CN114386467A - System and method for monitoring rail vehicle chassis fault abnormal sound based on acoustic characteristics - Google Patents
System and method for monitoring rail vehicle chassis fault abnormal sound based on acoustic characteristics Download PDFInfo
- Publication number
- CN114386467A CN114386467A CN202210110678.XA CN202210110678A CN114386467A CN 114386467 A CN114386467 A CN 114386467A CN 202210110678 A CN202210110678 A CN 202210110678A CN 114386467 A CN114386467 A CN 114386467A
- Authority
- CN
- China
- Prior art keywords
- signal
- monitoring
- sound
- time
- rail vehicle
- 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.)
- Pending
Links
- 230000002159 abnormal effect Effects 0.000 title claims abstract description 70
- 238000000034 method Methods 0.000 title claims abstract description 62
- 238000012544 monitoring process Methods 0.000 title claims abstract description 51
- 230000005236 sound signal Effects 0.000 claims abstract description 45
- 238000012545 processing Methods 0.000 claims abstract description 20
- 238000003062 neural network model Methods 0.000 claims abstract description 19
- 238000007405 data analysis Methods 0.000 claims abstract description 13
- 230000008569 process Effects 0.000 claims abstract description 13
- 238000009432 framing Methods 0.000 claims abstract description 12
- 230000003595 spectral effect Effects 0.000 claims description 21
- 238000001228 spectrum Methods 0.000 claims description 18
- 238000000605 extraction Methods 0.000 claims description 12
- 238000013528 artificial neural network Methods 0.000 claims description 7
- 238000012549 training Methods 0.000 claims description 7
- 238000004364 calculation method Methods 0.000 claims description 6
- 230000014759 maintenance of location Effects 0.000 claims description 6
- 210000002569 neuron Anatomy 0.000 claims description 5
- 238000007781 pre-processing Methods 0.000 claims description 5
- 230000002123 temporal effect Effects 0.000 claims description 5
- 239000000284 extract Substances 0.000 claims description 4
- 230000004927 fusion Effects 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 3
- 239000013598 vector Substances 0.000 claims description 3
- 238000005311 autocorrelation function Methods 0.000 claims description 2
- 238000003672 processing method Methods 0.000 claims description 2
- 238000010586 diagram Methods 0.000 description 16
- 230000006872 improvement Effects 0.000 description 13
- 238000001514 detection method Methods 0.000 description 7
- 238000012986 modification Methods 0.000 description 7
- 230000004048 modification Effects 0.000 description 7
- 230000008859 change Effects 0.000 description 4
- 238000012360 testing method Methods 0.000 description 4
- 230000001052 transient effect Effects 0.000 description 4
- 238000011161 development Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 239000000428 dust Substances 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000007500 overflow downdraw method Methods 0.000 description 1
- 238000005096 rolling process Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
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
- G06F2218/02—Preprocessing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Biomedical Technology (AREA)
- Data Mining & Analysis (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Signal Processing (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
Abstract
The invention relates to a rail vehicle chassis fault abnormal sound monitoring system and method based on acoustic characteristics. The system comprises a data signal acquisition module and a data analysis processing module; the data signal acquisition module is used for acquiring sound signals emitted in the running process of the vehicle; and the data analysis processing module is used for receiving the collected sound signals, converting the collected sound signals into digital signals, performing frame interception, extracting signal characteristics, and inputting the signal characteristics into the trained deep neural network model for classification and identification. The method comprises the steps that a sound signal sent out when a vehicle runs is picked up through a vehicle bottom data signal acquisition module; the collected sound signals are converted into digital signals through a data analysis processing module, short-time signal samples are obtained through framing and intercepting, signal sample characteristics are extracted and input into a neural network model for classification and identification, identification results are fused, and real-time monitoring of subway underframe equipment is achieved. The invention effectively improves the identification rate of obstacle monitoring.
Description
Technical Field
The invention relates to the field of rail transit fault monitoring, in particular to a system and a method for monitoring and identifying fault abnormal sound generated by subway vehicle chassis equipment, and particularly relates to a system and a method for monitoring the fault abnormal sound of a rail vehicle chassis based on acoustic characteristics.
Background
By the end of 2020, 244 rail transportation operation lines are opened in 45 cities in China, and the total length of the operation lines reaches 7969.7 km. Wherein the subway operation line reaches 6280.8km, accounting for 78.8%. Along with the rapid expansion of urban construction scale and population flow quantity, the operation load of subway vehicles is continuously increased, and higher requirements are also put forward for the fault detection of subway underframe equipment. The railway vehicle bogie mainly comprises a framework, wheel sets, a basic braking system, a central traction unit and other auxiliary devices, plays important roles of bearing, traction, running, braking and the like, and the health condition of each device directly influences the running safety of a train. The faults of the existing underframe equipment mainly comprise tread defects, equipment looseness of a bogie, faults of a motor and a gear box, faults of a bearing and the like. In the normal running process of the railway vehicle, the generated sound signal is relatively stable. When the vehicle has equipment failure, factors such as sound, audio frequency and the like in the process of traveling change, and the abnormal sound signal is generated. Therefore, the development of the fault abnormal sound monitoring research of the vehicle chassis equipment has important value for ensuring the normal operation of the rail vehicle and has great significance for the development of the fault diagnosis method of the mechanical equipment.
The conventional rail vehicle fault detection method is usually offline detection, which means that in a vehicle running gap, a worker diagnoses equipment through patrol or by using a professional detection instrument. The purpose of identifying faults can be achieved by off-line detection, but the requirement of real-time performance is not met. With the development of science and technology, engineers and technicians invented various rail vehicle fault monitoring methods, which are roughly divided into the following: 1. a vibration signal monitoring method; 2. temperature signal monitoring; 3. image information monitoring method; 4. and (4) a sound signal monitoring method. A train gearbox fault monitoring method based on vibration signals is introduced in the document Yuan-Qing, Li Qian, Yangliang and the like high-speed train gearbox line test and abnormal vibration analysis [ J ] railway rolling stock, 2016,36(1):24-29 ]. The document 'temperature distribution research of high-speed train fault axle box bearings' (Thomu, WangYijie, Chenguangdong, and the like.) temperature distribution research of high-speed train fault axle box bearings [ J ]. railway academy, 2016,38(7):50-56.) monitors potential faults of train axle boxes by using temperature information, and judges whether faults exist according to temperature changes of equipment under different working states. However, the temperature change is often in the later stage of equipment failure, and the temperature change standards of various kinds of equipment are not uniform, so that early failure early warning cannot be achieved. In the document "application of a 360-degree appearance fault image detection system for a subway train" (Luchi wave, Liguiqi, Gupeng, and the like. application of a 360-degree appearance fault image detection system for a subway train [ J ]. the transportation manager world, 2020(3):46-48.) images of train equipment are acquired by mounting a camera, and fault monitoring is performed according to a machine vision technology. However, the monitoring method based on the image is greatly influenced by the ambient illumination condition, and is easily shielded by dust, foreign matters and the like, the image information cannot monitor the internal fault of the equipment, and the data on the surface of the equipment only can not reach the target of global monitoring, so that erroneous judgment and missing judgment are easily caused.
The fault monitoring method based on the sound signals is that the sound signals sent out in the running process of the rail vehicle are collected through an acoustic sensor, and faults are early warned in time according to the change of environmental sounds. Compared with a vibration signal, the sound is collected by using a non-contact sensor, and all surrounding equipment data can be acquired by only a small number of microphones, so that the complexity and the cost of the system are reduced, and the limitation problem of other monitoring methods is solved. By utilizing the characteristic that abnormal sound can be generated under the condition of chassis equipment failure, the subway chassis equipment failure abnormal sound monitoring method based on acoustic feature fusion is provided, a Linear Prediction Coefficient (LPC) and an improved time spectrum kurtosis feature (MTSK) are extracted and fused by collecting sound signals emitted by a subway vehicle, a neural network model is built to monitor the equipment failure abnormal sound in real time, and the method has a good application prospect.
Disclosure of Invention
The invention aims to overcome the problems that a large number of sensors need to be installed on operating equipment in a contact mode, early fault early warning cannot be carried out, and the influence of the environment is large, so that the system and the method for monitoring the abnormal noise of the chassis fault of the railway vehicle based on the acoustic characteristics are provided.
In order to solve the technical problems, the technical scheme of the invention provides a rail vehicle chassis fault abnormal sound monitoring system based on acoustic characteristics; the system comprises a data signal acquisition module and a data analysis processing module:
the data signal acquisition module is used for acquiring sound signals emitted in the running process of the rail vehicle;
the data analysis processing module is used for receiving the sound signals collected by the data signal collection module, converting the received sound signals into digital signals for preprocessing, extracting signal characteristics, and inputting the signal characteristics into a trained deep neural network model for abnormal sound recognition;
the system firstly collects sound signals emitted in the running process of a rail vehicle through a data signal collecting module carried on a microphone at the bottom of a carriage, then transmits the sound signals to a data analyzing and processing module, the data analyzing and processing module extracts abnormal sound signal characteristics through a modern signal processing method, then inputs the extracted signal characteristics into a neural network model for classification and identification, and fuses identification results, wherein the identification results are the probability for judging the abnormal sound of a fault, and therefore monitoring of the abnormal sound of equipment is achieved.
The invention provides a rail vehicle chassis fault abnormal sound monitoring method based on acoustic characteristics, which comprises the following steps:
(1) microphones are distributed near a bogie of the rail vehicle and in the vehicle, and a data signal acquisition module is carried to pick up a sound signal emitted by the rail vehicle in the advancing process and transmit the sound signal to a data analysis processing module;
(2) the data analysis processing module converts the acquired sound signals into digital signals for preprocessing, frames and intercepts continuous sound signals into short-time signal samples, extracts the characteristics of the signal samples, inputs the signal characteristics extracted in real time into a neural network model for classification and identification, outputs identification results and finally fuses the identification results, so that the real-time monitoring of the abnormal sound of the vehicle chassis equipment is realized;
the neural network model is trained, and the training method comprises the following steps: and extracting characteristic vectors of various signal samples from the data set containing background noise samples and fault abnormal sound samples of the normal running of the rail vehicle, and training the deep neural network model.
As an improvement of the above technical solution, when performing feature extraction on a signal sample, the method includes LPC feature extraction and MTSK feature extraction.
As another improvement of the above technical solution, when performing LPC feature extraction, the method obtains a Yule-Walker equation by calculating an autocorrelation matrix after performing windowing and framing processing on an original signal, as shown in the following formula:
wherein p is the linear prediction order, aiIs a linear prediction coefficient, R is an autocorrelation function;
and solving the LPC coefficient by using a Durbin recursion algorithm.
As another improvement of the above technical solution, when performing MTSK feature extraction, the method calculates a temporal spectral kurtosis of a signal sample; the formula for the temporal spectral kurtosis is as follows:
wherein,
for the original signalThe time spectral kurtosis of a frame after windowing and framing, t is the signal time, f is the frequency of the sound signal,
for k-order spectral distance time dimension information after windowing and framing an original signal, the formula is as follows:
wherein the signal Xn(t, f) is the short-time Fourier transform (STFT) of a frame after windowing and framing the original signal,<·>trepresents a short-time average operator at time t, the length of time depending on the window length;
and when the second-order spectral distance is calculated, eliminating the part with the highest amplitude in each sub-band, and taking a retention coefficient alpha, wherein the retention coefficient alpha is the proportion of the rest part in one sub-band.
For a sample containing abnormal sound, the amplitude of the transient impact part is far larger than that of a surrounding signal, and when transient components are eliminated, the second spectral distance is obviously reduced. For the stable background noise, the amplitude difference in the frequency domain is small, and the variation of the second-order spectral distance is also small. According to the different sensibility of the background noise and the abnormal sound impact on the characteristics, the improved method improves the anti-noise performance of characteristic parameters and is beneficial to finding abnormal impact signals under the strong background noise. Meanwhile, the characteristics of the seam impact and the abnormal sound impact are distributed on different frequency domains, seam signals are mainly concentrated on low frequency, abnormal sound signals are concentrated on high frequency, and the frequency band distribution range is wide. According to the characteristic, the recognition efficiency of the abnormal sound signal characteristic is greatly improved.
As a further improvement of the above technical solution, the method constructs a fully-connected neural network having three hidden layers, each hidden layer contains 64 neurons, and ReLU is used as an activation function.
As a further improvement of the above technical solution, the ReLU is defined as: ReLU ═ max (0, x), where x is the neuron calculation result, and the parameters of the other neural networks are set as: the value of Batch _ size is 32, the value of Epoch is 200, and the cross entropy loss function is used as the loss function.
As a further improvement of the above technical solution, when the method fuses the recognition results, a D-S evidence theory algorithm is used to perform decision-level fusion on the recognition results of a plurality of samples within the period of time in which abnormal sound occurs, thereby improving the accuracy of recognition.
The rail vehicle chassis fault abnormal sound monitoring system and method based on the acoustic characteristics have the advantages that:
(1) the fault monitoring is carried out by using an acoustic means, so that the limitation problem of the traditional monitoring method can be solved, and the real-time monitoring of equipment faults can be realized.
(2) Because the background noise energy generated in the subway operation process is large, abnormal sound is easily submerged in strong background noise. In order to improve the accuracy of abnormal sound monitoring, a new characteristic parameter, namely improved time spectrum kurtosis, is provided. The method for extracting the spectral kurtosis is improved, the characteristic parameters of abnormal sound components can be obviously improved, the characteristic parameters of stable noise cannot be changed too much, and the characteristic parameters are combined with LPC (linear predictive coding) characteristics and are jointly used as the input parameters of a classifier, so that the abnormal sound monitoring performance under the condition of low signal-to-noise ratio is improved.
(3) The neural network model shows good performance in the aspects of processing the problems of nonlinearity, large data volume and the like. In order to improve the accuracy and generalization capability of the monitoring model, a fully-connected neural network model containing three hidden layers is constructed. Compared with the traditional classification model, the neural network has higher accuracy.
Drawings
FIG. 1 is a flow chart of a rail vehicle chassis fault abnormal sound monitoring method based on acoustic characteristics according to the invention;
fig. 2 is waveform diagrams of three types of samples collected by the data signal collection module according to the present invention, wherein fig. 2(a) is a waveform diagram of a rail joint, fig. 2(b) is a waveform diagram of a background noise, and fig. 2(c) is a waveform diagram of a fault abnormal sound;
fig. 3 is a time spectrum kurtosis diagram of three types of sound signals of a track seam, background noise and abnormal noise before and after a kurtosis calculation is improved, wherein fig. 3(a) is a time spectrum kurtosis diagram of the track seam before the improvement, fig. 3(b) is a time spectrum kurtosis diagram of the track seam after the improvement, fig. 3(c) is a time spectrum kurtosis diagram of the background noise before the improvement, fig. 3(d) is a time spectrum kurtosis diagram of the background noise after the improvement, fig. 3(e) is a time spectrum kurtosis diagram of the abnormal noise before the improvement, and fig. 3(f) is a time spectrum kurtosis diagram of the abnormal noise after the improvement;
FIG. 4 is a flowchart illustrating the steps of performing the LPC feature extraction in accordance with the present invention;
fig. 5 is a schematic structural diagram of the neural network model constructed by the present invention.
Detailed Description
The technical solution provided by the present invention is further described below with reference to the following embodiments and the accompanying drawings.
1. The time length of the fault abnormal sound is about 60-150ms, the background noise is segmented according to the time length of 300ms, then the fault abnormal sound is randomly superimposed into the background noise in the time domain, and normalization processing is uniformly performed to form a data set sample, the embodiment shows time domain waveforms of three types of samples shown in fig. 2, wherein fig. 2(a) is a track seam waveform diagram, fig. 2(b) is a background noise waveform diagram, and fig. 2(c) is a fault abnormal sound waveform diagram.
And further, extracting LPC and MTSK features of the three types of signal samples, and training a deep neural network model by using the extracted feature vectors.
As shown in fig. 1, it is a flowchart of a method for monitoring abnormal noise of a rail vehicle chassis fault based on acoustic characteristics according to the present invention; the method comprises the following steps:
(1) microphones are distributed near a bogie of the rail vehicle and in the vehicle, and a data signal acquisition module is carried to pick up a sound signal emitted by the rail vehicle in the advancing process and transmit the sound signal to a data analysis processing module;
(2) the data analysis processing module converts the collected sound signals into digital signals for data preprocessing: framing and intercepting continuous sound signals into short-time signal samples; and then, carrying out LPC and MTSK feature extraction on the signal sample, inputting the signal features extracted in real time into a trained neural network model for classification and identification, outputting an identification result, finally carrying out decision-level fusion on the identification result by using a D-S evidence theory algorithm, and outputting the result to realize real-time monitoring on the abnormal sound of the vehicle chassis equipment.
2. Improved temporal spectral kurtosis calculation
Kurtosis is a statistic reflecting the distribution characteristics of signals, and the calculation formula is as follows:
in the formula,
is the mean value of the signal, and the spectral kurtosis characteristic of the signal can be calculated from the frequency domain by a short-time Fourier transform (STFT) method. Let signal X (t, f) be STFT for X (t), w (n) be an analysis window:
define the empirical spectral distance of order k for X (t, f) as:
SkX(f)＝<|X(t,f)k|>
where < > represents the time averaging operator, t represents the signal time, and f represents the signal frequency. By STFT transformation, the spectral kurtosis can be defined as the normalized fourth-order empirical spectral distance:
windowing and framing the original signal, providing information of k-order spectral distance time dimension:
wherein<·>tRepresenting a short time period at time tAveraging operator, the length of time depends on the window length. Finally, the formula for the temporal spectral kurtosis is as follows:
the improvement method comprises the following steps: when calculating the second-order spectral distance, the part with higher amplitude in the sub-band is removed, and a retention coefficient alpha is taken, wherein the retention coefficient alpha is the proportion of the rest part in the sub-band. For a sample containing abnormal sound, the amplitude of the transient impact part is far larger than that of a surrounding signal, and when transient components are eliminated, the second spectral distance is obviously reduced. For the stable background noise, the amplitude difference in the frequency domain is small, and the variation of the second-order spectral distance is also small. According to the different sensibility of the background noise and the abnormal sound impact on the characteristics, the improved method improves the anti-noise performance of characteristic parameters and is beneficial to finding abnormal impact signals under the strong background noise. Meanwhile, the characteristics of the seam impact and the abnormal sound impact are distributed on different frequency domains, seam signals are mainly concentrated on low frequency, abnormal sound signals are concentrated on high frequency, and the frequency band distribution range is wide. According to the characteristic, the purpose of abnormal sound monitoring can be achieved.
As shown in fig. 3, a time spectrum kurtosis map of three types of sound signals of a track seam, a background noise and a fault abnormal sound before and after a modification is calculated for the kurtosis, wherein fig. 3(a) is a time spectrum kurtosis map of the track seam before the modification, fig. 3(b) is a time spectrum kurtosis map of the track seam after the modification, fig. 3(c) is a time spectrum kurtosis map of the background noise before the modification, fig. 3(d) is a time spectrum kurtosis map of the background noise after the modification, fig. 3(e) is a time spectrum kurtosis map of the fault abnormal sound before the modification, and fig. 3(f) is a time spectrum kurtosis map of the fault abnormal sound after the modification.
3. Calculation of LPC
As shown in fig. 4, a flowchart of the steps of the LPC feature extraction according to the present invention includes the following steps: windowing and framing the received original signal, calculating an autocorrelation matrix to obtain a Yule-Walker equation, and solving the LPC coefficient by using a Durbin recursion algorithm.
There is a correlation between the sound signals, so the current value can be approximated by a weighted linear combination of the past values. The error between the original sound values and the linear prediction values is minimized to determine a set of LPC parameters. The sound signal at time n is x (n), and its linear prediction value and prediction error can be expressed as follows:
from the mean square error criterion, the Yule-Walker equation can be obtained as follows:
solution a of the equationiNamely LPC coefficient, can be obtained by Durbin recursion algorithm.
In consideration of the fact that the operation environment of the subway is complex, abnormal sound caused by faults and sound signals can be transformed in the normal running process, the LPC (Linear predictive coding) features are extracted, and the abnormal sound caused by faults is identified together by combining MTSK (maximum Transmission Shift keying).
4. Model training and testing
Fig. 5 is a schematic structural diagram of the neural network model constructed according to the present invention.
We constructed a fully-connected neural network with three hidden layers, each containing 64 neurons, using ReLU as the activation function, which is defined as:
ReLU＝max(0,x)
the parameters of the input neural network are set as follows: and the Batch _ size is 32, the Epoch is 200, the loss function uses a cross entropy loss function, the received extracted signal characteristics are compared and identified when the equipment runs, and the identification result is output. Through training and testing, the identification accuracy reaches 100% at 20km/h and 98.2% and 88.0% at 40km/h and 60km/h respectively according to the test result of the model at 20 km/h. The result shows that the method can effectively identify the abnormal fault sound. Through comparison of results of different characteristics, the acoustic characteristic fusion method is proved to be capable of effectively improving the recognition rate of monitoring, and particularly, the recognition result of abnormal sound is greatly improved through MTSK characteristics.
As can be seen from the above detailed description of the invention, the system and the method for monitoring the abnormal noise of the chassis of the railway vehicle based on the acoustic features provided by the invention meet the requirement that a small number of non-contact sensors are mounted on running equipment, and realize real-time monitoring of the occurrence of the fault through acoustic feature recognition under the condition that the monitoring equipment is greatly influenced by the environment, thereby effectively improving the monitoring recognition rate.
Finally, it should be noted that the above embodiments are only used for illustrating the technical solutions of the present invention and are not limited. Although the present invention has been described in detail with reference to the embodiments, it will be understood by those skilled in the art that various changes may be made and equivalents may be substituted without departing from the spirit and scope of the invention as defined in the appended claims.
Claims (8)
1. A rail vehicle chassis fault abnormal sound monitoring system based on acoustic characteristics comprises a data information acquisition module and a data analysis processing module,
the data signal acquisition module is used for acquiring sound signals emitted in the running process of the rail vehicle;
the data analysis processing module is used for receiving the sound signals collected by the data signal collection module, converting the received sound signals into digital signals for preprocessing, extracting signal characteristics, and inputting the signal characteristics into a trained deep neural network model for abnormal sound recognition;
the system firstly collects sound signals emitted in the running process of a rail vehicle through a data signal collecting module carried on a microphone at the bottom of a carriage, then transmits the sound signals to a data analyzing and processing module, the data analyzing and processing module extracts abnormal sound signal characteristics through a modern signal processing method, then inputs the extracted signal characteristics into a neural network model for classification and identification, and fuses identification results, wherein the identification results are the probability for judging the abnormal sound of a fault, and therefore monitoring of the abnormal sound of equipment is achieved.
2. A rail vehicle chassis fault abnormal sound monitoring method based on acoustic characteristics comprises the following steps:
(1) microphones are distributed near a bogie of the rail vehicle and in the vehicle, and a data signal acquisition module is carried to pick up a sound signal emitted by the rail vehicle in the advancing process and transmit the sound signal to a data analysis processing module;
(2) the data analysis processing module converts the acquired sound signals into digital signals for preprocessing, frames and intercepts continuous sound signals into short-time signal samples, extracts the characteristics of the signal samples, inputs the signal characteristics extracted in real time into a neural network model for classification and identification, outputs identification results and finally fuses the identification results, so that the real-time monitoring of the abnormal sound of the vehicle chassis equipment is realized;
the neural network model is trained, and the training method comprises the following steps: and extracting characteristic vectors of various signal samples from the data set containing background noise samples and fault abnormal sound samples of the normal running of the rail vehicle, and training the deep neural network model.
3. The method for monitoring the abnormal noise of the chassis of the railway vehicle based on the acoustic features as claimed in claim 2, wherein the method comprises LPC feature extraction and MTSK feature extraction when the signal samples are subjected to feature extraction.
4. The method for monitoring the abnormal noise of the chassis of the rail vehicle based on the acoustic features as claimed in claim 3, wherein when the LPC feature extraction is carried out, after windowing and framing are carried out on an original signal, a Yule-Walker equation is obtained by calculating an autocorrelation matrix, and the equation is shown as the following formula:
wherein p is the linear prediction order, aiIs a linear prediction coefficient, R is an autocorrelation function;
and solving the LPC coefficient by using a Durbin recursion algorithm.
5. The method for monitoring the abnormal noise of the chassis of the railway vehicle based on the acoustic features as claimed in claim 3, wherein when the method is used for extracting the MTSK features, the time spectrum kurtosis of the signal samples is calculated; the formula for the temporal spectral kurtosis is as follows:
wherein,
in order to obtain the time spectrum kurtosis of a certain frame after windowing and framing the original signal, t is the signal time, f is the frequency of the sound signal,
for k-order spectral distance time dimension information after windowing and framing an original signal, the formula is as follows:
wherein the signal Xn(t, f) is the Fourier transform of a frame after windowing and framing the original signal,<·>trepresents a short-time average operator at time t, the length of time depending on the window length;
when calculating the second-order spectral distance, sorting all frame data in each sub-band according to the size, eliminating the part with the highest amplitude in each sub-band, and taking a retention coefficient alpha, wherein the retention coefficient alpha is the proportion occupied by the rest part in one sub-band.
6. The method for rail vehicle chassis fault and abnormal sound monitoring based on the acoustic features as claimed in claim 2, wherein the method constructs a fully-connected neural network with three hidden layers, each hidden layer comprises 64 neurons, and ReLU is used as an activation function.
7. The method of claim 6, wherein the ReLU is defined as: ReLU ═ max (0, x), where x is the neuron calculation result, and the parameters of the other neural networks are set as: the value of Batch _ size is 32, the value of Epoch is 200, and the cross entropy loss function is used as the loss function.
8. The method for monitoring the abnormal noise of the chassis of the railway vehicle based on the acoustic features as claimed in claim 2, wherein when the method is used for fusing the recognition results, a D-S evidence theory algorithm is used for carrying out decision-level fusion on the recognition results of a plurality of samples in the period of time during which the abnormal noise occurs.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210110678.XA CN114386467A (en) | 2022-01-29 | 2022-01-29 | System and method for monitoring rail vehicle chassis fault abnormal sound based on acoustic characteristics |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210110678.XA CN114386467A (en) | 2022-01-29 | 2022-01-29 | System and method for monitoring rail vehicle chassis fault abnormal sound based on acoustic characteristics |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114386467A true CN114386467A (en) | 2022-04-22 |
Family
ID=81204485
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210110678.XA Pending CN114386467A (en) | 2022-01-29 | 2022-01-29 | System and method for monitoring rail vehicle chassis fault abnormal sound based on acoustic characteristics |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114386467A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115290133A (en) * | 2022-06-30 | 2022-11-04 | 苏州经贸职业技术学院 | Method and system for monitoring track structure at joint of light rail platform |
| CN116204821A (en) * | 2023-04-27 | 2023-06-02 | 昆明轨道交通四号线土建项目建设管理有限公司 | Vibration evaluation method and system for rail transit vehicle |
-
2022
- 2022-01-29 CN CN202210110678.XA patent/CN114386467A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115290133A (en) * | 2022-06-30 | 2022-11-04 | 苏州经贸职业技术学院 | Method and system for monitoring track structure at joint of light rail platform |
| CN116204821A (en) * | 2023-04-27 | 2023-06-02 | 昆明轨道交通四号线土建项目建设管理有限公司 | Vibration evaluation method and system for rail transit vehicle |
| CN116204821B (en) * | 2023-04-27 | 2023-08-11 | 昆明轨道交通四号线土建项目建设管理有限公司 | Vibration evaluation method and system for rail transit vehicle |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114386467A (en) | System and method for monitoring rail vehicle chassis fault abnormal sound based on acoustic characteristics | |
| CN104236911B (en) | A kind of train bogie bearing military service process monitoring and fault diagnosis system and method | |
| CN111351664B (en) | Bearing temperature prediction and alarm diagnosis method based on LSTM model | |
| Ye et al. | Shock detection of rotating machinery based on activated time-domain images and deep learning: An application to railway wheel flat detection | |
| CN105258785A (en) | Acoustic monitoring station, wheel detection method and train bearing fault passive acoustic diagnosis system | |
| CN110789566B (en) | Track defect monitoring method and monitoring equipment based on axle box acceleration signal | |
| CN113486874B (en) | Rail corrugation feature identification method based on wheel-rail noise wavelet packet decomposition | |
| CN112082782B (en) | Urban rail transit train running gear fault digital diagnosis system and diagnosis method thereof | |
| CN107423692A (en) | A kind of rail corrugation fault detection method based on wavelet-packet energy entropy | |
| Cheng et al. | Slow feature analysis-aided detection and diagnosis of incipient faults for running gear systems of high-speed trains | |
| CN113988326A (en) | Subway equipment maintenance optimization method and system | |
| CN107782548B (en) | Rail vehicle part detection system | |
| CN111055881A (en) | Wheel-rail interface damage evolution monitoring method based on noise signals | |
| Shi et al. | Wayside acoustic fault diagnosis of train bearing based on improved linear discriminant analysis | |
| Yan et al. | Abnormal noise monitoring of subway vehicles based on combined acoustic features | |
| CN113447270B (en) | Acoustic monitoring system and method for traction motor bearing of railway vehicle | |
| CN116467570B (en) | Heavy-duty locomotive coupler swing angle quantitative identification method based on data driving | |
| CN115356133B (en) | Bogie operation condition online monitoring, real-time classifying and identifying algorithm | |
| CN111089738B (en) | Wheel-rail matching state identification method based on noise signals | |
| CN115062867A (en) | Bearing fault prediction method based on Teager energy spectrum and improved CNN model | |
| Wang et al. | Wavelet Integrated CNN With Dynamic Frequency Aggregation for High-Speed Train Wheel Wear Prediction | |
| CN113672859A (en) | Switch point machine fault acoustic diagnosis system | |
| CN109211556B (en) | Rail vehicle spare part detecting system | |
| CN113343887A (en) | Multi-sensor mixed fault signal blind separation method based on edge calculation and machine learning | |
| Shen et al. | Research on Vehicle Identification Algorithm Based on Sound Signal |
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 |