Data-fused steel rail welding seam disease detection method and device and medium

Abstract

The invention relates to the technical field of data fusion, and provides a data fusion steel rail welding seam disease detection method and device and a medium, and the method comprises the steps: collecting the vibration acceleration of an axle box part and the wheel-rail force data of a wheel pair part; performing initial alignment on a first waveform peak position corresponding to the vibration acceleration and the wheel-rail force data; calculating the similarity between the initially aligned wheel-rail force data and the vibration acceleration, and performing fine alignment on a waveform window corresponding to the wheel-rail force data and a waveform window corresponding to the vibration acceleration according to the similarity; according to the accurately aligned wheel-rail force data and vibration acceleration, extracting a steel rail welding seam position, and extracting target vibration acceleration data and target wheel-rail force data corresponding to the steel rail welding seam position; and performing channel fusion on the target vibration acceleration data and the target wheel-rail force data, and performing steel rail welding seam disease detection on the fused data by using a neural network model to obtain the disease degree of the steel rail welding seam. According to the invention, the accuracy of steel rail welding seam disease detection can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data fusion, and in particular to a data fusion rail weld defect detection method, equipment and medium. Background Art

[0002] With the development trend of increasing mileage, speed and density of high-speed railway, the railway infrastructure is gradually changing from new to old, and the subsequent safety maintenance will be the top priority of future railway management. Rail welds are the weak structure of rails. Defects generated during the welding process are prone to develop into fatigue cracks, and the heat-affected zone is prone to rail breakage, which will seriously affect the normal operation and safety of trains. Therefore, the detection of rail weld defects has become increasingly important.

[0003] Existing rail weld defect detection methods mainly rely on signal processing and machine learning technology, such as training algorithms and models to achieve automatic analysis and judgment of weld images and data. In practical applications, machine learning only relies on a single data source such as axle box acceleration, which may lead to limited detection accuracy, reduced reliability, and failure to provide more detailed defect information, resulting in low accuracy in rail weld defect detection. Summary of the invention

[0004] In view of the above problems, embodiments of the present invention provide a data fusion rail weld defect detection method, device and medium.

[0005] In a first aspect, an embodiment of the present invention provides a data fusion rail weld defect detection method, comprising:

[0006] Collect vibration acceleration at the axle box and wheel-rail force data at the wheelset;

[0007] Initially aligning the first waveform peak position corresponding to the vibration acceleration with the first waveform peak position corresponding to the wheel-rail force data;

[0008] Calculating the similarity between the wheel-rail force data after initial alignment and the vibration acceleration after initial alignment according to a preset sampling window, and precisely aligning the waveform window corresponding to the wheel-rail force data after initial alignment with the waveform window corresponding to the vibration acceleration after initial alignment according to the similarity;

[0009] Extracting the rail weld position according to the wheel-rail force data after precise alignment and the vibration acceleration after precise alignment, and extracting the target vibration acceleration data and target wheel-rail force data corresponding to the rail weld position;

[0010] The target vibration acceleration data and the target wheel-rail force data are channel-fused to obtain fused data, and a pre-built neural network model is used to perform rail weld defect detection on the fused data to obtain the degree of rail weld defect.

[0011] According to an embodiment of the present invention, before the initial alignment of the first waveform peak position corresponding to the vibration acceleration with the first waveform peak position corresponding to the wheel-rail force data, the method further includes:

[0012] The vibration acceleration and the wheel-rail force data are screened according to a preset collection distance to obtain the vibration acceleration and the wheel-rail force data at the collection distance;

[0013] Downsampling the vibration acceleration and wheel-rail force data at the acquisition distance respectively, and configuring overlapping data and missing data for the downsampled vibration acceleration and wheel-rail force data respectively;

[0014] The first waveform peak position corresponding to the configured vibration acceleration is extracted; and the first waveform peak position corresponding to the configured wheel-rail force data is extracted.

[0015] According to an embodiment of the present invention, the step of initially aligning the first waveform peak position corresponding to the vibration acceleration with the first waveform peak position corresponding to the wheel-rail force data includes:

[0016] The first waveform peak position corresponding to the wheel-rail force data is used as the reference peak position;

[0017] Adjusting the first waveform peak position corresponding to the vibration acceleration according to the reference peak position;

[0018] Aligning the first waveform peak position corresponding to the adjusted vibration acceleration with the first waveform peak position corresponding to the wheel-rail force data;

[0019] The vibration acceleration and the wheel-rail force data are aligned according to the first waveform peak position after alignment.

[0020] According to an embodiment of the present invention, the calculating the similarity between the wheel-rail force data after the initial alignment and the vibration acceleration after the initial alignment according to a preset sampling window includes:

[0021] Using the wheel-rail force data after initial alignment as reference data, and obtaining a reference sampling frequency corresponding to the reference data;

[0022] Downsampling the aligned acceleration data according to the reference sampling frequency;

[0023] Splitting the aligned wheel-rail force data window according to the number of sampling points in the sampling window to obtain a reference window;

[0024] Dividing the window of the sampled acceleration data according to the number of sampling points in the sampling window to obtain a matching window;

[0025] The similarities between the reference window and the matching window are calculated one by one.

[0026] According to an embodiment of the present invention, the step of finely aligning the waveform window corresponding to the wheel-rail force data after the initial alignment with the waveform window corresponding to the vibration acceleration after the initial alignment according to the similarity comprises:

[0027] The waveform window corresponding to the wheel-rail force data after the initial alignment is used as the reference window, and the waveform window corresponding to the vibration acceleration after the initial alignment is used as the matching window;

[0028] Select the matching window with the highest similarity corresponding to each benchmark window;

[0029] Determining the window position of the matching window using a preset maximum value index function;

[0030] Performing linear interpolation correction on the matching window according to the window position to obtain an acceleration correction window corresponding to the matching window;

[0031] The reference window is aligned with the acceleration correction window.

[0032] According to an embodiment of the present invention, the method of extracting the rail weld position according to the wheel-rail force data after precise alignment and the vibration acceleration after precise alignment includes:

[0033] Extract the wheel-rail force data after fine alignment and the convex waveform corresponding to the vibration acceleration after fine alignment;

[0034] The position corresponding to the raised waveform is used as the rail weld position.

[0035] According to an embodiment of the present invention, the channel fusion of the target vibration acceleration data and the target wheel-rail force data to obtain fused data includes:

[0036] Extracting a first channel position of target vibration acceleration data in a preset first channel;

[0037] Extracting the second channel position of the target wheel-rail force data in the preset second channel;

[0038] The first channel position and the second channel position are aligned, and the target vibration acceleration data and the target wheel-rail force data are superimposed according to the aligned channel positions to obtain fused data.

[0039] According to an embodiment of the present invention, the use of a pre-built neural network model to perform rail weld defect detection on the fused data to obtain the degree of rail weld defect includes:

[0040] Performing a convolution operation on the fused data using a convolution kernel in the neural network model;

[0041] Inputting the convolved fused data into the residual module in the neural network model;

[0042] Perform global averaging on the residual fusion data;

[0043] The classification layer in the neural network model is used to classify the damage degree of the globally averaged fusion data to obtain the damage degree of the rail weld.

[0044] In a second aspect, an embodiment of the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the multi-fuel cell system fan control method described in the above aspect.

[0045] In a third aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the rail weld defect detection method using data fusion described in the above aspects.

[0046] Compared with the prior art, the above technical solution of the present invention has the following beneficial effects:

[0047] The embodiments of the present invention align the vibration acceleration data collected by the sensor installed at the axle box and the wheel-rail force data collected by the strain gauges on the wheelset. First, an initial alignment is performed based on the first peak position of the welded joint, and second, a fine alignment is performed based on the correlation between the vibration acceleration and the wheel-rail force. The position of the rail weld is extracted, and the vibration acceleration and wheel-rail force data at the weld position are intercepted and then input into a neural network based on a deep learning-based "end-to-end" recognition algorithm. Finally, the health level is output to make a decision. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0049] Figure 1 A flow chart showing the detection of rail weld defects based on a vehicle-mounted method according to the first embodiment of the present invention is shown;

[0050] Figure 2 A working flow chart showing a rail weld defect detection method using data fusion according to a second embodiment of the present invention is shown;

[0051] Figure 3 A schematic diagram showing simulation data of vibration acceleration and wheel-rail force data according to the second embodiment of the present invention;

[0052] Figure 4 A schematic diagram showing simulation data after downsampling of vibration acceleration and wheel-rail force data according to the second embodiment of the present invention;

[0053] Figure 5 A schematic diagram showing simulation data after configuration data overlap and loss according to the second embodiment of the present invention;

[0054] Figure 6 A schematic diagram showing the alignment of data in the first stage of the second embodiment of the present invention is shown;

[0055] Figure 7 A schematic diagram showing linear interpolation correction according to Embodiment 2 of the present invention is shown;

[0056] Figure 8 A schematic diagram showing the alignment data in the second stage of the second embodiment of the present invention is shown;

[0057] Fig. 9 A schematic diagram showing the structure of a neural network model according to the second embodiment of the present invention is shown;

[0058] Fig.10a A schematic diagram showing a change in the first accuracy rate of the third embodiment of the present invention;

[0059] Fig.10b A schematic diagram showing the loss change of the first model of the third embodiment of the present invention;

[0060] Fig.11a A schematic diagram showing a change in the second accuracy rate of the third embodiment of the present invention;

[0061] Fig.11b A schematic diagram showing the loss change of the second model of the third embodiment of the present invention;

[0062] Fig.12 A schematic diagram showing the detection accuracy after network training before and after data fusion according to the third embodiment of the present invention;

[0063] Fig.13 A schematic diagram of the composition structure of an electronic device for implementing the rail weld defect detection method using data fusion according to the fourth embodiment of the present invention is shown. DETAILED DESCRIPTION

[0064] The present disclosure is further described below in conjunction with the embodiments shown in the accompanying drawings.

[0065] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0066] The present invention proposes a data fusion rail weld defect detection method based on data fusion technology. Based on deep learning theory and combined with multi-category classification methods, a 1DResnet neural network is built to realize the detection and identification of rail weld joint defects. Compared with traditional methods, the rail weld defect detection technology based on data fusion is more efficient and reduces human subjectivity. It has great potential and application prospects in the application of rail weld defect detection.

[0067] Embodiment 1

[0068] like Figure 1 As shown in the figure, it is a flow chart of rail weld defect detection based on vehicle-mounted method, that is, the rail weld defect detection is realized by processing and analyzing the vibration acceleration data collected by the sensor installed in the axle box and the wheel-rail force data collected by the strain gauge on the wheelset. First, the data is collected and then aligned. There are two stages here. One is the initial alignment based on the first peak position of the welded joint, and the other is the fine alignment based on the correlation between the vibration acceleration and the wheel-rail force. Then, the position of the rail weld is extracted, and the vibration acceleration and wheel-rail force data of the weld position are intercepted and input into the neural network of the "end-to-end" recognition algorithm based on deep learning. Finally, the health level is output to make a decision.

[0069] Embodiment 2

[0070] like Figure 2 As shown, the present invention proposes a data fusion rail weld defect detection method, comprising the following steps:

[0071] S1. Collect the vibration acceleration of the axle box and the wheel-rail force data of the wheelset.

[0072] In the embodiment of the present invention, the vibration acceleration data of the axle box part is collected by the sensor installed at the axle box part, and the wheel-rail force data is collected by the strain gauge on the wheelset, wherein the vibration acceleration of the axle box part refers to the acceleration related to the vibration frequency, which is used to describe the vibration intensity or vibration level at a specific frequency; the wheel-rail force data on the wheelset part generally refers to the measurement data of the contact force or interaction force between the track and the vehicle wheelset, which is used to evaluate the safety of train operation, the wear condition of the track facilities and the interaction between the wheelset and the track. The wheel-rail force data includes the longitudinal force between the wheel and the rail during the acceleration, deceleration and braking of the train, the lateral force on the wheelset when the track passes through a curve, and the vertical force between the wheel and the rail, that is, the pressure and support force between the wheelset and the track and the stress distribution of the wheel-rail contact surface.

[0073] Furthermore, the axle box acceleration and wheel-rail force are two high-frequency sampling data that can reflect the short-wave track unevenness information and are interrelated. Therefore, these two data sources are selected to detect rail weld defects. However, since it is difficult for the two sets of data sources to sample at the same time and frequency, the overall waveform will be offset, and the acquisition system and frequency are different, resulting in stretching or compression inside the collected data, resulting in mileage errors between the two sets of data. If the error is ignored and the original data is corrected, it will lead to the problem of too much calculation. Due to the lack of curve information, the original waveform is also difficult to show the overall relationship between the two data, so that it is impossible to establish a reference benchmark when evaluating track smoothness. Therefore, data alignment is required before data fusion.

[0074] S2. Preliminarily aligning the first waveform peak position corresponding to the vibration acceleration with the first waveform peak position corresponding to the wheel-rail force data.

[0075] In the embodiment of the present invention, the initial alignment refers to the first alignment, that is, the initial alignment is performed based on the first peak position of the welding joint to correct the mileage error. Before the initial alignment is performed, the first peak position needs to be obtained.

[0076] In the embodiment of the present invention, before the first waveform peak position corresponding to the vibration acceleration is initially aligned with the first waveform peak position corresponding to the wheel-rail force data, the method further includes:

[0077] The vibration acceleration and the wheel-rail force data are screened according to a preset collection distance to obtain the vibration acceleration and the wheel-rail force data at the collection distance;

[0078] Downsampling the vibration acceleration and wheel-rail force data at the acquisition distance respectively, and configuring overlapping data and missing data for the downsampled vibration acceleration and wheel-rail force data respectively;

[0079] The first waveform peak position corresponding to the configured vibration acceleration is extracted; and the first waveform peak position corresponding to the configured wheel-rail force data is extracted.

[0080] In detail, based on the vibration acceleration and wheel-rail force data, the data at the preset collection distance needs to be simulated, such as Figure 3 As shown in FIG. 1 , it is a schematic diagram of the simulation data of vibration acceleration and wheel-rail force data. The vibration acceleration and wheel-rail force data of 200 m are simulated with a sampling frequency of 10 kHz, and the waveform diagram corresponding to the acceleration data and the waveform diagram corresponding to the wheel-rail force data at a distance of 200 meters are obtained; the simulation data based on the vibration acceleration and wheel-rail force data needs to be downsampled and the data overlap and missing are set. The sampling frequency of the acceleration data is set to 5 kHz, and the sampling frequency of the wheel-rail force is set to 2 kHz. The downsampling is used to reduce the amount of data while maintaining the key features and trends of the data, such as Figure 4 As shown, it is a schematic diagram of the down-sampled simulation data corresponding to the vibration acceleration and wheel-rail force data, where the vibration acceleration data is the simulation data down-sampled under the condition that the sampling frequency is set to 5kHz, wherein the first peak position coordinates corresponding to the acceleration data are X=13.29, Y=25.8, and the wheel-rail force data is the simulation data down-sampled under the condition that the sampling frequency is set to 2kHz, wherein the first peak position coordinates corresponding to the wheel-rail force data are X=15.42, Y=149.8.

[0081] Specifically, the original vibration acceleration and wheel-rail force data have no missing data and overlapping data. In order to simulate the real data situation, data overlap or data missing may occur. Therefore, data overlap and missing data should be configured on the downsampled data, such as Figure 5 As shown in FIG. 1 , it is a schematic diagram of simulation data after configuring data overlap and missing. For example, in the acceleration data waveform diagram, there are missing data and overlapping data. Figure 5 The first box represents missing data, the second box represents overlapping data, and Figure 5 The corresponding first waveform peak position of the acceleration is X=13.29, Y=25.8, and the first waveform peak position of the wheel-rail force is X=15.42, Y=149.8.

[0082] Furthermore, after simulating the data and configuring the data overlaps and missing, the first stage of data alignment is performed, and the first stage of alignment is to perform initial alignment based on the first peak position of the weld joint.

[0083] In the embodiment of the present invention, the step of initially aligning the first waveform peak position corresponding to the vibration acceleration with the first waveform peak position corresponding to the wheel-rail force data includes:

[0084] The first waveform peak position corresponding to the wheel-rail force data is used as the reference peak position;

[0085] Adjusting the first waveform peak position corresponding to the vibration acceleration according to the reference peak position;

[0086] Aligning the first waveform peak position corresponding to the adjusted vibration acceleration with the first waveform peak position corresponding to the wheel-rail force data;

[0087] The vibration acceleration and the wheel-rail force data are aligned according to the first waveform peak position after alignment.

[0088] In detail, the first waveform peak position X=15.42, Y=149.8 corresponding to the wheel-rail force data is used as the reference peak position, and the first waveform peak position corresponding to the vibration acceleration is adjusted according to the reference peak position of the wheel-rail force, that is, the first waveform peak position X=13.29, Y=25.8 corresponding to the vibration acceleration is adjusted to X=15.46, Y=25.27, and the first waveform peak position X=15.46, Y=25.27 corresponding to the adjusted vibration acceleration is aligned with the first waveform peak position X=15.42, Y=149.8 corresponding to the wheel-rail force data in the first stage, as shown in FIG. Figure 6 As shown, it is a schematic diagram of the first stage alignment data, and the first waveform peak position X=15.46, Y=25.27 corresponding to the vibration acceleration is aligned with the first waveform peak position X=15.42, Y=149.8 corresponding to the wheel-rail force data, and only the position of the first waveform peak is aligned.

[0089] Furthermore, through the first stage of alignment, the overall misalignment of the original data waveform is eliminated, but the problem of local uneven distribution of the data waveform is not solved. Therefore, it is necessary to perform a second stage of alignment, that is, to perform a fine alignment based on the correlation between the vibration acceleration and the wheel-rail force data.

[0090] S3. Calculate the similarity between the wheel-rail force data after initial alignment and the vibration acceleration after initial alignment according to a preset sampling window, and precisely align the waveform window corresponding to the wheel-rail force data after initial alignment with the waveform window corresponding to the vibration acceleration after initial alignment according to the similarity.

[0091] In the embodiment of the present invention, the similarity refers to the matching degree between the wheel-rail force data and the vibration acceleration data calculated in the sampling window.

[0092] In the embodiment of the present invention, the calculating the similarity between the wheel-rail force data after the initial alignment and the vibration acceleration after the initial alignment according to the preset sampling window includes:

[0093] Using the wheel-rail force data after initial alignment as reference data, and obtaining a reference sampling frequency corresponding to the reference data;

[0094] Downsampling the aligned acceleration data according to the reference sampling frequency;

[0095] Splitting the aligned wheel-rail force data window according to the number of sampling points in the sampling window to obtain a reference window;

[0096] Dividing the window of the sampled acceleration data according to the number of sampling points in the sampling window to obtain a matching window;

[0097] The similarities between the reference window and the matching window are calculated one by one.

[0098] In detail, when calculating the similarity of two waveforms, it is necessary to ensure that the number of sampling points of the two data in the window length is consistent. Therefore, the acceleration data is downsampled with the sampling frequency of the wheel-rail force data after initial alignment as the benchmark data, and the window length is set to 5000 sampling points. The wheel-rail force data is used as the benchmark window for window division, and the acceleration is used as the matching window to match the benchmark window. The step size between matching data windows is 1 sampling point, where downsampling is used to reduce the sampling rate or sampling frequency of the data, which involves extracting a part of the data points from the high-frequency signal to create a low-frequency equivalent signal, and the number of data sampling points in the benchmark window is consistent with the number of data sampling points in the matching window.

[0099] Specifically, the Pearson correlation coefficient is used to calculate the similarity of the entire acceleration matching window based on the wheel-rail force data window. For each wheel-rail force data window, the Pearson correlation coefficient between it and all possible windows in the acceleration signal is calculated. The Pearson correlation coefficient measures the linear correlation between two variables, and its value range is -1 (complete negative correlation) to +1 (complete positive correlation).

[0100] Furthermore, after the similarity is determined, the reference window needs to be aligned with the matching window, and then subsequent processing is performed based on the aligned windows to solve the problem of local uneven distribution of the data waveform.

[0101] In the embodiment of the present invention, the fine alignment refers to the second stage alignment, that is, the fine alignment is performed based on the correlation between the vibration acceleration and the wheel-rail force data.

[0102] In the embodiment of the present invention, the step of finely aligning the waveform window corresponding to the wheel-rail force data after the initial alignment with the waveform window corresponding to the vibration acceleration after the initial alignment according to the similarity includes:

[0103] The waveform window corresponding to the wheel-rail force data after the initial alignment is used as the reference window, and the waveform window corresponding to the vibration acceleration after the initial alignment is used as the matching window;

[0104] Select the matching window with the highest similarity corresponding to each benchmark window;

[0105] Determining the window position of the matching window using a preset maximum value index function;

[0106] Performing linear interpolation correction on the matching window according to the window position to obtain an acceleration correction window corresponding to the matching window;

[0107] The reference window is aligned with the acceleration correction window.

[0108] In detail, the waveform window corresponding to the wheel-rail force data after the initial alignment is used as the reference window, and the waveform window corresponding to the vibration acceleration after the initial alignment is used as the matching window, and the maximum value index function is used to determine the position of the acceleration window with the highest similarity to each reference window.

[0109] Specifically, after completing the matching, it can be found that the acceleration window path that best matches the wheel-rail force data is not continuous, indicating that there are overlapping or missing parts in the waveforms of adjacent acceleration windows. If the overlapping and missing parts of the segments are ignored, directly splicing the matching window data will cause repetition and distortion of the axle box acceleration waveform. Therefore, before data splicing, the missing and overlapping parts should be taken into account, and the acceleration data window should be interpolated and stretched, and then the corrected acceleration window should be aligned with the reference window, such as Figure 7 As shown in the figure, it is a schematic diagram of linear interpolation correction, with the wheel-rail force data as the reference window and the acceleration data as the matching window. After the matching is completed, it can be found that there is a segment missing in matching A3, and there is a segment overlap and segment missing in matching A4, then window correction is required, then data splicing, waveform compression, and waveform stretching operations are performed to obtain the corrected acceleration data, and then the corrected acceleration data is aligned with the reference window.

[0110] For example, Figure 8 As shown in the figure, it is a schematic diagram of the second stage alignment data. The wheel-rail force data is used as the reference data to match the home acceleration data. During the matching process, overlaps and omissions will occur. For example, in the first waveform of acceleration, data overlaps occur in 10-40 and 30-60, and data omissions occur between 80-90. Therefore, it is necessary to perform linear interpolation correction on the acceleration data and then align the corrected acceleration data with the wheel-rail force data.

[0111] Furthermore, the position of the rail weld can be accurately identified after the two-stage alignment, and then the position of the rail weld can be analyzed to obtain the degree of damage of the rail weld.

[0112] S4. Extracting the rail weld position according to the wheel-rail force data after precise alignment and the vibration acceleration after precise alignment, and extracting the target vibration acceleration data and target wheel-rail force data corresponding to the rail weld position.

[0113] In the embodiment of the present invention, the rail weld position refers to the connection between two rail sections. When laying rails, they are usually composed of several shorter rail sections. In order to keep the railway line flat and ensure the safe operation of trains, these rails need to be welded at the connection.

[0114] In the embodiment of the present invention, the method of extracting the rail weld position according to the wheel-rail force data after precise alignment and the vibration acceleration after precise alignment includes:

[0115] Extract the wheel-rail force data after fine alignment and the convex waveform corresponding to the vibration acceleration after fine alignment;

[0116] The position corresponding to the raised waveform is used as the rail weld position.

[0117] In detail, both the wheel-rail force data after fine alignment and the vibration acceleration data after fine alignment have prominent positions. In the wheel-rail force data after fine alignment, there are usually bulges or mutations caused by the rail welds. These bulges usually reflect the changes in the wheel-rail contact surface or the vibration sensor position at the welds. Similarly, in the vibration acceleration data, changes in vibration modes or peaks of specific frequencies may occur at the welds. By using signal processing or pattern recognition technology, the bulge waveforms in the wheel-rail force data and vibration acceleration data can be identified. These bulges usually appear as mutations or extreme points in the data. Once the bulge waveform is identified, its position can be directly corresponded to the position of the rail weld, and can be located according to the track position information after fine alignment to ensure that the specific position of the weld is accurately marked.

[0118] Specifically, the position of the rail weld is accurately extracted after two-stage data alignment, and the vibration acceleration data and wheel-rail force data of the weld position are intercepted, namely, the target vibration acceleration data and the target wheel-rail force data.

[0119] Furthermore, multi-source data fusion technology can integrate incomplete information collected from multiple different data sources, and perform corresponding processing and fusion processing, so that the advantages of different data can complement each other and finally obtain a data result with decision-making significance, thereby weakening the uncertainty components existing in the data source, helping users to obtain effective fusion judgments and accurate comprehensive measurements, so that it is easier to make reasonable judgments and decisions, thereby fusing the target vibration acceleration data and the target wheel-rail force data.

[0120] S5. Channel-fusing the target vibration acceleration data and the target wheel-rail force data to obtain fused data, and using a pre-built neural network model to perform rail weld defect detection on the fused data to obtain the degree of rail weld defect.

[0121] In the embodiment of the present invention, the fused data refers to the data after the target vibration acceleration data and the target wheel-rail force data are fused. By collecting the two data sources of axle box acceleration and wheel-rail force data to detect rail weld defects, the detection accuracy is improved, the defect information can be more accurately identified, and the user can be helped to make decisions.

[0122] In the embodiment of the present invention, the target vibration acceleration data and the target wheel-rail force data are channel-fused to obtain fused data, including:

[0123] Extracting a first channel position of target vibration acceleration data in a preset first channel;

[0124] Extracting the second channel position of the target wheel-rail force data in the preset second channel;

[0125] The first channel position and the second channel position are aligned, and the target vibration acceleration data and the target wheel-rail force data are superimposed according to the aligned channel positions to obtain fused data.

[0126] In detail, the preset first channel refers to the channel corresponding to the target vibration acceleration data, and the preset second channel refers to the channel corresponding to the target wheel-rail force data. The position of the target vibration acceleration data in the channel and the position of the target wheel-rail force data in the channel are identified to ensure that the first channel position and the second channel position are aligned in time or space so that they can correspond correctly. Once the alignment is completed, the vibration acceleration data at the first channel position and the wheel-rail force data at the second channel position are superimposed to achieve channel fusion and obtain fused data.

[0127] Furthermore, the fused data can be analyzed based on the pre-built neural network model to obtain the degree of damage at the rail weld location based on the fused data.

[0128] In the embodiment of the present invention, the degree of damage of the rail weld means that if the weld is 0-0.1mm, it is healthy; if the weld is 0.1-0.3mm, it is slightly damaged; if the weld is less than 0mm or greater than 0.3mm, it is seriously damaged.

[0129] In the embodiment of the present invention, the use of a pre-built neural network model to perform rail weld defect detection on the fused data to obtain the degree of rail weld defect includes:

[0130] Performing a convolution operation on the fused data using a convolution kernel in the neural network model;

[0131] Inputting the convolved fused data into the residual module in the neural network model;

[0132] Perform global averaging on the residual fusion data;

[0133] The classification layer in the neural network model is used to classify the damage degree of the globally averaged fusion data to obtain the damage degree of the rail weld.

[0134] In detail, the wheel vibration is a one-dimensional time series, and the use of a typical two-dimensional convolutional neural network model will destroy the one-dimensional sequence data structure. Therefore, a 1DResnet neural network was constructed, which is mainly composed of residual modules. In deep neural networks, due to the layer-by-layer multiplication of gradients during the back-propagation process, when the number of network layers is large, the gradient will become very small or even disappear, causing the network update to become very slow or even stagnant. The residual block structure introduces a jump connection to directly add the network output of the previous layer to the network input of the current layer, thereby transferring information along the depth of the network. This idea comes from the concept of residual learning, which is to learn more effective mappings by fitting residuals, such as Fig. 9 As shown in the figure, it is a structural diagram of the neural network model. The 1DResnet neural network consists of an input layer, a convolutional layer, a ResNet module, a pooling layer and an output layer, and the ResNet module consists of 5 residual modules.

[0135] Specifically, the position of the rail weld is accurately extracted after two-stage data alignment, and the vibration acceleration data and wheel-rail force data of the weld position are intercepted. The initial data of the two sources are channel-fused and used as the input of the network. After entering the network, the data is first convolved with a small convolution kernel and then input into the subsequent residual module. After passing through the first layer of the residual network, the feature length of the data remains unchanged, and the feature data is halved each time it passes through a residual layer. After passing through 5 residual modules, global average pooling is performed, and finally the softmax layer is used to detect and identify rail weld joint defects and obtain the corresponding defect degree of the rail weld.

[0136] Embodiment 3

[0137] In order to understand the present invention more clearly, the reliability of the 1DResnet neural network model constructed by the embodiment of the present invention is further explained through a second embodiment below.

[0138] In the embodiment of the present invention, in order to verify the reliability of the network model, 11664 sets of wheel acceleration data and 11664 sets of wheel-rail force data obtained by the established vehicle-track vertical coupling model and rail joint irregularity disease model are used to verify the network model. 80% of the data is taken from the simulation data as the training set, and the remaining 20% ​​is used as the test set to test the performance of the model. The training results are as follows: Fig.10a and 10b As shown, Fig.10a As shown, it is a schematic diagram of the change of the first accuracy rate, from Fig.10a It can be seen that the overall accuracy is on an upward trend and gradually converges, the test set does not show overfitting, and the accuracy of the model can reach 99.8%; Fig.10b As shown, it is a schematic diagram of the loss change of the first model, from Fig.10b It can be seen that the overall loss change is on a downward trend and gradually converges, and the loss value of the model is close to zero.

[0139] Furthermore, in order to verify that data fusion can improve the detection accuracy, the acceleration data is input into the network for training separately. The training results are as follows: Fig.11a and Fig.11b As shown, Fig.11a As shown, it is a schematic diagram of the second accuracy change, Fig.11a It can be seen that the training process is not very stable, but the overall accuracy is on the rise. When the training rounds reach 500 times, the accuracy of the test set reaches 98.5%; Fig.11b As shown, it is a schematic diagram of the change of the loss of the second model, from Fig.11b It can be seen that the overall loss change is a downward trend and gradually converges. When the training round reaches 500 times, the loss value of the model is close to zero. Fig.12 As shown in Figure 2, it is a schematic diagram of the detection accuracy after network training before and after data fusion. Fig.12 It can be seen that when only acceleration data is input, the model accuracy is 0.985, while when acceleration is integrated with wheel-rail force, the model accuracy is 0.998, and the detection accuracy after data fusion is increased by 1.3%, indicating that data fusion can greatly improve the accuracy of disease detection.

[0140] The embodiments of the present invention align the vibration acceleration data collected by the sensor installed at the axle box and the wheel-rail force data collected by the strain gauges on the wheelset. First, an initial alignment is performed based on the first peak position of the welded joint, and second, a fine alignment is performed based on the correlation between the vibration acceleration and the wheel-rail force. The position of the rail weld is extracted, and the vibration acceleration and wheel-rail force data at the weld position are intercepted and then input into a neural network based on a deep learning-based "end-to-end" recognition algorithm. Finally, the health level is output to make a decision.

[0141] Embodiment 4

[0142] like Fig.13 As shown, this embodiment also provides a computer electronic device, which may include a processor 10, a memory 11, a communication bus 12 and a communication interface 13, and may also include a computer program stored in the memory 11 and executable on the processor 10, such as a data-fused rail weld defect detection program.

[0143] In some embodiments, the processor 10 may be composed of an integrated circuit, for example, a single packaged integrated circuit, or a plurality of packaged integrated circuits with the same or different functions, including one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and combinations of various control chips. The processor 10 is the control core (ControlUnit) of the electronic device, and uses various interfaces and lines to connect various components of the entire electronic device, and executes or executes programs or modules stored in the memory 11 (for example, a rail weld defect detection program that executes data fusion, etc.), and calls data stored in the memory 11 to execute various functions of the electronic device and process data.

[0144] The memory 11 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, mobile hard disk, multimedia card, card-type memory (for example: SD or DX memory, etc.), magnetic memory, disk, optical disk, etc. In some embodiments, the memory 11 can be an internal storage unit of an electronic device, such as a mobile hard disk of the electronic device. In other embodiments, the memory 11 can also be an external storage device of an electronic device, such as a plug-in mobile hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device. Further, the memory 11 can also include both an internal storage unit of the electronic device and an external storage device. The memory 11 can not only be used to store application software and various types of data installed in the electronic device, such as the code of the rail weld disease detection program of data fusion, but also can be used to temporarily store data that has been output or is to be output.

[0145] The communication bus 12 may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. The bus is configured to realize connection and communication between the memory 11 and at least one processor 10, etc.

[0146] The communication interface 13 is used for communication between the above-mentioned electronic device and other devices, including a network interface and a user interface. Optionally, the network interface may include a wired interface and / or a wireless interface (such as a WI-FI interface, a Bluetooth interface, etc.), which is generally used to establish a communication connection between the electronic device and other electronic devices. The user interface may be a display (Display), an input unit (such as a keyboard (Keyboard)), and optionally, the user interface may also be a standard wired interface, a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, and an OLED (Organic Light-Emitting Diode, organic light-emitting diode) touch device, etc. Among them, the display may also be appropriately referred to as a display screen or a display unit, which is used to display information processed in the electronic device and to display a visual user interface.

[0147] The figure only shows an electronic device with components. Those skilled in the art will understand that the structure shown in the figure does not constitute a limitation on the electronic device, and may include fewer or more components than shown in the figure, or combine certain components, or arrange the components differently.

[0148] For example, although not shown, the electronic device may also include a power source (such as a battery) for supplying power to each component. Preferably, the power source may be logically connected to the at least one processor 10 through a power management device, so that the power management device can realize functions such as charging management, discharging management, and power consumption management. The power source may also include one or more DC or AC power sources, recharging devices, power failure detection circuits, power converters or inverters, power status indicators, and other arbitrary components. The electronic device may also include a variety of sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be repeated here.

[0149] It should be understood that the embodiment is for illustration only and the scope of the patent application is not limited to this structure.

[0150] The data fusion rail weld defect detection program stored in the memory 11 of the electronic device is a combination of multiple instructions. When running in the processor 10, it can achieve:

[0151] Collect vibration acceleration at the axle box and wheel-rail force data at the wheelset;

[0152] Initially aligning the first waveform peak position corresponding to the vibration acceleration with the first waveform peak position corresponding to the wheel-rail force data;

[0153] Calculating the similarity between the wheel-rail force data after initial alignment and the vibration acceleration after initial alignment according to a preset sampling window, and precisely aligning the waveform window corresponding to the wheel-rail force data after initial alignment with the waveform window corresponding to the vibration acceleration after initial alignment according to the similarity;

[0154] Extracting the rail weld position according to the wheel-rail force data after precise alignment and the vibration acceleration after precise alignment, and extracting the target vibration acceleration data and target wheel-rail force data corresponding to the rail weld position;

[0155] The target vibration acceleration data and the target wheel-rail force data are channel-fused to obtain fused data, and a pre-built neural network model is used to perform rail weld defect detection on the fused data to obtain the degree of rail weld defect.

[0156] Specifically, the specific implementation method of the processor 10 for the above instructions can refer to the description of the relevant steps in the corresponding embodiment of the accompanying drawings, which will not be repeated here.

[0157] Furthermore, if the module / unit integrated in the electronic device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. The computer-readable storage medium can be volatile or non-volatile. For example, the computer-readable medium may include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a mobile hard disk, a magnetic disk, an optical disk, a computer memory, and a read-only memory (ROM).

[0158] Embodiment 5

[0159] This embodiment provides a storage medium storing a computer program. When the computer program is executed by a processor, the steps of the rail weld defect detection method using data fusion as described above are implemented.

[0160] These program codes can also be loaded onto a computer or other programmable data processing device so that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions executed on the computer or other programmable device for implementing the process. Figure 1 The steps of a specified function in a process or multiple processes.

[0161] Storage media include permanent and non-permanent, removable and non-removable media, and can be implemented by any method or technology to store information. Information can be computer-readable instructions, data structures, modules of programs or other data. Examples of storage media can include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices or any other non-transmission medium that can be used to store information that can be accessed by a computing device.

[0162] In the several embodiments provided by the present invention, it should be understood that the disclosed devices, media and methods can be implemented in other ways. For example, the device embodiments described above are only illustrative, for example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation.

[0163] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0164] In addition, each functional module in each embodiment of the present invention may be integrated into one processing unit, each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of hardware plus software functional modules.

[0165] It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0166] Therefore, no matter from which point of view, the embodiments should be regarded as illustrative and non-restrictive, and the scope of the present invention is not limited only according to the above description, and it is intended that all changes within the meaning and scope of equivalent elements within the scope of protection are included in the present invention.

[0167] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0168] In addition, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices stated in the system claim can also be implemented by one unit or device through software or hardware. The words first, second, etc. are used to indicate names, and do not indicate any particular order.

[0169] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A data fusion rail weld defect detection method, characterized in that: The method comprises: Collect vibration acceleration at the axle box and wheel-rail force data at the wheelset; Initially aligning the first waveform peak position corresponding to the vibration acceleration with the first waveform peak position corresponding to the wheel-rail force data; Calculating the similarity between the wheel-rail force data after initial alignment and the vibration acceleration after initial alignment according to a preset sampling window, and precisely aligning the waveform window corresponding to the wheel-rail force data after initial alignment with the waveform window corresponding to the vibration acceleration after initial alignment according to the similarity; Extracting the rail weld position according to the wheel-rail force data after precise alignment and the vibration acceleration after precise alignment, and extracting the target vibration acceleration data and target wheel-rail force data corresponding to the rail weld position; The target vibration acceleration data and the target wheel-rail force data are channel-fused to obtain fused data, and a pre-built neural network model is used to perform rail weld defect detection on the fused data to obtain the degree of rail weld defect.

2. The data fusion rail weld defect detection method according to claim 1, characterized in that: Before the first waveform peak position corresponding to the vibration acceleration is initially aligned with the first waveform peak position corresponding to the wheel-rail force data, the method further includes: The vibration acceleration and the wheel-rail force data are screened according to a preset collection distance to obtain the vibration acceleration and the wheel-rail force data at the collection distance; Downsampling the vibration acceleration and wheel-rail force data at the acquisition distance respectively, and configuring overlapping data and missing data for the downsampled vibration acceleration and wheel-rail force data respectively; The first waveform peak position corresponding to the configured vibration acceleration is extracted; and the first waveform peak position corresponding to the configured wheel-rail force data is extracted.

3. The data fusion rail weld defect detection method according to claim 2, characterized in that: The step of initially aligning the first waveform peak position corresponding to the vibration acceleration with the first waveform peak position corresponding to the wheel-rail force data comprises: The first waveform peak position corresponding to the wheel-rail force data is used as the reference peak position; Adjusting the first waveform peak position corresponding to the vibration acceleration according to the reference peak position; Aligning the first waveform peak position corresponding to the adjusted vibration acceleration with the first waveform peak position corresponding to the wheel-rail force data; The vibration acceleration and the wheel-rail force data are aligned according to the first waveform peak position after alignment.

4. The data fusion rail weld defect detection method according to claim 1, characterized in that: The calculating the similarity between the wheel-rail force data after the initial alignment and the vibration acceleration after the initial alignment according to the preset sampling window includes: Using the wheel-rail force data after initial alignment as reference data, and obtaining a reference sampling frequency corresponding to the reference data; Downsampling the aligned acceleration data according to the reference sampling frequency; Splitting the aligned wheel-rail force data window according to the number of sampling points in the sampling window to obtain a reference window; Dividing the window of the sampled acceleration data according to the number of sampling points in the sampling window to obtain a matching window; The similarities between the reference window and the matching window are calculated one by one.

5. The data fusion rail weld defect detection method according to claim 1, characterized in that: The step of finely aligning the waveform window corresponding to the wheel-rail force data after the initial alignment with the waveform window corresponding to the vibration acceleration after the initial alignment according to the similarity comprises: The waveform window corresponding to the wheel-rail force data after the initial alignment is used as the reference window, and the waveform window corresponding to the vibration acceleration after the initial alignment is used as the matching window; Select the matching window with the highest similarity corresponding to each benchmark window; Determining the window position of the matching window using a preset maximum value index function; Performing linear interpolation correction on the matching window according to the window position to obtain an acceleration correction window corresponding to the matching window; The reference window is aligned with the acceleration correction window.

6. The data fusion rail weld defect detection method according to claim 1, characterized in that: The method of extracting the rail weld position according to the wheel-rail force data after precise alignment and the vibration acceleration after precise alignment includes: Extract the wheel-rail force data after fine alignment and the convex waveform corresponding to the vibration acceleration after fine alignment; The position corresponding to the raised waveform is used as the rail weld position.

7. The data fusion rail weld defect detection method according to claim 1, characterized in that: The channel fusion of the target vibration acceleration data and the target wheel-rail force data to obtain fused data includes: Extracting a first channel position of target vibration acceleration data in a preset first channel; Extracting the second channel position of the target wheel-rail force data in the preset second channel; The first channel position and the second channel position are aligned, and the target vibration acceleration data and the target wheel-rail force data are superimposed according to the aligned channel positions to obtain fused data.

8. The data fusion rail weld defect detection method according to claim 1, characterized in that: The method of using the pre-built neural network model to perform rail weld defect detection on the fused data to obtain the degree of rail weld defect includes: Performing a convolution operation on the fused data using a convolution kernel in the neural network model; Inputting the convolved fused data into the residual module in the neural network model; Perform global averaging on the residual fusion data; The classification layer in the neural network model is used to classify the damage degree of the globally averaged fusion data to obtain the damage degree of the rail weld.

9. A computer device comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the rail weld defect detection method using data fusion as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the rail weld defect detection method using data fusion described in any one of claims 1 to 8 are implemented.

Images