• 专利附录
  • 专利说明
  • 权利要求
  • 类似技术
  • 同族专利
  • 引用文献
  • 法律状态
  • 优先权
    • 专利摘要:
    According to embodiments of the present disclosure, there is provided a diagnostic system for diagnosing faults in a railway point switch. The diagnostic system comprises: an input arranged to receive a waveform associated with operating the railway point switch during an event; a feature extraction module arranged to extract characteristic features representing a shape of the waveform; and a classification module arranged to apply the extracted characteristic features to logic rules for classifying the features according to an operating behaviour of the railway point switch during the event, classify the operating behaviour of the railway point switch based on the application of the extracted characteristic features to the logic rules, and generate a signal indicating the classified operating behaviour.
    公开号:EP3702238A1
    申请号:EP20157277.3
    申请日:2020-02-13
    公开日:2020-09-02
    发明作者:Michael Newman；David TICKEM；Daniel SHAYLER
    申请人:Thales Holdings UK PLC；
    IPC主号:B61L5-00
    专利说明:
    [0001] The present subject-matter relates to railway point switches, and in particular to systems and methods for diagnosing the operating behaviour of the point switches. The operating behaviour being diagnosed may relate to normal operation or one or more fault types. Background
    [0002] Railway point switches operate by swinging between different lines depending on the direction in which a train is headed. Condition monitoring may be employed to monitor operating behaviour of the point switch, in-use. Conventional condition monitoring typically monitors the behaviour of an electric motor when operating the point switch. For example, average current drawn from an electric motor may be monitored and, when the average current during a swing increases beyond a threshold value, an alarm or warning is sent to a maintenance terminal to advise that the point switch requires investigation to determine if there is a fault.
    [0003] Such monitoring means are not reliable at detecting all fault types, for instance faults exhibiting oscillations about a set value, which fault types would not necessarily fall below a threshold value. Also, such means of fault detection require an operator ultimately to assess the waveform in order to diagnose the fault, and so require manual investigation depending on the knowledge and experience of the operator.
    [0004] Accordingly, there is a need to provide improved condition monitoring. Summary
    [0005] According to an aspect of the present disclosure, there is provided a diagnostic system for diagnosing faults in a railway point switch. The diagnostic system comprises: an input arranged to receive a waveform associated with operating the railway point switch during an event; a feature extraction module arranged to extract characteristic features representing a shape of the waveform; and a classification module arranged to apply the extracted characteristic features to logic rules for classifying the features according to an operating behaviour of the railway point switch during the event, classify the operating behaviour of the railway point switch based on the application of the extracted features to the logic rules, and generate a signal indicating the classified operating behaviour.
    [0006] By extracting characteristic features representing a shape of the waveform, more detail is provided than merely obtaining an average current value. In this way, the shape of the waveform enables classification of the operating behaviour, for example, normal operation or a fault type, to be diagnosed when classifying the waveform. Accordingly, an automatic classification system is provided which obviates the need for manual investigation of the behaviour of a point switch by a trained operator.
    [0007] The characteristic features may be extracted using wavelet approximation. The wavelet approximation is beneficial compared to other reduction techniques since it retains the time resolution of the waveform so that the shape of the wave is maintained. The classification module may include a supervised machine learning algorithm for generating the logic rules for classifying the operating behaviour of the railway point switch.
    [0008] The supervised machine learning algorithm may comprise a Random Forrest. Use of the Random Forrest reduces overfitting of the data and improves classification accuracy compared to other methods.
    [0009] Classifying the operating behaviour of the railway point switch may include selecting subsets of features from the extracted features, generating probabilities that the subsets of the extracted features correspond to each of the feature sets, and identifying a most likely operating behaviour based on the generated probabilities.
    [0010] The event may be a swing of the point switch.
    [0011] The waveform may represent current drawn during the event versus a time duration of the event. Monitoring current is beneficial since it requires unobtrusive sensors to measure the current, which sensors do not interfere with operating the point switch.
    [0012] The extracted characteristic features may include average current for each period.
    [0013] The waveform may be received in real-time.
    [0014] According to a further aspect of the present disclosure, there is provided a training module for a diagnostic system, the diagnostic system for diagnosing faults in a railway point switch. The training module may comprise: an input arranged to receive a plurality of waveforms, each waveform associated with an operating behaviour of the railway point switch during an event; a feature extraction module arranged to extract characteristic features representing a shape of the waveforms; and a construction module arranged to construct a plurality of feature sets based on the characteristic features extracted from the waveforms, each feature set corresponding to an operating behaviour of the railway point switch.
    [0015] The features may be extracted using periodic time series data extraction.
    [0016] The feature sets may form part of a supervised machine learning algorithm.
    [0017] The supervised machine learning algorithm may comprise a Random Forrest.
    [0018] The features sets may form trees of the Random Forrest.
    [0019] The construction module may construct the feature sets by comparing extracted characteristic features between the plurality of waveforms, and disregard substantially identical features.
    [0020] The waveform may represent historical, in-service operating behaviour of the railway point switch. It is beneficial to utilise historical in-service waveforms since they will more accurately represent the in-service fault detection waveforms than other waveform types, for example waveforms constructed by industry experts or by simulators. The waveforms provided by in-service may also have a higher degree of representative variation due to other factors such as environmental changes. The degree to which environment effects the equipment is difficult to measure through simulation.
    [0021] The feature extraction module may be arranged to normalise the waveforms prior to extracting the features. In this way, seasonal variations affecting the waveforms may be negated.
    [0022] The foregoing diagnostic system may include the foregoing training module.
    [0023] According to a further aspect of the present disclosure, there is provided a method of diagnosing faults in a railway point switch. The method comprises: receiving a waveform associated with operating the railway point switch during an event; extracting characteristic features from the waveform, the characteristic features representing a shape of the waveform; applying the extracted characteristic features to logic rules for classifying the features according to an operating behaviour of the railway point switch during the event; classifying the operating behaviour of the railway point switch based on the application of the extracted characteristic features to the logic rules; and generating a signal indicating the classified operating behaviour.
    [0024] According to a further aspect of the present disclosure, there is provided a method of constructing a plurality of feature sets for classifying operating behaviour of a railway point switch. The method comprises: receiving a plurality of waveforms, each waveform associated with an operating behaviour of the railway point switch; extracting characteristic features from the waveforms, the characteristic features representing a shape of the waveform; and constructing a plurality of feature sets based on the characteristic features extracted from the waveforms, each feature set corresponding to an operating behaviour of the railway point switch; and generating logic rules for each feature set for classifying a real-time waveform. Brief Description of the Figures
    [0025] Aspects and embodiments of the disclosed subject-matter are best described with reference to the accompanying figures, of which:Figure 1 shows a railway point switch; Figures 2a and 2b show traces associated with operation of the railway point switch from Figure 1; Figure 3 shows a block diagram of a diagnostic system according to an embodiment; Figure 4 shows a schematic representing a wavelet reduction method for extracting features from waveforms using the diagnostic system from Figure 3; Figure 5 shows a schematic representation of subdivisions of a waveform obtained by the wavelet reduction method from Figure 4; Figure 6 shows a graph of a waveform and an corresponding waveform approximation obtained using the wavelet reduction method; Figure 7 shows a similar view to Figure 6 of two further waveform approximations obtained using the wavelet reduction method; Figure 8 shows a flow chart of operating a training module shown in Figure 3; and Figure 9 shows a flow chart of operating the diagnostic system from Figure 3. Detailed Description of the Embodiments
    [0026] Figure 1 shows a schematic of a railway switch 101 used to guide a train from a first track 103 to a second track 105 and a processor 109 for executing operating instructions to operate the railway switch 101, in accordance with some embodiments of the present disclosure.
    [0027] The railway switch 101 is one example of an electromechanical system in the railway infrastructure and is driven by an electric motor 107. When activated, the railway switch 101 moves switch blades (not shown) between the first track 103 and the second track 105 slightly, such that a train approaching the intersection between the first track 103 and the second track 105 is diverted onto the second track 105.
    [0028] The railway switch 101 comprises the electric motor 107 and a gear-mechanism (not shown). The electric motor 107 transforms electrical energy into mechanical energy and generates a rotational motion that is used to move the switch blades between the first track 103 and the second track 105. The gear mechanism may reduce the angular velocity of the motor and amplify the torque applied by the motor to the switch blades. In addition, it may transform the rotational motion into a translational motion. The switch blade mechanism includes springs and dampers configured to control the motion of the switch blades between the first track 103 and the second track 105. Furthermore, a lock pin or the like may be used to connect the electric motor 107 via a drive rod with the switch blades.
    [0029] In embodiments of the present disclosure, measurements of the electrical usage of the electric motor 107 are made each time the railway switch 101 is operated to switch the points, i.e., move the switch blades. The measurement may be carried out by measuring one of several electrical usage parameters; these parameters may include, for example, the current being drawn by the motor, the voltage drop across the machine, or the power transferred by the machine. Measurements may be made using a suitable meter (current meter, voltmeter, power meter etc.) In the example shown, the measurement is carried out by measuring current drawn by the electric motor during the operation. In some embodiments, the measurement(s) is output in the form of a signal trace or the like to an operating processor 109. The operating processor 109 may be located proximal to the railway switch 101 or be in a remote location. The processor may use periodic time series extraction to generate individual traces 111a, 111b (Figure 2) from the measured electrical parameter.
    [0030] With reference to Figures 2a and 2b, the signal trace may comprise a trace 111a, 111b showing change in the electrical usage parameter associated with the electric motor 107 with time during operation of the railway switch 101. The electrical parameter used in the traces 111a, 111b, shown in Figures 2a and 2b, is the electrical current drawn when operating the point switch 101 during an event, for instance a swing event. In particular, the two traces 111, 111b are traces relating to two different fault types when the railway point switch 101 is operating erroneously.
    [0031] In the first trace 111a, relating to the first fault type, the drawn current is seen to overshoot and then quickly decay and oscillate unevenly about a set-point value. The peak in current may be described as a load impulse. In the second trace 111b, relating to the second fault type, the load impulse quickly decays and the current oscillates unevenly about a first set-point value, before rising again to oscillate unevenly about a second set-point value. Both traces 111a, 111b end abruptly when the current suddenly decays to zero amps.
    [0032] With reference to Figure 3, the processor 109 (see Figure 1) is used to diagnose the fault types associated with each waveform in the first and second traces 111a, 111b. In particular, a diagnostic system 113 is stored on a memory (not shown) and is executed by the processor 109. The diagnostic system 113 includes a training module 115 stored as electronic data in the memory. The training module 115 includes an input 119, a feature extraction module 121, and a construction module 123. The processor 109 may be located at trackside or remotely. In the event of the remotely located processor 109, data is sent wirelessly, for instance by Global Systems for Mobile communications (GSM), at which point the waveform may be applied to the classifier.
    [0033] The input 119 received a plurality of traces 117 representing historical in-service waveforms similar to those shown in Figures 2a and 2b. Alternatively, simulated or manually constructed waveforms may be used, but historical in-service waveforms are preferred since they represent the point switch in a non-ideal environment and are more comparable to the in-service waveforms such as those shown in Figures 2a and 2b.
    [0034] The traces 117 may be obtained by performing periodic time series data extraction on a stream of data from the railway switch 101 (Figure 1).
    [0035] The feature extraction module 121 first normalises the waveforms 117 to remove any seasonal variations due to weather changes, for example. Next, the feature extraction module 121 performs pre-processing on the waveforms 117 to prepare the waveforms for classification.
    [0036] With reference to Figure 4, a trace may be pre-processed by wavelet approximation. Wavelet approximation, unlike many other possible transforms, is advantageous as it retains its time resolution and so the shape of the wave is maintained. Once reduced, the resulting waveform may be considered as an approximation of the original waveform. The approximation includes a plurality of time divisions, each including a mean value (as described below). The mean values in the time divisions may be used to obtain the characteristic features.
    [0037] The wavelet approximation reduces the wave to a number of residual mean values using the following formula: M 1 = 1 n ∑ i = 1 X i
    [0038] In formula (1), M is the mean value, n is the number of values in a section, and X is the numerical value.
    [0039] With reference to Figure 4, the full wave may be taken as a first "section", shown at a first level 140 of division. Formula (1) above is used to determine a mean for this first "section".
    [0040] The wave is then divided into a number of sections at a lower level, in this case the second level 142. The maximum number of subdivisions is equal to 2n in order for the wave to be represented uniformly, and this will produce (2n+1-1) wavelet components.
    [0041] According to Figure 4, the second level 142 has two "sections" into which the wave is divided. The divisions may be equal in time duration, such that the wave is divided into a first half and a second half each having equal time duration of half of the total time of the wave.
    [0042] Figure 4 shows three levels resulting in four sections into which the wave has been divided. The fourth section in the third level 144 shows formula (2), which is: M / 4 4= 1 / n 4 ∑ i = / 3 ∗ n4 n X i − M 1 − M / 2 2
    [0043] As can be seen from formula (2), a value in a lower level section is taken and the mean value, M, of a higher level section is subtracted from it. In this way, the mean value for the lower level section is determined using all values in that section. For instance, M1 is the mean of the entire wave, M1/2 and M2/2 are then the mean values of the first and second halves of the wave after the mean has been subtracted, and so on. This process can be repeated until either there are a certain maximum number of subdivisions or until there are a maximum number of wavelet components.
    [0044] With reference to Figure 5, the maximum number of subdivisions is eight, and the maximum number of wavelet components is fifteen.
    [0045] With reference to Figure 6, an original trace can be approximated using the wavelet described above using a maximum number of subdivisions of 16 wavelets. The original waveform is shown on the left of Figure 6 and the approximation is shown on the right of Figure 6. It can be seen from Figure 6 that the wavelet approximation follows the shape of the original trace whilst removing some of the resolution in current (mA) readings.
    [0046] Figure 7 shows two examples of wavelets with a different number of maximum subdivisions. Two primary effects can be seen as the number of sub-divisions used decreases. Firstly, the variation of the approximation from the original waveform increases, representing the removal of information when forming the wavelet. The second is that any high frequency changes (as can be seen at -180 and -270 in the example) are not well represented by the wavelet approximation. Sudden changes such as these do not represent significant characteristics for the majority of the fault symptoms that would have been selected for classification. However, this is not true in the case of certain fault types, for instance "motor brush wear" fault symptom as this has high frequency oscillations as a principal property for its identification.
    [0047] The construction module 123 uses the features to construct a plurality of feature sets each representing a fault type and then trains a supervised machine learning algorithm to classify the fault type associated with the trace. The supervised machine learning algorithm may be a Random Forrest. The feature sets are constructed by comparing the features from the plurality of traces 117 and identifying distinguishing features from among the features of the traces. The distinguishing features may be identified by comparing features between traces 117 and disregarding substantially identical features.
    [0048] During training, the construction module 123 looks at all values that are generated by the Random Forrest for a particular feature. The construction module 123 then generates logic rules to classify. The logic rules classify a fault according to how the feature of a real-time trace compares to the number from the logic rule.
    [0049] The feature sets for each fault type, or operating behaviour of the railway point switch 101, are then used to form the trees of the Random Forrest.
    [0050] Still with reference to Figure 3, the diagnostic system 113 may include an input 125, a feature extraction module 127, and a classification module 129.
    [0051] The input 125 receives the traces 111a, 111b. The feature extraction module 127 uses the same pre-processing as described above for the training module 115, and so duplicated description will be omitted. Once features have been extracted from a real-time trace 111a, 111b, those features are classified using the logic rule from the training phase. For instance, if the logic rule has a value of 10.16, the feature of the incoming trace 111a, 111b may be classified as a fault of class 1, 4, or 5, out of 6 potential feature classes, if the feature value from the trace 111a, 111b is higher than 10.16. Similarly, the feature of the incoming trace 111a, 111b may be classified as a fault of class 2, 3, or 6, out of 6 potential feature classes, if the feature value from the trace 111a, 111b, is lower than 10.16.Each tree in the Random Forrest provides a decision about which operating behaviour it believes is occurring. For example, the first tree may explain that it is operating normally, and would provide no other information regarding the probabilities of other faults. The decision of all the individual trees is then counted up, and thus the overall probability is obtained.
    [0052] By way of an illustrative example, if there are ten trees, and three of those ten trees predicted that the point switch was operating normally, three predicted fault A, and 4 predicted fault B. In this case, the overall probabilities given from the classifier would be 30% Normal, 30% fault A, and 40% fault B.
    [0053] It should be noted that the operating behaviours relating to the feature sets are not necessarily representative of fault types, since they could also represent the point switch 101 operating normally.
    [0054] Once the trace 111a, 111b has been classified according to an operating behaviour, a signal is generated that can be sent to an operator terminal (not shown) for retrieval by an operator to investigate. The signal may include a message detailing metadata associated with the fault, such as time and date of the event, and a classification label including a description of the classification that is suspected. Probabilities of each of the classes alongside the overall classification label may also be provided.
    [0055] With reference to Figure 8, operation of the training module may be described in terms of a method. Firstly, a plurality of traces 117 is received at step 200. The traces 117 may be pre-processed at step 202. Pre-processing firstly normalises the traces 117.
    [0056] Next, features are extracted from the readings in step 204. The features may represent the average current values during each period. Next, the extracted features are compared to features of other traces to determine distinguishing features from each trace. The distinguishing features may be obtained by ignoring substantially identical features, and retaining those features that are different to the features of other traces. The distinguishing features are collated as a feature set. Next, a Random Forrest is constructed at step 206 by creating a set of trees to classify the features from a trace 117. The number of trees in the Random Forrest is predetermined.
    [0057] With reference to Figure 9, operation of the diagnostic system may be described in terms of a method. Firstly, a real-time, in-service trace 111a, 111b is received at step 300. The trace is pre-processed step 302. Next, features are extracted from the periodic readings in step 304. The features may represent the average current values during each period. Next, the trace may be classified at step 306 using the Random Forrest. In one embodiment, the extracted features are classified using the Random Forrest which applies a value of a feature to the logic rules created during the training phase to generate a likelihood that the trace 111a, 111b is associated with an operating behaviour represented by each feature set. Subsequently, the classification of the operating behaviour is estimated based on the probabilities. Once the trace 111a, 111b has been classified, a signal is generated at step 308 indicating the operating behaviour of the railway point switch. The signal may be sent to an operator terminal to describe operating behaviour to an operator to aid fault diagnosis. In this way, there is reduced burden on the operator since a fault may be diagnosed more accurately and quickly and without input from an expert, as would usually be the case. In addition, a greater proportion of true positive alarms may be provided so that operators' time can be more efficiently used. Drawbacks associated with conventional systems include a high proportion of false alarms which produce a burden on the operators as they have to spend time investigating alarms that are not indicative of actual faults.
    [0058] The present disclosure is best understood in view of the following clauses. Clauses
    [0059] Clause 1. A diagnostic system for diagnosing faults in a railway point switch, the diagnostic system comprising: an input arranged to receive a waveform associated with operating the railway point switch during an event; a feature extraction module arranged to extract characteristic features representing a shape of the waveform; and a classification module arranged to apply the extracted characteristic features to logic rules for classifying the features according to an operating behaviour of the railway point switch during the event, classify the operating behaviour of the railway point switch based on the application of the extracted characteristic features to the logic rules, and generate a signal indicating the classified operating behaviour. Clause 2. The diagnostic system of Clause 1, wherein the characteristic features are extracted using wavelet approximation. Clause 3. The diagnostic system of Clause 1 or Clause 2, wherein the classification module includes a supervised machine learning algorithm for generating the logic rules for classifying the operating behaviour of the railway point switch. Clause 4. The diagnostic system of Clause 3, wherein the supervised machine learning algorithm comprises a Random Forrest. Clause 5. The diagnostic system of Clause 4, wherein classifying the operating behaviour of the railway point switch includes selecting subsets of features from the extracted features, generating probabilities that the subsets of the extracted features correspond to each of the feature sets, and identifying a most likely operating behaviour based on the generated probabilities. Clause 6. The diagnostic system of Clause 5, wherein the event is a swing of the point switch. Clause 7. The diagnostic system of any preceding clause, wherein the waveform represents current drawn during the event versus a time duration of the event. Clause 8. The diagnostic system of Clause 7 wherein the extracted characteristic features include average current for each period. Clause 9. The diagnostic system of any preceding clause wherein the waveform is received in real-time. Clause 10. A training module for a diagnostic system, the diagnostic system for diagnosing faults in a railway point switch, the training module comprising: an input arranged to receive a plurality of waveforms, each waveform associated with an operating behaviour of the railway point switch during an event; a feature extraction module arranged to extract characteristic features representing a shape of the waveforms; and a construction module arranged to construct a plurality of feature sets based on the characteristic features extracted from the waveforms, each feature set corresponding to an operating behaviour of the railway point switch. Clause 11. The training module of Clause 10, wherein the features are extracted using wavelet approximation. Clause 12. The training module of Clause 10 or Clause 11, wherein the feature sets form part of a supervised machine learning algorithm. Clause 13. The training module of Clause 12, wherein the supervised machine learning algorithm comprises a Random Forrest. Clause 14. The training module of Clause 13, wherein the features sets form trees of the Random Forrest. Clause 15. The training module of Clause 14, wherein the construction module constructs the feature sets by comparing extracted characteristic features between the plurality of waveforms, and disregarding substantially identical features. Clause 16. The training module of any one of Clauses 10 to 14, wherein the waveform represents historical, in-service operating behaviour of the railway point switch. Clause 17. The training module of Clause 15, wherein the feature extraction module is arranged to normalise the waveforms prior to extracting the features. Clause 18. The diagnostic system of any of Clauses 1 to 9 comprising the training module of any of Clauses 10 to 16. Clause 19. A method of diagnosing faults in a railway point switch, the method comprising: receiving a waveform associated with operating the railway point switch during an event; extracting characteristic features from the waveform, the characteristic features representing a shape of the waveform; applying the extracted characteristic features to logic rules for classifying the features according to an operating behaviour of the railway point switch during the event; classifying the operating behaviour of the railway point switch based on the application of the extracted characteristic features to the logic rules; and generating a signal indicating the classified operating behaviour of the railway point switch. Clause 20. A method of constructing a plurality of feature sets for classifying operating behaviour of a railway point switch, the method comprising: receiving a plurality of waveforms, each waveform associated with an operating behaviour of the railway point switch; extracting characteristic features from the waveforms, the characteristic features representing a shape of the waveform; and constructing a plurality of feature sets based on the characteristic features extracted from the waveforms, each feature set corresponding to an operating behaviour of the railway point switch, and generating logic rules for each feature set for classifying a real-time waveform.
    权利要求:
    Claims (15)
    [0001] A diagnostic system for diagnosing faults in a railway point switch, the diagnostic system comprising:
    an input arranged to receive a waveform associated with operating the railway point switch during an event;
    a feature extraction module arranged to extract characteristic features representing a shape of the waveform; and
    a classification module arranged to apply the extracted characteristic features to logic rules for classifying the features according to an operating behaviour of the railway point switch during the event, classify the operating behaviour of the railway point switch based on the application of the extracted characteristic features to the logic rules, and generate a signal indicating the classified operating behaviour.
    [0002] The diagnostic system of Claim 1, wherein the characteristic features are extracted using wavelet approximation.
    [0003] The diagnostic system of Claim 1 or Claim 2, wherein the classification module includes a supervised machine learning algorithm for generating the logic rules for classifying the operating behaviour of the railway point switch.
    [0004] The diagnostic system of Claim 3, wherein the supervised machine learning algorithm comprises a Random Forrest.
    [0005] The diagnostic system of Claim 4, wherein classifying the operating behaviour of the railway point switch includes selecting subsets of features from the extracted features, generating probabilities that the subsets of the extracted features correspond to each of the feature sets, and identifying a most likely operating behaviour based on the generated probabilities, and optionally wherein the event is a swing of the point switch.
    [0006] The diagnostic system of any preceding claim, wherein the waveform represents current drawn during the event versus a time duration of the event, and/or wherein the waveform is received in real-time.
    [0007] The diagnostic system of Claim 6 wherein the extracted characteristic features include average current for each period.
    [0008] A training module for a diagnostic system, the diagnostic system for diagnosing faults in a railway point switch, the training module comprising:
    an input arranged to receive a plurality of waveforms, each waveform associated with an operating behaviour of the railway point switch during an event;
    a feature extraction module arranged to extract characteristic features representing a shape of the waveforms; and
    a construction module arranged to construct a plurality of feature sets based on the characteristic features extracted from the waveforms, each feature set corresponding to an operating behaviour of the railway point switch.
    [0009] The training module of Claim 8, wherein the features are extracted using wavelet approximation.
    [0010] The training module of Claim 8 or Claim 9, wherein the feature sets form part of a supervised machine learning algorithm.
    [0011] The training module of Claim 10, wherein the supervised machine learning algorithm comprises a Random Forrest, and optionally wherein the features sets form trees of the Random Forrest, and optionally wherein the construction module constructs the feature sets by comparing extracted characteristic features between the plurality of waveforms, and disregarding substantially identical features.
    [0012] The training module of any one of Claims 8 to 11, wherein the waveform represents historical, in-service operating behaviour of the railway point switch, wherein the feature extraction module is arranged to normalise the waveforms prior to extracting the features.
    [0013] The diagnostic system of any of Claims 1 to 7 comprising the training module of any of Claims 8 to 12.
    [0014] A method of diagnosing faults in a railway point switch, the method comprising:
    receiving a waveform associated with operating the railway point switch during an event;
    extracting characteristic features from the waveform, the characteristic features representing a shape of the waveform;
    applying the extracted characteristic features to logic rules for classifying the features according to an operating behaviour of the railway point switch during the event;
    classifying the operating behaviour of the railway point switch based on the application of the extracted characteristic features to the logic rules; and
    generating a signal indicating the classified operating behaviour of the railway point switch.
    [0015] A method of constructing a plurality of feature sets for classifying operating behaviour of a railway point switch, the method comprising:
    receiving a plurality of waveforms, each waveform associated with an operating behaviour of the railway point switch;
    extracting characteristic features from the waveforms, the characteristic features representing a shape of the waveform; and
    constructing a plurality of feature sets based on the characteristic features extracted from the waveforms, each feature set corresponding to an operating behaviour of the railway point switch, and
    generating logic rules for each feature set for classifying a real-time waveform.
    类似技术:
    公开号 | 公开日 | 专利标题
    Ahmad et al.2019|A reliable technique for remaining useful life estimation of rolling element bearings using dynamic regression models
    Tsui et al.2015|Prognostics and health management: A review on data driven approaches
    US20210011791A1|2021-01-14|Abnormality detection system, abnormality detection method, abnormality detection program, and method for generating learned model
    Romano et al.2014|Automated detection of pipe bursts and other events in water distribution systems
    Soualhi et al.2014|Bearing health monitoring based on Hilbert–Huang transform, support vector machine, and regression
    Singleton et al.2014|Extended Kalman filtering for remaining-useful-life estimation of bearings
    Javed et al.2017|State of the art and taxonomy of prognostics approaches, trends of prognostics applications and open issues towards maturity at different technology readiness levels
    US9558601B2|2017-01-31|System and method for detecting vehicle system faults
    US10977568B2|2021-04-13|Information processing apparatus, diagnosis method, and program
    Niu et al.2010|Intelligent condition monitoring and prognostics system based on data-fusion strategy
    Chen et al.2008|Fault detection and diagnosis for railway track circuits using neuro-fuzzy systems
    CN104091070B|2017-05-17|Rail transit fault diagnosis method and system based on time series analysis
    Zhang et al.2013|Fault diagnosis and prognosis using wavelet packet decomposition, Fourier transform and artificial neural network
    US6567752B2|2003-05-20|General method for tracking the evolution of hidden damage or other unwanted changes in machinery components and predicting remaining useful life
    CN100465842C|2009-03-04|Data analysis system and method
    CA2433941C|2012-12-11|Adaptive modeling of changed states in predictive condition monitoring
    EP3186598B1|2018-06-27|Monitoring of a device having a movable part
    JP5284503B2|2013-09-11|Diagnostic system and method for predictive condition monitoring
    CN103699698B|2017-03-29|A kind of being based on improves Bayesian rail transit fault identification method and system
    EP1579288B1|2017-03-15|System state monitoring using recurrent local learning machine
    Asr et al.2017|Diagnosis of combined faults in Rotary Machinery by Non-Naive Bayesian approach
    Asada et al.2013|An algorithm for improved performance of railway condition monitoring equipment: Alternating-current point machine case study
    CN102208028B|2013-06-19|Fault predicting and diagnosing method suitable for dynamic complex system
    EP1374167B1|2010-05-12|Real-time spatio-temporal coherence estimation for autonomous mode identification and invariance tracking
    Chandrashekhar et al.2009|Structural damage detection using modal curvature and fuzzy logic
    同族专利:
    公开号 | 公开日
    GB2581390A|2020-08-19|
    GB2581390B|2021-03-03|
    GB201902157D0|2019-04-03|
    引用文献:
    公开号 | 申请日 | 公开日 | 申请人 | 专利标题
    法律状态:
    2020-07-31| STAA| Information on the status of an ep patent application or granted ep patent|Free format text: STATUS: REQUEST FOR EXAMINATION WAS MADE |
    2020-07-31| PUAI| Public reference made under article 153(3) epc to a published international application that has entered the european phase|Free format text: ORIGINAL CODE: 0009012 |
    2020-09-02| AK| Designated contracting states|Kind code of ref document: A1 Designated state(s): AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR |
    2020-09-02| 17P| Request for examination filed|Effective date: 20200213 |
    2020-09-02| AX| Request for extension of the european patent|Extension state: BA ME |
    2021-03-15| REG| Reference to a national code|Ref country code: DE Ref legal event code: R079 Ref document number: 602020000703 Country of ref document: DE Free format text: PREVIOUS MAIN CLASS: B61L0005020000 Ipc: G06N0005000000 |
    2021-04-12| GRAP| Despatch of communication of intention to grant a patent|Free format text: ORIGINAL CODE: EPIDOSNIGR1 |
    2021-04-12| STAA| Information on the status of an ep patent application or granted ep patent|Free format text: STATUS: GRANT OF PATENT IS INTENDED |
    2021-04-21| RIC1| Information provided on ipc code assigned before grant|Ipc: G06F 17/14 20060101ALI20210315BHEP Ipc: B61L5/02 20060101ALI20210315BHEP Ipc: B61L 27/00 20060101ALI20210315BHEP Ipc: G06N 20/20 20190101ALI20210315BHEP Ipc: G06N7/00 20060101ALI20210315BHEP Ipc: G06N5/00 20060101AFI20210315BHEP |
    2021-05-05| INTG| Intention to grant announced|Effective date: 20210413 |
    2021-08-24| GRAS| Grant fee paid|Free format text: ORIGINAL CODE: EPIDOSNIGR3 |
    2021-09-10| STAA| Information on the status of an ep patent application or granted ep patent|Free format text: STATUS: THE PATENT HAS BEEN GRANTED |
    2021-09-10| GRAA| (expected) grant|Free format text: ORIGINAL CODE: 0009210 |
    2021-10-13| REG| Reference to a national code|Ref country code: GB Ref legal event code: FG4D |
    2021-10-13| AK| Designated contracting states|Kind code of ref document: B1 Designated state(s): AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR |
    2021-10-15| REG| Reference to a national code|Ref country code: CH Ref legal event code: EP |
    2021-11-04| REG| Reference to a national code|Ref country code: DE Ref legal event code: R096 Ref document number: 602020000703 Country of ref document: DE |
    2021-11-10| REG| Reference to a national code|Ref country code: IE Ref legal event code: FG4D |
    2021-11-15| REG| Reference to a national code|Ref country code: AT Ref legal event code: REF Ref document number: 1438727 Country of ref document: AT Kind code of ref document: T Effective date: 20211115 |
    2021-12-20| REG| Reference to a national code|Ref country code: PT Ref legal event code: SC4A Ref document number: 3702238 Country of ref document: PT Date of ref document: 20211220 Kind code of ref document: T Free format text: AVAILABILITY OF NATIONAL TRANSLATION Effective date: 20211214 |
    2021-12-31| REG| Reference to a national code|Ref country code: FI Ref legal event code: FGE |
    2022-01-10| REG| Reference to a national code|Ref country code: DK Ref legal event code: T3 Effective date: 20220105 |
    2022-01-25| REG| Reference to a national code|Ref country code: SE Ref legal event code: TRGR |
    2022-01-26| REG| Reference to a national code|Ref country code: NL Ref legal event code: FP |
    2022-01-31| PGFP| Annual fee paid to national office [announced via postgrant information from national office to epo]|Ref country code: RO Payment date: 20211231 Year of fee payment: 3 Ref country code: FR Payment date: 20211224 Year of fee payment: 3 |
    2022-02-10| REG| Reference to a national code|Ref country code: LT Ref legal event code: MG9D |
    2022-02-14| REG| Reference to a national code|Ref country code: NO Ref legal event code: T2 Effective date: 20211013 |
    优先权:
    申请号 | 申请日 | 专利标题
    [返回顶部]