FORM 2
THE PATENTS ACT, 1970
(39 of 1970)
&
THE PATENTS RULES, 2003
COMPLETE SPECIFICATION
[See section 10, Rule 13]
SOUNDNESS DIAGNOSIS APPARATUS AND SOUNDNESS DIAGNOSIS
METHOD;
MITSUBISHI ELECTRIC CORPORATION, A CORPORATION ORGANISED
AND EXISTING UNDER THE LAWS OF JAPAN, WHOSE ADDRESS IS 7-3,
MARUNOUCHI 2-CHOME, CHIYODA-KU, TOKYO 100-8310, JAPAN
THE FOLLOWING SPECIFICATION PARTICULARLY DESCRIBES THE
INVENTION AND THE MANNER IN WHICH IT IS TO BE PERFORMED.
2
DESCRIPTION
TITLE OF THE INVENTION:
SOUNDNESS DIAGNOSIS APPARATUS AND SOUNDNESS DIAGNOSIS
METHOD5
Field
[0001] The present disclosure relates to a soundness
diagnosis apparatus that diagnoses soundness of a device
and a soundness diagnosis method.10
Background
[0002] Conventionally, in railway vehicles and the like,
sensors are used to detect states of devices mounted on the
railway vehicles and the like, and anomaly diagnosis is15
performed by using sensor data in order to promptly detect
occurrence of anomalies. Patent Literature 1 discloses a
technique of accurately diagnosing an anomaly of a brake in
consideration of various conditions such as a travel
section when an object to be diagnosed is a brake mounted20
on a railway vehicle.
Citation List
Patent Literature
[0003] Patent Literature 1: Japanese Patent Application25
Laid-open No. 2020-093770
Summary of Invention
Problem to be solved by the Invention
[0004] However, according to the above-described30
conventional technique, when a device to be diagnosed is a
device manually operated, such as a brake mounted on a
railway vehicle, data detected by a sensor is liable to
3
exhibit individual variations depending on characteristics
of each person that operate the device. Therefore, there
has been a problem in that when diagnosing the soundness of
the device using data including individual variations, the
accuracy of diagnosis decreases.5
[0005] The present disclosure has been made in view of
the above, and an object of the present disclosure is to
obtain a soundness diagnosis apparatus capable of reducing
the decrease in accuracy when diagnosing the soundness of
the device.10
Means to Solve the Problem
[0006] To solve the above problems and achieve an
object, the present disclosure is directed to a soundness
diagnosis apparatus to diagnose soundness of a device. The15
apparatus includes: a data loading unit to acquire
operating data of the device for a diagnosis target period;
a feature amount data generation unit to sample, as sample
data, a target data segment for feature amount data from
the operating data on a basis of a physical characteristic20
of the device, and to generate the feature amount data by
using the sample data; an inference unit to perform
soundness diagnosis on the feature amount data that is
sequentially generated by using a learned model obtained
through model learning of a normal state of the device; and25
a visualization unit to visualize a transition of a
soundness diagnosis result obtained by the inference unit.
Effects of the Invention
[0007] The soundness diagnosis apparatus of the present30
disclosure can achieve an effect that the decrease in
accuracy when diagnosing the soundness of a device can be
reduced.
4
Brief Description of Drawings
[0008] FIG. 1 is a first diagram illustrating a
configuration example of a soundness diagnosis apparatus
according to an embodiment.5
FIG. 2 is a flowchart illustrating operation of the
soundness diagnosis apparatus according to the embodiment.
FIG. 3 is a flowchart illustrating operation of
generating feature amount data by a feature amount data
generation unit of the soundness diagnosis apparatus10
according to the embodiment.
FIG. 4 is a diagram illustrating an example of a
configuration of processing circuitry when the processing
circuitry included in the soundness diagnosis apparatus
according to the embodiment is implemented by a processor15
and a memory.
FIG. 5 is a diagram illustrating an example of a
configuration of processing circuitry when the processing
circuitry included in the soundness diagnosis apparatus
according to the embodiment includes dedicated hardware.20
FIG. 6 illustrates a second diagram of a configuration
example of a soundness diagnosis apparatus according to an
embodiment.
Description of Embodiments25
[0009] Hereinafter, a soundness diagnosis apparatus and
a soundness diagnosis method according to an embodiment of
the present disclosure will be described in detail with
reference to the drawings.
[0010] Embodiment.30
FIG. 1 is a first diagram illustrating a configuration
example of a soundness diagnosis apparatus 20 according to
the present embodiment. The soundness diagnosis apparatus
5
20 is an apparatus that uses operating data 31 relating to
a device mounted on a railway vehicle 10 to diagnose
soundness of the device mounted on the railway vehicle 10.
The operating data 31 is data detected by a sensor (not
illustrated) or the like mounted on the railway vehicle 105
and indicating an operating state of the device. In the
present embodiment, specifically, a case will be described
in which the device mounted on the railway vehicle 10 is a
brake system 11. The brake system 11 is a system that
includes a brake cylinder (not illustrated) and controls a10
braking force by air pressure.
[0011] The configuration and operation of the soundness
diagnosis apparatus 20 will be described. As illustrated
in FIG. 1, the soundness diagnosis apparatus 20 includes a
condition setting unit 21, a data loading unit 22, a15
feature amount data generation unit 23, a learning unit 24,
an inference unit 25, and a visualization unit 26. The
condition setting unit 21 and the visualization unit 26 are
included in an operation user interface (UI) 27. FIG. 2 is
a flowchart illustrating the operation of the soundness20
diagnosis apparatus 20 according to the present embodiment.
[0012] The condition setting unit 21 receives settings
of various conditions for the data loading unit 22 and the
feature amount data generation unit 23 from a user 40 of
the soundness diagnosis apparatus 20, and sets the25
conditions to the data loading unit 22 and the feature
amount data generation unit 23 (step S11). Specifically,
the condition setting unit 21 receives, from the user 40, a
diagnosis target period that is a target period for which
the data loading unit 22 acquires the operating data 31.30
Furthermore, the condition setting unit 21 receives, from
the user 40, settings of various conditions used when the
feature amount data generation unit 23 generates feature
6
amount data 32 on the basis of a physical characteristic of
the brake system 11. The condition setting unit 21 is, for
example, an interface such as a mouse or a keyboard. When
the condition setting unit 21 and the visualization unit 26
are integrated as the operation UI 27, the condition5
setting unit 21 may be a touch panel or the like.
[0013] The data loading unit 22 acquires the operating
data 31 of the brake system 11 for the diagnosis target
period (step S12). In the example of FIG. 1, the data
loading unit 22 acquires the operating data 31 for the10
diagnosis target period from the operating data 31 for the
entire period detected by the railway vehicle 10. However,
the present invention is not limited thereto. The data
loading unit 22 may acquire the operating data 31 for the
entire period output from the railway vehicle 10, and may15
extract the operating data 31 for the diagnosis target
period from the operating data 31 for the entire period.
Note that the operating data 31 for the entire period
detected by the railway vehicle 10 may be stored in a
storage unit, and the data loading unit 22 may read the20
operating data 31 from the storage unit. The data loading
unit 22 outputs the operating data 31 of the brake system
11 for the diagnosis target period to the feature amount
data generation unit 23.
[0014] The feature amount data generation unit 2325
samples a target data segment for the feature amount data
32 from the operating data 31 as sample data on the basis
of the physical characteristic of the device, that is, the
brake system 11, and generates the feature amount data 32
by using the sample data thus sampled (step S13).30
Operation of generating the feature amount data 32 by the
feature amount data generation unit 23 will be described in
detail. FIG. 3 is a flowchart illustrating the operation
7
of generating the feature amount data 32 by the feature
amount data generation unit 23 of the soundness diagnosis
apparatus 20 according to the present embodiment.
[0015] The feature amount data generation unit 23
detects a brake release timing at the start of moving of5
the railway vehicle 10 from data of brake cylinder pressure
included in the operating data 31 on the basis of, for
example, vehicle speed information and brake notch
information of the railway vehicle 10 (step S21).
[0016] In general, in order to perform soundness10
diagnosis on the brake system 11, it is necessary to
generate the feature amount data 32 corresponding to the
physical characteristic of the brake system 11. The
feature amount data 32 needs to constantly exhibit an
identical behavior, and the tendency of the behavior needs15
to change little by little along with a change in
soundness, that is, the deterioration progress. The
operating data 31 of the brake system 11 is constantly
acquired during traveling of the railway vehicle 10 and the
value thereof variously changes due to complicated control20
during traveling. Therefore, it is necessary to sample a
feature amount having a characteristic of the above-
described deterioration progress from the operating data
31. The brake system 11 of the railway vehicle 10 is
mainly used at the time of deceleration, stop, and the25
like. However, the brake system 11 is liable to exhibit,
in its operation, characteristics of drivers of the railway
vehicle 10, that is, individual variations. Furthermore,
even with an identical brake level, the braking force
changes depending on the difference in occupancy of the30
railway vehicle 10. Therefore, in the present embodiment,
focusing on the time point when the railway vehicle 10
starts moving, that is, when the brake is released, data of
8
a portion where the brake cylinder pressure decreases with
the release of the brake is sampled and used as a target
data for the feature amount data 32. The portion where the
brake cylinder pressure decreases with the release of the
brake is a brake cylinder (BC) pressure falling portion5
where the air in the brake cylinder is released. For the
falling of the BC pressure, individual variations are
minimal between drivers, and the falling of the BC pressure
exhibits an identical behavior as the physical
characteristic of the brake system 11, and thus meets the10
requirement of the feature amount data 32 having the
characteristic of the above-described deterioration
progress.
[0017] The feature amount data generation unit 23
samples a prescribed time series range including the brake15
release timing at the start of moving of the railway
vehicle 10, that is, a time point when the BC pressure
falls, from the data of the brake cylinder pressure
included in the operating data 31, as sample data (step
S22). The feature amount data generation unit 23 may20
sample the sample data with the time point when the BC
pressure falls as a start point, or may sample the sample
data with a prescribed time point that is before the time
point when the BC pressure falls as a start point.
[0018] The feature amount data generation unit 2325
performs data cleansing of the sample data sampled (step
S23). The feature amount data generation unit 23 can
eliminate elements that become noise when calculating the
feature amount data 32, by removing irregular sample data
that clearly behaves differently from those indicating30
change due to aging.
[0019] The feature amount data generation unit 23
performs filtering to extract, from the sample data after
9
being subjected to data cleansing, sample data that matches
set conditions (step S24). The feature amount data
generation unit 23 narrows down environmental conditions
during traveling of the railway vehicle 10 to unify the
environmental conditions during traveling of the railway5
vehicle 10, that is, to unify analysis conditions in the
soundness diagnosis. For example, the feature amount data
generation unit 23 performs filtering on the basis of the
position of a brake notch at the time of stop, the
occupancy of the railway vehicle 10, the speed of brake10
release, and the like. Note that it is also expected that
the occupancy exhibits different tendencies depending on
the routes in which the railway vehicle 10 is used.
Therefore, the feature amount data generation unit 23 may
change the condition used when performing filtering15
depending on the route in which the railway vehicle 10 is
used. Furthermore, the feature amount data generation unit
23 may perform filtering for classification into a
plurality of conditions instead of narrowing down to one
condition so as to unify the analysis conditions in the20
soundness diagnosis. For example, when filtering is
performed on the basis of the occupancy of the railway
vehicle 10, the feature amount data generation unit 23 may
perform filtering so as to categorize the occupancy into
groups such as less than 30%, 30% or more and less than25
70%, and 70% or more.
[0020] The feature amount data generation unit 23
generates the feature amount data 32 by processing the
sample data after being subjected to filtering (step S25).
The feature amount data generation unit 23 processes the30
sample data after being subjected to filtering on the basis
of domain knowledge, that is, the physical characteristic
of the brake system 11 that is a device to be diagnosed.
10
For example, the feature amount data generation unit 23 may
obtain a temporary difference by performing first-order
differentiation, may obtain a cumulative sum by performing
first-order integration, or may combine a plurality of
feature amounts. The feature amount data generation unit5
23 outputs the generated feature amount data 32 to the
learning unit 24 and the inference unit 25. Note that the
feature amount data generation unit 23 may store the
generated feature amount data 32 in a storage unit. In
this case, the learning unit 24 and the inference unit 2510
read the feature amount data 32 from the storage unit.
[0021] In this manner, the feature amount data
generation unit 23 samples, on the basis of the vehicle
speed information and the brake notch information of the
railway vehicle 10, a range including a timing at which the15
brake is released at the start of moving of the railway
vehicle 10 as sample data from data of the brake cylinder
pressure included in the operating data 31 . It can also
be said that the feature amount data generation unit 23
samples a range including a timing at which the brake of20
the railway vehicle 10 is released from the operating data
31 as sample data. The feature amount data generation unit
23 performs data cleansing of the sample data thus sampled,
and performs filtering to unify environmental conditions
during traveling of the railway vehicle 10. The feature25
amount data generation unit 23 generates, on the basis of
the physical characteristic of the brake system 11, the
feature amount data 32 by processing the sample data after
being subjected to filtering.
[0022] Returning to the description of the soundness30
diagnosis apparatus 20. The learning unit 24 performs
model learning of a normal state by using the feature
amount data 32 acquired during a normal time among the
11
feature amount data 32 (step S14). The learning unit 24
obtains a learned model 33 as a result of the model
learning. The feature amount data 32 acquired during the
normal time may be based on data obtained during the time
of a test in a factory where the railway vehicle 10 is5
manufactured, or may be based on data obtained during the
initial stage of introduction from when the railway vehicle
10 starts its operation for a prescribed period. The
learning unit 24 performs model learning of the normal
state by using machine learning. For example, an outlier10
detection method can be used as a model of machine
learning. More specifically, the outlier detection method
includes an algorithm such as a one class support vector
machine (OCSVM), but is not limited thereto. A method
other than the outlier detection method, such as deep15
learning, may be used as the model of machine learning.
The learning unit 24 outputs the learned model 33 obtained
as a result of the model learning to the inference unit 25.
Note that the learning unit 24 may store the learned model
33 in a storage unit. In this case, the inference unit 2520
reads the learned model 33 from the storage unit.
[0023] The inference unit 25 performs soundness
diagnosis on the feature amount data 32 that is
sequentially generated by using the learned model 33
obtained through model learning of the normal state of the25
device, that is, by using the learned model 33 obtained as
a result of the model learning by the learning unit 24
(step S15). The inference unit 25 may perform soundness
diagnosis on the feature amount data 32 that is
sequentially generated at a prescribed interval, for30
example, every week, or may perform soundness diagnosis at
a timing specified and desired by the user 40 for
confirmation. In the soundness diagnosis method, for
12
example, the inference unit 25 calculates a degree of
deviation from the normal state, with respect to the
feature amount data 32 that is sequentially generated, as a
score, and normalizes the degree of deviation from the
normal state that has been scored to obtaine a soundness5
diagnosis result 34. When utilizing the OCSVM, the
inference unit 25 calculates a distance from a boundary
plane of the learned model 33 to each feature amount data
32, and normalizes the distance. The inference unit 25
outputs the soundness diagnosis result 34 obtained as a10
result of performing the soundness diagnosis, to the
visualization unit 26. Note that the inference unit 25 may
store the soundness diagnosis result 34 in a storage unit.
In this case, the visualization unit 26 reads the soundness
diagnosis result 34 from the storage unit.15
[0024] The visualization unit 26 visualizes the
transition of the soundness diagnosis result 34 obtained by
the inference unit 25 (step S16). For example, the
visualization unit 26 utilizes scatter diagrams, line
graphs, or the like to visualize tendency of transition of20
the soundness diagnosis result 34 based on multiple feature
amount data 32 by chronologically plotting the soundness
diagnosis result 34 from the past to the present.
[0025] Here, the soundness diagnosis apparatus 20 is
assumed to be used by, as the user 40, an engineer or the25
like of a device manufacturer that manufactures the device
mounted on the railway vehicle 10, such as the brake system
11 illustrated in the example of FIG. 1. The user 40 of
the device manufacturer estimates a future tendency of
soundness from the tendency of transition of the soundness30
diagnosis result 34 from the past to the present, and
recommends maintenance of the brake system 11, replacement
of the device, and the like to the railway company at an
13
optimum timing. The user 40 may determine the timing of
the maintenance of the brake system 11, the timing of the
replacement of the device, and the like by setting a
plurality of thresholds to the soundness diagnosis result
34, for example. At this time, the user 40 makes a5
determination by confirming statistical results of the
soundness diagnosis result 34 from both a microscopic
perspective based on each sample point and a macroscopic
perspective based on the regular soundness diagnosis result
34, such as on a monthly basis. For example, when10
maintenance has been performed by a railway company on the
railway vehicle 10, it is also assumed that the tendency of
the feature amount data 32 obtained by the soundness
diagnosis apparatus 20 may differ before and after the
maintenance. In general, when maintenance has been15
performed by the railway company on the railway vehicle 10,
it is considered that the feature amount data 32 tends to
be improved. When the tendency of the feature amount data
32 has been changed since a certain time point, the user 40
assumes that the maintenance has been performed by the20
railway company, and confirms the soundness diagnosis
result 34. When information on maintenance of the railway
vehicle 10 can be obtained from the railway company, the
user 40 confirms the soundness diagnosis result 34 on the
basis of the information on maintenance.25
[0026] Regarding the installation place of the soundness
diagnosis apparatus 20, the soundness diagnosis apparatus
20 may be installed in the device manufacturer described
above. Alternatively, the operation UI 27 of the
soundness diagnosis apparatus 20 may be installed in the30
device manufacturer and the remaining portion may be
installed in the railway company that operates the railway
vehicle 10 or in the railway vehicle 10. When the portion
14
such as the operation UI 27 is installed in the device
manufacturer and the remaining portion is installed in the
railway company that operates the railway vehicle 10 or in
the railway vehicle 10, the operation UI 27 may be a
terminal device such as a tablet. The soundness diagnosis5
apparatus 20 can remotely diagnose soundness of the brake
system 11 even if the soundness diagnosis apparatus 20 is
installed in a place different from the place where the
brake system 11 to be diagnosed is installed. Since the
operation UI 27 is at hand, regardless of where the10
remaining portion of the soundness diagnosis apparatus 20
other than the operation UI 27 is installed, the user 40 of
the soundness diagnosis apparatus 20 can set conditions to
the data loading unit 22 and the feature amount data
generation unit 23 by using the condition setting unit 21,15
and confirm the soundness diagnosis result 34 visualized by
the visualization unit 26.
[0027] Note that the condition setting unit 21 receives
settings of various conditions for the data loading unit 22
and the feature amount data generation unit 23 from the20
user 40, and sets the conditions to the data loading unit
22 and the feature amount data generation unit 23.
However, the present invention is not limited thereto. The
condition setting unit 21 may receive settings of
conditions regarding the learning method in the learning25
unit 24, the timing of soundness diagnosis in the inference
unit 25, and the like, from the user 40, and may set the
conditions to the learning unit 24 and the inference unit
25.
[0028] Next, a hardware configuration of the soundness30
diagnosis apparatus 20 according to the present embodiment
will be described. In the soundness diagnosis apparatus
20, the condition setting unit 21 is an interface that
15
receives operation from the user 40. In the visualization
unit 26, a portion that displays the soundness diagnosis
result 34 is a display such as a liquid crystal display
(LCD). Among the data loading unit 22, the feature amount
data generation unit 23, the learning unit 24, the5
inference unit 25, and the visualization unit 26, portions
other than the portion that displays the soundness
diagnosis result 34 are implemented by processing
circuitry. The processing circuitry may be a memory that
stores a program and a processor that executes the program10
stored in the memory, or may be dedicated hardware. The
processing circuitry is also referred to as a control
circuit.
[0029] FIG. 4 is a diagram illustrating an example of a
configuration of processing circuitry 90 when the15
processing circuitry included in the soundness diagnosis
apparatus 20 according to the present embodiment is
implemented by a processor 91 and a memory 92. The
processing circuitry 90 illustrated in FIG. 4 is a control
circuit and includes the processor 91 and the memory 92.20
If the processing circuitry 90 includes the processor 91
and the memory 92, each function of the processing
circuitry 90 is implemented by software, firmware, or a
combination of software and firmware. Software or firmware
is described as a program and stored in the memory 92. In25
the processing circuitry 90, the processor 91 reads and
executes the program stored in the memory 92 to implement
each function. That is, the processing circuitry 90
includes the memory 92 for storing the program that causes
the processing of the soundness diagnosis apparatus 20 to30
be resultantly executed. It can also be said that this
program is a program for causing the soundness diagnosis
apparatus 20 to execute each function implemented by the
16
processing circuitry 90. This program may be provided by a
storage medium in which the program is stored, or may be
provided by other means such as a communication medium.
[0030] The program described above can also be said as a
program that causes the soundness diagnosis apparatus 20 to5
execute: an acquisition step of acquiring, by the data
loading unit 22, the operating data 31 of the device for
the diagnosis target period; a generation step of sampling,
by the feature amount data generation unit 23, a target
data segment for the feature amount data 32 from the10
operating data 31 as sample data on the basis of the
physical characteristic of the device, and of generating
the feature amount data 32 by using the sample data; a
learning step of performing, by the learning unit 24, model
learning of the normal state by using the feature amount15
data 32 acquired during a normal time among the feature
amount data 32; an inference step of performing, by the
inference unit 25, soundness diagnosis on the feature
amount data 32 that is sequentially generated by using the
learned model 33 obtained as a result of the model learning20
by the learning unit 24; and a visualizing step of
visualizing, by the visualization unit 26, a transition of
the soundness diagnosis result 34 obtained by the inference
unit 25.
[0031] Here, the processor 91 is, for example, a central25
processing unit (CPU), a processing device, an arithmetic
device, a microprocessor, a microcomputer, a digital signal
processor (DSP), or the like. Furthermore, the memory 92
corresponds to a nonvolatile or volatile semiconductor
memory such as a random access memory (RAM), a read only30
memory (ROM), a flash memory, an erasable programmable ROM
(EPROM), or an electrically EPROM (EEPROM (registered
trademark)), a magnetic disk, a flexible disk, an optical
17
disk, a compact disk, a mini disk, a digital versatile disc
(DVD), a Blu-ray disc, a hard disk drive (HDD), or the
like.
[0032] FIG. 5 is a diagram illustrating an example of a
configuration of processing circuitry 93 when the5
processing circuitry included in the soundness diagnosis
apparatus 20 according to the present embodiment includes
dedicated hardware. For example, the processing circuitry
93 illustrated in FIG. 5 corresponds to a single circuit, a
composite circuit, a programmed processor, a parallel10
programmed processor, an application specific integrated
circuit (ASIC), a field programmable gate array (FPGA), or
a combination thereof. The processing circuitry 93 may be
partially implemented by dedicated hardware, and partially
implemented by software or firmware. In this manner, the15
processing circuitry 93 can implement the above-described
functions by using dedicated hardware, software, firmware,
or a combination thereof.
[0033] As has been described above, according to the
present embodiment, with respect to the brake system 1120
that is a device mounted on the railway vehicle 10, the
soundness diagnosis apparatus 20 samples the data of the BC
pressure falling portion in which the brake cylinder
pressure decreases at the time point when the railway
vehicle 10 starts moving, that is, the time point when the25
brake is released, from the operating data 31. Then, the
soundness diagnosis apparatus 20 performs data cleansing,
filtering, and the like to generate the feature amount data
32. The soundness diagnosis apparatus 20 learns data
acquired during the normal time by machine learning by30
using the feature amount data 32, and diagnoses the
soundness of the brake system 11 on the basis of the degree
of deviation from the data acquired during the normal time.
18
As a result, the soundness diagnosis apparatus 20 is less
affected by individual variations between drivers, even for
the device that is liable to exhibit, in its operation,
individual variations between drivers of the railway
vehicle 10, and can reduce the decrease in accuracy when5
diagnosing the soundness of the device.
[0034] Furthermore, the soundness diagnosis apparatus 20
can obtain the feature amount data 32 on the basis of the
operating data 31 that is already accessible. Therefore,
the soundness diagnosis apparatus 20 can perform soundness10
diagnosis without additionally introducing a special
sensor, a detection device, or the like.
[0035] The user 40 of the soundness diagnosis apparatus
20 can remotely confirm the current soundness of the brake
system 11 without inspecting the actual brake system 11 of15
the railway vehicle 10 on site, and can determine the
timing of the maintenance and the timing of the replacement
of the device.
[0036] Note that, in the present embodiment, a case
where the device to be diagnosed, the soundness of which is20
diagnosed by the soundness diagnosis apparatus 20, is the
brake system 11 has been specifically described. However,
the device to be diagnosed, the soundness of which is
diagnosed by the soundness diagnosis apparatus 20, is not
limited to the brake system 11. In the present embodiment,25
the data of the BC pressure falling portion at the start of
moving of the railway vehicle 10 is sampled as the target
data for the feature amount data 32 of the brake system 11.
However, even for devices operated by individuals, as long
as there is a timing at which data exhibits minimal30
individual variations and similar behavior can be detected
as a physical characteristic, the soundness diagnosis
apparatus 20 can be applied.
19
[0037] Moreover, in the present embodiment, the
configuration in which the soundness diagnosis apparatus 20
includes the learning unit 24 has been specifically
described, but the configuration of the soundness diagnosis
apparatus 20 is not limited thereto. For example, if it is5
not necessary to perform model learning of the normal state
for every device to be diagnosed or every user 40, a
soundness diagnosis apparatus 20A may have a configuration
not including the learning unit 24 as illustrated in FIG.
6. FIG. 6 illlustrates a second diagram of a configuration10
example of the soundness diagnosis apparatus 20A according
to the present embodiment. In this case, model learning of
the normal state is performed in advance by using the
feature amount data 32 acquired during a normal time among
the feature amount data 32 in a learning unit (not15
illustrated) provided outside the soundness diagnosis
apparatus 20A, and the learned model 33 obtained as a
result of the model learning is stored in a storage unit
inside the soundness diagnosis apparatus 20A. The
inference unit 25 can read the learned model 33 obtained20
through model learning of the normal state of the device
from the storage unit, and can perform soundness diagnosis
on the feature amount data 32 that is sequentially
generated by using the read learned model 33.
[0038] The configurations described in the above25
embodiment are just examples and can be combined with other
known techniques. The embodiments can be combined with
each other and the configurations can be partially omitted
or changed without departing from the gist.
30
Reference Signs List
[0039] 10 railway vehicle; 11 brake system; 20, 20A
soundness diagnosis apparatus; 21 condition setting unit;
20
22 data loading unit; 23 feature amount data generation
unit; 24 learning unit; 25 inference unit; 26
visualization unit; 27 operation UI; 31 operating data;
32 feature amount data; 33 learned model; 34 soundness
diagnosis result; 40 user.5
21
We Claim :
[Claim 1] A soundness diagnosis apparatus to diagnose
soundness of a device, the apparatus comprising:
a data loading unit to acquire operating data of the
device for a diagnosis target period;5
a feature amount data generation unit to sample, as
sample data, a target data segment for feature amount data
from the operating data on a basis of a physical
characteristic of the device, and to generate the feature
amount data by using the sample data;10
an inference unit to perform soundness diagnosis on
the feature amount data that is sequentially generated by
using a learned model obtained through model learning of a
normal state of the device; and
a visualization unit to visualize a transition of a15
soundness diagnosis result obtained by the inference unit.
[Claim 2] The soundness diagnosis apparatus according to
claim 1, comprising
a learning unit to perform model learning of the20
normal state by using the feature amount data acquired
during a normal time among the feature amount data to
obtain the learned model as a result of the model learning.
[Claim 3] The soundness diagnosis apparatus according to25
claim 2, wherein
the learning unit performs model learning of the
normal state by using machine learning, and
the inference unit calculates a degree of deviation
from the normal state, with respect to the feature amount30
data that is sequentially generated, as a score, and
normalizes the degree of deviation from the normal state
that is scored as the soundness diagnosis result.
22
[Claim 4] The soundness diagnosis apparatus according to
any one of claims 1 to 3, wherein
the device is a brake system mounted on a railway
vehicle, and
the feature amount data generation unit samples, as5
the sample data, a range including a timing at which a
brake of the railway vehicle is released from the operating
data.
[Claim 5] The soundness diagnosis apparatus according to10
any one of claims 1 to 3, wherein
the device is a brake system mounted on a railway
vehicle, and
the feature amount data generation unit samples, as
the sample data, a range including a timing at which a15
brake is released at a start of moving of the railway
vehicle from data of brake cylinder pressure included in
the operating data on the basis of vehicle speed
information and brake notch information of the railway
vehicle.20
[Claim 6] The soundness diagnosis apparatus according to
any one of claims 1 to 5, wherein
the device is a brake system mounted on a railway
vehicle, and25
the feature amount data generation unit performs data
cleansing of the sample data that is sampled, and performs
filtering to unify an environmental condition during
traveling of the railway vehicle.
30
[Claim 7] The soundness diagnosis apparatus according to
claim 6, wherein
the feature amount data generation unit generates the
23
feature amount data by processing the sample data after
being subjected to filtering on the basis of the physical
characteristic of the device.
[Claim 8] The soundness diagnosis apparatus according to5
any one of claims 1 to 7, wherein
the visualization unit visualizes a tendency of
transition of the soundness diagnosis result by
chronologically plotting the soundness diagnosis result
from past to present.10
[Claim 9] The soundness diagnosis apparatus according to
any one of claims 1 to 8, wherein
the soundness diagnosis apparatus is installed in a
place different from the device, and remotely diagnoses the15
soundness of the device.
[Claim 10] The soundness diagnosis apparatus according
to claim 9, comprising
a condition setting unit to receive, from a user of20
the soundness diagnosis apparatus, the diagnosis target
period that is a target period for which the data loading
unit acquires the operating data, and settings of various
conditions used when the feature amount data generation
unit generates the feature amount data based on the25
physical characteristic of the device, wherein
the user of the soundness diagnosis apparatus sets the
conditions by using the condition setting unit to confirm
the soundness diagnosis result visualized by the
visualization unit.30
[Claim 11] A soundness diagnosis method using a
soundness diagnosis apparatus to diagnose soundness of a
24
device, the method comprising:
an acquisition step of acquiring, by a data loading
unit, operating data of the device for a diagnosis target
period;
a generation step of sampling, by a feature amount5
data generation unit, a target data segment for feature
amount data from the operating data as sample data on a
basis of a physical characteristic of the device, and of
generating the feature amount data by using the sample
data;10
an inference step of performing, by an inference unit,
soundness diagnosis on the feature amount data that is
sequentially generated by using a learned model obtained
through model learning of a normal state of the device; and
a visualizing step of visualizing, by a visualization15
unit, a transition of a soundness diagnosis result obtained
by the inference unit.
[Claim 12] The soundness diagnosis method according to
claim 11, comprising20
a learning step of performing, by a learning unit,
model learning of the normal state by using the feature
amount data acquired during a normal time among the feature
amount data to obtain the learned model as a result of the
model learning.25
[Claim 13] The soundness diagnosis method according to
claim 12, wherein
in the learning step, the learning unit performs model
learning of the normal state by using machine learning, and30
in the inference step, the inference unit calculates a
degree of deviation from the normal state, with respect to
the feature amount data that is sequentially generated, as
25
a score, and normalizes the degree of deviation from the
normal state that is scored as the soundness diagnosis
result.
[Claim 14] The soundness diagnosis method according to5
any one of claims 11 to 13, wherein
the device is a brake system mounted on a railway
vehicle, and
in the generation step, the feature amount data
generation unit samples, as the sample data, a range10
including a timing at which a brake of the railway vehicle
is released from the operating data.
[Claim 15] The soundness diagnosis method according to
any one of claims 11 to 13, wherein15
the device is a brake system mounted on a railway
vehicle, and
in the generation step, the feature amount data
generation unit samples, as the sample data, a range
including a timing at which a brake is released at a start20
of moving of the railway vehicle from data of brake
cylinder pressure included in the operating data on the
basis of vehicle speed information and brake notch
information of the railway vehicle.
25
[Claim 16] The soundness diagnosis method according to
any one of claims 11 to 15, wherein
the device is a brake system mounted on a railway
vehicle, and
in the generation step, the feature amount data30
generation unit performs data cleansing of the sample data
that is sampled, and performs filtering to unify an
environmental condition during traveling of the railway
26
vehicle.
[Claim 17] The soundness diagnosis method according to
claim 16, wherein
in the generation step, the feature amount data5
generation unit generates the feature amount data by
processing the sample data after being subjected to
filtering on the basis of the physical characteristic of
the device.
10
[Claim 18] The soundness diagnosis method according to
any one of claims 11 to 17, wherein
in the visualizing step, the visualization unit
visualizes a tendency of transition of the soundness
diagnosis result by chronologically plotting the soundness15
diagnosis result from past to present.
[Claim 19] The soundness diagnosis method according to
any one of claims 11 to 18, wherein
the soundness diagnosis apparatus is installed in a20
place different from the device, and remotely diagnoses the
soundness of the device.
[Claim 20] The soundness diagnosis method according to
claim 19, comprising25
a condition setting step of receiving, by a condition
setting unit from a user of the soundness diagnosis
apparatus, the diagnosis target period that is a target
period for which the data loading unit acquires the
operating data, and settings of various conditions used30
when the feature amount data generation unit generates the
feature amount data based on the physical characteristic of
the device, wherein
27
the user of the soundness diagnosis apparatus sets the
conditions by using the condition setting unit, and
confirms the soundness diagnosis result visualized by the
visualization unit.