Method and device for monitoring the driving behaviour of rail vehicles

The invention relates to a method and a device for monitoring the driving behavior of rail vehicles and the diagnosis of components of rail vehicles.

It is known to monitor the driving behavior of rail vehicles during their intended use by recording and evaluating operating parameters of the rail vehicle that represent the driving behavior. It is known here, for example, to record the acceleration behavior of individual components of the rail vehicle by means of acceleration sensors arranged on the components. Since the existing wheel / rail contact of the rail vehicles causes the rail vehicle and therefore the individual components to vibrate, it is possible to infer irregularities by monitoring the actual vibration behavior. It is known to evaluate the individual acceleration signals in the frequency, statistical and / or time domain. Such methods can, for example, from<patcit id="pcit0001" dnum="WO0029270A"><text>WO 00/29270</text></patcit>, <patcit id="pcit0002" dnum="JP07174616B"><text>JP 07174616</text></patcit> or <patcit id="pcit0003" dnum="DE19826220"><text>DE 198 26 220</text></patcit> can be removed. In the known methods, however, it is disadvantageous that unforeseen external and / or internal influences lead to a change in the acceleration behavior of individual components, which then lead to a corresponding signal triggering. Often these external and internal influences are not identical to the events actually to be monitored, for example damage to individual components, so that false alarms cannot be ruled out with the known monitoring methods. For reasons of system acceptance, however, the maximum possible importance must be assigned to each alarm, so that the operating behavior of the rail vehicles is impaired as a result of the number of false alarms. For example, diagnostic systems with different relevance of the diagnostic results to one another are known from the automotive industry, such as diagnostic systems for brake lining wear, tire pressure or engine fire. With these diagnostic resp. Monitoring methods can also lead to the above-mentioned false alarms, since the operating parameters recorded there are fed in a one-dimensional manner to a single monitoring level and the evaluation methods or algorithms applied in this one monitoring level cannot compensate for any interfering influences affecting the measurement value acquisition.

The invention is based on the object of creating a method and a device of the generic type, by means of which reliable monitoring of the driving behavior of rail vehicles is possible with minimal triggering of false alarms.

According to the invention, this object is achieved by a method with the features mentioned in claim 1 and a device with the features mentioned in claim 15. In that at least some of the recorded operating parameters (measurement signals) are assigned in parallel to a plurality of monitoring levels of a monitoring hierarchy and at least two operating parameters (measurement signals) assigned to the same monitoring level are linked to one another in order to achieve event-dependent statements that characterize the driving behavior and the component diagnosis, depending on the assignment to one of the monitoring levels, the operating parameters (measurement signals) in the monitoring levels are subjected to the same and / or different evaluation algorithms, it is advantageously possible to carry out an optimized monitoring and diagnosis of the rail vehicle. In particular, it is possible Due to the combination options of the selection of the individual operating parameters (measurement signals) or the different combination options with regard to the assignment to the monitoring levels - and thus the selection of the evaluation algorithms - to recognize impending damage developments well before reaching critical states and thus the resulting time gain on the one hand for initiation of countermeasures and on the other hand for self-examination to use with regard to the exclusion of false positives. The same evaluation algorithms and / or different evaluation algorithms, for example in the frequency range and / or time range and / or in the statistics range, can be used in the different monitoring levels.

In particular, it is also essential to the invention that the signals corresponding to the different operating parameters in the different monitoring levels can, if necessary, be subjected to different evaluation algorithms. This allows the evaluation of the signals to be adapted to the different monitoring levels in a simple manner, with optimal evaluation algorithms being used in accordance with a necessary reaction time for recognizing relevant events. It is also particularly possible to link at least two operating parameters (measurement signals), which can be the same or different, with one another within a monitoring level.

Preferred configurations of the invention result from the features mentioned in the subclaims.

The invention is explained in more detail below in an exemplary embodiment with reference to the accompanying drawings. Show it:<dl id="dl0001"><dt>Figure 1</dt><dd>schematically a monitoring system of a rail vehicle;</dd><dt>Figures 2 to 5</dt><dd>Block diagrams of the signal processing within individual monitoring levels of the monitoring system;</dd><dt>Figure 6</dt><dd>an overall block diagram of the signal processing paths and</dd><dt>Figure 7</dt><dd>schematically the data flow in the monitoring system within a rail vehicle.</dd></dl>

<figref idref="f0001">Figure 1</figref> shows schematically a monitoring system 10 for a rail vehicle. The rail vehicle can, for example, be a car of a high-speed train, for example an intermediate car, a control car, a restaurant car or a powered end car. Of course, the subject matter of the invention can also be transferred to other rail vehicles, for example freight cars, coaches, traction vehicles or the like.

The monitoring system 10 comprises three monitoring levels 12, 14 and 16. The monitoring levels 12, 14 and 16 build on one another hierarchically. The first monitoring level 12 contains general diagnoses for the detection of damage, errors or the like in the initial stage, which can be remedied in the context of maintenance, repairs and / or maintenance without any time critically.

In the monitoring level 14, advanced damage and / or errors are recognized which could possibly lead to dangerous driving situations of the rail vehicle.

The monitoring level 16 contains what is known as a unit protection system, which detects, for example, loss of contact between wheel / rail, including the extreme case of a rail vehicle derailment.

The monitoring system 10 is provided with signals designated 18 overall. The signals 18 are supplied, among other things, by acceleration sensors arranged on individual components of the rail vehicle. The acceleration sensors can, for example, be assigned to a wheel set, wheel set bearing, bogie, car body, spring / damper systems. These acceleration sensors detect the accelerations acting on the respective components as a result of the wheel / rail contact. The accelerations can be recorded and evaluated in all three spatial directions, i.e. in the direction of travel, vertical to the direction of travel and transverse to the direction of travel.

The signals 18 also include an actual speed v of the rail vehicle, a mileage of the rail vehicle and temperature signals that are proportionally measured wheelset bearing temperatures. If necessary, further physical quantities that characterize the operating parameters of the rail vehicle can also be fed to the monitoring system. The signals 18 thus all represent momentary dynamic measured variables of the monitored rail vehicle.

In each of the monitoring levels 12, 14 and 16, the signals 18 fed to the monitoring system 10 are subjected to different evaluation algorithms. This affects both the type of evaluation and the time span of the evaluation. This will be discussed in greater detail using the following figures. The signals 18 are evaluated to determine whether events to be assigned to the monitoring levels 12, 14 or 16 are recognized. In the most uncritical monitoring level 12, for example, there is a processing time up to a time span t₁ of, for example, 300 s. When a relevant event is detected in the monitoring level 12, a message 12 'is sent to maintenance or the like. The message 12 'can be sent for example by means of remote data transmission.

In the monitoring level 14, the signals 18 are processed with a processing time span t₂, which is for example up to 30 s. Because of the more critical damage development to be assigned to the monitoring level 14, the processing time is faster here. When an event to be assigned to the monitoring level 14 is recognized, a message 14 'is issued which, for example, triggers an onboard message in the rail vehicle, for example to a train driver, a train control center or the like.

In the most critical monitoring level 16, processing takes place with a correspondingly adapted processing time t₃, which is, for example, up to 0.1 s. If an event to be assigned to the monitoring level 16 is recognized, a message 16 'is issued which directly or indirectly triggers an intervention in the driving operation of the rail vehicle, for example rapid braking or the like.

It becomes clear that, due to different degrees of damage escalation, staggered reports 12 ', 14' and 16 'are provided in accordance with monitoring levels 12, 14, 16. While the events to be assigned to the monitoring level 12 are not time-critical, events to be assigned to the monitoring level 14 lead directly to messages to an internal communication network of the rail vehicle (in the case of the train formation, to the communication network of the train) with possible suggestions for countermeasures to be initiated by the train crew. The messages 16 ', however, trigger immediate countermeasures.

The monitoring system 10 is also fed control or reference variables which, for example, have a minimum speed v_min of the rail vehicle and a linking algorithm 20 (linking logics 44, 68, 82, <figref idref="f0002 f0003 f0004 f0005 f0006">Figures 2 to 6</figref>) of the signals 18 according to the processing of the signals in the individual monitoring levels 12, 14 and 16 respectively.

With regard to the evaluation algorithms to be used in the individual monitoring levels 12, 14 and 16, the following should be noted. In the monitoring level 12 for initiating maintenance measures, procedures in the frequency range are effective. Due to the signal statistics improved by averaging procedures, these have the potential to track the natural frequencies of running gears and rail vehicles during travel and to recognize changes caused by damage or wear at an early stage. The excitation influences originating from the track superstructure do not interfere with the detection and assessment of the frequency position.

For the detection of dangerous driving situations in the monitoring level 14, for example, statistical methods are used which, in particular in the case of periodically occurring shock pulses, provide information on errors in the wheelset, wheelset bearings, damper systems or the like. By means of these evaluation algorithms, escalating developments can be recognized at short notice in order to initiate appropriate countermeasures (message 14 ').

The evaluation algorithms used in the monitoring level 16 are able to detect spontaneous derailments. Time series analyzes (histogram analyzes) are carried out in different speed classes.

<figref idref="f0002">Figure 2</figref> shows the evaluation algorithms of the signals 18 in the monitoring level 12 in a block diagram. Sixteen acceleration signals of individual components of the rail vehicle are present as output signals. This can be, for example, acceleration signals from the wheelset bearing housing, bogie or the like. The acceleration channels are read in with a sampling rate that is adapted to the highest frequency of interest in time processing, for example 32 kHz. They are then digitally filtered and sampled down (step 21). This limits the frequency band and minimizes the computing effort for the evaluation algorithms. The evaluation then takes place with a Hanning window (step 22) in order to reduce distortions in the frequency domain. The amplitude and phase spectra of the signals are then determined by means of a spectral analysis 22. This is done, for example, by means of a Fast Fourier Transformation (FFT) with a resolution of 0.5 Hz. With a frequency range of interest from 0 to 1024 Hz, this leads to 4096 required support points for the Fast Fourier transformation, so that a detection window is created of 2 s (4096 support points through f_sample = 2 s) for a single fast Fourier transformation. Over a period of 20 s, individual spectra are generated overlapping in the time domain, for example with a degree of overlap of 50%, so that the number of individual spectra to be averaged is nineteen. This process is restarted with an acquisition cycle of 30 s. The degree of overlap of the individual spectra can be varied within the limits between 30% and 70%, for example, deviating from the preferably 50%. As a result of the spectral analysis 22, nineteen spectra, for example, are available for the channels 16, of which the spectral auto and cross powers are determined (step 24) and then averaged (step 26) by suitable combination (step 30). The averaged phase spectra 34 are determined on the basis of the cross powers. Averaged coherence spectra 32 are determined on the basis of the auto and cross powers. The result of this evaluation is therefore eight features M₁ (corresponds to the coherence spectra 32), eight features M₂ (corresponds to phase spectra 34) and sixteen features M₃ (corresponds to the amplitude spectra 28) are available. Thus there are a total of thirty-two features 36 (8 x M₁ + 8 x M₂ + 16 x M₃) to disposal.

The spectra 28, 32 and 34 are determined and temporarily stored over a period of, for example, 120 s (or optionally 180 s, 240 s, 300 s or 360 s). The traveling speed v of the rail vehicle is measured once at the beginning and once at the end of each 20s detection window and is also stored temporarily. The smallest possible change in speed is used as a criterion for the final archiving of a spectrum originating from the set time window.

A selection circuit 38 can be used to define frequency bands to be monitored (monitoring bands) within the present amplitude spectra 28, phase spectra 34 and coherence spectra 32 by defining lower and upper band limits. Various parameters are evaluated within these defined frequency bands. First, the frequency of the maximum (peak) within the frequency band is determined. It is then checked whether the peak lies within the upper and lower frequency limit (parameters 1 and 2). If the peak leaves this frequency range, an alarm bit is set. The amplitude value of the peak must lie within the defined amplitude limit (parameters 3 and 4). If the range set by the upper and lower amplitude limit is left, an alarm bit is also set. The -3dB point is also monitored in the amplitude spectrum. This is the frequency line whose amplitude is attenuated by 3 dB above the peak. This -3dB point appears in the spectrum both to the left and to the right of the peak. The lower and upper frequency limits can be set for both -3dB points (parameters 5 to 8). This provides information about the curve shape around the peak area. If the -3dB point leaves the range defined by the limits, an alarm bit is also set.

Like the block diagram in <figref idref="f0002">Figure 2</figref> As can be seen, a check 40 of the averaged spectra is thus possible, with only parameters 1 to 4 being activated for coherence spectra 32 and phase spectra 34, while parameters 1 to 8 are activated for amplitude spectra 28. The monitoring 40 for exceeding the limit value is only applied to the amplitude spectra 28 provided for archiving. The maximum resultant alarm signals (alarm bits) 42 (256 alarm signals A₁, 256 alarm signals A₂, 1024 alarm signals A₃) are fed to a logic logic 44, which combines the alarm signals 42 with one another and, if the conditions are met, leads to an alarm 46, for example in the set 300s cycle.

It becomes clear that in the evaluation algorithms used in the monitoring level 12, in particular, spectral analyzes and correlations are used.

<figref idref="f0003">Figure 3</figref> shows the evaluation algorithms of the signals 18 in the monitoring level 14 in a block diagram. The acceleration signals of individual components of the rail vehicle are again present as output signals. The four bearing signals from the bogies of a rail vehicle are preferably evaluated here, so that eight of the acceleration signals 18 are evaluated. These are split into two sub-bands via a filter (splitter) 48. The lower subband (low) 50 covers a range from, for example, 0 to 400 Hz, while the upper subband (high) 52 covers the frequency range from, for example, 0.4 to 16 kHz. The mean value and the variance (steps 51 and 53, respectively) are calculated from the samples located in the respective sub-bands 50 and 52 over a period of 1 s. The respective 1-second kurtosis value is then calculated from this (steps 54 and 56). These 1-second kurtosis values ​​are then averaged over a period of 30 s (steps 58 and 60, respectively). The result of this evaluation is therefore eight features M₄ (Kurtosis values ​​in the low subband) and eight characteristics M₅ (Kurtosis values ​​in the high subband) are available. Thus there are another sixteen features 36 '(8 x M₄ + 8 x M₅) to disposal.

A check then takes place for each of the averaged kurtosis values ​​to determine whether limit values ​​have been exceeded (steps 62 or 66). If one of the kurtosis values ​​exceeds the specifiable limit value, an alarm bit 42 'can be generated in each case. So there are eight alarm signals A here₄ and eight alarm signals A₅ to disposal. These are routed via a link logic 68 which links the alarm signals 42 'with one another and, if the conditions are met, leads to an alarm 46', for example in the set 30s cycle.

It becomes clear that in the evaluation algorithms used in the monitoring level 14, in particular statistical methods such as kurtosis are used.

<figref idref="f0004">Figure 4</figref> shows the evaluation algorithms of the signals 18 in the monitoring level 16 in a block diagram. The sixteen acceleration signals of individual components of the rail vehicle are again present as output signals. A crest factor is determined for eight of these acceleration signals and a histogram analysis is carried out for a further eight (the same or the other) acceleration signals. As with the previous evaluation algorithms, the signals are again sampled with a sample frequency of 32 kHz.

When determining the crest factor (step 70), the individual values ​​are recorded in 10 ms time windows, that is to say in bursts of 320 individual values. In this time window, the peak value and the RMS value are determined and the quotient of the peak value and the RMS value (crest factor) is formed. The result of this evaluation is therefore eight features M₆ (corresponds to the crest factors) are available.

In parallel, the histogram analysis (step 72) takes place for eight acceleration signals, in particular for the four bearing signals of the bogies of the rail vehicle. The acquisition also takes place in 10 ms time windows, i.e. in bursts of 320 values. The largest positive and smallest negative value are selected from these 320 values ​​per time window (step 74) and entered in the histogram 76. This means that two values ​​occur per 10 ms time window for each monitored acceleration signal. With the histogram analysis, the frequency of certain acceleration values ​​is monitored in four speed classes. A histogram 76 is created for each speed class and signal.

The crest factor analysis thus gives eight features M₆ and thirty-two features M by histogram analysis₇ to disposal. Thus there are a total of forty features 36 ″ (8 x M₆ + 32 x M₇) to disposal.

The individual features are monitored for exceeding a threshold value. The features obtained as a result of the histogram analysis are monitored for exceeding a symmetrical threshold value. This symmetrical threshold value can be set as a parameter. Since the speed is recorded in 1s steps, the values ​​recorded in these 100x10ms time windows belong to the same speed class. The size of the acceleration values ​​is monitored depending on the speed class. If an adjustable maximum value is exceeded or the threshold value of the crest factor is exceeded, a signal alarm is given. The maximum resulting from this check 78 or 80 alarm signals (alarm bits) 42 ″ (eight alarm signals A.₆, thirty-two alarms A₇) are fed to a logic logic 82 which links the alarm signals 42 ″ to one another and, if the conditions are met, leads to an alarm 46 ″, for example in the set 10 ms cycle.

It becomes clear that in the evaluation algorithms used in the monitoring level 16, in particular crest factor or histogram analyzes are used.

<figref idref="f0005">Figure 5</figref> shows in a block diagram the evaluation algorithms of the temperature signals, which are available in the total amount of signals 18. The temperatures are monitored in the monitoring level 14, because so-called hot runners can be detected via the temperature monitoring, which can arise as a result of bearing damage in the early stages.

The temperature signals corresponding to the measured temperature of the individual bearing half-shells are recorded in 1s steps and averaged over a time window of 30 s (step 84). These eight averaged temperature values ​​thus available form the characteristics M₈. Furthermore, differences (step 86) between the averaged temperature signals of a bogie and the respective temperature of an individual bearing half-shell are determined, so that a further four features M₉ (Bogie 1) and four other features M₁₀ (Bogie 2) each with a limit temperature ΔT = T_individually - T_medium are available accordingly. A total of sixteen additional features 36 ′ are thus available in the monitoring level 14.

The features 36 'are then checked for exceeding a limit value of a bogie-specific, specifiable maximum temperature. Another monitoring criterion is the difference between the individual temperature of a bearing and the average temperature of all bearing half-shells of a bogie, which must also not exceed a certain limit value. The bearing-specific temperature deviation ΔT determined in the signal processing is monitored for a threshold value that can be set as a parameter. Furthermore, the bogie-specific maximum temperature T is_Max programmable as alarm parameters. According to these predeterminable threshold values, the check 88 is carried out to determine whether the temperature T_individually &gt; T_Max or the temperature T_individually &gt; T_medium + ΔT_border is. The maximum alarm signals 42 'resulting from the checking of the parameters (eight alarm signals A₈, four alarm signals A₉, four alarm signals A₁₀) are fed to the logic logic 68, which combines the alarm signals with one another and, if the conditions are met, leads to an alarm 46 ', for example in the set 30s cycle.

<figref idref="f0006">Figure 6</figref> shows in a block diagram the entire monitoring system 10 with the evaluation algorithms in the monitoring levels 12, 14 and 16. Overall, the following summarized signal processing or determination of features, alarm bits and alarms triggered by the linking logic results. These procedures have already been explained in detail with reference to the preceding figures.

With two bogies per car with eight channels each, there are sixteen acceleration channels per system. The speed v is recorded once for the entire system. There are four temperature channels per bogie, i.e. eight temperature channels per wagon. The compilation in<figref idref="f0007">Figure 7</figref> consists of the diagnostic tasks of the <figref idref="f0003 f0004 f0005 f0006">Figures 3 to 6</figref> together and shows again the selected form of the reporting concept via logical connections in the three different signal processing scenarios with response times adapted to the task.

The check for limit value violations is carried out with frequency range monitoring after spectrum selection has been made in adjustable time windows (120 or 180/240/300/360 seconds). The monitoring is applied to sixteen acceleration signals as well as eight coherence and phase combinations per vehicle. For the amplitude spectra there are eight monitoring bands and eight parameters (2 x amplitude, 2 x frequency, 4 x attenuation), i.e. 1024 alarm bits per vehicle, for the coherence / phase signals (2 x amplitude / 2 x frequency) there are 256 alarm bits per vehicle available. A maximum of eight freely selectable alarm bits can be linked for frequency monitoring and trigger an external alarm after being checked by a timer. The timing element ensures that an alarm condition is met only via the link result of two successive stored measurements (spectrum determinations) in order to forward an alarm to the external.

With the measurement data acquisition in the time domain, the data acquisition and signal alarm check for histogram and crest factor take place every 10 ms. This procedure is applied to the eight position signals of a vehicle. Forty alarm bits are set per vehicle.

With kurtosis monitoring, new value pairs are available for monitoring every 30 seconds. When used on the two frequency ranges (high / low) of the eight bearing signals of a vehicle, sixteen alarm bits are available.

The monitoring of the bearing temperatures is directly related to the bearing vibration monitoring by means of kurtosis. Each storage temperature is checked for two alarm conditions (T._Max and ΔT_border). This means that sixteen alarm bits are available per vehicle. The result of the storage temperature monitoring is also available in 30s intervals like the kurtosis monitoring, i.e. when linking kurtosis and temperature alarms, no management of different time intervals is necessary. The detected alarm bits are written into a kurtosis temperature signal alarm file, which is updated every 30 seconds. The signal alarms are linked in the logic editor and trigger an alarm when the condition is met.

<figref idref="f0007">Figure 7</figref> shows schematically the structure of the monitoring system 10. The monitoring system 10 has a modular structure, so that separate commissioning (of the individual monitoring levels 12, 14 and 16) as well as a quick exchange of individual components and the possibility of individual adaptation to other conditions is possible. The signals 18 of the sensors (not shown) are adapted to the input areas of analog / digital converters 102 via signal conditioning 100. The analog / digital converters 102 are implemented according to the required number of channels of the signals to be processed and the conversion rate of 32 kSamples per second. The analog / digital converters are connected to a digital signal processor (DSP) 104, in which the signal processing or processing, the feature extraction, the limit value comparisons and the alarming - is implemented - as explained in the previous figures. A flash memory 106 for loading the program and an external memory 108 for storing the data are assigned to the digital signal processor 104. The sequence control of the digital signal processor 104 is carried out by a computing system 110 to which a flash memory disc 112 is assigned for temporary storage of the measurement data. Furthermore, a hard disk memory 114 is assigned to the computing system 110. This equipment makes it possible to write the data to the flash memory 112 during the intended use of the rail vehicle and to transfer the data (copy) to the hard disk memory 114 only when the rail vehicle is at a standstill. This increases the data security of the monitoring system 10. The size of the flash memory 112 must be adapted to the maximum travel time of the rail vehicle and can be rewritten after each transfer to the hard disk memory 114.

It becomes clear that the system solution enables safe operation of the monitoring system 10 in the mobile area while complying with increased requirements for robustness, stability and energy efficiency.