CN114047259A - Method for detecting multi-scale steel rail damage defects based on time sequence - Google Patents
Method for detecting multi-scale steel rail damage defects based on time sequence Download PDFInfo
- Publication number
- CN114047259A CN114047259A CN202111260216.8A CN202111260216A CN114047259A CN 114047259 A CN114047259 A CN 114047259A CN 202111260216 A CN202111260216 A CN 202111260216A CN 114047259 A CN114047259 A CN 114047259A
- Authority
- CN
- China
- Prior art keywords
- data
- time series
- ultrasonic
- steel rail
- time
- 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
- 229910000831 Steel Inorganic materials 0.000 title claims abstract description 60
- 239000010959 steel Substances 0.000 title claims abstract description 60
- 230000007547 defect Effects 0.000 title claims abstract description 22
- 238000000034 method Methods 0.000 title claims description 22
- 239000000523 sample Substances 0.000 claims abstract description 19
- 238000001514 detection method Methods 0.000 claims abstract description 18
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 17
- 238000005070 sampling Methods 0.000 claims abstract description 17
- 230000011218 segmentation Effects 0.000 claims abstract description 9
- 230000004927 fusion Effects 0.000 claims abstract description 6
- 238000000605 extraction Methods 0.000 claims abstract description 4
- 238000009499 grossing Methods 0.000 claims description 13
- 238000002604 ultrasonography Methods 0.000 claims description 6
- 238000011176 pooling Methods 0.000 claims description 4
- 230000009466 transformation Effects 0.000 claims description 4
- 238000007781 pre-processing Methods 0.000 claims description 2
- 230000006870 function Effects 0.000 abstract description 4
- 238000012545 processing Methods 0.000 abstract description 4
- 238000013135 deep learning Methods 0.000 abstract description 3
- 230000006872 improvement Effects 0.000 abstract description 3
- 230000008569 process Effects 0.000 description 5
- 238000013528 artificial neural network Methods 0.000 description 4
- 238000009826 distribution Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000012549 training Methods 0.000 description 3
- 230000005856 abnormality Effects 0.000 description 2
- 230000003321 amplification Effects 0.000 description 2
- 238000007689 inspection Methods 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 238000003199 nucleic acid amplification method Methods 0.000 description 2
- 102100037651 AP-2 complex subunit sigma Human genes 0.000 description 1
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 description 1
- 101000806914 Homo sapiens AP-2 complex subunit sigma Proteins 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000001902 propagating effect Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000012731 temporal analysis Methods 0.000 description 1
- 238000000700 time series analysis Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
- G01N29/44—Processing the detected response signal, e.g. electronic circuits specially adapted therefor
- G01N29/4445—Classification of defects
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
- G01N29/44—Processing the detected response signal, e.g. electronic circuits specially adapted therefor
- G01N29/4481—Neural networks
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
- G01N29/44—Processing the detected response signal, e.g. electronic circuits specially adapted therefor
- G01N29/46—Processing the detected response signal, e.g. electronic circuits specially adapted therefor by spectral analysis, e.g. Fourier analysis or wavelet analysis
Abstract
The invention is suitable for the technical improvement field of signal processing and deep learning, and provides a multi-scale steel rail damage defect detection method based on a time sequence, which comprises the following steps: s1, carrying out data segmentation and pretreatment according to the original track ultrasonic data of the echo channel of the ultrasonic probe type; s2, setting a sampling coefficient to sample the preprocessed segmentation data to obtain sampling data of different scales; s3, carrying out feature extraction and feature fusion on the collected rail damage data with different scales; s4, inputting the fused features into a deep convolutional neural network to connect with a softmax function to classify the time series; and S5, obtaining a rail damage list according to the recognition result of the convolutional neural network. By utilizing multi-scale steel rail ultrasonic time series signal division, the damage detection rate can be obviously improved, and the missing report rate can be greatly reduced.
Description
Technical Field
The invention belongs to the technical improvement field of signal processing and deep learning, and particularly relates to a multi-scale steel rail damage defect detection method based on a time sequence.
Background
In the process of inspection operation of the large-scale steel rail flaw detection vehicle, the advantages of high sample collection efficiency, high running speed, large detection area and the like are widely applied to the routine steel rail line inspection work. But the ultrasonic damage data of the steel rail has long playback time and large data volume, and the culture difficulty of the damage approver is high, the period is long, and the like. Ultrasonic flaw detection of steel rails requires a great deal of time and labor. Therefore, the intelligent identification method of the steel rail ultrasonic signal can greatly improve the detection efficiency of the steel rail flaw, effectively save the labor cost of steel rail flaw detection and effectively shorten the flaw detection period.
In the current extensive steel rail flaw detection intelligent detection system, a B-display image is analyzed and drawn according to steel rail ultrasonic data, and from the angle of the image, the steel rail flaw is detected by using computer vision. Thus, the image forming of the abnormal part of the steel rail can be intuitively sensed, but the interference of noise is difficult to eliminate for the steel rail ultrasonic data with insufficient data quality.
Disclosure of Invention
The invention aims to provide a method for detecting multi-scale steel rail damage defects based on a time sequence, and aims to solve the problem that the steel rail damage defects can be detected more accurately and efficiently by monitoring and evaluating the quality of steel rails through a signal processing and deep learning technology.
The invention is realized in such a way that a method for detecting the multi-scale steel rail damage defects based on time series comprises the following steps:
s1, carrying out data segmentation and pretreatment according to the original track ultrasonic data of the echo channel of the ultrasonic probe type;
s2, setting a sampling coefficient to sample the preprocessed segmentation data to obtain sampling data of different scales;
s3, carrying out feature extraction and feature fusion on the collected rail damage data with different scales;
s4, inputting the fused features into a deep convolutional neural network to connect with a softmax function to classify the time series;
and S5, obtaining a rail damage list according to the recognition result of the convolutional neural network.
The further technical scheme of the invention is as follows: and in the step S3, the smooth characteristic and the frequency domain characteristic of each group of ultrasonic time series information are extracted.
The further technical scheme of the invention is as follows: reducing the effects of noise by acquiring smoothing features in ultrasound data acquisitionWherein f issmRepresenting the smoothed ultrasound time series information; (X) X*,X*An ultrasonic time series signal representing an arbitrary combination of channels and slices; x is the number ofjAn echo point representing the ultrasonic wave output when the time is j; l denotes the step size of smoothing.
The further technical scheme of the invention is as follows: when the frequency domain characteristics are extracted, because the ultrasonic signals of the steel rail belong to a discrete time sequence, Z transformation is adopted,Z∈Rxthe discrete time series is transformed into the complex frequency domain, where Z is a complex variable, which may be represented in the form of Z ═ Z | ejω(ii) a x (n) represents a time series, and n represents a time; rxConvergence field of X (Z).
The further technical scheme of the invention is as follows: unifying the feature scale of the input deep convolutional neural network in the step S4 and adding the feature dimension includes the following steps:
s41, designing and using incompatible convolution kernel k in the first layerconvPerforming convolution operation on the time sequence;
and S42, reducing the dimension of the feature by using a pooling layer with the same output dimension of the convolution layer.
The further technical scheme of the invention is as follows: the data segmentation and the display data of the preprocessed steel rail ultrasonic data at the present stage in the step S1 are presented in an image mode, and are divided into a rail head part, a rail waist part and a rail bottom part according to a time sequence according to different ultrasonic wave output information; the ultrasonic time-series data is divided by channels.
The further technical scheme of the invention is as follows: in step S2, downsampling the original time series data using a downsampling coefficient σ to form multi-scale time series data, and performing sliding smoothing on the multi-scale time series analysis using a sliding average operator to obtain a sliding smoothing feature fsmObtaining complex frequency domain characteristics f of multi-scale time sequence by time-frequency transformationfr。
The further technical scheme of the invention is as follows: and the fusion characteristics are used as input data and input into the deep convolutional neural network, and the weight of the deep convolutional neural network is continuously adjusted through forward iteration and backward iteration, so that the loss value is minimum finally.
The invention has the beneficial effects that: the ultrasonic signal of the steel rail is regarded as a time sequence signal, and the flaw of the flaw in a certain time period can be accurately positioned through the adjustment of time scale. And the multi-scale steel rail ultrasonic time series signal division is utilized, so that the damage detection rate can be obviously improved, and the missing report rate can be greatly reduced. On the basis of multi-scale time sequence data, higher-level smooth features, complex frequency domain features and the like are continuously designed, so that the influence of noise in the ultrasonic data of the steel rail on a damage result can be effectively avoided, and the time when the damage occurs can be accurately judged. The false alarm rate of damage identification can be effectively reduced.
Drawings
Fig. 1 is a flowchart of a method for detecting a multi-scale steel rail flaw based on a time series according to an embodiment of the present invention.
Detailed Description
As shown in fig. 1, the flow chart of the method for detecting a multi-scale steel rail defect based on a time series provided by the invention is detailed as follows:
aiming at the problems of the existing steel rail damage defect detection, the invention regards the steel rail ultrasonic data as a time sequence, uses down-sampling to respectively generate time sequences with different scales aiming at the time sequence, carries out feature extraction on the time sequences with different scales, and then uses a deep convolution neural network to realize the steel rail damage detection.
The ultrasonic data of the steel rail is obtained by continuously transmitting ultrasonic signals by three groups of probe wheels in the one-time operation process of a large-scale steel rail flaw detection vehicle and reflecting according to foreign matters on the steel rail. Wherein, the probe wheel comprises a rail head, a rail waist, a rail bottom and a bottom wave disappearance, and the total number of the probe wheels is 20 groups of probes with different frequencies and directions. Because the probe continuously receives the transmitted ultrasonic signal in one operation, the ultrasonic signal of the steel rail can be regarded as a time sequence signal.
The original rail ultrasonic signals are mixed with each other due to the fact that waves coming out of the rail head, the rail web and the rail bottom exist, and different characteristics exist for detecting rail damage defects. Therefore, the invention separates and cuts the original rail ultrasonic echo data according to the ultrasonic probe type, namely different echo channels. Is recorded as Xcm={xc1,xc2,…,xcmIn which X iscmRepresents the time sequence of the wave emergence of the mth channel c, with m ∈ [0,19 ]],xctnAnd a signal echo point of the wave of the channel c at the nth time is shown. The initial complete rail ultrasonic wave output signal can be expressed as X ═ { X ═ Xc1,Xc2,…,Xcm}. According to different ultrasonic signals, the complete ultrasonic signal is subdivided into railhead ultrasonic time series signals XheadUltrasonic time series signal X of rail webwebAnd the ultrasonic time series signal X of the rail bottombase. And the ultrasonic wave has time characteristics according to the wave emergence of the steel rail, namely, the certain damage in the steel rail does not exist on a certain wave emergence, and the ultrasonic wave emergence which is mutually connected should exist in a certain time interval. Therefore, the invention adjusts the time scale of the time sequence, and the time scale can be divided into Tr=[2,22,…,2n]N is equal to R. That is, when n is 1, the time size data of 2 units is used as a time unit data of the present invention to form a group of data, and so on, that is, the original data is cut into different segments according to different time scales. According to different damage positions, n has different values. For example, at the railhead ultrasonic echo time series nmostThe maximum value is 11, namely the range of not more than one steel rail is met. Time series data X for single channel wavecmAnd fused ultrasound time-series data Xhead、Xweb、XbaseTime scale adjustment is performed.
And setting a sampling coefficient sigma to sample ultrasonic time sequences of the rail head, the rail web and the rail bottom, wherein the form of the rail damage defect in a time domain is continuous and tends to spread. Therefore, setting the sampling coefficient σ does not affect the high-frequency and low-frequency distributions of the time-series signal. Generally, σ can be taken as a space [1,2,4,8,16], and when σ is 1, it means that time-series sampling is not performed, and the original time series is kept unchanged; when σ -2 means that the time sequence is sampled at 2 times the sampling rate, and so on. The maximum sampling rate is set to 16 because an excessive sampling coefficient may cause a large amount of time series information to be lost, so that the time series signal loses the original time distribution. Therefore, each group of time series comprises ultrasonic time series information of a single channel and segmentation and fusion ultrasonic time series information of the rail head, the rail waist and the rail bottom, and 5 groups of new time series information can be generated. Different sizes, different sampling frequencies, and the generated time data sequence can provide more abundant features.
For multi-scale time series information, it is hoped that characteristics which are rich and can reflect real states of time series can be extracted for detecting the damage abnormality in the steel rail. In the present invention, the related features include a smooth feature of a time-series signal, a complex frequency domain feature, and the like. The features were chosen and described as follows:
in the ultrasonic data of the steel rail, scattered noise signals with high-frequency characteristics may be generated near the defect of normal wave-out damage due to the influence of factors such as hardware interference in the acquisition process, untimely probe adjustment in the acquisition process and the like. Is provided withWherein f issmRepresenting the smoothed ultrasound time series information; (X) X*,X*Representing ultrasound time of arbitrary channel and slice combinationsA sequence signal; x is the number ofjAn echo point representing the ultrasonic wave output when the time is j; l represents the step size of smoothing, i.e., how long the time range is set as a smoothing interval. Through smooth feature acquisition, the influence of sharp echo time sequence information can be reduced, and the damage showing the trend really is more obvious. According to different positions of the rail damage defects, setting different thresholds for l, such as the rail head part, and setting l to be the length of the whole rail; for the web portion l can be provided with the length of one screw hole.
The complex frequency domain features are also important features of the time series signals, and certain limiting conditions in the frequency domain can be converted into the complex frequency domain according to corresponding conditions, and the features of the signals or the systems are directly observed in the complex frequency domain. Because the ultrasonic signals of the steel rail belong to a discrete time sequence, Z transformation is adopted,Z∈Rxthe discrete time series are transformed into the complex frequency domain. Where Z is a complex variable, the representation may be in the form of Z ═ Z | ejω(ii) a x (n) represents a time series, and n represents a time; rxConvergence field of X (Z).
Further, on the basis of the multi-size time series information features, the feature dimension is further increased in order to make the feature scale of the input deep convolution neural network uniform. Based on multi-size time series information characteristics, the first layer is designed to use different convolution kernels kconvPerforming convolution operations on time series, e.g. here different convolution kernels kconvMay include underlying statistical calculations of convolution kernels such as local means, variances, etc. Normalization is realized as long as the output dimensionality of the convolution layer is ensured to be the same, and then the dimensionality reduction is carried out on the characteristics by using a pooling layer. This step can also enable amplification of multi-size time series features. And finally, inputting the normalized and fused features into a deep convolutional neural network, and finally connecting a softmax function to realize the classification of the time sequence.
In the ultrasonic time series information of the steel rail, because the wave of the flaw in the steel rail is continuous and has obvious flaw tendency, the characteristic of the presented time series can be obviously different, for example, the echo frequency of a certain channel is suddenly increased. And combining a method for adjusting the time scale, after the ultrasonic time sequence identification of all the steel rails is completed, reflecting that the damage abnormality occurs at the time t through the time sequence identification result, and finally determining the damage type according to the identification result of the deep convolutional neural network to obtain a final damage list.
On the basis of multi-scale time sequence data, higher-level smooth features, complex frequency domain features and the like are continuously designed, so that the influence of noise in the ultrasonic data of the steel rail on a damage result can be effectively avoided, and the time when the damage occurs can be accurately judged. The false alarm rate of damage identification can be effectively reduced.
Step S1, cutting and preprocessing the ultrasonic time sequence of the steel rail
For an original ultrasonic data signal X, the original signal X can be split into X according to channels according to the difference of the outgoing wave probes of the large-scale steel rail flaw detection vehiclecmFusing channel signals into railhead ultrasonic time sequence data X according to the properties of the wave outlet channelheadUltrasonic time series data X of rail webwebAnd ultrasonic time-series data X of the rail bottombase。
Adjusting railhead time series data XheadTime series data X of rail webwebTime series data X of rail bottombaseTime scale of (1) according to Tr=[2,22,…,2n]And (4) scale, representing the original time sequence into a data segment according to the new time scale.
Step S2, multi-scale division of ultrasonic time series of steel rails
On the basis of the sliced data, for each kind of data including single-channel time-series data and fused time-series data, the sliced data is sampled using a sampling rate σ ═ 1,2,4,8,16, respectively. And obtaining sampling data of different scales.
Step S3, designing multi-scale time series signal characteristics
And (5) aiming at the multi-size sampling data obtained in the step (S2), carrying out characteristic design on the ultrasonic time series signals of the single channel and the fused time series signals of the rail head, the rail waist and the rail bottom. The method extracts the smooth characteristic and the frequency domain characteristic of each group of ultrasonic time sequence information.
ByAnd calculating the smooth characteristic of each group of ultrasonic time series, namely multiplying the average value of the wave number of a certain section on the basis of a certain time series signal f (x), and outputting the average value as a smoothed time series. The smoothing step length l determines the length of the time series concerned when the smoothing feature is obtained, and the smoothing step lengths l with different lengths are combined to obtain the features of the time series under more receptive fields. The smoothing feature can remove the effects of sharp noise, eliminating the effects of noise that are short in duration.
ByConverting a signal in the time domain into a signal characteristic f in the complex frequency domainfr. Complex frequency domain feature ffrCharacterised by the convergence domain R of the given signalxDefinitively, when Z is 1, the above formula can be changed toIf the initial signal is defined as a single-sided signal, it is the signalThat is, when the time-series signal x (n) is finite, the response of the signal will not become infinite, and the signal is stable. The characteristics of the Z transformation, such as linear additivity, can be mixed into a new characteristic which can describe the original signal.
Step S4, amplification and normalization of different scale features
In the features obtained in step S3, since the initial data scale is different, the feature scale generated may also be different, and a special convolution kernel k is designed for each featureconvFurther processing the feature obtained in step S3 to generateThe feature graphs are consistent in scale. Convolution kernel k in the inventionconvConvolution kernels such as local mean, variance and the like are used. And connecting a maximum pooling layer to reduce the dimension of the data.
Step S5, designing, training and recognizing deep convolution neural network
And (4) inputting the multi-dimensional features generated in the step (S4) into the deep convolutional neural network in a hierarchical connection mode, wherein at the present stage, the wide deep convolutional neural networks are various, and the invention tries to use a VGG-Net network and a Res-Net network. Especially Res-Net network, can utilize the residual block to avoid the problem that the gradient disappears in the course of training.
For data annotation, since the data is already segmented at step S1 and the multi-scale features are generated according to step S2 and step S3, the data segment is annotated by manual annotation, and the label is the corresponding damage category.
Sending the data with the labels into a deep convolution neural network in a batch mode, automatically initializing according to the distribution characteristics of the data to generate a loss hyper-curved surface during initialization, and generating a cross entropy loss functionWherein y isiLabel representing a specimen, piAnd (4) representing the prediction probability of the sample, and calculating the loss value of the characteristic weight of each layer through continuous forward propagation. And reversely propagating the network once according to the loss value, and correcting the characteristic weight. And continuously carrying out training iteration, wherein the loss value can be iterated to the global optimum along the direction of the maximum gradient of the loss curved surface. When the loss value is small and stable, it indicates that the optimal loss value has been iterated. Finally, softmax is connected for identifying the defect type of the damage.
After the deep convolutional neural network model is trained, the new sample identification process is that the new sample is subjected to forward propagation once, the network weight is not calculated and adjusted through backward propagation, and the identification result is directly output by the softmax function.
And step S6, generating a rail damage defect list, and recording the identification result to generate the damage defect list.
The ultrasonic signal of the steel rail is regarded as a time sequence signal, and the flaw of the flaw in a certain time period can be accurately positioned through the adjustment of the time scale. And the multi-scale steel rail ultrasonic time series signal division is utilized, so that the damage detection rate can be obviously improved, and the missing report rate can be greatly reduced.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents and improvements made within the spirit and principle of the present invention are intended to be included within the scope of the present invention.
Claims (8)
1. A multi-scale steel rail damage defect detection method based on a time sequence is characterized by comprising the following steps:
s1, carrying out data segmentation and pretreatment according to the original track ultrasonic data of the echo channel of the ultrasonic probe type;
s2, setting a sampling coefficient to sample the preprocessed segmentation data to obtain sampling data of different scales;
s3, carrying out feature extraction and feature fusion on the collected rail damage data with different scales;
s4, inputting the fused features into a deep convolutional neural network to connect with a softmax function to classify the time series;
and S5, obtaining a rail damage list according to the recognition result of the convolutional neural network.
2. The method for detecting the multi-scale steel rail damage defect based on the time series according to claim 1, wherein the smooth feature and the frequency domain feature of each group of ultrasonic time series information are extracted in the step S3.
3. The method for detecting the flaw of the steel rail on the basis of the time series and the multiple scales according to the claim 2, wherein the influence of noise is reduced by acquiring the smooth feature in the ultrasonic data acquisition, and the method is characterized in thatWherein f issmRepresenting the smoothed ultrasound time series information; (X) X*,X*An ultrasonic time series signal representing an arbitrary combination of channels and slices; x is the number ofjAn echo point representing the ultrasonic wave output when the time is j; l denotes the step size of smoothing.
4. The method for detecting the multi-scale steel rail flaw based on the time series according to the claim 3, wherein when the frequency domain features are extracted, because the steel rail ultrasonic signals belong to the discrete time series, Z transformation is adopted,Z∈Rxthe discrete time series is transformed into the complex frequency domain, where Z is a complex variable, which may be represented in the form of Z ═ Z | ejω(ii) a x (n) represents a time series, and n represents a time; rxConvergence field of X (Z).
5. The method for detecting the multi-scale steel rail defect based on the time series according to claim 4, wherein the step S4 is characterized in that the step S4 includes the following steps of unifying the characteristic dimension of the input deep convolutional neural network and increasing the characteristic dimension:
s41, designing and using incompatible convolution kernel k in the first layerconvPerforming convolution operation on the time sequence;
and S42, reducing the dimension of the feature by using a pooling layer with the same output dimension of the convolution layer.
6. The method for detecting the multi-scale steel rail damage defect based on the time series according to claim 5, wherein the displayed data at the present stage of the data segmentation and preprocessing of the ultrasonic data of the steel rail in the step S1 is presented in an image mode and is divided into a rail head part, a rail waist part and a rail bottom part according to the time series according to different ultrasonic wave output information; the ultrasonic time-series data is divided by channels.
7. The method for detecting defects of a multi-scale steel rail according to claim 6, wherein the step S2 is performed by downsampling the original time series data using a downsampling coefficient σ to form multi-scale time series data, and the sliding smoothing is performed by using a sliding average operator to analyze the multi-scale time series data to obtain a sliding smoothing feature fsmObtaining complex frequency domain characteristics f of multi-scale time sequence by time-frequency transformationfr。
8. The method for detecting the multi-scale steel rail damage defect based on the time sequence according to claim 7, wherein the fusion features are used as input data and input into the deep convolutional neural network, and the weight of the deep convolutional neural network is continuously adjusted through forward and backward iteration to finally minimize the loss value.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111260216.8A CN114047259A (en) | 2021-10-28 | 2021-10-28 | Method for detecting multi-scale steel rail damage defects based on time sequence |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111260216.8A CN114047259A (en) | 2021-10-28 | 2021-10-28 | Method for detecting multi-scale steel rail damage defects based on time sequence |
Publications (1)
Publication Number | Publication Date |
---|---|
CN114047259A true CN114047259A (en) | 2022-02-15 |
Family
ID=80206186
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111260216.8A Pending CN114047259A (en) | 2021-10-28 | 2021-10-28 | Method for detecting multi-scale steel rail damage defects based on time sequence |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114047259A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115294376A (en) * | 2022-04-24 | 2022-11-04 | 西京学院 | Weld defect detection method based on fusion of ultrasonic waveform and ultrasonic image characteristics |
Citations (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2003322643A (en) * | 2002-04-30 | 2003-11-14 | Sumitomo Metal Ind Ltd | Quality inspection method in welded steel pipe welded section |
US20040193064A1 (en) * | 2000-05-30 | 2004-09-30 | Vladimir Shusterman | Multi-scale analysis and representation of physiological and health data |
JP2005351660A (en) * | 2004-06-08 | 2005-12-22 | Toshiba Corp | Ultrasonic flaw detection method and ultrasonic flaw detector |
US20080004527A1 (en) * | 2006-04-05 | 2008-01-03 | Coleman D Jackson | High-resolution ultrasound spectral and wavelet analysis of vascular tissue |
CN103175897A (en) * | 2013-03-13 | 2013-06-26 | 西南交通大学 | High-speed turnout damage recognition method based on vibration signal endpoint detection |
CN110217263A (en) * | 2019-05-16 | 2019-09-10 | 北京巨安永业科技有限公司 | A kind of double track four-wheel failure detector based on ultrasound |
CN110967409A (en) * | 2018-09-29 | 2020-04-07 | 东莞灵虎智能科技有限公司 | Method for extracting and identifying original data features based on multi-channel ultrasonic flaw detection |
CN111563455A (en) * | 2020-05-08 | 2020-08-21 | 南昌工程学院 | Damage identification method based on time series signal and compressed convolution neural network |
CN111637964A (en) * | 2020-05-12 | 2020-09-08 | 西南交通大学 | Rail corrugation identification method |
CN111855810A (en) * | 2020-07-20 | 2020-10-30 | 济南大学 | Rail foot damage identification method and system based on recurrent neural network |
CN112200225A (en) * | 2020-09-23 | 2021-01-08 | 西南交通大学 | Steel rail damage B display image identification method based on deep convolutional neural network |
WO2021068848A1 (en) * | 2019-10-09 | 2021-04-15 | 山东大学 | Tunnel structure disease multi-scale measurement and intelligent diagnosis system and method |
CN112903813A (en) * | 2021-01-14 | 2021-06-04 | 北京安铁软件技术有限公司 | Railway track ultrasonic automatic flaw detection method |
CN112946081A (en) * | 2021-02-09 | 2021-06-11 | 武汉大学 | Ultrasonic imaging method based on defect multi-feature intelligent extraction and fusion |
US20210232813A1 (en) * | 2020-01-23 | 2021-07-29 | Tongji University | Person re-identification method combining reverse attention and multi-scale deep supervision |
CN113435404A (en) * | 2021-07-14 | 2021-09-24 | 深圳市比一比网络科技有限公司 | Safe auxiliary driving method and system for shunting electric bus based on radar and image detection |
CN113533516A (en) * | 2021-07-12 | 2021-10-22 | 北京交通大学 | Damage detection method in steel rail based on Lamb wave time field method |
-
2021
- 2021-10-28 CN CN202111260216.8A patent/CN114047259A/en active Pending
Patent Citations (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20040193064A1 (en) * | 2000-05-30 | 2004-09-30 | Vladimir Shusterman | Multi-scale analysis and representation of physiological and health data |
JP2003322643A (en) * | 2002-04-30 | 2003-11-14 | Sumitomo Metal Ind Ltd | Quality inspection method in welded steel pipe welded section |
JP2005351660A (en) * | 2004-06-08 | 2005-12-22 | Toshiba Corp | Ultrasonic flaw detection method and ultrasonic flaw detector |
US20080004527A1 (en) * | 2006-04-05 | 2008-01-03 | Coleman D Jackson | High-resolution ultrasound spectral and wavelet analysis of vascular tissue |
CN103175897A (en) * | 2013-03-13 | 2013-06-26 | 西南交通大学 | High-speed turnout damage recognition method based on vibration signal endpoint detection |
CN110967409A (en) * | 2018-09-29 | 2020-04-07 | 东莞灵虎智能科技有限公司 | Method for extracting and identifying original data features based on multi-channel ultrasonic flaw detection |
CN110217263A (en) * | 2019-05-16 | 2019-09-10 | 北京巨安永业科技有限公司 | A kind of double track four-wheel failure detector based on ultrasound |
WO2021068848A1 (en) * | 2019-10-09 | 2021-04-15 | 山东大学 | Tunnel structure disease multi-scale measurement and intelligent diagnosis system and method |
US20210232813A1 (en) * | 2020-01-23 | 2021-07-29 | Tongji University | Person re-identification method combining reverse attention and multi-scale deep supervision |
CN111563455A (en) * | 2020-05-08 | 2020-08-21 | 南昌工程学院 | Damage identification method based on time series signal and compressed convolution neural network |
CN111637964A (en) * | 2020-05-12 | 2020-09-08 | 西南交通大学 | Rail corrugation identification method |
CN111855810A (en) * | 2020-07-20 | 2020-10-30 | 济南大学 | Rail foot damage identification method and system based on recurrent neural network |
CN112200225A (en) * | 2020-09-23 | 2021-01-08 | 西南交通大学 | Steel rail damage B display image identification method based on deep convolutional neural network |
CN112903813A (en) * | 2021-01-14 | 2021-06-04 | 北京安铁软件技术有限公司 | Railway track ultrasonic automatic flaw detection method |
CN112946081A (en) * | 2021-02-09 | 2021-06-11 | 武汉大学 | Ultrasonic imaging method based on defect multi-feature intelligent extraction and fusion |
CN113533516A (en) * | 2021-07-12 | 2021-10-22 | 北京交通大学 | Damage detection method in steel rail based on Lamb wave time field method |
CN113435404A (en) * | 2021-07-14 | 2021-09-24 | 深圳市比一比网络科技有限公司 | Safe auxiliary driving method and system for shunting electric bus based on radar and image detection |
Non-Patent Citations (3)
Title |
---|
刘恩东: "基于BP神经网络的管道缺陷识别", 《电子技术与软件工程》, no. 02 * |
崔万里 等: "EMD-ISOMAP高速列车小幅蛇行异常特征提取", 《中国测试》, no. 12 * |
田英 等: "基于EWT多尺度模糊熵和PSO-ELM列车轮损伤检测", 《组合机床与自动化加工技术》, no. 06, 20 June 2020 (2020-06-20) * |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115294376A (en) * | 2022-04-24 | 2022-11-04 | 西京学院 | Weld defect detection method based on fusion of ultrasonic waveform and ultrasonic image characteristics |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108804740B (en) | Long-distance pipeline pressure monitoring method based on integrated improved ICA-KRR algorithm | |
CN110405537B (en) | Method for establishing guide rail precision prediction model based on deep learning | |
CN103546853A (en) | Speaker abnormal sound detecting method based on short-time Fourier transformation | |
Lin et al. | Comparison of passive acoustic soniferous fish monitoring with supervised and unsupervised approaches | |
CN110967409A (en) | Method for extracting and identifying original data features based on multi-channel ultrasonic flaw detection | |
CN115239632B (en) | Rail surface damage detection method integrating inspection image and ultrasonic image | |
CN108680245A (en) | Whale globefish class Click classes are called and traditional Sonar Signal sorting technique and device | |
CN112674780A (en) | Automatic atrial fibrillation signal detection method in electrocardiogram abnormal signals | |
CN114799610B (en) | Welding quality real-time detection method and system based on inverse Fourier transform and self-encoder | |
CN114047259A (en) | Method for detecting multi-scale steel rail damage defects based on time sequence | |
CN115824519B (en) | Comprehensive diagnosis method for valve leakage faults based on multi-sensor information fusion | |
CN111695465B (en) | Pipe network fault diagnosis and positioning method and system based on pressure wave mode identification | |
CN113140229A (en) | Sound detection method based on neural network, industrial acoustic detection system and method | |
CN111353640B (en) | Method for constructing wind speed prediction model by combination method | |
CN113640380A (en) | Multi-stage classification method and system for rail damage detection | |
CN114227382B (en) | Cutter damage monitoring system and method based on novel capsule network | |
CN116935892A (en) | Industrial valve anomaly detection method based on audio key feature dynamic aggregation | |
CN117332293A (en) | Truck overload detection method based on sound Mel frequency characteristics | |
CN110346032A (en) | A kind of Φ-OTDR vibration signal end-point detecting method combined based on constant false alarm with zero-crossing rate | |
CN113720910B (en) | Intelligent rail defect detection method and system based on ultrasonic signals | |
CN109884483A (en) | Insulating tube type busbar shelf depreciation acoustics on-line monitoring method and device | |
CN115223044A (en) | End-to-end three-dimensional ground penetrating radar target identification method and system based on deep learning | |
CN114093385A (en) | Unmanned aerial vehicle detection method and device | |
CN114881938A (en) | Grain size detection method and system based on wavelet analysis and neural network | |
Kartik et al. | A Deep learning based system to predict the noise (disturbance) in audio files |
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 |