CN115200695A - Abnormal sound wave characteristic identification-based safety early warning method and system for truck band-type brake - Google Patents
Abnormal sound wave characteristic identification-based safety early warning method and system for truck band-type brake Download PDFInfo
- Publication number
- CN115200695A CN115200695A CN202210826826.8A CN202210826826A CN115200695A CN 115200695 A CN115200695 A CN 115200695A CN 202210826826 A CN202210826826 A CN 202210826826A CN 115200695 A CN115200695 A CN 115200695A
- Authority
- CN
- China
- Prior art keywords
- sound wave
- data
- abnormal
- time
- band
- 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
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01H—MEASUREMENT OF MECHANICAL VIBRATIONS OR ULTRASONIC, SONIC OR INFRASONIC WAVES
- G01H11/00—Measuring mechanical vibrations or ultrasonic, sonic or infrasonic waves by detecting changes in electric or magnetic properties
- G01H11/06—Measuring mechanical vibrations or ultrasonic, sonic or infrasonic waves by detecting changes in electric or magnetic properties by electric means
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/24—Querying
- G06F16/248—Presentation of query results
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/25—Integrating or interfacing systems involving database management systems
Abstract
The invention provides a safety early warning method and system for a truck band-type brake based on abnormal sound wave characteristic identification. On one hand, sound wave spectrum processing is carried out on the data collected in real time based on sound wave signal characteristic analysis, an improved template matching algorithm is adopted, an abnormal band-type brake vehicle sound wave recognition algorithm is compiled by combining SVM machine learning design, and whether the sound wave data received in real time has the characteristics of band-type brake sound waves or not is judged. On the other hand, the analysis, summarization and query of relevant contracting brake information such as vehicle real-time running sound wave visualization, sound wave characteristic real-time extraction, abnormal alarm prompting, statistical report output and the like are realized by comprehensively utilizing a JAVA development platform, an SQLServer background database and MATLAB modular programming, and the method has important significance for guaranteeing the safety of goods train running and marshalling operation. The scheme of the invention is beneficial to effectively discovering the potential safety hazard of the band-type brake of the railway freight vehicle in the transportation process in time, and avoiding the safety problems of steel rail abrasion, vehicle overturn and the like.
Description
Technical Field
The invention belongs to the field of traffic engineering, and relates to a safety early warning method and system for a truck band-type brake based on abnormal sound wave characteristic identification.
Background
In the daily transportation production operation process of a marshalling station, dangerous conditions such as 'dragging' and 'jumping' of a train on a track can be caused by abnormal brake locking of the truck, so that the tread of the wheel is seriously abraded, strained and peeled, and the damage to the driving safety is large. For hump disassembly, when an abnormal band-type brake vehicle passes through a curve or special line sections such as turnout frog and switch rail in the sliding process, the vehicle is easy to get out of line or even overturn. Therefore, the automatic identification and early warning of the band-type brake vehicle are very important.
With the great development of the crossing type of the railway and the continuous improvement of the informatization degree, various computer information systems are widely applied to various transport stations. The prior automatic identification method for the railway wagon contracting brake vehicle mainly comprises two methods, namely thermal imaging identification and a delay relay alarm circuit, wherein the former method is a ground auxiliary system, and the latter method is a vehicle-mounted auxiliary system. The thermal imaging equipment can only focus on a local area for detection, and the structures of train bodies corresponding to different freight train types are different, so that the method is difficult to meet the band-type brake detection requirements of different train types; the relay delay circuit alarming method needs to transform the goods vehicle in the aspects of mechanical structure and electronic circuit, has huge cost as a vehicle-mounted system, has large randomness of truck marshalling, and is difficult to unify related standards.
Disclosure of Invention
Aiming at various defects of the prior art, the invention provides a wagon brake safety early warning method and system based on abnormal sound wave characteristic identification, aiming at preventing safety accidents such as vehicle side collision, derailment and the like.
In order to achieve the purpose, the technical scheme of the invention is as follows:
a safety early warning method for a truck band-type brake based on abnormal sound wave feature recognition comprises the following steps:
s1: trackside historical sound wave data acquisition and analysis
The method comprises the steps that rail-side historical sound wave collection is achieved through a wagon brake sound wave collection hardware system, numbering is conducted according to date and time, and sound wave files are transmitted and stored to a server; the method specifically comprises the following steps:
s1-1: historical data collection
After the acoustic wave data acquisition position is selected, a field acoustic wave data acquisition system is arranged, abnormal band-type brake acoustic wave data of trucks of various types are acquired, and vehicle operation background acoustic wave data are acquired;
s1-2: fourier transform-based acoustic feature extraction
Obtaining the frequency distribution characteristics of various types of sound wave signals by adopting fast Fourier transform;
taking an energy matrix obtained after FFT as an initial sample, wherein the energy matrix is in a row frequency axis and a column time axis, and each moment corresponds to a sound sample of a column vector, namely the frequency distribution of each moment corresponding to one sound sample;
performing linear change on the original sound wave spectrum energy data by adopting a Min-Max normalization method;
the frequency distribution after Fourier transform is 0-6000 Hz, the frequency distribution is equally divided into 512 frequency sections, and the initial data sample is divided into 60 frequency sections according to the frequency interval of 100 Hz;
after Fourier transformation, 15-17 sound samples exist every second, namely band-type brake abnormal data every second can provide frequency vector data at 16 moments on average;
counting characteristic parameters contained in sound vector data corresponding to each moment, extracting 6 peak values and energy in corresponding 6 peak value frequency ranges, and totaling 14 parameters;
the obtained characteristic matrix is used as an input sample matrix of a data matching algorithm after being transformed;
s1-3: acoustic data sample analysis and template storage
Numbering the sound wave types according to the types of the railway freight vehicles and the types of the background sound waves; combining all sound wave data samples, and storing the sample mean value of the same type of sound waves as template data in a local database;
s1-4: designing an identification classifier, and generating the classifier and corresponding structural data by adopting machine learning training on the basis of the acquired trackside historical data samples;
s2, real-time running sound wave data processing of the freight train comprises the following steps:
s2-1: real-time acoustic data acquisition
Acquiring real-time abnormal contracting brake sound wave data of the truck in the same way as the S1-1, storing the acquired sound wave data in an SQL (structured query language) server database according to a certain code rate, and synchronously and visually displaying the current real-time acquired sound wave characteristics according to 14 sound wave characteristic parameters;
s2-2: real-time acoustic data processing
Obtaining the frequency spectrum energy distribution of the original sound wave by adopting Fourier transform in S1-2 based on the input real-time vehicle running sound wave data;
s3: the acoustic wave identification based on feature matching comprises the following steps:
s3-1: dynamic buckling template matching
Extracting an approximate template according to the similarity distance calculation and the similarity calculation;
s3-2: vector machine identification
2) Classifier activation and outcome determination
(1) Extracting 3 closest sound wave types of the real-time sound wave samples according to the template matching result based on the dynamic bending, wherein the sound wave types are assumed to be a, b and c;
(2) if any value from 1 to 5 does not appear in a \ b \ c, the current sound wave data cannot be matched with abnormal brake data of various trucks, and the early warning result is output as a normal type;
(3) if at least one value from 1 to 5 exists in a \ b \ c, the corresponding classifier needs to be activated, including: an a-b classifier, an a-c classifier and a b-c classifier;
(4) judging a classification result, and loading the actually measured data to output the classification result by combining the classifier determined in the step (3);
s4: early warning and recording
S4-1: early warning of truck band-type brake
If the current real-time sound wave type is judged to be abnormal band-type sound waves after the classifier is activated in S3, alarm information is output and voice broadcasting prompt is carried out;
s4-2: vehicle safety record
And generating a safety report according to the monitored abnormal brake condition of the truck, and generating a related report according to related indexes such as occurrence frequency and frequency according to different statistical time periods.
Further, in step S1-1, the background sound wave includes: the sound waves of vibration of the passing rail gap when the vehicle normally operates on the line, the sound waves of operation of each model of an adjacent line train, the sound waves of whistling and rumbling of a locomotive, and the sound waves of operation of the locomotive in and out of a warehouse.
Further, in the step S1-2, 200Hz is selected as the minimum frequency peak interval.
Further, in step S1-4, the classifiers are 20 groups, and the combination and assignment of the acoustic wave types involved in the classifiers are as follows:
| classifier | Acoustic wave combination | Output value | Sorting device | Acoustic wave combination | Output value |
| 1 | 1 and 4 | 1、-1 | 11 | 4 and 8 | 1、-1 |
| 2 | 2 and 5 | 1、-1 | 12 | 2 and 9 | 1、-1 |
| 3 | 1 and 6 | 1、-1 | 13 | 3 and 9 | 1、-1 |
| 4 | 1 and 11 | 1、-1 | 14 | 3 and 10 | 1、-1 |
| 5 | 4 and 6 | 1、-1 | 15 | 4 and 10 | 1、-1 |
| 6 | 4 and 11 | 1、-1 | 16 | 4 and 11 | 1、-1 |
| 7 | 3 and 6 | 1、-1 | 17 | 1 and 10 | 1、-1 |
| 8 | 2 and 7 | 1、-1 | 18 | 1 and 11 | 1、-1 |
| 9 | 3 and 7 | 1、-1 | 19 | 2 and 6 | 1、-1 |
| 10 | 1 and 8 | 1、-1 | 20 | 5 and 10 | 1、-1 |
And each group of classifiers is trained by adopting a characteristic matrix corresponding to the type of the sound wave to generate corresponding classifiers and classification rules.
Further, in the step 2-1, the current real-time acquired sound wave characteristics are synchronously visualized and displayed.
Further, the step S3-1 specifically includes the following steps:
4) Similarity distance calculation
Extracting sound wave fragments with m time length and 11 types of sound wave templates in a database to calculate the similar distance, wherein the calculation formula is as shown in formula (1):
where δ is the similarity distance between the real-time acoustic segment and the template, t i And f i Respectively representing the time of a sample point in the template and the corresponding acoustic amplitude parameter, t j And f j Respectively correspond to the real-time sound wave segmentsTime and amplitude parameters of (1);
5) Similarity calculation
A calculation formula (2) of the similarity is constructed:
In the formula d sk Similarity of the sound wave segment s and the template k; delta sk The similarity distance between the real-time sound wave segment s and the template k is calculated according to the formula (1); s k A similar distance threshold value of the kth type sound wave sample;
6) Approximate template extraction
And according to the similarity calculation result, selecting the sound wave types corresponding to the 3 highest similarity values from the 11 similarities, and extracting corresponding template sound wave data and sound wave characteristic data to serve as the basis of subsequent identification.
Further, the decision rule in the classifier classification result decision is as follows:
The invention also provides a freight car band-type brake safety early warning system based on abnormal sound wave characteristic identification, which is used for realizing the freight car band-type brake safety early warning method based on the abnormal sound wave characteristic identification and comprises the following steps: the system comprises a field sound wave acquisition module, a data processing and analyzing module and a foreground information publishing module; the field sound wave acquisition module is used for acquiring pre-investigated trackside historical sound wave data and acquiring field real-time sound wave data; the data processing and analyzing module comprises an audio processing unit, a band-type brake sound wave characteristic parameter extracting unit and a mode identifying unit, wherein the audio processing unit obtains the frequency spectrum energy distribution of original sound waves by adopting Fourier transform based on input vehicle running sound wave data, outputs the frequency spectrum energy distribution data to the band-type brake characteristic parameter extracting unit, pre-classifies historical sound wave data samples, and stores the frequency spectrum distribution vector mean value of the same type of sound wave samples in a local database as template data; the band-type brake sound wave characteristic parameter extraction unit is used for extracting each frequency spectrum distribution characteristic parameter by combining sound wave frequency spectrum data output by the audio processing unit; the pattern recognition unit is used for recognizing whether the current sound wave is abnormal or not by comparing various sound wave spectrum characteristics stored in a local database on the basis of the sound wave spectrum characteristics extracted in real time; a foreground information issuing module; the foreground information issuing module comprises a real-time waveform feature display unit, an abnormal band-type brake alarm unit and a time-sharing data information statistical unit; the real-time waveform feature display unit is used for visualizing the acquired sound wave according to a certain code rate and synchronously displaying the features of the band-type brake sound wave according to the result of the early-stage investigation and analysis; the abnormal band-type brake alarm unit is used for outputting the collected sound wave data to a data processing and analyzing module with a built-in recognition algorithm through a data intercommunication interface, and if the abnormal sound wave data is recognized through a mode, outputting alarm information and carrying out voice broadcasting prompt; the time-sharing data information statistical unit is used for generating a safety report for the abnormal condition monitored by the vehicle and generating a related report according to the statistical time-sharing according to related indexes such as occurrence frequency and frequency.
Furthermore, the pattern recognition unit comprises a template matching unit and an SVM classifier, the template matching unit is used for realizing matching of real-time sound waves in the sound wave template, the SVM classifier utilizes machine learning to extract abnormal band-type brake sound wave characteristics, reasonable time step lengths are set by combining sampling frequency after sound wave processing, and whether forward historical sound wave waveform data of each time step length have the characteristics of band-type brake sound waves or not is judged.
The invention has the beneficial effects that:
1. the constructed safety early warning system for the contracting brake of the truck carries out real-time non-contact acquisition on the sound wave information of the passing vehicle on a peak top platform through hardware such as a sound pickup, an audio acquisition card and the like, carries out template matching on the acquired running sound wave and 11 types of sound wave samples in a training database through a built-in sound wave characteristic processing algorithm and an identification algorithm on the basis of not interfering the operation efficiency of a hump, and judges whether the sound wave of the current vehicle is abnormal or not in real time. The test result of the embodiment shows that the alarm rate of the system algorithm to the abnormal band-type brake sound wave is 97.5%, the vehicle type identification rate is 79.5%, the reliability is high, the relevant information can be recorded in a visual mode such as a graph and a report form and then submitted to the safety management personnel of the actual transportation organization of the station, and the safety detection task of the transportation production of the station can be well assisted. The scheme of the invention is beneficial to timely and effectively processing the potential safety hazard of the band-type brake of the railway freight vehicle in the transportation operation process, and avoiding the safety problems of rail abrasion, vehicle overturn and the like.
2. On the basis of comparing, selecting and analyzing various intelligent recognition algorithms, the method autonomously writes an algorithm to extract 14 spectrum characteristic parameters of various sound wave samples, comprehensively adopts template matching and an SVM training classifier to design a recognition algorithm, determines algorithm processes, mechanisms and rules, and verifies the feasibility of the algorithm. The data interconnection and intercommunication among hardware, software and servers are realized by technologies such as SQL database access, FFT signal processing, java programming and the like and by applying visualization and modular programming technologies, and a system platform is developed to automatically monitor abnormal brake vehicles in the daily transportation production operation process. The research result of the invention has stronger applicability and stability, the developed auxiliary monitoring system can effectively provide decision service for station production safety management, compared with the existing thermal imaging technology and electronic relay technology, the invention has the technical characteristics of low cost and strong scene adaptability, and the related systems pass field tests in the Shanghai railway bureau Tu lake east station.
3. The realization of three goals of safety management standardization, inspection and treatment normalization and field operation standardization is promoted. Compared with the existing goods train band-type brake early warning method, the invention forms a safety early warning system with reliability, accuracy and easiness in use. In the aspect of reliability, the system constructs an outdoor trackside acquisition equipment subsystem comprising a pickup, an audio acquisition card and a direct-current power supply, has high component precision, reliable parameters and good transmission quality, and can continuously and stably work in an indoor client server subsystem; on the aspect of accuracy, the voiceprint data of the composite abnormal band-type brake characteristic judged by comprehensively using a sonic wave characteristic processing algorithm, a machine learning algorithm and a matching identification algorithm are detected and identified, and the alarm accuracy is improved; on the aspect of usability, the B/S framework design adopted by the system of the invention provides a good human-computer interaction interface, and the convenience and the usability of operation and maintenance are emphasized. The invention can solve the problem of realizing the safety early warning of the railway freight car band-type brake with low cost and high efficiency, and can be applied to the working process of freight car running state monitoring and station marshalling-off operation management.
Drawings
Fig. 1 is a schematic working flow of the freight car band-type brake safety early warning method based on abnormal sound wave feature identification provided by the invention.
Fig. 2 is a hardware configuration of the field simple sound wave acquisition system of the present invention.
Fig. 3 is a schematic diagram of the selection of the acquisition position of the field acoustic test according to the invention.
Fig. 4 is a schematic diagram of the spectrum characteristics of the sound wave based on the fast fourier transform.
Fig. 5 is a schematic diagram of the screening of the characteristic value of the peak frequency of the sound wave.
FIG. 6 is a schematic diagram of the matching principle of the dynamic bending template of the present invention.
Fig. 7 is a schematic diagram of the safety early warning system for the contracting brake of the truck based on the abnormal sound wave feature recognition provided by the invention.
Fig. 8 shows a main interface scheme of the platform of the early warning system for the freight car band-type brake.
FIG. 9 is an alarm processing interface of the warning system of the present invention.
FIG. 10 is a report statistical analysis interface of the warning system of the present invention.
Fig. 11 is an interface for managing the authority of the warning system according to the present invention.
FIG. 12 shows the frequency distribution of acoustic wave templates of normal vehicle types and abnormal band-type brakes of JSQ and tank vehicles in the embodiment of the invention.
Fig. 13 is a graph showing the frequency distribution of acoustic templates for 6 normal background acoustic wave types in example 6 of the present invention.
Detailed Description
The technical solutions provided by the present invention will be described in detail with reference to specific examples, which should be understood that the following specific embodiments are only illustrative and not limiting the scope of the present invention. Additionally, the steps illustrated in the flow charts of the figures may be performed in a computer system such as a set of computer-executable instructions and, although a logical order is illustrated in the flow charts, in some cases, the steps illustrated or described may be performed in an order different than here.
The invention provides a safety early warning method for a truck band-type brake based on abnormal sound wave characteristic identification, which has a flow as shown in the attached figure 1 and comprises the following steps:
s1: trackside historical sound wave data acquisition and analysis
The method has the advantages that the railway side historical sound wave collection is realized through the wagon brake sound wave collection hardware system, numbering is carried out according to date and time, and the sound wave file is transmitted and stored to the server, so that follow-up retrieval and analysis are facilitated.
S1-1: historical data collection
As a large amount of historical data is needed as a basis for a subsequent sound wave detection and identification algorithm, characteristic parameter templates of different types of sound waves are extracted through test data to serve as a basis for realizing real-time judgment of the abnormal contracting brake state. Therefore, various sound wave data acquisition tests need to be firstly carried out, and the types and data characteristics of sound waves possibly generated in the process of truck operation or shunting operation are analyzed, so that the sound waves are used as the basis of the S2 data processing and identifying step.
(1) Data acquisition test
The step adopts the simple on-site acoustic wave data acquisition system shown in fig. 2 for acquisition, and the specific test scheme is designed as follows:
1) Acoustic data acquisition location selection
In order to ensure the consistency between the acquired data position and the later equipment arrangement position, the acoustic data acquisition position is selected at a position 100 meters away from a hook lifting room annunciator when the freight train marshalling station arrives, and the specific arrangement position is shown in figure 3.
2) Collecting abnormal band-type brake sound wave data of trucks of various types
In order to reduce the influence on the daily operation of the marshalling station, the traffic flow is small in the time period from 1 o 'clock to 3 o' clock in the afternoon, and the field operation instructor commands the brake of the test vehicle and the train to move in the time period. Because the test vehicle is empty, in order to simulate the situation that the vehicle loaded with goods actually has an abnormal band-type brake, the wind brake is adopted during the brake operation.
3) Collecting vehicle operation background sound wave data
The background sound waves comprise track gap passing vibration sound waves generated when the vehicle normally operates on the line, operation sound waves of various types of adjacent trains, locomotive whistling and rumbling sound waves, locomotive in-out garage operation sound waves and the like. The sound waves do not need to select time periods and test vehicles, and relevant types of sound wave data can be recorded in combination with the field environment according to the sample size requirement.
S1-2: fourier transform-based acoustic feature extraction
(1) Fourier transform
Since the invention is based on the state recognition of sound wave data, the time sampling frequency of a sound pickup in the system is 48000Hz, and the duration of different types of sound waves is different from 1s to 20s, the data volume of a single sound wave sample is larger. In the face of such discrete, non-periodic acoustic signals, the present invention employs Fast Fourier Transform (FFT) to reveal the frequency distribution characteristics of each type of acoustic signal, as shown in fig. 4. Fig. 4 (a) is an initial acoustic electric signal, fig. 4 (b) is a frequency spectrum characteristic after FFT conversion, wherein 0.5-2s is a vehicle whistle, 2-3.5 is a coupler sound when the vehicle is started, a band-type brake starts to appear from far to near after 6 seconds, band-type brake sound waves are intensively distributed in 9 to 19s, and meanwhile, the frequency distribution of different sound waves is obviously different.
(2) Establishment and extraction of acoustic wave characteristic vector
The sound wave characteristic vector is the key of the method, and the vector capable of reflecting the sound wave characteristic is extracted from the complex and discrete initial sound wave sample data, so that the identification efficiency is improved, and the real-time performance of early warning is improved. Through repeated design, the invention provides the following characteristic vector extraction steps:
1) Taking an energy matrix obtained after FFT as an initial sample, wherein the energy matrix is arranged in a row frequency axis and a column time axis, and each time corresponds to a sound sample of a column vector, namely each time corresponds to the frequency distribution of one sound sample.
2) Because the sound wave acquisition equipment is influenced by weather conditions such as working temperature, humidity, wind speed and the like, the energy distribution difference of the same type of sound waves acquired under different working conditions is obvious, and in order to eliminate the difference of the same type of sound waves, a Min-Max normalization method is adopted to linearly change the original sound wave frequency spectrum energy data.
3) The frequency distribution after Fourier transform is 0-6000 Hz, and is equally divided into 512 frequency zones, for further statistical analysis, the initial data sample is divided into 60 frequency zones according to the frequency interval of 100 Hz.
4) After Fourier transformation, 15-17 sound samples exist every second, namely, band-type brake abnormal data per second can provide frequency vector data at 16 moments on average.
5) And (3) counting the characteristic parameters contained in the sound vector data corresponding to each moment, and extracting 6 peak values and the energy in the corresponding 6 peak value frequency bands to obtain 14 parameters in total.
6) And the obtained feature matrix is used as an input sample matrix of a data matching algorithm after being transposed.
The peak value extraction in the step 5) is a difficulty. In order to ensure the independence between the screened frequency characteristics, each frequency peak value should meet certain interval requirements, and 200Hz is selected as the minimum frequency peak value interval by observing a large amount of sound wave frequency distribution data. The frequency peak screening is schematically shown in the following figure. As shown in fig. 5, the first 3 mutually independent frequency peaks can be screened according to the peak energy level, and if the 4 th frequency peak is screened according to the energy level, the point 1 will become the 4 th frequency peak. However, in reality, the 2 nd peak to the left of point 1 is too close to the point 1, and both actually belong to the same peak, so it is necessary to exclude the possibility that point 1 is the 4 th peak. Based on this constraint, point 2 is then screened as the 4 th frequency peak. The screening of the 5 th and 6 th frequency peaks also needs to follow the requirement of minimum frequency interval to ensure the independence between the characteristic peaks.
The feature vector extraction matrix formed by the invention is shown in table 1, and 14 sound wave feature parameters are represented by parameters x1 to x 14.
TABLE 1 14 characteristic parameter indices and corresponding sample data storage
S1-3: acoustic data sample analysis and template storage
(1) Sample classification
As the railway freight vehicles have 5 types of common tank cars, gondola cars, boxcars, JSQ cars and flatcars, through field tests, 4 groups of effective abnormal band-type brake data of the tank cars, 8 groups of the gondola cars, 8 groups of the boxcars, 10 groups of the JSQ cars and 10 groups of the flatcars are collected. In a similar way, the normal background sound waves are 6 in total, 8 groups of cross-track vibration sound waves, 10 groups of train running sound waves of each vehicle type of adjacent lines, 4 groups of locomotive whistle sound waves, 4 groups of locomotive engine roaring sound waves, 4 groups of locomotive in-out warehouse running sound waves and 4 groups of vehicle braking hooking sound waves are collected. In order to facilitate the subsequent sound wave identification, the sound wave types are numbered as follows:
tank car band-type brake sound wave 1, gondola car band-type brake sound wave 2, box car band-type brake sound wave 3, JSQ car band-type brake sound wave 4, flatcar band-type brake sound wave 5, cross rail gap vibrations sound wave 6, each motorcycle type train operation sound wave 7 of adjacent line, locomotive whistle sound wave 8, locomotive engine rumble sound wave 9, locomotive business turn over storehouse operation sound wave 10, vehicle braking hook-hiding sound wave 11.
(2) Template storage
Combining 11 sound wave data samples, storing the average value of the sound wave samples of the same type in a local database as template data, wherein the time length of each sound wave template is m (in the invention, the value of m is 16, and the corresponding time range is 1s, namely 16 time sequence samples exist in each second after data processing).
S1-4: designing recognition classifier, based on the collected trackside historical data sample, using machine learning training to generate classifier and corresponding structural data
Classifier design
A Support Vector Machine (SVM) is a traditional binary classification method, and the invention relates to multi-classification recognition of band-type brake sound waves of various types of trucks. The SVM multi-classification algorithm can be divided into two types of 1 to 1 and 1 to more. The 1-to-1 classification speed is low and the reliability is high, and the 1-to-multi classification speed is high and the precision is low. According to the type of the test data collected in S1-2, the scene of the invention relates to 11 types of sound wave data (including 5 types of vehicle abnormal brake data and 6 types of common operation background sound wave data), and if 11 types of sound wave data are compared pairwise, the 11 types of sound wave data need to be compared
An SVM classifier is used for improving efficiency and emphasizing the key basisThe reduction operation is appropriately performed. Moreover, some classifiers are not necessary, for example, if the template matching result does not contain abnormal data, the template is a normal sound wave type template, i.e., on the premise of not activating the abnormal template, further identification is not necessary basically. However, some interferences between a normal sound wave type and an abnormal sound wave type, such as a rail-seam vibration sound wave and a hook-breaking sound wave, need to be considered; and similarities between abnormal band-type brake sound waves, such as between JSQ car band-type brake sound waves and tank car band-type brake sound waves, and between flatcar band-type brake and gondola car band-type brake sound waves. Based on the above rules, the present invention designs the following 20 sets of classifiers, as shown in table 2.
Table 2 20 set of classifiers relating to acoustic wave type combinations and assignments
| Classifier | Acoustic wave combination | Output value | Sorting device | Acoustic wave combination | Output value |
| 1 | 1 and 4 | 1、-1 | 11 | 4 and 8 | 1、-1 |
| 2 | 2 and 5 | 1、-1 | 12 | 2 and 9 | 1、-1 |
| 3 | 1 and 6 | 1、-1 | 13 | 3 and 9 | 1、-1 |
| 4 | 1 and 11 | 1、-1 | 14 | 3 and 10 | 1、-1 |
| 5 | 4 and 6 | 1、-1 | 15 | 4 and 10 | 1、-1 |
| 6 | 4 and 11 | 1、-1 | 16 | 4 and 11 | 1、-1 |
| 7 | 3 and 6 | 1、-1 | 17 | 1 and 10 | 1、-1 |
| 8 | 2 and 7 | 1、-1 | 18 | 1 and 11 | 1、-1 |
| 9 | 3 and 7 | 1、-1 | 19 | 2 and 6 | 1、-1 |
| 10 | 1 and 8 | 1、-1 | 20 | 5 and 10 | 1、-1 |
And each group of classifiers is trained by adopting the characteristic matrix corresponding to the sound wave type to generate corresponding classifiers and classification rules.
S2, processing sound wave data of real-time running of the freight train;
s2-1: real-time acoustic data acquisition
And acquiring real-time abnormal band-type brake sound wave data of the truck in the same manner as the S1-1. The acquired sound wave data are stored in an SQL server database according to a certain code rate, and the sound wave characteristics acquired in real time at present are synchronously and visually displayed according to the 14 sound wave characteristic parameters in the table 1, so that the current waveform and the abnormal waveform can be conveniently compared.
S2-2: real-time acoustic data processing
And (3) obtaining the spectral energy distribution of the original sound wave by adopting (1) Fourier transform (FFT) in S1-2 based on the input real-time vehicle running sound wave data.
S3: feature matching based acoustic wave identification
The method is characterized by performing feature matching based on two types of data streams, wherein one type is identification classifier structure data and sample data obtained by performing feature extraction and machine learning training through a Fourier transform-based acoustic feature extraction algorithm according to pre-investigated historical band-type brake acoustic wave and background acoustic wave data acquired by an S1 test; the other type is sound wave data obtained by preprocessing audio acquired in real time in the S2 field. And performing feature matching in the step, wherein the feature matching adopts a recognition algorithm module, and comprises the steps of comparing real-time data with sample data, screening a matching template, activating a classifier and outputting a recognition result, and judging whether to issue alarm information or not according to the recognition result. Whether an alarm is generated or not, relevant data and information can be recorded in a local database, and subsequent data analysis and management work is facilitated.
In order to avoid the limitation of a single algorithm, the invention innovatively provides an intelligent classification algorithm integrating template matching and machine learning so as to realize the identification of the abnormal brake sound wave of the truck.
S3-1: dynamic buckling template matching
The traditional template matching algorithm is that n sub-templates are extracted from band-type brake sound wave data of different types of trucks based on test data, whether each template appears in a section of frequency spectrum needs to be analyzed in data analysis, similarity among data samples is calculated, generally cosine distance or Pearson correlation coefficient is adopted to calculate similarity, and a threshold range of similarity is trained by combining a large amount of similar sample data. And during real-time audio analysis, data conversion and processing are performed firstly, similarity calculation is performed, and if the corresponding distance value is within the threshold range, a brake early warning is sent out.
However, the application scene of the invention is focused on comparing the similarity of the sound wave frequency sequences, and the difference of the degree of tightness of brake disc brake contracting of the freight train and the difference of vehicle speed can cause the difference of frequency composition and frequency band distribution when the same type of vehicle is contracting. Therefore, the similarity between sample data cannot be effectively judged by pairwise comparison in the traditional template matching, and the template matching of the acoustic wave samples is realized based on the dynamic bending thought. The improved template matching process of the present invention is analyzed in conjunction with fig. 6.
1) And calculating the similar distance. Because real-time sound wave input data are generally composite sound waves, different sound wave types exist in a monitoring period, sound wave fragments with the time length of m and 11 types of sound wave templates in a database need to be extracted for similar distance calculation, and a calculation formula is shown as a formula (1).
Where δ is the similarity distance between the real-time acoustic segment and the template, t i And f i Respectively representing the time of a sample point in the template and the corresponding acoustic amplitude parameter, t j And f j The time and amplitude parameters in the real-time sound wave segment correspond, respectively.
2) And calculating the similarity. According to the similarity distance calculation formula (1), the similarity distance delta between any real-time segment s and the template k can be obtained sk . Because training thresholds of different templates are different, the smaller the similarity distance is, the closer the actually measured segment is to the observed template is, and a calculation formula (2) of the similarity is constructed.
In the formula d sk Similarity of the sound wave segment s and the template k; delta. For the preparation of a coating sk The similarity distance between the real-time sound wave segment s and the template k is calculated according to the formula (1); s k Is a similar distance threshold of the kth type sound wave sample, if the measured distance is within the threshold, the characteristic of the current sound wave template is represented, but the measured distance does not mean that the characteristic can be determinedThe acoustic wave template is determined as the current type because the condition that the same acoustic wave sample simultaneously meets the thresholds of the multiple types of acoustic wave templates exists.
3) And extracting an approximate template. And according to the similarity calculation result, selecting the sound wave types corresponding to 3 highest similarity values from the 11 similarities, and extracting corresponding template sound wave data and sound wave characteristic data to serve as the basis of subsequent identification.
S3-2: support vector machine identification
1) Classifier activation and outcome determination
Through dynamic bending template matching, the similar distance of real-time sound wave data corresponding to 11 sound wave samples is calculated, and the activation and result judgment process of the SVM classifier provided by the invention is as follows:
(1) extracting 3 closest sound wave types of the real-time sound wave samples according to the template matching result based on the dynamic bending, wherein the sound wave types are assumed to be a, b and c;
(2) if any value from 1 to 5 does not appear in a \ b \ c, the current sound wave data cannot be matched with abnormal brake data of various trucks, and the early warning result is output as a normal type;
(3) if at least one value from 1 to 5 exists in a \ b \ c, the corresponding classifier needs to be activated, and the method comprises the following steps: an a-b classifier, an a-c classifier and a b-c classifier;
(4) and (4) judging a classification result, loading the actually measured data by combining the classifier determined in the step (3) and outputting the classification result, and constructing the following judgment rules for the method because the classification result has differences:
S4: early warning and recording
S4-1: early warning of truck band-type brake
And if the current real-time sound wave type is judged to be abnormal band-type sound waves after the activation of the classifier in the S3, the system outputs alarm information and carries out voice broadcast prompting. Because the real-time acoustic waveform visualization is carried out in the S2-1, after the alarm is given, the on-duty manager can more clearly confirm whether the current waveform is abnormal for the second time, the reliability of alarm prompt is ensured from two aspects of machines and workers, the invalid alarm is avoided, and the false alarm rate is reduced.
S4-2: vehicle safety record
The system can generate a safety report according to the monitored abnormal brake-contracting condition of the truck, and generate related reports according to related indexes such as occurrence frequency and frequency according to different statistical time periods such as weeks, months, years and the like, so that station managers can analyze the problems existing in the current production of transport vehicles, and can make operation and maintenance work of related equipment facilities in time.
In order to realize the freight car band-type brake safety early warning method based on the abnormal sound wave characteristic identification, the invention designs an early warning system platform, and the design requirements comprise the following 8 items:
(1) login and authority management of system users and the like.
(2) And storing and managing abnormal and normal sound wave data.
(3) The system is internally provided with an automatic band-type brake vehicle sound wave identification algorithm, judges abnormal conditions based on a real-time waveform file and gives voice alarm information.
(4) And (5) auditing the abnormal vehicle alarm information by the manager on duty, and performing secondary confirmation.
(5) The system synchronously visualizes the real-time waveform data and the band-type brake characteristic index data.
(6) And (4) carrying out classified management on abnormal waveform data of the truck band-type brake.
(7) The system generates a safety report according to the abnormal conditions monitored by the operation vehicle, and generates a statistical analysis report, a histogram and other analysis tables of the abnormal band-type brake vehicle according to the week, zhou Yue and time intervals.
(8) And the system user refers to the historical information and calls a historical analysis statistical result.
Specifically, the invention provides a freight car band-type brake safety early warning system based on abnormal sound wave feature recognition, which is structured as shown in figure 7, and comprises 3 basic modules of field sound wave acquisition, data processing analysis and foreground information release, wherein data and information are exchanged among the modules through a local database. The local database is responsible for coordinating and sharing information data among all the modules while storing various data according to different functional requirements. The composition and interaction relationship of each sub-module will be described in detail below.
(1) On-site sound wave acquisition module
The invention constructs a set of simple on-site sound wave data acquisition module, which comprises 4 devices, namely a direct current power supply, a trackside sound pickup, an audio acquisition card and a storage server. The field sound wave collection module is shown in fig. 2, and further comprises a transmission cable.
Wherein:
the direct current power supply is a portable mobile power supply 30000mA and 12V and is responsible for supplying power to the pickup beside the rail;
the sound wave sampling precision of the trackside sound pickup is 48000Hz, the working environment temperature is-40-75 ℃, and the main function is to transmit the collected sound wave electric signal data to the audio acquisition card through a cable;
the audio acquisition card is a drive-free installation acquisition box, is inserted on the server and is directly powered by the server, and the portable sound card has the main functions of receiving sound wave electric signals transmitted by a sound pick-up and converting the electric signals into digital signals to be transmitted to the storage server;
the storage server chip is an Intel Xeon E2224G notebook, has 8G internal memory and 1T hard disk, and has the main functions of storing, analyzing and managing sound wave data acquired in field implementation and supplying power to the audio acquisition card.
The field sound wave acquisition module is used for acquiring pre-investigated trackside historical sound wave data (including abnormal band-type brake sound waves and background sound wave data) and is also used for acquiring field sound wave data in real time. The on-site sound wave acquisition module can realize the collection work of the sound waves generated by the running of the vehicle, and can number the collected sound wave data samples according to the date and time, and transmit and store the sound wave files into the server, so that the follow-up retrieval and analysis are facilitated.
(2) Data processing and analyzing module
The data processing and analyzing module consists of an audio processing part, a band-type brake characteristic parameter extraction part and a pattern recognition part 3:
(1) the audio processing unit is used for obtaining the frequency spectrum energy distribution of the original sound wave by adopting Fourier transform (FFT) based on the input vehicle running sound wave data and outputting the frequency spectrum energy distribution data to the band-type brake characteristic parameter extraction unit; the fourier transform (FFT) details are described in the method of the invention. Meanwhile, the audio processing unit pre-classifies historical sound wave data samples, the historical sound wave data samples are divided into 11 sound wave types according to historical data acquisition serial number records in the S1, and the audio processing unit is combined with the 11 sound wave data samples, and the frequency spectrum distribution vector mean value of the similar sound wave samples is used as template data to be stored in a local database.
(2) The band-type brake sound wave characteristic parameter extraction unit is used for extracting 14 spectrum distribution characteristic parameters by combining sound wave spectrum data output by the audio processing unit, and a specific extraction algorithm is explained in the method S1-2.
(3) And the mode identification unit is used for identifying whether the current sound wave is abnormal or not by comparing various sound wave spectrum characteristics stored in the local database on the basis of the sound wave spectrum characteristics extracted in real time. Both the relevant recognition results and the data need to be stored in a local database. The mode recognition unit comprises a template matching unit and an SVM (support vector machine) classifier, the template matching unit is used for realizing matching of real-time sound waves in the sound wave template, the SVM classifier utilizes machine learning to extract abnormal band-type brake sound wave characteristics, reasonable time step length (0.5 s) is set by combining sampling frequency after sound wave processing, and whether forward historical sound wave waveform data of each time step length have the characteristics of band-type brake sound waves or not can be judged. Both the relevant recognition results and the data need to be stored in a local database.
(3) Foreground information issuing module
The foreground information issuing module comprises 3 main units, namely a real-time waveform characteristic display unit, an abnormal band-type brake alarm unit and a time-sharing data information statistical unit, and the main functions of the units are as follows:
(1) the real-time waveform feature display unit is used for visualizing the acquired sound wave waveforms according to a certain code rate, synchronously displaying the features of the band-type brake sound waves according to the results of early investigation and analysis (14 sound wave feature parameters in the table 1), conveniently comparing the current waveforms with abnormal waveforms, and after an alarm is given out, enabling an on-duty manager to more clearly confirm whether the current waveforms are abnormal for the second time, so that the reliability of alarm prompt is ensured from two aspects of a machine and a worker.
(2) And the abnormal band-type brake alarm unit outputs the collected sound wave data to a data processing and analyzing module with a built-in recognition algorithm through a data intercommunication interface, and if the abnormal band-type brake alarm unit is recognized through a mode, alarm information is output and voice broadcasting prompt is carried out.
(3) And the time-sharing data information statistical unit is used for generating a safety report according to the abnormal conditions monitored by the vehicles and generating a related report according to related indexes such as occurrence frequency and frequency according to a statistical time period, so that the station managers can conveniently find the existing problems.
As shown in fig. 8, the early warning system main body interface designed by the present invention can be divided into 4 large sections, which are respectively a title section, a menu section, a waveform display section and a feature data display section, and the basic descriptions of the sections are as follows:
(1) Title block: the Chinese and English names of the system.
(2) A menu section block: the function menu necessary for the system comprises three functions of band-type brake monitoring, historical recording and authority management. The selectable function items of the band-type brake monitoring pull-down secondary menu comprise a waveform display function, an alarm processing function and a monitoring starting function; the selectable function items of the drop-down secondary menu of the historical record layout block comprise a historical data viewing function and a statistical form generating function.
(3) Waveform display block: displaying a sound wave form according to a signal input by the audio acquisition equipment in real time;
(4) The characteristic data display block: and displaying 14 characteristic index values of the monitored sound wave sample at the corresponding moment, and calculating the characteristic index values every 0.5 second by the background, wherein the characteristic data dynamically changes along with the calculation.
In order to facilitate the operation and management of the system, the invention further designs an alarm processing function interface, a statistical report function interface and an authority management function interface.
The alarm processing function interface mainly displays four types of basic information, namely an abnormal data number, abnormal data occurrence time, an abnormal sound wave type and alarm operation, as shown in figure 9. The alarm operation supports playing, storing, deleting and the like of abnormal sound wave data, and can provide reference basis for field management personnel to confirm alarm information and update a historical training database in a later period.
The information mainly displayed by the report statistical analysis interface comprises a statistical chart and a statistical table, and the time information such as the starting and ending years and months can be set in the uppermost statistical period by self, as shown in fig. 10. The statistical chart can be displayed in a selected mode by a line chart, a bar chart and a text box; the information displayed by the statistical table is the time distribution of various abnormal sound wave alarms in the whole day in the statistical period, including the time range, the occurrence frequency and the proportion, is favorable for searching the high-occurrence period of the band-type brake in the daily work plan, and improves the alertness of the running operation of the truck in the Gao Duiying period.
The information mainly displayed by the authority management function interface includes basic information of the work number, name, account number and password of the administrator for verifying the identity when logging in the system, as shown in fig. 11. Meanwhile, the basic operation of the authority management supports editing and deleting of the account and the password, and supports adding of a new account and searching of the job number.
In order to fully demonstrate the scientificity and the accuracy of the scheme, the truck band-type brake safety early warning system formed by the invention is preliminarily tested in a railway marshalling station, a test scene mainly aims at analyzing the band-type brake conditions of two truck types, namely a tank truck type and a JSQ truck type, and test sample data are shown in a table 3.
TABLE 3 actual measurement of various types of acoustic wave sample data
| Sample set | Type of sound | Extraction sample size (full sample energy matrix) |
| 1 | JSQ abnormal band-type brake sound wave | 513×812 |
| 2 | JSQ Normal operation Acoustic wave | 513×219 |
| 3 | Abnormal band-type brake sound wave of tank car | 513×499 |
| 4 | Sound wave for normal operation of tank car | 513×313 |
| 5 | Rail gap vibration sound wave | 513×37 |
| 6 | Sound wave for adjacent train running | 513×375 |
| 7 | Whistling sound wave for locomotive | 513×147 |
| 8 | Rolling sound wave for locomotive | 513×260 |
| 9 | Sound wave for locomotive entering and exiting garage | 513×141 |
| 10 | Sound wave for braking cloacan hook of vehicle | 513×57 |
According to the extracted sample energy matrix, firstly normalizing according to columns, then solving a mean value according to rows to obtain template vectors of various sound waves, and firstly analyzing the frequency distribution conditions of various sound wave samples according to the template vectors, wherein the frequency distributions of sound wave templates with normal JSQ models and normal tank cars and abnormal band-type brakes are shown in an attached figure 12. The JSQ vehicle and the tank vehicle can operate normally with sound waves which are relatively close to each other and have basically consistent variation trend of frequency energy distribution curves, and the variation trend is concentrated in a low frequency range through visual analysis; frequency energy distribution of abnormal band-type brake sound waves of the JSQ vehicle is dispersed, and frequency distribution is biased to a middle-high frequency range; the frequency energy distribution of the abnormal band-type brake sound waves of the tank car is also relatively dispersed, the frequency distribution is also biased to a middle and high frequency range, but the peak position of the abnormal band-type brake sound waves is obviously different from that of the JSQ abnormal band-type brake sound waves.
Similarly, the frequency distribution curves of the 6 background acoustic wave templates are shown in fig. 13. According to analysis, the sound wave frequencies of adjacent line train operation, vehicle brake hooks, locomotive rolling and the like take low frequencies as main components, wherein the low frequencies of the sound waves of the adjacent line train operation are particularly prominent, and the total energy occupation ratio below 300Hz is up to 21.6%; the energy distribution curve of the rail gap vibration sound wave template has a plurality of discrete peak values, the characteristics are obvious, and the frequency peak sections are mostly concentrated below 3000 Hz; the howling sound wave generated when the vehicle is put in and out of the sliding device has a plurality of peak values in frequency bands of about 500Hz and 2000-3000 Hz, and the frequency distribution is mainly medium frequency.
According to the 14 sound wave characteristic parameters provided by the invention in S2, the distribution of the template characteristic parameters corresponding to the 10 sound wave sample sets in the present embodiment is shown in table 4. According to the characteristic parameter extraction result, the energy distribution of the band-type brakes (sample sets 1 and 3) of the tank truck and the JSQ truck is relatively close, particularly the percentage of 6 peak energy is uniformly distributed between 2.1% and 2.7%, and the energy peaks of the two types are mainly in a middle-high frequency range. Meanwhile, the frequency of sound wave energy generated by the vehicle rail-gap-crossing vibration (sample set 5) and the second vehicle braking inertia (sample set 10) is the highest, wherein the frequency characteristic of the vehicle rail-gap-crossing vibration sound wave is closer to the frequency characteristic of the tank car abnormal band-type brake sound wave (sample set 3).
TABLE 4 template characteristic parameter distribution
Based on the acoustic wave data, part of the data in each sample is extracted as a test set, and a classification recognition algorithm is called to obtain a test result, as shown in table 5. Analysis shows that for each type of sound wave, the identification result of the same sound wave test set may contain different sound wave types, for example, a JSQ abnormal band brake may be identified as a tank car band brake or rail gap vibration, a locomotive whistle may be identified as a tank car band brake, and the like. According to the identification result, the average alarm rate of the algorithm to the abnormal sound waves is 97.5%, and the identification rate is 79.5%; for other normal background sound wave types, the average alarm rate is 2.3%, and the identification rate is not considered, because the algorithm temporarily does not design an identification code for a normal sound wave template for improving the efficiency, and because the identification is not needed as long as the normal sound wave is detected. Through analysis of sample test results, the template matching and SVM recognition method provided by the invention is considered to have higher feasibility and reliability.
TABLE 5 analysis of test sample identification results
It should be noted that the above-mentioned contents only illustrate the technical idea of the present invention, and the protection scope of the present invention is not limited thereby, and it is obvious to those skilled in the art that several modifications and decorations can be made without departing from the principle of the present invention, and these modifications and decorations fall within the protection scope of the claims of the present invention.
Claims (9)
1. A safety early warning method for a truck band-type brake based on abnormal sound wave feature recognition is characterized by comprising the following steps:
s1: trackside historical sound wave data acquisition and analysis
The method comprises the steps that rail-side historical sound wave collection is achieved through a wagon brake sound wave collection hardware system, numbering is conducted according to date and time, and sound wave files are transmitted and stored to a server; the method specifically comprises the following steps:
s1-1: historical data collection
After the acoustic wave data acquisition position is selected, a field acoustic wave data acquisition system is arranged, abnormal band-type brake acoustic wave data of trucks of various types are acquired, and vehicle operation background acoustic wave data are acquired;
s1-2: fourier transform-based acoustic feature extraction
Obtaining the frequency distribution characteristics of various types of sound wave signals by adopting fast Fourier transform;
taking an energy matrix obtained after FFT as an initial sample, wherein the energy matrix is in a row frequency axis and a column time axis, and each moment corresponds to a sound sample of a column vector, namely the frequency distribution of each moment corresponding to one sound sample;
performing linear change on the original sound wave frequency spectrum energy data by adopting a Min-Max normalization method;
the frequency distribution after Fourier transform is 0-6000 Hz, the frequency distribution is equally divided into 512 frequency sections, and the initial data sample is divided into 60 frequency sections according to the frequency interval of 100 Hz;
after Fourier transformation, 15-17 sound samples exist every second, namely band-type brake abnormal data every second can provide frequency vector data at 16 moments on average;
counting characteristic parameters contained in sound vector data corresponding to each moment, extracting 6 peak values and energy in corresponding 6 peak value frequency ranges, and totaling 14 parameters;
the obtained characteristic matrix is used as an input sample matrix of a data matching algorithm after being transformed;
s1-3: acoustic data sample analysis and template storage
Numbering the sound wave types according to the types of the railway freight vehicles and the types of the background sound waves; combining all sound wave data samples, and storing the sample mean value of the same type of sound waves as template data in a local database;
s1-4: designing an identification classifier, and generating the classifier and corresponding structural data by adopting machine learning training on the basis of the acquired trackside historical data sample;
s2, real-time running sound wave data processing of the freight train comprises the following steps:
s2-1: real-time acoustic data acquisition
Acquiring real-time abnormal contracting brake sound wave data of the truck in the same way as the S1-1, storing the acquired sound wave data in an SQL server database according to a certain code rate, and synchronously and visually displaying the current real-time acquired sound wave characteristics according to 14 sound wave characteristic parameters;
s2-2: real-time acoustic data processing
Obtaining the frequency spectrum energy distribution of the original sound wave by adopting Fourier transform in S1-2 based on the input real-time vehicle running sound wave data;
s3: the acoustic wave identification based on feature matching comprises the following steps:
s3-1: dynamic buckling template matching
Extracting an approximate template according to the similarity distance calculation and the similarity calculation;
s3-2: vector machine identification
1) Classifier activation and outcome determination
(1) Extracting 3 closest sound wave types of the real-time sound wave samples according to the template matching result based on the dynamic bending, wherein the sound wave types are assumed to be a, b and c;
(2) if any value from 1 to 5 does not appear in a \ b \ c, the current sound wave data cannot be matched with abnormal brake data of various trucks, and the early warning result is output as a normal type;
(3) if at least one value from 1 to 5 exists in a \ b \ c, the corresponding classifier needs to be activated, and the method comprises the following steps: an a-b classifier, an a-c classifier and a b-c classifier;
(4) judging a classification result, and loading the actually measured data to output the classification result by combining the classifier determined in the step (3);
s4: early warning and recording
S4-1: early warning of truck band-type brake
If the current real-time sound wave type is judged to be abnormal band-type sound waves after being activated by the classifier in the S3, outputting alarm information and carrying out voice broadcasting prompt;
s4-2: vehicle safety record
And generating a safety report according to the monitored abnormal brake condition of the truck, and generating a related report according to different statistical time periods and related indexes such as occurrence frequency and frequency.
2. The safety early warning method for the contracting brake of the truck based on the abnormal sound wave feature recognition of claim 1, wherein in the step S1-1, the background sound wave comprises: the sound waves of vibration of the passing rail gap when the vehicle normally operates on the line, the sound waves of operation of each model of an adjacent line train, the sound waves of whistling and rumbling of a locomotive, and the sound waves of operation of the locomotive in and out of a warehouse.
3. The abnormal sound wave feature identification-based freight car brake safety early warning method according to claim 1, wherein 200Hz is selected as a minimum frequency peak interval in the step S1-2.
4. The abnormal sound wave feature recognition-based freight car band-type brake safety early warning method according to claim 1, wherein in the step S1-4, the classifiers are 20 groups, and the sound wave types involved in the classifiers are combined and assigned as follows:
and each group of classifiers is trained by adopting the characteristic matrix corresponding to the sound wave type to generate corresponding classifiers and classification rules.
5. The abnormal sound wave feature identification based safety precaution method for the brake of the freight car according to claim 1, wherein in the step 2-1, the sound wave features collected in real time at present are synchronously and visually displayed.
6. The abnormal sound wave feature recognition-based safety early warning method for the contracting brake of the truck as claimed in claim 1, wherein the step S3-1 specifically comprises the following steps:
1) Similarity distance calculation
Extracting sound wave fragments with m time length and 11 types of sound wave templates in a database to calculate the similar distance, wherein the calculation formula is as shown in formula (1):
where δ is the similarity distance between the real-time acoustic segment and the template, t i And f i Respectively representing the time of a sample point in the template and the corresponding acoustic amplitude parameter, t j And f j Respectively corresponding to the time and amplitude parameters in the real-time sound wave segment;
2) Similarity calculation
A calculation formula (2) of the similarity is constructed:
in the formula d sk Similarity of the sound wave segment s and the template k; delta sk The similarity distance between the real-time sound wave segment s and the template k is calculated according to the formula (1); s is k A similar distance threshold value of the kth type sound wave sample;
3) Approximate template extraction
And according to the similarity calculation result, selecting the sound wave types corresponding to the 3 highest similarity values from the 11 similarities, and extracting corresponding template sound wave data and sound wave characteristic data to serve as the basis of subsequent identification.
7. The abnormal sound wave feature recognition-based safety early warning method for the contracting brake of the truck as claimed in claim 1, wherein the judgment rule in the classifier classification result judgment is as follows:
comprehensively outputting abnormal band-type brake early warning by an algorithm;
8. The utility model provides a freight train band-type brake safety precaution system based on unusual sound wave feature recognition which characterized in that for realize above-mentioned freight train band-type brake safety precaution method based on unusual sound wave feature recognition, includes: the system comprises a field sound wave acquisition module, a data processing and analyzing module and a foreground information publishing module; the on-site sound wave acquisition module is used for acquiring pre-investigated trackside historical sound wave data and acquiring on-site real-time sound wave data; the data processing and analyzing module comprises an audio processing unit, a band-type brake sound wave characteristic parameter extracting unit and a mode identifying unit, wherein the audio processing unit obtains the frequency spectrum energy distribution of original sound waves by adopting Fourier transform based on input vehicle running sound wave data, outputs the frequency spectrum energy distribution data to the band-type brake characteristic parameter extracting unit, pre-classifies historical sound wave data samples, and stores the frequency spectrum distribution vector mean value of the same type of sound wave samples in a local database as template data; the band-type brake sound wave characteristic parameter extraction unit is used for extracting each frequency spectrum distribution characteristic parameter by combining sound wave frequency spectrum data output by the audio processing unit; the pattern recognition unit is used for recognizing whether the current sound wave is abnormal or not by comparing various sound wave spectrum characteristics stored in a local database on the basis of the sound wave spectrum characteristics extracted in real time; a foreground information issuing module; the foreground information issuing module comprises a real-time waveform characteristic display unit, an abnormal band-type brake alarm unit and a time-sharing data information statistical unit; the real-time waveform feature display unit is used for visualizing the acquired sound wave according to a certain code rate and synchronously displaying the features of the band-type brake sound wave according to the result of the early-stage investigation and analysis; the abnormal band-type brake alarm unit is used for outputting the collected sound wave data to a data processing analysis module with a built-in recognition algorithm through a data intercommunication interface, and if the abnormal sound wave data exist through pattern recognition, outputting alarm information and carrying out voice broadcast prompting; the time-sharing data information statistical unit is used for generating a safety report for the abnormal condition monitored by the vehicle and generating a related report according to the statistical time-sharing according to related indexes such as occurrence frequency and frequency.
9. The truck band-type brake safety early warning system based on abnormal sound wave feature recognition according to claim 8, wherein the pattern recognition unit comprises a template matching unit and an SVM classifier, the template matching unit is used for realizing matching of real-time sound waves in a sound wave template, the SVM classifier utilizes machine learning to extract abnormal band-type brake sound wave features, reasonable time step lengths are set according to sampling frequency after sound wave processing, and whether forward historical sound wave waveform data of each time step length have the features of band-type brake sound waves or not is judged.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210826826.8A CN115200695A (en) | 2022-07-14 | 2022-07-14 | Abnormal sound wave characteristic identification-based safety early warning method and system for truck band-type brake |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210826826.8A CN115200695A (en) | 2022-07-14 | 2022-07-14 | Abnormal sound wave characteristic identification-based safety early warning method and system for truck band-type brake |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115200695A true CN115200695A (en) | 2022-10-18 |
Family
ID=83580429
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210826826.8A Pending CN115200695A (en) | 2022-07-14 | 2022-07-14 | Abnormal sound wave characteristic identification-based safety early warning method and system for truck band-type brake |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115200695A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115881079A (en) * | 2023-02-16 | 2023-03-31 | 山东铁路投资控股集团有限公司 | Noise early warning method, system, equipment and storage medium in railway track construction |
-
2022
- 2022-07-14 CN CN202210826826.8A patent/CN115200695A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115881079A (en) * | 2023-02-16 | 2023-03-31 | 山东铁路投资控股集团有限公司 | Noise early warning method, system, equipment and storage medium in railway track construction |
| CN115881079B (en) * | 2023-02-16 | 2023-05-23 | 山东铁路投资控股集团有限公司 | Noise early warning method, system, equipment and storage medium in railway track construction |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Sun et al. | Sound based fault diagnosis for RPMs based on multi-scale fractional permutation entropy and two-scale algorithm | |
| CN202038323U (en) | Online operation status detection display system for urban mass transit | |
| Sun et al. | Fault diagnosis for train plug door using weighted fractional wavelet packet decomposition energy entropy | |
| CN109189019B (en) | Standardized monitoring system for locomotive crew member value taking | |
| Kyriakidis et al. | Data based framework to identify the most significant performance shaping factors in railway operations | |
| CN105416340B (en) | Train reminding and pre-warning system | |
| Kyriakidis et al. | Railway accidents caused by human error: historic analysis of UK railways, 1945 to 2012 | |
| Lin et al. | Quantitative causal analysis of mainline passenger train accidents in the United States | |
| Balouchi et al. | Development of railway track condition monitoring from multi-train in-service vehicles | |
| CN113486874B (en) | Rail corrugation feature identification method based on wheel-rail noise wavelet packet decomposition | |
| Ning et al. | Feature recognition of small amplitude hunting signals based on the MPE-LTSA in high-speed trains | |
| CN115200695A (en) | Abnormal sound wave characteristic identification-based safety early warning method and system for truck band-type brake | |
| Zhang et al. | Analysis of human-factor-caused freight train accidents in the United States | |
| CN109142978B (en) | Active monitoring method for running state of railway contact net | |
| Barmada et al. | Clustering techniques applied to a high-speed train pantograph–catenary subsystem for electric arc detection and classification | |
| CN205524319U (en) | Braking state and warning playback device based on on --spot deposit data | |
| Sun et al. | Condition monitoring and fault diagnosis strategy of railway point machines using vibration signals | |
| CN110263718A (en) | Vehicle enters and leaves section detection method and detection system | |
| US20240075970A1 (en) | System and method for continuous welded rail risk modeling | |
| CN114118470A (en) | Intelligent management and control method and system for production and operation of full-automatic driving vehicle base | |
| Wang et al. | Development of metro track geometry fault diagnosis convolutional neural network model based on car-body vibration data | |
| Harrison et al. | At the limit? Using operational data to estimate train driver human reliability | |
| CN205157057U (en) | Infrared detecting system of rail train | |
| Shadfar et al. | Detection of rail local defects using in-service trains | |
| Ye et al. | An acoustic-based recognition algorithm for the unreleased braking of railway wagons in marshalling yards |
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 |