Bogie abnormality detection system and bogie abnormality detection method

Abstract

In order to provide a bogie abnormality detection system capable of simply detecting on the vehicle side, without depending on an acoustic sensor installed on the ground, an abnormality of a vehicle which is in a noisy traveling state, the bogie abnormality detection system that detects an abnormality of a bogie of the vehicle traveling on a track comprises: a vibration sensor or an acoustic sensor that senses the vibration of the bogie and outputs vibration data of the bogie; and a vehicle information control device that uses the vibration data and the traveling state information of the vehicle as inputs. The vehicle information control device has: an extraction unit that extracts, from the vibration data, on the basis of the traveling state information, the vibration data at least either during linear zone traveling or during coasting operation as vibration information; and a detection unit that detects, using a predictive model calculated by machine learning in which the vibration information and abnormality information of the bogie are used as training data, an abnormality of the bogie from the vibration information extracted by the extraction unit.

Description

Title of Invention: BOGIE ABNORMALITY DETECTION SYSTEM AND BOGIE ABNORMALITY DETECTION METHOD Technical Field

[0001] The present invention relates to a bogie abnormality detection system and a bogie abnormality detection method used when an abnormality occurs in a bogie of a railway or track vehicle . Background Art

[0002] Conventionally, as a method for inspecting an occurrence of such an abnormality as cracking, chipping, breaking, and peeling in a bogie of a railway vehicle, a visual inspection and a hammering test are performed. However, in these inspections performed in a stationary state, a point different from a point vibrating (a point where a load is applied) during traveling is hammered; therefore, a state of a traveling vehicle cannot be observed. Further, since the inspections are performed by a human, the accuracy inevitably varies .

[0003] To solve this problem, Nonpatent Literature 1 and Patent Literature 1 disclose a technology for detecting any abnormality in a bogie using an acoustic sensor installed by the side of a railway line. Citation List Patent Literature

[0004] Patent Literature 1: Japanese Unexamined Patent Application Publication No. 2020-73366 Patent Literature 2: Japanese Unexamined Patent Application Publication No. 2019-167079 Nonpatent Literature

[0005] Nonpatent Literature 1: https: / / www.nikkei.com / article DGXZQOHD225BY0S0A221C2000000 / Nikkei Inc. West Japan Railway Company Prevents Accident by "Distinguishing Shinkansen's Traveling Sound by AI" Summary of Invention Technical Problem

[0006] The method disclosed in Patent Literature 1 is limited to a detection of an abnormality of a bogie of a vehicle passing by the proximity of an acoustic sensor; in addition, detection by an acoustic sensor is applicable only to such a line as an elevated railway of a Shinkansen involving relatively little noise and it is difficult to apply this method to a local line.

[0007] In the method disclosed in Patent Literature 2, meanwhile, the thrust of an actuator is varied between a linear zone and a curved zone depending on the vibration of resonance frequency of a bogie; but just by combining the methods disclosed in these two Patent Literatures, vibration data of a curved zone involving much noise cannot be removed and it is difficult to detect an abnormality of a bogie.

[0008] In consideration of the foregoing, consequently, the present invention has been made and it is an object of the present invention to provide a technology for detecting an abnormality of a vehicle in a traveling state without dependence on an acoustic sensor installed on the ground. Solution to Problem

[0009] To solve the above problem, one of typical bogie abnormality detection systems according to the present invention is a system that detects an abnormality in a bogie of a vehicle traveling on a track. The system includes: a vibration sensor or an acoustic sensor that senses vibration of a bogie and outputs vibration data of the bogie; and a vehicle information control device that uses vibration data and traveling state information of a vehicle as inputs. The vehicle information control device includes: an extraction unit that extracts, as vibration information, vibration data at least either during linear zone traveling or during coasting operation from vibration data based on traveling state information; and a detection unit that detects an abnormality of a bogie from vibration information extracted by the extraction unit using a predictive model calculated by machine learning in which vibration information and abnormality information of the bogie as training data. Advantageous Effects of Invention

[0010] According to the present invention, the accuracy of detection of an abnormality of a bogie can be enhanced by: determining the time of at least either one of coasting operation and linear zone traveling; filtering frequencies; and using a predictive model calculated by machine learning to easily separate only a characteristic frequency. Other problems, configurations, and effects than described above will be apparent from following description of embodiments . Brief Description of Drawings

[0011] FIG. 1 is a drawing illustrating an example of an overall configuration of a bogie abnormality detection system in an embodiment of the present invention. FIG. 2 is a drawing illustrating an example of a configuration of a vehicle information control device and a relation with other devices. FIG. 3 is a drawing illustrating an example of a configuration of a predictive data calculation device and a relation with other devices. FIG. 4 is a drawing illustrating an example of a configuration of machine learning by Deep Learning using a Sigmoid function. FIG. 5 is a drawing illustrating a calculation graph of a multiplication part. FIG. 6 is a drawing illustrating a calculation graph of an addition part. FIG. 7 is a drawing illustrating a calculation graph of a Sigmoid part. FIG. 8 is a drawing illustrating of a spectrogram of sound data sampled in a vehicle during actual station to station traveling. Description of Embodiments

[0012] Hereafter, a description will be given to a mode for carrying out the present invention as an embodiment with reference to the drawings. The present invention is not limited to this embodiment. In the drawings, an identical part will be marked with an identical reference numeral and symbol. Embodiment

[0013] FIG. 1 is a drawing illustrating an example of an overall configuration of a bogie abnormality detection system in an embodiment of the present invention. First, a description will be given to a mode of processing until transmission of information from the vehicle side to the ground side.

[0014] A vibration sensor or an acoustic sensor S2 mounted in a vehicle SI senses vibration of a bogie to acquire vibration data and outputs (transmits) the vibration data to a vehicle information control device S4.

[0015] A traveling information transmission device S3 acquires the following pieces of information as traveling state information of the vehicle and outputs (transmits) the acquired information to the vehicle information control device S4, the acquired information including: positional information such as information about traveling, speed, rotational speed of a wheel, GPS information, a signal from a ground coil, and a kilometrage, that enables a traveling position of the vehicle to be determined; a state (power running, coasting, braking, and the like) of a master controller; and braking and driving state information, such as driving state information of a driving device of a vehicle motor and braking state information of a brake.

[0016] Based on the traveling state information received from the traveling information transmission device S3, the vehicle information control device S4 filters vibration data inputted from the vibration sensor or acoustic sensor S2.

[0017] When during this filtering, information of at least either one of a period of time and a zone during which vehicles pass by each other is included in the traveling state information, vibration data acquired during the period of time or zone contains vibration data produced by the passing-by; therefore, an influence of vibration produced by passing-by can be removed by removing the vibration data by filtering.

[0018] Vibration information acquired by filtering is transmitted, together with traveling state information, to a ground-side communication device S6 through a vehicle-side communication device S5. Subsequently, a description will be given to a mode of processing until transmission of processed information received on the ground side to the vehicle side. The ground-side communication device S6 transmits information received from the vehicle-side communication device S5 to a predictive data calculation device S7.

[0020] The predictive data calculation device S7 manages information on a vehicle-by-vehicle basis, and, when any problem or the like is found in a bogie as a result of maintenance performed at a vehicle maintenance and repair shop S8, stores the information related thereto and performs machine learning between the stored information and information received from the ground-side communication device S 6 . Information of machine learning performed at the predictive data calculation device S7 is transmitted to the vehicle-side communication device S5 through the ground-side communication device S6.

[0021] A description will be given to a mode of processing on the vehicle side based on information received from the ground side . The vehicle-side communication device S5 transmits information of machine learning received from the ground-side communication device S6 to the vehicle information control device S4.

[0022] The vehicle information control device S4 filters vibration data received from the vibration sensor or acoustic sensor S2 using traveling state information received from the traveling information transmission device S3 and computes the vibration data using information of machine learning performed on the ground side. When as the result of the computation, any abnormality is found, the presence of the abnormality is communicated (notified of) to a crew S10 in the vehicle SI or a maintenance personnel Sil manned on the ground or the like.

[0023] Subsequently, a description will be given to operation of the vehicle-side vehicle information control device S4. FIG. 2 is a drawing illustrating an example of a configuration of the vehicle information control device S4 and a relation with other devices. A traveling state calculation unit S44 determines whether the present traveling state is coasting operation or linear zone traveling from traveling state information received from the traveling information transmission device S3 and outputs determination information.

[0024] With respect to determination of coasting operation among the determinations, not only coasting period of time but also a zone in which driving state information of a driving device of a vehicle motor takes not more than a predetermined threshold value may be determined to be equivalent to coasting operation . With respect to determination of linear zone traveling, the zone need not be a perfect linear zone and a determination may be made by providing a curvature with a threshold value.

[0025] This determination information is stored in a storage unit S45 together with such traveling state information as speed and acceleration and is transmitted also to a vibration extraction unit S41.

[0026] The vibration extraction unit S41 filters vibration data using vibration data from the vibration sensor or acoustic sensor S2 and determination information from the traveling state calculation unit S44. By this filtering, vibration data at the time of at least either one of coasting operation and linear zone traveling is extracted as vibration information and the vibration data is transmitted to a predictive model confirmation unit S42 and the storage unit S45. This vibration information is stored in the storage unit S45.

[0027] Meanwhile, a predictive model storing unit S46 stores a predictive model, received from a ground-side predictive data calculation device S7 through the ground-side communication device S6 and the vehicle-side communication device S5, in the storage unit S45. The predictive model storing unit S46 reads traveling state information and vibration information from the storage unit S45 and transmits these pieces of information to the ground-side communication device S6 through the vehicleside communication device S5.

[0028] The predictive model confirmation unit S42 reads a predictive model and traveling state information stored in the storage unit S45 and combines the predictive model and traveling state information with vibration information received from the vibration extraction unit S41 to determine whether the vibration state involves an abnormality in a bogie to detect the abnormality. When the presence of an abnormality in a bogie is detected, this detection of the abnormality is notified to a crew or the like through an abnormal state display unit S43.

[0029] As a basis for detection of the presence of an abnormality in a bogie, whether the presence of an abnormality in a bogie is determined a predetermined number of times or the presence of an abnormality in a bogie is determined for a certain period of time may be taken into account. An example of calculation at the predictive model confirmation unit S42 will be described later in the section of <Calculation Example Using Deep Learning>.

[0030] Subsequently, a description will be given to operation of the ground-side predictive data calculation device S7. FIG. 3 is a drawing illustrating an example of a configuration of the predictive data calculation device S7 and a relation with other devices. Traveling state information and vibration information received from the ground-side communication device S6 are stored in the storage unit S75 through a received data storing unit S74 on a vehicle-by-vehicle basis.

[0031] Further, inspection information indicating which bogie of a vehicle an abnormality such as crevice, cracking, and breaking has occurred and other like items are stored in the storage unit S75 from the maintenance and repair shop S8 through a storage data input unit S72.

[0032] The predictive model calculation unit S73 uses inspection information, traveling state information, and vibration information stored in the storage unit S75 on a vehicle-by-vehicle basis to build a predictive model. An example of predictive model building will be described later in the section of CGalculation Example Using Deep Learning>. A built predictive model is stored in the storage unit S75.

[0033] A predictive model transmission unit S71 transmits the predictive model stored in the storage unit S75 to the vehicle-side communication device S5 through the ground-side communication device S6.

[0034] cCalculation Example Using Deep Learning> As an example of determination methods using machine learning, a description will be given to an example of calculation utilizing Deep Learning. Of the varied information cited above, only some may be utilized or information may be limited only to those obtained during linear zone traveling. The determination method is not limited to a technique by Deep Learning and a technique based on Bayesian statistics using conjugate distribution or MCMC method, a technique such as SVM and t-test capable of distinguishing statistical normal data and other data, or any other like technique may be used. As an example of calculation using Deep Learning, in the present embodiment, a Sigmoid function is used; however, the present invention is not limited to this and any other technique such ReLU may be used.

[0035] FIG. 4 is a drawing illustrating an example of a configuration of machine learning by Deep Learning using a Sigmoid function. First, at an input data layer S91, information referenced to when generating vibration information is inputted.

[0036] Table 1 shows an example of input information. In the example shown in Table 1, with respect to frequency X (X=l, 2, 3, ...), vibration information is subjected to Fourier transform and the magnitude of vibration divided by a certain frequency is taken as input data. [Table 1] Frequency 1 Frequency Frequency Frequency 4 Acceleration Speed Curvature Gradient 0 . 07 0 . 08 0.30 0 . 02 . . . 0 . 7 70 0 . 1 0

[0037] In Table 1, as additional data, acceleration, speed, a curvature of a traveling zone, and gradient data are taken as input information. Only some in Table 1 may be added as long as vibration information is inputted as basic data.

[0038] Aside from the information shown in Table 1, information that can be referenced to when vibration is produced may be added, such information including: other traveling state information, occupancy rate, wind speed, wind direction (including those obtained by vector decomposing to the frontward and sideward of a vehicle), humidity, temperature, rainfall, and the like. Irrelevant data may be added unless the data has an influence on a calculation result.

[0039] Subsequently, at a normalization layer S92, input data is normalized. With respect to normalization, the maximum data of input data (hereafter, referred to as "input vector") may be taken as 1 and normalization may be based on relative magnitude or normalization may be performed based on the possible maximum value in each piece of data. In the present embodiment, a normalized input data matrix is set as X = (xl, x2, x3, ... xn).

[0040] Subsequently, at an Affine layer S93, a matrix W of weighted signals is set to [Formula 1] W = J w 1 h e 2 1, w 3 1 , '■ * • , w rn. 1 | S U- .1 2 : W 2 2 , 'W 3 2 j s ’ * 4 W 2 $ $ vx in. Mi 2 is , v- 3 n .. e ’ • >v- And a bias matrix B is set to B = (bl, b2, b3, ..., bm) and the elements of a result matrix X is set to A = (al, a2, a3, ..., am) then, the result matrix A is obtained by the following formula : A = XW+B

[0041] As an example, when one element of the result matrix A is taken out, al = wl1*xl+wl2*x2 + ... +wln*xn+bl

[0042] With respect to the values of the matrixes W, B constituting the above-mentioned formula, numeric values may be put in advance as initial values, or random numeric values may be generated. In the above-mentioned example, al is still a linear expression and a complicated shape cannot be Gxpros s g d .

[0043] Subsequently, at a Sigmoid layer S94, the values are transformed into a non-linear value using a Sigmoid function. An example of a Sigmoid function includes the following function : Y = 1 / (1+exp (—x)) A value passed through a Sigmoid function is normalized between -1 and 1; therefore, the value can be utilized as input data to an Affine layer S95 at the next stage.

[0044] The above-mentioned processing (Affine layer S95, Sigmoid layer S96, ...) is repeated and the repeatedly processed data makes output data from an output data layer S97.

[0045] Subsequently, as processing at a predictive model calculation unit S73, with respect to the above output data, a training data matrix L is generated. In this example, the training data matrix L is set to L = (11, 12, 13, ..., Im) 11 is taken as a parameter representing no bogie abnormality; 12 is taken as a parameter representing an abnormality in a bogie part a; 13 is taken as a parameter representing an abnormality in a bogie part [3; ...

[0046] Here, 1 is taken as a case where an individual parameter applies and 0 is taken as a case where an individual parameter does not apply. In this case, for example, when a bogie does not involve an abnormality, a training data matrix L is set to L = (1, 0, 0, ..., 0) and when an abnormality occurs only in a bogie part a, a training data matrix L is set to L = (0, 1, 0, 0, ..., 0) (however, "..." is all 0.)

[0047] With respect to training data, a result of a regular inspection or a theoretical value may be used. When a result of a regular inspection is used, for example, training data may be applied to vibration information after the previous regular inspection.

[0048] When a theoretical value is used, for example, the following method may be used: a width of frequency produced from a bogie is predetermined; data of a width of allowable frequency and frequencies deviating therefrom are inputted as input data; and output data to be training data is caused to be learnt as 0 when a frequency is allowable and 1 when a frequency deviates.

[0049] Training data and output data are compared with each other. When there is a difference, it is reflected in a weighted signal and a bias in the Affine layer at the previous stage. As methods applicable in this case, a gradient method, an error backpropagation method, and the like are known.

[0050] In relation to the present embodiment, a description will be given to an example in which a weighted signal and a bias are modified using an error backpropagation method. In the error backpropagation method, a chain rule of differential is used. For example, when a multiplication part an addition part, and a Sigmoid part of a calculation formula are notated by a notation method called calculation graph, the notations respectively shown in FIG. 5, FIG. 6, and FIG. 7 are obtained. In each drawing, the notation above the arrow of the calculation graph is that obtained by calculating output data and the notation below the arrow of the calculation graph is that obtained from output data using an error backpropagation method. As shown in FIG. 5 to FIG. 7, a weighted signal and a bias value can be caused to be learnt so that correct output data is outputted.

[0051] Training data may be generated based on a failure part found in a vehicle inspection, such as a monthly inspection, an important part inspection, a general inspection, and an extra inspection or may be generated based on simulation data or the like.

[0052] As mentioned above, the predictive model calculation unit S73 transmits a matrix of a predictive model learnt using the input data layer S91 to the output data layer S97 to the vehicle-side communication device S5 through the ground-side communication device S6. This matrix of a predictive model is used at the vehicle-side predictive model confirmation unit S42 to determine any abnormality.

[0053] According to the present invention, information provided in the vehicle information control device S4 can be used to easily sort data with less noise and the accuracy of detection of an abnormality in a bogie can be enhanced. Since data with less noise can be sorted, sound with less noise can be extracted with a vibration sensor or an acoustic sensor additionally mounted in a vehicle without installing a vibration meter directly in a bogie.

[0054] FIG. 8 is a drawing showing a spectrogram of sound data sampled in a vehicle during actual station to station traveling. Between 130s and 250s during which coasting and linear zone traveling are performed, characteristic frequencies can be sampled and intense frequencies are distributed in a stripe pattern. However, it is found that in other spectrograms during curved zone traveling and power running, characteristic frequencies cannot be found. [ 0055] Up to this point, a description has been given to an embodiment of the present invention but the present invention is not limited to the above-mentioned embodiment and can be variously modified without departing from the subject matter of the present invention. List of Reference Signs

[0056] SI: vehicle, S2: vibration sensor or acoustic sensor, S3: traveling information transmission device, S4: vehicle information control device, S5: vehicle-side communication device, S6: ground-side communication device, S7: predictive data calculation device, S8: vehicle maintenance and repair shop S10: crew, Sil: maintenance personnel, S41: vibration extraction unit, S42: predictive model confirmation unit S43: abnormal state display unit, S44: traveling state calculation unit, S45: storage unit, S46: predictive model storing unit, S71: predictive model transmission unit S72: maintenance data input unit, S73: predictive model calculation unit, S74: received data storing unit, S75: storage unit, S91: input data layer, S92: normalization layer, S93, S95: Affine layer, S94, S96: Sigmoid layer, S97: output data layer

Claims

1. A bogie abnormality detection system that detects an abnormality of a bogie of a vehicle traveling on a track comprising:a vibration sensor or an acoustic sensor that senses vibration of the bogie and outputs vibration data of the bogie; anda vehicle information control device that uses the vibration data and traveling state information of the vehicle as inputs, whereinthe vehicle information control device includes:an extraction unit that, based on the traveling state information, extracts vibration data at least either during liner zone traveling or during coasting operation from the vibration data as vibration information; anda detection unit that detects an abnormality of the bogie from the vibration information extracted by the extraction unit using a predictive model calculated by machine learning in which the vibration information and abnormality information of the bogie are used as training data.

2. The bogie abnormality detection system according to Claim 1, whereinthe traveling state information contains driving state information of a motor driving device of the vehicle, and the extraction unit extracts the vibration data of a zone in which the driving state information is not more than a predetermined threshold value as the vibration information.

3. The bogie abnormality detection system according to Claim 1 or 2, whereinthe traveling state information contains information of at least either one of a period of time and a zone in which the vehicle passes by another vehicle, andthe extraction unit removes, from information of at least either one of the period of time and the zone, the vibration data at least either during the period of time or during the zone.

4. The bogie abnormality detection system according to any of Claims 1 to 3, whereinwhen a presence of an abnormality of the bogie is determined for a certain period of time or a certain number of times, the detection unit detects an abnormality of the bogie.

5. A bogie abnormality detection method comprising the steps of:sensing vibration of a bogie of a vehicle traveling on a track and acquiring vibration data of the bogie;based on traveling state information of the vehicle, extracting vibration data at least either during linear zone traveling or during coasting operation from the vibration data as vibration information;calculating a predictive model by machine learning in which the vibration information and abnormality information of the bogie are used as training data; andusing the predictive model to detect an abnormality of the bogie from the vibration information.The bogie abnormality detection method according to Claim 5, whereinthe traveling state information contains driving state information of a motor driving device of the vehicle, andthe vibration data during a zone in which the driving state information is not more than a predetermined threshold value is extracted as the vibration information.

7. The bogie abnormality detection method according to Claim 5 or 6, whereinthe traveling state information contains information of at least either one of a period of time and a zone in which the vehicle passes by another vehicle, andthe vibration data at least either during the period of time or during the zone is removed from information of at least either one of the period of time and the zone.

8. The bogie abnormality detection method according to any of Claims 5 to 7, whereinwhen a presence of an abnormality of the bogie is determined for a certain period of time or a certain number of times, an abnormality of the bogie is detected.INTERNATIONAL SEARCH REPORT International application No. PCT / JP2023 / 045324A. CLASSIFICATION OF SUBJECT MATTER B60L 3 / W(2019.01)i; B61L 25 / 04(2006.01)1 FI: B60L3 / 00 N; B61L25 / 04 According to International Patent Classification (IPC) or to both national classification and IPC B. FIELDS SEARCHED Minimum documentation searched (classification system followed by classification symbols) B60L3 / 00; B61L25 / 04 Documentation searched other than minimum documentation to the extent that such documents are included in the fields searched Published examined utility model applications of Japan 1922-1996 Published unexamined utility model applications of Japan 1971-2024 Registered utility model specifications of Japan 1996-2024 Published registered utility model applications of Japan 1994-2024 Electronic data base consulted during the international search (name of data base and, where practicable, search terms used) C. DOCUMENTS CONSIDERED TO BE RELEVANT Category* Citation of document, with indication, where appropriate, of the relevant passages Relevant to claim No. X JP 2021-46191 A (NABTESCO CORP.) 25 March 2021 (2021-03-25) paragraphs [0028]-[0029], [0095]-[0126], [0332], fig. 1-2, 9-12 1-8 | | Further documents are listed in the continuation of Box C. | V | See patent family annex. * Special categories of cited documents: “A” document defining the general state of the art which is not considered to be of particular relevance “D” document cited by the applicant in the international application ‘4E” earlier application or patent but published on or after the international filing date *4L” document which may throw doubts on priority claim(s) or which is cited to establish the publication date of another citation or other special reason (as specified) “O” document referring to an oral disclosure, use, exhibition or other means “P” document published prior to the international filing date but later than the priority date claimed “T” later document published after the international filing date or priority date and not in conflict with the application but cited to understand the principle or theory underlying the invention “X” document of particular relevance; the claimed invention cannot be considered novel or cannot be considered to involve an inventive step when the document is taken alone “Y” document of particular relevance; the claimed invention cannot be considered to involve an inventive step when the document is combined with one or more other such documents, such combination being obvious to a person skilled in the art document member of the same patent family Date of the actual completion of the international search Date of mailing of the international search report 07 February 2024 20 February 2024 Name and mailing address of the ISA / JP Authorized officer Japan Patent Office (ISA / JP) 3-4-3 Kasumigaseki, Chiyoda-ku, Tokyo 100-8915 Japan Telephone No.Form PCT / ISA / 210 (second sheet) (July 2022)INTERNATIONAL SEARCH REPORT Information on patent family members International application No. PCT / JP2023 / 045324Patent document cited in search report Publication date (day / month / year) Patent family member)s) Publication date (day / month / year) JP 2021-46191 A 25 March 2021 US 2021 / 0078619 Al paragraphs [0062]-[0063], [0139]-[0170], [0382], fig. 1-2, 9-12 EP 3792141 Al CN 112498413 A

Images