by digital-analog fusion

Abstract

The invention discloses a cross-working-condition fault diagnosis method and system for a planetary gearbox driven by digital-analog fusion, and the method comprises the steps: collecting a vibration signal, a rotating speed signal and a torque signal of the planetary gearbox, and carrying out the preprocessing of the vibration signal; monitoring the preprocessed data, and judging whether the working condition is changed or not; when the working condition changes, updating corresponding parameters of the kinetic model according to the acquired rotating speed signal and torque signal data, and generating simulation data matched with the current working condition; dividing the simulation data into an analog signal, a source domain signal and a target domain signal, inputting the analog signal, the source domain signal and the target domain signal into the first model, and training parameters in the first model; excessive classification is effectively avoided, unnecessary calculation burden is reduced, and it is ensured that the classification result can better reflect the real physical characteristics of the data; and meanwhile, through preprocessing, the capability of extracting cross-working-condition fault features by the first model is enhanced, and the robustness and generalization capability of classification of the first model are high.

Description

Technical Field

[0001] The present invention relates to the technical field of planetary gear fault diagnosis, and in particular to a method and system for cross-operating condition fault diagnosis of a planetary gearbox driven by digital-analog fusion. Background Art

[0002] Currently, many invention patents have focused on using data-driven methods to diagnose planetary gearbox faults, but these methods often require a large amount of labeled data. In actual industrial environments, especially when working conditions change, high-quality labeled data is very limited. In order to overcome the above shortcomings, researchers began to explore the combination of dynamic models and data-driven methods, and applied them to gearbox fault diagnosis to make up for the shortcomings of data-driven methods. The dynamic simulation process avoids unnecessary noise interference. The excitation characteristics in the simulation signal are usually more obvious, but the actual measured vibration signal is often interfered by background noise. Therefore, the final effect of the fusion model depends largely on the correlation between the simulation data and the real data, and the detection sensitivity and accuracy may be low when diagnosing faults across working conditions. Summary of the invention

[0003] In view of the deficiencies in the prior art, the object of the present invention is to provide a cross-operating condition fault diagnosis method and system for a planetary gearbox driven by digital-analog fusion.

[0004] In order to achieve the above object, the present invention provides the following technical solutions:

[0005] A cross-operating condition fault diagnosis method for a planetary gearbox driven by digital-analog fusion includes the following steps:

[0006] Collect the vibration signal, speed signal and torque signal of the planetary gearbox and pre-process the vibration signal;

[0007] Monitor the preprocessed data to determine whether the working conditions have changed;

[0008] When the working condition changes, the corresponding parameters of the dynamic model are updated according to the acquired speed signal and torque signal data, and simulation data matching the current working condition is generated;

[0009] The simulation data is divided into a simulation signal, a source domain signal and a target domain signal and inputted into the first model, and the parameters in the first model are trained;

[0010] The real-time rotation speed signal and torque signal of the planetary gearbox are collected and input into the trained first model to perform fault diagnosis on the planetary gearbox.

[0011] In the present invention, preferably, the specific steps of training the parameters in the first model include:

[0012] Inputting the simulation signal, the source domain signal and the target domain signal into the feature extractor in the first model to obtain features corresponding to the simulation signal, the source domain signal and the target domain signal;

[0013] The features corresponding to the source domain signal are input into the label predictor of the first model, and the label predictor outputs the prediction loss L of the source sample y ;

[0014] The features corresponding to the simulated signal and the source domain signal are input into the subdomain discriminator of the first model, and the subdomain discriminator outputs the loss L of subdomain discrimination de ;

[0015] The features corresponding to the source domain signal and the target domain signal are input into the global discriminator of the first model, and the loss L of the global discrimination is output. d ;

[0016] By predicting the loss L y , loss L de and loss L d Back propagation updates the weight parameters θ of the feature extractor in the first model f , the weight parameter θ of the label predictor y , the weight parameter θ of the global discriminator d And the weight parameter θ of the subdomain discriminator de ;

[0017] After repeatedly iterating the first model until a set number of iterations is met, the first model is saved.

[0018] In the present invention, preferably, a piezoelectric acceleration sensor is used to monitor the vibration signal of the planetary gearbox, and a speed torque meter is used to monitor the speed signal and torque signal of the planetary gearbox.

[0019] In the present invention, preferably, the vibration signal is preprocessed in the following specific steps:

[0020] Perform FFT analysis on the original vibration signal to determine the processing frequency band;

[0021] Perform time-domain bandpass filtering according to frequency band;

[0022] Perform Hilbert transform on the time domain signal after bandpass filtering, calculate the amplitude of the analytical signal obtained in the previous step, and obtain the envelope curve;

[0023] Calculate the FFT of the envelope curve to obtain the envelope spectrum.

[0024] In the present invention, preferably, the dynamic model adopts a six-degree-of-freedom gear dynamic model.

[0025] In the present invention, preferably, the generating of simulation data matching the current working condition specifically comprises the steps of:

[0026] The dynamics model generates simulation signals;

[0027] A sliding time window is used to collect 4096 points as one sample every 400 points, and a total of 900 samples are saved as a simulation signal data set.

[0028] The cross-operating-condition fault diagnosis system of a planetary gearbox driven by digital-analog fusion includes an acquisition unit, a preprocessing unit, an operating-condition monitoring unit, a dynamic model and a first model.

[0029] The acquisition unit includes a piezoelectric acceleration sensor and a speed torque meter, wherein the piezoelectric acceleration sensor is used to acquire a vibration signal of the planetary gearbox, and the speed torque meter is used to acquire a speed signal and a torque signal of the planetary gearbox;

[0030] The preprocessing unit is used to preprocess the signal collected by the collection unit;

[0031] The working condition monitoring unit is used to determine the working condition of the planetary gearbox according to the signal processed by the preprocessing unit;

[0032] When the working condition of the dynamic model changes, the dynamic model generates a simulation signal matching the current working condition according to the changed signal, and divides the simulation signal into a simulation signal, a source domain signal and a target domain signal for training the first model;

[0033] The first model includes a feature extractor, a label predictor, a subdomain discriminator and a global discriminator. The feature extractor receives preprocessed data and generates corresponding features. The features are sent to the label predictor, the subdomain discriminator and the global discriminator respectively to generate the final fault classification of the travel gearbox.

[0034] In the present invention, preferably, the first model is trained using the simulation signal generated by the dynamic model, that is, the simulation signal, the source domain signal and the target domain signal are input into the feature extractor to obtain features corresponding to the simulation signal source domain signal and the target domain signal;

[0035] The features corresponding to the source domain signal are input into the label predictor, and the label predictor outputs the prediction loss L of the source sample y ;

[0036] The features corresponding to the simulated signal and the source domain signal are input into the subdomain discriminator, and the subdomain discriminator outputs the loss L of subdomain discrimination de ;

[0037] The features corresponding to the source domain signal and the target domain signal are input into the global discriminator, and the loss L of the global discrimination is output.d ;

[0038] By predicting the loss L y , loss L de and loss L d Back propagation updates the weight parameters θ of the feature extractor of the first model f , the weight parameter θ of the label predictor y , the weight parameter θ of the global discriminator d , and the weight parameter θ of the subdomain discriminator de ;

[0039] After repeatedly iterating the first model until a set number of iterations is met, the first model is saved.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] According to the method of the present invention, when the operating condition changes, the first model can diagnose the fault type. During the training of the first model, the dynamic model is used to supplement the problem of insufficient label samples of the current operating condition, which not only effectively avoids over-classification and reduces unnecessary computational burden, but also ensures that the classification results can better reflect the true physical properties of the data. At the same time, by performing envelope preprocessing on the collected data, the first model's ability to extract cross-operating condition fault features is effectively enhanced, and the robustness of classification and the generalization ability of the overall first model are improved by aligning the conditional distribution of the sub-domain discriminator. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is a flow chart of the cross-operating-condition fault diagnosis method for a planetary gearbox driven by digital-analog fusion as described in the present invention.

[0043] Figure 2 It is a structural schematic diagram of the kinetic model described in the present invention.

[0044] Figure 3 This is a schematic diagram comparing the accuracy of the method described in the present invention with that of different methods in cross-operating condition fault diagnosis. DETAILED DESCRIPTION

[0045] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0046] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used herein in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items.

[0047] See also Figure 1 A preferred embodiment of the present invention provides a cross-operating condition fault diagnosis method for a planetary gearbox driven by digital-analog fusion, which mainly determines the operating condition of the planetary gearbox by monitoring the time domain signal of the vibration sensor and the speed and torque signal of the speed and torque meter. When the operating condition changes, the first model can diagnose the fault type, and the fault type is specifically a healthy gear, a sun gear crack fault, and a planetary gear crack fault. During training, the first model uses a dynamic model to supplement the problem of insufficient label samples of the current operating condition, which not only effectively avoids over-classification and reduces unnecessary computational burden, but also ensures that the classification results can better reflect the real physical properties of the data; at the same time, by performing envelope preprocessing on the collected data, the first model's ability to extract cross-operating condition fault features is effectively enhanced, and the robustness of classification and the generalization ability of the overall first model are improved by aligning the conditional distribution of the subdomain discriminator.

[0048] The specific steps of this method are:

[0049] S1. Collect the vibration signal, speed signal and torque signal of the planetary gearbox and pre-process the vibration signal. The specific pre-processing steps are as follows:

[0050] Perform FFT analysis on the original vibration signal to determine the processing frequency band;

[0051] Perform time-domain bandpass filtering according to frequency band;

[0052] Perform Hilbert transform on the time domain signal after bandpass filtering, calculate the amplitude of the analytical signal obtained in the previous step, and obtain the envelope curve;

[0053] The FFT of the envelope curve is calculated to obtain the envelope spectrum. The fault feature extraction capability of the first model is enhanced through this preprocessing process.

[0054] S2. Monitor the preprocessed data to determine whether the working conditions have changed.

[0055] S3. When the working condition changes, the corresponding parameters of the dynamic model are updated according to the acquired speed signal and torque signal data, that is, the speed and torque parameters of the sun gear at the input end of the dynamic model are updated, and then the updated dynamic model is used to run and generate a simulation signal that matches the current working condition, and a sliding time window is used to collect 4096 points as a sample every 400 points, and a total of 900 samples are saved as a simulation signal data set. The simulation data generated by the dynamic model can make up for the problem of insufficient existing data.

[0056] S4. The simulation data is divided into a simulation signal, a source domain signal, and a target domain signal, and input into the first model, and the parameters in the first model are trained:

[0057] Inputting the simulation signal, the source domain signal and the target domain signal into the feature extractor in the first model to obtain features corresponding to the simulation signal, the source domain signal and the target domain signal;

[0058] The features corresponding to the source domain signal are input into the label predictor of the first model, and the label predictor outputs the prediction loss L of the source sample y ;

[0059] The features corresponding to the simulated signal and the source domain signal are input into the subdomain discriminator of the first model, and the subdomain discriminator outputs the loss L of subdomain discrimination de ;

[0060] The features corresponding to the source domain signal and the target domain signal are input into the global discriminator of the first model, and the loss L of the global discrimination is output. d , ensuring that the label predictor loss L y The simultaneous domain discriminator loss L is as small as possible de , L d As big as possible,

[0061] By predicting the loss L y , loss L de and loss L de Back propagation updates the weight parameters θ of the feature extractor of the first model f , the weight parameter θ of the label predictor y , the weight parameter θ of the global discriminator d , and the weight parameter θ of the subdomain discriminator de ;

[0062] After repeatedly iterating the first model until a set number of iterations is met, the first model is saved.

[0063] S5. Collect the real-time rotation speed signal and torque signal of the planetary gearbox and input them into the trained first model to perform fault diagnosis on the planetary gearbox. The first model outputs a clustering result, and determines the fault type of the planetary gearbox according to the clustering result.

[0064] In this embodiment, a piezoelectric acceleration sensor is used to monitor the vibration signal of the planetary gearbox, and a speed torque meter is used to monitor the speed signal and torque signal of the planetary gearbox.

[0065] In this embodiment, if Figure 2 As shown, the dynamic model adopts a six-degree-of-freedom gear dynamic model. The six-degree-of-freedom model can fully describe the motion behavior of the gear transmission system, including the rotation and displacement of the gears. Compared with the traditional two-dimensional or simplified models, it can more accurately simulate the dynamic characteristics of the gear meshing process, especially the complex interactions of multi-degree-of-freedom systems.

[0066] Another preferred embodiment of the present invention provides a cross-operating-condition fault diagnosis system for a planetary gearbox driven by digital-analog fusion, comprising a collection unit, a preprocessing unit, an operating-condition monitoring unit, a dynamic model and a first model.

[0067] The acquisition unit includes a piezoelectric acceleration sensor and a speed torque meter, wherein the piezoelectric acceleration sensor is used to acquire a vibration signal of the planetary gearbox, and the speed torque meter is used to acquire a speed signal and a torque signal of the planetary gearbox;

[0068] The preprocessing unit is used to preprocess the signal collected by the collection unit;

[0069] The working condition monitoring unit is used to determine the working condition of the planetary gearbox according to the signal processed by the preprocessing unit;

[0070] When the working condition of the dynamic model changes, the dynamic model generates a simulation signal matching the current working condition according to the changed signal, and divides the simulation signal into a simulation signal, a source domain signal and a target domain signal for training the first model;

[0071] The first model includes a feature extractor, a label predictor, a subdomain discriminator and a global discriminator. The feature extractor receives preprocessed data and generates corresponding features. The features are sent to the label predictor, the subdomain discriminator and the global discriminator respectively to generate the final fault classification of the travel gearbox.

[0072] In this embodiment, the first model is trained using the simulation signal generated by the dynamic model, that is, the simulation signal, the source domain signal and the target domain signal are input into the feature extractor to obtain features corresponding to the simulation signal source domain signal and the target domain signal;

[0073] The features corresponding to the source domain signal are input into the label predictor, and the label predictor outputs the prediction loss L of the source sample y ;

[0074] The features corresponding to the simulated signal and the source domain signal are input into the subdomain discriminator, and the subdomain discriminator outputs the loss L of subdomain discrimination de ;

[0075] The features corresponding to the source domain signal and the target domain signal are input into the global discriminator, and the loss L of the global discrimination is output. de ;

[0076] By predicting the loss L y , loss L de and loss L de Back propagation updates the weight parameters θ of the feature extractor in the first model f , the weight parameter θ of the label predictor y , the weight parameter θ of the global discriminator d And the weight parameter θ of the subdomain discriminator de ;

[0077] After repeatedly iterating the first model until a set number of iterations is met, the first model is saved.

[0078] In this embodiment, the first model adopts a domain adversarial neural network (DANN). By adding a subdomain discriminator, the conditional distribution alignment of simulation data and measured data is achieved through adversarial training. This method solves the diagnostic errors caused by data differences.

[0079] In this embodiment, a visualization unit is also included, and the visualization unit visualizes the output data of the first model.

[0080] like Figure 3 As shown, in this embodiment, experimental verification is carried out at speeds of 1200r / min, 1500r / min, and 1800r / min, and loads of 40N·m, 80N·m, and 120N·m. The planetary gearbox experimental data is divided into cases A, B, C, D, and E, which also contain 900 samples. Taking the migration problem A→B as an example, A represents the labeled source domain data, and B represents the unlabeled target domain data. The experimental conditions are shown in Table 1 below, and compared with the other three methods, as shown in Figure 2 As shown in the figure, the results show that the proposed method performs well in cross-operating fault diagnosis. Compared with 1D-CNN, the average diagnostic accuracy of the proposed method is improved by 8.82%, compared with DANN and MK-MMD, the average diagnostic accuracy is improved by 1.65% and 1.77% respectively, and compared with the Time-Domain method, the average diagnostic accuracy of the proposed method is improved by 3.69%.

[0081] Table 1:

[0082]

[0083] In some other preferred embodiments of the present invention, a computer-readable storage medium is provided, storing a computer program. When the computer program is executed by a processor, the processor executes the steps of the method described in the above embodiment.

[0084] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, and other media that can store program codes.

[0085] The above description is a detailed description of the preferred feasible embodiments of the present invention, but the embodiments are not intended to limit the scope of the patent application of the present invention. All equivalent changes or modified changes completed under the technical spirit suggested by the present invention should fall within the patent scope covered by the present invention.

Claims

1. A cross-operating fault diagnosis method for a planetary gearbox driven by digital-analog fusion, characterized in that: Includes steps: Collect the vibration signal, speed signal and torque signal of the planetary gearbox and pre-process the vibration signal; Monitor the preprocessed data to determine whether the working conditions have changed; When the working condition changes, the corresponding parameters of the dynamic model are updated according to the acquired speed signal and torque signal data, and simulation data matching the current working condition is generated; The simulation data is divided into a simulation signal, a source domain signal and a target domain signal and inputted into the first model, and the parameters in the first model are trained; The real-time rotation speed signal and torque signal of the planetary gearbox are collected and input into the trained first model to perform fault diagnosis on the planetary gearbox.

2. The cross-operating condition fault diagnosis method for a planetary gearbox driven by digital-analog fusion according to claim 1 is characterized in that: The specific steps of training the parameters in the first model include: Inputting the simulation signal, the source domain signal and the target domain signal into the feature extractor in the first model to obtain features corresponding to the simulation signal, the source domain signal and the target domain signal; The features corresponding to the source domain signal are input into the label predictor of the first model, and the label predictor outputs the prediction loss L of the source sample y ; The features corresponding to the simulated signal and the source domain signal are input into the subdomain discriminator of the first model, and the subdomain discriminator outputs the loss L of subdomain discrimination de ; The features corresponding to the source domain signal and the target domain signal are input into the global discriminator of the first model, and the loss L of the global discrimination is output. d ; By predicting the loss L y , loss L de and loss L d Back propagation updates the weight parameters θ of the feature extractor in the first model f , the weight parameter θ of the label predictor y , the weight parameter θ of the global discriminator d , and the weight parameter θ of the subdomain discriminator de ; After repeatedly iterating the first model until a set number of iterations is met, the first model is saved.

3. The cross-operating condition fault diagnosis method for a planetary gearbox driven by digital-analog fusion according to claim 2 is characterized in that: A piezoelectric acceleration sensor is used to monitor the vibration signal of the planetary gearbox, and a speed torque meter is used to monitor the speed signal and torque signal of the planetary gearbox.

4. The cross-operating condition fault diagnosis method for a planetary gearbox driven by digital-analog fusion according to claim 1 is characterized in that: Specific steps for preprocessing vibration signals: Perform FFT analysis on the original vibration signal to determine the processing frequency band; Perform time-domain bandpass filtering according to frequency band; Perform Hilbert transform on the time domain signal after bandpass filtering, calculate the amplitude of the analytical signal obtained in the previous step, and obtain the envelope curve; Calculate the FFT of the envelope curve to obtain the envelope spectrum.

5. The cross-operating condition fault diagnosis method for a planetary gearbox driven by digital-analog fusion according to claim 4 is characterized in that: The dynamic model adopts a six-degree-of-freedom gear dynamic model.

6. The cross-operating condition fault diagnosis method for a planetary gearbox driven by digital-analog fusion according to claim 1 is characterized in that: The generating of simulation data matching the current working condition specifically comprises the following steps: The dynamics model generates simulation signals; A sliding time window is used to collect 4096 points as one sample every 400 points, and a total of 900 samples are saved as a simulation signal data set.

7. A planetary gearbox cross-operating condition fault diagnosis system driven by digital-analog fusion, used to implement the planetary gearbox cross-operating condition fault diagnosis method driven by digital-analog fusion as claimed in any one of claims 1 to 6, characterized in that: It includes a collection unit, a preprocessing unit, a working condition monitoring unit, a dynamic model and a first model. The acquisition unit includes a piezoelectric acceleration sensor and a speed torque meter, wherein the piezoelectric acceleration sensor is used to acquire a vibration signal of the planetary gearbox, and the speed torque meter is used to acquire a speed signal and a torque signal of the planetary gearbox; The preprocessing unit is used to preprocess the signal collected by the collection unit; The working condition monitoring unit is used to determine the working condition of the planetary gearbox according to the signal processed by the preprocessing unit; When the working condition of the dynamic model changes, the dynamic model generates a simulation signal matching the current working condition according to the changed signal, and divides the simulation signal into a simulation signal, a source domain signal and a target domain signal for training the first model; The first model includes a feature extractor, a label predictor, a subdomain discriminator and a global discriminator. The feature extractor receives preprocessed data and generates corresponding features. The features are sent to the label predictor, the subdomain discriminator and the global discriminator respectively to generate the final fault classification of the travel gearbox.

8. The cross-operating-condition fault diagnosis system for planetary gearboxes driven by digital-analog fusion according to claim 7 is characterized in that: The first model is trained using the simulation signal generated by the dynamic model, that is, the simulation signal, the source domain signal and the target domain signal are input into the feature extractor to obtain features corresponding to the simulation signal source domain signal and the target domain signal; The features corresponding to the source domain signal are input into the label predictor, and the label predictor outputs the prediction loss L of the source sample y ; The features corresponding to the simulated signal and the source domain signal are input into the subdomain discriminator, and the subdomain discriminator outputs the loss L of subdomain discrimination de ; The features corresponding to the source domain signal and the target domain signal are input into the global discriminator, and the loss L of the global discrimination is output. d ; By predicting the loss L y , loss L de and loss L d Back propagation updates the weight parameters θ of the feature extractor in the first model f , the weight parameter θ of the label predictor y , the weight parameter θ of the global discriminator d And the weight parameter θ of the subdomain discriminator de ; After repeatedly iterating the first model until a set number of iterations is met, the first model is saved.

Images