CN112800855A - Non-invasive real-time fault monitoring method for train bogie - Google Patents
Non-invasive real-time fault monitoring method for train bogie Download PDFInfo
- Publication number
- CN112800855A CN112800855A CN202110003825.9A CN202110003825A CN112800855A CN 112800855 A CN112800855 A CN 112800855A CN 202110003825 A CN202110003825 A CN 202110003825A CN 112800855 A CN112800855 A CN 112800855A
- Authority
- CN
- China
- Prior art keywords
- vibration
- fault
- signal
- bogie
- label
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01H—MEASUREMENT OF MECHANICAL VIBRATIONS OR ULTRASONIC, SONIC OR INFRASONIC WAVES
- G01H17/00—Measuring mechanical vibrations or ultrasonic, sonic or infrasonic waves, not provided for in the preceding groups
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M17/00—Testing of vehicles
- G01M17/08—Railway vehicles
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T90/00—Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation
Abstract
The invention discloses a non-intrusive real-time fault monitoring method for a train bogie, which comprises the following steps: acquiring pre-acquired vibration data M corresponding to each key component and pre-acquired vibration data C corresponding to the total signal measuring points; extracting the signal characteristics of the bogie vibration data and establishing a bogie vibration signal characteristic library; obtaining a multi-label fault identification model S (F)C) (ii) a Real-time vibration data CR at the set signal measurement point is collected, and signal characteristics F in CR are extractedCR(ii) a Executing a multi-objective optimization algorithm to obtain a non-dominated solution set NS; acquiring N groups of independent variables with minimum corresponding optimization target values in the NS as N groups of fault identification preliminary results; using trained model S (F)C) Identifying and outputting the collected signal CRAnd taking the group of the preliminary fault identification results with the largest intersection with the multi-label classification identification results in the N groups of the preliminary fault identification results as final fault identification results. The invention provides a non-invasive real-time fault monitoring method for a train bogie, which does not need a large number of sensors and has high precision and stability.
Description
Technical Field
The invention belongs to the field of train part identification and fault monitoring, and particularly relates to a non-intrusive real-time fault monitoring method for a train bogie.
Background
In recent years, high-speed trains are continuously developed in China, a train bogie is used as a core component of the high-speed train, the stability of the train bogie is related to the running safety of the high-speed train, and component detection and fault identification aiming at the bogie are also paid extensive attention.
The bogie fault monitoring method mainly comprises real-time monitoring and parking maintenance, the parking maintenance frequency is low, and the real-time stability and safety of train operation cannot be guaranteed. Therefore, a real-time monitoring method is generally adopted to monitor the bogie failure.
The existing bogie real-time fault monitoring is mainly realized by mounting a large number of sensors in an intrusive mode. For example, for fault monitoring of a key component prone to fault on a bogie, patent publication No. CN105403420A proposes a train bogie fault monitoring method with multiple sensors integrated, in which multiple sensors are mainly used to simultaneously obtain multiple vibration signals, thereby realizing multipoint fault monitoring. For another example, for fault monitoring of rotating components such as bearings, a patent with publication number CN103018046A proposes a fault monitoring method for a bogie bearing of a motor train of a high-speed motor train, so as to avoid rapid expansion of bearing faults due to high-speed operation of the motor train; the patent with publication number CN104236911A proposes a train bogie service process monitoring and fault diagnosis system, which can monitor the bogie rotating shaft and the bearing in real time.
The method can realize fault monitoring of key parts of the bogie, but all the methods are invasive methods, a large number of sensors are required to be installed to realize real-time measurement at the same time, the cost is high, the redundancy is large, and the arrangement of the sensors can influence the structural optimization and the light weight design of the bogie to a certain extent.
Disclosure of Invention
The invention aims to provide a non-intrusive real-time fault monitoring method for a train bogie, aiming at the defects that a large number of sensors are required to be installed in an intrusive fault monitoring method for the bogie in the prior art, and the non-intrusive real-time fault monitoring method for the bogie does not need a large number of sensors, and is high in precision and stability.
In order to solve the technical problems, the technical scheme adopted by the invention is as follows:
a non-intrusive real-time fault monitoring method for a train bogie is characterized by comprising the following steps:
step 1, vibration data of a total signal measuring point and z key components on a bogie are pre-collected, and pre-collected vibration data M corresponding to each key component and pre-collected vibration data C corresponding to the total signal measuring point are obtained;
step 2, extracting signal characteristics including time domain information, frequency information and fault type information in bogie vibration data based on M and C, and establishing a bogie vibration signal characteristic library; the characteristic library of the vibration signal of the bogie comprises FMAnd FCWherein F isMAs a signal characteristic at z critical components, FCSignal features at the total signal measurement points;
step 3, using FCPerforming multi-label classification training to obtain a multi-label fault recognition model S (F)C) (ii) a Wherein, FCAs input to the multi-tag recognition model, FCThe fault type information in (2) is used as the output of the multi-label identification model;
step 4, collecting real-time vibration data CR at a total signal measuring point, and extracting signal characteristics F containing time domain information and frequency information in the CRCR;
Step 5, the number of key parts which are started simultaneously is the same as that of the key partsSetting an optimization target by taking a key part label of time starting and a fault type label of a key component as independent variables and based on FCRExecuting a multi-objective optimization algorithm to obtain a non-dominated solution set NS; acquiring N groups of independent variables with minimum corresponding optimization target values in the NS as N groups of fault identification preliminary results;
step 6, using the model S (F) trained in step 3C) And (4) identifying the signal CR collected in the step (4), outputting a multi-label classification identification result, comparing the multi-label classification identification result with the N groups of fault identification preliminary results obtained in the step (5), and taking one group of fault identification preliminary results with the largest intersection with the multi-label classification identification result in the N groups of fault identification preliminary results as a fault identification final result.
As a preferable mode, in step 1, the pre-collection condition includes single-vibration-source vibration and multi-vibration-source vibration, the single-vibration-source vibration refers to vibration of a single key part in a normal or known abnormal state, and the multi-vibration-source vibration refers to vibration of a plurality of key parts and corresponding working states when combined randomly.
In a preferred embodiment, in step 1, M ═ L is providedM,m(t)],C=[LC,c(t)]M (t) is the vibration signal value corresponding to each key component, c (t) is the vibration signal value corresponding to the total signal measuring point, LMTag values and L for each key componentM=[lM,lt,lg,ld],LCIs a label value L corresponding to the total signal measurement pointC=[lm,lt,lg,ld],lMNumber labels for each key part,/tIs a time stamp tag,/gAs a fault type label,/dFor vibration signal direction labels, /)mAre key component tags that are activated simultaneously.
As a preferable mode, the step 2 includes:
step 201, performing FFT on m, (t) and c (t) to obtain 1-J harmonic signals corresponding to m (t), and obtaining 1-J harmonic amplitude sets A corresponding to m (t)MAnd phase set BM(ii) a FFT conversion is performed on c (t) to obtain 1 to c (t) corresponding to c (t)J harmonic signal to obtain 1-J harmonic amplitude set A corresponding to c (t)CAnd phase set BC;
Step 202, calculating corresponding data in M and C obtained under the condition of single vibration source vibration in step 1 to obtain the phase difference of each harmonic order of each key component relative to the total signal measuring pointAnd the amplitude ratio u of each harmonic order of each key component relative to the total signal measurement pointM;
Step 203, establishing a bogie vibration signal feature library; the characteristic library of the vibration signal of the bogie comprises FMAnd FCWherein, in the step (A),FC=[LC,AC,BC]。
as a preferable mode, the step 3 includes:
step 301, adding LCInmAnd lgConverting into a binary label value;
step 302, adding FCDividing the training set into a training set and a testing set;
step 303, constructing a multi-label fault recognition model by using a RankingSVM model and learning, wherein the input of the multi-label fault recognition model is ACAnd BCThe output of the multi-label fault identification model is LCThe binary label value in (1); selecting the model with the highest precision on the test set as a trained multi-label fault recognition model S (F)C)。
Preferably, in the step 5, the minimization of the difference between the sum of the vibration signal characteristics of the single vibration sources and the total vibration signal characteristic is one of the optimization objectives; and minimizing the weighted variance of the harmonic order difference values of the vibration signals of the single vibration sources into the second optimization target.
Further, still include: and updating the multi-label fault identification model and the bogie vibration signal feature library based on the confirmed completely correct fault identification final result.
Further, still include:
step 7, combining the time stamp signal t and F in the vibration signal feature libraryCObtaining a database T of the characteristic time sequence of the total vibration signalc(t),Tc(t)={LC(t),AC(t),BC(t) }; and predicting the fault state of the bogie at a certain future time by using historical data in the total vibration signal characteristic time sequence database and adopting an extreme learning machine method.
Compared with the prior art, the invention has the following beneficial effects:
1) the non-invasive method is adopted to carry out real-time fault monitoring on the train bogie and key components, and only vibration signals of a total signal measuring point are required to be obtained during actual monitoring, so that sensor redundancy is avoided, the cost is saved, and the structure optimization and lightweight design of the train bogie are facilitated.
2) A fault diagnosis model is established based on a minimum phasor characteristic error and minimum phasor characteristic variance method, the similarity of each harmonic phasor characteristic of a total signal with a single-vibration-source total signal and a single-vibration-source signal is analyzed, a plurality of optimization targets are obtained, multi-target optimization is executed to obtain a plurality of groups of bogie component fault preliminary identification schemes, and the effectiveness and accuracy of fault identification can be guaranteed.
3) The multi-label identification and verification method of the total vibration signal is provided, the precision and the stability of fault monitoring can be further improved by combining a preliminary identification result, and the non-invasive train bogie component identification and fault monitoring with high precision and high stability are realized.
4) The online training and the feature library updating are realized by adopting a sample updating method, so that the wide adaptability, the accuracy and the stability of the non-invasive bogie component identification and fault monitoring can be further improved.
Drawings
FIG. 1 is a flow chart of an embodiment of the present invention.
Detailed Description
As shown in fig. 1, the train bogie non-intrusive real-time fault monitoring method of the invention comprises the following steps:
step 1, intrusive pre-collecting vibration data of train bogie
The bogie is the most important part of the train, and the safety and the performance of the bogie determine the safety and the stability of the train operation. The main parts and fault-prone parts of the bogie comprise a vehicle body device, a framework, a spring shock absorber, a foundation brake device and a driving mechanism. The required training data of the invention is pre-collected mainly by adopting an experimental method, a vibration sensor is arranged on z key components (comprising a wheel set bearing, a cross beam joint point, a gear box suspension point and the like) and a total signal measuring point (generally positioned on a framework or a side frame) of the train bogie, and the vibration data of the bogie is pre-collected by changing experimental conditions.
The specific vibration signal pre-acquisition scheme is as follows:
(1) and acquiring a single vibration source vibration signal, wherein the single vibration source vibration refers to the vibration of a single key part in a normal or known abnormal state. And obtaining vibration signals M and C of the corresponding key components and the total signal measuring point by respectively simulating the normal operation state and the abnormal operation states of each key component.
(2) The multi-vibration-source vibration signal acquisition refers to the vibration of a plurality of key components and corresponding working states when combined randomly. And obtaining vibration signals M and C of the corresponding key components and the total signal measuring point by simultaneously simulating the normal running state and the abnormal running state of a plurality of key components.
The storage method of the vibration signal may be expressed as M ═ LM,m(t)],C=[LC,c(t)]In the formula, m (t) is the vibration signal value corresponding to each key component, c (t) is the vibration signal value corresponding to the total signal measuring point, LMTag values and L for each key componentM=[lM,lt,lg,ld],LCIs a label value L corresponding to the total signal measurement pointC=[lm,lt,lg,ld],lMA label for the number of each key component for distinguishing the key component to which the signal corresponds,/tIs a time stamp tag,/gFor fault type labels, for marking different fault types,/dFor vibration signal direction labels, for recording the direction of the vibration signal,/mThe key component tags, which are activated simultaneously, are used to mark the key component that is vibrating in the total vibration signal.
In the signal acquisition process, assignment of signal labels can be completed according to the ID of the sensor which transmits back signals, fault types need to be marked manually, each signal is repeatedly acquired for 5 times, namely 5 samples, the total sampling time is 1 minute, and finally pre-acquisition data are measured and mainly used for non-invasive preliminary training of the multi-label fault recognition model.
Step 2, extracting the vibration signal characteristics of the bogie and establishing a characteristic library
The time-frequency domain characteristics of the vibration signals of different components and the phase differences of the vibration signals of different components have uniqueness and distinguishability, the signal characteristics of the same component under different fault conditions also have distinguishability, the time-frequency domain analysis is carried out on the vibration signals, and the classification accuracy of the recognition model can be improved by extracting corresponding signal characteristics.
Extracting signal characteristics including time domain information, frequency information and fault type information in bogie vibration data based on M and C, and establishing a bogie vibration signal characteristic library; the characteristic library of the vibration signal of the bogie comprises FMAnd FCWherein F isMAs a signal characteristic at z critical components, FCSignal features at the total signal measurement points; the method specifically comprises the following steps:
step 201, performing time-frequency domain analysis on the vibration signal
F (t) is used for replacing vibration signals m (t) and c (t) measured by key components and total signal measuring points obtained in the step 1, FFT conversion is carried out on each vibration signal, and sub-harmonic signals { f (f) and { f (f) of 1-J (J is selected according to needs, for example, J is 16) of the vibration signals are obtained1(t),f2(t),…,fJ(t) }, and at the same time, obtaining corresponding harmonic amplitude information A ═ a1(t),a2(t),…,aJ(t) } and harmonic phase informationWherein A and B respectively represent 1 to J harmonicsSet of amplitude values and set of phases, a andrespectively, amplitude and phase values, and subscripts indicate harmonic orders.
Through the above step 201, the 1-J harmonic amplitude set A corresponding to m (t) is obtainedMAnd phase set BMObtaining a 1-J harmonic amplitude set A corresponding to c (t)CAnd phase set BC。
Step 202, calculating the phase difference and amplitude ratio of the signals measured by the sensors at each part relative to the signals at the total measuring point
Calculating corresponding data in M and C obtained under the condition of single vibration source vibration in the step 1, and calculating phase difference of each harmonic of a kth measured single vibration source vibration signal M (t) and a total measuring point vibration signal C (t)Sum-amplitude ratio ukiWherein i represents the harmonic frequency, and the phase differences measured for multiple times are averaged to obtain the phase difference of each harmonic order of the sensor signal of the key component relative to the total signal measuring pointAveraging the amplitude ratios measured for multiple times to obtain the amplitude ratio u of each harmonic order of the key component relative to the total signal measuring pointM. Wherein the content of the first and second substances,uM=[u1,u2,…,uJ],n represents the number of pre-acquisitions under single-vibration-source vibration conditions at a single critical component.
Step 203, constructing a bogie vibration signal feature library
Establishing a bogie vibration signal feature library; the bogie vibration signal feature library comprises vibration signals of different bogie key partsCharacteristic value F ofMAnd the characteristic value F of the vibration signal at the measuring point of the total signalCWherein, in the step (A),FC=[LC,AC,BC]。
step 3, training a non-invasive multi-label fault recognition model
In this step, F is usedCPerforming multi-label classification training to obtain a multi-label fault recognition model S (F)C) (ii) a Wherein, FCAs input to the multi-tag recognition model, FCAs output of the multi-tag identification model. The method specifically comprises the following steps:
step 301, adding LCInmAnd lgConversion to a binary label value: lmAnd after the label is binarized, the label is changed into a 0-1 vector, wherein the value of the opened key part of the bogie corresponding to the label is 1, and otherwise, the value is 0. lgAnd after the label is binarized, the label is changed into a 0-1 vector, wherein the corresponding label value of the fault type of the opened bogie key component is 1, and otherwise, the fault type is 0.
Step 302, adding FCDivided into training and test sets. The training needs to use the total signal sample under all experimental conditions, and 80% of the total vibration signal sample is used as a training set and 20% is used as a testing set.
Step 303, constructing a multi-label fault recognition model by using a RankingSVM model, wherein the input of the multi-label fault recognition model is ACAnd BCThe output of the multi-label fault identification model is LCThe binarized tag value in (1). Learning the relation between the label value and the vibration signal characteristic under different conditions, wherein the RankingSVM adopts a Gaussian kernel as a kernel function, kernel function parameters are determined by 5-fold cross validation, and a model with the highest precision on a test set is selected as a trained multi-label fault recognition model S (F)C). After the model training is completed, the model is directly used for fault monitoring of real-time running train data.
Step 4, non-invasive real-time acquisition and feature extraction of vibration signals of total measuring points of the bogie
And on the actual train running bogie, only installing a vibration sensor at the total signal measuring point, acquiring the vibration signal CR at the total signal measuring point of the bogie in real time, recording the timestamp of the real-time acquired data, transmitting the total vibration signal data to the data storage and processing platform in real time, wherein the data transmission interval is 1 minute, and transmitting in a 4G mode.
Performing real-time signal feature extraction on a data processing platform in the same manner as the step 2 to obtain a real-time total vibration signal feature matrix FCR。
Step 5, bogie component based on non-invasive vibration decomposition and fault primary identification method
The step 5 specifically comprises the following steps:
step 501, selecting a multi-objective optimization method and reasonably selecting corresponding hyper-parameters: and (3) determining the number of the searched particles to be 50, the maximum iteration number to be 50 and the archive size to be 25 by adopting a multi-objective particle swarm optimization model. The model uses a single vibration source vibration signal characteristic library and a key component vibration signal characteristic library under the condition of multi-vibration source vibration.
Step 502, optimizing variables into different component and fault types:
(1) the discrete independent variable can be expressed as
X=[D,(lm1,lm2,…,lmD),(lg1,lg2,…,lgD)]Wherein D represents the number of key sites that are activated simultaneously, (l)m1,lm2,…,lmD) Key site tags indicating simultaneous activation (l)g1,lg2,…,lgD) A fault type tag indicating the critical components that are simultaneously activated.
(2) Corresponding continuous independent variable is
x=[d,(Lm1,Lm2,…,LmD),(Lg1,Lg2,…,LgD)];
Wherein d is [0.5, z +0.5) ] Lmi∈(0,1),Lgi∈(0,1),
And D ═ D +1/2],lmi=[Lmi],lgi=[Lgi]I.e. D is rounded up by D, lmi、lgiAre respectively Lmi、LgiRounding of (d) to get the whole.
Step 503, setting an optimization target to minimize the difference between the sum of the vibration signal characteristics of the single vibration sources and the total vibration signal characteristic as one of the optimization targets; and minimizing the weighted variance of the harmonic order difference values of the vibration signals of the single vibration sources into the second optimization target. The optimization objective function is as follows:
wherein the content of the first and second substances,a j-th harmonic phasor value representing the real-time total signal, the phasor consisting of its respective amplitude and phase;representing the j-th harmonic phasor of a vibration signal of a total measuring point when the ith key component vibrates under a pre-collected single vibration source;representing the j-th harmonic phasor of the vibration signal at the ith key component when the ith key component vibrates under the pre-acquisition single vibration source, wherein the phasor is obtained by the transformation ratio amplitude A of the vibration source signalMc=AMU, phase and phase difference; w is ajA is the total number of vibration parts as the weight of each order harmonic.
And step 504, executing multi-objective optimization, and recording the iteration number It as 1. And calculating the optimization function values of all the search results, and selecting a non-dominated solution to store in the file.
And 505, updating the search path to generate a new bogie component and fault combination scheme.
Step 506, if It is less than the maximum iteration number, returning to step 504; otherwise, the multi-objective optimization algorithm is ended, and the non-dominated solution set NS in the final archive is output.
Step 507, the 5 sets of argument values { L } with the minimum corresponding optimization target value in the non-dominated solution set NS are selectedCR1,LCR2,LCR3,LCR4,LCR5As a preliminary result of truck non-intrusive fault identification.
Step 6, non-invasive bogie each part and fault final identification model
Step 601, adopting the RankingSVM multi-label recognition model S (F) trained in step 3C) Identifying the real-time total vibration signal characteristic value acquired in the step 4, and outputting a multi-label classification identification result:
LCR0=S(FCR) In the formula, LCRAnd representing the multi-label bogie fault identification and fault judgment results.
Step 602, adopting the multi-label classification result to identify the 5 groups of fault initial results { L } obtained in step 5CR1,LCR2,LCR3,LCR4,LCR5And (4) verifying and screening, and selecting a preliminary fault identification result which is the same as the fault identification and fault judgment results of the bogie in the step 601 or has the largest intersection as a final fault identification result of the non-invasive bogie.
[I,N]=maxi=1,2,3,4,5{sum(LCR0=LCRi)}
LCR=LCRI
In the formula, I represents and LCR0The subscript value of the primary recognition result with the largest intersection, N, represents the intersectionNumber of collection elements, LCRIIs represented by the formulaCR0And (5) the primary recognition result with the largest intersection.
Step 603, due to the difference of vibration signals of all parts of different train bogies, the multi-label identification model and the characteristic library need to be updated regularly. After the fault identification is completed, the maintainers confirm that a completely correct identification result and the corresponding total vibration signal characteristic matrix are used as new samples to be stored in the data storage module, when the number of accumulated new samples reaches 500, the total vibration signal sample data to be collected and the new samples are integrated, the step 3 is executed, and the multi-label fault identification model and the steering frame vibration signal characteristic library are updated.
Step 7, non-intrusive bogie fault prediction
Decomposing accumulated data based on the non-invasive vibration characteristics of the steps 5 and 6 to gradually form a total vibration signal characteristic time sequence database Tc(t),Tc(t)={LC(t),AC(t),BC(t) }, in which LCCombining the time stamp signal t to form LC(t),ACCombining the time stamp signal t to form AC(t),BCCombining the time stamp signals t to form BC(t) of (d). And the extreme learning machine is adopted to carry out non-invasive prediction on the vibration characteristic time sequence, so that the change condition of the vibration characteristic of the bogie in the future can be found, and the non-invasive fault prediction and early warning are realized.
Step 701, resampling the vibration signal characteristic time sequence of the total measuring point and the vibration characteristic time sequences of all the parts, and increasing the minimum interval of time sequence data to 1 hour.
Step 702, predicting the vibration characteristics of the total signal measuring point at the T + n moment by using the historical data and adopting an extreme learning machine to obtain Tc(t+n)。
And 703, repeating the step 5 and the step 6, and identifying faults of all parts of the bogie at the moment t + n to realize fault prediction.
And step 704, performing important monitoring and inspection on the part which is predicted to generate the fault according to the prediction result, and preventing the fault from occurring.
While the present invention has been described with reference to the embodiments shown in the drawings, the present invention is not limited to the embodiments, which are illustrative and not restrictive, and it will be apparent to those skilled in the art that various changes and modifications can be made therein without departing from the spirit and scope of the invention as defined in the appended claims.
Claims (8)
1. A non-intrusive real-time fault monitoring method for a train bogie is characterized by comprising the following steps:
step 1, vibration data of a total signal measuring point and z key components on a bogie are pre-collected, and pre-collected vibration data M corresponding to each key component and pre-collected vibration data C corresponding to the total signal measuring point are obtained;
step 2, extracting signal characteristics including time domain information, frequency information and fault type information in bogie vibration data based on M and C, and establishing a bogie vibration signal characteristic library; the characteristic library of the vibration signal of the bogie comprises FMAnd FCWherein F isMAs a signal characteristic at z critical components, FCSignal features at the total signal measurement points;
step 3, using FCPerforming multi-label classification training to obtain a multi-label fault recognition model S (F)C) (ii) a Wherein, FCAs input to the multi-tag recognition model, FCThe fault type information in (2) is used as the output of the multi-label identification model;
step 4, collecting real-time vibration data CR at a total signal measuring point, and extracting signal characteristics F containing time domain information and frequency information in the CRCR;
Step 5, setting an optimization target by taking the number of the key parts which are started simultaneously, the labels of the key parts which are started simultaneously and the labels of the fault types of the key components as independent variables, and setting the optimization target based on FCRExecuting a multi-objective optimization algorithm to obtain a non-dominated solution set NS; acquiring N groups of independent variables with minimum corresponding optimization target values in the NS as N groups of fault identification preliminary results;
step 6, using the model S (F) trained in step 3C) And (4) identifying the signal CR collected in the step (4), outputting a multi-label classification identification result, comparing the multi-label classification identification result with the N groups of fault identification preliminary results obtained in the step (5), and taking one group of fault identification preliminary results with the largest intersection with the multi-label classification identification result in the N groups of fault identification preliminary results as a fault identification final result.
2. The method for non-intrusive real-time fault monitoring of the train bogie according to claim 1, wherein in the step 1, the pre-collection conditions comprise single-vibration-source vibration and multi-vibration-source vibration, the single-vibration-source vibration refers to vibration of a single key part in a normal or known abnormal state, and the multi-vibration-source vibration refers to vibration of a plurality of key parts and corresponding working states in random combination.
3. The method for non-intrusive real-time fault monitoring of a train bogie as defined in claim 2, wherein in step 1, M ═ LM,m(t)],C=[LC,c(t)]M (t) is the vibration signal value corresponding to each key component, c (t) is the vibration signal value corresponding to the total signal measuring point, LMTag values and L for each key componentM=[lM,lt,lg,ld],LCIs a label value L corresponding to the total signal measurement pointC=[lm,lt,lg,ld],lMNumber labels for each key part,/tIs a time stamp tag,/gAs a fault type label,/dFor vibration signal direction labels, /)mAre key component tags that are activated simultaneously.
4. The method for non-intrusive real-time fault monitoring of a train bogie as set forth in claim 3, wherein said step 2 comprises:
step 201, performing FFT on m, (t) and c (t) to obtain 1-J harmonic signals corresponding to m (t), and obtaining 1-J harmonic amplitude sets A corresponding to m (t)MAnd phase set BM(ii) a FFT conversion is carried out on c (t) to obtain the correspondence of c (t)Obtaining the 1-J harmonic amplitude set A corresponding to c (t)CAnd phase set BC;
Step 202, calculating corresponding data in M and C obtained under the condition of single vibration source vibration in step 1 to obtain the phase difference of each harmonic order of each key component relative to the total signal measuring pointAnd the amplitude ratio u of each harmonic order of each key component relative to the total signal measurement pointM;
5. the method for non-intrusive real-time fault monitoring of a train bogie as set forth in claim 4, wherein said step 3 comprises:
step 301, adding LCInmAnd lgConverting into a binary label value;
step 302, adding FCDividing the training set into a training set and a testing set;
step 303, constructing a multi-label fault recognition model by using a RankingSVM model and learning, wherein the input of the multi-label fault recognition model is ACAnd BCThe output of the multi-label fault identification model is LCThe binary label value in (1); selecting the model with the highest precision on the test set as a trained multi-label fault recognition model S (F)C)。
6. The method for non-intrusive real-time fault monitoring of a train bogie as set forth in claim 4, wherein in the step 5, the minimization of the difference between the sum of the vibration signal characteristics of the single vibration sources and the total vibration signal characteristic is taken as one of the optimization objectives; and minimizing the weighted variance of the harmonic order difference values of the vibration signals of the single vibration sources into the second optimization target.
7. The method for non-intrusive real-time fault monitoring of a train bogie as defined in any of claims 1 to 6, further comprising: and updating the multi-label fault identification model and the bogie vibration signal feature library based on the confirmed completely correct fault identification final result.
8. The method for non-intrusive real-time fault monitoring of a train bogie as defined in any of claims 4 to 6, further comprising:
step 7, combining the time stamp signal t and F in the vibration signal feature libraryCObtaining a database T of the characteristic time sequence of the total vibration signalc(t),Tc(t)={LC(t),AC(t),BC(t) }; and predicting the fault state of the bogie at a certain future time by using historical data in the total vibration signal characteristic time sequence database and adopting an extreme learning machine method.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110003825.9A CN112800855B (en) | 2021-01-04 | 2021-01-04 | Non-invasive real-time fault monitoring method for train bogie |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110003825.9A CN112800855B (en) | 2021-01-04 | 2021-01-04 | Non-invasive real-time fault monitoring method for train bogie |
Publications (2)
Publication Number | Publication Date |
---|---|
CN112800855A true CN112800855A (en) | 2021-05-14 |
CN112800855B CN112800855B (en) | 2022-07-12 |
Family
ID=75807914
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110003825.9A Active CN112800855B (en) | 2021-01-04 | 2021-01-04 | Non-invasive real-time fault monitoring method for train bogie |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112800855B (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113358212A (en) * | 2021-06-21 | 2021-09-07 | 重庆理工大学 | Electromechanical fault diagnosis method and system based on relative harmonic order and modeling method |
CN114235446A (en) * | 2021-11-18 | 2022-03-25 | 浙江众合科技股份有限公司 | Online intelligent diagnosis system applied to train bogie |
CN116577716A (en) * | 2023-07-06 | 2023-08-11 | 西安高压电器研究院股份有限公司 | Current sensor vibration characteristic testing method, related equipment and related system |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108257044A (en) * | 2017-09-19 | 2018-07-06 | 济南大学 | A kind of non-intrusion type load decomposition method based on steady-state current model |
WO2019051606A1 (en) * | 2017-09-15 | 2019-03-21 | Tandemlaunch Inc | System and method for classifying passive human-device interactions through ongoing device context awareness |
WO2019153388A1 (en) * | 2018-02-12 | 2019-08-15 | 大连理工大学 | Power spectral entropy random forest-based aeroengine rolling bearing fault diagnosis method |
CN110161343A (en) * | 2019-06-12 | 2019-08-23 | 中南大学 | A kind of non-intrusion type real-time dynamic monitoring method of intelligence train exterior power receiving device |
CN110751385A (en) * | 2019-10-08 | 2020-02-04 | 威胜集团有限公司 | Non-invasive load identification method, terminal device and storage medium |
CN112115643A (en) * | 2020-09-15 | 2020-12-22 | 中南大学 | Smart train service life non-invasive prediction method |
CN112149230A (en) * | 2020-09-27 | 2020-12-29 | 中南大学 | Method for predicting comfort deterioration of wind-induced train of strong wind railway |
-
2021
- 2021-01-04 CN CN202110003825.9A patent/CN112800855B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2019051606A1 (en) * | 2017-09-15 | 2019-03-21 | Tandemlaunch Inc | System and method for classifying passive human-device interactions through ongoing device context awareness |
CN108257044A (en) * | 2017-09-19 | 2018-07-06 | 济南大学 | A kind of non-intrusion type load decomposition method based on steady-state current model |
WO2019153388A1 (en) * | 2018-02-12 | 2019-08-15 | 大连理工大学 | Power spectral entropy random forest-based aeroengine rolling bearing fault diagnosis method |
CN110161343A (en) * | 2019-06-12 | 2019-08-23 | 中南大学 | A kind of non-intrusion type real-time dynamic monitoring method of intelligence train exterior power receiving device |
CN110751385A (en) * | 2019-10-08 | 2020-02-04 | 威胜集团有限公司 | Non-invasive load identification method, terminal device and storage medium |
CN112115643A (en) * | 2020-09-15 | 2020-12-22 | 中南大学 | Smart train service life non-invasive prediction method |
CN112149230A (en) * | 2020-09-27 | 2020-12-29 | 中南大学 | Method for predicting comfort deterioration of wind-induced train of strong wind railway |
Non-Patent Citations (1)
Title |
---|
HUI LIU 等: "An improved non-intrusive load disaggregation algorithm and its application", 《SUSTAINABLE CITIES AND SOCIETY》 * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113358212A (en) * | 2021-06-21 | 2021-09-07 | 重庆理工大学 | Electromechanical fault diagnosis method and system based on relative harmonic order and modeling method |
CN114235446A (en) * | 2021-11-18 | 2022-03-25 | 浙江众合科技股份有限公司 | Online intelligent diagnosis system applied to train bogie |
CN116577716A (en) * | 2023-07-06 | 2023-08-11 | 西安高压电器研究院股份有限公司 | Current sensor vibration characteristic testing method, related equipment and related system |
CN116577716B (en) * | 2023-07-06 | 2023-10-20 | 西安高压电器研究院股份有限公司 | Current sensor vibration characteristic testing method, related equipment and related system |
Also Published As
Publication number | Publication date |
---|---|
CN112800855B (en) | 2022-07-12 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN112800855B (en) | Non-invasive real-time fault monitoring method for train bogie | |
CN103335617B (en) | A kind of railway track geometric deformation detection method based on vibration signal | |
CN102520697B (en) | Onsite information preprocessing method of remote cooperative diagnosis | |
CN111623868B (en) | Convolutional neural network construction method for rail corrugation identification | |
CN102944416B (en) | Multi-sensor signal fusion technology-based fault diagnosis method for wind turbine blades | |
CN110376003A (en) | Intelligent whole train service life prediction technique and its system based on BIM | |
CN105022912A (en) | Rolling bearing fault prediction method based on wavelet principal component analysis | |
CN106803101A (en) | Odometer method for diagnosing faults based on HMM | |
Khosravi et al. | Reducing the positional errors of railway track geometry measurements using alignment methods: A comparative case study | |
CN109374631A (en) | A kind of tunnel state evaluating method | |
CN112507479B (en) | Oil drilling machine health state assessment method based on manifold learning and softmax | |
Liu et al. | Domain adaptation digital twin for rolling element bearing prognostics | |
Hoelzl et al. | On-board monitoring for smart assessment of railway infrastructure: A systematic review | |
CN111504647A (en) | AR-MSET-based performance degradation evaluation method for rolling bearing | |
CN114705432B (en) | Method and system for evaluating health state of explosion-proof motor bearing | |
CN112816052B (en) | Train compartment vibration monitoring method, vibration signal characteristic library establishment and application method | |
CN107977679A (en) | Method based on frequency response function and operation response characteristic diagnosis of complex device initial failure | |
JP7417342B2 (en) | Method for monitoring vibrations in train cabins, construction and application method of vibration signal feature library | |
Yu et al. | A new method for quantitative estimation of rolling bearings under variable working conditions | |
Rahman et al. | Deep learning model for railroad structural health monitoring via distributed acoustic sensing | |
CN109598048A (en) | A kind of lubrication degradation prediction technique of track vehicle door system | |
Zeng et al. | Rail break prediction and cause analysis using imbalanced in-service train data | |
Li et al. | Fault diagnosis of rolling bearing using a refined composite multiscale weighted permutation entropy | |
CN115600695A (en) | Fault diagnosis method of metering equipment | |
CN112069621B (en) | Method for predicting residual service life of rolling bearing based on linear reliability index |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |