U.S. patent application number 16/792952 was filed with the patent office on 2020-10-22 for abnormality detection device of fuel vapor escape prevention system.
This patent application is currently assigned to TOYOTA JIDOSHA KABUSHIKI KAISHA. The applicant listed for this patent is TOYOTA JIDOSHA KABUSHIKI KAISHA. Invention is credited to Yosuke HASHIMOTO, Akihiro KATAYAMA, Harufumi MUTO, Kazuki TSURUOKA.
Application Number | 20200332747 16/792952 |
Document ID | / |
Family ID | 1000004673346 |
Filed Date | 2020-10-22 |
View All Diagrams
United States Patent
Application |
20200332747 |
Kind Code |
A1 |
MUTO; Harufumi ; et
al. |
October 22, 2020 |
ABNORMALITY DETECTION DEVICE OF FUEL VAPOR ESCAPE PREVENTION
SYSTEM
Abstract
At the time of stopping operation of the vehicle, the pressures
inside of the fuel tank (5) and inside of the canister (6) detected
at every constant time are stored in the storage device. A learned
neural network using the pressures inside the fuel tank (5) and
inside the canister (6) for each fixed time stored in the storage
device and the atmospheric pressure as input parameters of the
neural network and using a case where perforation occurs in the
system causing fuel vapor to leak as a truth label is stored. At
the time of stopping operation of the vehicle, a perforation
abnormality causing fuel vapor to leak is detected from these input
parameters by using the learned neural network.
Inventors: |
MUTO; Harufumi;
(Miyoshi-shi, JP) ; KATAYAMA; Akihiro;
(Toyota-shi, JP) ; HASHIMOTO; Yosuke;
(Nagakute-shi, JP) ; TSURUOKA; Kazuki;
(Toyota-shi, JP) |
|
Applicant: |
Name |
City |
State |
Country |
Type |
TOYOTA JIDOSHA KABUSHIKI KAISHA |
Toyota-shi |
|
JP |
|
|
Assignee: |
TOYOTA JIDOSHA KABUSHIKI
KAISHA
Toyota-shi
JP
|
Family ID: |
1000004673346 |
Appl. No.: |
16/792952 |
Filed: |
February 18, 2020 |
Current U.S.
Class: |
1/1 |
Current CPC
Class: |
F02M 25/0809 20130101;
G05B 13/027 20130101; F02M 25/0836 20130101; F02M 25/0854 20130101;
F02M 25/0872 20130101 |
International
Class: |
F02M 25/08 20060101
F02M025/08; G05B 13/02 20060101 G05B013/02 |
Foreign Application Data
Date |
Code |
Application Number |
Apr 16, 2019 |
JP |
2019-077760 |
Claims
1. An abnormality detection device of a fuel vapor escape
prevention system comprising: a canister formed with a fuel vapor
chamber and atmospheric pressure chamber at the two sides of an
activated carbon layer, the fuel vapor chamber being on the one
hand communicated with an inside space above a fuel level of a fuel
tank and on the other hand communicated through a purge control
valve with an intake passage of an engine, a flow path switching
valve able to selectively connect the atmospheric pressure chamber
to the atmosphere and a suction pump, and a pressure sensor
detecting pressure at an inside of the fuel tank and inside of the
canister, wherein at the time of stopping operation of the vehicle,
processing for detection of an abnormality is performed to generate
a valve closing instruction making the purge control valve close, a
switching instruction switching a switched position of the flow
path switching valve to a switched position at which the
atmospheric pressure chamber is connected to the suction pump, and
a pump operation instruction making the suction pump operate to
make the inside of the fuel tank and inside of the canister a
negative pressure, at the time the processing for detection of an
abnormality is performed, a pressures at the inside of the fuel
tank and inside of the canister detected by the pressure sensor at
every fixed time are stored in a storage device, a learned neural
network learned in weights using the pressures at the inside of the
fuel tank and inside of the canister at every fixed time stored in
the storage device and at least the atmospheric pressure when the
processing for detection of an abnormality is performed as input
parameters of the neural network and using a case where perforation
occurs in the system causing leakage of fuel vapor as a truth label
is stored, and at the time of stopping operation of the vehicle, a
perforation abnormality causing fuel vapor to leak is detected from
said input parameters by using the learned neural network.
2. The abnormality detection device of a fuel vapor escape
prevention system according to claim 1, wherein the processing for
detection of an abnormality includes processing for generating a
valve opening instruction making the purge control valve open after
generating a valve closing instruction of the purge control valve,
a learned neural network learned in weights using the pressures at
the inside of the fuel tank and inside of the canister at every
fixed time stored in the storage device and at least the
atmospheric pressure when the processing for detection of an
abnormality is performed as input parameters of the neural network
and using a case where when perforation occurs in the system
causing leakage of fuel vapor, a case where a valve opening
abnormality occurs in which the purge control valve continues
opened, and a case where a valve closing abnormality occurs in
which the purge control valve continues closed as truth labels,
respectively, is stored, and, at the time of stopping operation of
the vehicle, a perforation abnormality causing fuel vapor to leak,
a valve opening abnormality of the purge control valve, and a valve
closing abnormality of the purge control valve are detected from
the input parameters by using the learned neural network.
3. The abnormality detection device of a fuel vapor escape
prevention system according to claim 1, wherein the input
parameters are comprised of the pressures at the inside of the fuel
tank and inside of the canister at every fixed time stored in the
storage device, the atmospheric pressure when the processing for
detection of an abnormality is performed, and a remaining amount of
a fuel in the fuel tank when the processing for detection of an
abnormality is performed.
4. The abnormality detection device of a fuel vapor escape
prevention system according to claim 1, wherein the input
parameters are comprised of the pressures at the inside of the fuel
tank and inside of the canister at every fixed time stored in the
storage device, the atmospheric pressure when the processing for
detection of an abnormality is performed, a remaining amount of a
fuel in the fuel tank when the processing for detection of an
abnormality is perforated, a temperature of the fuel in the fuel
tank, and a parameter showing a capacity of the suction pump.
5. An abnormality detection device of a fuel vapor escape
prevention system comprising: a canister formed with a fuel vapor
chamber and atmospheric pressure chamber at the two sides of an
activated carbon layer, the fuel vapor chamber being on the one
hand communicated with an inside space above a fuel level of a fuel
tank and on the other hand communicated through a purge control
valve with an intake passage of an engine, a flow path switching
valve able to selectively connect the atmospheric pressure chamber
to the atmosphere and a suction pump, a passage from the flow path
switching valve toward the atmospheric pressure chamber and a
suction passage from the flow path switching valve toward the
suction pump being connected by a reference pressure detection
passage having a restricted opening, and a pressure sensor arranged
in the suction passage from the flow path switching valve toward
the suction pump, at the time of stopping operation of the vehicle,
processing for detection of an abnormality is performed to generate
a valve closing instruction making the purge control valve close, a
pump operation instruction making the suction pump operate to make
an inside of the fuel tank and inside of the canister a negative
pressure while maintaining a switched position of the flow path
switching valve at a switched position where the atmospheric
pressure chamber is connected to the atmosphere when a
predetermined time elapses after stopping operation of the vehicle,
a switching instruction switching the switched position of the flow
path switching valve to a switched position at which the
atmospheric pressure chamber is connected to the suction pump after
generation of the pump operation instruction, and a valve opening
instruction making the purge control valve open after the
generation of the switching instruction, at the time the processing
for detection of an abnormality is performed, a pressures at the
inside of the fuel tank and inside of the canister detected by the
pressure sensor at every fixed time are stored in a storage device,
a learned neural network learned in weights using the pressures at
the inside of the fuel tank and inside of the canister at every
fixed time stored in the storage device and at least the
atmospheric pressure when the processing for detection of an
abnormality is performed as input parameters of the neural network
and using a case where perforation occurs in the system causing
leakage of fuel vapor as a truth label is stored, and, at the time
of stopping operation of the vehicle, a perforation abnormality
causing fuel vapor to leak is detected from said input parameters
by using the learned neural network.
6. The abnormality detection device of a fuel vapor escape
prevention system described in claim 5, wherein a learned neural
network learned in weights using the pressures at the inside of the
fuel tank and inside of the canister at every fixed time stored in
the storage device and at least the atmospheric pressure when the
processing for detection of an abnormality is performed as input
parameters of the neural network and using a case where perforation
occurs in the system causing leakage of fuel vapor, a case where a
valve opening abnormality occurs in which the purge control valve
continues opened, a case where a valve closing abnormality occurs
in which the purge control valve continues closed, a case where an
abnormality occurs in the pressure sensor, a case where a switching
abnormality occurs in which the switched position of the flow path
switching valve is maintained at a switched position connecting the
atmospheric pressure chamber to the atmosphere, a case where a
switching abnormality occurs in which the switched position of the
flow path switching valve is maintained at a switched position
connecting the atmospheric pressure chamber to the suction pump, a
case where an abnormality occurs in which the suction pump
continues operating, and a case where an abnormality occurs in
which the suction pump continues stopped, respectively, as truth
labels is stored, and at the time of stopping operation of the
vehicle, a perforation abnormality causing fuel vapor to leak, the
valve opening abnormality of the purge control valve, the valve
closing abnormality of the purge control valve, an abnormality of
the pressure sensor, the switching abnormality of the flow path
switching valve, and an abnormality of the suction pump are
detected from the input parameters by using the learned neural
network.
7. The abnormality detection device of a fuel vapor escape
prevention system according to claim 5, wherein the input
parameters are comprised of the pressures at the inside of the fuel
tank and inside of the canister at every fixed time stored in the
storage device and the atmospheric pressure when the processing for
detection of an abnormality is performed and a remaining amount of
a fuel in the fuel tank when the processing for detection of an
abnormality is performed.
8. The abnormality detection device of a fuel vapor escape
prevention system according to claim 5, wherein the input
parameters are comprised of the pressures at the inside of the fuel
tank and inside of the canister at every fixed time stored in the
storage device and the atmospheric pressure when the processing for
detection of an abnormality is performed, a remaining amount of a
fuel in the fuel tank when the processing for detection of an
abnormality is performed, a temperature of the fuel in the fuel
tank, and a parameter showing a capacity of the suction pump.
Description
FIELD
[0001] The present invention relates to an abnormality detection
device of a fuel vapor escape prevention system.
BACKGROUND
[0002] In an internal combustion engine, to prevent fuel vapor from
escaping to the outside atmosphere, in the past use has been made
of a fuel vapor escape prevention system comprising a canister
formed with a fuel vapor chamber and atmospheric pressure chamber
at the two sides of an activated carbon layer, making the fuel
vapor chamber on the one hand communicate with an inside space
above a fuel level of a fuel tank and on the other hand connecting
it through a purge control valve to an intake passage of the
engine. In such a fuel vapor escape prevention system, for example,
if the walls of a fuel vapor flow pipe connecting the fuel vapor
chamber and purge control valve of the canister are perforated by a
hole, fuel vapor will end up escaping through the hole to the
outside atmosphere.
[0003] Therefore, known in the art is a diagnosis device designed
to detect a pressure at the inside space above the fuel level at
the inside of the fuel tank (simply referred to as the "pressure
inside the fuel tank") and diagnose if an abnormality has occurred
in the fuel vapor escape prevention system from a change of the
pressure inside the fuel tank, for example, if the walls of the
fuel vapor flow pipe are perforated by a hole (for example, see
Japanese Unexamined Patent Publication No. 2004-44396).
[0004] In this diagnosis device, by making the purge control valve
open in the state where the atmospheric pressure chamber of the
canister is cut off from the atmosphere when the vehicle is being
driven steadily and stably, the pressure inside the fuel tank is
lowered below the atmospheric pressure, then by making the purge
control valve close, the inside of the fuel tank is made to be in a
scaled state. At this time, for example, if the walls of the fuel
vapor flow pipe are perforated by a hole, the pressure inside the
fuel tank will rise a little at a time. Therefore, in this
diagnosis device, when the pressure inside the fuel tank increases
more than a fixed value after the inside of the fuel tank is made
in a sealed state, for example, it is judged that the walls of the
fuel vapor flow pipe are perforated by a hole.
SUMMARY
[0005] In this case, if the diameter of the hole formed in the
walls of the fuel vapor flow pipe is small, the amount of rise of
the pressure inside the fuel tank will become small. On the other
hand, the pressure inside the fuel tank also fluctuates due to
other factors, for example, the temperature of the fuel inside the
fuel tank. Therefore, if it is judged that the walls of the fuel
vapor flow pipe are perforated by a hole when the pressure inside
the fuel tank increases more than a fixed value, there is the risk
of mistaken judgment.
[0006] The present invention provides an abnormality detection
device of a fuel vapor escape prevention system using a neural
network to, for example, detect perforation of the walls of a fuel
vapor flow pipe and able to accurately detect such perforation of
the walls of the fuel vapor flow pipe even if, at this time, the
diameter of the hole is small.
[0007] That is, according to the present invention, there is
provided an abnormality detection device of a fuel vapor escape
prevention system comprising: [0008] a canister formed with a fuel
vapor chamber and atmospheric pressure chamber at the two sides of
an activated carbon layer, the fuel vapor chamber being on the one
hand communicated with an inside space above a fuel level of a fuel
tank and on the other hand communicated through a purge control
valve with an intake passage of an engine, [0009] a flow path
switching valve able to selectively connect the atmospheric
pressure chamber to the atmosphere and a suction pump, and [0010] a
pressure sensor detecting pressure at an inside of the fuel tank
and inside of the canister, wherein [0011] at the time of stopping
operation of the vehicle, processing for detection of an
abnormality is performed to generate a valve closing instruction
making the purge control valve close, a switching instruction
switching a switched position of the flow path switching valve to a
switched position at which the atmospheric pressure chamber is
connected to the suction pump, and a pump operation instruction
making the suction pump operate to make the inside of the fuel tank
and inside of the canister a negative pressure, [0012] at the time
the processing for detection of an abnormality is performed, a
pressures at the inside of the fuel tank and inside of the canister
detected by the pressure sensor at every fixed time are stored in a
storage device, [0013] a learned neural network learned in weights
using the pressures at the inside of the fuel tank and inside of
the canister at every fixed time stored in the storage device and
at least the atmospheric pressure when the processing for detection
of an abnormality is performed as input parameters of the neural
network and using a case where perforation occurs in the system
causing leakage of fuel vapor as a truth label is stored, and
[0014] at the time of stopping operation of the vehicle, a
perforation abnormality causing fuel vapor to leak is detected from
the above mentioned input parameters by using the learned neural
network.
[0015] Furthermore, according to the present invention, there is
provided an abnormality detection device of a fuel vapor escape
prevention system comprising: [0016] a canister formed with a fuel
vapor chamber and atmospheric pressure chamber at the two sides of
an activated carbon layer, the fuel vapor chamber being on the one
hand communicated with an inside space above a fuel level of a fuel
tank and on the other hand communicated through a purge control
valve with an intake passage of an engine, [0017] a flow path
switching valve able to selectively connect the atmospheric
pressure chamber to the atmosphere and a suction pump, a passage
from the flow path switching valve toward the atmospheric pressure
chamber and a suction passage from the flow path switching valve
toward the suction pump being connected by a reference pressure
detection passage having a restricted opening, and [0018] a
pressure sensor arranged in the suction passage from the flow path
switching valve toward the suction pump, [0019] at the time of
stopping operation of the vehicle, processing for detection of an
abnormality is perfainted to generate a valve closing instruction
making the purge control valve close, a pump operation instruction
making the suction pump operate to make an inside of the fuel tank
and inside of the canister a negative pressure while maintaining a
switched position of the flow path switching valve at a switched
position where the atmospheric pressure chamber is connected to the
atmosphere when a predetermined time elapses after stopping
operation of the vehicle, a switching instruction switching the
switched position of the flow path switching valve to a switched
position at which the atmospheric pressure chamber is connected to
the suction pump after generation of the pump operation
instruction, and a valve opening instruction making the purge
control valve open after the generation of the switching
instruction, [0020] at the time the processing for detection of an
abnormality is performed, a pressures at the inside of the fuel
tank and inside of the canister detected by the pressure sensor at
every fixed time are stored in a storage device, [0021] a learned
neural network learned in weights using the pressures at the inside
of the fuel tank and inside of the canister at every fixed time
stored in the storage device and at least the atmospheric pressure
when the processing for detection of an abnormality is performed as
input parameters of the neural network and using a case where
perforation occurs in the system causing leakage of fuel vapor as a
truth label is stored, and, [0022] at the time of stopping
operation of the vehicle, a perforation abnormality causing fuel
vapor to leak is detected from the above mentioned input parameters
by using the learned neural network.
ADVANTAGEOUS EFFECTS OF INVENTION
[0023] By using the pressures at the inside of the fuel tank and
inside of the canister detected by the pressure sensor at every
fixed time and at least the atmospheric pressure as input
parameters of the neural network for learning the weights of the
neural network, for example, it is possible to accurately detect
perforation of the walls of the fuel vapor flow pipe even if a
small diameter hole is formed in the walls of the fuel vapor flow
pipe.
BRIEF DESCRIPTION OF DRAWINGS
[0024] FIG. 1 is an overall view of a fuel vapor escape prevention
system.
[0025] FIG. 2A and FIG. 2B are enlarged views schematically showing
the pump module shown in FIG. 1.
[0026] FIG. 3 is a perspective view of the gravity sensor.
[0027] FIG. 4 is a view showing a remaining amount of fuel in the
fuel tank.
[0028] FIG. 5 is a view showing a change in the system internal
pressure.
[0029] FIG. 6 is a view showing one example of a neural
network.
[0030] FIG. 7 is a view showing processing for detection of an
abnormality and a change in the system internal pressure.
[0031] FIG. 8 is a view showing processing for detection of an
abnormality and a change in the system internal pressure.
[0032] FIG. 9 is a view showing processing for detection of an
abnormality and a change in the system internal pressure.
[0033] FIG. 10 is a view showing processing for detection of an
abnormality and a change in the system internal pressure.
[0034] FIG. 11 is a view showing processing for detection of an
abnormality and a change in the system internal pressure.
[0035] FIG. 12 is a view showing a neural network used in an
example of the present invention.
[0036] FIG. 13 is a view showing a list of input parameters.
[0037] FIG. 14 is a view showing indicators of performance of the
suction pump.
[0038] FIG. 15 is a view showing a list of output values.
[0039] FIG. 16 is a view showing a training data set.
[0040] FIG. 17 is a view for explaining a learning method.
[0041] FIG. 18 is a flow chart for performing processing for
detection of an abnormality.
[0042] FIG. 19 is a flow chart for performing processing for
learning.
[0043] FIG. 20 is a flow chart for reading data into an electronic
control unit.
[0044] FIG. 21 is a view showing an example of a neural
network.
[0045] FIG. 22 is a flow chart for the detection of an
abnormality.
[0046] FIG. 23A and FIG. 23B are enlarged views schematically
showing modifications of the pump module shown in FIG. 1.
[0047] FIG. 24 is a view showing processing for detection of an
abnormality and a change in the system internal pressure.
[0048] FIG. 25 is a view showing processing for detection of an
abnormality and a change in the system internal pressure.
[0049] FIG. 26 is a view showing processing for detection of an
abnormality and a change in the system internal pressure.
[0050] FIG. 27 is a view showing processing for detection of an
abnormality and a change in the system internal pressure.
[0051] FIG. 28 is a view showing processing for detection of an
abnormality and a change in the system internal pressure.
[0052] FIG. 29 is a view showing processing for detection of an
abnormality and a change in the system internal pressure.
[0053] FIG. 30 is a view showing processing for detection of an
abnormality and a change in the system internal pressure.
[0054] FIG. 31 is a view showing processing for detection of an
abnormality and a change in the system internal pressure.
[0055] FIG. 32 is a view showing processing for detection of an
abnormality and a change in the system internal pressure.
[0056] FIG. 33 is a view showing a neural network used in another
example according to the present invention.
[0057] FIG. 34 is a view showing a list of output values.
[0058] FIG. 35 is a flow chart for performing processing for
detection of an abnormality.
[0059] FIG. 36 is a view showing a neural network used in another
example according to the present invention.
DESCRIPTION OF EMBODIMENTS
Overall Configuration of Internal Combustion Engine
[0060] FIG. 1 shows an overall view of a fuel vapor escape
prevention system. Referring to FIG. 1, 1 indicates an engine body,
2 a surge tank, 3 an intake duct, 4 a throttle valve, 5 a fuel
tank, and 6 a canister. At the fuel tank 5, a fuel level gauge 7
for detecting a height of a fuel level in the fuel tank 5 and a
temperature sensor 8 for detecting a temperature of the fuel in the
fuel tank are attached. On the other hand, the canister 6 has an
activated carbon layer 9 and a fuel vapor chamber 10 and
atmospheric pressure chamber 11 respectively arranged at the two
sides of the activated carbon layer 9. Note that, in FIG. 1, the
fuel vapor chamber 10 is formed from a first fuel vapor chamber 10a
and a second fuel vapor chamber 10b, but this fuel vapor chamber 10
may also be configured from a single common fuel vapor chamber.
[0061] As shown in FIG. 1, the first fuel vapor chamber 10a is
communicated through a fuel vapor flow pipe 12 to an inside space
above a fuel level of the fuel tank 5, while the second fuel vapor
chamber 10b is connected through a fuel vapor flow pipe 13 with the
inside of the surge tank 2, that is, the inside of the intake
passage. Inside this fuel vapor flow pipe 13, a purge control valve
14 is arranged. On the other hand, the atmospheric pressure chamber
11 is communicated through a small volume activated carbon layer 15
and suction pump module 16 to an atmosphere communication pipe
17.
[0062] In FIG. 1, 20 shows an electronic control unit for
controlling engine operation and the fuel vapor escape prevention
system. As shown in FIG. 1, the electronic control unit 20 is
comprised of a digital computer provided with a storage device 22,
that is, a memory 22, a CPU (microprocessor) 23, input port 24, and
output port 25, which are connected with each other by a
bidirectional bus 21. At the input port 24, an output signal of the
fuel level gauge 7, an output signal of the temperature sensor 8,
an output signal of an atmospheric pressure sensor 30 for detecting
the atmospheric pressure, and an output signal of a driving
start/stop switch 31 are input through respectively corresponding
AD converters 26.
[0063] In a hybrid engine provided with an electric motor as a
drive source, if the driving start/stop switch 31 is set to ON,
operation of the vehicle by the engine or electric motor is
started, while if the driving start/stop switch 31 is set to OFF,
operation of the vehicle by the engine or electric motor is
stopped. On the other hand, in an engine not provided with an
electric motor as a drive source, if the driving start/stop switch
31 is set to ON, the engine is started and operation of the vehicle
is started, while if the driving start/stop switch 31 is set to
OFF, the engine is stopped and operation of the vehicle is
stopped.
[0064] Further, as shown in FIG. 1, at an accelerator pedal 32, a
load sensor 33 generating an output voltage proportional to an
amount of depression of the accelerator pedal 32 is connected. The
output voltage of the load sensor 33 is input through a
corresponding AD converter 26 to the input port 24. Furthermore, at
the input port 24, a crank angle sensor 34 generating an output
pulse every time a crankshaft rotates by for example 30.degree. is
connected. In the CPU 33, the engine speed is calculated based on
the output signal of the crank angle sensor 34. On the other hand,
the output port 25 is connected through corresponding drive
circuits 27 to the purge control valve 14, the suction pump module
16, and an actuator for the throttle valve 4.
[0065] FIG. 2A and FIG. 2B are enlarged views schematically showing
the suction pump module 16 shown in FIG. 1. Referring to FIG. 2A
and FIG. 2B, the suction pump module 16 is provided with a suction
pump 40 and a flow path switching valve 42 driven by an actuator
41. Furthermore, the suction pump module 16 has an atmospheric
pressure chamber connection path 43 connected through the small
volume activated carbon layer 15 to the atmospheric pressure
chamber 11, an atmosphere communication pipe connection path 44
connected through an atmosphere communication pipe 17 to the
atmosphere, an atmosphere communication path 45 extending from the
atmosphere communication pipe connection path 44 to the flow path
switching valve 42, and a suction passage 46 extending from the
suction pump 40 toward the flow path switching valve 42. A pressure
sensor 47 is arranged in the suction passage 46.
[0066] Inside the flow path switching valve 42, a first passage 48
able to connect the atmospheric pressure chamber connection path 43
and the atmosphere communication path 45 as shown in FIG. 2A and a
second passage 49 able to connect the atmospheric pressure chamber
connection path 43 and the suction passage 46 as shown in FIG. 2B
are formed. The flow path switching valve 42 is normally, as shown
in FIG. 2A, held at a position where the atmospheric pressure
chamber connection path 43 is connected through the first passage
48 to the atmosphere communication path 45. At this time, the
atmospheric pressure chamber 11 is communicated to the atmosphere
through the small volume activated carbon layer 15, atmospheric
pressure chamber connection path 43, first passage 48, atmosphere
communication path 45, atmosphere communication pipe connection
path 44, and atmosphere communication pipe 17. As shown in FIG. 2A,
the switched position of the flow path switching valve 42 at which
the atmospheric pressure chamber 11 is communicated with the
atmosphere is referred to as the "normal position".
[0067] On the other hand, when detecting an abnormality of the fuel
vapor escape prevention system, the flow path switching valve 42 is
switched to a position at which the atmospheric pressure chamber
connection path 43 is connected to the suction passage 46 through
the second passage 49 as shown in FIG. 2B. The position where the
atmospheric pressure chamber 11 is connected to the suction passage
46 as shown in FIG. 2B is referred to as the "test position". If at
this time operating the suction pump 40, the air inside the
atmospheric pressure chamber 11 is sucked in through the small
volume activated carbon layer 15, atmospheric pressure chamber
connection path 43, second passage 49, and suction passage 46. If
at this time the purge control valve 14 is closed, the pressure at
the inside of the fuel tank 5, the inside of the canister 6, the
inside of the fuel vapor flow pipe 12, and the inside of the fuel
vapor flow pipe 13 between the canister 6 and the purge control
valve 14 falls. Note that, below, the pressure at the inside of the
fuel tank 5, the inside of the canister 6, the inside of the fuel
vapor flow pipe 12, and the inside of the fuel vapor flow pipe 13
between the canister 6 and the purge control valve 14 will be
simply referred to as the "pressure of the inside of the fuel tank
5 and the inside of the canister 6". Further, in this embodiment
according to the present invention, the pressure detected by the
pressure sensor 47 will be referred to as the "pressure inside the
fuel vapor escape prevention system", that is, the system internal
pressure. Therefore, when the pressure of the inside of the fuel
tank 5 and the inside of the canister 6 is detected by the pressure
sensor 47, the pressure of the inside of the fuel tank 5 and the
inside of the canister 6 becomes the system internal pressure.
[0068] Next, referring to FIG. 3 and FIG. 4, the method of
accurately finding the remaining amount of fuel in the fuel tank 5
will be simply explained using the case of using a gravity sensor
as an example. As shown in FIG. 3, the gravity sensor 60 is
comprised of a square-shaped frame 61, a cross-shaped sheet 63
having four sheet-shaped arms 62 supported by the frame 61, and a
weight 64 attached to a center part of the cross-shaped sheet 63.
At the arms 62 of the cross-shaped sheet 63, strain gauges are
respectively attached. This gravity sensor 60 is attached to the
vehicle so that the frame 61 is positioned inside a horizontal
plane when the vehicle is stopped on a horizontal surface.
[0069] Gravity acts on the weight 64 in the vertical direction, so
force downward in the vertical direction acts at the center of the
cross-shaped sheet 63. At this time, if the frame 61 is positioned
in a horizontal plane, the amounts of strain of the arms 62 are the
same. Therefore, when the vehicle is stopped on a horizontal
surface, the amounts of strain of the arms 62 become the same. As
opposed to this, if the vehicle is stopped inclined with respect to
the horizontal plane, the amounts of strain of the arms 62 become
different values. The direction of inclination and amount of
inclination of the vehicle with respect to the horizontal plane can
be learned from the differences in amounts of strain of the arms
62. On the other hand, the height of the fuel level inside of the
fuel tank 5 when the vehicle is positioned on a flat surface can be
learned from the detected value of the fuel level gauge 7.
[0070] Therefore, no matter what shape the fuel tank 5, the
remaining amount of fuel in the fuel tank 5 can be learned from the
detected value of the fuel level gauge 7 and the direction of
inclination and amount of inclination of the vehicle with respect
to the horizontal plane detected by the gravity sensor 60.
Therefore, in this example, as shown in FIG. 4, the relationship
among the amount of inclination F in the longitudinal direction of
the vehicle detected by the gravity sensor 60, the amount of
inclination S in the lateral direction of the vehicle, the detected
value H of the fuel level gauge 7, and the remaining amount M of
the fuel at the inside of the fuel tank 5 is found in advance by
experiments and the remaining amount M of fuel at the inside of the
fuel tank 5 is found based on the relationship shown in FIG. 4.
[0071] Next, the basic thinking of the present invention will be
explained. As will be understood from FIG. 1 and FIG. 2B, if
closing the purge control valve 14, switching the flow path
switching valve 42 to the test position, and in that state
operating the suction pump 40, the pressure in the fuel tank 5, in
the canister 6, in the fuel vapor flow pipe 12, and in the fuel
vapor flow pipe 13 between the canister 6 and the purge control
valve 14, that is, the system internal pressure, will fall. The
change in the system internal pressure at this time is shown in
FIG. 5. Note that, in FIG. 5, the point A shows when the suction
action by the suction pump 40 is started while the point B shows
when there is no longer a pressure drop.
[0072] Now then, in FIG. 5, a solid line shows the case where the
fuel vapor escape prevention system is normal, while the broken
line, for example, shows the case where the walls of the fuel vapor
flow pipe 12 are perforated by a hole or where the walls of the
fuel vapor pipe 13 between the canister 6 and the purge control
valve 14 are perforated by a hole. If the walls of the fuel vapor
flow pipe 12 or 13 are perforated by a hole, outside air flows in
from the hole, so the pressure at the point B, as shown by the
broken line, becomes higher than that at the normal time shown by
the solid line. Therefore, it becomes possible to judge if the
walls of the fuel vapor flow pipe 12 or 13 are perforated by a hole
from the magnitude of the pressure at the point B.
[0073] On the other hand, a dash dot line of FIG. 5 shows the case
where the walls of the fuel vapor flow pipe 12 or 13 are not
perforated by a hole, but for some reason of another, the amount of
evaporation of fuel per unit time in the fuel tank 5 increases, for
example, when the temperature of the fuel in the fuel tank 5 is
higher than the case shown by the solid line. In this case as well,
in the same way as when the walls of the fuel vapor flow pipe 12 or
13 are perforated by a hole shown by the broken line, the pressure
at the point B becomes higher than that at the normal time shown by
the solid line. Therefore, if judging that the walls of the fuel
vapor flow 12 or 13 are perforated by a hole just because the
pressure at the point B becomes higher than the normal time shown
by the solid line, a mistaken judgment would be caused.
[0074] However, if the walls of the fuel vapor circulation pipe 12
or 13 are perforated by a hole, that is, in the case shown by the
broken line, compared with the cases shown by the solid line and
the dash and dot line, the degree of drop of pressure will become
smaller in the middle of the drop in pressure, and the overall
shape of the curve of the drop differs between the case shown by
the broken line and the cases shown by the solid line and the dash
and dot line. Therefore, if finding the overall shape of the curve
of the pressure drop, it becomes possible to accurately judge if
the walls of the fuel vapor flow pipe 12 or 13 are perforated by a
hole from the differences in the overall shape of the curve of the
pressure drop. Therefore, in the present invention, the system
internal pressure is detected at every fixed time, and a neural
network is used to judge if an abnormality occurs in the fuel vapor
escape prevention system based on the system internal pressure
detected at every fixed time.
[0075] Summary of Neural Network
[0076] As explained above, in the embodiment according to the
present invention, a neural network is used to judge if an
abnormality occurs in the fuel vapor escape prevention system.
Therefore, first, a neural network will be briefly explained. FIG.
6 shows a simple neural network. The circle marks in FIG. 6 show
artificial neurons. In the neural network, these artificial neurons
are usually called "nodes" or "units" (in the present application,
they are called "nodes"). In FIG. 12, L=1 shows an input layer, L=2
and L=3 show hidden layers, and L=4 shows an output layer. Further,
in FIG. 6, x.sub.1 and x.sub.2 show output values from nodes of the
input layer (L=1), y.sub.1 and y.sub.2 show output values from the
nodes of the output layer (L=4), z.sup.(2).sub.1, z.sup.(2).sub.2,
and z.sup.(2).sub.3 show output values from the nodes of one hidden
layer (L=2), and z.sup.(3).sub.1, z.sup.(3).sub.2, and
z.sup.(3).sub.3 show output values from the nodes of another hidden
layer (L=3). Note that, the numbers of hidden layers may be made
one or any other numbers, while the number of nodes of the input
layer and the numbers of nodes of the hidden layers may also be
made any numbers. Further, the number of nodes of the output layer
may be made a single node, but may also be made a plurality of
nodes.
[0077] At the nodes of the input layer, the inputs are output as
they are. On the other hand, the output values x.sub.1 and x.sub.2
of the nodes of the input layer are input at the nodes of the
hidden layer (L=2), while the respectively corresponding weights
"w" and biases "b" are used to calculate sum input values "u" at
the nodes of the hidden layer (L=2). For example, a sum input value
u.sub.k calculated at a node shown by z.sup.(2).sub.k (k=1, 2, 3)
of the hidden layer (L=2) in FIG. 6 becomes as shown in the
following equation:
[ Equation 1 ] U k = m = 1 n ( x m w km ) + b k ##EQU00001##
[0078] Next, this sum input value u.sub.k is converted by an
activation function "f" and is output from a node shown by
z.sup.(2).sub.k of the hidden layer (L=2) as an output value
z.sup.(2).sub.k (=f(u.sub.k)). On the other hand, the nodes of the
hidden layer (L=3) receive as input the output values
z.sup.(2).sub.1, z.sup.2).sub.2, and z.sup.(2).sub.3 of the nodes
of the hidden layer (L=2). At the nodes of the hidden layer (L=3),
the respectively corresponding weights "w" and biases "b" are used
to calculate the sum input values "u" (.SIGMA.zw+b). The sum input
values "u" are similarly converted by an activation function and
output from the nodes of the hidden layer (L=3) as the output
values z.sup.(3).sub.1, z.sup.(3).sub.2, and z.sup.(3).sub.3. As
this activation function, for example, a Sigmoid function .sigma.
is used.
[0079] On the other hand, at the nodes of the output layer (L=4),
the output values z.sup.(3).sub.1, z.sup.(3).sub.2, and
z.sup.(3).sub.3 of the nodes of the hidden layer (L=3) are input.
At the nodes of the output layer, the respectively corresponding
weights "w" and biases "b" are used to calculate the sum input
values "u" (.SIGMA.zw+b) or just the respectively corresponding
weights "w" are used to calculate the sum input values "u"
(.SIGMA.zw). For example, at the regression problem, at the nodes
of the output layer, an identity function is used, therefore, from
the nodes of the output layer, the sum input values "u" calculated
at the nodes of the output layer are output as they are as the
output values "y".
[0080] Learning in Neural Network
[0081] Now then, if designating the teacher data showing the truth
values of the output values "y" of the neural network, that is, the
truth data, as y.sub.t, the weights "w" and biases "b" in the
neural network are learned using the error back propagation
algorithm so that the difference between the output values "y" and
the teacher data, that is, the truth data y.sub.t, becomes smaller.
This error backpropagation algorithm is known. Therefore, the error
backpropagation algorithm will be explained simply below in its
outlines. Note that, a bias "b" is one kind of weight "w", so
below, a bias "b" will be also be included in what is referred to
as a weight "w". Now then, in the neural network such as shown in
FIG. 6, if the weights at the input values u.sup.(L) to the nodes
of the layers of L=2, L=3, or L=4 are expressed by w.sup.(L), the
differential due to the weights w.sup.(L) of the error function E,
that is, the slope .differential.E/.differential.w.sup.(L), can be
rewritten as shown in the following equation:
[Equation 2]
[0082]
.differential.E/.differential.w.sup.(L)=(.differential.E/.differen-
tial.u.sup.(L))(.differential.u.sup.(L)/.differential.w.sup.(L))
(1)
where, z.sup.(L-1).differential.w.sup.(L)=.differential.u.sup.(L),
so if (.differential.E/.differential.u.sup.(L))=.delta..sup.(L),
the above equation (1) can be shown by the following equation:
[Equation 3]
[0083]
.differential.E/.differential.w.sup.(L)=.delta..sup.(L)z.sup.(L-1)
(2)
[0084] where, if u.sup.(L) fluctuates, fluctuation of the error
function E is caused through the change in the sum input value
u.sup.(L+1) of the following layer, so .delta..sup.(L) can be
expressed by the following equation:
[ Equation 4 ] .delta. ( L ) = ( .differential. E / .differential.
u ( L ) ) = k = 1 k ( .differential. E / .differential. u k ( L + 1
) ) ( .differential. k ( L + 1 ) / .differential. u ( L ) ) ( k = 1
, 2 ) ( 3 ) ##EQU00002##
where, if expressing z.sup.(L)=f(u.sup.(L)), the input value
u.sub.k.sup.(L+1) appearing at the right side of the above equation
(3) can be expressed by the following formula:
[ Equation 5 ] Input values u k ( L + 1 ) = k = 1 k w k ( L + 1 ) z
( L ) = k = 1 k w k ( L + 1 ) f ( u ( L ) ) ( 4 ) ##EQU00003##
where, the first term (.differential.E/.differential.u.sup.(L+1))
at the right side of the above equation (3) is .delta..sup.(L+1),
and the second term
(.differential.u.sub.k.sup.(L+1)/.differential.u.sup.(L)) at the
right side of the above equation (3) can be expressed by the
following equation:
[Equation 6]
[0085]
.differential.(w.sub.k.sup.(L+1)z.sup.(L))/.differential.u.sup.(L)-
=w.sub.k.sup.(L+1).differential.f(u.sup.(L))/.differential.u.sup.(L)=w.sub-
.k.sup.(L+1)f(u.sup.(L)) (5)
[0086] Therefore, .delta..sup.(L) is shown by the following
formula.
[ Equation 7 ] .delta. ( L ) = k = 1 k w k ( L + 1 ) .delta. ( L +
1 ) f ' ( u L - 1 ) That is .delta. ( L - 1 ) = k = 1 k w k ( L )
.delta. ( L ) f ' ( u ( L - 1 ) ) ( 6 ) ##EQU00004##
[0087] That is, if .delta..sup.(L+1) is found, .delta..sup.(L) can
be found.
[0088] Now then, if there is a single node of the output layer
(L=4), teacher data, that is, truth data y.sub.t, is found for a
certain input value, and the output values from the output layer
corresponding to this input value are "y", if the square error is
used as the error function, the square error E is found by
E=1/2(y-y.sub.t).sup.2. In this case, at the node of the output
layer (L=4), the output values "y" become f(u.sup.(L)), therefore,
in this case, the value of .delta..sup.(L) at the node of the
output layer (L=4) becomes like in the following equation:
[Equation 8]
[0089]
.delta..sup.(L)=.differential.E/.differential.u.sup.(L)=(.differen-
tial.E/.differential.y)
(.differential.y/.differential.u.sup.(L))=(y-y.sub.t)f(u.sup.(L))
(7)
[0090] In this case, in the regression problem, as explained above,
f(u.sup.(L)) is an identity function and f(u.sup.(L1)=1. Therefore,
this leads to .delta..sup.(L)=y-y.sub.t and .delta..sup.(L) is
found.
[0091] If .delta..sup.(L) is found, the above equation (6) is used
to find the .delta..sup.(L-1) of the previous layer. The .delta.'s
of the previous layer are successively found in this way. Using
these values of .delta.'s, from the above equation (2), the
differential of the error function E, that is, the slope
.differential.E/.differential.w.sup.(L), is found for the weights
"w". If the slope .differential.E/.differential.w.sup.(L) is found,
this slope .differential.E/.differential.w.sup.(L) is used to
update the weights "w" so that the value of the error function E
decreases. That is, the weights "w" are learned.
[0092] On the other hand, in the classification problem, at the
time of learning, the output values y.sub.1, y.sub.2 . . . from the
output layer (L=4) are input to a softmax layer. If defining the
output values from the softmax layer as y.sub.1', y.sub.2' . . .
and the corresponding truth labels as y.sub.t1, y.sub.t2 . . . as
the error function E, the following cross entropy error E is
used.
[ Equation 9 ] Cross entropy error E = - k = 1 n y tk log y k ' ( "
n " is number of nodes of output layer ) ( 8 ) ##EQU00005##
[0093] In this case as well, the values of .delta..sup.(L) at the
nodes of the output layer (L=4) become
.delta..sup.(L)=y.sub.k-y.sub.tk (k=1, 2 . . . n). From the values
of these .delta..sup.(L), the .delta..sup.(L-1) of the previous
layers are found using the above equation (6).
Embodiments of Present Invention
[0094] First, referring to FIG. 7, the processing for detection of
an abnormality performed at the time of stopping operation of the
vehicle for judging if an abnormality occurs in the fuel vapor
escape prevention system will be explained. FIG. 7 shows a valve
closing instruction for making the purge control valve 14 close, a
valve opening instruction for making the purge control valve 14
open, a switching instruction switching the flow path switching
valve 42 to the normal position, a switching instruction for
switching the flow path switching valve 42 to the test position, an
instruction for operating the suction pump 40 and an instruction
for stopping the suction pump 40, and a change in the system
internal pressure detected by the pressure sensor 47. Note that, in
this example, the system internal pressure shows the pressure at
the inside of the fuel tank 5 and the inside of the canister 6.
[0095] In FIG. 7, to shows the time of stopping operation of the
vehicle. At this time, the valve closing instruction making the
purge control valve 14 close, the switching instruction switching
the flow path switching valve 42 to the normal position, and the
instruction for stopping the suction pump 40 are issued. FIG. 7
shows the times when the purge control valve 14, flow path
switching valve 42, and suction pump 40 are operating normally
based on these instructions. Therefore, at the time of stopping
operation of the vehicle, the purge control valve 14 is made to
close, the flow path switching valve 42 is switched to the normal
position, and the suction pump 40 is made to stop. This state is
continued for a fixed time period from the time of stopping
operation of the vehicle to the time t.sub.1. During this fixed
time period, the suction pump 40 continues stopped, so the suction
action by the suction pump 40 is not performed. Therefore, the
system internal pressure detected by the pressure sensor 47 becomes
atmospheric pressure.
[0096] Next, if reaching the time t.sub.1, the switching
instruction switching the flow path switching valve 42 to the test
position and the instruction for operating the suction pump 40 are
issued. On the other hand, the valve closing instruction continues
to be issued to the purge control valve 14. At this time, the
inside of the fuel tank 5, the inside of the canister 6, the inside
of the fuel vapor flow pipe 12, and the inside of the fuel vapor
flow pipe 13 between the canister 6 and the purge control valve 14
form a sealed space isolated from the atmosphere. If the suction
pump 40 is made to operate in such a state, the air inside this
sealed space is gradually sucked in by the suction pump 40. As a
result, the pressure at the inside of the fuel tank 5 and the
inside of the canister 6, that is, the system internal pressure
detected by the pressure sensor 47, gradually falls. Next, around
when reaching the time t.sub.2, no pressure drop occurs any longer
and the system internal pressure hovers at the dropped state.
[0097] If reaching the time t.sub.2, the valve opening instruction
making the purge control valve 14 open is issued. On the other
hand, at this time, the flow path switching valve 42 is maintained
at the test position and the suction pump 40 continues to be
operated. Therefore, at this time, the suction action by the
suction pump 40 is continued, but the purge control valve 14 is
made to open, so the system internal pressure rapidly rises and
becomes atmospheric pressure. Next, if reaching the time t.sub.3,
the purge control valve 14, flow path switching valve 42, and
suction pump 40 are returned to the states at the time of stopping
operation of the vehicle. That is, if reaching the time t.sub.3,
the valve closing instruction making the purge control valve 14
close, the switching instruction switching the flow path switching
valve 42 to the normal position, and the instruction for stopping
the suction pump 40 are issued.
[0098] On the other hand, if an abnormality occurs in the fuel
vapor escape prevention system, the pattern of change in the system
internal pressure detected by the pressure sensor 47 becomes a
pattern of change different from the pattern of change in the
system internal pressure at the normal time shown in FIG. 7. Next,
this will be explained while referring to FIG. 8 to FIG. 10. Note
that, FIG. 8 to FIG. 10, in the same way as FIG. 7, show a valve
closing instruction making the purge control valve 14 close and a
valve opening instruction making the purge control valve 14 open, a
switching instruction switching the flow path switching valve 42 to
the normal position and a switching instruction switching the flow
path switching valve 42 to the test position, an instruction for
operating the suction pump 40 and an instruction for stopping the
suction pump 40, and a change in the system internal pressure
detected by the pressure sensor 47. Further, in FIG. 8 to FIG. 10,
the broken lines show the patterns of change in the system internal
pressure at the normal time shown in FIG. 7.
[0099] The solid line in FIG. 8 show the pattern of change in the
system internal pressure in the case, for example, where the walls
of the fuel vapor flow pipe 12 or 13 are perforated by a small
hole. In this case, outside air continues to flow into the system
through the small hole, so the system internal pressure does not
fall to the system internal pressure at the normal time. Therefore,
the pattern of change in the system internal pressure in this case
becomes a pattern of change different from the pattern of change in
the system internal pressure at the normal time.
[0100] On the other hand, the solid line of FIG. 9 shows when a
valve opening abnormality occurs in which the purge control valve
14 continues opened even if the valve closing instruction making
the purge control valve 14 close is issued. At this time, the
inside the system continues to be communicated with the outside air
through the purge control valve 14, so as shown by the solid line
of FIG. 9, the system internal pressure is maintained at the
atmospheric pressure. Further, the solid line of FIG. 10 shows when
a valve closing abnormality occurs in which the purge control valve
14 continues closed even if the valve opening instruction making
the purge control valve 14 open is issued. At this time, as shown
by the solid line of FIG. 9, even if the time t.sub.2 is passed,
the system internal pressure is maintained as is at negative
pressure.
[0101] In this way, if an abnormality occurs, the pattern of change
in the system internal pressure becomes a pattern of change
different from the pattern of change in the system internal
pressure at the normal time. Therefore, in the embodiment according
to the present invention, to learn the differences in the pattern
of change in the system internal pressure, the system internal
pressure is detected at every fixed time. Next, this will be
explained while referring to FIG. 11. Note that, FIG. 11, like FIG.
7, shows a valve closing instruction making the purge control valve
14 close and a valve opening instruction making the purge control
valve 14 open, a switching instruction switching the flow path
switching valve 42 to the normal position and a switching
instruction switching the flow path switching valve 42 to the test
position, an instruction for operating the suction pump 40 and an
instruction for stopping the suction pump 40, and changes in the
system internal pressure detected by the pressure sensor 47.
Further, the solid line in FIG. 11 shows the pattern of change in
the system internal pressure at the normal time shown in FIG.
7.
[0102] Referring to FIG. 11, in the time period xt from the time
t.sub.1 to the time t.sub.3, the system internal pressure is
detected at every fixed time .DELTA.t by the pressure sensor 47. In
FIG. 11, x.sub.1, x.sub.2 . . . x.sub.n-1, and x.sub.n show the
system internal pressures detected by the pressure sensor 47 at
every fixed time .DELTA.t . The system internal pressures x.sub.1,
x.sub.2 . . . X.sub.n-1, and x.sub.n at every fixed time .DELTA.t
detected by the pressure sensor 47 are stored once in the storage
device. In the embodiment according to the present invention, an
abnormality judgment estimation model able to estimate if an
abnormality occurs in the fuel vapor escape prevention system based
on the system internal pressures x.sub.1, x.sub.2 . . . x.sub.n-1,
and x.sub.n at every constant time .DELTA.t stored once in the
storage device by using a neural network is prepared. Note that, in
the embodiment, what is detected by the pressure sensor 47 is the
pressure of the inside of the fuel tank 5 and the inside of the
canister 6. Therefore, in this embodiment, the pressure sensor 47
does not necessarily have to be placed inside the suction passage
46. It can be arranged at any location at which the pressure at the
inside of the fuel tank 5 and the inside of the canister 6 can be
detected.
[0103] Next, the neural network used for preparing the abnormality
judgment estimation model will be explained while referring to FIG.
12. Referring to FIG. 12, in this neural network 70 as well, in the
same way as the neural network shown in FIG. 6, L=1 shows an input
layer, L=2 and L=3 show hidden layers, and L=4 shows an output
layer. As shown in FIG. 12, the input layer (L=1) is comprised of
n+k number of nodes. "n" number of input values x.sub.1, x.sub.2 .
. . x.sub.n-1, and x.sub.n and "k" number of input values xx.sub.1,
xx.sub.2 . . . xx.sub.k-1 and xx.sub.k are input to the nodes of
the input layer (L=1). In this case, the "n" number of input values
x.sub.1, x.sub.2 . . . x.sub.n-1, and x.sub.n are the system
internal pressures x.sub.1, x.sub.2 . . . x.sub.n-1, and x.sub.n at
every fixed time .DELTA.t shown in FIG. 11.
[0104] On the other hand, FIG. 12 illustrates the hidden layer
(L=2) and hidden layer (L=3), but the numbers of these hidden
layers may also be made a single layer or any other number of
layers. Further, the number of nodes of these hidden layers may
also be made any number of nodes. Further, in this embodiment, the
number of nodes of the output layer (L=4) is made four nodes and
the output values from the nodes of the output layer (L=4) are
shown by y.sub.1', y.sub.2', y.sub.3', and y.sub.4'. These output
value y.sub.1', y.sub.2', y.sub.3', and y.sub.4' arc sent to the
softmax layer SM where they are converted to the respectively
corresponding output values y.sub.1, y.sub.2, y.sub.3, and y.sub.4.
The total of these output values y.sub.1, y.sub.2, y.sub.3, and
y.sub.4 is 1. The output values y.sub.1, y.sub.2, y.sub.3, and
y.sub.4 express the ratios to 1.
[0105] Next, the input values xx.sub.1, xx.sub.2 . . . xx.sub.k-1,
and xx.sub.k in FIG. 12 will be explained while referring to the
list shown in FIG. 13. Now then, as explained above, in the
embodiment according to the present invention, a neural network is
used to estimate whether an abnormality occurs in the fuel vapor
escape prevention system based on the system internal pressures
x.sub.1, x.sub.2 . . . X.sub.n-1, and x.sub.n at every fixed time
.DELTA.t . In this regard, however, the system internal pressure
changes due to, for example, the atmospheric pressure and other
external factors and the remaining amount of fuel in the fuel tank
5 and other internal factors. Therefore, when estimating if an
abnormality occurs in the fuel vapor escape prevention system, it
is necessary to consider the effects of these factors.
[0106] FIG. 13 lists the input parameters to the neural network 70
acting as such factors. Note that, FIG. 13 lists input parameters
having a strong influence on changes in the system internal
pressure as essential input parameters, lists input parameters
which have a strong influence on changes in the system internal
pressure, though not to the extent of the essential input
parameters, as large influence input parameters, and lists input
parameters which have an influence on changes in the system
internal pressure, though not to the extent of the large influence
input parameters, as auxiliary input parameters.
[0107] As shown in FIG. 13, the system internal pressure x.sub.1,
x.sub.2 . . . x.sub.n-1 and atmospheric pressure xx.sub.1 are made
essential input parameters. In this case, the system internal
pressures x.sub.1, x.sub.2 . . . x.sub.n-1, and x.sub.n are
naturally essential input parameters. On the other hand, if the
atmospheric pressure changes, the system internal pressure also
changes accordingly. Therefore, the atmospheric pressure xx.sub.1
is made an essential input parameter.
[0108] On the other hand, when the system internal pressure becomes
a negative pressure due to the suction action by the suction pump
40, if the fuel in the fuel tank 5 evaporates, the system internal
pressure rises. In this case, the greater the amount of evaporation
of the fuel per unit time, the greater the amount of change in the
system internal pressure. On the other hand, the amount of
evaporation of the fuel per unit time is proportional to the
remaining amount of fuel in the fuel tank 5. Therefore, the greater
the remaining amount of fuel in the fuel tank 5, the greater the
influence given to the system internal pressure. Therefore, as
shown in FIG. 13, the remaining amount xx.sub.2 of fuel in the fuel
tank 5 is made a large influence input parameter.
[0109] Further, if the temperature of the fuel in the fuel tank 5
rises, the amount of evaporation of the fuel per unit time
increases, so the temperature of the fuel in the fuel tank 5 also
has an influence on the system internal pressure. However, this
influence on the system internal pressure is smaller than the
remaining amount of fuel in the fuel tank 5, so as shown in FIG.
13, the temperature of the fuel xx.sub.3 in fuel tank 5 is made an
auxiliary input parameter. Further, the performance of the suction
pump 40 also has quite a bit of an influence on the system internal
pressure, therefore, as shown in FIG. 13, the characteristic value
of flow rate xx.sub.4 of the suction pump 40 is made an auxiliary
input parameter.
[0110] FIG. 14 shows the relationship between this characteristic
value of flow rate of the suction pump 40 and the maximum pump flow
rate of the suction pump 40. As shown in FIG. 14, the
characteristic value of flow rate of the suction pump 40 is made
1.0 when the maximum pump flow rate of the suction pump 40 used as
a reference is G, and the characteristic value of flow rate of each
suction pump 40 is determined in accordance with the maximum pump
flow rate of each suction pump 40. The characteristic value of flow
rate of the suction pump 40 is a parameter showing the capacity of
the suction pump 40. If the capacity of the suction pump 40 is
high, the characteristic value of flow rate of the suction pump 40
becomes high, while if the capacity of the suction pump 40 is low,
the characteristic value of flow rate of the suction pump 40
becomes low.
[0111] FIG. 15 shows a list of what kinds of states the output
values y.sub.1', y.sub.2', y.sub.3', and y.sub.4' and output values
y.sub.1, y.sub.2, y.sub.3, and y.sub.4 shown in FIG. 12 show. As
will be understood from FIG. 15, the output value y.sub.1' and
output value y.sub.1 show a perforation abnormality in which the
walls of the fuel vapor circulation pipe 12 or 13 are perforated by
a small hole, the output value y.sub.2' and output value y.sub.2
show a valve opening abnormality in which the purge control valve
14 continues opened, the output value y.sub.3' and output value
y.sub.3 show a valve closing abnormality in which the purge control
valve 14 continues closed, and the output value y.sub.4' and output
value y.sub.4 show the normal time.
[0112] Now then, as the input values x.sub.1, x.sub.2 . . .
n.sub.n-1, and x.sub.n and the input values xx.sub.1, xx.sub.2 . .
. xx.sub.k-1, and xx.sub.k of the neural network 70 shown in FIG.
12, the values of only the essential input parameters shown in FIG.
13, that is, only the system internal pressures x.sub.1, x.sub.2, .
. . x.sub.n-1, and x.sub.n and atmospheric pressure xx.sub.1, can
be used. Of course, in addition to the values of the essential
input parameters, it is possible to make the values of the large
influence input parameters the input values xx.sub.1, xx.sub.2 . .
. xx.sub.k-1, and xx.sub.k and possible to make the values of the
large influence input parameters and the values of the auxiliary
input parameters in addition to the values of the essential input
parameters the input values xx.sub.1, xx.sub.2 . . . xx.sub.k-1,
and xx.sub.k . Note that, below, the embodiment will be explained
using as an example the case of making the values of the large
influence input parameters and the values of the auxiliary input
parameters in addition to the values of the essential input
parameters the input values xx.sub.1, xx.sub.2 . . . xx.sub.k-1,
and xx.sub.k (in this example, k=4).
[0113] FIG. 16 shows a training data set prepared using the input
values xx.sub.1, xx.sub.2 . . . xx.sub.k-1, and xx.sub.k, the input
values xx.sub.1, xx.sub.2 . . . xx.sub.k-1, and xx.sub.k, and
training data, that is, the truth labels yt. In this FIG. 16, the
input values xx.sub.1, xx.sub.2 . . . xx.sub.k-1, and xx.sub.k show
the system internal pressure for each fixed time .DELTA.t. This
system internal pressure is detected by the pressure sensor 47.
Further, in FIG. 16, the input values xx.sub.1, xx.sub.2 . . .
xx.sub.k-1, and xx.sub.k, that is, the input values xx.sub.1,
xx.sub.2, xx.sub.3, and xx.sub.4, respectively show the atmospheric
pressure, the remaining amount of fuel in the fuel tank 5, the
temperature of the fuel in the fuel tank 5, and the characteristic
value of flow rate of the suction pump 40, that is, the capacity of
the suction pump 40. In this case, the atmospheric pressure is
detected by the atmospheric pressure sensor 30, the remaining
amount of fuel in the fuel tank 5 is detected by the fuel level
gauge 7 and the gravity sensor 60, the temperature of the fuel in
the fuel tank 5 is detected by the temperature sensor 8, and the
characteristic value of flow rate of the suction pump 40 is
calculated based on the relationship shown in FIG. 14.
[0114] On the other hand, in FIG. 16, yt.sub.1 . . . yt.sub.S (in
this example, s=4) show the training data, that is, the truth
labels, for the output values y.sub.1', y.sub.2', y.sub.3', and
y.sub.4' and output values y.sub.1, y.sub.2, y.sub.3, and y.sub.4
shown in FIG. 15. That is, in FIG. 16, yt.sub.1 shows the truth
label when a perforation abnormality occurs in which the walls of
the fuel vapor flow pipe 12 or 13 are perforated by a small hole,
yt.sub.2 shows the truth label where a valve opening abnormality
occurs in which the purge control valve 14 continues opened,
yt.sub.3 shows the truth label where a valve closing abnormality
occurs in which the purge control valve 14 continues closed, and
yt.sub.4 shows the truth label at the normal time.
[0115] In this case, for example, when a perforation abnormality
occurs in which the walls of the vapor circulation pipe 12 or 13
are perforated by a small hole, only the truth label yt.sub.1 is
made 1, while the remaining truth labels yt.sub.2, yt.sub.3, and
yt.sub.4 are all made zero. Similarly, when a valve opening
abnormality occurs in which the purge control valve 14 continues
opened, only the truth label yt.sub.2 is made 1, while the
remaining truth labels yt.sub.1, yt.sub.3, and yt.sub.4 are all
made zero. When a valve closing abnormality occurs in which the
purge control valve 14 continues closed, only the truth label
yt.sub.3 is made 1, while the remaining truth labels yt.sub.1,
yt.sub.2, and yt.sub.4 are all made zero, and when normal, only the
truth label yt.sub.4 is made 1 and the remaining truth labels
yt.sub.1, yt.sub.2, and yt.sub.3 are all made zero.
[0116] On the other hand, as shown in FIG. 16, in this training
data set, "m" number of data showing the relationship between the
input values xx.sub.1, xx.sub.2 . . . xx.sub.n-1, x.sub.n and the
input values xx.sub.1, xx.sub.2 . . . xx.sub.k-1, xx.sub.k and
truth labels yt are acquired. For example, at the No. 2 data, the
acquired input values x.sub.12, x.sub.22 . . . x.sub.n-12, and
x.sub.n2, input values xx.sub.12, xx.sub.22 . . . xx.sub.k-12, and
xx.sub.k2, and truth labels yt.sub.12 . . . yt.sub.S2 are listed,
while at the No. m-1 data, the input values x.sub.1m-1, x.sub.2m-1
. . . x.sub.n-1m-1, and x.sub.nm-1, the input values xx.sub.1m-1,
xx.sub.2m-1 . . . . xx.sub.k-1m-1and xx.sub.km-1, and the truth
labels yt.sub.Sm-1 of the acquired input parameters are listed.
[0117] Next, the method of preparation of the training data set
shown in FIG. 16 will be explained. FIG. 17 shows one example of
the method of preparation of the training data set. Referring to
FIG. 17, the engine body 1, fuel tank 5, canister 6, etc. shown in
FIG. 1 are placed inside a sealed test chamber 80 able to be
adjusted in internal pressure. In order to estimate if an
abnormality occurs in the fuel vapor escape prevention system by
using a test control device 81, processing for detection of an
abnormality is performed operating the purge control valve 14, flow
path switching valve 42, and suction pump 40 in a predetermined
order of operation. At this time, the state of the fuel vapor
escape prevention system is successively changed to a state where a
perforation abnormality occurs in which the walls of the fuel vapor
flow pipe 12 or 13 are perforated by a small hole, a state where a
valve opening abnormality occurs in which the purge control valve
14 continues opened, a state where a valve closing abnormality
occurs in which the purge control valve 14 continues closed, and a
normal state. In the changed states, the combination of atmospheric
pressure, the remaining amount of fuel in the fuel tank 5, the
temperature of the fuel in the fuel tank 5, and the characteristic
value of flow rate of the suction pump 40 is successively changed
while processing for detection of an abnormality is repeatedly
performed.
[0118] While this processing for detection of an abnormality is
being performed, the data required for preparing the training data
set is acquired. FIG. 18 shows the routine for processing for
detection of an abnormality performed in the test control device 81
for performing this processing for detection of an abnormality.
This routine for processing for detection of an abnormality is
performed by interruption at every fixed time .DELTA.t shown in
FIG. 11. Note that, in the routine shown in FIG. 18, "t" shows the
point of time found when defining the time t.sub.0 when the engine
stops operating at FIG. 17 as zero and using this time t.sub.0 as
the starting point.
[0119] Referring to FIG. 18, first, at step 100, it is judged if
the time "t" is the time t.sub.1 shown in FIG. 11. When the time
"t" is before the time t.sub.1 shown in FIG. 11, the processing
cycle is ended. As opposed to this, when it is judged that the time
"t" is not before the time t.sub.1 shown in FIG. 11, the routine
proceeds to step 101 where it is judged if the time "t" reaches the
time t.sub.1 shown in FIG. 11. When the time "t" reaches the time
t.sub.1 shown in FIG. 11, the routine proceeds to step 102 where a
switching instruction switching the flow path switching valve 42 to
the test position is issued, then the routine proceeds to step 103
where an instruction for operating the suction pump 40 is issued.
Next, at step 104, the acquisition and storage of the system
internal pressure x.sub.n detected by the pressure sensor 47 are
started. .DELTA.t this time, the system internal pressure x.sub.n
is stored inside the test control device 81.
[0120] On the other hand, when at step 101 it is judged that the
time "t" is not the time t.sub.1 shown in FIG. 11, the routine
proceeds to step 105 where the system internal pressure x,, is
acquired and stored in the test control device 81. That is, as
shown in FIG. 11, the system internal pressure x.sub.n is acquired
at every fixed time .DELTA.t while the system internal pressure
x.sub.t, acquired at every fixed time .DELTA.t is stored inside the
test control device 81. Next, at step 106, it is judged if the time
"t" becomes the time t.sub.2 shown in FIG. 11. When the time "t"
becomes the time t.sub.2 shown in FIG. 11, the routine proceeds to
step 107 where a valve opening instruction making the purge control
valve 14 open is issued.
[0121] On the other hand, when at step 106 it is judged that the
time "t" is not the time t.sub.2 shown in FIG. 11, the routine
proceeds to step 108 where it is judged if the time "t" becomes the
time t.sub.3 shown in FIG. 11. When the time "t" is not the time
t.sub.3 shown in FIG. 11, the processing cycle is ended. As opposed
to this, when it is judged that the time "t" becomes the time
t.sub.3 shown in FIG. 11, the routine proceeds to step 109 where a
valve closing instruction making the purge control valve 14 close
is issued. Next, at step 110, a switching instruction switching the
flow path switching valve 42 to the normal position is issued.
Next, at step 111, an instruction for stopping the suction pump 40
is issued. Next, at step 112, the action of acquiring and storing
the system internal pressure x.sub.n detected by the pressure
sensor 47 is stopped. Next, at step 113, the processing for
detection of an abnormality is ended. In the test control device 81
shown in FIG. 17, if the processing for detection of an abnormality
ends, the next processing for detection of an abnormality is
started.
[0122] In this way, the system internal pressure x.sub.n for each
fixed time .DELTA.t when the combination of the atmospheric
pressure, the remaining amount of fuel in the fuel tank 5, the
temperature of the fuel in the fuel tank 5, and the characteristic
value of flow rate of the suction pump 40 is changed in each state
of a state where a perforation abnormality occurs in which the
walls of the fuel vapor flow pipe 12 or 13 are perforated by a
small hole, a state where a valve opening abnormality occurs in
which the purge control valve 14 continues opened, a state where a
valve closing abnormality occurs in which the purge control valve
14 continues closed, and a normal state are stored in the test
control device 81. That is, the No. 1 to No. "m" input values
x.sub.1m, x.sub.2m . . . x.sub.nm-1, and x.sub.nm, the input values
xx.sub.1m, xx.sub.2m . . . xx.sub.km-1, and xx.sub.km, and the
truth labels yt.sub.sm(m=1, 2, 3 . . . m) of the training data set
shown in FIG. 16 are stored inside the test control device 81.
[0123] If a training data set such as shown in FIG. 16 is prepared
in this way, electronic data of the prepared training data set is
used to learn the weights of the neural network 70 shown in FIG.
12. In the example shown in FIG. 17, a learning apparatus 82 for
learning the weights of the neural network is provided. As this
learning apparatus 82, a personal computer can also be used. As
shown in FIG. 17, this learning apparatus 82 is provided with a CPU
(microprocessor) 83 and a storage device, that is, the memory 84.
In the example shown in FIG. 17, the numbers of nodes of the neural
network shown in FIG. 12 and the electronic data of the prepared
training data set are stored in the memory 84 of the learning
apparatus 82 and the weights of the neural network are learned at
the CPU 83.
[0124] FIG. 19 shows the routine for processing for learning the
weights of a neural network performed at the learning apparatus 82.
Referring to FIG. 19, first, at step 200, the data of the training
data set for the neural network 70 stored in the memory 84 of the
learning apparatus 82 is read in. Next, at step 201, the number of
nodes of the input layer (L=1), the numbers of nodes of the hidden
layer (L=2) and hidden layer (L=3), and the number of nodes of the
output layer (L=4) of the neural network 70 are read in. Next, at
step 202, the neural network 70 such as shown in FIG. 12 is
prepared based on these numbers of nodes.
[0125] Next, at step 203, the weights of the neural network 70 are
learned. At this step 203, first, the No. 1 input values x.sub.1,
x.sub.2 . . . x.sub.n-1, and x.sub.n and input values xx.sub.1,
xx.sub.2 . . . xx.sub.k-1, and xx.sub.k of FIG. 16 are input to the
nodes of the input layer of the neural network 70 (L=1). At this
time, the output values y.sub.1', y.sub.2', y.sub.3', and y.sub.4'
are output from the nodes of the output layer of the neural network
70. These output value y.sub.1', y.sub.2', y.sub.3', and y.sub.4'
are sent to the softmax layer SM and converted to respectively
corresponding output values y.sub.1, Y.sub.2, y.sub.3, and y.sub.4.
Next, using these output value y.sub.1, y.sub.2, y.sub.3, and
y.sub.4 and truth labels yt.sub.1-yt.sub.S, the above-mentioned
cross entropy error E is calculated. The weights of the neural
network 70 are learned using the error backpropagation algorithm so
that the cross entropy error E becomes smaller.
[0126] If the weights of the neural network 70 finish being learned
based on the No. 1 data of FIG. 16, next the weights of the neural
network 70 are learned based on the No. 2 data of FIG. 16 using the
error backpropagation algorithm. Similarly, the weights of the
neural network 70 are successively learned until the No. "m" data
of FIG. 16. When the weights of the neural network 70 have finished
being learned for all of the No. 1 to No. "m" data of FIG. 16, the
routine proceeds to step 204.
[0127] At step 204, it is judged if the cross entropy error E
becomes a preset error setting or less. When it is judged that the
cross entropy error E does not become the preset error setting or
less, the routine returns to step 203 where, again, learning of the
weights of the neural network 70 is performed based on the training
data set shown in FIG. 16. Next, learning of the weights of the
neural network 70 is continued until the cross entropy error E
becomes the preset error setting or less. When at step 204 it is
judged that the cross entropy error E becomes the preset error
setting or less, the routine proceeds to step 205 where the learned
weights of the neural network 70 are stored in the memory 84 of the
learning apparatus 82. In this way, an abnormality judgment
estimation model able to estimate if an abnormality occurs in the
fuel vapor escape prevention system is prepared.
[0128] In the embodiment according to the present invention, the
thus prepared abnormality judgment estimation model of the fuel
vapor escape prevention system is used to diagnose a fault in the
fuel vapor escape prevention system of a commercially available
vehicle. To this end, the abnormality judgment estimation model of
the fuel vapor escape prevention system is stored in the electronic
control unit 20 of the commercially available vehicle. FIG. 20
shows a routine for reading data into the electronic control unit
performed at the electronic control unit 20 of a commercially
available vehicle so as to store the abnormality judgment
estimation model of the fuel vapor escape prevention system in the
electronic control unit 20.
[0129] If referring to FIG. 20, first, at step 300, the number of
nodes of the input layer (L=1) of the neural network 7, the numbers
of nodes of the hidden layer (L=2) and hidden layer (L=3), and the
number of nodes of the output layer (L=4) shown in FIG. 12 are read
into the memory 22 of the electronic control unit 20. Next, at step
301 a neural network 71 such as shown in FIG. 21 is prepared based
on these numbers of nodes. As will be understood from FIG. 21, in
this neural network 71, the softmax layer is removed. Note that, in
this case, the neural network 71 may also be provided with the
softmax layer 71 such as shown in FIG. 12. Next, at step 302, the
learned weights of the neural network 70 are read into the memory
22 of the electronic control unit 20. Due to this, the abnormality
judgment estimation model of the fuel vapor escape prevention
system is stored in the electronic control unit 20 of the
commercially available vehicle.
[0130] Next, referring to FIG. 22, the routine for detection of an
abnormality of the fuel vapor escape prevention system performed in
a commercially available vehicle will be explained. This routine is
also performed by interruption at every fixed time. Referring to
FIG. 22, first, at step 400, it is judged if the vehicle has
stopped being driven based on the output signal of the driving
start/stop switch 31. When the vehicle has not stopped being
driven, the processing cycle is ended. As opposed to this, when it
is judged that the vehicle has stopped being driven, the routine
proceeds to step 401 where it is judged if a detection permission
flag permitting detection of an abnormality of the fuel vapor
escape prevention system is set. When the routine first proceeds to
step 401 after the vehicle stops being driven, the detection
permission flag is not set, so the routine proceeds to step 402. At
step 402, the processing for detection of an abnormality shown in
FIG. 18 is performed.
[0131] If this processing for detection of an abnormality is
performed, the system internal pressure x.sub.n is acquired for at
every fixed time .DELTA.t and the system internal pressure x.sub.n
acquired at every fixed time .DELTA.t is stored in the memory 22 of
the electronic control unit 20. Next, at step 403, it is judged if
the processing for detection of an abnormality has ended. When the
processing for detection of an abnormality has not ended, the
processing cycle is ended. As opposed to this, when it is judged
that the processing for detection of an abnormality has ended, the
routine proceeds to step 404 where the detection permission flag is
set. If the detection permission flag is set, at the next
processing cycle, the routine proceeds from step 401 to step
405.
[0132] At step 405, the system internal pressures x.sub.1, x.sub.2
. . . x.sub.n-1, and x.sub.n for each fixed time .DELTA.t stored in
the memory 22 of the electronic control unit 20 are read in. Next,
at step 406, the input values xx.sub.1, xx.sub.2 . . . xx.sub.k-1,
and xx.sub.k stored in the memory 22 of the electronic control unit
20 are read in. Next, at step 407, the system internal pressures
x.sub.nx.sub.1, x.sub.2 . . . x.sub.n-1, and x.sub.n for each fixed
time .DELTA.t and the input values xx.sub.1, xx.sub.2 . . .
xx.sub.k-1, and xx.sub.k are input to the nodes of the input layer
(L=1) of the neural network 71 shown in FIG. 21. At this time,
output value y.sub.1', y.sub.2', y.sub.3', and y.sub.4' are output
from the nodes of the output layer of the neural network 70. Due to
this, at step 408, the output values y.sub.1', y.sub.2', y.sub.3',
and y.sub.4' are acquired.
[0133] Next, at step 409, the largest output value y.sub.i' is
selected from the acquired output values y.sub.1', y.sub.2',
y.sub.3', and y.sub.4'. At this time, it is estimated that the
abnormal state shown in FIG. 15 corresponding to this largest
output value y.sub.i' occurs. Therefore, at step 410, it is judged
that the abnormal state shown in FIG. 15 corresponding to this
largest output value y.sub.i' occurs and, for example, a warning
light showing that the abnormal state shown in FIG. 15
corresponding to this largest output value y.sub.i' occurs is
turned on. Next, at step 411, the detection of an abnormality is
ended.
[0134] In this way, in the abnormality detection device of a fuel
vapor escape prevention system according to the present invention,
the fuel vapor escape prevention system is provided with the
canister 6 at which the fuel vapor chamber 10 and atmospheric
pressure chamber 11 are formed at the two sides of the activated
carbon layer 9. The fuel vapor chamber 10 is on the one hand
communicated with the inside space above the fuel level of the fuel
tank 5 and is on the other hand communicated with the inside of the
intake passage of the engine through the purge control valve 14.
Furthermore, the fuel vapor escape prevention system is provided
with the flow path switching valve 42 able to selectively connect
the atmospheric pressure chamber 11 to the atmosphere and the
suction pump 40 and the pressure sensor 47 detecting the pressure
at the inside of the fuel tank 5 and the inside of the canister 6.
At the time of stopping operation of the vehicle, processing for
detection of an abnormality generating a valve closing instruction
making the purge control valve 14 close, a switching instruction
switching the switched position of the flow path switching valve 42
to the switched position where the atmospheric pressure chamber 11
is connected to the suction pump 40, and a pump operation
instruction making the suction pump 40 operate so as to make the
inside of the fuel tank and the inside of the canister 6 a negative
pressure is performed. When this processing for detection of an
abnormality is performed, the pressure at the inside of the fuel
tank 5 and the inside of the canister 6 detected at every fixed
time by the pressure sensor 40 is stored in the storage device. The
learned neural network learned in weights using the pressure at the
inside of the fuel tank 5 and inside of the canister 6 at every
fixed time stored in the storage device and at least the
atmospheric pressure when the processing for detection of an
abnormality is performed as input parameters of the neural network
and using a case where perforation occurs in the system causing
leakage of fuel vapor as a truth label is stored, and, at the time
of stopping operation of the vehicle, a perforation abnormality
causing fuel vapor to leak is detected from the input parameters by
using the learned neural network.
[0135] In this case, in the embodiment according to the present
invention, the above-mentioned processing for detection of an
abnormality includes processing for generating a valve opening
instruction making the purge control valve 14 open after generating
the valve closing instruction of the purge control valve 14, a
learned neural network learned in weights using the pressure at the
inside of the fuel tank 5 and inside of the canister 6 at every
fixed time stored in the storage device and at least the
atmospheric pressure when the processing for detection of an
abnormality is performed as input parameters of the neural network
and using a case where perforation occurs in the above-mentioned
system causing leakage of fuel vapor, a case where a valve opening
abnormality occurs in which the purge control valve 14 continues
opened, and a case where a valve closing abnormality occurs in
which the purge control valve 14 continues closed as truth labels
is stored, and, at the time of stopping operation of the vehicle, a
perforation abnormality causing fuel vapor to leak, a valve opening
abnormality of the purge control valve 14, and a valve closing
abnormality of the purge control valve 14 are detected from the
input parameters by using the learned neural network.
[0136] Further, in this embodiment according to the present
invention, the above-mentioned input parameters are comprised of
the pressures at the inside of the fuel tank 5 and the inside of
the canister 6 at every fixed time stored in the storage device and
the atmospheric pressure when processing for detection of an
abnormality is performed and the remaining amount of fuel in the
fuel tank 5 when processing for detection of an abnormality is
performed. Alternatively, in the embodiment according to the
present invention, the above-mentioned input parameters are
comprised of the pressures at the inside of the fuel tank 5 and the
inside of the canister 6 at every fixed time stored in the storage
device and the atmospheric pressure when processing for detection
of an abnormality is performed, the remaining amount of fuel in the
fuel tank 5 when processing for detection of an abnormality is
performed, the temperature of the fuel of the fuel tank 5, and a
parameter showing the capacity of the suction pump 40.
[0137] FIG. 23A to FIG. 36 show another embodiment configured so as
to detect further more abnormal states. In this another embodiment,
as the suction pump module 16 shown in FIG. 1, a suction pump
module 16 schematically shown in FIG. 23A and FIG. 23B is used.
Referring to FIG. 23A and FIG. 23B, in the suction pump module 16
shown in FIG. 23A and FIG. 23B, a reference pressure detection
passage 50 connecting the atmospheric pressure chamber connection
path 43 and the suction passage 46 is added to the suction pump
module 16 shown in FIG. 2A and FIG. 2B. Inside this reference
pressure detection passage 50, a restricted opening 51 is
provided.
[0138] Next, processing for detection of an abnormality performed
using the suction pump module 16 shown in FIG. 23A and FIG. 23B at
the time of stopping operation of the vehicle will be explained.
FIG. 24 shows, similar to FIG. 7, a valve closing instruction
making the purge control valve 14 close and a valve opening
instruction making the purge control valve 14 open, a switching
instruction switching the flow path switching valve 42 to the
normal position and a switching instruction switching the flow path
switching valve 42 to the test position, an instruction for
operating the suction pump 40 and an instruction for stopping the
suction pump 40, and a change in the system internal pressure
detected by the pressure sensor 47.
[0139] In FIG. 24, to shows the time of stopping operation of the
vehicle. At this time, a valve closing instruction making the purge
control valve 14 close, a switching instruction switching the flow
path switching valve 42 to the normal position, and an instruction
for stopping the suction pump 40 are issued. FIG. 24 shows when the
purge control valve 14, flow path switching valve 42, and suction
pump 40 are operating normally based on these instructions.
Therefore, at the time of stopping operation of the vehicle, the
purge control valve 14 is made to close, the flow path switching
valve 42 is switched to the normal position, and the suction pump
40 is made to stop. This state is maintained for a constant time
from the time of stopping operation of the vehicle to the time
t.sub.1. During this constant period, the suction pump 40 continues
stopped, so no action of suction by the suction pump 40 is
performed. Therefore, the system internal pressure detected by the
pressure sensor 47 becomes atmospheric pressure.
[0140] Next, if reaching the time t.sub.2, an instruction for
operating the suction pump 40 is issued. At this time, the flow
path switching valve 42 has been switched to the normal position
shown in FIG. 23A. Therefore, if the suction pump 40 is operated,
outside air is sucked in from the atmosphere communication pipe 17
through the atmosphere communication pipe connection path 44,
atmosphere communication path 45, first passage 48, atmosphere
communication chamber connection path 43, reference pressure
detection passage 50 having the restricted opening 51, and suction
passage 46. At this time, since the restricted opening 51 is
provided, the inside of the suction passage 46 becomes a negative
pressure. Therefore, the system internal pressure detected by the
pressure sensor 47 falls as shown in FIG. 24.
[0141] In this case, the changes in the system internal pressure
detected by the pressure sensor 47 show the changes in the system
internal pressure when perforated by a hole of the same diameter as
the diameter of the restricted opening 51. Therefore, the changes
in the system internal pressure at this time become the reference
for judging whether the fuel vapor escape prevention system is
perforated by a hole. Therefore, the passage 50 will be referred to
as the "reference pressure detection passage". Next, if reaching
the time t.sub.2, a switching instruction switching the flow path
switching valve 42 to the test position shown in FIG. 23B is
issued. If the flow path switching valve 42 is switched to the test
position, since, at this time, the pressure at the inside of the
fuel tank 5 and the inside of the canister 6 is atmospheric
pressure, the pressure at the inside of the suction passage 46
rises. Therefore, if reaching the time t.sub.2, the system internal
pressure detected by the pressure sensor 47 rises as shown in FIG.
24. Next, if reaching the time t.sub.3, it starts to fall due to
the action of suction of air by the suction pump 40 from the inside
of the fuel tank 5 and the inside of the canister 6. Next, around
when reaching the time t.sub.4, the pressure no longer falls and
the system internal pressure remains in the fallen state.
[0142] If reaching the time t.sub.4, a valve opening instruction
making the purge control valve 14 open is issued. On the other
hand, at this time, the flow path switching valve 42 is maintained
at the test position and the suction pump 40 continues to be
operated. Therefore, at this time, the suction action due to the
suction pump 40 is continued, but the purge control valve 14 is
made to open, so the system internal pressure rapidly rises and
becomes atmospheric pressure. Next, if reaching the time t.sub.5,
the purge control valve 14, flow path switching valve 42, and
suction pump 40 are returned to the states at the time of stopping
operation of the vehicle. That is, if reaching the time t.sub.5, a
valve closing instruction making the purge control valve 14 close,
a switching instruction switching the flow path switching valve 42
to the normal position, and an instruction for stopping the suction
pump 40 are issued.
[0143] On the other hand, in this embodiment, as shown in FIG. 24,
from the time of stopping operation of the vehicle, that is, at the
time period xt from the time t.sub.0 to the time t.sub.5, the
system internal pressure is detected at every fixed time .DELTA.t
by the pressure sensor 47. In FIG. 24, x.sub.1, x.sub.2 . . .
x.sub.n-1, and x.sub.n show the system internal pressures detected
by the pressure sensor 47 at every fixed time .DELTA.t. The system
internal pressures x.sub.1, x.sub.2 . . . x.sub.n-1, and x.sub.n at
every fixed time .DELTA.t detected by the pressure sensor 47 are
stored once in the storage device. In this embodiment as well, an
abnormality judgment estimation model able to estimate whether an
abnormality occurs in the fuel vapor escape prevention system based
on the system internal pressures x.sub.1, x.sub.2 . . . x.sub.n-1,
and x.sub.n at every fixed time .DELTA.t stored by the storage
device by using a neural network is prepared.
[0144] In this embodiment as well, if an abnormality occurs in the
fuel vapor escape prevention system, the pattern of change in the
system internal pressure detected by the pressure sensor 47 becomes
a pattern of change different from the pattern of change in the
system internal pressure at the normal time shown in FIG. 24. Next,
this will be explained with reference to FIG. 25 to FIG. 32. Note
that, FIG. 25 to FIG. 32, in the same way as FIG. 24, shows a valve
closing instruction making the purge control valve 14 close and
valve opening instruction making the purge control valve 14 open, a
switching instruction switching the flow path switching valve 42 to
the normal position and a switching instruction switching the flow
path switching valve 42 to the test position, an instruction for
operating the suction pump 40 and an instruction for stopping the
suction pump 40, and a change in the system internal pressure
detected by the pressure sensor 47. Further, in FIG. 25 to FIG. 32,
the broken lines show the patterns of change in the system internal
pressure at the normal time shown in FIG. 24.
[0145] The solid line of FIG. 25 shows the pattern of change in the
system internal pressure detected by the pressure sensor 47 in the
case where an abnormality occurs in the pressure sensor 47. If an
abnormality occurs in the pressure sensor 47 and the output of the
pressure sensor 47 becomes no longer stable, as shown between the
time t.sub.0 and the time t.sub.1, the system internal pressure
detected by the pressure sensor 47 is not maintained at the
atmospheric pressure but repeatedly fluctuates. Therefore, the
pattern of change in the system internal pressure in this case as
well becomes a pattern of change different from the pattern of
change in the system internal pressure at the normal time.
[0146] The solid line of FIG. 26 shows the pattern of change in the
system internal pressure detected by the pressure sensor 47 in the
case where an abnormality where the suction pump 40 continues
operating occurs in which the suction pump 40 continues operating
even if an instruction to stop the suction pump 40 is issued. As
will be understood from FIG. 26, between the time t.sub.0 and the
time t.sub.2, the flow path switching valve 42 is switched to the
normal position shown in FIG. 23A. Therefore, if the suction pump
40 is made to operate at this time, the outside air is sucked in
from the atmosphere communication pipe 17 through the atmosphere
communication pipe connection path 44, atmosphere communication
path 45, first passage 48, atmosphere communication chamber
connection path 43, reference pressure detection passage 50 having
the restricted opening 51, and suction passage 46. At this time,
the inside of the suction passage 46 becomes a negative pressure.
Therefore, at this time, the system internal pressure detected by
the pressure sensor 47 is a low value such as shown in FIG. 26.
Therefore, the pattern of change in the system internal pressure in
this case as well becomes a pattern of change different from the
pattern of change in the system internal pressure at the normal
time.
[0147] The solid line of FIG. 27 shows the pattern of change in the
system internal pressure detected by the pressure sensor 47 in the
case where an abnormality where the suction pump 40 stops operating
occurs in which the suction pump 40 continues stopped even if an
instruction for operating the suction pump 40 is issued. At this
time, the entirety of the inside of the fuel vapor escape
prevention system becomes the atmospheric pressure, so as shown in
FIG. 27, the system internal pressure detected by the pressure
sensor 47 is maintained at atmospheric pressure. Therefore, the
pattern of change in the system internal pressure in this case as
well becomes a pattern of change different from the pattern of
change in the system internal pressure at the normal time.
[0148] The solid line of FIG. 28 shows the pattern of change in the
system internal pressure detected by the pressure sensor 47 in the
case where an abnormality where the flow path switching valve 42
sticks at the test position occurs in which the flow path switching
valve 42 continues to stick at the test position shown in FIG. 23B.
In this case, if the suction pump 40 is made to operate at the time
t.sub.1, outside air is sucked in from the atmosphere communication
pipe 17 through the atmosphere communication pipe connection path
44, atmosphere communication path 45, first passage 48, atmosphere
communication chamber connection path 43, reference pressure
detection passage 50 having the restricted opening 51, and suction
passage 46. Accordingly, the system internal pressure detected by
the pressure sensor 47 rapidly falls as shown in FIG. 28 and is
maintained at the fallen state. Therefore, the pattern of change in
the system internal pressure in this case as well becomes a pattern
of change different from the pattern of change in the system
internal pressure at the normal time.
[0149] The solid line of FIG. 29 shows the pattern of change in the
system internal pressure detected by the pressure sensor 47 if the
abnormality where the flow path switching valve 42 sticks at the
normal position occurs in which the flow path switching valve 42
continues to stick to the normal position shown in FIG. 23A. In
this case, if, at the time t.sub.1, the suction pump 40 is made to
operate, the suction action of the air at the inside of the fuel
tank 5 and the inside of the canister 6 is started. Therefore, the
system internal pressure detected by the pressure sensor 47
gradually falls such as shown in FIG. 29. Therefore, the pattern of
change in the system internal pressure in this case as well becomes
a pattern of change different from the pattern of change in the
system internal pressure at the normal time.
[0150] The solid line of FIG. 30, for example, shows the pattern of
changes in the system internal pressure if the walls of the fuel
vapor flow pipe 12 or 13 are perforated by a small hole. In this
case, the outside air continues to flow inside the system through
the small hole, so the system internal pressure does not fall down
to the system internal pressure at the normal time and therefore
the pattern of change in the system internal pressure in this case
as well becomes a pattern of change different from the pattern of
change in the system internal pressure at the normal time.
[0151] The solid line of FIG. 31 shows when a valve opening
abnormality occurs in which the purge control valve 14 continues
opened even if a valve closing instruction making the purge control
valve 14 close is issued. At this time, the pressure at the inside
of the fuel tank 5 and the inside of the canister 6 is atmospheric
pressure. Therefore, at the time t.sub.2, even if the flow path
switching valve 42 is switched from the normal position shown in
FIG. 23A to the test position shown in FIG. 23B and the suction
pump 40 is made to operate, the system internal pressure detected
by the pressure sensor 47, as shown in FIG. 31, rises to
atmospheric pressure and is maintained at atmospheric pressure.
Therefore, the pattern of change in the system internal pressure in
this case as well becomes a pattern of change different from the
pattern of change in the system internal pressure at the normal
time.
[0152] The solid line in FIG. 32 shows when a valve closing
abnormality of the purge control valve 14 continuing closed even if
a valve opening instruction making the purge control valve 14 open
is issued. At this time, at the time 14, even if a valve opening
instruction making the purge control valve 14 open is issued, the
purge control valve 14 continues closed, so the inside of the fuel
tank 5 and the inside of the canister 6 are maintained at a
negative pressure and the system internal pressure detected by the
pressure sensor 47 is maintained at a low state such as shown in
FIG. 32. Therefore, the pattern of change in the system internal
pressure in this case also becomes a pattern of change different
from the pattern of change in the system internal pressure at the
normal time.
[0153] If an abnormality occurs in this way, the pattern of change
in the system internal pressure becomes a pattern of change
different from the pattern of change in the system internal
pressure at the normal time. Therefore, in this embodiment as well,
as shown in FIG. 24, in the time period st from the time t.sub.0 to
the time t.sub.5, the system internal pressure is detected at every
fixed time .DELTA.t by the pressure sensor 47. In FIG. 24, x.sub.1,
x.sub.2 . . . x.sub.n-1, and x.sub.n show the system internal
pressure detected at every fixed time .DELTA.t by the pressure
sensor 47. The system internal pressures x.sub.1, x.sub.2 . . .
x.sub.n-1, and x.sub.n at every fixed time .DELTA.t detected by the
pressure sensor 47 are stored once in the storage device. In this
embodiment as well, an abnormality judgment estimation model able
to estimate whether an abnormality occurs in the fuel vapor escape
prevention system based on the system internal pressures x.sub.1,
x.sub.2 . . . x.sub.n-1, and x.sub.n at every fixed time .DELTA.t
stored by the storage device by using a neural network is
prepared.
[0154] Next, the neural network 72 used for preparation of the
abnormality judgment estimation model will be explained while
referring to FIG. 33. If referring to FIG. 33, in this neural
network 72 as well, in the same way as the neural network shown in
FIG. 12, L=1 shows an input layer, L=2 and L=3 show hidden layers,
and L=4 shows an output layer. As shown in FIG. 33, the input layer
(L=1) is comprised of n+k number of nodes while the "n" number of
input values x.sub.1, x.sub.2 . . . x.sub.n-1, and x.sub.n and the
"k" number of input values xx.sub.1, xx.sub.2 . . . xx.sub.k-1, and
xx.sub.k are input to the nodes of the input layer (L=1). In this
case, the "n" number of input values x.sub.1, x.sub.2 . . .
x.sub.n-1, and x.sub.n are system internal pressures x.sub.1,
x.sub.2 . . . x.sub.n-1, and x.sub.n for each fixed value .DELTA.t
shown in FIG. 24.
[0155] On the other hand, FIG. 33 describes the hidden layer (L=2)
and hidden layer (L=3), but the number of these hidden layers may
also be made a single layer or any other number of layers. Further,
the number of nodes of these hidden layers may also be made any
number of nodes. Further, in this embodiment, the number of nodes
of the output layer (L=4) is made nine nodes and the output values
from the nodes of the output layer (L=4) are shown by y.sub.1',
y.sub.2', y.sub.3', y.sub.4', y.sub.5', y.sub.6', y.sub.7',
y.sub.8', and y.sub.9'. These output values y.sub.1', y.sub.2',
y.sub.3', y.sub.4', y.sub.5', y.sub.6', y.sub.7', y.sub.8', and
y.sub.9' are sent to the softmax layer SM where they are converted
to the respectively corresponding output values y.sub.1, y.sub.2,
y.sub.3, y.sub.4, y.sub.5, y.sub.6, y.sub.7, y.sub.8, and y.sub.9.
The total of these output values y.sub.1, y.sub.2, y.sub.3,
y.sub.4, y.sub.5, y.sub.6, y.sub.7, y.sub.8, and y.sub.9 is 1. The
output values y.sub.1, y.sub.2, y.sub.3, y.sub.4, y.sub.5, y.sub.6,
y.sub.7, y.sub.8, , and y.sub.9 express the ratios to 1.
[0156] The input values xx.sub.1, xx.sub.2 . . . xx.sub.k-1, and
xx.sub.k in FIG. 33 are the same as the input values xx.sub.1,
xx.sub.2 . . . xx.sub.k-1, and xx.sub.k in FIG. 12. These input
values xx.sub.1, xx.sub.2 . . . xx.sub.k-1, and xx.sub.k were
already explained while referring to the list shown in FIG. 13, so
the explanations of these input values xx.sub.1, xx.sub.2 . . .
xx.sub.k-1, and xx.sub.k will be omitted.
[0157] FIG. 34 shows a list of what kinds of states the output
values y.sub.1', y.sub.2', y.sub.3', y.sub.4', y.sub.5', y.sub.6',
y.sub.7', y.sub.8', and y.sub.9' and output values y.sub.1,
y.sub.2, y.sub.3, y.sub.4, y.sub.5, y.sub.6, y.sub.7, y.sub.8, and
y.sub.9 shown in FIG. 33 show. As will be understood from FIG. 34,
the output value y.sub.1' and output value y.sub.1 show a
perforation abnormality in which the walls of the fuel vapor flow
pipe 12 or 13 are perforated by a small hole, the output value
y.sub.2' and output value y.sub.2 show a valve opening abnormality
in which the purge control valve 14 continues opened, the output
value y.sub.3' and output value y.sub.3 show a valve closing
abnormality in which the purge control valve 14 continues closed,
the output value y.sub.4' and output value y.sub.4 show an
abnormality of the pressure sensor 47, the output value y.sub.5'
and output value y.sub.5 show an abnormality of the flow path
switching valve 42 sticking at the normal position, the output
value y.sub.6' and output value y.sub.6 show an abnormality of the
flow path switching valve 42 sticking at the test position, the
output value y.sub.7' and output value y.sub.7 show an abnormality
in which the suction pump 40 continues operating, the output value
y.sub.8' and output value y.sub.8 show an abnormality in which the
suction pump 40 continues to stop operating, and the output value
y.sub.9' and output value y.sub.9 show the normal times.
[0158] Now then, in this embodiment as well, as the input values
x.sub.1, x.sub.2 . . . x.sub.n-1, and x.sub.n and the input values
xx.sub.1, xx.sub.2 . . . xx.sub.k-1, and xx.sub.k of the neural
network 72 shown in FIG. 33, the values of only the essential input
parameters shown in FIG. 13, that is, only the system internal
pressures x.sub.1, x.sub.2 . . . x.sub.n-1, and x.sub.n atmospheric
pressure xx.sub.1, can be used. Of course, in addition to the
values of the essential input parameters, it is possible to make
the values of the large influence input parameters the input values
xx.sub.1, xx.sub.2 . . . xx.sub.k-1, and xx.sub.k and possible to
make the values of the large influence input parameters and the
values of the auxiliary input parameters in addition to the values
of the essential input parameters the input values xx.sub.1,
xx.sub.2 . . . xx.sub.k-1, xx.sub.k. Note that, below, this
embodiment will be explained using as an example the case of making
the values of the large influence input parameters and the values
of the auxiliary input parameters in addition to the values of the
essential input parameters the input values xx.sub.1, xx.sub.2 . .
. xx.sub.k-1, xx.sub.k (in this example, k=4).
[0159] In this embodiment as well, first, the input values x.sub.1,
x.sub.2 . . . x.sub.n-1, and x.sub.n, the input values xx.sub.1,
xx.sub.2 . . . xx.sub.k-1, and xx.sub.k, and the training data,
that is, the truth labels yt, are used to prepare the training data
set shown in FIG. 16. In this embodiment as well, in FIG. 16, the
input values x.sub.1, x.sub.2 . . . x.sub.n-1, and x.sub.n show the
system internal pressures for each fixed time .DELTA.t detected by
the pressure sensor 47, and the input values xx.sub.1, xx.sub.2 . .
. xx.sub.k'11, and xx.sub.k, that is, the input values xx.sub.1,
xx.sub.2, xx.sub.3, and xx.sub.4 show the atmospheric pressure, the
remaining amount of fuel in the fuel tank 5, the temperature of the
fuel in the fuel tank 5, and the characteristic value of flow rate
of the suction pump 40.
[0160] On the other hand, in FIG. 16, yt.sub.1 . . . yt.sub.S (in
this example, s=9) show the training data for the output values
y.sub.1', y.sub.2', y.sub.3', y.sub.4', y.sub.5', y.sub.6',
y.sub.7', y.sub.8', and y.sub.9' and the output values y.sub.1,
y.sub.2, y.sub.3, y.sub.4, y.sub.5, y.sub.6, y.sub.7, y.sub.8, and
y.sub.9 shown in FIG. 34, that is, the truth labels. That is, in
FIG. 16, yt.sub.1 shows the truth label when a perforation
abnormality occurs in which the walls of the fuel vapor flow pipe
12 or 13 are perforated by a small hole, yt.sub.2 shows the truth
label when a valve opening abnormality occurs in which the purge
control valve 14 continues opened, yt.sub.3 shows the truth label
when a valve closing abnormality occurs in which the purge control
valve 14 continues closed, yt.sub.4 shows the truth label when an
abnormality occurs in the pressure sensor 47, yt.sub.5 shows the
truth label when an abnormality occurs in which the flow path
switching valve 42 sticks at the normal position, yt.sub.6 shows
the truth label when an abnormality occurs in which the flow path
switching valve 42 sticks at the test position, yt.sub.7 shows the
truth label when an abnormality occurs in which the suction pump 40
continues operating, yt.sub.8 shows the truth label when an
abnormality occurs in which the suction pump 40 continues to stop
operating, and yt.sub.9 shows a truth label at the normal time.
[0161] In this case, for example, when a perforation abnormality
occurs in which the walls of the vapor flow pipe 12 or 13 are
perforated by a small hole, only the truth label yt1 is made 1,
while the remaining truth labels yt.sub.2, yt.sub.3, yt.sub.4,
yt.sub.5, yt.sub.6, yt.sub.7, yt.sub.8, and yt.sub.9 are all made
zero. Similarly, when a valve opening abnormality occurs in which
the purge control valve 14 continues opened, only the truth label
yt.sub.2 is made 1, while the remaining truth labels yt.sub.1,
yt.sub.3, yt.sub.4, yt.sub.5, yt.sub.6, yt.sub.7, yt.sub.8, and
yt.sub.9 are all made zero. When a valve closing abnormality occurs
in which the purge control valve 14 continues closed, only the
truth label yt.sub.3 is made 1, while the remaining truth labels
yt.sub.1, yt.sub.2, yt.sub.4, yt.sub.5, yt.sub.6, yt.sub.7,
yt.sub.8, and yt.sub.9 are all made zero. When an abnormality
occurs in the pressure sensor 47, only the truth label yt.sub.4 is
made 1, while the remaining truth labels yt.sub.1, yt.sub.2,
yt.sub.3, yt.sub.5, yt.sub.6, yt.sub.7, yt.sub.8, and yt.sub.9 are
all made zero. When an abnormality occurs in which the flow path
switching valve 42 sticks at the normal position, only the truth
label yt.sub.5 is made 1, while the remaining truth labels
yt.sub.2, yt.sub.3, yt.sub.4, yt.sub.6, yt.sub.7, yt.sub.8, and
yt.sub.9 are all made zero. When an abnormality occurs in which the
flow path switching valve 42 sticks at the test position, only the
truth label yt.sub.6 is made 1, while the remaining truth labels
yt.sub.1, yt.sub.2, yt.sub.3, yt.sub.4, yt.sub.5, yt.sub.7,
yt.sub.8, and yt.sub.9 are all made zero. When an abnormality
occurred in which the suction pump 40 continues operating, only the
truth label yt.sub.7 is made 1, while the remaining truth labels
yt.sub.1, yt.sub.2, yt.sub.3, yt.sub.4, yt.sub.5, yt.sub.6,
yt.sub.8, and yt.sub.9 are all made zero. When an abnormality
occurs in which the suction pump 40 continues to stop operating,
only the truth label yt.sub.8 is made 1, while the remaining truth
labels yt.sub.1, yt.sub.2, yt.sub.3, yt.sub.4, yt.sub.5, yt.sub.6,
yt.sub.7, and yt.sub.9 are all made zero. When at the normal time,
only the truth label yt.sub.9 is made 1, while the remaining truth
labels yt.sub.1, yt.sub.2, yt.sub.3, yt.sub.4, yt.sub.5, yt.sub.6,
yt.sub.7, and yt.sub.8 are all made zero.
[0162] On the other hand, as shown in FIG. 16, in this training
data set, "m" number of data showing the relationship with the
input values x.sub.1, x.sub.2 . . . x.sub.n-1, and x.sub.n the
input values xx.sub.1, xx.sub.2 . . . xx.sub.k-1, and xx.sub.k and
the truth labels yt is acquired. For example, the No. 2 data lists
the acquired input values x.sub.12, x.sub.22 . . . x.sub.n-12, and
x.sub.n2, input values xx.sub.12, xx.sub.22 . . . xx.sub.k-12, and
xx.sub.k2, and truth labels yt.sub.12 . . . yt.sub.S2, the No. m-1
data lists the acquired input values x.sub.1m-1, x.sub.2m-1 . . .
x.sub.n-1m-1, and x.sub.nm-1, input values xx.sub.1m-1, xx.sub.2m-1
. . . xx.sub.k-1m-1, and xx.sub.km-1, and truth labels yt.sub.Sm-1
of the input parameters.
[0163] This training data set is also prepared by a method similar
to the method already explained with reference to FIG. 17. That is,
in this case as well, the test control device 81 shown in FIG. 17
is used to estimate whether an abnormality occurs in the fuel vapor
escape prevention system by performing processing for detection of
an abnormality in accordance with a predetermined order of
operation of the purge control valve 14, flow path switching valve
42, and suction pump 40. At this time, the state of the fuel vapor
escape prevention system is successively changed to a state where a
perforation abnormality occurs in which the walls of the fuel vapor
flow pipe 12 or 13 are perforated by a small hole, a state where a
valve opening abnormality occurred in which the purge control valve
14 continues opened, a state where a valve closing abnormality
occurs in which the purge control valve 14 continues closed, a
state where an abnormality occurred in the pressure sensor 47, a
state where an abnormality occurs in which the flow path switching
valve 42 sticks at the normal position, a state where an
abnormality occurs in which the flow path switching valve 42 sticks
at the test position, a state where an abnormality occurs in which
the suction pump 40 continues operating, a state where an
abnormality occurs in which the suction pump 40 continues to stop
operating, and a notional state. In the changed states, the
combination of atmospheric pressure, the remaining amount of fuel
in the fuel tank 5, the temperature of the fuel in the fuel tank 5,
and the characteristic value of flow rate of the suction pump 40 is
successively changed while processing for detection of an
abnormality is repeatedly performed.
[0164] While this processing for detection of an abnormality is
being performed, the data required for preparing the training data
set is acquired. FIG. 35 shows the routine for processing for
detection of an abnormality performed in the test control device 81
for performing this processing for detection of an abnormality.
This routine for processing for detection of an abnormality routine
is performed by interruption at every fixed time .DELTA.t shown in
FIG. 24. Note that, in the routine shown in FIG. 35, "t" shows the
point of time found when defining the time t.sub.0 when the engine
stops operating at FIG. 24 as zero and using this time t.sub.0 as
the starting point.
[0165] Referring to FIG. 35, first, at step 500, the system
internal pressures x.sub.n detected by the pressure sensor 47 are
acquired and stored. That is, as shown in FIG. 24, the system
internal pressures x. are acquired at every fixed time .DELTA.t.
The system internal pressures x.sub.n acquired at every fixed time
.DELTA.t are stored in the test control device 81. Next, at step
501, it is judged if the time "t" is the time t.sub.1 shown in FIG.
24. When the time "t" is not the time t.sub.1 shown in FIG. 24, the
routine proceeds to step 502 where it is judged if the time "t" is
the time t.sub.2 shown in FIG. 24. When the time "t" is not the
time t.sub.2 shown in FIG. 24, the routine proceeds to step 503
where it is judged if the time "t" is the time t.sub.4 shown in
FIG. 24. When the time "t" is not the time t.sub.4 shown in FIG.
24, the routine proceeds to step 504 where it is judged if the time
"t" is the time t.sub.5 shown in FIG. 24. When the time "t" is not
the time is shown in FIG. 24, the processing cycle is ended.
[0166] On the other hand, when at step 501 it is judged that the
time "t" is the time t.sub.1 shown in FIG. 24, the routine proceeds
to step 505 where an instruction for operating the suction pump 40
is issued. Next, the processing cycle is ended. Further, when at
step 502 it is judged that the time "t" is the time t.sub.2 shown
in FIG. 24, the routine proceeds to step 506 where a switching
instruction switching the flow path switching valve 42 to the test
position is issued. Next, the processing cycle is ended. Further,
when at step 503 it is judged that the time "t" is the time t.sub.4
shown in FIG. 24, the routine proceeds to step 507 where a valve
opening instruction making the purge control valve 14 open is
issued. Next, the processing cycle is ended.
[0167] On the other hand, when, at step 504, it is judged that the
time "t" is the time is shown in FIG. 24, the routine proceeds to
step 508 where a valve closing instruction causing the purge
control valve 14 to close is issued. Next, at step 509, a switching
instruction switching the flow path switching valve 42 to the
normal position is issued. Next, at step 510, an instruction for
stopping the suction pump 40 is issued. Next, at step 511, the
action of acquiring and storing the system internal pressure
x.sub.n detected by the pressure sensor 47 is stopped. Next, at
step 512, the processing for detection of an abnormality is ended.
At the test control device 81 shown in FIG. 17, if the processing
for detection of an abnormality ends, the next processing for
detection of an abnormality is started.
[0168] In this way, the system internal pressure x.sub.n for each
fixed time .DELTA.t when the combination of the atmospheric
pressure, the remaining amount of fuel in the fuel tank 5, the
temperature of the fuel in the fuel tank 5, and the characteristic
value of flow rate of the suction pump 40 is changed in each state
of a state where a perforation abnormality occurs in which the
walls of the fuel vapor flow pipe 12 or 13 are perforated by a
small hole, a state where a valve opening abnormality occurs in
which the purge control valve 14 continues opened, a state where a
valve closing abnormality occurs in which the purge control valve
14 continues closed, a state where an abnormality occurs in the
pressure sensor 47, a state where an abnormality of the flow path
switching valve 42 sticking at the normal position occurs, a state
where an abnormality occurs in which the flow path switching valve
42 sticks at the test position, a state where an abnormality occurs
in which the suction pump 40 continues operating, a state where an
abnormality occurs in which the suction pump 40 continues to stop
operating, and a normal state are stored in the test control device
81. That is, the No. 1 to No. "m" input values x.sub.1m, x.sub.2m .
. . x.sub.nm-1, and x.sub.nm, the input values xx.sub.1, xx.sub.2m
. . . xx.sub.km-1, and xx.sub.km, and the truth label
yt.sub.sm(m=1, 2, 3 . . . m) of the training data set shown in FIG.
16 are stored inside the test control device 81.
[0169] If a training data set is prepared in this way, electronic
data of the prepared training data set is used to learn the weights
of the neural network 72 shown in FIG. 33. In this case, the
weights of the neural network 72 shown in FIG. 33 are also learned
by the learning apparatus 82 shown in FIG. 17 using the routine for
processing for learning weights of the neural network shown in FIG.
19. If the weights of the neural network 72 finish being learned,
the learned weights of the neural network 72 are stored in the
memory 84 of the learning apparatus 82. In this way, an abnormality
judgment estimation model able to estimate whether an abnormality
occurs in the fuel vapor escape prevention system is prepared.
[0170] In this embodiment as well, the thus prepared abnormality
judgment estimation model of the fuel vapor escape prevention
system is used to diagnose a fault of the fuel vapor escape
prevention system in a commercially available vehicle. To this end,
this abnormality judgment estimation model of the fuel vapor escape
prevention system is stored in the electronic control unit 20 of
the commercially available vehicle using a routine for reading data
into the electronic control unit shown in FIG. 20. That is,
referring to FIG. 20, first, at step 300, the number of nodes of
the input layer (L=1) of the neural network 72 shown in FIG. 33,
the numbers of nodes of the hidden layer (L=2) and hidden layer
(L=3), and the number of nodes of the output layer (L=4) are read
into the memory 22 of the electronic control unit 20. Next, at step
301, the neural network 73 such as shown in FIG. 36 is prepared
based on these numbers of nodes. As will be understood from FIG.
36, in this neural network 73, the softmax layer is removed. Note
that, in this case, the neural network 73 may also be provided with
the softmax layer 71 such as shown in FIG. 33. Next, at step 302,
the learned weights of the neural network 72 are read into the
memory 22 of the electronic control unit 20. Due to this, the
abnormality judgment estimation model of the fuel vapor escape
prevention system is stored in the electronic control unit 20 of
the commercially available vehicle.
[0171] In this embodiment as well, as the routine for detection of
an abnormality of the fuel vapor escape prevention system performed
at a commercially available vehicle, the routine shown in FIG. 22
is used. In this embodiment, even when using the routine shown in
FIG. 22, step 400 to step 406 are no different from the content
explained above, so the explanations of step 400 to step 406 will
be omitted. On the other hand, regarding step 407 and on, this
differs somewhat from the content explained before, so only step
407 and on will be explained in brief.
[0172] That is, at step 407, system internal pressures
x.sub.nx.sub.1, x.sub.2 . . . x.sub.n-1, and x. for each fixed time
.DELTA.t and input values xx.sub.1, xx.sub.2 . . . xx.sub.k-1, and
xx.sub.k are input to the nodes of the input layer (L=1) of the
neural network 73 shown in FIG. 36. At this time, output values
y.sub.1', y.sub.2', y.sub.3', y.sub.4', y.sub.5', y.sub.6',
y.sub.7', y.sub.8', and y.sub.9' are output from the nodes of the
output layer of the neural network 70. Due to this, at step 408,
the output values y.sub.1', y.sub.2', y.sub.3', y.sub.4', y.sub.5',
y.sub.6', y.sub.7', y.sub.8', and y.sub.9' are acquired.
[0173] Next, at step 409, the largest output value y.sub.i' is
selected from among the acquired output values y.sub.1', y.sub.2',
y.sub.3', y.sub.4', y.sub.5', y.sub.6', y.sub.7', y.sub.8', and
y.sub.9'. At this time, it is estimated that the abnormal state
shown in FIG. 34 corresponding to this largest output value
y.sub.i' occurs. Therefore, at step 410, it is judged that the
abnormal state shown in FIG. 34 corresponding to this largest
output value y.sub.1' occurs and, for example, a warning light
showing the abnormal state shown in FIG. 34 corresponding to the
largest output value y.sub.i' occurs is turned on. Next, at step
411, the detection of an abnormality is ended.
[0174] In this way, in an abnormality detection device of a fuel
vapor escape prevention system according to another embodiment of
the present invention, the fuel vapor escape prevention system is
provided with the canister 6 at which the fuel vapor chamber 10 and
the atmospheric pressure chamber 11 are formed at the two sides of
the activated carbon layer 9. The fuel vapor chamber 10 is on the
one hand communicated with the inside space above the fuel level of
the fuel tank 5 and is on the other hand communicated through the
purge control valve 14 with the intake passage of the engine.
Furthermore, the fuel vapor escape prevention system is provided
with the flow path switching valve 42 able to selectively connect
the atmospheric pressure chamber 11 with the atmosphere and suction
pump 40. The passage 43 from the flow path switching valve 42
toward the atmospheric pressure chamber 11 and the suction passage
46 from the flow path switching valve 42 toward the suction pump 40
are connected by the reference pressure detection passage 50 having
the restricted opening 51. Inside the suction passage 46 from the
flow path switching valve 42 toward the suction pump 40, the
pressure sensor 47 is arranged. At the time of stopping operation
of the vehicle, processing for detection of an abnormality is
performed generating a valve closing instruction causing the purge
control valve 14 to close, a pump operation instruction making the
suction pump 40 operate to make the inside of the fuel tank 5 and
inside of the canister 6 a negative pressure while maintaining the
switched position of the flow path switching valve 42 at a switched
position where the atmospheric pressure chamber 11 is connected to
the atmosphere when a predetermined time elapses after stopping
operation of the vehicle, a switching instruction switching the
switched position of the flow path switching valve 42 after
generation of the pump operation instruction to a switched position
at which the atmospheric pressure chamber 11 is connected to the
suction pump 40, and a valve opening instruction making the purge
control valve 14 open after the generation of the switching
instruction. At the time the processing for detection of an
abnormality is performed, the pressure at the inside of the fuel
tank 5 and inside of the canister 6 detected by the pressure sensor
47 at every fixed time are stored in the storage device, a learned
neural network learned in weights using the pressures at the inside
of the fuel tank 5 and inside of the canister 6 at every fixed time
stored in the storage device and at least the atmospheric pressure
when the processing for detection of an abnormality is performed as
input parameters of the neural network and using a case where
perforation occurs in the system causing leakage of fuel vapor as a
truth label is stored, and, at the time of stopping operation of
the vehicle, a perforation abnormality causing fuel vapor to leak
is detected from the input parameters by using the learned neural
network.
[0175] In this case, in this embodiment according to the present
invention, a learned neural network learned in weights using the
pressures at the inside of the fuel tank 5 and inside of the
canister 6 at every fixed time stored in the storage device and at
least the atmospheric pressure when the processing for detection of
an abnormality is performed as input parameters of the neural
network and using a case where perforation occurs in the fuel vapor
escape prevention system causing leakage of fuel vapor, a case
where a valve opening abnormality occurs in which the purge control
valve 14 continues opened, a case where a valve closing abnormality
occurs in which the purge control valve 14 continues closed, a case
where an abnormality occurs in the pressure sensor 47, a case where
a switching abnormality occurs in which the switched position of
the flow path switching valve 42 is maintained at a switched
position connecting the atmospheric pressure chamber 11 to the
atmosphere, a case where a switching abnormality occurs in which
the switched position of the flow path switching valve 42 is
maintained at a switched position connecting the atmospheric
pressure chamber 11 to the suction pump 40, a case where an
abnormality occurs in which the suction pump 40 continues
operating, and a case where an abnormality occurs in which the
suction pump 40 continues stopped as truth labels is stored, and,
at the time of stopping operation of the vehicle, a perforation
abnormality causing fuel vapor to leak, a valve opening abnormality
of the purge control valve, a valve closing abnormality of the
purge control valve, an abnormality of the pressure sensor, a
switching abnormality of the flow path switching valve, and an
abnormality of the suction pump are detected from the input
parameters by using the learned neural network.
[0176] Furthermore, in this case, in this embodiment according to
the present invention, the above-mentioned input parameters are
comprised of the pressures at the inside of the fuel tank 5 and
inside of the canister 6 at every fixed time stored in the storage
device, the atmospheric pressure when processing for detection of
an abnormality is performed, and the remaining amount of fuel in
the fuel tank 5 when processing for detection of an abnormality is
performed. Further, in this case, in this embodiment according to
the present invention, the above-mentioned input parameters are
comprised of the pressures at the inside of the fuel tank 5 and
inside of the canister 6 at every fixed time stored in the storage
device, the atmospheric pressure when processing for detection of
an abnormality is performed, the remaining amount of fuel in the
fuel tank 5 when processing for detection of an abnormality is
performed, the temperature of the fuel in the fuel tank 5, and a
parameter showing the capacity of the suction pump 40.
* * * * *