JP2020176625A - Abnormality detection device of fuel vapor escape prevention system - Google Patents

Abstract

To accurately detect an abnormality of a fuel vapor escape prevention system.SOLUTION: At the time of stopping operation of a vehicle, pressures inside a fuel tank (5) and inside a canister (6) that are detected at constant time intervals are stored in a storage device. A learned neural network is stored, which is learned in weights, using the pressures inside the fuel tank (5) and inside the canister (6) at constant time intervals 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. 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.SELECTED DRAWING: Figure 22

Description

The present invention relates to an abnormality detection device for a fuel vapor emission prevention system.

In an internal combustion engine, in order to prevent fuel vapor from flowing out into the outside air, a canister in which a fuel vapor chamber and an atmospheric pressure chamber are formed on both sides of the activated coal layer is provided, and the fuel vapor chamber is used as fuel on one side. Conventionally, a fuel vapor emission prevention system that communicates with the internal space above the fuel liquid level of the tank and is connected to the intake passage of the engine via a purge control valve has been used. In such a fuel vapor emission prevention system, for example, when a hole is made in the pipe wall of the fuel vapor flow pipe connecting the fuel vapor chamber of the canister and the purge control valve, the fuel vapor is introduced into the outside air through this hole. It leaks to.

Therefore, the pressure in the internal space above the fuel liquid level in the fuel tank (simply called the pressure in the fuel tank) is detected, and is there an abnormality in the fuel vapor emission prevention system due to the change in the pressure in the fuel tank? For example, a diagnostic device that diagnoses whether or not a hole has been formed in the pipe wall of a fuel steam flow pipe is known (see, for example, Patent Document 1). With this diagnostic device, the pressure inside the fuel tank drops below atmospheric pressure by opening the purge control valve while the atmospheric pressure chamber of the canister is shut off from the atmosphere during stable steady operation. Then, the inside of the fuel tank is sealed by closing the purge control valve. At this time, for example, if there is a hole in the pipe wall of the fuel vapor flow pipe, the pressure in the fuel tank gradually increases. Therefore, in this diagnostic device, when the pressure inside the fuel tank rises above a certain value after the inside of the fuel tank is sealed, it is determined that, for example, there is a hole in the pipe wall of the fuel vapor flow pipe. To.

Japanese Unexamined Patent Publication No. 2004-44396

In this case, if the diameter of the hole formed in the pipe wall of the fuel vapor flow pipe is small, the amount of pressure increase in the fuel tank becomes small. On the other hand, the pressure in the fuel tank also fluctuates due to other factors, for example, the temperature of the fuel in the fuel tank. Therefore, since the pressure in the fuel tank becomes higher than a certain value, there is a risk of making a false judgment if it is determined that a hole has been formed in the pipe wall of the fuel vapor flow pipe.
In the present invention, for example, a hole is made in the pipe wall of the fuel vapor flow pipe by using a neural network, and even if the diameter of the hole made at this time is small, the hole is accurately opened in the pipe wall of the fuel vapor flow pipe. It is an object of the present invention to provide an abnormality detection device of a fuel vapor emission prevention system capable of detecting such a situation.

That is, according to the present invention, the canisters in which the fuel steam chamber and the atmospheric pressure chamber are formed on both sides of the activated coal layer are provided, and the fuel steam chamber is on the one hand in the internal space above the fuel liquid level of the fuel tank. A flow path switching valve that communicates and, on the other hand, is connected to the intake passage of the engine via a purge control valve and can selectively connect the atmospheric pressure chamber to the atmosphere and the suction pump, and in the fuel tank and In the abnormality detection device of the fuel vapor discharge prevention system equipped with a pressure sensor that detects the pressure in the canister, the valve closing command for closing the purge control valve and the switching position of the flow path switching valve are set when the vehicle is stopped. , Anomaly detection processing is performed to generate a switching command to switch the atmospheric pressure chamber to the switching position connected to the suction pump and a pump operation command to operate the suction pump to make the fuel tank and the canister negative pressure. , When the abnormality detection process is being performed, the pressure in the fuel tank and the canister detected by the pressure sensor at regular intervals is stored in the storage device, and in the fuel tank at regular intervals stored in the storage device. Weight learning using at least atmospheric pressure as the input parameter of the neural network when the pressure in the canister and the abnormality detection process are performed, and the correct label when there is a hole in the above system to leak fuel vapor. The trained neural network in which the fuel was performed is stored, and when the vehicle is stopped, the trained neural network is used to detect a perforated abnormality that leaks fuel vapor from the above input parameters. Abnormality detection device is provided.

Further, according to the present invention, a canister in which a fuel steam chamber and an atmospheric pressure chamber are formed on both sides of the activated coal layer is provided, and the fuel steam chamber is, on the one hand, in the internal space above the fuel liquid level of the fuel tank. On the other hand, it is connected to the intake passage of the engine via a purge control valve, and is equipped with a flow path switching valve that can selectively connect the atmospheric pressure chamber to the atmosphere and the suction pump. The passage from the flow path switching valve to the atmospheric pressure chamber and the suction passage from the flow path switching valve to the suction pump are connected by a reference pressure detection passage having a throttle portion, and suction from the flow path switching valve to the suction pump. In the abnormality detection device of the fuel vapor discharge prevention system in which a pressure sensor is arranged in the passage, a valve closing command for closing the purge control valve when the vehicle is stopped and switching of the flow path switching valve after the vehicle is stopped. A pump operation command that activates the suction pump to create negative pressure in the fuel tank and canister when a preset time elapses while maintaining the position in a switching position where the atmospheric pressure chamber is connected to the atmosphere. After the pump operation command is issued, the switching position of the flow path switching valve is switched to the switching position where the atmospheric pressure chamber is connected to the suction pump, and after the switching command is issued, the purge control valve is opened. When the abnormality detection process for generating a command is performed and the abnormality detection process is being performed, the pressure in the fuel tank and the canister detected by the pressure sensor at regular intervals is stored in the storage device and stored in the storage device. The stored pressure in the fuel tank and in the canister at regular intervals and at least atmospheric pressure when the abnormality detection process is performed are used as the input parameters of the neural network, and a hole is created in the above system to leak fuel vapor. The trained neural network in which the weights have been trained is stored with the correct answer label as the answer label, and when the vehicle is stopped, the trained neural network is used to leak fuel vapor from the above input parameters. An abnormality detection device for a fuel vapor emission prevention system that detects a space abnormality is provided.

By learning the weight of the neural network using the pressure in the fuel tank and the canister and at least the atmospheric pressure detected by the pressure sensor at regular intervals as the input parameters of the neural network, for example, the pipe wall of the fuel vapor flow pipe. In addition, even if a hole with a small diameter is opened, it is possible to accurately detect that the hole is formed in the pipe wall of the fuel vapor flow pipe.

FIG. 1 is an overall view of the fuel vapor emission prevention system. 2A and 2B are enlarged views illustrating the pump module shown in FIG. FIG. 3 is a perspective view of the gravity sensor. FIG. 4 is a diagram showing the remaining amount of fuel in the fuel tank. FIG. 5 is a diagram showing changes in the internal pressure of the system. FIG. 6 is a diagram showing an example of a neural network. FIG. 7 is a diagram showing the abnormality detection process and the change in the system internal pressure. FIG. 8 is a diagram showing the abnormality detection process and the change in the system internal pressure. FIG. 9 is a diagram showing the abnormality detection process and the change in the system internal pressure. FIG. 10 is a diagram showing the abnormality detection process and the change in the system internal pressure. FIG. 11 is a diagram showing the abnormality detection process and the change in the system internal pressure. FIG. 12 is a diagram showing a neural network used in an embodiment according to the present invention. FIG. 13 is a diagram showing a list of input parameters. FIG. 14 is a diagram showing an index of the performance of the suction pump. FIG. 15 is a diagram showing a list of output values. FIG. 16 is a diagram showing a training data set. FIG. 17 is a diagram for explaining a learning method. FIG. 18 is a flowchart for executing the abnormality detection process. FIG. 19 is a flowchart for executing the learning process. FIG. 20 is a flowchart for reading data into the electronic control unit. FIG. 21 is a diagram showing an example of a neural network. FIG. 22 is a flowchart for detecting an abnormality. 23A and 23B are enlarged views illustrating a modified example of the pump module shown in FIG. 1. FIG. 24 is a diagram showing the abnormality detection process and the change in the system internal pressure. FIG. 25 is a diagram showing the abnormality detection process and the change in the system internal pressure. FIG. 26 is a diagram showing the abnormality detection process and the change in the system internal pressure. FIG. 27 is a diagram showing the abnormality detection process and the change in the system internal pressure. FIG. 28 is a diagram showing the abnormality detection process and the change in the system internal pressure. FIG. 29 is a diagram showing the abnormality detection process and the change in the system internal pressure. FIG. 30 is a diagram showing the abnormality detection process and the change in the system internal pressure. FIG. 31 is a diagram showing the abnormality detection process and the change in the system internal pressure. FIG. 32 is a diagram showing the abnormality detection process and the change in the system internal pressure. FIG. 33 is a diagram showing a neural network used in another embodiment according to the present invention. FIG. 34 is a diagram showing a list of output values. FIG. 35 is a flowchart for executing the abnormality detection process. FIG. 36 is a diagram showing a neural network used in another embodiment according to the present invention.

<Overall configuration of internal combustion engine>
FIG. 1 shows an overall view of the fuel vapor emission prevention system. Referring to FIG. 1, 1 is an engine body, 2 is a surge tank, 3 is an intake duct, 4 is a throttle valve, 5 is a fuel tank, and 6 is a canister. The fuel tank 5 is provided with a fuel level gauge 7 for detecting the height of the fuel liquid level in the fuel tank 5 and a temperature sensor 8 for detecting the temperature of the fuel in the fuel tank. On the other hand, the canister 6 has an activated carbon layer 9, a fuel vapor chamber 10 and an atmospheric pressure chamber 11 arranged on both sides of the activated carbon layer 9, respectively. In FIG. 1, the fuel vapor chamber 10 is formed of the first fuel vapor chamber 10a and the second fuel vapor chamber 10b, and the fuel vapor chamber 10 is composed of one common fuel vapor chamber. You can also.

As shown in FIG. 1, the first fuel steam chamber 10a communicates with the internal space above the fuel liquid level of the fuel tank 5 via the fuel steam flow pipe 12, and the second fuel steam chamber 10b It is connected to the surge tank 2 via the fuel steam flow pipe 13, that is, into the intake passage. A purge control valve 14 is arranged in the fuel vapor flow pipe 13. On the other hand, the atmospheric pressure chamber 11 is connected to the atmospheric communication pipe 17 via a small volume activated carbon layer 15 and a suction pump module 16.

In FIG. 1, 20 shows an electronic control unit for controlling the operation of the engine and the fuel vapor emission prevention system. As shown in FIG. 1, the electronic control unit 20 comprises a digital computer and is connected to each other by a bidirectional bus 21, that is, a memory 22, a CPU (microprocessor) 23, an input port 24, and a storage device 22. It includes an output port 25. The output signal of the fuel level gauge 7, the output signal of the temperature sensor 8, the output signal of the atmospheric pressure sensor 30 for detecting the atmospheric pressure, and the output signal of the operation start / stop switch 31 correspond to the input port 24, respectively. It is input via the AD converter 26.

In a hybrid engine equipped with an electric motor as a drive source, the operation of the vehicle by the engine or the electric motor is started when the operation start / stop switch 31 is turned on, and the engine or the electric motor is turned off when the operation start / stop switch 31 is turned off. The operation of the vehicle is stopped. On the other hand, in an engine that does not have an electric motor as a drive source, the engine is started when the operation start / stop switch 31 is turned on to start the operation of the vehicle, and the engine is started when the operation start / stop switch 31 is turned off. Is stopped and the operation of the vehicle is stopped.

Further, as shown in FIG. 1, a load sensor 33 that generates an output voltage proportional to the amount of depression of the accelerator pedal 32 is connected to the accelerator pedal 32, and the output voltage of the load sensor 33 is the corresponding AD converter 26. It is input to the input port 24 via. Further, a crank angle sensor 34 that generates an output pulse every time the crankshaft rotates, for example, 30 ° is connected to the input port 24. 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 to the actuators of the purge control valve 14, the suction pump module 16 and the throttle valve 4 via the corresponding drive circuit 27.

2A and 2B show enlarged views illustrating the suction pump module 16 shown in FIG. Referring to FIGS. 2A and 2B, the suction pump module 16 includes a suction pump 40 and a flow path switching valve 42 driven by an actuator 41. Further, the suction pump module 16 has an atmospheric communication pipe connecting passage 43 connected to the atmospheric chamber 11 via a small volume activated coal layer 15 and an atmospheric communication pipe connecting passage 43 connected to the atmosphere via an atmospheric communication pipe 17. It has an atmospheric communication passage 45 extending from the atmospheric communication pipe connecting path 44 toward the flow path switching valve 42, and a suction passage 46 extending from the suction pump 40 toward the flow path switching valve 42, and is inside the suction passage 46. A pressure sensor 47 is attached to the.

Inside the flow path switching valve 42, there is a first passage 48 capable of connecting the atmospheric pressure chamber connecting passage 43 and the atmospheric communication passage 45 as shown in FIG. 2A, and an atmospheric pressure chamber connecting passage as shown in FIG. 2B. A second passage 49 that can connect the 43 and the suction passage 46 is formed. The flow path switching valve 42 is usually held at a position where the atmospheric pressure chamber connecting path 43 is connected to the atmospheric communication passage 45 via the first passage 48, as shown in FIG. 2A, and at this time, the atmospheric pressure is maintained. The chamber 11 communicates with the atmosphere through a small-volume activated coal layer 15, an atmospheric chamber connecting passage 43, a first passage 48, an atmospheric communication passage 45, an atmospheric communication pipe connecting passage 44, and an atmospheric communication pipe 17. As shown in FIG. 2A, the switching position of the flow path switching valve 42 in which the atmospheric pressure chamber 11 communicates with the atmosphere is referred to as a normal position.

On the other hand, when anomaly detection of the fuel vapor discharge prevention system is performed, the flow path switching valve 42 is connected to the suction passage 46 via the second passage 49 by connecting the atmospheric pressure chamber connecting passage 4 as shown in FIG. 2B. Can be switched to position. As shown in FIG. 2B, the position where the atmospheric pressure chamber 11 is connected to the suction passage 46 is referred to as a test position. At this time, when the suction pump 40 is operated, the air in the atmospheric pressure chamber 11 is sucked through the small volume activated coal layer 15, the atmospheric pressure chamber connecting path 4, the second passage 49 and the suction passage 46, and at this time, When the purge control valve 14 is closed, the pressure in the fuel tank 5, the canister 6, the fuel steam flow pipe 12, and the fuel steam flow pipe 13 between the canister 6 and the purge control valve 14 decreases. Hereinafter, the pressure in the fuel tank 5, the canister 6, the fuel steam flow pipe 12, and the fuel steam flow pipe 13 between the canister 6 and the purge control valve 14 is simply applied to the fuel tank 5 and the canister 6. It is called the pressure of. Further, in the embodiment according to the present invention, the pressure detected by the pressure sensor 47 is referred to as the pressure in the fuel vapor emission prevention system, that is, the system internal pressure. Therefore, when the pressure in the fuel tank 5 and the canister 6 is detected by the pressure sensor 47, the pressure in the fuel tank 5 and the canister 6 becomes the system internal pressure.

Next, with reference to FIGS. 3 and 4, an accurate method of obtaining the remaining amount of fuel in the fuel tank 5 will be briefly described by taking the case of using a gravity sensor as an example. As shown in FIG. 3, the gravity sensor 60 is attached to a square frame 61, a cross-shaped thin plate 63 having four plate-shaped arms 62 supported by the frame 61, and a central portion of the cross-shaped thin plate 63. A strain gauge is attached to each arm 62 of the cross-shaped thin plate 63, which is composed of a weight 64. The gravity sensor 60 is attached to the vehicle so that the frame 61 is located in the horizontal plane when the vehicle is stopped on the horizontal plane.

Since gravity acts in the vertical direction on the weight 64, a downward force in the vertical direction acts on the center of the cross-shaped thin plate 63. At this time, if the frame 61 is located in the horizontal plane, the strain amount of each arm 62 is the same. Therefore, when the vehicle is stopped on the horizontal plane, the strain amount of each arm 62 is the same. On the other hand, when the vehicle is tilted with respect to the horizontal plane and stopped, the strain amount of each arm 62 becomes a different value, and the tilt direction and tilt amount of the vehicle with respect to the horizontal plane can be known from the difference in the strain amount of each arm 62. .. On the other hand, the height of the fuel liquid level in the fuel tank 5 when the vehicle is located in the horizontal plane can be known from the detected value of the fuel level gauge 7.

Therefore, regardless of the shape of the fuel tank 5, the amount of fuel remaining in the fuel tank 5 is determined from the detected value of the fuel level gauge 7 and the tilt direction and tilt amount of the vehicle with respect to the horizontal plane detected by the gravity sensor 60. You will know the amount. Therefore, in this example, as shown in FIG. 4, the tilt amount F in the front-rear direction of the vehicle detected by the gravity sensor 60, the tilt amount S in the lateral direction of the vehicle S, and the detected value H of the fuel level gauge 7 are used. , The relationship with the remaining amount M of fuel in the fuel tank 5 is obtained in advance by an experiment, and the remaining amount M of fuel in the fuel tank 5 is obtained based on the relationship shown in FIG.

Next, the basic concept of the present invention will be described. As can be seen from FIGS. 1 and 2B, when the suction pump 40 is operated with the purge control valve 14 closed and the flow path switching valve 42 switched to the test position, the fuel in the fuel tank 5, the canister 6, and the fuel are operated. The pressure in the steam flow pipe 12 and in the fuel steam flow pipe 13 between the canister 6 and the purge control valve 14, that is, the pressure inside the system is reduced. The change in the system internal pressure at this time is shown in FIG. In FIG. 5, point A indicates when the suction action by the suction pump 40 is started, and point B indicates when the pressure drop no longer occurs.

By the way, in FIG. 5, the solid line shows the case where the fuel vapor emission prevention system is normal, and the broken line shows the case where a hole is made in the wall of the fuel vapor flow pipe 12 or between the canister 6 and the purge control valve 14. The case where a hole is made in the pipe wall of the fuel vapor flow pipe 13 of the above is shown. When a hole is made in the pipe wall of the fuel vapor flow pipe 12 or 13, the outside air flows in from the hole, so that the pressure at the point B becomes higher than the normal time shown by the solid line as shown by the broken line. Therefore, it is possible to determine whether or not a hole has been formed in the pipe wall of the fuel vapor flow pipe 12 or 13 from the magnitude of the pressure at the point B.

On the other hand, in the one-point chain line of FIG. 5, there is no hole in the pipe wall of the fuel steam flow pipe 12 or 13, but when the amount of fuel evaporation per unit time in the fuel tank 5 increases for some reason, For example, the case where the temperature of the fuel in the fuel tank 5 is higher than the case shown by the solid line is shown. In this case as well, the pressure at point B is higher than the normal time shown by the solid line, as in the case where the pipe wall of the fuel vapor flow pipe 12 or 13 shown by the broken line is punctured. Therefore, even if the pressure at point B is higher than the normal time shown by the solid line, it is judged that a hole is formed in the pipe wall of the fuel vapor flow pipe 12 or 13, and a misjudgment occurs.

However, when the wall of the fuel vapor flow pipe 12 or 13 is punctured, that is, when it is indicated by a broken line, the pressure is in the middle of the pressure drop as compared with the case where it is indicated by the solid line and the alternate long and short dash line. The degree of descent of is small, and the overall shape of the descent curve is different between the case shown by the broken line and the case shown by the solid line and the alternate long and short dash line. Therefore, when the overall shape of the pressure drop curve is obtained, it is possible to accurately determine whether or not a hole has been formed in the pipe wall of the fuel vapor flow pipe 12 or 13 from the difference in the overall shape of the pressure drop curve. Therefore, in the present invention, the system internal pressure is detected at regular time intervals, and based on the detected system internal pressure at regular time intervals, a neural network is used to determine whether or not an abnormality has occurred in the fuel vapor emission prevention system. I have to.
<Overview of neural network>

As described above, in the embodiment according to the present invention, a neural network is used to determine whether or not an abnormality has occurred in the fuel vapor emission prevention system. Therefore, the neural network will be briefly described first. FIG. 6 shows a simple neural network. The circles in FIG. 6 represent artificial neurons, and in neural networks, these artificial neurons are usually referred to as nodes or units (referred to as nodes in the present application). In FIG. 6, L = 1 indicates an input layer, L = 2 and L = 3 indicate a hidden layer, and L = 4 indicates an output layer, respectively. Further, in FIG. 6, x ₁ and x ₂ indicate output values from each node of the input layer (L = 1), and y ₁ and y ₂ are outputs from each node of the output layer (L = 4). The values are shown, and z ⁽²⁾ _1, z ⁽²⁾ ₂ and z ⁽²⁾ ₃ show the output values from each node of the hidden layer (L = 2), and z ⁽³⁾ _1, z. ⁽³⁾ ₂ and z ⁽³⁾ ₃ indicate the output value from each node of the hidden layer (L = 3). The number of hidden layers can be one or any number, and the number of nodes in the input layer and the number of nodes in the hidden layer can also be any number. Further, the number of nodes in the output layer may be one or a plurality of nodes.

次いで、この総入力値ｕ_ｋは活性化関数ｆにより変換され、隠れ層 ( L＝２) のｚ^（２） _ｋで示されるノードから、出力値ｚ^（２） _ｋ(= f (ｕ_ｋ)) として出力される。一方、隠れ層 ( L＝３) の各ノード には、隠れ層 ( L＝２) の各ノードの出力値ｚ^（２） _１、ｚ^（２） _２ およびｚ^（２） _３が入力され、隠れ層 ( L＝3 ) の各ノードでは、夫々対応する重みｗおよびバイアスｂを用いて総入力値ｕ（Σｚ・ｗ＋ｂ）が算出される。この総入力値ｕは同様に活性化関数により変換され、隠れ層 ( L＝3 ) の各ノードから、出力値ｚ^（３） _１、ｚ^（３） _２ およびｚ^（３） _３として出力される、この活性化関数としては、例えば、シグモイド関数σが用いられる。 The input is output as it is at each node of the input layer. On the other hand, each node of the hidden layer (L = 2), the output value x ₁ and x ₂ of each node in the input layer is inputted, in each node of the hidden layer (L = 2), respectively corresponding weight w and The total input value u is calculated using the bias b. For example, the total input value _{u k} calculated in the node indicated by the hidden layer (L = 2) of _{^{z (2) k (k =}} 1,2,3) in FIG. 6, the following equation.

Then, the total input value _{u k} is converted by the activation function f, the hidden layer (L = 2) of ^{z ₍₂₎} from the node indicated by _k, the output value _{^{z (2) k (= f}} (u k) ) Is output. On the other hand, the output values z ⁽²⁾ _1, z ⁽²⁾ ₂ and z ⁽²⁾ ₃ of each node of the hidden layer (L = 2) are input to each node of the hidden layer (L = 3), and are hidden. At each node of the layer (L = 3), the total input value u (Σz · w + b) is calculated using the corresponding weights w and bias b, respectively. This total input value u is similarly converted by the activation function and output as output values z ⁽³⁾ _1, z ⁽³⁾ ₂ and z ⁽³⁾ ₃ from each node of the hidden layer (L = 3). , For example, a sigmoid function σ is used as this activation function.

On the other hand, the output values z ⁽³⁾ _1, z ⁽³⁾ ₂ and z ⁽³⁾ ₃ of each node of the hidden layer (L = 3) are input to each node of the output layer (L = 4) and output. At each node of the layer, the total input value u (Σz · w + b) is calculated using the corresponding weights w and the bias b, respectively, or the total input value u (Σz · w) is calculated using only the corresponding weights w. w) is calculated. For example, in the regression problem, an identity function is used at the node of the output layer, and therefore, the node of the output layer outputs the total input value u calculated at the node of the output layer as the output value y as it is.
<Learning in neural networks>

ここで、ｚ^{（Ｌ−１）}・∂ｗ^（Ｌ）＝ ∂ｕ^（Ｌ）であるので、（∂Ｅ/∂ｕ^（Ｌ））＝δ^（Ｌ）とすると、上記（１）式は、次式でもって表すことができる。

Now, assuming that the teacher data indicating the correct answer value of the output value y of the neural network, that is, the correct answer data is y _t , each weight w and the bias b in the neural network are the output value y and the teacher data, that is, the correct answer data y _t. It is learned using the error backpropagation method so that the difference between the two and is small. This backpropagation method is well known, and therefore the outline of the back propagation method will be briefly described below. Since the bias b is a kind of the weight w, it will be referred to as the weight w including the bias b below. By the way, in the neural network as shown in FIG. 6, when the weight at the input value u ^(L) to the node of each layer of L = 2, L = 3 or L = 4 is expressed by w ^(L) , the error function E The derivative by the weight w ^(L) , that is, the gradient ∂E / ∂w ^(L) can be rewritten by the following equation.

Here, since z ^(L-1) and ∂w ^(L) = ∂u ^(L) , if (∂E / ∂u ^(L) ) = δ ^(L) , the above equation (1) is It can be expressed by the following equation.

ここで、z^（Ｌ）＝ｆ(u^（Ｌ）) と表すと、上記（３）式の右辺に現れる入力値ｕ_k ^{（Ｌ＋１）}は、次式で表すことができる。

ここで、上記（３）式の右辺第１項（∂Ｅ/∂ｕ^{（Ｌ＋１）}）はδ^{（Ｌ＋１）}であり、上記（３）式の右辺第２項（∂u_ｋ ^{（Ｌ＋１）} /∂u^（Ｌ））は、次式で表すことができる。

従って、δ^（Ｌ）は、次式で示される。

即ち、δ^{（Ｌ＋１）}が求まると、δ^（Ｌ）を求めることができることになる。 Here, when u ^(L) fluctuates, the error function E fluctuates through a change in the total input value u ^{(L + 1)} of the next layer, so δ ^(L) can be expressed by the following equation.

Here, when expressed as ^{z (L) = f (u} (L)), the input values u _k appearing in the right side of the above (3) ^{(L + 1)} can be expressed by the following equation.

Here, the first term (∂E / ∂u ^{(L + 1)} ) on the right side of the above equation (3) is δ ^{(L + 1)} , and the second term (∂u _k ^{(L + 1)} / ∂ ^{) on} the right side of the above equation (3). u ^(L) ) can be expressed by the following equation.

Therefore, δ ^(L) is expressed by the following equation.

That is, when δ ^{(L + 1)} is obtained, δ ^(L) can be obtained.

この場合、回帰問題では、前述したように、ｆ(u^（Ｌ）) は恒等関数であり、ｆ’(u^（Ｌｌ）) ＝ １となる。従って、δ^（Ｌ）＝ｙ−ｙ_ｔ となり、δ^（Ｌ）が求まる。 Now, a node is one of the output layer (L = 4), the teacher data for a certain input value, i.e., has been required correct answer data y _t, the output value from the output layer to the input value In the case of y, when the squared error is used as the error function, the squared error E is obtained by E = 1/2 (y−y _t ) ² . In this case, the output value y = f (u ^(L) ) at the node of the output layer (L = 4), and therefore, in this case, the value of δ ^(L) at the node of the output layer (L = 4). Is expressed by the following equation.

In this case, in the regression problem, as described above, f (u ^(L) ) is an identity function, and f'(u ^(Ll) ) = 1. ^{Therefore, δ (L) = y-} y t becomes, [delta] ^(L) is obtained.

When δ ^(L) is obtained, δ ^{(L-1) of the} front layer can be obtained using the above equation (6). In this way, the δ of the presheaf is sequentially obtained, and using the values of these δ, the derivative of the error function E for each weight w, that is, the gradient ∂E / ∂w ^(L) , is obtained from the above equation (2 ^). Is asked. When the gradient ∂E / ∂w ^{(L) is} obtained, the weight w is updated by using this gradient ∂E / ∂w ^{(L) so} that the value of the error function E decreases. That is, the weight w is learned.

この場合も、出力層 ( L＝４) の各ノードにおけるδ^（Ｌ）の値は、δ^（Ｌ）＝ｙ_k
−ｙ_tk （ｋ＝１，２・・・ｎ）となり、これらδ^（Ｌ）の値から上式（６）を用いて前層のδ^{（Ｌ−１）}が求まる。
＜本発明による実施例＞ On the other hand, in the classification problem, at the time of learning, the output values y ₁ , y ₂ ... From the output layer (L = 4) are input to the softmax layer, and the output values from the softmax layer are y ₁ ', y. ₂ '..., assuming that the corresponding correct labels are y _t1, y _t2 ..., The following cross entropy error E is used as the error function E.

In this case as well, the value of δ ^(L) at each node of the output layer (L = 4) is δ ^(L) = y _k.
-Y _tk (k = 1, 2, ... n), and δ ^{(L-1) of the} previous layer can be obtained from these values of δ ^(L) using the above equation (6).
<Example according to the present invention>

First, with reference to FIG. 7, an abnormality detection process performed to determine whether or not an abnormality has occurred in the fuel vapor emission prevention system when the vehicle is stopped will be described. FIG. 7 shows
A valve closing command for closing the purge control valve 14, a valve opening command for opening the purge control valve 14, a switching command for switching the flow path switching valve 42 to the normal position, and a switching command for switching the flow path switching valve 42 to the test position. The operation command of the suction pump 40, the stop command of the suction pump 40, and the change in the system internal pressure detected by the pressure sensor 47 are shown. In this example, the system internal pressure indicates the pressure in the fuel tank 5 and the canister 6.

In FIG. 7, t ₀ indicates when the operation of the vehicle is stopped. At this time, a valve closing command for closing the purge control valve 14, a switching command for switching the flow path switching valve 42 to the normal position, and a stop of the suction pump 40 are shown. A directive has been issued. FIG. 7 shows when the purge control valve 14, the flow path switching valve 42, and the suction pump 40 are operating normally based on these commands. Therefore, when the operation of the vehicle is stopped, the purge control valve 14 is closed, the flow path switching valve 42 is switched to the normal position, and the suction pump 40 is stopped. This condition is a predetermined period continues from when the operation stop of the vehicle until time t _1. Since the suction pump 40 continues to be stopped during this fixed period, the suction action by the suction pump 40 is not performed, and therefore, the system internal pressure detected by the pressure sensor 47 is atmospheric pressure.

Then, when the time t ₁ is reached, a switching command for switching the flow path switching valve 42 to the test position and an operation command for the suction pump 40 are issued. On the other hand, a valve closing command is continuously 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 steam flow pipe 12, and the inside of the fuel steam flow pipe 13 between the canister 6 and the purge control valve 14 are closed spaces isolated from the outside air. Since the suction pump 40 is activated, the air in this closed space is gradually sucked by the suction pump 40, and as a result, the pressure in the fuel tank 5 and the canister 6, that is, the pressure sensor 47 is detected. The internal pressure of the system gradually decreases. Then, when the time t ₂ is reached, the pressure drop does not occur, and the system internal pressure stagnates in a lowered state.

When the time t ₂ is reached, the valve opening command for opening the purge control valve 14 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 operate. Therefore, at this time, the suction action by the suction pump 40 is continued, but the purge control valve 14 is opened, so that the internal pressure of the system rises sharply and becomes atmospheric pressure. Then, upon reaching the time t _3, the purge control valve 14, the passage switching valve 42 and the suction pump 40 is returned to the state upon stopping of the vehicle. That is, when reaching the time t _3, the valve closing command for closing the purge control valve 14, a switching command for switching the flow path switching valve 42 to the normal position, and a stop command of the suction pump 40 is issued.

On the other hand, when an abnormality occurs in the fuel vapor emission prevention system, the change pattern of the system internal pressure detected by the pressure sensor 47 becomes a change pattern different from the normal system internal pressure change pattern shown in FIG. 7. Next, this will be described with reference to FIGS. 8 to 10. In addition, in FIGS. 8 to 10, similarly to FIG. 7, the valve closing command for closing the purge control valve 14, the valve opening command for opening the purge control valve 14, and the flow path switching valve 42 are set to the normal positions. The switching command for switching, the switching command for switching the flow path switching valve 42 to the test position, the operation command for the suction pump 40, the stop command for the suction pump 40, and the change in the system internal pressure detected by the pressure sensor 47 are shown. Further, in FIGS. 8 to 10, the broken line indicates the change pattern of the system internal pressure at the normal time shown in FIG. 7.

The solid line in FIG. 8 shows, for example, the change pattern of the system internal pressure when a small hole is formed in the pipe wall of the fuel vapor flow pipe 12 or 13. In this case, since the outside air continues to flow into the system through the small hole, the system internal pressure does not drop to the normal system internal pressure. Therefore, the change pattern of the system internal pressure in this case is the normal system internal pressure. The change pattern is different from the change pattern of.

On the other hand, the solid line in FIG. 9 indicates a valve opening abnormality in which the purge control valve 14 continues to open even when a valve closing command for closing the purge control valve 14 is issued. At this time, since the inside of the system continues to communicate with the atmosphere through the purge control valve 14, the internal pressure of the system is maintained at atmospheric pressure as shown by the solid line in FIG. The solid line in FIG. 10 indicates a valve closing abnormality in which the purge control valve 14 continues to close even when a valve opening command for opening the purge control valve 14 is issued. At this time, as shown by the solid line in FIG. 9, the system pressure even after the time t ₂ is maintained at negative pressure.

As described above, the change pattern of the system internal pressure becomes a change pattern different from the normal change pattern of the system internal pressure when an abnormality occurs. Therefore, in the embodiment according to the present invention, in order to learn the difference in the change pattern of the system internal pressure, the system internal pressure is detected at regular intervals. Next, this will be described with reference to FIG. .. Note that, as in FIG. 7, FIG. 11 shows a valve closing command for closing the purge control valve 14, a valve opening command for opening the purge control valve 14, and a switching command for switching the flow path switching valve 42 to the normal position. A switching command for switching the flow path switching valve 42 to the test position, an operation command for the suction pump 40, a stop command for the suction pump 40, and a change in the system internal pressure detected by the pressure sensor 47 are shown. Further, the solid line in FIG. 11 shows the change pattern of the system internal pressure at the normal time shown in FIG. 7.

Referring to FIG. 11, in the period xt time t ₃ from the time t _1, at regular time intervals Delta] t, the system pressure is detected by the pressure sensor 47. In FIG. 11, x ₁ , x _{2 ...} X _n-1 , x _n indicate the system internal pressure detected by the pressure sensor 47 at intervals of Δt for a certain period of time. The system internal pressures x ₁ , x _{2 ...} X _n-1 , x _{n at} regular time Δt detected by the pressure sensor 47 are temporarily stored in the storage device. In the embodiment according to the present invention, the fuel vapor is discharged by using a neural network based on the system internal pressures x ₁ , x _{2 ...} x _n-1 , x _n once stored in the storage device at every Δt for a certain period of time. An anomaly determination estimation model is created that can estimate whether or not an abnormality has occurred in the prevention system. In this embodiment, the pressure in the fuel tank 5 and the canister 6 is detected by the pressure sensor 47. Therefore, in this embodiment, the pressure sensor 47 does not necessarily have to be arranged in the suction passage 46, and can be arranged in any place where the pressure in the fuel tank 5 and the canister 6 can be detected.

Next, the neural network used to create this abnormality determination estimation model will be described with reference to FIG. Referring to FIG. 12, in this neural network 70 as well as the neural network shown in FIG. 6, L = 1 is an input layer, L = 2 and L = 3 are hidden layers, and L = 4 is an output layer, respectively. Shown. As shown in FIG. 14, the input layer (L = 1) consists of n + k nodes, and n input values x ₁ , x _{2 ...} x _n-1 , x _n and k input values xx. ₁ , xx _{2 ... x} x _k-1 , xx _k are input to each node of the input layer (L = 1). In this case, the n input values x ₁ , x _{2 ...} x _n-1 , x _n are the system internal pressures x ₁ , x _{2 ...} x _n-1 , for each fixed time Δt shown in FIG. x _n .

On the other hand, although the hidden layer (L = 2) and the hidden layer (L = 3) are shown in FIG. 12, the number of layers of these hidden layers can be one or any number, and these can be set. The number of hidden layer nodes can also be arbitrary. Further, in this embodiment, the number of nodes in the output layer (L = 4) is set to 4, and the output values from the nodes in the output layer (L = 4) are y ₁ ', y ₂ ', and y ₃ It is indicated by', y ₄ '. These output values _{y 1 ', y 2',} y 3 ', y 4' is fed to Softmax layer SM, are converted respectively to a corresponding output value _{y 1, y 2, y 3} , y 4. The sum of these output values y ₁ , y ₂ , y ₃ , and y ₄ is 1, and each output value y ₁ , y ₂ , y ₃ , and y ₄ represents a ratio to 1.

Next, the input values xx ₁ , xx _{2 ... X} x _k-1 , and xx _k in FIG. 12 will be described with reference to the list shown in FIG. As described above, in the embodiment according to the present invention, the fuel vapor emission prevention system is based on the system internal pressure x ₁ , x _{2 ...} x _n-1 , x _n at regular time Δt intervals using a neural network. It is estimated whether or not there is an abnormality in the system. However, the internal pressure of the system changes under the influence of external factors such as atmospheric pressure and internal factors such as the remaining amount of fuel in the fuel tank 5, and therefore, whether or not an abnormality in the fuel vapor emission prevention system has occurred. It is necessary to consider the effects of these factors when estimating the fuel.

FIG. 13 lists the input parameters to the neural network 70 that serve as these factors. Note that, in FIG. 13, input parameters that strongly affect the change in the system internal pressure are listed as essential input parameters, and although not as much as the essential input parameters, the input parameters that strongly affect the change in the system internal pressure are listed. However, the input parameters that affect the change in the system internal pressure are listed as auxiliary input parameters, although they are listed as the input parameters that have a large influence.

As shown in FIG. 13, system internal pressure x ₁ , x _{2 ...} x _n-1 , x _n and atmospheric pressure xx ₁ are essential input parameters. In this case, it is natural that the system internal pressures x ₁ , x _{2 ...} X _n-1 , and x _n are essential input parameters. On the other hand, if the atmospheric pressure changes, the internal pressure of the system also changes accordingly. Therefore, atmospheric pressure xx ₁ is an essential input parameter.

On the other hand, when the internal pressure of the system becomes negative due to the suction action of the suction pump 40, the internal pressure of the system rises when the fuel in the fuel tank 5 evaporates. In this case, the larger the amount of fuel evaporated per unit time, the more the system. The amount of change in internal pressure becomes large. On the other hand, the amount of fuel evaporated per unit time is proportional to the remaining amount of fuel in the fuel tank 5. Therefore, the larger the remaining amount of fuel in the fuel tank 5, the greater the influence on the internal pressure of the system. Therefore, as shown in FIG. 13, the remaining amount of fuel xx ₂ in the fuel tank 5 is an input parameter having a large influence.

Further, as the fuel temperature in the fuel tank 5 increases, the amount of fuel evaporated per unit time increases, so that the fuel temperature in the fuel tank 5 also affects the internal pressure of the system. However, since the effect on the system internal pressure is smaller than the remaining amount of fuel in the fuel tank 5, the fuel temperature xx ₃ in the fuel tank 5 is used as an auxiliary input parameter as shown in FIG. Further, the performance of the suction pump 40 also affects the internal pressure of the system to some extent, and therefore, as shown in FIG. 13, the flow rate characteristic value xx ₄ of the suction pump 40 is used as an auxiliary input parameter.

FIG. 14 shows the relationship between the flow rate characteristic value of the suction pump 40 and the maximum pump flow rate of the suction pump 40. As shown in FIG. 14, the flow rate characteristic value of the suction pump 40 is 1.0 when the maximum pump flow rate of the suction pump 40 which is a certain reference is G, and the maximum pump of each suction pump 40. The flow rate characteristic value of the suction pump 40 is determined according to the flow rate. The flow rate characteristic value of the suction pump 40 indicates an index indicating the capacity of the suction pump 40. When the capacity of the suction pump 40 is high, the flow characteristic value of the suction pump 40 is high, and when the capacity of the suction pump 40 is low, the flow characteristic value is high. The flow rate characteristic value of the suction pump 40 becomes low.

15, the output value _{y 1} as shown in FIG. _{12 ', y 2', y} 3 ', y 4' if and output values _{y 1, y 2, y 3} , y 4 represents the what state The list is shown. As can be seen from Figure 15, the output value y _{1 'and} the output value y ₁ shows a perforated abnormalities small hole in the pipe wall of the fuel vapor flow tube 12 or 13 is open, the output value y _2' and output value y ₂ shows a valve opening defect of the purge control valve 14 continues to open, the output value y _{3 'and} the output value y ₃ is shows a closing abnormality that the purge control valve 14 continues closed cage, the output value y _{4 'and} the output value y ₄ shows the normal.

Now, as the input values x ₁ , x _{2 ...} x _n-1 , x _n and the input values xx ₁ , xx _{2 ... x} x _k-1 , xx _k of the neural network 70 shown in FIG. Only the values of the essential input parameters shown in 13, i.e., system internal pressure x ₁ , x _{2 ...} x _n-1 , x _n and atmospheric pressure xx ₁ can be used. Of course, in addition to the values of the required input parameters, the values of the input parameters that have a large influence can be the input values xx ₁ , xx _{2 ... x} xx _k-1 , xx _k, and the required input parameters In addition to the values, the values of the input parameters having a large influence and the values of the auxiliary input parameters can be set to the input values xx ₁ , xx _{2 ... X} xx _k-1 , xx _k . In addition to the essential input parameter values, the input parameter values that have a large influence and the auxiliary input parameter values are also input values xx ₁ , xx _{2 ... x} xx _k-1 , xx _k (this). In the example, an example according to the present invention will be described by taking the case of k = 4) as an example.

FIG. 16 shows the input values x ₁ , x _{2 ...} x _n-1 , x _n , the input values xx ₁ , xx _{2 ... x} x _k-1 , xx _k, and the teacher data, that is, the correct label. It shows a training data set created using yt. In FIG. 16, the input values x ₁ , x _{2 ...} X _n-1 , x _n indicate the system internal pressure at every Δt for a certain period of time, and this system internal pressure is detected by the pressure sensor 47. Further, in FIG. 16, the input values xx ₁ , xx _{2 ... X} xx _k-1 , xx _k are, that is, the input values xx ₁ , xx ₂ , xx _{3, and} xx ₄ are atmospheric pressure and fuel tank, respectively. It shows the remaining amount of fuel in 5, the fuel temperature in the fuel tank 5, and the flow characteristic value of the suction pump 40, that is, an index showing 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, and the fuel temperature in the fuel tank 5 is detected by the temperature sensor 8. Then, the flow rate characteristic value of the suction pump 40 is calculated based on the relationship shown in FIG.

On the other hand, in FIG. 16, (in this _{example, s = 4) yt 1 ···} yt S , the output value _{y 1} as shown in each of FIG _{15 ', y 2', y} 3 ', y 4' and the output value The teacher data for y ₁ , y ₂ , y ₃ , and y ₄ , that is, the correct label is shown. That is, in FIG. 16, yt ₁ indicates the correct label when there is a perforated abnormality in which a small hole is formed in the pipe wall of the fuel steam flow pipe 12 or 13, and yt ₂ indicates the purge control valve. 14 shows the correct label when a valve opening abnormality occurs in which the valve keeps opening, and yt ₃ shows the correct answer label when a valve closing abnormality occurs in which the purge control valve 14 keeps closing. , Yt ₄ indicate the correct label when it is normal.

In this case, for example, when there is a perforated abnormality in which a small hole is formed in the pipe wall of the steam flow pipe 12 or 13, only the correct answer label yt ₁ is set to 1, and the remaining correct answer labels yt _2, yt _3, yt ₄ is all zero. Similarly, when a valve opening abnormality occurs in which the purge control valve 14 continues to open, only the correct answer label yt ₂ is set to 1, and the remaining correct answer labels yt _1, yt _{3, and} yt ₄ are all set to zero. , When a valve closing abnormality occurs in which the purge control valve 14 continues to close, only the correct answer label yt ₃ is set to 1, and the remaining correct answer labels yt _1, yt _{2, and} yt ₄ are all set to zero, which is normal. When it is time, only the correct label yt ₄ is set to 1, and the remaining correct labels yt _1, yt _{2, and} yt ₃ are all set to zero.

On the other hand, as shown in FIG. 16, in this training data set, the input values x ₁ , x _{2 ...} x _n-1 , x _n and the input values xx ₁ , xx _{2 ... x} x _k-1 , M pieces of data representing the relationship between xx _k and the correct label yt have been acquired. For example, in the second data (No. 2), the acquired input values x ₁₂ , x _{22 ...} x _n-12 , x _n2 and the input values xx ₁₂ , xx _{22 ...} xx _k-12 , Xx _k2 and correct labels yt _{12 ...} yt _S2 are listed, and in the m-1st data (No. m-1), the input value x _{1 m-1} of the acquired input parameter, x _{2m-1 ...} x _n-1m-1 , x _nm-1 , input values xx _1m-1 , xx _{2m-1 ...} xx _k-1m-1 , xx _km-1 , and correct label yt _Sm-1 is listed.

Next, a method of creating the training data set shown in FIG. 16 will be described. FIG. 17 shows an example of how to create a training data set. Referring to FIG. 17, the engine body 1, the fuel tank 5, the canister 6, and the like shown in FIG. 1 are installed in a closed test chamber 80 in which the internal pressure can be adjusted, and the test control device 81 is used in the fuel vapor emission prevention system. In order to estimate whether or not an abnormality has occurred, an abnormality detection process is performed in which the purge control valve 14, the flow path switching valve 42, and the suction pump 40 are operated according to a predetermined operation order. At this time, the state of the fuel steam discharge prevention system is a state in which a small hole is opened in the pipe wall of the fuel steam flow pipe 12 or 13, and a valve opening abnormality in which the purge control valve 14 continues to open is generated. Is occurring, the purge control valve 14 continues to close, a valve closing abnormality occurs, and the normal state is sequentially changed, and in each of the changed states, the atmospheric pressure and the remaining fuel in the fuel tank 5 The abnormality detection process is repeatedly performed while sequentially changing the combination of the amount, the fuel temperature in the fuel tank 5, and the flow rate characteristic value of the suction pump 40.

While this anomaly detection process is taking place, the data needed to create the training dataset is acquired. FIG. 18 shows an abnormality detection processing routine executed in the test control device 81 to perform this abnormality detection processing. This abnormality detection processing routine is executed by an interrupt at every Δt for a certain period of time shown in FIG. In the routine shown in FIG. 18, t represents the time obtained by setting the time t ₀ when the operation of the engine is stopped in FIG. 17 as zero and starting from this time t ₀ .

With reference to FIG. 18, first, in step 100, it is determined whether or not the time t is before the time t ₁ shown in FIG. When the time t is before the time t ₁ shown in FIG. 11, the processing cycle is terminated. On the other hand, when it is determined that the time t is not before the time t ₁ shown in FIG. 11, the process proceeds to step 101 to determine whether or not the time t has reached the time t ₁ shown in FIG. Will be done. When the time t reaches the time t ₁ shown in FIG. 11, the process proceeds to step 102, a switching command for switching the flow path switching valve 42 to the test position is issued, and then the process proceeds to step 103 to operate the suction pump 40. A command is issued. Then, in step 104, acquisition and storage of the system internal pressure _xn detected by the pressure sensor 47 is started. At this time, the system internal pressure _xn is stored in the test control device 81.

On the other hand, when it is determined in step 101 that the time t is not the time t ₁ shown in FIG. 11, the process proceeds to step 105, and the system internal pressure _xn is acquired and stored in the test control device 81. That is, as shown in FIG. 11, the system internal pressure _xn is acquired every Δt for a certain period of time, and the system internal pressure _xn acquired every Δt for a certain period of time is stored in the test control device 81. Next, in step 106, it is determined whether or not the time t has reached the time t ₂ shown in FIG. When the time t reaches the time t ₂ shown in FIG. 11, a valve opening command is issued to proceed to step 107 to open the purge control valve 14.

On the other hand, in step 106, when it is determined that the time t is not the time t ₂ shown in FIG. 11, the process proceeds to step 108 to determine whether or not the time t has reached the time t ₃ shown in FIG. Will be done. When the time t is not the time t ₃ shown in FIG. 11, the processing cycle is terminated. In contrast, the time t is, if it is determined to have become a time t ₃ when shown in Figure 11, the routine proceeds to step 109, the closing command for closing the purge control valve 14 is issued.
Next, in step 110, a switching command for switching the flow path switching valve 42 to the normal position is issued. Next, in step 111, a stop command for the suction pump 40 is issued. Then, in step 112, the acquisition and storage of the system internal pressure _xn detected by the pressure sensor 47 is stopped. Then, in step 113, the abnormality detection process is completed. In the test control device 81 shown in FIG. 17, when the abnormality detection process is completed, the next abnormality detection process is started.

In this way, a state in which a small hole is formed in the pipe wall of the fuel steam flow pipe 12 or 13 is generated, a state in which the purge control valve 14 continues to open, and a state in which a valve opening abnormality occurs, purge The atmospheric pressure, the remaining amount of fuel in the fuel tank 5, the fuel temperature in the fuel tank 5, and the suction pump 40 in each state of the valve closing abnormality in which the control valve 14 continues to close and the normal state. The system internal pressure _{xn at} every Δt for a certain period of time when the combination of the flow rate characteristic values of the above is changed is stored in the test control device 81. That is, the No. 1 of the training data set shown in FIG. 1 to No. Input values of _m x _{1 m} , x _{2 m ...} x _nm-1 , x _nm , input values xx _{1 m} , xx _{2 m ...} xx _km-1 , xx _km , and correct label yt _sm (m = 1, 2) , 3 ... m) are stored in the test control device 81.

When the training data set as shown in FIG. 16 is created in this way, the weights of the neural network 70 shown in FIG. 12 are learned using the electronic data of the created training data set. In the example shown in FIG. 17, a learning device 82 for learning the weight of the neural network is provided. A personal computer can also be used as the learning device 82. As shown in FIG. 17, the learning device 82 includes a CPU (microprocessor) 83 and a storage device, that is, a memory 84. In the example shown in FIG. 17, the number of nodes of the neural network shown in FIG. 12 and the electronic data of the created training data set are stored in the memory 84 of the learning device 82, and the weight of the neural network is learned in the CPU 83. Will be.

FIG. 19 shows a neural network weight learning processing routine performed by the learning device 82.
Referring to FIG. 19, first, in step 200, each data of the training data set for the neural network 70 stored in the memory 84 of the learning device 82 is read. Then, in step 201, the number of nodes in the input layer (L = 1), the number of nodes in the hidden layer (L = 2) and the hidden layer (L = 3), and the number of nodes in the output layer (L = 4) of each neural network 70. Is read, and then in step 202, a neural network 70 as shown in FIG. 12 is created based on the number of these nodes.

Next, in step 203, the weight of the neural network 70 is learned. In this step 203, first, the first (No. 1) input values x ₁ , x _{2 ...} x _n-1 , x _n and the input values xx ₁ , xx _{2 ... x} x _{k in FIG. -1} , xx _k are input to each node of the input layer (L = 1) of the neural network 70. At this time, output values are output from each node of the output layer of the neural network 70 as y ₁ ', y ₂ ', y ₃ ', y ₄ ', and these output values y ₁ ', y ₂ ', y ₃ '. , _{y 4} 'is fed to softmax layer SM, it is converted into an output value _{y 1, y 2, y 3} , y 4 respectively corresponding to. Next, the cross entropy error E described above is calculated from these output values y ₁ , y ₂ , y ₃ , y _4, and the correct answer labels yy _{1 ...} yt _S , so that the cross entropy error E becomes smaller. The weights of the neural network 70 are learned by using the error back propagation method described above.

When the learning of the weight of the neural network 70 based on the first (No. 1) data in FIG. 16 is completed, then the learning of the weight of the neural network 70 based on the second (No. 2) data in FIG. 16 is performed. , The error back propagation method is used. Similarly, the weights of the neural network 70 are learned sequentially up to the mth (No. m) in FIG. When the learning of the weights of the neural network 70 is completed for all of the first (No. 1) to the mth (No. m) in FIG. 16, the process proceeds to step 204.

In step 204, it is determined whether or not the cross entropy error E is equal to or less than a preset setting error. When it is determined that the cross entropy error E is not less than or equal to the preset setting error, the process returns to step 203, and the weight learning of the neural network 70 is performed again based on the training data set shown in FIG. It is said. Next, learning of the weight of the neural network 70 is continued until the cross entropy error E becomes equal to or less than the preset setting error. When it is determined in step 204 that the cross entropy error E is equal to or less than the preset setting error, the process proceeds to step 205, and the learned weight of the neural network 70 is stored in the memory 84 of the learning device 82. In this way, an abnormality determination estimation model capable of estimating whether or not an abnormality has occurred in the fuel vapor emission prevention system is created.

In the embodiment according to the present invention, the failure diagnosis of the fuel vapor emission prevention system in the commercial vehicle is performed by using the abnormality determination estimation model of the fuel vapor emission prevention system thus created. Therefore, the fuel vapor emission prevention is performed. The abnormality determination estimation model of the system is stored in the electronic control unit 20 of the commercial vehicle. FIG. 20 shows a data reading routine to the electronic control unit performed in the electronic control unit 20 in order to store the abnormality determination estimation model of the fuel vapor emission prevention system in the electronic control unit 20 of a commercial vehicle.

Referring to FIG. 20, first, in step 300, the number of nodes in the input layer (L = 1) of the neural network 70 shown in FIG. 12, the hidden layer (L = 2) and the hidden layer (L = 3). The number of nodes and the number of nodes in the output layer (L = 4) are read into the memory 22 of the electronic control unit 20, and then, in step 301, a neural network 71 as shown in FIG. 21 is created based on these numbers of nodes. Ru. As can be seen from FIG. 21, the softmax layer is removed in this neural network 71. In this case, the neural network 71 may be provided with the softmax layer SM as shown in FIG. Next, in step 302, the learned weight of the neural network 70 is read into the memory 22 of the electronic control unit 20. As a result, the abnormality determination estimation model of the fuel vapor emission prevention system is stored in the electronic control unit 20 of the commercial vehicle.

Next, an abnormality detection routine of the fuel vapor emission prevention system executed in a commercial vehicle will be described with reference to FIG. This routine is also executed by interrupts at regular intervals. With reference to FIG. 22, first, in step 400, it is determined whether or not the operation of the vehicle has been stopped based on the output signal of the operation start / stop switch 31. The processing cycle ends when the vehicle is not stopped. On the other hand, when it is determined that the operation of the vehicle has been stopped, the process proceeds to step 401, and it is determined whether or not the detection permission flag for permitting the abnormality detection of the fuel vapor emission prevention system is set. When the vehicle first proceeds to step 401 after the vehicle is stopped, the detection permission flag is not set, so the process proceeds to step 402. In step 402, the abnormality detection process shown in FIG. 18 is executed.

When this abnormality detection process is executed, the system internal pressure _xn is acquired every Δt for a certain period of time, and the system internal pressure _xn acquired every Δt for a certain period of time is stored in the memory 22 of the electronic control unit 20. Next, in step 403, it is determined whether or not the abnormality detection process is completed. When the abnormality detection process is not completed, the process cycle is terminated. On the other hand, when it is determined that the abnormality detection process is completed, the process proceeds to step 404 and the detection permission flag is set. When the detection permission flag is set, the process proceeds from step 401 to step 405 in the next processing cycle.

In step 405, the system internal pressures x ₁ , x _{2 ...} X _n-1 , x _n stored in the memory 22 of the electronic control unit 20 for each constant time Δt are read. Next, in step 406, the input values xx ₁ , xx _{2 ...} xx _k-1 , xx _k stored in the memory 22 of the electronic control unit 20 are read. Then, in step 407, the system internal pressures x _n x ₁ , x _{2 ...} x _n-1 , x _n and the input values xx ₁ , xx _{2 ... x} x _k-1 , xx _k are set every Δt for a certain period of time. It is input to each node of the input layer (L = 1) of the neural network 71 shown in FIG. From each node in the output layer in this case the neural network 70, the output values _{y 1 ', y 2',} y 3 ', y 4' is outputted, whereby the step 408 Oite, the output value _{y 1} ', y _{2 ', y 3', y} 4 ' is obtained.

Next, in step 409, the maximum output value y _i'is selected from the acquired output values y ₁ ', y ₂ ', y ₃ ', and y ₄ '. At this time, it is presumed that the abnormal state shown in FIG. 15 corresponding to the maximum output value y _i'has occurred. Therefore, in step 410, it is determined that the abnormal state shown in FIG. 15 corresponding to the maximum output value y _i'has occurred. For example, the abnormality shown in FIG. 15 corresponding to the maximum output value y _i'is determined. A warning light is lit to indicate that a condition has occurred. Then, in step 411, the abnormality detection is completed.

As described above, in the abnormality detection device of the fuel vapor emission prevention system according to the present invention, the fuel vapor emission prevention system includes a canister 6 in which a fuel steam chamber 10 and an atmospheric pressure chamber 11 are formed on both sides of the activated coal layer 9, respectively. On the one hand, the fuel steam chamber 10 communicates with the internal space above the fuel liquid level of the fuel tank 5, and on the other hand, it is connected to the intake passage of the engine via the purge control valve 14. Further, the fuel vapor emission prevention system includes a flow path switching valve 42 capable of selectively connecting the atmospheric pressure chamber 11 to the atmosphere and the suction pump 40, and a pressure sensor 47 for detecting the pressure in the fuel tank 5 and the canister 6. Is equipped with. A valve closing command for closing the purge control valve 14 when the vehicle is stopped, a switching command for switching the switching position of the flow path switching valve 42 to a switching position where the atmospheric pressure chamber 11 is connected to the suction pump 40, and fuel. An abnormality detection process is performed to generate a pump operation command for operating the suction pump 40 in order to make the inside of the tank and the inside of the canister 6 negative pressure. When this abnormality detection process is being performed, the pressures in the fuel tank 5 and the canister 6 detected by the pressure sensor 40 at regular time intervals are stored in the storage device, and are stored in the storage device at regular time intervals. The pressure in the fuel tank 5 and the canister 6 and at least atmospheric pressure when the abnormality detection process is performed are set as the input parameters of the neural network, and the correct answer label is when there is a hole in the above system to leak fuel vapor. As a result, the trained neural network in which the weight has been learned is stored, and when the vehicle is stopped, the trained neural network is used to detect a perforated abnormality that leaks fuel vapor from the above input parameters. Ru.

In this case, in the embodiment according to the present invention, the above-mentioned abnormality detection process includes a process of generating a valve opening command for opening the purge control valve 14 after the valve closing command for the purge control valve 14 is generated. The pressure in the fuel tank 5 and the canister 6 at regular intervals stored in the storage device and at least atmospheric pressure at the time of abnormality detection processing are used as input parameters of the neural network, and fuel vapor is leaked to the above-mentioned system. When a hole is formed, when a valve opening abnormality occurs in which the purge control valve 14 keeps opening, and when a valve closing abnormality occurs in which the purge control valve 14 keeps closing, the correct answer label is used. The trained neural network in which the weights have been learned is stored, and when the vehicle is stopped, the trained neural network is used to leak fuel vapor from the above input parameters, and the purge control valve 14 An abnormality in opening the valve and an abnormality in closing the purge control valve 14 are detected.

Further, in the embodiment according to the present invention, the above-mentioned input parameters are the pressure in the fuel tank 5 and the canister 6 stored in the storage device at regular intervals, the atmospheric pressure when the abnormality detection process is performed, and the abnormality. It consists of the remaining amount of fuel in the fuel tank 5 when the detection process is performed. Further, in the embodiment according to the present invention, the above-mentioned input parameters are the pressure in the fuel tank 5 and the canister 6 stored in the storage device at regular intervals, the atmospheric pressure when the abnormality detection process is performed, and the abnormality. It is composed of an index showing the remaining amount of fuel in the fuel tank 5 when the detection process is performed, the temperature of the fuel in the fuel tank 5, and the capacity of the suction pump 40.

23A-38 show another embodiment in which more abnormal conditions are detected. In this other embodiment, as the suction pump module 16 shown in FIG. 1, the suction pump module 16 graphically shown in FIGS. 23A and 23B is used. With reference to FIGS. 23A and 23B, in the suction pump module 16 shown in FIGS. 23A and 23B, with respect to the suction pump module 16 shown in FIGS. 2A and 2B, the atmospheric pressure chamber connecting passage 43 and the suction passage 46 A reference pressure detection passage 50 is added to connect the reference pressure detection passages 50, and a throttle portion 51 is provided in the reference pressure detection passage 50.

Next, the abnormality detection process performed when the vehicle is stopped using the suction pump module 16 shown in FIGS. 23A and 23B will be described. FIG. 24 shows a valve closing command for closing the purge control valve 14, a valve opening command for opening the purge control valve 14, a switching command for switching the flow path switching valve 42 to a normal position, and a flow, similar to those in FIG. A switching command for switching the path switching valve 42 to the test position, an operation command for the suction pump 40, a stop command for the suction pump 40, and a change in the system internal pressure detected by the pressure sensor 47 are shown.

In FIG. 24, t ₀ indicates when the operation of the vehicle is stopped. At this time, a valve closing command for closing the purge control valve 14, a switching command for switching the flow path switching valve 42 to the normal position, and a stop of the suction pump 40 are shown. A directive has been issued. FIG. 24 shows when the purge control valve 14, the flow path switching valve 42, and the suction pump 40 are operating normally based on these commands. Therefore, when the operation of the vehicle is stopped, the purge control valve 14 is closed, the flow path switching valve 42 is switched to the normal position, and the suction pump 40 is stopped. This condition is a predetermined period continues from when the operation stop of the vehicle until time t _1. Since the suction pump 40 continues to be stopped during this fixed period, the suction action by the suction pump 40 is not performed, and therefore, the system internal pressure detected by the pressure sensor 47 is atmospheric pressure.

Then, when the time t ₁ is reached, an operation command for the suction pump 40 is issued. At this time, the flow path switching valve 42 is switched to the normal position shown in FIG. 23A. Therefore, when the suction pump 40 operates, the outside air moves from the atmospheric communication pipe 17 to the atmospheric communication pipe connecting path 44, the atmospheric communication passage 45, It is sucked through the first passage 48, the air communication chamber connecting passage 43, the reference pressure detection passage 50 having the throttle portion 51, and the suction passage 46. At this time, since the throttle portion 51 is provided, the inside of the suction passage 46 becomes a negative pressure, and therefore, the system internal pressure detected by the pressure sensor 47 decreases as shown in FIG. 24.

In this case, the change in the system internal pressure detected by the pressure sensor 47 represents the change in the system internal pressure when a hole having the same diameter as the diameter of the throttle portion 51 is opened. Therefore, the change in the system internal pressure at this time is , It becomes a standard when judging whether or not a hole is opened in the fuel vapor emission prevention system. Therefore, the passage 50 is referred to as a reference pressure detection passage. Then, upon reaching the time t _2, the flow path switching valve 42, switching instruction for switching to the test position shown in FIG. 23B is emitted. When the flow path switching valve 42 is switched to the test position, the pressure inside the fuel tank 5 and the inside of the canister 6 are at atmospheric pressure, so that the pressure inside the suction passage 46 rises. Thus, the system pressure detected by the pressure sensor 47 reaches the time t _2, the as shown in Figure 24, elevated, then reaches the time t _3, the fuel tank 5 by the suction pump 40 and the canister 6 It begins to decrease due to the suction action of air from inside. Then, at a time reaching the time t _4, no longer occur pressure drop, the system pressure is stagnant in a state of reduced.

Upon reaching the time t _4, the valve opening command for opening the purge control valve 14 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 operate. Therefore, at this time, the suction action by the suction pump 40 is continued, but the purge control valve 14 is opened, so that the internal pressure of the system rises sharply and becomes atmospheric pressure. Then, upon reaching the time t _5, the purge control valve 14, the passage switching valve 42 and the suction pump 40 is returned to the state upon stopping of the vehicle. That is, when reaching the time t _5, the valve closing command for closing the purge control valve 14, a switching command for switching the flow path switching valve 42 to the normal position, and a stop command of the suction pump 40 is issued.

On the other hand, in this embodiment, as shown in FIG. 24, the internal pressure of the system is detected by the pressure sensor 47 every Δt for a certain period of time from the time when the vehicle is stopped, that is, during the period xt from time t ₀ to time t _5. Will be done. In FIG. 24, x ₁ , x _{2 ...} X _n-1 , x _n indicate the system internal pressure detected by the pressure sensor 47 at intervals of Δt for a certain period of time. The system internal pressures x ₁ , x _{2 ...} X _n-1 , x _{n at} regular time Δt detected by the pressure sensor 47 are temporarily stored in the storage device. Also in this embodiment, a fuel vapor emission prevention system is used by using a neural network based on the system internal pressure x ₁ , x _{2 ...} x _n-1 , x _n once stored in the storage device at every Δt for a certain period of time. An abnormality judgment estimation model that can estimate whether or not an abnormality has occurred is created.

Also in this embodiment, when an abnormality occurs in the fuel vapor emission prevention system, the change pattern of the system internal pressure detected by the pressure sensor 47 becomes a change pattern different from the normal system internal pressure change pattern shown in FIG. 24. .. Next, this will be described with reference to FIGS. 25 to 32. In addition, in FIGS. 25 to 32, as in FIG. 24, the valve closing command for closing the purge control valve 14, the valve opening command for opening the purge control valve 14, and the flow path switching valve 42 are set to the normal positions. The switching command for switching, the switching command for switching the flow path switching valve 42 to the test position, the operation command for the suction pump 40, the stop command for the suction pump 40, and the change in the system internal pressure detected by the pressure sensor 47 are shown. Further, in FIGS. 25 to 32, the broken line indicates the change pattern of the system internal pressure at the normal time shown in FIG. 24.

The solid line in FIG. 25 shows the change pattern of the system internal pressure detected by the pressure sensor 47 when an abnormality occurs in the pressure sensor 47. When an abnormality occurs in the pressure sensor 47 and the output of the pressure sensor 47 becomes unstable, the internal pressure of the system detected by the pressure sensor 47 is maintained at the atmospheric pressure as between the time t ₀ and the time t _1. It repeats fluctuations. Therefore, the change pattern of the system internal pressure in this case is different from the change pattern of the system internal pressure at the normal time.

The solid line in FIG. 26 shows the change pattern of the system internal pressure detected by the pressure sensor 47 when the suction pump 40 continues to operate even if a stop command is issued to the suction pump 40. Shown. As can be seen from FIG. 26, between time t ₀ and time t ₂ , the flow path switching valve 42 is switched to the normal position shown in FIG. 23A, so that the suction pump 40 is activated at this time. Then, the outside air passes from the atmosphere communication pipe 17 to the atmosphere communication pipe connection passage 44, the atmosphere communication passage 45, the first passage 48, the atmosphere communication chamber connection passage 43, the reference pressure detection passage 50 having the throttle portion 51, and the suction passage 46. It is sucked through. At this time, the inside of the suction passage 46 becomes a negative pressure, and therefore, the system internal pressure detected by the pressure sensor 47 at this time shows a low value as shown in FIG. Therefore, the change pattern of the system internal pressure in this case is also different from the change pattern of the system internal pressure at the normal time.

The solid line in FIG. 27 shows the change pattern of the system internal pressure detected by the pressure sensor 47 when the operation stop abnormality of the suction pump 40 continues to stop even if the operation command is issued to the suction pump 40. Shown. At this time, the entire inside of the fuel vapor emission prevention system becomes atmospheric pressure, so that the system internal pressure detected by the pressure sensor 47 is maintained at atmospheric pressure as shown in FIG. 27. Therefore, the change pattern of the system internal pressure in this case is also different from the change pattern of the system internal pressure at the normal time.

The solid line in FIG. 28 shows the change in the system internal pressure detected by the pressure sensor 47 when the flow path switching valve 42 continues to be fixed at the test position shown in FIG. 23B when the test position fixing abnormality of the flow path switching valve 42 occurs. Shows the pattern. In this case, when the suction pump 40 is actuated at time t _1, the air communicating pipe connecting road 44 outside air from the atmosphere communicating pipe 17, the air communication passage 45, the first passage 48, the air communicating chamber connecting passage 43, the diaphragm The system internal pressure, which is sucked through the reference pressure detection passage 50 having the portion 51 and the suction passage 46 and detected by the pressure sensor 47, rapidly drops and is maintained in the lowered state as shown in FIG. .. Therefore, the change pattern of the system internal pressure in this case is also different from the change pattern of the system internal pressure at the normal time.

The solid line in FIG. 29 shows the change in the system internal pressure detected by the pressure sensor 47 when the flow path switching valve 42 continues to be fixed at the normal position shown in FIG. 23A when the normal position fixing abnormality of the flow path switching valve 42 occurs. Shows the pattern. In this case, when the suction pump 40 is actuated at time t _1, the suction effect of air in and canister 6 fuel tank 5 is started, therefore, the system pressure detected by the pressure sensor 47, FIG. 28 As shown in, it gradually decreases. Therefore, the change pattern of the system internal pressure in this case is also different from the change pattern of the system internal pressure at the normal time.

The solid line in FIG. 30 shows, for example, the change pattern of the system internal pressure when a small hole is made in the pipe wall of the fuel vapor flow pipe 12 or 13. In this case, since the outside air continues to flow into the system through the small hole, the system internal pressure does not drop to the normal system internal pressure. Therefore, the change pattern of the system internal pressure in this case is also the normal system internal pressure. The change pattern is different from the change pattern of.

The solid line in FIG. 31 indicates a valve opening abnormality in which the purge control valve 14 continues to be opened even when a valve closing command for closing the purge control valve 14 is issued. At this time, the inside of the fuel tank 5 and the inside of the canister 6 are at atmospheric pressure. Thus, at time t _2, the channel switching valve 42 is switched to the test position shown in FIG. 23B from the normal position shown in FIG. 23A, also in this case the suction pump 40 have been actuated by the pressure sensor 47 The detected system internal pressure rises to atmospheric pressure and is maintained at atmospheric pressure as shown in FIG. Therefore, the change pattern of the system internal pressure in this case is also different from the change pattern of the system internal pressure at the normal time.

The solid line in FIG. 32 indicates a valve closing abnormality in which the purge control valve 14 continues to be closed even when a valve opening command for opening the purge control valve 14 is issued. At this time At time t _4, even if the valve opening command for opening the purge control valve 14 is issued, so continue to close the purge control valve 14, and in the canister 6 fuel tank 5 in the negative pressure The system internal pressure maintained and detected by the pressure sensor 47 is maintained low as shown in FIG. Therefore, the change pattern of the system internal pressure in this case is also different from the change pattern of the system internal pressure at the normal time.

When such an abnormality occurs, the change pattern of the system internal pressure becomes a change pattern different from the change pattern of the system internal pressure at the normal time. Therefore in this embodiment, as shown in FIG. 24, in the period xt time t ₅ from time t _0, at regular time intervals Delta] t, the system pressure is detected by the pressure sensor 47. In FIG. 24, x ₁ , x _{2 ...} X _n-1 , x _n indicate the system internal pressure detected by the pressure sensor 47 at intervals of Δt for a certain period of time. The system internal pressures x ₁ , x _{2 ...} X _n-1 , x _{n at} regular time Δt detected by the pressure sensor 47 are temporarily stored in the storage device. Also in this embodiment, the fuel vapor emission prevention system is once stored in the storage device using a neural network based on the system internal pressures x ₁ , x _{2 ...} x _n-1 , x _n at regular time Δt intervals. An abnormality judgment estimation model that can estimate whether or not an abnormality has occurred is created.

Next, the neural network 72 used to create this abnormality determination estimation model will be described with reference to FIG. 33. Referring to FIG. 33, also in this neural network 72, as in the neural network shown in FIG. 12, L = 1 is an input layer, L = 2 and L = 3 are hidden layers, and L = 4 is an output layer, respectively. Shown. As shown in FIG. 33, the input layer (L = 1) consists of n + k nodes, and n input values x ₁ , x _{2 ...} x _n-1 , x _n and k input values xx. ₁ , xx _{2 ... x} x _k-1 , xx _k are input to each node of the input layer (L = 1). In this case, the n input values x ₁ , x _{2 ...} x _n-1 , x _n are the system internal pressures x ₁ , x _{2 ...} x _{n-1 for} each fixed time Δt shown in FIG. 24. x _n .

On the other hand, although the hidden layer (L = 2) and the hidden layer (L = 3) are shown in FIG. 33, the number of layers of these hidden layers can be one or any number, and these can be set. The number of hidden layer nodes can also be arbitrary. Further, in this embodiment, the number of nodes in the output layer (L = 4) is set to 9, and the output values from the nodes in the output layer (L = 4) are y ₁ ', y ₂ ', and y ₃ It is indicated by', y ₄ ', y ₅ ', y ₆ ', y ₇ ', y ₈ ', y ₉ '. These output values _{y 1 ', y 2',} y 3 ', y 4', y 5 ', y 6', y 7 ', y 8', y 9 ' is sent to the Softmax layer SM, respectively It is converted to the corresponding output values y ₁ , y ₂ , y ₃ , y ₄ , y ₅ , y ₆ , y ₇ , y ₈ , y ₉ . The sum of these output values y ₁ , y ₂ , y ₃ , y ₄ , y ₅ , y ₆ , y ₇ , y ₈ , and y ₉ is 1, and each output value y ₁ , y ₂ , y ₃ , y ₄ , Y ₅ , y ₆ , y ₇ , y ₈ and y ₉ represent the ratio to 1.

Input value _{xx 1, xx 2 ··· x x} k-1, xx k in FIG. 33 is the same as the input value _{xx 1, xx 2 ··· x x} k-1, xx k in FIG. Since these input values xx ₁ , xx _{2 ... X} xx _k-1 , and xx _k have already been described with reference to the list shown in FIG. 13, these input values xx ₁ , xx _{2 ... x} The description of xx _-1 and xx _k will be omitted.

Figure 34 is the output value _{y 1} as shown in FIG. _{33 ', y 2', y} 3 ', y 4', y 5 ', y 6', y 7 ', y 8', y 9 ' and the output value y shows _{1, y 2, y 3,} y 4, y 5, or list _{y 6, y 7, y 8} , y 9 represents the what state. As can be seen from Figure 34, the output value y _{1 'and} the output value y ₁ shows a perforated abnormalities small hole in the pipe wall of the fuel vapor flow tube 12 or 13 is open, the output value y _2' and output value y ₂ shows a valve opening defect of the purge control valve 14 continues to open, the output value y _{3 'and} the output value y ₃ is shows a closing abnormality that the purge control valve 14 continues closed cage, the output value y _{4 'and} the output value y ₄ shows the abnormality of the pressure sensor 47, the output value y _5' and the output value y ₅ shows the normal position fixation abnormality of the passage switching valve 42 , the output value y _{6 'and} the output value y ₆ shows a test position fixation abnormality of the passage switching valve 42, the output value y _7' and the output value y ₇ may show a continuously operating anomalies of the suction pump 40 cage, the output value y _{8 'and} the output value y ₈ shows a deactivation abnormality of the suction pump 40, the output value y _9' and the output value y ₉ shows the normal.

By the way, also in this embodiment, the input values x ₁ , x _{2 ...} x _n-1 , x _n and the input values xx ₁ , xx _{2 ... x} x _{k-1 of the} neural network 72 shown in FIG. As xx _k , only the values of the essential input parameters shown in FIG. 13, that is, the system internal pressures x ₁ , x _{2 ...} x _n-1 , x _n and atmospheric pressure xx ₁ can be used. Of course, in addition to the values of the required input parameters, the values of the input parameters that have a large influence can be the input values xx ₁ , xx _{2 ... x} xx _k-1 , xx _k, and the required input parameters In addition to the values, the values of the input parameters having a large influence and the values of the auxiliary input parameters can be set to the input values xx ₁ , xx _{2 ... X} xx _k-1 , xx _k . In addition to the essential input parameter values, the input parameter values that have a large influence and the auxiliary input parameter values are also input values xx ₁ , xx _{2 ... x} xx _k-1 , xx _k (this). In the example, this embodiment will be described by taking the case of k = 4) as an example.

Also in this embodiment, first, the input values x ₁ , x _{2 ...} x _n-1 , x _n , the input values xx ₁ , xx _{2 ... x} x _k-1 , xx _k, and the teacher data, That is, the training data set shown in FIG. 16 is created using the correct label yt. Also in this embodiment, in FIG. 16, the input values x ₁ , x _{2 ...} X _n-1 , x _n indicate the system internal pressure at every Δt for a certain period of time detected by the pressure sensor 47, and the input values xx. ₁ , xx _{2 ... x} xx _k-1 , xx _k are the input values xx ₁ , xx ₂ , xx _3, xx ₄ , respectively, the atmospheric pressure, the remaining amount of fuel in the fuel tank 5, and the fuel. The fuel temperature in the tank 5 and the flow rate characteristic value of the suction pump 40 are shown.

On the other hand, in FIG. 16, yt _{1 ...} yt _S (s = 9 in this example) are output values y ₁ ', y ₂ ', y ₃ ', y ₄ ', y ₅ respectively shown in FIG. 34. _{', y 6', y 7} ', y 8', y 9 ' and the output values _{y 1, y 2, y 3} , y 4, y 5, y 6, y 7, the teacher data for _y 8, y _9, That is, the correct answer label is shown. That is, in FIG. 16, yt ₁ indicates the correct label when there is a perforated abnormality in which a small hole is formed in the pipe wall of the fuel steam flow pipe 12 or 13, and yt ₂ indicates the purge control valve. 14 indicates the correct label when a valve opening abnormality occurs in which the valve continues to open, and yt ₃ indicates the correct answer label when a valve closing abnormality occurs in which the purge control valve 14 continues to close. , Yt ₄ indicates the correct label when the pressure sensor 47 has an abnormality, and yt ₅ indicates the correct label when the normal position fixing abnormality of the flow path switching valve 42 has occurred. yt ₆ indicates the correct label when the test position sticking abnormality of the flow path switching valve 42 occurs, and yt ₇ indicates the correct label when the operation continuation abnormality of the suction pump 40 occurs. , Yt ₈ indicates the correct answer label when the suction pump 40 has an abnormal operation stop, and yy ₉ indicates the correct answer label when the suction pump 40 is in the normal state.

In this case, for example, when there is a perforated abnormality in which a small hole is formed in the pipe wall of the steam flow pipe 12 or 13, only the correct answer label yt ₁ is set to 1, and the remaining correct answer labels yt _2, yt _3, yt ₄ , yt _5, yt _6, yt _7, yt _{8 and} yt ₉ are all set to zero. Similarly, when a valve opening abnormality occurs in which the purge control valve 14 continues to open, only the correct answer label yt ₂ is set to 1, and the remaining correct answer labels yt _1, yt _3, yt ₄ , yt _5, yt. _6, yt _7, yt _{8 and} yt ₉ are all set to zero, and when a valve closing abnormality occurs in which the purge control valve 14 keeps closing, only the correct answer label yt ₃ is set to 1 and the remaining correct answer labels. yt _1, yt _2, yt ₄ , yt _5, yt _6, yt _7, yt _8, yt ₉ are all set to zero, and when an abnormality occurs in the pressure sensor 47, only the correct answer label yt ₄ is set to 1. At the same time, the remaining correct answer labels yt ₁ , yt _2, yt _3, yt _5, yt _6, yt _7, yt _{8 and} yt ₉ are all set to zero, and when the normal position sticking abnormality of the flow path switching valve 42 occurs. , Only the correct answer label yt ₅ is set to 1, and the remaining correct answer labels yt ₁ , yt _2, yt _3, yt ₄ , yt _6, yt _7, yt _{8 and} yt ₉ are all set to zero, and the flow path switching valve. When the test position sticking abnormality of 42 occurs, only the correct answer label yt ₆ is set to 1, and the remaining correct answer labels yt ₁ , yt _2, yt _3, yt ₄ , yt _5, yt _7, yt _8, yt. ₉ are all set to zero, and when the operation continuation abnormality of the suction pump 40 occurs, only the correct answer label yt ₇ is set to 1, and the remaining correct answer labels yt ₁ , yt _2, yt _3, yt ₄ , yt _{5 ,} Yt _6, yt _{8 and} yt ₉ are all set to zero, and when the suction pump 40 is abnormally stopped, only the correct answer label yt ₈ is set to 1 and the remaining correct answer labels yt ₁ , yt _2, yt _3, yt ₄ , yt _5, yt _6, yt _7, yt ₉ are all zero, and when normal, only the correct label yt ₉ is set to 1, and the remaining correct labels yt ₁ , yt _2, yt _3, yt ₄ , yt _5, yt _6, yt _7, yt ₈ are all set to zero.

On the other hand, as shown in FIG. 16, in this training data set, the input values x ₁ , x _{2 ...} x _n-1 , x _n and the input values xx ₁ , xx _{2 ... x} x _k-1 , M pieces of data representing the relationship between xx _k and the correct label yt have been acquired. For example, in the second data (No. 2), the acquired input values x ₁₂ , x _{22 ...} x _n-12 , x _n2 and the input values xx ₁₂ , xx _{22 ...} xx _k-12 , Xx _k2 and correct labels yt _{12 ...} yt _S2 are listed, and in the m-1st data (No. m-1), the input value x _{1 m-1} of the acquired input parameter, x _{2m-1 ...} x _n-1m-1 , x _nm-1 , input values xx _1m-1 , xx _{2m-1 ...} xx _k-1m-1 , xx _km-1 , and correct label yt _Sm-1 is listed.

This training data set is also created by a method similar to the method already described with reference to FIG. That is, also in this case, the purge control valve 14, the flow path switching valve 42, and the suction pump are used to estimate whether or not an abnormality has occurred in the fuel vapor emission prevention system by the test control device 81 shown in FIG. An abnormality detection process is performed in which the 40 is operated according to a predetermined operation order. At this time, the state of the fuel steam discharge prevention system is a state in which a small hole is opened in the pipe wall of the fuel steam flow pipe 12 or 13, and a valve opening abnormality in which the purge control valve 14 continues to open. , The purge control valve 14 keeps closing, a valve closing abnormality occurs, the pressure sensor 47 has an abnormality, and the flow path switching valve 42 has a normal position fixing abnormality. , The state where the test position fixing abnormality of the flow path switching valve 42 has occurred, the state where the operation continuation abnormality of the suction pump 40 has occurred, the state where the operation stop abnormality of the suction pump 40 has occurred, and the normal state have been sequentially changed. , In each of the changed states, the abnormality detection process is repeatedly performed while sequentially changing the combination of the atmospheric pressure, the remaining amount of fuel in the fuel tank 5, the fuel temperature in the fuel tank 5, and the flow rate characteristic value of the suction pump 40. Will be done.

While this anomaly detection process is taking place, the data needed to create the training dataset is acquired. FIG. 35 shows an abnormality detection processing routine executed in the test control device 81 to perform this abnormality detection processing. This abnormality detection processing routine is executed by an interrupt at every Δt for a certain period of time shown in FIG. 24. In the routine shown in FIG. 35, t represents a time obtained by setting the time t ₀ when the operation of the engine is stopped in FIG. 24 as zero and starting from this time t ₀ .

Referring to FIG. 35, first, in step 500, the system internal pressure _xn detected by the pressure sensor 47 is acquired and stored. That is, as shown in FIG. 24, the system internal pressure _xn is acquired every Δt for a fixed time, and the system internal pressure _xn acquired every Δt for a fixed time is stored in the test control device 81. Next, in step 501, it is determined whether or not the time t is the time t ₁ shown in FIG. 24. When the time t is not the time t ₁ shown in FIG. 24, the process proceeds to step 502, and it is determined whether or not the time t is the time t ₂ shown in FIG. 24. If the time t is not the time t ₂ shown in FIG. 24, the process proceeds to step 503, and it is determined whether or not the time t is the time t ₄ shown in FIG. 24. If the time t is not the time t ₄ shown in FIG. 24, the process proceeds to step 504, and it is determined whether or not the time t is the time t ₅ shown in FIG. 24. Time t, the processing cycle is ended when it is not the time t ₅ shown in FIG. 24.

On the other hand, when it is determined in step 501 that the time t is the time t ₁ shown in FIG. 24, the process proceeds to step 505 and an operation command for the suction pump 40 is issued. Then, the processing cycle is terminated. Further, in step 502, when it is determined that the time t is the time t ₂ shown in FIG. 24, the process proceeds to step 506, and a switching command for switching the flow path switching valve 42 to the test position is issued. Then, the processing cycle is terminated. Further, in step 503, when it is determined that the time t is the time t ₄ shown in FIG. 24, the process proceeds to step 507, and a valve opening command for opening the purge control valve 14 is issued. Then, the processing cycle is terminated.

On the other hand, in step 504, when it is determined that the time t is the time t ₅ shown in FIG. 24, the process proceeds to step 508, and a valve closing command for closing the purge control valve 14 is issued. Next, in step 509, a switching command for switching the flow path switching valve 42 to the normal position is issued. Then, in step 510, a stop command for the suction pump 40 is issued. Then, in step 511, the acquisition and storage of the system internal pressure _xn detected by the pressure sensor 47 is stopped. Then, in step 512, the abnormality detection process is completed. In the test control device 81 shown in FIG. 17, when the abnormality detection process is completed, the next abnormality detection process is started.

In this way, a state in which a small hole is formed in the pipe wall of the fuel steam flow pipe 12 or 13 is generated, a state in which the purge control valve 14 continues to open, a state in which a valve opening abnormality occurs, and a purge A state in which the control valve 14 continues to close, a state in which a valve closing abnormality occurs, a state in which an abnormality occurs in the pressure sensor 47, a state in which a normal position fixing abnormality occurs in the flow path switching valve 42, a state in which the flow path switching valve 42 Atmospheric pressure, fuel tank 5 in each state of test position sticking abnormality, suction pump 40 operation continuation abnormality, suction pump 40 operation stop abnormality, and normal state. The system internal pressure _{xn at} every Δt for a certain period of time when the combination of the remaining amount of fuel in the fuel tank 5, the fuel temperature in the fuel tank 5, and the flow rate characteristic value of the suction pump 40 is changed is stored in the test control device 81. That is, the No. 1 of the training data set shown in FIG. 1 to No. Input values of _m x _{1 m} , x _{2 m ...} x _nm-1 , x _nm , input values xx _{1 m} , xx _{2 m ...} xx _km-1 , xx _km , and correct label yt _sm (m = 1, 2) , 3 ... m) are stored in the test control device 81.

When the training data set is created in this way, the weights of the neural network 72 shown in FIG. 33 are trained using the electronic data of the created training data set. In this case, the learning of the weight of the neural network 72 shown in FIG. 33 is also performed by the learning device 82 shown in FIG. 17 using the learning processing routine of the weight of the neural network shown in FIG. When the learning of the weight of the neural network 72 is completed, the learned weight of the neural network 72 is stored in the memory 84 of the learning device 82. In this way, an abnormality determination estimation model capable of estimating whether or not an abnormality has occurred in the fuel vapor emission prevention system is created.

Also in this embodiment, the failure diagnosis of the fuel vapor emission prevention system in the commercial vehicle is performed using the abnormality determination estimation model of the fuel vapor emission prevention system created in this way. Therefore, the fuel vapor emission prevention system of this embodiment is also used. The abnormality determination estimation model is stored in the electronic control unit 20 of a commercial vehicle by using the data reading routine to the electronic control unit shown in FIG. That is, referring to FIG. 20, first, in step 300, the number of nodes in the input layer (L = 1) of the neural network 72 shown in FIG. 33, the hidden layer (L = 2), and the hidden layer (L = 3). ) And the number of nodes in the output layer (L = 4) are read into the memory 22 of the electronic control unit 20, and then, in step 301, a neural network 73 as shown in FIG. Created. As can be seen from FIG. 36, in this neural network 73, the softmax layer is removed. In this case, the neural network 73 may be provided with the softmax layer SM as shown in FIG. 33. Next, in step 302, the learned weight of the neural network 72 is read into the memory 22 of the electronic control unit 20. As a result, the abnormality determination estimation model of the fuel vapor emission prevention system is stored in the electronic control unit 20 of the commercial vehicle.

Also in this embodiment, the routine shown in FIG. 22 is used as the abnormality detection routine of the fuel vapor emission prevention system executed in the commercial vehicle. Even when the routine shown in FIG. 22 is used in this embodiment, the contents of steps 400 to 406 are the same as those described above, and therefore the description of steps 400 to 406 will be omitted. On the other hand, since step 407 and subsequent steps are slightly different from the contents previously described, only step 407 and subsequent steps will be briefly described.

That is, in step 407, the system internal pressures x _n x ₁ , x _{2 ...} x _n-1 , x _n and the input values xx ₁ , xx _{2 ... x} x _k-1 , xx _k are set every Δt for a certain period of time. , Is input to each node of the input layer (L = 1) of the neural network 73 shown in FIG. At this time, from each node of the output layer of the neural network 73, the output values y ₁ ', y ₂ ', y ₃ ', y ₄ ', y ₅ ', y ₆ ', y ₇ ', y ₈ ', y ₉ 'Is output, thereby in step 408, output values y ₁ ', y ₂ ', y ₃ ', y ₄ ', y ₅ ', y ₆ ', y ₇ ', y ₈ ', y ₉ '. Is obtained.

Then, in step 409, the maximum of the acquired output values y ₁ ', y ₂ ', y ₃ ', y ₄ ', y ₅ ', y ₆ ', y ₇ ', y ₈ ', y ₉ ' The output value y _i'is selected. At this time, it is presumed that the abnormal state shown in FIG. 34 corresponding to this maximum output value y _i'has occurred. Therefore, in step 410, it is determined that the abnormal state shown in FIG. 34 corresponding to the maximum output value y _i'has occurred, and for example, it is shown in FIG. 34 corresponding to the maximum output value y _i '. A warning light is lit to indicate that an abnormal condition has occurred. Then, in step 411, the abnormality detection is completed.

As described above, in the abnormality detection device of the fuel vapor emission prevention system according to another embodiment of the present invention, the fuel vapor emission prevention system is formed with the fuel steam chamber 10 and the atmospheric pressure chamber 11 on both sides of the activated coal layer 9, respectively. The canister 6 is provided, and the fuel steam chamber 10 communicates with the internal space above the fuel liquid level of the fuel tank 5 on the one hand and into the intake passage of the engine via the purge control valve 14 on the other hand. Has been done. Further, the fuel vapor discharge prevention system includes a flow path switching valve 42 capable of selectively connecting the atmospheric pressure chamber 11 to the atmosphere and the suction pump 40, and a passage from the flow path switching valve 42 to the atmospheric pressure chamber 11. 43 and the suction passage 46 from the flow path switching valve 42 to the suction pump 40 are connected by a reference pressure detection passage 50 having a throttle portion 51, and in the suction passage 46 from the flow path switching valve 42 to the suction pump 40. The pressure sensor 47 is arranged. While maintaining the valve closing command to close the purge control valve 14 when the vehicle is stopped and the switching position of the flow path switching valve 42 to the switching position where the atmospheric pressure chamber 11 is connected to the atmosphere after the vehicle is stopped. When a preset time elapses, a pump operation command for operating the suction pump 40 to make the inside of the fuel tank 5 and the inside of the canister 6 negative pressure, and after the pump operation command is generated, the flow path switching valve 42 An abnormality detection process is performed in which a switching command for switching the switching position to the switching position where the atmospheric pressure chamber 11 is connected to the suction pump 40 and a valve opening command for opening the purge control valve 14 are generated after the switching command is generated. .. When this abnormality detection process is being performed, the pressures in the fuel tank 5 and the canister 6 detected by the pressure sensor 47 at regular time intervals are stored in the storage device, and are stored in the storage device at regular time intervals. The pressure in the fuel tank 5 and the canister 6 and at least atmospheric pressure when the abnormality detection process is performed are set as the input parameters of the neural network, and the correct answer label is when there is a hole in the above system to leak fuel vapor. As, the trained neural network in which the weight has been learned is stored, and when the vehicle is stopped, this trained neural network is used to detect a perforated abnormality that leaks fuel vapor from the above input parameters. Will be done.

In this case, in the embodiment according to the present invention, the pressure in the fuel tank 5 and the canister 6 stored in the above-mentioned storage device at regular intervals and at least the atmospheric pressure when the abnormality detection process is performed are input to the neural network. As a parameter, the purge control valve 14 keeps opening when there is a hole in the fuel steam emission prevention system to leak fuel steam. When a valve opening abnormality occurs, the purge control valve 14 keeps closing. When a valve abnormality has occurred, when the pressure sensor 47 has an abnormality, or when a switching abnormality has occurred in which the switching position of the flow path switching valve 42 is maintained at the switching position that connects the atmospheric pressure chamber 11 to the atmosphere. , When the switching position of the flow path switching valve 42 is maintained at the switching position where the atmospheric pressure chamber 11 is connected to the suction pump 40, when there is a switching abnormality in which the suction pump 40 continues to operate, and A trained neural network in which weight learning has been performed is stored with each time when an abnormality in which the suction pump 40 continues to stop occurs as a correct answer label, and this trained neural network is used when the vehicle is stopped. , From the above input parameters, perforated abnormality that leaks fuel vapor, valve opening abnormality of purge control valve, valve closing abnormality of purge control valve, pressure sensor abnormality, flow path switching valve switching abnormality, and suction pump abnormality. Is detected.

Further, in this case, in the embodiment according to the present invention, the above-mentioned input parameters are the pressure in the fuel tank 5 and the canister 6 stored in the storage device at regular intervals, and the atmospheric pressure when the abnormality detection process is performed. And the remaining amount of fuel in the fuel tank 5 when the abnormality detection process is performed. Further, in this case, in the embodiment according to the present invention, the above-mentioned input parameters are the pressure in the fuel tank 5 and the canister 6 stored in the storage device at regular intervals, and the atmospheric pressure when the abnormality detection process is performed. It is composed of an index showing the remaining amount of fuel in the fuel tank 5 when the abnormality detection process is performed, the temperature of the fuel in the fuel tank 5, and the capacity of the suction pump 40.

1 Engine body 2 Surge tank 5 Fuel tank 6 Canister 9 Activated coal layer 10 Fuel steam chamber 11 Atmospheric pressure chamber 12, 13 Fuel steam flow pipe 14 Purge control valve 16 Suction pump module 20 Electronic control unit 42 Flow path switching valve 40 Suction pump 47 Pressure sensor 50 Reference pressure detection passage

Claims (8)

It is equipped with canisters in which a fuel steam chamber and an atmospheric pressure chamber are formed on both sides of the activated coal layer, and the fuel steam chamber communicates with the internal space above the fuel liquid level of the fuel tank on the one hand and on the other hand. Detects pressure in the fuel tank and canister, as well as a flow path switching valve that is connected to the intake passage of the engine via a purge control valve and can selectively connect the atmospheric pressure chamber to the atmosphere and the suction pump. In the abnormality detection device of the fuel vapor discharge prevention system equipped with the pressure sensor, the valve closing command for closing the purge control valve and the switching position of the flow path switching valve are set to the atmospheric pressure chamber when the vehicle is stopped. Abnormality detection processing is performed to generate a switching command to switch to the switching position connected to the suction pump and a pump operation command to operate the suction pump to make the inside of the fuel tank and the inside of the canister negative pressure. During the processing, the pressure in the fuel tank and the canister detected by the pressure sensor at regular intervals is stored in the storage device, and the pressure in the fuel tank and in the fuel tank at regular intervals stored in the storage device is stored. Weight learning is performed using the pressure inside the canister and at least atmospheric pressure when the anomaly detection process is performed as the input parameters of the neural network, and the correct label when there is a hole in the system that leaks fuel vapor. The trained neural network that has been performed is stored, and when the vehicle is stopped, the trained neural network is used to detect a perforated abnormality that leaks fuel steam from the above input parameters. Abnormality detection device. The abnormality detection process includes a process of generating a valve opening command for opening the purge control valve after the valve closing command of the purge control valve is generated, and the fuel stored in the storage device at regular intervals. The pressure inside the tank and the canister and at least the atmospheric pressure at the time when the abnormality detection process is performed are used as the input parameters of the neural network, and the purge control valve opens when there is a hole in the system that leaks fuel vapor. The trained neural network in which the weight has been learned is stored with the correct label when the valve opening abnormality that keeps valve is occurring and when the purge control valve keeps closing the valve closing abnormality. A request to detect a perforated abnormality that leaks fuel vapor, a valve opening abnormality of a purge control valve, and a valve closing abnormality of a purge control valve from the above input parameters using the learned neural network when the vehicle is stopped. Item 1. The abnormality detection device for the fuel vapor emission prevention system according to Item 1. The input parameters are the pressure in the fuel tank and the canister at regular intervals stored in the storage device, the atmospheric pressure when the abnormality detection process is performed, and the fuel when the abnormality detection process is performed. The abnormality detection device for the fuel vapor emission prevention system according to claim 1, which comprises the remaining amount of fuel in the tank. The input parameters are the pressure in the fuel tank and the canister at regular intervals stored in the storage device, the atmospheric pressure when the abnormality detection process is performed, and the fuel when the abnormality detection process is performed. The abnormality detection device for a fuel vapor emission prevention system according to claim 1, which comprises an index indicating the remaining amount of fuel in the tank, the temperature of the fuel in the fuel tank, and the capacity of the suction pump. It is equipped with canisters in which a fuel steam chamber and an atmospheric pressure chamber are formed on both sides of the activated coal layer, and the fuel steam chamber communicates with the internal space above the fuel liquid level of the fuel tank on one side and on the other side. Is connected to the intake passage of the engine via a purge control valve, and is equipped with a flow path switching valve that can selectively connect the atmospheric pressure chamber to the atmosphere and the suction pump. The passage to the pressure chamber and the suction passage from the flow path switching valve to the suction pump are connected by a reference pressure detection passage having a throttle portion, and the pressure sensor is arranged in the suction passage from the flow path switching valve to the suction pump. In the abnormality detection device of the fuel steam emission prevention system, the valve closing command for closing the purge control valve when the vehicle is stopped and the switching position of the flow path switching valve after the vehicle is stopped are set to the atmospheric pressure. A pump operation command and a pump operation command to operate a suction pump to create negative pressure in the fuel tank and canister when a preset time elapses while maintaining the chamber in a switching position connected to the atmosphere. A switching command for switching the switching position of the flow path switching valve to a switching position where the atmospheric pressure chamber is connected to the suction pump, and a valve opening command for opening the purge control valve after the switching command is generated. When the abnormality detection process is being performed, the pressure in the fuel tank and the canister detected by the pressure sensor at regular intervals is stored in the storage device, and the storage device stores the pressure. The pressure in the fuel tank and canister at regular intervals stored in the device and at least atmospheric pressure when the abnormality detection process is performed are used as input parameters of the neural network, and there is a hole in the system to leak fuel vapor. A trained neural network in which weights have been trained is stored with the time when it occurs as a correct answer label, and when the vehicle is stopped, the trained neural network is used to leak fuel vapor from the above input parameters. Abnormality detection device for fuel vapor emission prevention system that detects abnormalities with holes. A hole for leaking fuel vapor to the system by using the pressure in the fuel tank and the canister at regular intervals stored in the storage device and at least the atmospheric pressure at the time when the abnormality detection process is performed as input parameters of the neural network. When there is a gap, when the purge control valve keeps opening, when there is a valve opening abnormality, when the purge control valve keeps closing, when there is a valve closing abnormality, or when the pressure sensor has an abnormality. , The switching position of the flow path switching valve is maintained at the switching position that connects the atmospheric pressure chamber to the atmosphere. When a switching abnormality occurs, the switching position of the flow path switching valve connects the atmospheric pressure chamber to the suction pump. Weight learning is performed using the correct label when there is a switching abnormality that is maintained at the switching position, when there is an abnormality in which the suction pump continues to operate, and when there is an abnormality in which the suction pump continues to stop. The trained neural network that has been performed is stored, and when the vehicle is stopped, the trained neural network is used to obtain a perforated abnormality that leaks fuel vapor from the above input parameters, a valve opening abnormality of the purge control valve, and the like. The abnormality detection device for a fuel vapor discharge prevention system according to claim 5, which detects an abnormality in the purge control valve, an abnormality in the pressure sensor, an abnormality in switching the flow path switching valve, and an abnormality in the suction pump. The input parameters are the pressure in the fuel tank and the canister at regular intervals stored in the storage device, the atmospheric pressure when the abnormality detection process is performed, and the fuel when the abnormality detection process is performed. The abnormality detection device for the fuel vapor emission prevention system according to claim 5, which comprises the remaining amount of fuel in the tank. The input parameters are the pressure in the fuel tank and the canister at regular intervals stored in the storage device, the atmospheric pressure when the abnormality detection process is performed, and the fuel when the abnormality detection process is performed. The abnormality detection device for a fuel vapor emission prevention system according to claim 5, which comprises an index indicating the remaining amount of fuel in the tank, the temperature of the fuel in the fuel tank, and the capacity of the suction pump.

Images