CN112362368A

CN112362368A - Fault diagnosis method, device and system for train traction motor and readable medium

Info

Publication number: CN112362368A
Application number: CN202110045387.2A
Authority: CN
Inventors: 张艺菲; 郑杰; 韩三丰
Original assignee: Siemens Mobility Technologies Beijing Co Ltd
Current assignee: Siemens Mobility Technologies Beijing Co Ltd
Priority date: 2021-01-14
Filing date: 2021-01-14
Publication date: 2021-02-12

Abstract

The invention provides a fault diagnosis method, a device, a system and a readable medium for a train traction motor, wherein the method comprises the following steps: acquiring a vibration signal generated on the traction motor, wherein the vibration signal can represent the vibration state of the traction motor; acquiring an acoustic emission signal generated on the traction motor, wherein the acoustic emission signal can represent the acoustic emission state of the traction motor; extracting and obtaining at least one extracted feature from the vibration signal and the acoustic emission signal; inputting at least one extracted feature into at least one corresponding diagnosis model obtained through pre-training to obtain a fault diagnosis result of the traction motor output by the diagnosis model; the types of the extracted features correspond to diagnostic models, and each diagnostic model is obtained by training at least one extracted feature corresponding to the diagnostic model, which is extracted from historically acquired vibration signals generated on the traction motor and historically acquired acoustic emission signals generated on the traction motor. The scheme can improve the accuracy of fault diagnosis of the traction motor.

Description

Fault diagnosis method, device and system for train traction motor and readable medium

Technical Field

The invention relates to the technical field of rail transit, in particular to a fault diagnosis method, a fault diagnosis device, a fault diagnosis system and a readable medium for a train traction motor.

Background

With the development of electrical and mechanical technologies at home and abroad, a traction motor becomes a main power device, and especially plays a decisive role in the reliable operation of a train set, and sudden failure of the traction motor can cause the accidental stop of a train or even serious accidents. Therefore, early fault diagnosis of the traction motors is critical to improving the reliability of the train consist. The main failures of the traction motor can be classified into two major types, i.e., electrical failures and mechanical failures, wherein the mechanical failures mainly including bearing failures, eccentric failures, shaft bending, and the like are the most frequently occurring failures in the traction motor. Therefore, early diagnosis of mechanical faults of the traction motor is of great significance to guarantee safe operation of the train set.

Vibration analysis is one of the common technical means for motor fault diagnosis, however, the vibration analysis is susceptible to noise and resonance, so that the diagnosis accuracy cannot meet the requirement. In order to make up for the deficiency of vibration analysis, the acoustic emission technology with high signal-to-noise ratio becomes a more ideal substitute technology for the vibration analysis technology. However, the method for diagnosing the mechanical fault of the traction motor by using the acoustic emission technology has the defect that the acoustic emission signal cannot be repeatedly obtained through multiple times of loading in the detection process due to the irreversibility of the acoustic emission, so that the acquisition of the signal in each detection process is very important.

In summary, no matter the vibration technology or the acoustic emission technology, the problem of misjudging the state of the traction motor is easily caused in the process of detecting the mechanical fault, so that the accuracy of fault diagnosis of the traction motor is low.

Disclosure of Invention

The invention provides a fault diagnosis method, a fault diagnosis device, a fault diagnosis system and a readable medium for a train traction motor, which can improve the accuracy of fault diagnosis of the traction motor.

In a first aspect, an embodiment of the present invention provides a method for diagnosing a fault of a train traction motor, where the method includes:

acquiring a vibration signal generated on a traction motor, wherein the vibration signal can represent the vibration state of the traction motor;

acquiring an acoustic emission signal generated on a traction motor, which can characterize the acoustic emission state of the traction motor;

extracting and obtaining at least one extracted feature from the vibration signal and the acoustic emission signal;

inputting the at least one extracted feature into at least one corresponding diagnosis model obtained through pre-training, and obtaining a fault diagnosis result of the traction motor output by the diagnosis model;

the types of the extracted features correspond to the diagnosis models, and each diagnosis model is obtained by utilizing at least one extracted feature extracted from the vibration signals generated on the traction motor and the acoustic emission signals generated on the traction motor, wherein the extracted feature corresponds to the diagnosis model.

In one possible implementation, the step of obtaining the extracted features includes:

extracting performance characteristics of the vibration signals to obtain vibration characteristics;

extracting performance characteristics of the acoustic emission signals to obtain acoustic emission characteristics;

screening the obtained vibration characteristics to obtain at least one target vibration characteristic, wherein the target vibration characteristic is a vibration characteristic of which the corresponding characteristic value changes when the traction motor fails;

screening the obtained acoustic emission characteristics to obtain at least one target acoustic emission characteristic, wherein the target acoustic emission characteristic is an acoustic emission characteristic of which the corresponding characteristic value changes when the traction motor fails;

and fusing each target vibration characteristic and each target acoustic emission characteristic to obtain the extraction characteristic.

fusing the vibration signal and the acoustic emission signal to obtain a fused signal;

performing performance feature extraction on the fusion signal to obtain a primary fusion feature;

screening out primary fusion characteristics of which the characteristic values can change when the traction motor fails from the primary fusion characteristics;

and determining the screened primary fusion features as the extraction features.

In one possible implementation, the performance feature extracting step includes: the performance feature extraction is performed in a time domain range, the performance feature extraction is performed in a frequency domain range, and/or the performance feature extraction is performed in a time-frequency domain range.

In a possible implementation manner, in the case of performing performance feature extraction in the time domain, the performance feature extraction includes extracting at least one of the following:

the peak value, the mean value, the variance, the root mean square value and the standard deviation of the vibration signal; and/or the presence of a gas in the gas,

the peak value, the count, the average signal level, the rise time and the root mean square value of the acoustic emission signal.

In a possible implementation manner, the step of performing performance feature extraction in a frequency domain further includes:

setting at least one characteristic defect frequency band comparison set, wherein each characteristic defect frequency band comparison set comprises a fault category of the traction motor and a characteristic defect frequency band corresponding to the fault category;

the performance feature extraction in the frequency domain comprises the following steps:

carrying out time domain to frequency domain conversion on the obtained signal to obtain a signal frequency spectrum curve in a frequency domain range;

and acquiring the amplitude of the curve corresponding to each characteristic defect frequency band from the obtained signal spectrum curve according to the characteristic defect frequency band comparison set.

In one possible implementation, the performing performance feature extraction in the time-frequency domain includes:

performing wavelet packet transformation on the acquired signal;

and extracting the energy amplitude of each frequency band from the wavelet packet converted signal.

In one possible implementation manner, the method further includes:

and fusing the fault diagnosis results of the traction motor output by the plurality of diagnosis models to obtain the final fault diagnosis result of the traction motor.

In a second aspect, an embodiment of the present invention provides a fault diagnosis apparatus for a train traction motor, including:

a vibration signal acquisition module for acquiring a vibration signal generated on the traction motor, which can represent the vibration state of the traction motor;

an acoustic emission signal acquisition module for acquiring an acoustic emission signal generated at the traction motor, which is capable of characterizing an acoustic emission state of the traction motor;

the characteristic extraction module is used for extracting and obtaining at least one extraction characteristic from the vibration signal obtained by the vibration signal obtaining module and the acoustic emission signal obtained by the acoustic emission signal obtaining module;

the diagnosis result output module is used for inputting the at least one extracted feature obtained by the feature extraction module into at least one corresponding diagnosis model obtained by pre-training to obtain a fault diagnosis result of the traction motor output by the diagnosis model; the types of the extracted features correspond to the diagnosis models, and each diagnosis model is obtained by utilizing at least one extracted feature extracted from the vibration signals generated on the traction motor and the acoustic emission signals generated on the traction motor, wherein the extracted feature corresponds to the diagnosis model.

In one possible implementation, the feature extraction module includes:

a vibration feature extraction unit, which is used for extracting the performance feature of the vibration signal to obtain the vibration feature;

the acoustic emission characteristic extraction unit is used for extracting the performance characteristic of the acoustic emission signal to obtain an acoustic emission characteristic;

a vibration feature screening unit, configured to screen the vibration features obtained by the vibration feature extraction unit to obtain at least one target vibration feature, where the target vibration feature is a vibration feature in which a corresponding feature value changes when the traction motor fails;

the acoustic emission characteristic screening unit is used for screening the acoustic emission characteristics obtained by the acoustic emission characteristic extracting unit to obtain at least one target acoustic emission characteristic, wherein the target acoustic emission characteristic is an acoustic emission characteristic of which the corresponding characteristic value changes when the traction motor fails;

and the characteristic fusion unit is used for fusing each target vibration characteristic obtained by the vibration characteristic screening unit and each target acoustic emission characteristic obtained by the acoustic emission characteristic screening unit to obtain the extraction characteristic.

In another possible implementation manner, the feature extraction module includes:

the signal fusion unit is used for fusing the vibration signal and the acoustic emission signal to obtain a fusion signal;

a fusion signal feature extraction unit, for extracting the performance feature of the fusion signal obtained by the signal fusion unit to obtain the primary fusion feature;

a fused feature screening unit, configured to screen primary fused features, of which feature values change when the traction motor fails, from the primary fused features obtained by the fused signal feature extraction unit;

an extracted feature determining unit for determining the primary fusion feature screened by the fusion feature screening unit as the extracted feature.

In a third aspect, another embodiment of the present invention further provides a fault diagnosis apparatus for a train traction motor, including: at least one memory and at least one processor;

the at least one memory to store a machine readable program;

the at least one processor is configured to invoke the machine-readable program to perform the method of the first aspect.

In a fourth aspect, an embodiment of the present invention provides a system for diagnosing a fault of a train traction motor, including:

the vibration sensor is arranged on the traction motor and is used for acquiring vibration signals generated on the traction motor;

the acoustic emission sensor is arranged on the traction motor and is used for acquiring acoustic emission signals generated on the traction motor;

a fault diagnostic device of a train traction motor and communicatively connected to the vibration sensor and the acoustic emission sensor and configured to:

acquiring the vibration signal acquired by the vibration sensor;

acquiring the acoustic emission signal acquired by the acoustic emission sensor;

inputting the at least one extracted feature into at least one corresponding diagnosis model obtained through pre-training, and obtaining a fault diagnosis result of the traction motor output by the diagnosis model; the types of the extracted features correspond to the diagnosis models, and each diagnosis model is obtained by utilizing at least one extracted feature extracted from the vibration signals generated on the traction motor and the acoustic emission signals generated on the traction motor, wherein the extracted feature corresponds to the diagnosis model.

In one possible implementation, the fault diagnosis device of the train traction motor is configured to perform the following operations when obtaining the extracted features:

In another possible implementation, the fault diagnosis device of the train traction motor is further configured to perform the following operations when obtaining the extracted features:

In a fifth aspect, the present invention also provides a computer-readable medium, on which computer instructions are stored, and when executed by a processor, the computer instructions cause the processor to execute the method mentioned in the first aspect.

According to the technical scheme, after the vibration signals and the acoustic emission signals on the train traction motor are collected, the extracted features are obtained according to the vibration signals and the acoustic emission signals collected historically, and the diagnosis model is obtained through training according to the extracted features. And when the current vibration signal and acoustic emission signal of the train traction motor are acquired, extracting characteristics are obtained according to the current vibration signal and acoustic emission signal, and then the extracting characteristics can be input into the diagnosis model obtained by training to obtain the diagnosis result of the train traction motor. Therefore, the scheme combines the vibration signal and the acoustic emission signal to obtain data for training a diagnosis model, and further obtains the diagnosis model for carrying out fault diagnosis on the train traction motor. The traditional method generally adopts single type data to carry out fault diagnosis, and therefore the condition of misjudgment of the fault of the train traction motor can also occur. Therefore, the mode of fault diagnosis by combining the vibration signal and the acoustic emission signal provided by the scheme can improve the accuracy of fault diagnosis of the train traction motor.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained based on these drawings without creative efforts.

Fig. 1 is a flowchart of a method for diagnosing a fault of a traction motor of a train according to an embodiment of the present invention;

FIG. 2 is a flow diagram of a method of determining extracted features according to one embodiment of the present invention;

FIG. 3 is a flow diagram of another method for determining extracted features provided by one embodiment of the present invention;

fig. 4 is a schematic diagram of a fault diagnosis apparatus for a train traction motor according to an embodiment of the present invention;

fig. 5 is a schematic diagram of another fault diagnosis apparatus for a train traction motor according to an embodiment of the present invention;

fig. 6 is a schematic diagram of a fault diagnosis device for a traction motor of a train according to another embodiment of the present invention;

FIG. 7 is a schematic diagram of a fault diagnosis apparatus for a train traction motor including a memory and a processor according to an embodiment of the present invention;

FIG. 8 is a schematic diagram of a fault diagnostic system for a train traction motor according to one embodiment of the present invention;

fig. 9 is a flowchart of another method for diagnosing a fault in a traction motor of a train according to an embodiment of the present invention.

List of reference numerals

Detailed Description

As mentioned above, at present, when the fault diagnosis of the traction motor is performed, the fault diagnosis of the traction motor is generally performed by adopting a vibration analysis method, but because noise and resonance have a large influence on the vibration analysis, the acoustic emission technology with a high signal-to-noise ratio gradually replaces the vibration analysis technology, but the acoustic emission technology also has disadvantages. Therefore, whether the vibration analysis technology or the acoustic emission technology is adopted, the accuracy in fault diagnosis of the traction motor is low, and serious accidents can occur because fault information of the traction motor is not accurately obtained.

In the embodiment of the invention, the historical vibration signals of the vibration sensor and the acoustic emission signals of the acoustic emission sensor are considered to be simultaneously utilized, and the diagnosis model is obtained through training according to the model training data determined by the historical vibration signals and the historical acoustic emission signals, so that after the newly acquired vibration signals and acoustic emission signals are subjected to feature extraction and screening, the accurate diagnosis data of the traction motor can be obtained by utilizing the diagnosis model. According to the scheme, the faults of the traction motor are diagnosed by fusing the vibration sensor and the acoustic emission sensor, so that misjudgment caused by fault diagnosis by adopting single type data can be well avoided, and the accuracy of fault diagnosis of the traction motor of the train can be improved.

The method, the device, the system and the readable medium for diagnosing the fault of the train traction motor provided by the embodiment of the invention are described in detail below with reference to the accompanying drawings.

As shown in fig. 1, the present invention provides a method 100 for diagnosing a fault of a traction motor of a train, which may include the steps of:

step 101: acquiring a vibration signal generated on the traction motor, wherein the vibration signal can represent the vibration state of the traction motor;

step 102: acquiring an acoustic emission signal generated on the traction motor, wherein the acoustic emission signal can represent the acoustic emission state of the traction motor;

step 103: extracting and obtaining at least one extracted feature from the vibration signal and the acoustic emission signal;

step 104: inputting at least one extracted feature into at least one corresponding diagnosis model obtained through pre-training to obtain a fault diagnosis result of the traction motor output by the diagnosis model; the types of the extracted features correspond to diagnostic models, and each diagnostic model is obtained by training at least one extracted feature corresponding to the diagnostic model, which is extracted from historically acquired vibration signals generated on the traction motor and historically acquired acoustic emission signals generated on the traction motor.

After the vibration signals and the acoustic emission signals on the train traction motor are collected, extracted features are obtained according to the vibration signals and the acoustic emission signals collected historically, and a diagnosis model is obtained through training further according to the extracted features. And when the current vibration signal and acoustic emission signal of the train traction motor are acquired, extracting characteristics are obtained according to the current vibration signal and acoustic emission signal, and then the extracting characteristics can be input into the diagnosis model obtained by training to obtain the diagnosis result of the train traction motor. Therefore, the scheme combines the vibration signal and the acoustic emission signal to obtain data for training a diagnosis model, and further obtains the diagnosis model for carrying out fault diagnosis on the train traction motor. The traditional method generally adopts single type data to carry out fault diagnosis, and therefore the condition of misjudgment of the fault of the train traction motor can also occur. Therefore, the mode of fault diagnosis by combining the vibration signal and the acoustic emission signal provided by the scheme can improve the accuracy of fault diagnosis of the train traction motor.

In the embodiment of the invention, the vibration signal can be acquired by a vibration sensor arranged on the traction motor, and the acoustic emission sensor can be acquired by an acoustic emission sensor arranged on the traction motor. The vibration sensor and the acoustic emission sensor are both arranged on a driving end of a train traction motor to be diagnosed, the vibration sensor is used for acquiring vibration signals of the traction motor in real time, and the acoustic emission sensor is used for acquiring acoustic emission signals of the traction motor in real time. And the data used in the model training is the vibration signals and acoustic emission signals collected in a selected certain historical time period, and a fault diagnosis model of the train traction motor is constructed on the basis of the vibration signals and the acoustic emission signals. Further, when the model needs to be updated, the diagnostic model may be trained again based on the re-selected historical data.

In the embodiment of the present invention, the vibration sensor may include an acceleration sensor, a velocity sensor, a displacement sensor, and the like, and different sensors may be applied to different scenes. The frequency measured by the displacement sensor is 0-10 kHz, accurate low-frequency amplitude and phase can be given in the range, the speed sensor is 5-2 kHz and can generate strong signals for medium-frequency vibration, the acceleration sensor is 5-20 kHz and has strong signals in a high-frequency range, and in order to perform fault analysis in the high-frequency range, the acceleration sensor is generally used as a vibration sensor in fault diagnosis of a train traction motor. The traction motor is mainly made of metal materials with stable structure, the materials have small acoustic anisotropy and small acoustic attenuation coefficient, and the frequency band range is mostly 25 KHz-10 MHz, so that the resonant acoustic emission sensor is selected in practical application. Therefore, the low-frequency to high-frequency signal acquisition can be realized by combining the vibration sensor and the acoustic emission sensor, so that the accuracy of the diagnosis result of the traction motor is improved by signal analysis and processing.

In a possible implementation manner, based on the fault diagnosis method 100 for a train traction motor shown in fig. 1, when the extracted features are determined in step 103, a manner of performing feature extraction and screening on the vibration signal and the acoustic emission signal respectively, and then fusing the screened vibration features and acoustic emission features may be specifically adopted. As shown in fig. 2, the method may specifically include:

step 201: extracting performance characteristics of the vibration signals to obtain vibration characteristics;

step 202: extracting the performance characteristics of the acoustic emission signals to obtain acoustic emission characteristics;

step 203: screening the obtained vibration characteristics to obtain at least one target vibration characteristic, wherein the target vibration characteristic is a vibration characteristic of which the corresponding characteristic value can change when the traction motor fails;

step 204: screening the obtained acoustic emission characteristics to obtain at least one target acoustic emission characteristic, wherein the target acoustic emission characteristic is an acoustic emission characteristic of which the corresponding characteristic value changes when the traction motor fails;

step 205: and fusing the vibration characteristics of each target and the acoustic emission characteristics of each target to obtain the extracted characteristics.

In the embodiment of the invention, the performance characteristics of the vibration signal and the acoustic emission signal are respectively extracted, then the extracted characteristics are subjected to characteristic screening to screen out the characteristics of which the corresponding characteristic values can change when the traction motor fails, and the screened vibration characteristics and the acoustic emission characteristics are further fused to obtain the extracted characteristics. Therefore, the characteristics which change with the characteristic value generated when the traction motor breaks down can be more accurately screened out through characteristic extraction and characteristic screening in the scheme, the model built in the mode has more pertinence to fault diagnosis, the data processing amount can be reduced, and the model training efficiency and the execution efficiency of model diagnosis are improved. In addition, in the scheme provided by this embodiment, when model training is performed, the target vibration feature and the target acoustic emission feature are fused to obtain an extraction feature for performing model training, which is also helpful for solving the problem that the accuracy of the diagnosis result of a traction motor by using a single type of data is not high.

In a possible implementation manner, based on the method 100 for diagnosing a fault of a train traction motor shown in fig. 1, when determining the extracted features in step 103, it may be considered to fuse the acquired vibration signal and acoustic emission signal, and then extract and screen the features of the fused signals. As shown in fig. 3, the specific steps may include:

step 301: fusing the vibration signal and the acoustic emission signal to obtain a fusion signal;

step 302: performing performance feature extraction on the fusion signal to obtain primary fusion features;

step 303: screening out primary fusion characteristics of which the characteristic values can change when the traction motor fails from the primary fusion characteristics;

step 304: and determining the screened primary fusion features as extraction features.

In the embodiment of the invention, after the historically acquired vibration signals and acoustic emission signals are acquired, the historically acquired vibration signals and acoustic emission signals are firstly fused through a signal fusion algorithm or a preset signal fusion method, and then the fused signals are subjected to feature extraction and feature screening so as to screen out primary fusion features of which feature values can change when a traction motor fails, so that the primary fusion features can be determined as extraction features for model training. Further, after the currently acquired vibration signal and acoustic emission signal are acquired, signal fusion can be performed on the currently acquired vibration signal and acoustic emission signal, then feature extraction and screening can be performed on the fused signal, and subsequently, a fault diagnosis result can be obtained by inputting the screened extracted features into a diagnosis model.

In the embodiment of the present invention, unlike the method for obtaining extracted features shown in fig. 2, the method is to fuse the vibration signal and the acoustic emission signal, and the fusion may be performed by a preset fusion algorithm or a fusion method, and then further perform feature extraction and feature screening on the fused signal. When the feature screening is performed, some feature selection algorithms (such as a random forest algorithm) can be used for screening the extracted primary fusion features, so that the primary fusion features of which the corresponding feature values change when the traction motor fails can be determined as the extracted features. Therefore, the signal fusion mode is carried out firstly, the vibration signal and the acoustic emission signal are not required to be processed by the processor when the characteristic extraction and the characteristic screening are carried out, and therefore the memory of the processor can be released, and the execution efficiency of the processor is improved.

In one possible implementation manner, in the method 100 for diagnosing a fault of a train traction motor shown in fig. 2 and 3, the step of extracting the performance characteristics in step 201, step 202, and/or step 302 may specifically include one or more of the following steps: and extracting the performance characteristics in a time domain range, extracting the performance characteristics in a frequency domain range, and extracting the performance characteristics in the time-frequency domain range.

In the embodiment of the present invention, when performing feature extraction on a signal, feature extraction may be performed in multiple dimensions, such as a time domain range, a frequency domain range, and a time-frequency domain range. In practical application, not only can a proper performance dimension be selected for feature extraction according to the condition of the feature value corresponding to the fault type, for example, if the feature change of a certain fault type in the frequency domain can best reflect the fault type, the feature extraction can be selected in the frequency domain. In addition, multi-dimensional feature extraction can be selected, for example, feature extraction is performed on any two or more of a time domain range, a frequency domain range and a time-frequency domain range at the same time, so that the diagnosis accuracy of the diagnosis model can be improved through the multi-dimensional features.

In one possible implementation, as shown in fig. 2 and 3, in the case of performing performance feature extraction in the time domain, the performance feature extraction may include extracting at least one of the following:

peak, count, average signal level, rise time, root mean square value of the acoustic emission signal.

In the embodiment of the invention, when the traction motor fails, the acquired signal is correspondingly changed. And by extracting the peak value, the mean value, the variance, the root mean square value and the standard deviation of the vibration signal, the peak value, the count, the average signal level, the rise time, the root mean square value and the like of the acoustic emission signal, the situation that the signal changes, such as the intensity change (peak value) of the signal and the fluctuation situation (variance, the root mean square value and the standard deviation) of the signal, and the like can be analyzed through the characteristic quantities. Therefore, the parameters are used as the parameters for extracting the time domain characteristics, so that the characteristics which can reflect the change caused when the traction motor breaks down can be more accurately extracted, and the accuracy of the diagnosis model obtained by utilizing the characteristics for training is higher. Meanwhile, the precision of fault diagnosis of the train traction motor by using the diagnosis model is improved.

In a possible implementation manner, as shown in fig. 2 and fig. 3, when performing performance feature extraction in a frequency domain, the method may further include:

thus, the specific steps of extracting the features of the signal in the frequency domain may include:

In the embodiment of the present invention, when performing feature extraction on a signal in a frequency domain, a feature defect frequency band comparison set may be set first, where the feature defect frequency band comparison set should include a fault category of a traction motor and a feature defect frequency band corresponding to the category. And then, the signal can be changed from a time domain to a frequency domain, so that the amplitude corresponding to each characteristic defect frequency band can be obtained from the transformed signal spectrum curve according to the characteristic defect frequency band comparison set. Therefore, according to the scheme, the characteristic defect frequency band comparison set is constructed in advance, so that the amplitude values of the frequency bands corresponding to various faults of the traction motor can be extracted in a targeted manner when the characteristics are extracted in the frequency domain range, and the faults of the traction motor can be diagnosed more accurately according to the acquired characteristics.

In a possible implementation manner, as shown in fig. 2 and fig. 3, when performing performance feature extraction on a signal in a time-frequency domain, the performance feature extraction may be specifically implemented in the following manner:

performing wavelet packet transformation on the acquired signal;

In the embodiment of the invention, when the performance characteristics of the signal are extracted in the time-frequency domain range, the wavelet packet change of the signal is considered. Wavelet packet transformation can divide frequency band parts in multiple levels, further decompose high-frequency parts which are not subdivided in frequency band analysis, and adaptively select corresponding frequency bands according to the characteristics of analyzed signals to enable the corresponding frequency bands to be matched with signal frequency spectrums, so that time-frequency resolution is improved. Therefore, the time-frequency domain spectrogram with higher resolution can be obtained through wavelet packet transformation, and therefore the fault diagnosis precision of the traction motor can be improved by obtaining the amplitude of the corresponding frequency band from the time-frequency domain spectrogram with higher precision.

In a possible implementation manner, as shown in fig. 2 and fig. 3, in the method 100 for diagnosing faults of a train traction motor, after performing performance feature extraction on a plurality of performance dimensions in step 201, step 202, and/or step 302, that is, performing performance feature extraction in a time domain range, performing performance feature extraction in a frequency domain range, and performing performance feature extraction in a time-frequency domain range, a fault diagnosis model corresponding to each performance dimension may be constructed by using features of each performance dimension, and extracted features of each performance dimension may be diagnosed by using the fault diagnosis model corresponding to each performance dimension. Specifically, the process may include:

In the embodiment of the invention, the diagnostic model is obtained by training for each performance dimension by using the obtained extraction features under each performance dimension. Therefore, when the fault state of the traction motor is diagnosed, the extracted features obtained from each performance dimension can be correspondingly input into the fault diagnosis model corresponding to the performance dimension, so that a diagnosis result can be obtained through each fault diagnosis model, and finally, fusion analysis (for example, a voting decision mode) is performed on each diagnosis result to obtain a fault diagnosis result with the highest possibility. Therefore, the method can improve the precision of fault diagnosis in a decision fusion mode, and can optimize and update the fault diagnosis model with larger deviation between the diagnosis result and the real result according to each diagnosis result so as to further improve the precision of fault diagnosis on the traction motor.

As shown in fig. 4, an embodiment of the present invention provides a fault diagnosis apparatus 400 for a train traction motor, including:

a vibration signal acquisition module 401 for acquiring a vibration signal generated at the traction motor, which can represent a vibration state of the traction motor;

an acoustic emission signal acquisition module 402 for acquiring acoustic emission signals generated at the traction motor, which are capable of characterizing the acoustic emission state of the traction motor;

a feature extraction module 403, configured to extract and obtain at least one extracted feature from the vibration signal obtained by the vibration signal obtaining module 401 and the acoustic emission signal obtained by the acoustic emission signal obtaining module 402;

a diagnosis result output module 404, configured to input at least one extracted feature obtained by the feature extraction module 403 into a corresponding at least one pre-trained diagnosis model, so as to obtain a fault diagnosis result of the traction motor output by the diagnosis model; the types of the extracted features correspond to diagnostic models, and each diagnostic model is obtained by training at least one extracted feature corresponding to the diagnostic model, which is extracted from historically acquired vibration signals generated on the traction motor and historically acquired acoustic emission signals generated on the traction motor.

In an embodiment of the present invention, the vibration signal acquisition module 401 may be configured to perform step 101 in the above-described method embodiment, the acoustic emission signal acquisition module 402 may be configured to perform step 102 in the above-described method embodiment, the feature extraction module 403 may be configured to perform step 103 in the above-described method embodiment, and the diagnosis result output module 404 may be configured to perform step 104 in the above-described method embodiment.

In a possible implementation manner, based on the train traction motor fault diagnosis apparatus 400 shown in fig. 4, as shown in fig. 5, the feature extraction module 403 includes:

a vibration feature extraction unit 4031, configured to perform performance feature extraction on the vibration signal to obtain a vibration feature;

an acoustic emission feature extraction unit 4032, which is used for extracting the performance features of the acoustic emission signals to obtain acoustic emission features;

a vibration feature screening unit 4033, configured to screen the vibration features obtained by the vibration feature extraction unit 4031 to obtain at least one target vibration feature, where the target vibration feature is a vibration feature whose corresponding feature value changes when the traction motor fails;

an acoustic emission feature screening unit 4034, configured to screen the acoustic emission features obtained by the acoustic emission feature extraction unit 4032 to obtain at least one target acoustic emission feature, where the target acoustic emission feature is an acoustic emission feature in which a corresponding feature value changes when the traction motor fails;

and the feature fusion unit 4035 is configured to fuse the target vibration features obtained by the vibration feature screening unit 4033 with the target acoustic emission features obtained by the acoustic emission feature screening unit 4034 to obtain extracted features.

In an embodiment of the present invention, the vibration feature extraction unit 4031 may be configured to perform step 201 in the above-described method embodiment, the acoustic emission feature extraction unit 4032 may be configured to perform step 202 in the above-described method embodiment, the vibration feature screening unit 4033 may be configured to perform step 203 in the above-described method embodiment, the acoustic emission feature screening unit 4034 may be configured to perform step 204 in the above-described method embodiment, and the feature fusion unit 4035 may be configured to perform step 205 in the above-described method embodiment.

In another possible implementation manner, based on the train traction motor fault diagnosis apparatus 400 shown in fig. 4, as shown in fig. 6, the feature extraction module 403 may include:

a signal fusion unit 4036 for fusing the vibration signal and the acoustic emission signal to obtain a fusion signal;

a fused signal feature extraction unit 4037, configured to perform performance feature extraction on the fused signal obtained by the signal fusion unit 4036, to obtain a primary fused feature;

a fused feature screening unit 4038, configured to screen primary fused features, of which feature values change when the traction motor fails, from the primary fused features obtained by the fused signal feature extraction unit 4037;

an extracted feature determining unit 4039 configured to determine the primary fusion features screened by the fusion feature screening unit 4038 as extracted features.

In the embodiment of the present invention, the signal fusion unit 4036 may be configured to perform step 301 in the above-described method embodiment, the fused signal feature extraction unit 4037 may be configured to perform step 302 in the above-described method embodiment, the fused feature screening unit 4038 may be configured to perform step 303 in the above-described method embodiment, and the extracted feature determination unit 4039 may be configured to perform step 304 in the above-described method embodiment.

In one possible implementation manner, as shown in fig. 5 and fig. 6, when performing the performance feature extraction, the feature extraction module 403 is configured to perform the performance feature extraction in a time domain, perform the performance feature extraction in a frequency domain, and/or perform the performance feature extraction in a time-frequency domain.

In a possible implementation manner, as shown in fig. 5 and fig. 6, when the feature extraction module 403 performs performance feature extraction in a time domain, at least one of the following is specifically extracted:

In one possible implementation manner, as shown in fig. 5 and fig. 6, when the feature extraction module 403 performs performance feature extraction in the frequency domain, it is configured to perform the following operations:

performing wavelet packet transformation on the acquired signal;

In a possible implementation manner, based on the feature extraction module 403 shown in fig. 5 and 6, the apparatus 400 for diagnosing a fault of a train traction motor may further include:

and the diagnosis result fusion unit is used for fusing the fault diagnosis results of the traction motor output by the plurality of diagnosis models to obtain the final fault diagnosis result of the traction motor.

As shown in fig. 7, another fault diagnosis apparatus 700 for a train traction motor according to an embodiment of the present invention includes: at least one memory 701 and at least one processor 702;

at least one memory 701 for storing a machine-readable program;

at least one processor 702, coupled to the at least one memory 701, is configured to invoke a machine readable program to perform the method 100 for diagnosing a failure of a traction motor of a train as provided by any of the above embodiments.

As shown in fig. 8, an embodiment of the present invention further provides a system 800 for diagnosing a fault of a train traction motor, including: at least one vibration sensor 801, at least one acoustic emission sensor 802 and a train traction motor fault diagnosis device 700;

at least one vibration sensor 801 which is installed on the traction motor and is used for collecting vibration signals generated on the traction motor;

the acoustic emission sensor 802 is arranged on the traction motor and is used for acquiring acoustic emission signals generated on the traction motor;

a fault diagnosis device 700 of a train traction motor, communicatively connected to a vibration sensor 801 and an acoustic emission sensor 802, and configured to perform the following operations:

acquiring a vibration signal acquired by a vibration sensor 801;

acquiring an acoustic emission signal acquired by an acoustic emission sensor 802;

inputting at least one extracted feature into at least one corresponding diagnosis model obtained through pre-training to obtain a fault diagnosis result of the traction motor output by the diagnosis model; the types of the extracted features correspond to diagnostic models, and each diagnostic model is obtained by training at least one extracted feature corresponding to the diagnostic model, which is extracted from historically acquired vibration signals generated on the traction motor and historically acquired acoustic emission signals generated on the traction motor.

In the embodiment of the invention, the vibration sensor and the acoustic emission sensor are simultaneously arranged on the train traction motor to acquire the vibration signal and the acoustic emission signal, and the fault diagnosis device of the train traction motor performs model training through the vibration signal and the acoustic emission signal which are historically acquired by the vibration sensor and the acoustic emission sensor to obtain a diagnosis model. After the training of the diagnosis model is completed, the fault diagnosis device of the train traction motor performs fault diagnosis on the vibration signal and the acoustic emission signal which are currently acquired by the vibration sensor and the acoustic emission sensor by using the fault diagnosis model so as to determine the diagnosis result of the train traction motor. Therefore, the fault diagnosis system of the train traction motor provided by the scheme adopts a mode of combining the vibration sensor and the acoustic emission sensor, and solves the problem of low accuracy when a single type sensor carries out fault diagnosis on the train traction motor.

In one possible implementation, as in the system shown in fig. 8, the failure diagnosis device 700 of the train traction motor is configured to perform the following operations when acquiring the extracted features:

extracting the performance characteristics of the acoustic emission signals to obtain acoustic emission characteristics;

screening the obtained vibration characteristics to obtain at least one target vibration characteristic, wherein the target vibration characteristic is a vibration characteristic of which the corresponding characteristic value can change when the traction motor fails;

and fusing the vibration characteristics of each target and the acoustic emission characteristics of each target to obtain the extracted characteristics.

In another possible implementation manner, as in the system shown in fig. 8, the failure diagnosis device 700 of the train traction motor, when acquiring the extracted features, may further be configured to:

fusing the vibration signal and the acoustic emission signal to obtain a fusion signal;

performing performance feature extraction on the fusion signal to obtain primary fusion features;

and determining the screened primary fusion features as extraction features.

The present invention also provides a computer readable medium storing instructions for causing a machine to perform a train traction motor fault diagnosis method as herein described. Specifically, a system or an apparatus equipped with a storage medium on which software program codes that realize the functions of any of the above-described embodiments are stored may be provided, and a computer (or a CPU or MPU) of the system or the apparatus is caused to read out and execute the program codes stored in the storage medium.

In this case, the program code itself read from the storage medium can realize the functions of any of the above-described embodiments, and thus the program code and the storage medium storing the program code constitute a part of the present invention.

Examples of the storage medium for supplying the program code include a floppy disk, a hard disk, a magneto-optical disk, an optical disk (e.g., CD-ROM, CD-R, CD-RW, DVD-ROM, DVD-RAM, DVD-RW, DVD + RW), a magnetic tape, a nonvolatile memory card, and a ROM. Alternatively, the program code may be downloaded from a server computer via a communications network.

Further, it should be clear that the functions of any one of the above-described embodiments may be implemented not only by executing the program code read out by the computer, but also by causing an operating system or the like operating on the computer to perform a part or all of the actual operations based on instructions of the program code.

Further, it is to be understood that the program code read out from the storage medium is written to a memory provided in an expansion board inserted into the computer or to a memory provided in an expansion module connected to the computer, and then causes a CPU or the like mounted on the expansion board or the expansion module to perform part or all of the actual operations based on instructions of the program code, thereby realizing the functions of any of the above-described embodiments.

Another train traction motor fault diagnosis method 900 provided in the embodiment of the present invention is further described in detail below with reference to a train traction motor fault diagnosis system including a vibration sensor, an acoustic emission sensor, and a train traction motor fault diagnosis device. As shown in fig. 9, the method 900 for diagnosing a fault of a train traction motor includes:

step 901: historical vibration signals collected by the vibration sensor are obtained.

In the embodiment of the present invention, the vibration sensor may include an acceleration sensor, a velocity sensor, a displacement sensor, and the like, and different sensors may be applied to different scenes. The frequency measured by the displacement sensor is 0-10 kHz, accurate low-frequency amplitude and phase can be given in the range, the speed sensor is 5-2 kHz and can generate strong signals for medium-frequency vibration, the acceleration sensor is 5-20 kHz and has strong signals in a high-frequency range, and in order to perform fault analysis in the high-frequency range, the acceleration sensor is generally used as a vibration sensor in fault diagnosis of a train traction motor. The vibration sensor is arranged at the driving end of the train traction motor and used for diagnosing bearing faults, shaft bending faults, eccentric faults and the like in the mechanical faults of the motor.

The historical vibration signal is indicative of the vibration state of the traction motor to be fault diagnosed over a historical period of time that may be determined by human beings based on model training requirements. For example, in the vibration signals of a certain traction motor collected by an acceleration sensor for one year, the vibration signals in a certain month are selected as the steady-state or unsteady-state historical vibration signals applied to the training model.

Step 902: and acquiring historical acoustic emission signals acquired by the acoustic emission sensor.

In the embodiment of the invention, like a vibration sensor, an acoustic emission sensor is arranged at the driving end of a train traction motor and is used for diagnosing bearing faults, shaft bending faults, eccentric faults and the like in motor mechanical faults. As the traction motor is mainly made of metal materials with stable structures, the materials have small acoustic anisotropy, and the frequency band range is mostly 25 KHz-10 MHz, therefore, the resonant acoustic emission sensor is selected in practice. The historical acoustic emission signal represents the acoustic emission state of the traction motor to be subjected to fault diagnosis within a certain historical time period, and the historical time period can be determined manually according to model training requirements. The specific example is shown in the above step 901, and will not be described herein again.

Unlike vibration sensors, acoustic emission sensors are more sensitive to some non-collision and frictional fault detection. For example, the rotor is not centered, thermally bent, unbalanced and the like, and the faults are not greatly related to the vibration state of the motor, so that the acoustic emission sensor can well make up for the deficiency of the vibration sensor.

It is to be added here that the vibration sensors and acoustic emission sensors mounted on the traction motor can comprise a plurality of vibration sensors and a plurality of acoustic emission sensors during the signal acquisition. Therefore, the collected vibration signals and acoustic emission signals can be ensured to be more complete, and the problem that accurate signals are not collected due to low sensitivity or failure of a single vibration sensor or a single acoustic emission sensor can be solved.

Step 903: and determining the extracted features for model training according to the historical vibration signals and the historical acoustic emission signals.

In the embodiment of the present invention, the historical vibration signals and the historical acoustic emission signals respectively acquired in step 901 and step 902 are subjected to operations such as feature extraction and feature screening, so as to determine the extracted features for model training, which characterize the historical vibration signals and the historical acoustic emission signals.

In one possible implementation, step 903 may include the following steps:

step 903 a: extracting the characteristics of the historical vibration signals in a time domain, a frequency domain and/or a time-frequency domain range to obtain time domain vibration characteristics, frequency domain vibration characteristics and/or time-frequency domain vibration characteristics;

step 903 b: extracting the characteristics of the historical acoustic emission signals in a time domain, a frequency domain and/or a time-frequency domain range to obtain time-domain acoustic emission characteristics, frequency-domain acoustic emission characteristics and/or time-frequency domain acoustic emission characteristics;

step 903 c: performing characteristic screening on the time domain vibration characteristic, the frequency domain vibration characteristic and/or the time-frequency domain vibration characteristic to obtain at least one target vibration characteristic;

step 903 d: performing characteristic screening on the time-domain acoustic emission characteristic, the frequency-domain acoustic emission characteristic and/or the time-frequency-domain acoustic emission characteristic to obtain at least one target acoustic emission characteristic;

step 903 e: and fusing the vibration characteristics of each target and the acoustic emission characteristics of each target to obtain the extracted characteristics.

In steps 903a and 903b, feature extraction needs to be performed on the historical vibration signal and the historical acoustic emission signal, and in the feature extraction process of the signals, feature extraction needs to be performed in a time domain, a frequency domain and/or a time-frequency domain range. It should be noted that, when performing feature extraction on the vibration signal and the acoustic emission signal, feature extraction may be performed in any one or more of a time domain, a frequency domain, and a time-frequency domain, for example, feature extraction may be performed only on the time domain, only on the frequency domain, and only on the time-frequency domain. For another example, feature extraction may be performed in the time domain and the frequency domain, in the time domain and the time-frequency domain, or in the frequency domain and the time-frequency domain. For another example, feature extraction can be performed in the time domain, the frequency domain, and the time-frequency domain simultaneously.

When extracting data in the time domain, for each vibration signal, the extracted features may include: peak, mean, variance, root mean square, standard deviation, etc.; for each acoustic emission signal, the extracted features may include: peak, count, Average Signal Level (ASL), rise time, root mean square value, etc.

In the frequency domain data extraction, for each vibration and acoustic emission signal, an envelope spectrum analysis is performed in combination with Hilbert (Hilbert) transform and FFT (fast fourier transform). And extracting the maximum amplitude of the characteristic defect frequency band from the spectrum analysis, and establishing a frequency domain characteristic matrix.

As shown in table 1, when the feature extraction of the vibration signal and the acoustic emission signal is performed in the frequency domain, 8 kinds of common features shown in table 1, that is, an inner orbital ball passing frequency, an outer orbital ball passing frequency, a ball rotation frequency, a cage frequency, a shaft deviation or bending frequency, a mechanical loosening frequency, a static eccentricity, and a dynamic eccentricity frequency are extracted at corresponding frequency values according to frequency values at which a fault is known to occur in advance.

TABLE 1 characteristic diagnostic frequency of traction motors

Ball passing frequency of inner track	ƒ_{Inner part}
		Outer orbital ball through frequency	ƒ_{Outer cover}
Frequency of rotation of the ball	ƒ_Rotate
		Cage frequency	ƒ_{Holding rack}
Deviation or bending of axes	ƒ_r
		Mechanical loosening	(1,2,3,4 and 0.5, 1.5). times. ƒ_r
Static eccentricity	2×ƒ_s
		Dynamic eccentricity	2 x 1 x ƒ of slip frequency_r

Wherein, ƒ_rIs the rotational frequency, ƒ_sIs the line frequency.

Table 1 above is composed of 8 characteristic defect band comparison sets, and each fault type corresponds to one characteristic defect band comparison set. The characteristic defect frequency band comparison set can be constructed according to experience and can also be obtained by analyzing historical data. In this way, after the obtained signal is subjected to time-domain to frequency-domain conversion (such as hilbert transform and fast fourier transform) to obtain a signal spectrum curve in a frequency domain range, the amplitude of the curve corresponding to each characteristic defect band can be obtained from the obtained signal spectrum curve according to the characteristic defect band comparison set.

When the time-frequency domain feature extraction is carried out, the wavelet packet transformation is considered to be capable of carrying out multi-level division on the frequency band part, the high-frequency part which is not subdivided in the frequency band analysis is further decomposed, and the corresponding frequency band can be selected in a self-adaptive mode according to the feature of the analyzed signal to be matched with the signal frequency spectrum, so that the time-frequency resolution is improved. Therefore, the Wavelet Packet Decomposition (WPD) technology is applied to extract the time-frequency domain characteristics of any one or more vibration and acoustic emission signals, the energy ratio of each frequency band is calculated, and therefore the fault diagnosis precision of the traction motor can be improved by obtaining the amplitude of the corresponding frequency band from the time-frequency domain spectrogram with higher precision.

Due to various interferences often exist in the process of collecting the vibration signals and the acoustic emission signals, the obtained data are inaccurate and deviate from the real numerical value. Therefore, before feature extraction is carried out on the vibration signal and the acoustic emission signal, the collected vibration signal and the acoustic emission signal can be preprocessed, authenticity and usability of data are improved, randomness of the vibration signal is analyzed, and therefore a specific processing means is determined. The vibration signal of the equipment is often a disordered signal, and the operation state of the equipment needs to be judged directly through original data, and fault diagnosis is difficult to carry out. The acquired signals can be pre-processed. Common signal preprocessing algorithms include arithmetic mean, weighted mean, five-point cubic smoothing, sliding mean, median, fuzzy control, etc. In addition, the vibration signal and the acoustic emission signal may be subjected to denoising processing, for example, noise interference signals may be filtered by a threshold denoising method.

In step 903c and step 903d, when the extracted features are screened, the features of any one or more performance dimensions in the time domain, the frequency domain, and the time-frequency domain may be screened. For example, the features of the time domain range may be only screened, the features of the frequency domain range may be only screened, and the features of the time frequency domain range may be only screened. For another example, the features of the time domain and the frequency domain range may be screened, and the features of the frequency domain and the frequency domain range may be screened. For another example, the features of the time domain, the frequency domain and the time-frequency domain may be simultaneously screened.

In the feature screening process, vibration features and acoustic emission features of which corresponding feature values can change when the traction motor fails need to be extracted. For example, for the vibration signal of the time domain feature, features including a peak value, a mean value, a variance, a root mean square value, a standard deviation and the like are extracted, and state features which do not cause a fault possibly corresponding to the current state of the traction motor can be filtered from the features through expert knowledge analysis, so that feature screening can be realized. For another example, the size relationship between the extracted features and the feature threshold may be determined by setting the feature threshold, and the determined size relationship may be used as a screening criterion. Similarly, the same is true for feature screening in the frequency domain and the time-frequency domain. The acoustic emission feature is selected in accordance with the vibration feature, and the related features can be selected by expert knowledge, setting a threshold value and the like, which is not described herein.

In the process of carrying out data fusion on the features, the extracted different features can be used as columns, the data quantity of each type of features is used as rows, and a matrix is constructed, so that the screened features are fused and determined as the extracted features. For example, the characteristic peak 1, the mean, the variance, and the standard deviation extracted from the historical vibration signal, and the peak 2, the count, the rise time, and the root mean square value extracted from the historical acoustic emission signal respectively constitute columns of a matrix, and 100, that is, the data amount of the characteristic signal is extracted as rows for each characteristic signal, so that a 100 × 6 matrix is constituted as a characteristic matrix for extracting the characteristic.

It should be noted here that if the generated matrix has too large dimension, which may cause difficulty in data training and model building, the matrix may be reduced by using a reduction algorithm, such as PCA (principal component analysis), so that the matrix is in a suitable dimension.

In another possible implementation manner, step 903 may further include the following steps:

step 903A: fusing the historical vibration signal and the historical acoustic emission signal to obtain a fused signal;

step 903B: extracting the characteristics of the fusion signal in a time domain, a frequency domain and/or a time-frequency domain range to obtain a time domain fusion characteristic, a frequency domain fusion characteristic and/or a time-frequency domain fusion characteristic;

step 903C: performing feature screening on the time domain fusion features, the frequency domain fusion features and/or the time-frequency domain fusion features to obtain at least one target fusion feature, wherein the target fusion feature is a fusion feature of which the corresponding feature value can change when the traction motor fails;

step 903D: and determining each target fusion feature as an extraction feature.

In the embodiment of the present invention, different from the previous implementation, the implementation is to perform signal fusion on the historical vibration signal and the historical acoustic emission signal, and then perform feature extraction and feature screening on the fused signals. When signal fusion is carried out, signals can be fused through a fusion algorithm, and signals can also be subjected to superposition operation to obtain fusion signals. The processes of feature extraction and feature screening for signals are based on the same concept as the above implementation, and are not described herein again.

Step 904: and acquiring a first time-domain vibration characteristic, a first frequency-domain vibration characteristic, a first time-domain acoustic emission characteristic, a first frequency-domain acoustic emission characteristic and/or a first time-frequency-domain acoustic emission characteristic from the extracted characteristics.

In the embodiment of the invention, the first time domain vibration feature is obtained by performing feature extraction and screening on a historical vibration signal in a time domain range, the first frequency domain vibration feature is obtained by performing feature extraction and screening on the historical vibration signal in a frequency domain range, the first time frequency domain vibration feature is obtained by performing feature extraction and screening on the historical vibration signal in the time frequency domain range, and the feature values of the first time domain vibration feature, the first frequency domain vibration feature and the first time frequency domain vibration feature are all changed when the traction motor fails; the first time-domain acoustic emission characteristic is obtained by performing characteristic extraction and screening on the historical acoustic emission signals in a time-domain range, the first frequency-domain acoustic emission characteristic is obtained by performing characteristic extraction and screening on the historical acoustic emission signals in a frequency-domain range, the first time-domain acoustic emission characteristic is obtained by performing characteristic extraction and screening on the historical acoustic emission signals in the time-frequency-domain range, and the characteristic values of the first time-domain acoustic emission characteristic, the first frequency-domain acoustic emission characteristic and the first time-frequency-domain acoustic emission characteristic all change when the traction motor fails.

In the above process of determining and extracting the features, no matter the first way of first extracting and screening the features and then performing feature fusion, or the second way of first performing signal fusion and then performing feature extraction and screening, data processing is performed in the time domain, the frequency domain and/or the time-frequency domain, respectively.

Step 905: and performing model training by using at least one of the first time-domain vibration characteristic, the first frequency-domain vibration characteristic, the first time-domain acoustic emission characteristic, the first frequency-domain acoustic emission characteristic and the first time-frequency-domain acoustic emission characteristic to obtain at least one diagnostic model corresponding to the performance dimension and the signal type.

In the embodiment of the present invention, a corresponding diagnosis model needs to be constructed according to the characteristics of the vibration signal in the time domain, the frequency domain and/or the time-frequency domain acquired in step 904 and the characteristics of the acoustic emission signal in the time domain, the frequency domain and/or the time-frequency domain, so that the collected signals related to the traction motor fault can be analyzed by using the diagnosis models with different performance dimensions.

Step 906: and acquiring a current vibration signal acquired by the vibration sensor, wherein the current vibration signal is used for representing the current vibration state of the traction motor.

Step 907: and acquiring a current acoustic emission signal acquired by the acoustic emission sensor, wherein the current acoustic emission signal is used for representing the current acoustic emission state of the traction motor.

Step 908: and determining a current extraction characteristic according to the current vibration signal and the current acoustic emission signal, wherein the current extraction characteristic is used for representing the characteristics of the current vibration signal and the current acoustic emission signal.

Steps 906 to 908 are processes of collecting vibration signals and acoustic emission signals of the traction motor in the current state in real time, and performing feature extraction and feature screening on the current vibration signals and the current acoustic emission signals. It should be noted that the current vibration signal and acoustic emission signal processing (feature extraction, filtering and fusion) logic is consistent with the historical vibration signal and acoustic emission signal processing logic used in training the model. That is, the collection of the current vibration signal and the current acoustic emission signal is consistent with the way of collecting the historical vibration signal and the historical acoustic emission signal, and the feature extraction and feature screening processes of the current vibration signal and the current acoustic emission signal are consistent with the way of collecting the historical vibration signal and the historical acoustic emission signal (that is, the determination way of the current extracted feature is consistent with the determination way of the extracted feature in the model training), which is not described herein again.

Step 909: and correspondingly inputting each feature data contained in the current extracted features into the corresponding diagnosis model to respectively obtain diagnosis results.

In the embodiment of the present invention, the features obtained in each performance dimension included in the currently extracted features may be input into the fault detection model corresponding to the performance dimension and the signal class, so as to obtain the diagnosis results respectively, where the acoustic emission data is also the same. For example, the current vibration signal is extracted and screened in the frequency domain to obtain features, and the features are input into a fault detection model obtained by training the features obtained by extracting and screening the features in the frequency domain by using the historical vibration signal, so that a diagnosis result corresponding to the diagnosis model is obtained. Other features and fault diagnosis models are also derived from this.

Step 910: and counting the diagnosis results to obtain diagnosis information.

In the embodiment of the present invention, by counting the diagnosis results obtained in step 909, decision-making determination can be performed, so as to obtain the diagnosis information of the traction motor to be measured. Therefore, the situation that a certain type of signal is acquired wrongly or problems exist in data processing in a time domain, a frequency domain and/or a time-frequency domain can be avoided through the method. Therefore, the accuracy of fault diagnosis of the traction motor can be further improved through the method.

For example, the diagnostic results obtained by step 909 are the first group: bearing failure; second group: bending the shaft; third group: bearing failure; and a fourth group: bearing failure; and a fifth group: bearing failure; a sixth group: the bearing failed. Thus, the fault of the traction motor can be determined to be a bearing fault more probably through the decision of statistics. In addition, under careful consideration, the shaft bending fault of the traction motor can be detected according to the diagnosis result of the second group, or the root of the deviation of the second group fault diagnosis is found, and whether the second group diagnosis model needs to be retrained or not is judged.

It should be noted that not all steps and modules in the above flows and system structure diagrams are necessary, and some steps or modules may be omitted according to actual needs. The execution order of the steps is not fixed and can be adjusted as required. The system structure described in the above embodiments may be a physical structure or a logical structure, that is, some modules may be implemented by the same physical entity, or some modules may be implemented by a plurality of physical entities, or some components in a plurality of independent devices may be implemented together.

In the above embodiments, the hardware module may be implemented mechanically or electrically. For example, a hardware module may comprise permanently dedicated circuitry or logic (such as a dedicated processor, FPGA or ASIC) to perform the corresponding operations. A hardware module may also include programmable logic or circuitry (e.g., a general-purpose processor or other programmable processor) that may be temporarily configured by software to perform the corresponding operations. The specific implementation (mechanical, or dedicated permanent, or temporarily set) may be determined based on cost and time considerations.

While the invention has been shown and described in detail in the drawings and in the preferred embodiments, it is not intended to limit the invention to the embodiments disclosed, and it will be apparent to those skilled in the art that various combinations of the code auditing means in the various embodiments described above may be used to obtain further embodiments of the invention, which are also within the scope of the invention.

Claims

1. A method (100) for diagnosing a fault in a traction motor of a train, comprising:

2. The method of claim 1, wherein the step of obtaining the extracted features comprises:

3. The method of claim 1, wherein the step of obtaining the extracted features comprises:

4. A method according to claim 2 or 3, wherein the performance feature extraction step comprises: the performance feature extraction is performed in a time domain range, the performance feature extraction is performed in a frequency domain range, and/or the performance feature extraction is performed in a time-frequency domain range.

5. The method according to claim 4, wherein in the case of performing the performance feature extraction in the time domain, the performance feature extraction includes extracting at least one of:

6. The method of claim 4, wherein the step of performing performance feature extraction in the frequency domain further comprises:

7. The method of claim 4, wherein the performance feature extraction in the time-frequency domain comprises:

performing wavelet packet transformation on the acquired signal;

8. The method of claim 4, further comprising:

9. A fault diagnosis device (400) for a train traction motor, comprising:

a vibration signal acquisition module (401) for acquiring a vibration signal generated at the traction motor, which is capable of characterizing a vibration state of the traction motor;

an acoustic emission signal acquisition module (402) for acquiring an acoustic emission signal generated at the traction motor, which is capable of characterizing an acoustic emission state of the traction motor;

a feature extraction module (403) for extracting and obtaining at least one extracted feature from the vibration signal obtained by the vibration signal obtaining module (401) and the acoustic emission signal obtained by the acoustic emission signal obtaining module (402);

a diagnosis result output module (404) for inputting the at least one extracted feature obtained by the feature extraction module (403) into a corresponding at least one pre-trained diagnosis model to obtain a fault diagnosis result of the traction motor output by the diagnosis model; the types of the extracted features correspond to the diagnosis models, and each diagnosis model is obtained by utilizing at least one extracted feature extracted from the vibration signals generated on the traction motor and the acoustic emission signals generated on the traction motor, wherein the extracted feature corresponds to the diagnosis model.

10. The apparatus of claim 9, wherein the feature extraction module (403) comprises:

a vibration feature extraction unit (4031) for extracting the performance features of the vibration signals to obtain vibration features;

an acoustic emission feature extraction unit (4032) for extracting the performance features of the acoustic emission signal to obtain acoustic emission features;

a vibration feature screening unit (4033) for screening the vibration features obtained by the vibration feature extraction unit (4031) to obtain at least one target vibration feature, wherein the target vibration feature is a vibration feature of which a corresponding feature value changes when the traction motor fails;

an acoustic emission feature screening unit (4034) for screening the acoustic emission features obtained by the acoustic emission feature extraction unit (4032) to obtain at least one target acoustic emission feature, wherein the target acoustic emission feature is an acoustic emission feature whose corresponding feature value changes when the traction motor fails;

and the feature fusion unit (4035) is used for fusing each target vibration feature obtained by the vibration feature screening unit (4033) and each target acoustic emission feature obtained by the acoustic emission feature screening unit (4034) to obtain the extraction feature.

11. The apparatus of claim 9, wherein the feature extraction module (403) comprises:

a signal fusion unit (4036) for fusing the vibration signal and the acoustic emission signal to obtain a fusion signal;

a fused signal feature extraction unit (4037) for performing performance feature extraction on the fused signal obtained by the signal fusion unit (4036) to obtain a primary fused feature;

a fused feature screening unit (4038) for screening out primary fused features, of which feature values are changed when the traction motor fails, from the primary fused features obtained by the fused signal feature extraction unit (4037);

an extracted feature determination unit (4039) for determining the primary fusion feature screened by the fusion feature screening unit (4038) as the extracted feature.

12. A fault diagnosis device (700) for a train traction motor, comprising: at least one memory (701) and at least one processor (702);

the at least one memory (701) for storing a machine readable program;

the at least one processor (702) configured to invoke the machine readable program to perform the method of any of claims 1 to 8.

13. A fault diagnostic system (800) for a train traction motor, comprising:

at least one vibration sensor (801) mounted on the traction motor and used for acquiring vibration signals generated on the traction motor;

at least one acoustic emission sensor (802) mounted on the traction motor and configured to collect acoustic emission signals generated on the traction motor;

a train traction motor fault diagnostic device (700) communicatively connected to said vibration sensor (801) and said acoustic emission sensor (802) and configured to:

acquiring the vibration signal acquired by the vibration sensor (801);

acquiring the acoustic emission signal acquired by the acoustic emission sensor (802);

14. The system of claim 13, wherein the train traction motor fault diagnostic device (700) is configured to, in obtaining the extracted features, perform the following:

15. The system of claim 13, wherein the train traction motor fault diagnostic device (700) is further configured to, in obtaining the extracted features, perform the following:

16. Computer readable medium, characterized in that it has stored thereon computer instructions which, when executed by a processor, cause the processor to carry out the method of any one of claims 1 to 8.