Method for automatically evaluating orbital measurement data.

Abstract

The present invention relates to a method for automatically evaluating wavelet-transformed track measurement data (1) of a track geometry and / or ballast bed using a computing device. To enable the automatic generation of proposals for eliminating track errors, the method proposes to firstly wavelet-transform a measurement sequence of the track measurement data (1) to be evaluated, assigned to a given track section, with a number of wavelets of different wavelengths, and to generate from this wavelet transformation a kind of heat map, in which the wavelength and the wavelet transform over the position in the track are given as heat information, as well as a wavelet power density spectrum (3). Additionally, a signal intensity graph (4) for different wavelength ranges (D0, D1, D2, D3) is calculated, and accordingly, the local location (B, C, D, F), the spread (Δx) and the assigned wavelength range (E, D0, D1, D2, D3) of the dominant track error are identified from the heat map, in particular from the contours of the heat map, in order to determine the type, location, spread and magnitude of the track error.

Description

[Technical Field]

[0001] The present invention relates to a method for automatically evaluating wavelet transformed track measurement data of track geometry and / or ballast bed using a computing device. [Background technology]

[0002] The evaluation of measurement data acquired by a track inspection vehicle using wavelet transform is known in the prior art. For example, Patent Document 1 discloses a system for recognizing track irregularities, which evaluates data obtained from an acceleration sensor using wavelet transform. Patent Document 2 also discloses a method for detecting long-wavelength track irregularities, which evaluates data obtained from an angular acceleration sensor using wavelet transform. Patent Document 3 discloses another method for recognizing track irregularities, which detects track irregularities using Fourier transform of the wavelength at the base of the irregularity.

[0003] The majority of railway tracks are constructed as ballast superstructures. Sleepers are laid within the ballast. Due to the wheel forces of the trains running over it, the ballast is rounded and some parts are cracked and worn away. This causes irregular settlements within the ballast and shifts in the lateral geometry of the track. Settlements in the ballast bed cause errors in longitudinal elevation, cant (in curves), twist, gauge and alignment. These errors themselves increase the forces that act, which in turn act destructively on the ballast and foundation.

[0004] When these geometrical variations exceed predetermined limits or safety limits set by the railway authorities, maintenance work is planned and carried out. Track construction machines are deployed to eliminate and correct these geometrical track errors. Before track construction work begins, the geometrical track position of the track is determined, and correction values ​​are derived from this track position and transmitted to the machine that will carry out the corrective measures.

[0005] To reduce correction work, the machine is equipped with a tracking measuring system that checks the track position for compliance with pre-given tolerances (Patent Document 4).

[0006] A tamping machine with a fully hydraulic tamping drive (Patent Document 5) uses sensors integrated into the fully hydraulic tamping drive to determine the ballast characteristics during operation (Patent Document 6). By evaluating the force profile, squeeze speed, squeeze stroke and time course, ballast bed parameters such as ballast bed hardness, compaction force, ballast bed stiffness and ballast damping can be measured and presented for each tamped sleeper (Patent Document 7).

[0007] The degree of ballast contamination can be determined by these parameters. When the fines content in the ballast exceeds 30%, the track position can no longer be sustainably corrected by tamping. The ballast must then be replaced or cleaned. Ballast bed cleaning machines are used for this purpose.

[0008] Where there are significant stiffness differences in the track (e.g. at driven rail joints, at bridge or tunnel transitions), high wheel-rail forces cause the sleepers to strike the ballast bed, destroying the ballast (and rounding off the edges). Often, such areas are visible as white spots (lost rock dust). At these areas, the running dynamics lead to the ballast being pulverized, and these areas are manifested by loose mineral dust. These individual errors typically have an extent of a few meters but tend to spread further, causing the error amplitude to grow rapidly. This often results in safety-critical errors that must be eliminated immediately, or better, preventatively. These errors can be eliminated by a tamping machine working in individual error-resolving mode.

[0009] Differential settlement of the ballast bed causes floating sleepers, which typically occur at intervals of 2-5 m (D0 band). Floating sleepers can also be cleared by track tamping machines.

[0010] Error wavelengths in the 3-25 m range (D1 band) typically indicate track errors resulting from vehicle interaction (axle spacing within a bogie, bogie spacing or independent axle spacing, cushioning and damping behavior of the running gear). Settlement occurs due to the displacement, wear, and breakage of ballast particles. Track position errors in this wavelength range can be eliminated by track tamping machines. It is known that highly contaminated ballast beds exhibit high compaction forces. There are no moving gaps between the ballast bed particles because they are filled with fine material. This increases the compaction force that must be expended to move and compact the ballast. At the same time, this type of contaminated bed exhibits a reduced durability of the corrected track geometry due to weak friction and meshing between the ballast bed particles.

[0011] Long-wave errors between 25 and 70 meters (D2 band) are errors in the roadbed adjacent to the ballast. Typically, the cause is poor drainage or insufficient support properties (such as waterlogged loam or clay), resulting in relatively long-wave settlement. Blockages to drainage can occur, for example, through the construction of sound barriers. Sound barriers prevent water from draining from the trackbed. The longer the wave of the error, the more likely its cause is to be found further below the ballast layer. While these errors can certainly be counteracted with track tamping machines, the source of the error cannot be sustainably eliminated. Only trackbed cleaning or rehabilitation can achieve this. For this purpose, a protective layer is applied using trackbed improvement machines (or other non-mechanized methods). When track errors are due to poor drainage, drainage should be improved. This can be done by digging up trackside ditches or by cleaning and flushing drainage channels.

[0012] Errors with wavelengths greater than 70 m (D3 band) are attributable to poor foundations and insufficient bearing properties. Here, the use of subgrade improvement equipment can help, either by incorporating a load-distributing subgrade protection layer or by replacing the soil. Experience has shown that this type of error is often manifested by long-wave twisting.

[0013] Railroads classify track tracks into track classes, which are distinguished by the speed range within which the track operates. Each track class is assigned a unique limit for the standard deviation of track errors. This assignment is made for track geometry variables such as elevation, lateral tilt, twist, and gauge. Some limits are used for planning track work (which only needs to be performed within a specific time period), while others are critical and require immediate resolution or restriction of operation (even down to halt).

[0014] In contrast, measurement of ballast bed characteristics using a fully hydraulic tamping drive unit provides a very detailed and specific understanding of the ballast characteristics with respect to sleepers, thus enabling objective judgment.

[0015] Currently, track maintenance is planned based on track geometry measurements. Track inspection vehicles drive over the track at regular intervals and determine the geometric position of the track. The track position is then divided into sections, usually about 200 meters long, and the standard deviations of the height, alignment, cant and twist are determined. In addition to these statistical values, single individual errors are also measured. If the statistical values ​​exceed certain tolerances for comfort, maintenance work is planned and carried out.

[0016] The determination of track errors is then based on standard deviations or moving average values ​​of the measurement signals. The exact position, extent, type and cause of track errors remains unspecified or unclear. The planning and execution of track work is often based on pre-defined rules or the experience of the personnel. Ballast bed and foundation characteristics are usually not included in the determination, since objective measurement data often do not exist.

[0017] In most cases, a 200 m long section is evaluated and a track quality index (TQI) is provided for this section. The track quality index is often calculated from a weighted composite of the standard deviations of different track geometry parameters, or from the longitudinal elevation alone.

[0018] The drawback of these methods is that they are not based on an analysis of the causes of track errors and therefore often result in the application of inappropriate methods for correction, which results in increased maintenance costs. An incorrect method can, for example, lead to an increased and rapidly increasing number of tamping operations. The correct method might have been ballast bed cleaning. This not only leads to increased maintenance costs, but also has a detrimental effect on the lifespan of track components (ballast, rails, sleepers, etc.), increasing LCC. Ballast can be severely damaged after a long period of installation. A large proportion of fine particles and organic materials, or soil pushed up from the base, can fill the spaces between the ballast particles. In practical use, with this type of ballast structure, it has been found that the track position cannot be reliably corrected by track tamping machines.

[0019] In practical use, it is also known that individual errors occur randomly distributed within the track. Approximately 40% of these local defects can be remedied sustainably. 60% of these errors recur within a short time. Tracks with good ballast condition are tamped on average every four years. Individual errors indicating ballast destruction require maintenance measures approximately every one to three months. With each tamping, parts of the ballast are damaged by the tamping tool due to the high compaction forces. Therefore, long intervals between work cycles are economically significant.

[0020] Locations in the track with high ballast hardness (high stiffness) form high points in the track. The greater the differential stiffness variation in the track, the greater the force interaction between the wheels and rails, the higher the load on the track, and the faster the track geometry deteriorates. A single, short defect in the track has a tendency to propagate longitudinally in the track under the high dynamic forces acting on it, increasing the height of the track error and producing associated errors from active rail vehicles.

[0021] The application of artificial intelligence methods is prior art. The applied AI models can be classified into several different categories. A distinction is made between Artificial Neural Networks (ANN), Adaptive Neuro-Fuzzy Inference Systems (ANFIS), Decision Support Systems (DSS) and Artificial Learning models. Artificial intelligence models have the ability to map complex track position deterioration behavior, or the type, location, spread and wavelength of track position errors, with a high degree of accuracy. The AI ​​model must be trained with a training dataset. The AI ​​model is then checked with a test dataset.

[0022] By comparative LCC analysis, the costs of different maintenance methods (e.g., tamping instead of the necessary track cleaning, which continually shortens the maintenance cycle) can be compared with each other. To compare different maintenance strategies with each other, so-called standard elements or standard kilometers are determined, for example. For this, the expertise of railway engineers or determined real-world figures and costs are incorporated, for example. Standard kilometers are divided according to categories such as substructure quality, radius, traffic load, superstructure configuration, and number of tracks. [Prior art documents] [Patent documents]

[0023] [Patent Document 1] Chinese Patent Application Publication No. 111979859 [Patent Document 2] Chinese Patent No. 104032629 [Patent Document 3] Chinese Patent Application Publication No. 104947555 [Patent Document 4] Austrian Patent Application Publication No. 516278 [Patent Document 5] Austrian Patent Application Publication No. 513973 [Patent Document 6] Austrian Patent Application Publication No. 515801 [Patent Document 7] Austrian Patent Application Publication No. 520117 [Patent Document 8] Austrian Patent Invention No. 515801 Summary of the Invention

[0024] The problem underlying the present invention is to provide a method for automatically evaluating and analyzing a sequence of measured track measurement data, in particular of track geometry variables and / or ballast bed. The analysis preferably automatically determines the type, location, extent and magnitude of track errors. The method preferably further automatically generates suggestions for resolving the track errors and outputs a general assessment of the track condition.

[0025] The present invention solves the set problem with the features of independent claim 1. Advantageous developments of the invention are set out in the dependent claims.

[0026] The present invention is characterized in that first, a measurement sequence assigned to an arbitrary orbital section of the orbit measurement data to be evaluated is subjected to a wavelet transform using a plurality of wavelets each having a different wavelength, and from this wavelet transform, a kind of heat map is formed, in which the wavelength and the wavelet transform over the position in the orbit are given as heat information, and a wavelet power density spectrum is also formed. Additionally, signal intensity graphs are calculated for different wavelength regions, and accordingly, the local location, extent and assigned wavelength region of the dominant orbit error are identified from the heat map, particularly from the contours of the heat map, in order to determine the type, location, extent and magnitude of the orbit error.

[0027] For this purpose, the measured measurement sequences of track measurement data (e.g. longitudinal elevation, alignment, torsion, lateral elevation, ballast hardness, compaction force, ballast stiffness, ballast damping) are analyzed with wavelet and, optionally, fractal methods in terms of their local location and extent as well as their wavelength content. Importantly, unlike the Fourier transform, the wavelet transform preserves the position information of the errors, i.e., each individual error can subsequently be assigned a unique position within the track. Additionally, a wavelet power density spectrum is calculated, generating a kind of heat map. In the heat map, the location, extent, and magnitude of the track errors can be seen by color or by lines of equal intensity (isovalues).

[0028] Additionally, a log-log fractal graph can be generated, from which the orbit error is calculated as a function of wavelength (from the slope and position within the wavelength domain). The type, location, and extent of the orbit error are automatically calculated depending on the analyzed range of wavelengths contained in the measurement sequence. The strength and magnitude of the orbit error are estimated from the integral of the wavelet power density spectrum over the corresponding wavelength band. All of this is used to automatically evaluate the measured measurement sequence of orbit geometry variables and automatically determine the type, location, extent, and magnitude of the orbit error. From this analysis, suggestions for eliminating the orbit error are automatically generated. From this analysis, a general state determination of the observed orbit is also made.

[0029] The present invention provides an expert system that automatically assigns the type, location and extent of orbital errors using objective figures.For this evaluation, however, in addition to orbital position measurements, information about ballast is also available.The influence that different measurement quantities may have on each other is likely not captured by this expert system.

[0030] Therefore, it is desirable for this expert system to train an artificial intelligence that incorporates and takes these hidden relationships into its evaluation, thus leading to greater accuracy.

[0031] Wavelets originated from the idea of ​​dividing the track into shorter sections and using the Fourier transform to find the locations where shortwave track errors occur (short-path Fourier transform).

[0032] Unlike the sine and cosine functions of the Fourier transform, wavelets have locality in the wavelength spectrum and in the location spectrum. In simple terms, the wavelet transform acts as if the signal is filtered bit by bit with a bandpass filter of a specific bandwidth. Thus, a specific error wavelength region is found, localized. From a simple orbit error signal, wavelength is shown in two dimensions versus location. There are several different wavelet functions that can be applied. Typical ones are the Morlet wavelet and the Mexican hat.

[0033] A Mexican hat wavelet, for example, can be mathematically expressed as follows:

number

[0034] The wavelet transform is

number

[0035] With b as the shift factor the wavelet is moved through a function F(x) (e.g. the progression of the measured trackbed stiffness along the length of the track) and with a as the wavelength parameter (scale factor) the wavelength of the wavelet is changed, resulting in a two-dimensional diagram (the detected wavelength is plotted over the position b in the track).

[0036] By applying fractal theory, it is possible to calculate the so-called fractal index from the track measurement data curve of the track. For this purpose, for a specific length of railway track, for example 200 m, the length of the traverse that can be fitted to the measurement data curve is calculated with decreasing step size. For the traverse length,

number

[0037] If we logarithmize this equation, we get logL(d)=(1-D r )·log(d)+log(n) holds true.

[0038] A regression line is calculated (for each section) and plotted on the log-log diagram. The slope is always negative (the finer the section, the greater the traverse length). k=1-D r k...fractal number This becomes:

[0039] Research has shown that the different slopes of the regression lines can be assigned to wavelength regions and their causes.

[0040] According to European standards, typical wavelength ranges are divided into four categories to which the underlying causes of trajectories can be assigned according to practical experience and measurements. [Table 1] This table shows the classification of wavelength regions and the assignment of what causes this waveform.

[0041] The wavelength range D0 is currently not known or evaluated in most cases by electronic inspection and measurement runs. It is caused, among other things, by loose sleepers and by reactions between the sleepers and the rail fastenings. Sleepers that hit the ballast occur every 1.2 m, preferably every 3-3.6 m (i.e., twice and five to six times the typical sleeper pitch of 0.6 m).

[0042] Region D1 is a typical region where quasi-periodic track errors occur due to carbody and bogie motion. Dynamic loads acting on the rails lead to ballast degradation. The result is ballast wear and granule breakage. Maintenance measures include tamping or cleaning the ballast and replacing the waste with new ballast.

[0043] D2 occurs in the ballast-foundation mix area and within the foundation, which can also be improved by tamping and track cleaning.

[0044] Area D3 is attributable to foundation problems, and long-wave variations are often characterized by torsional variations. Area D3 is characterized by insufficient bearing capacity. Possible causes are poor drainage, poor soil quality (loam, clay, peat), unsuitable subgrade material, or a missing or too weak subgrade protection layer. This track error can be eliminated by eliminating drainage problems, subgrade improvement, subgrade rehabilitation, or soil replacement.

[0045] According to the invention, wavelet and fractal analysis is applied to the recorded measurement sequences, and the expert system thus created is used for training artificial intelligence.

[0046] The AI ​​model then automatically provides the location, extent and type of track errors, and also provides an indication of the overall quality state of the track and suggests optimal maintenance measures based on technical and economic calculations.

[0047] In the drawing, the invention is shown diagrammatically in one embodiment. [Brief explanation of the drawings]

[0048] [Figure 1] FIG. 10 is a diagram illustrating a Mexican hat wavelet. [Figure 2] FIG. 10 shows a fractal plot of the track error spectrum before and after trackbed cleaning. [Figure 3]FIG. 1 illustrates evaluation of a sequence of orbital error measurements by wavelet analysis. DETAILED DESCRIPTION OF THE INVENTION

[0049] Figure 1 shows an example of the shape of a so-called Mexican hat wavelet. For evaluation, the wavelet is moved through a measurement sequence. The same wavelength components of the signal match and generate a corresponding signal. Since this evaluation is performed in succession by moving several different wavelengths, a two-dimensional plot results.

[0050] Figure 2 shows the results of fractal analysis of a track section before and after track cleaning. While a clear impact is evident in the medium-wave regions 5 and 6 (2–15 m), the long-wave region remains virtually unaffected by ballast bed cleaning. The gentle slope in the long-wave region 7 suggests that the foundation has sufficient bearing capacity and is intact. Track cleaning has resulted in an improvement. However, track cleaning had no effect on the long-wave region. By tracking the evolution of the fractal count over track load or operating time, the remaining life or rate of deterioration of the ballast can be estimated. Similarly, a larger slope in the long-wave region indicates a foundation problem. Uniquely, fractal analysis can be performed for any track length, e.g., the entire track or even the entire track network. Fractal analysis provides numerical values ​​that are independent of the length of the pattern being analyzed.

[0051] Dirty ballast remains dirty even after tamping, and the condition of the foundation does not change much. Each line 5, 6, 7 indicates the error in the corresponding wavelength range. The steeper the line 5, 6, the greater the influence of the error. Fractal analysis is used as a second, independent method to determine the error wavelength band. While fractal analysis does provide an indication of the type and magnitude of the error, it does not provide an indication of its location. Fractal analysis determines these indications only for the section being evaluated and can prioritize them within this section to indicate the presence of a track error with this error wavelength component.

[0052] Figure 3 shows the evaluation of measurement signals by wavelet analysis. Measurement signal 1 (upper image area) can be geometrical measurements (longitudinal clearance, alignment, twist, gauge, lateral clearance) but also physical measurements (rail temperature, ballast bed hardness, ballast bed stiffness, ballast damping or compaction force at the end of tamping). Also visible in this image is the presentation of the TQI (track quality index), calculated, for example, from the weighted standard deviation of geometrical and physical measurements. The higher the TQI, the worse the track. The TQI value should also be presented track-grade-specific. High-speed track running at 300 km / h has tighter tolerances than freight track running at up to 80 km / h.

[0053] Below the signal curve, a two-dimensional heat map 2 calculated using wavelets (Morlet) is presented. The color representation of this heat map 2 allows for a clear visual representation of the error intensity, both in terms of its location and its extent. The vertical axis represents the wavelength, while the horizontal axis represents the position within the orbit. As an example, each 400 m section is evaluated, so that error wavelengths up to 200 m are still analyzable. The evaluation in the illustrated image considers error wavelengths up to 150 m, thus covering a total of four wavelength bands of interest, D0-D3. The error intensity appears in this heat map as contours or in the form of color grading. Thus, an error J in the region G of signal 1 appears correspondingly in heat map 2, in the region of the D1 band, at a wavelength of approximately 20 m. This is typically an error caused by ballast conditions and can be eliminated by tamping. This error may be present in the region of 250-350 m. Since there are also errors near 220 m, tamping between 200 and 350 m is the best option. Bands D2 and D3 show components that indicate drainage and bearing capacity problems in this region. This should be considered the original cause of the track errors in this region. The graph below displays the error intensities for the three wavelength bands D0-D2. This facilitates correlating the location and spread of the error with the error intensities. To the right of the heat map, the wavelet power density spectrum LD can be seen. Wavelength is plotted vertically, and power density is plotted horizontally. The graph also shows the wavelength bands. To determine the average power density in a wavelength band, an integral (denoted by A) is calculated. This is performed for all four bands. Finding the maximum is also important, as it indicates the dominant track error wavelength. Using this method, individual errors (with spread Δx, for example, in this image) can also be detected and reported with their location and spread.

[0054] The error type is assigned according to Table 1 shown above and the optimal maintenance is assigned accordingly. A comparative LCC analysis can be used to justify why, for example, tamping is more advantageous than track cleaning over this short region (as in this case).

[0055] Marking B shows a shortwave track error in the 40 m region, for example. This track error can be attributed to a floating sleeper. At the same time, the image in the PSD spectrum shows that the intensity is low and therefore tamping does not yet need to be carried out in this region. Marking F identifies a weak spot in the boundary layer at a wavelength around 35 m. Here, it would be recommended to investigate the effective drainage of this section.

Claims

1. 1. A method for automatically evaluating wavelet-transformed track measurement data (1) of a track geometry and / or ballast bed using a computing device, comprising: firstly wavelet-transforming a measurement sequence of the track measurement data (1) to be evaluated, assigned to a given track section, with a plurality of wavelets, each with a different wavelength; forming from the wavelet transform a kind of heat map, in which the wavelengths and the wavelet transform over positions in the track are given as heat information, and a wavelet power density spectrum (3); additionally calculating signal intensity graphs (4) for different wavelength ranges (D0, D1, D2, D3); and accordingly identifying from the heat map the local locations (B, C, D, F), extent and assigned wavelength ranges (E, D0, D1, D2, D3) of dominant track errors, in particular from the contours of the heat map, in order to determine the type, location, extent and magnitude of the track errors.

2. 2. A method according to claim 1, characterized in that an integral (A) of the power density over the wavelength, in particular over a wavelength range (D0-D3), is formed in the wavelet power density spectrum (3) in order to determine the intensity of the orbit error.

3. 3. The method according to claim 1, wherein the determined orbital errors are assigned to wavelength ranges (D0 to D3).

4. 4. The method according to claim 2 or 3, characterized in that the strength (signal strength, A) of the orbit error is also determined depending on the local spread and position of the heat information within the heat map (2, C, D, E).

5. 5. The method according to claim 1, wherein the location and magnitude of individual errors (D, Δx) are identified from the heat map.

6. 6. A method according to any one of claims 1 to 5, characterized in that the measurement sequence of the orbit measurement data (1) is analyzed with respect to its wavelength content by means of fractal analysis (Fig. 2), the orbit errors are automatically classified with respect to the wavelength range (D0, D1, D2, D3) depending on the analyzed range of wavelengths, the magnitude and intensity of the orbit errors are determined depending on the slope of the fractal line (5, 6, 7), and the orbit quality (TQI) is calculated.

7. 7. A method according to any one of claims 1 to 6, characterized in that the expert system thus formed is used to train an artificial intelligence, which automatically learns from the available data (1, 2, 3, 4, TQI) and identifies orbital errors.

8. 8. A method according to any one of claims 1 to 7, characterized in that the evaluation results in the generation of proposals for eliminating the orbital errors.

9. 9. The method according to claim 1, further comprising automatically compiling the results of said evaluation and suggestions for eliminating said orbital errors in a report (FIG. 3).

10. 10. The method according to claim 1, further comprising calculating and estimating the expected duration of the tamping operation from the results of said evaluation.

Images