Method for diagnosing and monitoring vehicles

Abstract

The invention relates to a method for diagnosing and monitoring vehicles, in particular a drive system of a rail vehicle, at least having the following steps: a) receiving an acoustic airborne sound signal by means of at least one acoustic sensor (1, 2, 3, 4, 5, 6, 7, 8); b) determining at least one feature from the airborne sound signal; and c) carrying out at least one step selected from a diagnosis and a monitoring process of at least one vehicle component on the basis of at least one feature determined in step b).

Description

[0001] Description

[0002] Diagnostic and monitoring procedures for vehicles

[0003] The present invention relates to a method for diagnosing and monitoring vehicles. The present invention particularly relates to a method by which faults and damage to a vehicle component, such as in particular a rail vehicle, can be detected.

[0004] Vehicles, especially rail vehicles, must be highly safe to operate. Accurate assessment and prediction of the technical condition of vehicles and vehicle components (e.g., chassis components) is therefore essential. In this context, effective and efficient maintenance and repair of vehicles and vehicle components is crucial.

[0005] Interval-based servicing is intended to keep the probability of failure as low as possible. However, this often results in service being performed that isn't actually necessary. To keep lifecycle costs as low as possible without increasing the risk of failures and operational disruptions, diagnostic and monitoring systems are sometimes used. These also form the basis for condition-based maintenance.

[0006] Common diagnostic and monitoring systems for trains, such as for their bogies and drive systems, come in various forms. For example, stationary systems are known which enable diagnostics in particular. These include systems placed along the track which contain vibration sensors and microphones. Mobile systems are also known, i.e. systems which are permanently installed in the train and travel with the train. Diagnostics are usually carried out using acceleration sensors. Monitoring is carried out using state-of-the-art temperature sensors on the bearings. Finally, handheld devices are known which are usually vibration-based and are used in particular as devices for measuring the condition in the depot during servicing.

[0007] AT 523862 A1 relates to a device and a method for diagnosis and monitoring of vehicles, in particular for chassis, in particular for rail vehicles, wherein at least one first sensor and at least one computing unit are provided, which are connected to one another in a signal-transmitting manner and can be arranged on or in a vehicle. It is proposed that the at least first sensor is designed as an acoustic sensor and at least one data transmission unit is provided for transmitting data processed in the at least one computing unit. This enables an analysis of the acoustic behavior of the vehicle or of vehicle components with regard to large frequency ranges, wherein analysis results can be forwarded for evaluation.

[0008] However, such solutions still have further potential for improvement.

[0009] The object of the present invention is to at least partially overcome the disadvantages known from the prior art. In particular, the object of the present invention is to provide a solution by which the diagnosis and monitoring of a vehicle, in particular a rail vehicle, can be improved.

[0010] According to the invention, the object is achieved at least in part by a method having the features of claim 1. According to the invention, the object is further achieved at least in part by a system having the features of claim 15 and by a vehicle having the features of claim 16. Preferred embodiments of the invention are described in the subclaims, in the description or the figures, wherein further features described or shown in the subclaims or in the description or the figures can represent an object of the invention, individually or in any combination, unless the context clearly indicates the opposite.

[0011] A method is described for carrying out at least one step selected from a diagnosis and monitoring for vehicles, in particular for a drive system of a rail vehicle, at least comprising the following method steps: a) receiving an acoustic airborne sound signal by at least one acoustic sensor; b) determining at least one feature from the airborne sound signal; and c) carrying out at least one step selected from a diagnosis and a monitoring of at least one vehicle component based on at least one feature determined in method step b).

[0012] Such a method has clear advantages over the state-of-the-art solutions.

[0013] The method described here is used to monitor a vehicle, particularly a rail vehicle, for damage and, if it does occur, to diagnose it. This allows service intervals to be maintained and unnecessary service units or maintenance activities to be avoided or at least reduced. At the same time, availability and reliability can be increased.

[0014] The method relates in particular to damage to the chassis and thus fundamentally to the vehicle's drive system. The invention makes it possible to monitor a large number of different components and diagnose their damage.

[0015] For this purpose, the method comprises at least the following steps.

[0016] According to method step a), an acoustic airborne sound signal is received by at least one acoustic sensor. In principle, any acoustic sensor or microphone can be used as the acoustic sensor.

[0017] A favorable solution is achieved by providing at least two sensors configured as a microphone array. Such a microphone array improves analysis results, as it enables precise localization of noise events and noise sources.

[0018] Using the acoustic sensor, a wide range of technical conditions of the vehicle or its components can be assessed. For example, wheel polygons, flat spots and chipping on wheels, bearing damage on axle or wheelset bearings, engine and transmission bearings, etc. can be detected. Furthermore, parameters can be identified that indicate the technical behavior of a spring, damper, or bearing.

[0019] A further advantage of the method according to the invention is that, compared to a diagnostic and monitoring device with acceleration sensors, a smaller number of sensors is required with the method according to the invention. This is because if an acoustic sensor is in the near field or environment of the respective components to be monitored, several components can be covered with one sensor. Furthermore, some technical conditions of the vehicle or the vehicle components are easier to distinguish from one another on the basis of acoustic information, or more precisely on the basis of airborne sound information, than on the basis of kinematic information.

[0020] It is advantageous if at least one acoustic sensor can be arranged in a chassis cavity. By arranging a sensor in a cavity (e.g., in the area between the wheels of a wheelset, between a first longitudinal member and a second longitudinal member of a chassis, or in a recess in a chassis frame structure), it is protected from environmental influences, such as disruptive wind noise, which could affect measurement results.

[0021] However, it can also be helpful if at least one sensor can be located on the outside of the landing gear. This allows noise from components located on the outside of the landing gear, such as roll dampers, to be detected and correctly identified with a high degree of certainty.

[0022] To record and evaluate noises in the area of ​​the chassis, it is advantageous if at least one sensor can be connected to a chassis frame.

[0023] If, for example, axle or wheelset bearing noises are to be recorded and analyzed, it is advantageous if at least one sensor can be connected to a first wheelset bearing housing or a first wheelset guiding device. The sensor can be arranged, for example, on an outer or inner side of the first wheelset bearing housing.

[0024] A further advantageous embodiment is achieved if at least one sensor can be connected to a motor or a transmission. By arranging a sensor, for example, on a motor, transmission, or common drive housing, or in the near field of the drive, engine and transmission noises, such as noise from a transmission bearing, can be detected and correctly assigned with a high degree of certainty.

[0025] According to feature b), at least one feature is further determined from the airborne sound signal. Such features are known, in particular, from speech recognition of acoustic signals and can be generated and processed using known computing methods. Accordingly, one or a plurality of corresponding features can be generated from a particularly digitized acoustic signal.

[0026] Particularly preferably, the so-called cepstrum can be generated as a corresponding feature, or the features determined in method step b) can at least comprise the cepstrum of the airborne sound signal. In a manner known per se to a person skilled in the art, the cepstrum can be obtained from the spectrum by forming the EFT of the logarithmized magnitude spectrum. More precisely, the cepstrum of a signal can be obtained by applying the inverse Fourier transform for time-discrete signals to the logarithmized magnitude of the time-discrete Fourier transform of the signal. For example, in a manner known per se, the cepstrum can be the result of the following calculation sequence: transformation of a signal from the time domain to the frequency domain; logarithmization of the spectral amplitudes; and transformation into the frequency domain, in which the independent variable again represents a time axis.

[0027] The cepstrum can be used in particular to analyze periodic structures in frequency spectra. Accordingly, the cepstrum can be used particularly advantageously in the method described here. This is because damage to vehicles, particularly to the chassis or drive system, often results in periodic noises that can be detected and diagnosed particularly advantageously using the cepstrum. In particular, the so-called mel-frequency cepstral coefficients (MFCCs) are extracted from airborne sound recordings. The MFCCs enable a compact representation of the spectrum of a signal by combining the cepstrum of a signal with an approximately logarithmic scaling of the frequency axis. Accordingly, this design can lead to reliable fault detection with low computational effort.More specifically, to calculate the MFCCs, the signal is first divided into individual windows by windowing. An N-valued discrete Fourier transform is then performed for each of these windows. The resulting power magnitude spectrum is then filtered with a Mel filter bank consisting of overlapping triangular filters. Finally, the MFCCs are generated by the inverse discrete cosine transform of the logarithmized and summed filter bank energies. The features thus obtained are then used to train a classifier, as described below.

[0028] Furthermore, it may be advantageous to extract LPC coefficients from the received airborne sound signal in process step b). Linear Predictive Coding (LPC) is a transformation method used, for example, in speech compression in mobile radio systems and can also offer advantages in the present application.

[0029] Another possibility applicable according to the invention is that, in process step b), PLP coefficients (Perceptual Linear Prediction) are extracted from the received airborne sound signal. This can particularly be advantageous in very good recognition rates.

[0030] It may also be advantageous for log-mel spectral data to be extracted from the received airborne sound signal in method step b), or for amplitude modulation spectral data to be extracted from the received airborne sound signal in method step b). Such features can also have advantages in a method according to the invention. The amplitude modulation spectrum is particularly suitable, for example, for detecting fluctuations in the signal envelope.

[0031] However, the present invention is in no way limited to the features described above and determinable in step b), so that these are to be regarded as exemplary and not limiting.

[0032] In light of the above, it can be particularly advantageous if at least one sensor has a digitization unit. This measure allows digitization of measurement signals to be carried out directly by means of the first sensor, whereby digital signals do not have to be transmitted to the computing unit. However, it is particularly advantageous in principle that the acoustic signal is digitized regardless of the location of the digitization, so that a digitization unit can also be a unit connected to the sensor.

[0033] The method described here further comprises the further method step c ), namely the implementation of at least one step selected from a diagnosis and a monitoring of at least one vehicle component based on at least one feature determined in method step b ).

[0034] In other words, the feature previously known in particular from speech recognition and determined from airborne sound, also referred to as a feature, is used to perform at least one step selected from a diagnosis and monitoring of at least one vehicle component. Thus, using the feature(s), it is possible to examine occurring acoustic signals for corresponding error states. This allows for reliable testing and diagnosis of the vehicle.

[0035] The method described here allows an analysis of the acoustic behavior of the vehicle or of vehicle components with regard to large frequency ranges, whereby analysis results can be transmitted for evaluation, for example via cable or radio, to a driver's cab or to a control, maintenance or repair station outside the vehicle.

[0036] Status information can be output accordingly. For example, a message can be output when an error is detected. More precise information can also be given about the component to which the error relates or what type of error has occurred. A prioritization information can also be output, which provides or makes it available information about whether the error is safety-critical or whether operation is possible at least for a limited time and / or distance. If no error status is detected, a message can be output indicating that error-free operation is possible.

[0037] There are also advantages compared to state-of-the-art solutions.

[0038] Known solutions based on the analysis of structure-borne sound signals are detected using acceleration sensors. These sensors must be mounted directly on the component housing. Accordingly, restrictions regarding positioning must be observed. In contrast, the present invention is based on the detection and processing of airborne sound signals. Compared to the known solutions, this solution is less intrusive because the sensors for recording the airborne sound do not have to be mounted directly on the component housing, but can simply be positioned in the immediate vicinity of the component.For this reason, the described method is more universally applicable, as there are fewer restrictions regarding sensor placement and additional installations, such as the attachment of additional mounting plates that could impair the operating behavior or mechanical integrity of the components, are not required. The present invention is therefore also independent of the design of the component to be monitored.

[0039] There is also the potential to cover multiple components with one microphone or an array of microphones, which can further improve the application possibilities.

[0040] Compared to a vibration-based diagnostic system, it can be assumed that an airborne sound-based solution can be implemented much more cost-effectively due to the freedom in positioning the sensors and the lower mechanical requirements outside the bogie as well as the potential to cover several components.

[0041] In addition, by using features that are also known from speech recognition, acoustic signals that can indicate an error condition, which can also be referred to as status indicators, can be recognized particularly safely and reliably.

[0042] The risk of operating the vehicle with a potentially safety-critical fault can thus be significantly reduced. It is also helpful if the condition indicators are transmitted from the vehicle to at least one infrastructure facility outside the vehicle. This allows maintenance and repair measures to be planned and carried out in a timely manner in the infrastructure facility (e.g., in a vehicle depot, in an operations center, etc.). This promotes condition-based maintenance and repair.

[0043] To detect fault conditions, such as rolling bearing damage, features are used that have already been widely used in speech recognition, speaker recognition, or the classification of acoustic scenes. However, these features have not been used in the prior art to detect fault conditions in a vehicle using airborne sound recordings.

[0044] It may further be preferred that the method comprises the further method steps: d) training a classifier with the at least one feature using training data and validation data or defining a classifier; wherein e) method step c) is carried out using the classifier.

[0045] The classifier serves, in particular, to assign the detected acoustic signals to a fault or damage based on at least one characteristic, such as the determined cepstrum. Alternatively, if such characteristics are absent, it can be concluded that no fault condition exists and the vehicle therefore does not require service.

[0046] The classifier can, for example, be defined based on a model and then applied. Preferably, however, the classifier can be trained based on the recorded features. For this purpose, training data and validation data can be collected, and the classifier trained and validated. After validation of the classifier, it can then be applied accordingly in a condition diagnosis.

[0047] By using a classifier, the recognition of features or acoustic signals that may indicate a fault condition can be further improved.

[0048] A relatively simple classifier can be used to obtain the characteristics, also known as features.

[0049] The number of parameters to be optimized and the training data required for their optimization are thus small. This enables a fast and cost-effective training process and simultaneously reduces the risk of overfitting the classifier. These aspects represent decisive economic and technical advantages over solutions that either rely on a large number of different features or that pursue an end-to-end approach based on deep learning.

[0050] It may be preferred if the classifier comprises at least one artificial neural network. In particular, using an artificial neural network allows for effective training, thus ensuring that any errors that occur are reliably detected.

[0051] For example, it can be advantageous if at least one neural network used comprises a multilayer perceptron (MLP), more precisely preferably a fully-connected feedforward multilayer perceptron (MLP). The MLP consists of an input layer, two hidden layers and an output layer. The width of the input layer and the output layer are fixed by the number of features used and the number of classes. The width of the hidden layers can be freely chosen. The first and second hidden layers are each followed by a freely selectable activation function. The output layer is followed by a softmax function. The network parameters are optimized by backpropagating an error to be minimized in a supervised learning process. This is done by a gradient descent method such as the Adam optimization algorithm. The cross-entropy loss serves as the error or cost function.This allows for particularly adaptable testing and diagnosis and thus allows for the most comprehensive detection of errors.

[0052] Method step a) can preferably be carried out using at least two acoustic sensors. This measure allows measurement signals from the first sensor and the second sensor to be evaluated in combination, thereby improving the accuracy and reliability of the evaluation.

[0053] Furthermore, the use of two or more acoustic sensors allows for the direction and / or distance of the sound source relative to the sensors to be determined. This allows not only the type of faulty component but also its precise location to be determined. Diagnosis can thus be performed even more precisely.

[0054] More preferably, a microphone array or an array of a plurality of sensors can be used, as already described above.

[0055] Furthermore, it is helpful if at least one sensor can be connected underfloor to a car body of the vehicle. More preferably, at least one sensor can be arranged on the bogie or underbody of the car body of a rail vehicle. This embodiment enables particularly advantageous production. This is because a computing unit can then be arranged in the car body and can serve as an evaluation unit, so that the spatial proximity means that the sensor data can be evaluated without complex peripherals. Instead, short signal paths are achieved between the sensor and the computing unit, thus achieving a high connection quality. Furthermore, replacing the chassis as well as retrofitting and conversion processes are easy to carry out, since no signal connections need to be provided between the chassis and the car body.

[0056] By connecting a sensor to the car body under the floor, a suspended arrangement of the first sensor is also made possible, whereby the sensor protrudes, for example, into the space between the running gear.

[0057] For further advantages and technical features of the method, reference is hereby made to the description of the system, the vehicle, the figures and the description of the figures, and vice versa.

[0058] Also described is a system for carrying out at least one step selected from a diagnosis and monitoring for vehicles, in particular for a drive system of a rail vehicle, wherein at least one acoustic sensor and at least one computing unit are provided, which are connected to one another in a signal-transmitting manner, wherein it is further provided that the computing unit is designed to carry out a method as described above.

[0059] The device described here is thus designed in particular to carry out a method as described above. This can be achieved in a manner understandable to a person skilled in the art by implementing appropriate software on the computing unit or a memory thereof. The computing unit can preferably have a classifier for this purpose. This essentially results in the advantages described above. This enables reliable detection and diagnosis of fault states in the vehicle. This makes reliable monitoring and diagnosis of fault states possible.

[0060] For further advantages and technical features of the system, reference is made to the description of the method, the vehicle, the figures and the description of the figures and vice versa.

[0061] Also described is a vehicle, in particular a rail vehicle, which is characterized in that the vehicle comprises a previously described system for carrying out at least one step selected from a diagnosis and monitoring for vehicles.

[0062] From the above, it is clear that the advantages described above essentially result. This enables reliable detection and diagnosis of vehicle fault conditions. This allows for reliable monitoring and diagnosis of fault conditions.

[0063] For further advantages and technical features of the vehicle, reference is hereby made to the description of the method, the system, the figures and the description of the figures, and vice versa.

[0064] Further details, features, and advantages of the subject matter of the invention emerge from the dependent claims and from the following description of the figures. The figures show: Fig. 1: a schematic plan view of a section of an exemplary chassis of a rail vehicle with acoustic sensors arranged in a chassis space of an exemplary first embodiment of a device according to the invention;

[0065] Fig. 2 A schematic plan view of a section of an exemplary chassis of a rail vehicle with acoustic sensors arranged on the chassis outer sides of an exemplary second embodiment of a device according to the invention; and

[0066] Fig. 3 A schematic side view of a section of an exemplary rail vehicle with acoustic sensors arranged on the underside of a car body of an exemplary third embodiment of a device according to the invention; and

[0067] Fig. 4 A flow chart for an exemplary embodiment of a method according to the invention including the use of a trained classifier.

[0068] Fig. 1 shows a schematic floor plan of a section of an exemplary running gear of a rail vehicle with acoustic sensors arranged on or at the running gear of an exemplary first embodiment of a device according to the invention for diagnosing and monitoring the rail vehicle, with which a corresponding method according to the invention can be carried out. The running gear can generally be selected and can comprise, for example, internally mounted or externally mounted running gears.

[0069] A first sensor 1, a second sensor 2, a third sensor 3, a fourth sensor 4, a fifth sensor 5, a sixth sensor 6, a seventh sensor 7 and an eighth sensor 8 are provided, which are arranged in a chassis space 9 which is delimited by a first wheel 10 and a second wheel 11 of a first wheel set 12 of the chassis.The first sensor 1 is screwed to a first longitudinal member 14 of a chassis frame 16 of the chassis, the second sensor 2 to a second longitudinal member 15 of the chassis frame 16, the third sensor 3 to a first wheelset bearing housing 17 of the chassis, the fourth sensor 4 to a second wheelset bearing housing 18 of the chassis, the fifth sensor 5 to a first swing arm 19 of a first wheelset guiding device 21 of the chassis, the sixth sensor 6 to a second swing arm 20 of a second wheelset guiding device 22 of the chassis, the seventh sensor 7 to a motor housing of a motor 23 of the chassis and the eighth sensor 8 to a gearbox housing of a gearbox 24 of the chassis.

[0070] The motor 23 and the transmission 24 are connected to the chassis frame 16, and the transmission 24 is coupled to the first wheel set 12. A clutch 25 is arranged between the motor 23 and the transmission 24 to transmit drive forces and torques.

[0071] The first wheelset 12 is coupled to the chassis frame 16 via the first wheelset bearing housing 17, a first primary spring 26 and the first wheelset guide device 21 as well as a second wheelset bearing, the second wheelset bearing housing 18, a second primary spring 27 and the second wheelset guide device 22.

[0072] The chassis has a second wheelset (not shown) which is connected to the chassis frame 16 via further wheelset bearings, further wheelset bearing housings, further primary springs and further wheelset guide devices, which are also not shown.

[0073] The first sensor 1, the second sensor 2, the third sensor 3, the fourth sensor 4, the fifth sensor 5, the sixth sensor 6 and the seventh sensor 7 are designed as measuring microphones. The eighth sensor 8 is designed as a microphone array, i.e. as a microphone group, with a first measuring cell 28, a second measuring cell 29 and a third measuring cell 30. The measuring microphones, the first measuring cell 28, the second measuring cell 29 and the third measuring cell 30 are designed, for example, as omnidirectional condenser free-field microphones with permanent polarization, but can basically have a selectable design or operating mode.

[0074] Furthermore, the first sensor 1 has a first

[0075] digitization unit 31 , the second sensor 2 a second

[0076] digitization unit 32 , the third sensor 3 a third

[0077] digitization unit 33 , the fourth sensor 4 a fourth

[0078] digitization unit 34 , the fifth sensor 5 a fifth

[0079] digitization unit 35 , the sixth sensor 6 a sixth

[0080] Digitization unit 36 ​​, the seventh sensor 7 a seventh digitization unit 37 and the eighth sensor 8 an eighth digitization unit 38 , in which a digitization 39 of acoustic measuring signals is carried out.

[0081] The first sensor 1 is arranged in the acoustic near field of the first wheel 10, and the second sensor 2 is arranged in the acoustic near field of the second wheel 11. Wheel noises are recorded by the first sensor 1 and the second sensor 2 in order to detect wheel polygons, flat spots, etc.

[0082] The third sensor 3 and the fifth sensor 5 are arranged in the acoustic near field of the first wheelset bearing and the first primary spring 26, the fourth sensor 4 and the sixth sensor 6 are arranged in the acoustic near field of the second wheelset bearing and the second primary spring 27.

[0083] The third sensor 3, the fourth sensor 4, the fifth sensor 5, and the sixth sensor 6 record axle bearing noises to detect bearing damage. Furthermore, spring parameters of the first primary spring 26 and the second primary spring 27 can be identified. The seventh sensor 7 is arranged in the acoustic near field of the clutch 25 to record clutch noises and detect clutch damage.

[0084] The first measuring cell 28 of the eighth sensor 8 is arranged in the acoustic near field of an invisible pinion, the second measuring cell 29 in the acoustic near field of an invisible gear stage and the third measuring cell 30 in the acoustic near field of an invisible large gear. The eighth sensor 8 is used to record gear noises in order to detect gear damage. The microphone system consisting of the first measuring cell 28, the second measuring cell 29 and the third measuring cell 30 makes it possible to localize the gear noises. If, for example, the first measuring cell 28 records acoustic measurement signals whose sound pressure levels are greater than a limit value defined for the measurement signals of the first measuring cell 28, this indicates damage to the pinion.

[0085] The first digitization unit 31, the second digitization unit 32, the third digitization unit 33, the fourth digitization unit 34, the fifth digitization unit 35, the sixth digitization unit 36, the seventh digitization unit 37 and the eighth digitization unit 38 have antennas (not shown in Fig. 1), by means of which measurement signals recorded by the first sensor 1, the second sensor 2, the third sensor 3, the fourth sensor 4, the fifth sensor 5, the sixth sensor 6, the seventh sensor 7 and the eighth sensor 8 are transmitted to a computing unit 40, also not shown in Fig. 1.

[0086] The computing unit 40 is arranged in a car body 41 of the rail vehicle (not shown in Fig. 1) and is connected via a cable to a radio-based data transmission unit 42 (not shown in Fig. 1) in the roof area of ​​the car body 41. However, the computing unit 40 can also be arranged under the floor of the car body 41.

[0087] The first sensor 1, the second sensor 2, the third sensor 3, the fourth sensor 4, the fifth sensor 5, the sixth sensor 6, the seventh sensor 7 and the eighth sensor 8 are supplied with electricity via batteries not visible in Fig. 1.

[0088] The computing unit 40 and the data transmission unit 42 are powered by a power supply device of the rail vehicle (not shown in Fig. 1) or can also be supplied with energy in another way, for example battery-based.

[0089] In the computing unit 40, an evaluation of the digitized measurement signals is carried out, as described in connection with Fig. 4. The result data from this evaluation are transmitted via the cable to the data transmission unit 42 and from there sent via radio to a maintenance facility (not shown in Fig. 1), i.e., to an infrastructure facility outside the rail vehicle.

[0090] In Fig. 2, a schematic floor plan of a section of an exemplary chassis of a rail vehicle with acoustic sensors arranged on the chassis outer sides of an exemplary second embodiment of a device according to the invention for diagnosis and monitoring of the rail vehicle is disclosed.

[0091] A first sensor 1 and a second sensor 2, which are configured as microphones and thus as acoustic sensors, are provided on the outer side of a chassis frame 16. The first sensor 1 is laterally connected to a first longitudinal member 14 of the chassis frame 16 via a first digitization unit 31, and the second sensor 2 is laterally connected to the first longitudinal member 14 via a second digitization unit 32.

[0092] The first longitudinal member 14 is connected via a cross member 43 to a second longitudinal member (not shown in Fig. 2) which is arranged opposite the first longitudinal member 14. A first wheelset 12 and a second wheelset 13 are coupled to the chassis frame 16 via a first wheelset bearing 44 and a second wheelset bearing 45 and, not shown in Fig. 2, via a third wheelset bearing and a fourth wheelset bearing. Furthermore, a first primary spring 26, a second primary spring 27 and, not shown in Fig. 2, a third primary spring and a fourth primary spring are provided between the chassis frame 16 on the one hand and the first wheelset 12 and the second wheelset 13 on the other hand. A first secondary spring 46 and a second secondary spring 47 (not shown in Fig. 2) are arranged between the chassis frame 16 and a car body 41 of the rail vehicle (not shown in Fig. 2). 2 second secondary springs, not shown, which are connected to the cross member 43, are arranged.

[0093] In the exemplary second embodiment of a device according to the invention, a chassis space 9 is provided between the first longitudinal member 14 and the second longitudinal member. However, no sensors are provided in the chassis space 9 in the exemplary second embodiment.

[0094] The first sensor 1 detects noises from the first axle box bearing 44 and the first primary spring 26. The second sensor 2, synchronously with the first sensor 1, detects noises from the second axle box bearing 45 and the second primary spring 27. This allows bearing and spring damage to be detected and spring parameters to be identified.

[0095] First measurement signals from the first sensor 1 and second measurement signals from the second sensor 2 are evaluated in a computing unit 40 arranged in the car body 41, as disclosed, for example, in Fig. 3 for an exemplary third embodiment of a device according to the invention. Due to their advantageously time-synchronous detection, a phase relationship between the first measurement signals and the second measurement signals is taken into account, whereby noise events such as driving over rail joints can be detected and categorized.

[0096] With regard to an electricity supply and data transmission to the computing unit 40, the first sensor 1 and the second sensor 2 are designed, for example, in the same way as those sensors which are described in connection with Fig. 1. The computing unit 40 is, as described in connection with Fig. 1 for an exemplary first embodiment of a device according to the invention, connected to a data transmission unit 42 which is arranged in the car body 41 and via which diagnostic and monitoring data formed from the first measurement signals and the second measurement signals are sent to an infrastructure facility outside the rail vehicle.

[0097] In Fig. 3, a schematic side view, for example greatly simplified compared to Fig. 1, of a section of an exemplary rail vehicle with a chassis and with acoustic sensors arranged underfloor on an underside of a car body 41 of an exemplary third embodiment of a device according to the invention is shown.

[0098] The running gear has a running gear frame 16, which is coupled to the car body 41 via a first secondary spring 46 and a second secondary spring (not shown in Fig. 3). A first wheelset 12 and a second wheelset 13 are coupled to the running gear frame 16 via wheelset bearings, wheelset guide devices, and primary springs (not shown in Fig. 3).

[0099] A first sensor 1 and a second sensor 2 are provided, which are designed as acoustic sensors and are arranged so as to project from above into a space between the wheels of the first wheel set 12 and the second wheel set 13 of the chassis, or between two longitudinal members of the chassis frame 16, i.e., into a chassis intermediate space 9. The chassis intermediate space 9 is bounded at the top by the upper limits of the wheels and at the bottom by the lower limits of the wheels.

[0100] However, it is also possible that the first sensor 1 and the second sensor 2 are provided in the region of the chassis frame 16, for example in recesses of the chassis frame 16, and the chassis intermediate space 9 is limited upwards by an upper edge of the chassis frame and downwards by a lower edge of the chassis frame.

[0101] Furthermore, a third sensor 3 and a fourth sensor 4 are shown, which are arranged directly beneath the car body 41. This demonstrates the variability of the positioning of the acoustic sensors within the scope of the present invention.

[0102] The first sensor 1 is connected to a first holder 47, and the second sensor 2 is connected to a second holder 48. The first holder 47 and the second holder 48 are connected to the underside of the car body 41. It can be seen that the embodiment according to Figure 3 has internally mounted wheel sets 12, 13.

[0103] The first sensor 1 is arranged in the area of ​​a first wheel-rail contact of the first wheelset 12, the second sensor 2 in the area of ​​a second wheel-rail contact of the second wheelset 13. As a result, wheel noises are recorded synchronously by the first sensor 1 as first measurement signals and by the second sensor 2 as second measurement signals, digitized by means of a first digitization unit 31 of the first sensor 1 and a second digitization unit 32 of the second sensor 2, and then transmitted to a computing unit 40 arranged in the car body 41.

[0104] The first digitization unit 31 and the second digitization unit 32 are designed as analog-to-digital converters.

[0105] For transmitting the digitized first measurement signals and the digitized second measurement signals, the first digitization unit 31 has a first antenna 49. The second digitization unit 32 is connected to the computing unit 40 via a first cable 52, which is guided in the hollow second holder 48. The computing unit 40 has a second antenna 50 for receiving radio signals emitted by the first antenna 49.

[0106] In the computing unit 40, the digitized first measurement signals and the digitized second measurement signals are filtered and analyzed, as described in connection with Fig. 4. Status indicators thus determined are transmitted from a data storage unit 55 connected to the computing unit 40 as result data via a second cable 53, a data bus 56 designed as a multi-function vehicle bus (MVB), and a third cable 54 to a data transmission unit 42.

[0107] The result data is also transmitted via the data bus 56 to a driver's cab (not shown) of the rail vehicle, where a display based on the condition indicators indicates to a train driver whether the wheels are damaged, worn, or fault-free. According to the invention, however, it is also conceivable to transmit the result data to the data transmission unit 42 without interposing the data bus 56.

[0108] The data transmission unit 42 is arranged in the roof area of ​​the car body 41 and has a third antenna 51. The status indicators are transmitted via the third antenna 51 to an infrastructure facility (not shown) outside the rail vehicle.

[0109] The infrastructure facility is designed as a maintenance station for the rail vehicle and maintenance and / or repair measures are planned there (e.g. the procurement of spare parts is initiated) on the basis of the transmitted condition indicators, i.e. on the basis of diagnostic and / or monitoring information.

[0110] For training, data (training and validation data) is collected, based on which the detection / classification method is designed and optimized. The resulting method can then be used in the application phase 58 for detection / classification.

[0111] Figure 4 shows a flowchart describing a method according to the invention for monitoring and diagnosing the vehicle. Figure 4 shows an exemplary embodiment in which a classifier is first trained and then applied to carry out the described method.

[0112] Figure 4 shows, on the left, the training phase 57 for training the classifier, and on the right, the application phase 58, in which the classifier is applied. In the training phase 57, data is first collected 59, with the collected data being divided into training data 60 and validation data 61. Feature extraction 62, 63 takes place to extract the features from the training data 60 and validation data 62. Based on the training data 60 and the extracted features, the classifier is trained 64.

[0113] The resulting data is incorporated into validation 65 and allows validation 65 of the classifier. If necessary, the training phase, including validation, is repeated. An iterative approach is also possible.

[0114] With reference to the application phase 58, data collection 66 initially takes place. This is carried out according to the invention by method step a) and thus by receiving an acoustic airborne sound signal by at least one acoustic sensor (1, 2, 3, 4, 5, 6, 7, 8). For this purpose, the provided microphones or acoustic sensors, i.e. in particular the first sensor 1, the second sensor 2, the third sensor 3, the fourth sensor 4, the fifth sensor 5, the sixth sensor 6, the seventh sensor 7 and the eighth sensor 8, are used.

[0115] Furthermore, feature extraction 67 is performed, wherein in this step, according to method step b), at least one feature is determined from the airborne sound signal. The feature is, in particular, a so-called feature known from speech recognition.

[0116] Based on the feature or features determined in this way, an application 68 of the classifier can be carried out, wherein the data from the validation 65 or from the validated classifier additionally flow into the application 68 of the classifier. A condition diagnosis 69 can be carried out by the application 68 of the classifier. Such a condition diagnosis 69 comprises, according to method step c), carrying out at least one step selected from a diagnosis and a monitoring of at least one vehicle component based on at least one feature determined in method step b).

[0117] Accordingly, the monitoring and diagnosis of a component in the non-limiting embodiment described here are achieved by means of a feature-based classification approach based on the analysis of airborne sound recordings. The features used correspond to the state of the art for tasks in the areas of speech recognition, speaker recognition and the classification of acoustic scenes. For example, but not limited to, the so-called Mel-Frequency Cepstral Coefficients (MFCCs) are extracted from airborne sound recordings. The MFCCs enable a compact representation of the spectrum of a signal by combining the cepstrum of a signal with an approximately logarithmic scaling of the frequency axis.

[0118] The cepstrum of a signal is obtained by applying the inverse Fourier transform for discrete-time signals to the logarithm of the discrete-time Fourier transform of the signal. To calculate the MFCCs, the signal is first divided into individual windows by windowing. An N-valued discrete Fourier transform is then performed for each of these windows. The power magnitude spectrum is then filtered with a Mel filter bank consisting of overlapping triangular filters. Finally, the MFCCs are generated by the inverse discrete cosine transform of the logarithm of the filter bank energies. The features obtained in this way are then used to train an artificial neural network that acts as a classifier. The classifier comprises, for example and in no way limiting, at least one neural network, such as a fully connected feedforward multi-layer perceptron (MLP).The MLP consists of an input layer, two hidden layers, and an output layer. The width of the input layer and the output layer are fixed, for example by the number of features used and the number of classes. The width of the hidden layers can be freely chosen. The first and second hidden layers are each followed by a freely selectable activation function. The output layer is followed by a softmax function. The network parameters are optimized by backpropagating an error to be minimized in a supervised learning process. This is done using a gradient descent method such as the Adam optimization algorithm. The cross-entropy loss serves as the error or cost function.

[0119] Although the invention has been illustrated and described in detail by the preferred embodiment, the invention is not limited to the disclosed examples and other variations can be derived therefrom by those skilled in the art without departing from the scope of the invention.

Claims

Patent claims 1. Method for carrying out at least one step selected from a diagnosis and monitoring for vehicles, in particular for a drive system of a rail vehicle, at least comprising the following method steps: a) receiving an acoustic airborne sound signal by at least one acoustic sensor (1, 2, 3, 4, 5, 6, 7, 8); b) determining at least one feature from the airborne sound signal; and c) carrying out at least one step selected from a diagnosis and monitoring of at least one vehicle component based on at least one feature determined in method step b).

2. Method according to claim 1, characterized in that the features determined in method step b) comprise at least the cepstrum of the airborne sound signal.

3. Method according to claim 2, characterized in that in method step b) Mel-Frequency Cepstral coefficients are extracted from the received airborne sound signal.

4. Method according to one of claims 1 to 3, characterized in that in method step b) LPC coefficients are extracted from the received airborne sound signal.

5. Method according to one of claims 1 to 4, characterized in that in method step b) PLP coefficients are extracted from the received airborne sound signal. Method according to one of claims 1 to 5, characterized in that in method step b) log-mel spectral data is extracted from the received airborne sound signal. Method according to one of claims 1 to 6, characterized in that in method step b) amplitude modulation spectral data is extracted from the received airborne sound signal. Method according to one of claims 1 to 7, characterized in that the method comprises the further method steps: d) training a classifier with the at least one feature using training data and validation data or defining a classifier; wherein e) method step c) is carried out using the classifier. Method according to claim 8, characterized in that at least one artificial neural network is used as the classifier. Method according to claim 9, characterized in that the neural network comprises a multilayer perceptron.Method according to one of claims 1 to 10, characterized in that method step a) is carried out using at least two acoustic sensors (1, 2, 3, 4, 5, 6, 7, 8). Method according to claim 11, characterized in that the at least two acoustic sensors are designed as a microphone array. Method according to one of claims 1 to 12, characterized in that at least one acoustic sensor (1, 2, 3, 4, 5, 6, 7, 8) is arranged on the bogie or underbody of the car body of a rail vehicle. Method according to one of claims 1 to 13, characterized in that at least one acoustic sensor (1, 2, 3, 4, 5, 6, 7, 8) comprises a digitization unit (31, 32, 33, 34, 35, 36, 37, 38). System for carrying out at least one step selected from a diagnosis and monitoring for vehicles, in particular for a drive system of a rail vehicle, wherein at least one acoustic sensor (1, 2, 3, 4, 5, 6, 7, 8) and at least one computing unit (40) are provided, which are connected to one another in a signal-transmitting manner, characterized in that the computing unit (40) is designed to carry out a method according to one of claims 1 to 14.Vehicle, in particular rail vehicle, characterized in that the vehicle comprises a system for performing at least one step selected from a diagnosis and monitoring for vehicles according to claim 15.

Images