Combined approach for evaluating a data record of an axle counter

Abstract

A computer-implemented method for evaluating a data set from an axle counter (100) is described, comprising: i) determining a plurality of peaks in the data set to provide a peak evaluation (110); ii) determining a plurality of shapes in the data set to provide a shape evaluation (120); and iii) comparing (130) the peak evaluation (110) and the shape evaluation (120) to provide a validated evaluation of the data set.

Description

Technical area

[0001] The invention relates to a computer-implemented method for evaluating a data set from an axle counter or the sensor of an axle counter. In particular, a peak-based evaluation and a shape-based evaluation are compared, thereby allowing validation of the evaluation of the data set. Furthermore, the invention relates to a data processing device, the axle counter, and a computer program product.

[0002] The invention can thus relate to the technical field of evaluation (in particular the validation thereof) of data sets of an axle counter or axle counter sensor, in particular with regard to rail vehicles. Technical background

[0003] Axle counters are well known in the field of rail-bound vehicles and rail infrastructure and form an essential part of the control system, for example to determine whether a wheel (of a train) is passing a track and how many wheels have passed. This can be of crucial importance for many applications, e.g. track vacancy detection, and in particular for safety within a rail infrastructure. As a rule, sections of a track are each equipped with two axle counter sensors (so-called counting points), one at the beginning and the other at the end of the section. Whether a track section is determined to be free or occupied depends in particular on the number of wheels detected by the two axle counter sensors. The higher-level evaluation unit (the actual axle counter) evaluates the signals detected by the sensors and uses this to derive the condition of the track section.Only if a section is actually recognized as free may a train pass through this section of track.

[0004] In the context of this document, the term "axle counter" can encompass both the unit "sensor (or counting point) and evaluation unit" or refer only to the evaluation unit (the actual axle counter). If the sensor is to be referred to specifically, the term "axle counter sensor" can be used; if the evaluation unit is to be referred to specifically, the term "axle counter evaluation unit" can be used.

[0005] However, particularly in the rail vehicle sector, particularly high safety and reliability requirements are placed on them. Accordingly, in numerous applications, evaluated axle counter data sets, especially for rail traffic automation, can only be used if particularly high reliability can be demonstrated.

[0006] To date, an axle counter cannot achieve 100% counting accuracy in practice. This is due, on the one hand, to the complicated / complex signals captured by the axle counter sensors, and, on the other hand, to signal processing methods that are not precise enough to handle complex cases. This is less of a safety issue, since the counts always err on the side of caution, but rather an availability issue.

[0007] Currently, wheel counting (or rather, the evaluation of the data set) is carried out very conservatively to ensure the safety of rail vehicles. For example, a section of a track is considered "occupied" rather than "free" if the counting accuracy is not 100% certain.

[0008] Another limitation of conventional axle counters is that, although they can detect the presence of wheels with a high degree of certainty, even state-of-the-art products can only provide such results at the expense of false positive detections (meaning that even though no wheel passes the sensor, a signal is detected (e.g., by metal masses affecting the sensor) and is counted by the sensor as a passing wheel).

[0009] Furthermore, a conventional axle counter cannot further classify characteristics of a passing rail vehicle (e.g. magnetic track brakes, wheels, bogies, or whether it is a freight train, passenger train, etc.).

[0010] Figure 6shows an example of an analog data set 200 (data set 210 represents the digitized signal) recorded by an axle counting sensor (an axle counting sensor or counting point consists of two sensor subsystems). In this example, four bogies of a rail vehicle have passed the sensor. Since each bogie in this example has two axles (each with two wheels, of which the one on the sensor side is always detected), two signals 201 (or peaks, pulses) are always located close together (peak cluster). In other words, the high signals correspond to the actual wheels. Furthermore, in the second and third bogie 202, lower signals can be observed between the high signals. These are usually caused by magnetic track brakes.

[0011] Figure 7shows a detailed view of a signal cluster of a two-axle bogie 202, wherein a low signal of a magnetic rail brake 204 is located between two high signals of the wheels 203.

[0012] Conventionally, the single-thresholding method is used to process axle counter signals. This method is Figure 8This is clearly illustrated. A predefined threshold is configured for each axle counter sensor. After recording data set 200, the recorded signals are examined at a specific threshold of an axle counter: Values ​​above the threshold are counted as wheel / axle, values ​​below are not, or only values ​​above the threshold generate a digital signal (reference numeral 210). Values ​​above this fixed threshold are mapped to "1" and values ​​below the threshold to "0". The value "1" indicates a point in time at which a wheel passes an axle counter. Thus, after truncation, data series 210 ideally shows only wheel signals. The advantage of this method is that it is fast enough to provide real-time detection of the presence of wheels.

[0013] However, the conventional method can have a critical disadvantage: it fails to take context into account. If a wheel signal does not reach the threshold, it cannot be detected (false negative). If the dataset contains noise or a non-wheel signal, e.g., caused by a magnetic track brake, this signal can exceed the threshold, and an additional, non-existent wheel is counted (false positive).

[0014] If the signals are close to the threshold, it can be extremely difficult to determine whether a wheel is involved or not. These are also the main reasons for "false positive" and "false negative" detections. Such errors can occur quite frequently and lead to inaccurate evaluations of the axle counter data records. This can then lead, for example, to a track section being considered occupied, disrupting train traffic until it is manually cleared.

[0015] Figure 9shows an example of a "counted one too many" error (false positive) according to the conventional "single threshold" method. Within the box, the first line shows the signal detected by the first sensor subsystem, and the second line shows the signal detected by the second sensor subsystem. These are the two sensors in an axle counter. The third and fourth lines are the results of a conventional single threshold algorithm, which displays the digitized threshold signals of the first and second sensor subsystems, respectively (corresponding to the evaluation of the analog signals from the first and second lines, respectively). The horizontal line in the first line indicates the specified threshold (trigger threshold).In the box example, for a single threshold, the algorithm incorrectly considers the result to be three wheels on a two-axle bogie due to the peak in the middle (caused by a magnetic track brake between the two wheels).

[0016] Figure 10 shows an example of a "one undercounted" error (false negative) according to the conventional "single threshold" method. As in Figure 9Rows one to four show the signals from the first sensor subsystem and the second sensor subsystem, along with their threshold values. In the example of the box shown, in the second row, there are two high signals from wheels and one lower signal caused by a magnetic track brake. The horizontal line is the set threshold. However, in the corresponding fourth row, which shows the threshold signal from the second sensor subsystem, no wheel is detected. This is because the values ​​of the signals in the second row are too low and do not exceed the threshold.

[0017] The disadvantages outlined above are not consistent with safe and efficient rail transport. Furthermore, these disadvantages may hinder the desired advancement of automation in rail transport (while maintaining high safety standards). Summary of the invention

[0018] There may be a need to evaluate a data set from an axle counter in an efficient and reliable manner or to validate the evaluation.

[0019] A method for evaluating a data set of an axle counter, a device for data processing, an axle counter, and a computer program product are described below.

[0020] According to a first aspect of the invention, a method, in particular a computer-implemented method, for evaluating (or validating the evaluation) of a data set of an axle counter (or sensor of an axle counter) is described, the method comprising: i) Determining a plurality of peaks (in particular local maxima) in the data set to provide a peak evaluation (e.g., a data set with extracted and classified peaks); ii) Determining a plurality of shapes (in particular, extracted signal shapes, further in particular compared with reference shapes) in the data set to provide a shape evaluation (e.g., a data set with determined and classified signal shapes); and iii) Comparing the peak evaluation and the shape evaluation (e.g., comparing the corresponding data sets) to provide a (validated) (overall) evaluation of the data set.

[0021] According to a second aspect of the invention, a data processing device is described which has at least one processor and which is configured to carry out at least one of the methods described above.

[0022] According to a third aspect of the invention, an axle counter is described which comprises: i) at least one sensor or sensor device (e.g. a dual sensor system) for recording a data set (or detecting (wheel) signals over time; and ii) a device for data processing (or an evaluation device, the actual axle counter) as described above.

[0023] In particular, the sensor device and the data processing device are coupled to each other (e.g. wired or wireless).

[0024] According to a fourth aspect of the invention, a computer program product is described which has instructions which, when the program is executed by a computer, cause the computer to carry out the method described above.

[0025] According to an exemplary embodiment, the invention can be based on the idea that a data set from an axle counter can be evaluated efficiently and reliably if two specific evaluation methods are compared with each other, so that a validated evaluation of the data set is obtained. The specific evaluation methods are, on the one hand, a peak-based approach and, on the other hand, a shape-based approach. The peak-based approach can refer to the extraction of peaks (e.g., local maxima based on specific peak criteria) from the data set. The shape-based approach, on the other hand, can refer to the extraction of specific signal shapes (e.g., based on comparisons with reference shapes). Both approaches are described in further detail below.

[0026] Experiments with real data sets have shown that both methods have their own strengths and weaknesses when dealing with specific data. In one example, the peak-based approach can efficiently and accurately locate where the signal is located. In contrast, in one example, the shape-based approach can be more error-prone when determining signal boundaries, especially for unusual signal clusters. For example, if a waveform is exceptionally long, the entire curve may not be detected when segmenting the waveform. In such a case, the peak-based approach can help ensure that the waveform determination includes the missing parts. In other words, the peak-based approach can particularly reliably determine the positions of wheel signals and thus compensate for certain inaccuracies of the shape-based approach.

[0027] At the same time, the peak-based approach can have the disadvantage that the height of a peak between (two) predefined thresholds can be ambiguous; e.g., a signal caused by a passing wheel but with only a low signal height can be incorrectly classified as a magnetic track brake. The shape-based approach, on the other hand, considers the overall shape of the signal clusters and not just a single signal feature such as the peak height. As a result, the shape-based approach tends to be more robust in correctly classifying the wheel signals, even when, for example, an ambiguous peak height is present, thus compensating for any inaccuracies of the peak-based approach.

[0028] The inventors have thus recognized that a combination of the two approaches described can be surprisingly efficient and reliable, because the strengths of the approaches can be combined and the weaknesses compensated.

[0029] For example, the peak evaluation and the shape evaluation can be compared (matched) to determine where they match and where they do not. "Comparing" here can mean, for example, that a waveform (or peak cluster) must match the peaks in the order "wheel peak - wheel peak" or "wheel peak - magnetic track brake peak - wheel peak." Matching signals can be considered reliable (validated). Non-matching signals can be corrected accordingly, for example, waveforms based on peak positions. Correcting waveform segments can involve, for example, creating a new curve, merging two curves, extending a curve, shortening a curve, or even deleting a curve.

[0030] Based on the corrected shapes, a particularly reliable evaluation can be obtained in a short time, which can also enable classification of the acquired signals. The described method can be implemented computationally particularly quickly and cost-effectively and can be implemented directly into existing systems. Exemplary implementation examples

[0031] According to one embodiment, determining the plurality of peaks comprises extracting local maxima from the data set. This approach can be particularly efficient for identifying peak signals and peak clusters and incorporating them into the analysis.

[0032] According to a further embodiment, determining the plurality of peaks further comprises determining at least one peak criterion. One or more criteria can be considered to extract peaks (local maxima) from the data set. The criteria can be selected such that, for example, only wheel signals and no magnetic track brake signals are reliably extracted.

[0033] According to another embodiment, a peak criterion comprises at least one of peak height, peak width, peak prominence, peak time (in the data set), and peak slope. The term "peak prominence" in this context can refer to how prominent the peak is, considering the intrinsic height, and / or how prominent the peak is compared to other peaks. One or more of these peak criteria can enable efficient and reliable extraction (and classification) of peaks for peak analysis. For example, thresholds can be set for one or more of the peak criteria.

[0034] According to a further embodiment, two or more threshold values ​​are used to determine the at least one peak criterion. Using two or more threshold values ​​per peak criterion can enable particularly reliable peak extraction (and classification).

[0035] According to a further embodiment, determining the plurality of shapes, in particular curve shapes, comprises determining at least one reference shape, and comparing at least one signal, in particular a signal shape, of the data set with the at least one reference shape. Comparing with a reference shape (in particular a reference shape library) can enable rapid and accurate extraction and classification of the (signal) shapes.

[0036] According to a further embodiment, the reference shape comprises, for example, one of the following: one wheel, two wheels, two wheels and a magnetic track brake between them, a single-axle bogie, a two-axle bogie, a wagon, or a train. This is only a selection of common shapes in the axle counter dataset; more can be added as needed.

[0037] According to another embodiment, the comparison comprises a Dynamic Time Warping (DTW) algorithm. Such an implementation using an established method can be implemented quickly and directly and can lead to surprisingly accurate results. In particular, the DTW algorithm can be used to determine a measure of difference between a measured shape in the data set and a reference shape. One or more thresholds can be introduced for this measure of difference.

[0038] In the context of this document, "DTW" can specifically refer to an algorithm that enables the optimal alignment of two shapes (e.g., in time series) so that the matching regions of the two shapes can be found and transformed to each other. Based on this alignment, the similarity of these shapes can be quantified in the following way (also often referred to as costs): In the standard version of DTW, the cost function (the effort) results solely from the difference between the function values ​​of projected pairs of values ​​between the measured shape and the reference shape.

[0039] According to another embodiment, the method further comprises extracting a signal shape from the data set. For example, the data set can be automatically divided into segments, each segment comprising a shape (signal, signal cluster, curve). This allows for efficient selection of a measured shape for comparison with a reference shape.

[0040] According to another embodiment, extracting a signal shape further comprises normalizing the signal shape before comparing it with the reference shape. This enables a particularly flexible comparison.

[0041] According to another embodiment, extracting a waveform further comprises segmenting the data set using boundary detection to thereby determine the shape boundaries of the waveform. This approach can be an efficient and accurate means of extracting the measured shapes for comparison.

[0042] If a (wave) shape is exceptionally long and its values ​​fluctuate greatly, it may be that only part of a shape is extracted, resulting in an incomplete shape. Such long waveforms can occur, for example, due to train stops or a reduction in train speed. In such a case, the acquired signal may show a waveform that is exceptionally long (see, for example, Figure 5 In this case, the combination with peak analysis can enable an accurate definition of the shape boundary.

[0043] According to a further embodiment, the method further comprises: pre-processing the data set to remove noise, in particular by means of smoothing and / or down-sampling.

[0044] Especially in the case of peak analysis, the original signal, with its many small fluctuations (or noise), can lead to many unwanted small peaks. Therefore, preprocessing to remove these peaks can be a significant advantage for reliably extracting peaks. To filter out noise while still retaining the important rad peaks, smoothing (average smoothing) or downsampling can be applied to the original signal. This, at best, results in a smoothed signal in which most of the small fluctuations have disappeared.

[0045] According to a further embodiment, the evaluation, in particular the determination of the plurality of peaks and / or the determination of the plurality of shapes, further comprises: classifying at least one signal of the data set.

[0046] Possible classes can include, for example, wheel, axle, magnetic track brake, single-axle bogie, double-axle bogie, and unknown signal. Significantly more efficient evaluation can be achieved if the meaning of the respective peaks / shapes is also "understood" and taken into account. Furthermore, classification can also be used advantageously for other applications in rail transport automation.

[0047] A further classification of the passing components (wheels, (active) magnetic track brakes, bogies, etc.) could be an important prerequisite for the increased safety requirements of the next generation of train control systems.

[0048] For example, peaks whose peak criterion(s) exceed one or more thresholds can be classified as wheels. Sometimes peaks are very high but have low prominence (usually short but high signals). Therefore, it is advantageous to consider multiple peak criteria for classification, especially considering the respective curve shape.

[0049] According to a further embodiment, the method further comprises: transferring the signal classification back to the data set (mapping), in particular for further validating the data set. After the peak evaluation, the shape evaluation, and the validation of the overall evaluation, this can be transferred back to the data set, e.g., as a plausibility check. This can further increase reliability and security. If there are (still) inconsistent results during the transfer back, these can finally be classified as "unknown signals."

[0050] According to a further embodiment, classifying a signal further comprises: considering at least one further signal in the vicinity of the signal to be classified, in particular a plurality of further signals in the neighborhood ("heuristic"). By including one or more further signals or signal clusters from the environment, the classification of an unknown signal can be determined. This is, for example, in Figure 4 where an unknown signal with four peaks can be identified in the overall context as two consecutive two-axle bogies. In another example (see Figure 5 ) a longer distance between two peaks can be identified as a stop / slowdown of the train.

[0051] According to a further embodiment, the method is used in connection with rail-bound vehicles and / or rail infrastructure.

[0052] The application of axle counters in rail transport has already been discussed in detail above. Furthermore, the rail vehicle sector has particularly high safety and reliability requirements and corresponding standards. Therefore, the use of the method for validating evaluations can be particularly important and advantageous in this context.

[0053] According to a further embodiment, the method further comprises providing reference shapes as training data to an artificial intelligence (AI) algorithm. Using a plurality of reference shapes (e.g., simulated and / or measured data sets), an AI algorithm can be trained and become increasingly better at comparing reference shapes and measured shapes. This can be done with human interaction (machine learning) or by means of the algorithm itself (deep learning).

[0054] In the context of this document, the term "AI" can specifically refer to computer-based approaches designed to mimic the cognitive functions of a human mind, especially learning and problem-solving. A variety of different mathematical algorithms and computational models have been developed to implement AI functionalities, such as machine learning, neural networks, especially deep learning, genetic algorithms, kernel regression, etc. The primary purpose of these approaches can be seen as improving an existing algorithm by training it with training data, resulting in a learning effect and improving the algorithm's problem-solving ability over time. This can occur with or without human intervention (e.g., tweaking).

[0055] According to a further embodiment, the described method can be implemented efficiently with short computing times in established software (e.g. C, C++, Python).

[0056] In an exemplary embodiment, aspects of the present invention can also be described as follows: Experiments with example data sets show that the use of the pure peak detection algorithm already outperforms the conventional "single thresholding" method.

[0057] After combining peak and shape detection, due to the internal matching of peaks and shapes and the automatic correction of mismatched shapes (validation), the described method can be considered more robust and powerful than conventional approaches.

[0058] It should be noted that embodiments of the invention have been described with reference to various subject matters. In particular, some embodiments have been described with reference to method claims, while other embodiments have been described with reference to apparatus claims. However, a person skilled in the art will appreciate from the foregoing and the following description that, unless otherwise stated, in addition to any combination of features belonging to one type of subject matter, any combination of features relating to different subject matters is also deemed to be disclosed by this document. This applies in particular also to features of the method claims and features of the apparatus claims.

[0059] The above-defined aspects and further aspects of the present invention will become apparent from the examples of embodiments to be described below and will be explained with reference to the examples of embodiments. The invention will be described in more detail below with reference to embodiments to which, however, the invention is not limited. Short description of the drawings

[0060] Figure 1 shows determining a plurality of peaks to obtain a peak evaluation according to an exemplary embodiment of the invention. Figure 2 shows determining a plurality of shapes to obtain a shape evaluation according to an exemplary embodiment of the invention. Figure 3 shows a first comparison of the peak evaluation and the shape evaluation to validate the evaluations, according to an exemplary embodiment of the invention. Figure 4shows a second comparison of the peak evaluation and the shape evaluation to validate the evaluations, according to an exemplary embodiment of the invention. Figure 5 shows a third comparison of the peak evaluation and the shape evaluation to validate the evaluations, according to an exemplary embodiment of the invention. The Figures 6 and 7 each show a data record of an axle counter. Figures 8 to 10 illustrate a conventional threshold-based method and its disadvantages. Figure 11 shows an axle counter according to an exemplary embodiment of the invention. Detailed description of the drawings

[0061] The representations in the drawings are schematic. It should be noted that in different figures, similar or identical elements or features are provided with the same reference numerals or with reference numerals that differ from the corresponding reference numerals only within the first digit. To avoid unnecessary repetition, elements or features that have already been explained with reference to a previously described embodiment will not be explained again at a later point in the description.

[0062] Furthermore, spatially relative terms such as "front" and "back," "top" and "bottom," "left" and "right," etc., are used to describe the relationship of one element to another, as illustrated in the figures. Thus, the spatially relative terms may apply to orientations used that differ from the orientation illustrated in the figures. Obviously, these spatially relative terms refer only to simplify the description and to the orientation shown in the figures and are not necessarily limiting, since a device according to an embodiment of the invention may assume orientations other than those illustrated in the figures, particularly when used.

[0063] Figure 1shows determining a plurality of peaks to obtain a peak evaluation 110 according to an exemplary embodiment of the invention. In this example, peaks were extracted from the data set and correctly classified as wheels 111 and magnetic track brakes 112. Based on the observation that wheels and magnetic track brakes lead to local peaks (maxima) in the axle counter data set, all peaks were first extracted and then classified as "wheel," "magnetic track brake," or "unknown signal."

[0064] In an exemplary embodiment, a four-step approach is used to obtain the peak evaluation 110: 1. Extract all relevant local maxima from the dataset (these usually correspond to wheels or magnetic track brakes). 2. Calculate peak criteria for these local maxima, e.g., height, width, prominence, etc. 3. Classify the peaks as "wheels," "magnetic track brakes," or "unknown." 4. Map the classification back to the dataset.

[0065] Figure 2 shows the determination of a plurality of shapes to obtain a shape evaluation 120 according to an exemplary embodiment of the invention. In the example shown, two different (signal) shapes were correctly determined: i) two-axle bogie without a magnetic track brake 121 and ii) two-axle bogie with a magnetic track brake 122.

[0066] Determining shapes is not about extracting local maxima (as in peak detection), but rather about finding specific signal shapes (pattern recognition), e.g., using a curve recognition algorithm. An important observation here is that certain signals generally follow certain shapes. For example, the curves of two wheels are represented by the shapes of two high signal peaks, while two wheels with a magnetic track brake between them are represented by the shapes of two high signal peaks with a low signal in between.

[0067] Based on this observation, the shape approach can be based on the idea of ​​classifying such shapes or curves according to their shape. Here, it is not important that the shapes are identical (e.g., to a reference shape), but rather that they are similar; sometimes slightly longer or shorter, sometimes slightly higher or lower. Such a comparison between signal shape and reference shape, bridging the variations, can be performed particularly efficiently, for example, using DTW.

[0068] In an exemplary embodiment, a four-step approach is used to obtain the shape evaluation: 1. Define typical shapes (reference shapes), e.g., two wheels or two wheels with magnetic track brakes. These typical shapes can be used as reference shapes for DTW, which compares the reference shapes with the measured shapes from the dataset. 2. When extracting the shapes from the dataset (segmentation), local shape boundaries are determined; this can be further supported by heuristic approaches. 3. Each extracted shape segment can first be normalized (with respect to length) so that it is comparable to the reference shapes (using DTW). 4. Based on the comparison using DTW (warping distance can be obtained), the extracted shapes can be classified. If the differences from the reference shapes are too large (distance too large), the "unknown signal" classification can be used.

[0069] Figure 3shows a first comparison of the peak evaluation 110 and the shape evaluation 120 to validate the evaluations, according to an exemplary embodiment of the invention. By means of the comparison, a validated (and combined) evaluation 130 can be obtained. This can make the evaluation more stable and accurate, allowing even challenging cases to be reliably captured. In an exemplary example, the evaluation and validation of a data set of approximately 60 seconds in length can be processed in 0.02-0.15 seconds.

[0070] As described above, the peak-based approach is designed to determine where peaks are located and then classify them. This can lead to uncertainty regarding false positives / negatives. The shape-based approach is designed to locate shapes (especially curves) and classify them using reference shapes. In complex cases, uncertainty can occur, particularly at shape boundaries. In the example shown, all peak clusters 132 were correctly identified. Only peak cluster 131 has an unusual shape and is therefore classified as an "unknown signal." However, using the peak-based approach, it can be concluded that a wheel-magnetic track brake-wheel signal shape is present.

[0071] Figure 4shows a second comparison of the peak evaluation 110 and the shape evaluation 120 to validate the evaluations, according to an exemplary embodiment of the invention. In this case, a challenging peak cluster 133 is present: two two-axle bogies were detected within a short distance (presumably very short wagons or particularly high speed). The shape-based approach will not detect the shape with four peaks. However, using the peak-based approach, it can be determined that four wheel signals are present, so that the shape evaluation 120 can be corrected (re-segmentation).

[0072] Figure 5shows a third comparison of the peak evaluation 110 and the shape evaluation 120 in order to validate the evaluations 130, according to an exemplary embodiment of the invention. Between the wheel signals 141 and 142, there is a particularly large distance 140 (e.g., brief standstill, particularly low speed), which the shape-based approach cannot detect. However, using the peak-based approach, it can be determined that two wheel signals from the same bogie are present, so that the shape evaluation 120 can be corrected. Furthermore, in this case, correction can also be made using heuristics, whereby neighboring signals are taken into account in order to correctly map the peaks and shapes to the data set.

[0073] Figure 11shows an axle counter 100 according to an exemplary embodiment of the invention. It has at least one sensor device 102 (this can also be a dual-sensor system) that is attached to a rail 101 and can detect the data sets described above. In this example, the sensor device is connected via cable (alternatively wirelessly) to an evaluation device 103. This evaluation device 103 is configured to receive the measured data set and to evaluate and validate it according to the method described above.

[0074] It should be noted that the term "comprising" does not exclude other elements or steps, and the use of the article "a" does not exclude a plurality. Elements described in connection with different embodiments may also be combined. It should also be noted that reference signs in the claims should not be construed to limit the scope of the claims. Regardless of the grammatical gender of a particular term, persons of male, female, or other gender identities are included.

Claims

1. A computer-implemented method for evaluating a data set of an axle counter (100), the method comprising: determining a plurality of peaks in the data set to provide a peak evaluation (110); determining a plurality of shapes in the data set to provide a shape evaluation (120); and comparing (130) the peak evaluation (110) and the shape evaluation (120) to provide a validated evaluation of the data set.

2. The method of claim 1, wherein determining the plurality of peaks (110) comprises: extracting local maxima (111) from the data set.

3. The method according to claim 2, wherein determining the plurality of peaks (110) further comprises: determining at least one peak criterion, in particular at least one of peak height, peak width, peak prominence, peak time, peak slope, further in particular wherein two or more threshold values ​​are used to determine the at least one peak criterion.

4. The method according to one of the preceding claims, wherein determining the plurality of shapes (120), in particular curve shapes, comprises: determining at least one reference shape, in particular one of bogie-two-wheels (121), bogie-two-wheels-one-magnetic-rail-brake (122), single-axle bogie, two-axle bogie; and comparing at least one signal, in particular a signal shape, of the data set with the at least one reference shape.

5. The method of claim 4, wherein the comparing comprises a Dynamic Time Warping, DTW, algorithm.

6. The method according to claim 4 or 5, further comprising: extracting a signal shape from the data set, in particular further comprising: normalizing the signal shape before comparing it with the reference shape; and / or segmenting the data set by boundary determination to thereby determine the shape boundaries of at least one signal shape.

7. The method according to one of the preceding claims, further comprising: pre-processing the data set to remove noise, in particular by means of smoothing and / or down-sampling.

8. The method according to one of the preceding claims, wherein the evaluation, in particular the determination of the plurality of peaks (110) and / or the determination of the plurality of shapes (120), further comprises: classifying at least one signal of the data set, in particular as at least one of wheel (111), magnetic rail brake (112), single-axle bogie, two-axle bogie (121, 122), unknown signal (131).

9. The method according to claim 8, further comprising: transferring the signal classification back to the data set, in particular for further validating the data set.

10. The method according to one of the preceding claims, wherein classifying a signal further comprises: considering at least one further signal in the vicinity of the signal to be classified, in particular a plurality of further signals in the vicinity.

11. The method according to one of the preceding claims, wherein the method is used in connection with rail-bound vehicles and / or rail infrastructure.

12. The method according to any one of the preceding claims, further comprising: providing reference shapes as training data to an artificial intelligence, AI, algorithm.

13. A data processing device (103) comprising at least one processor and configured to carry out the method according to any one of claims 1 to 12.

14. An axle counter (100) comprising: a sensor device (102) for receiving a data set; and a data processing device (103) according to claim 13, which is coupled to the sensor device (102).

15. A computer program product comprising instructions which, when executed by a computer, cause the computer to carry out the method according to any one of claims 1 to 13.

Images