JP6673519B1 - Internal combustion engine state detection system, data analysis device, and vehicle - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a state detection system for an internal combustion engine in which the detection accuracy of a predetermined state such as misfire or variation in air-fuel ratio among cylinders is improved. A control device 70 is a state detection system for an internal combustion engine, which is applied to an internal combustion engine 10 mounted on a vehicle and detects a predetermined operating state of the internal combustion engine 10 with variations in combustion states among cylinders. . The control device 70 includes a CPU 72, which is an execution device, and a storage device 76. The storage device 76 stores mapping data 76a that is data that defines a detection mapping that outputs the value of the combustion state variable by combining the rotational waveform variable and the road surface state variable based on the parameters learned by machine learning. . The CPU 72 executes an acquisition process for acquiring the values of the rotation waveform variable and the road surface state variable, and the internal combustion engine 10 has a predetermined operating state based on the output value of the detection map with the respective values of the rotation waveform variable and the road surface state variable as inputs. Determination processing for determining whether or not [Selection diagram] Fig. 1

Description

  The present invention relates to a state detection system for an internal combustion engine mounted on a vehicle for detecting an operating state accompanied by a variation in combustion state between cylinders, for example, a misfire or an air-fuel ratio between cylinders, and a state detection system for the same. The present invention relates to a data analysis device and a vehicle constituting a system.

  In the internal combustion engine, when the combustion state among the cylinders varies due to misfire, a variation in the air-fuel ratio between the cylinders, or the like, the rotational fluctuation of the crankshaft increases. Therefore, it is possible to detect a misfire, a variation in the air-fuel ratio between cylinders, and the like, based on the rotation fluctuation pattern of the crankshaft. For example, Patent Literature 1 discloses a hierarchy configured to input time-series data of the rotation speed of a crankshaft of an internal combustion engine sampled at a predetermined cycle to an input layer and output information on a misfired cylinder from an output layer. A misfire detection system for detecting misfire using a neural network model is described.

JP-A-4-91348

  By the way, when the road surface has irregularities, vibration occurs in the vehicle, and this vibration is transmitted to the crankshaft. That is, vibration from the road surface is superimposed on the rotational behavior of the crankshaft. Therefore, in a running vehicle, the detection accuracy of misfire or the like based on the pattern of the rotation fluctuation of the crankshaft decreases.

Hereinafter, means for solving the above-described problems and the effects thereof will be described.
An internal combustion engine state detection system for solving the above-mentioned problem is applied to an internal combustion engine mounted on a vehicle, and detects a predetermined operating state of the internal combustion engine with variation in combustion state between cylinders. It is a state detection system. The state detection system includes a rotation waveform variable that is a variable including information on a difference between cylinders of a rotation speed of a crankshaft of the internal combustion engine during a generation period of a combustion torque of each cylinder, and a state of a road surface on which the vehicle is traveling. A map in which a road surface state variable, which is a variable related to the input, and a combustion state variable, which is a variable relating to the degree of variation in the combustion state between the cylinders, are output, and the rotational waveform variable and the rotation waveform variable based on a parameter learned by machine learning. When a mapping that outputs the value of the combustion state variable by a joint operation of road surface state variables is used as a detection mapping, a storage device that stores mapping data that is data that defines the detection mapping, and a rotational behavior of the crankshaft. Acquire the value of the rotation waveform variable based on the output of the sensor that detects The internal combustion engine obtains the value of the road surface state variable, and the internal combustion engine based on the output value of the detection map that receives the respective values of the rotational waveform variable and the road surface state variable obtained through the obtaining process. An execution device for executing a determination process of determining whether or not the vehicle is in the predetermined operation state.

  If the combustion state of each cylinder varies, a difference occurs in the combustion torque of each cylinder. Therefore, there is a difference between cylinders in the rotation speed of the crankshaft during the period in which the combustion torque of each cylinder is generated. Such information on the difference in the rotational speed of the crankshaft between the cylinders can be obtained from the output of a sensor that detects the rotational behavior of the crankshaft. If the difference in the rotational speed of the crankshaft between the cylinders during the period in which the combustion torque occurs is simply caused by the variation in the combustion state of each cylinder, the misfire or the cylinder-to-cylinder It is possible to accurately detect the operating state of the internal combustion engine accompanied by the variation in the combustion state between the cylinders, such as the air-fuel ratio imbalance. On the other hand, in the above-described configuration, in consideration of the fact that the variation in the combustion state between the cylinders causes a difference between the cylinders in the rotation speed of the crankshaft during the period during which the combustion torque is generated, the rotation including the information on the difference between the cylinders at the same rotation speed Waveform variables are included in the input to the detection mapping.

  Also, in an internal combustion engine mounted on a vehicle, when the road surface on which the vehicle is running has irregularities, the vehicle generates vibration, and this vibration is transmitted to the crankshaft. Therefore, the state of the road surface affects the rotation behavior of the crankshaft. On the other hand, in the above configuration, the road surface state variable relating to the road surface state is also included in the input of the detection mapping. In the above configuration, the presence or absence of the above-mentioned predetermined operating state is determined based on the value of the combustion state variable calculated by the combined operation of the rotation waveform variable based on the parameter learned by machine learning and the road surface state variable. . The parameters here can be learned based on whether or not the above-mentioned predetermined operating state has occurred when each of the above-described rotational waveform variables and road surface state variables takes various values. Therefore, according to the above configuration, it is possible to calculate the combustion state variable in consideration of the influence of the state of the road on which the vehicle is traveling. Therefore, according to the above configuration, the detection accuracy of the predetermined operating state in the internal combustion engine mounted on the vehicle can be improved.

  The above-described state detection system can be configured as a system that detects a state where a misfire has occurred. Further, the present invention can be configured as a system that detects a state in which the air-fuel ratio varies between cylinders.

Further, for example, as the road surface state variable in the above-described state detection system, a variable indicating whether the road surface on which the vehicle is traveling is a paved road or an unpaved road can be used.
In one aspect of the state detection system, when the execution device determines that the internal combustion engine is in the predetermined operation state by the determination processing, the execution device operates hardware to switch to the predetermined operation state. Execute the coping process for coping. As the coping process, for example, a process of notifying the occupant that the internal combustion engine is in a predetermined operation state, or a process of eliminating or reducing the predetermined operation state may be performed.

  In one aspect of the state detection system, the determination process calculates the combustion state variable which is an output value of the detection mapping to which the values of the rotational waveform variable and the road surface state variable acquired through the acquisition process are input. The execution device includes a first execution device mounted on the vehicle and a second execution device mounted on a device different from the vehicle. Then, the first execution device executes the acquisition process and a vehicle-side reception process of receiving a signal based on a calculation result of the output value calculation process. The second execution device executes the output value calculation process and an external transmission process of transmitting a signal based on a calculation result of the output value calculation process to the vehicle.

  In such a case, the output value calculation processing with a high calculation load is performed by a device different from the vehicle. Therefore, the calculation load on the vehicle can be reduced. As a component of the state detection system in such a case, a data analysis device including the second execution device and the storage device and a vehicle including the first execution device can be considered.

FIG. 1 is a diagram illustrating a configuration of a drive system of a vehicle equipped with a state detection system and an internal combustion engine to which the first embodiment is applied. 5 is a flowchart showing a procedure of a misfire detection process in the embodiment. 5 is a flowchart showing a procedure of a road surface state variable calculation process in the embodiment. The figure which shows the structure of the state detection system of 2nd Embodiment. 9A and 9B are flowcharts illustrating a procedure of a process executed by the state detection system according to the second embodiment. 9 is a flowchart illustrating a procedure of a misfire detection process according to the third embodiment. 13 is a time chart showing mapping input variables in the third embodiment. 13A and 13B are flowcharts illustrating a procedure of a process executed by the state detection system according to the fourth embodiment. 15 is a flowchart illustrating a procedure of an imbalance detection process according to the fifth embodiment. 15 is a flowchart illustrating a procedure of an imbalance handling process according to the fifth embodiment. (A) And (b) is a flowchart which shows the procedure of the process which the state detection system of 6th Embodiment performs.

(1st Embodiment)
Hereinafter, a first embodiment of a state detection system for an internal combustion engine will be described with reference to FIGS.

  In the internal combustion engine 10 mounted on the vehicle VC shown in FIG. 1, a throttle valve 14 is provided in the intake passage 12. The air sucked from the intake passage 12 flows into the combustion chambers 18 of the cylinders # 1 to # 4 when the intake valve 16 opens. The internal combustion engine 10 is provided with a fuel injection valve 20 that injects fuel so as to be exposed to the combustion chamber 18 and an ignition device 22 that generates spark discharge. In the combustion chamber 18, a mixture of air and fuel is provided for combustion, and energy generated by the combustion is taken out as rotational energy of the crankshaft 24. The air-fuel mixture supplied to the combustion is discharged to the exhaust passage 28 as exhaust gas with the opening of the exhaust valve 26. A catalyst 30 having oxygen storage capacity is provided in the exhaust passage 28. The exhaust passage 28 communicates with the intake passage 12 via an EGR passage 32. The EGR passage 32 is provided with an EGR valve 34 for adjusting the cross-sectional area of the passage.

  The rotational power of the crankshaft 24 is transmitted to the intake-side camshaft 42 via the intake-side variable valve timing device 40 and is transmitted to the exhaust-side camshaft 46 via the exhaust-side variable valve timing device 44. The variable intake-side valve timing device 40 changes the relative rotational phase difference between the intake-side camshaft 42 and the crankshaft 24. The exhaust-side variable valve timing device 44 changes a relative rotational phase difference between the exhaust-side camshaft 46 and the crankshaft 24.

  An input shaft 66 of a transmission 64 can be connected to the crankshaft 24 of the internal combustion engine 10 via a torque converter 60. The torque converter 60 includes a lock-up clutch 62. When the lock-up clutch 62 is engaged, the crankshaft 24 and the input shaft 66 are connected. A drive wheel 69 is mechanically connected to an output shaft 68 of the transmission 64. In the present embodiment, the transmission 64 is a stepped transmission that can change the speed ratio from the first speed to the sixth speed.

  The crankshaft 24 is coupled to a crank rotor 50 provided with a tooth portion 52 indicating each of a plurality of rotation angles (here, 34 rotation angles) of the crankshaft 24. Basically, the crank rotor 50 is provided with the tooth portions 52 at intervals of 10 ° CA, but has one missing tooth portion 54 where the interval between adjacent tooth portions 52 is 30 ° CA. Is provided. This is for indicating the reference rotation angle of the crankshaft 24.

  The control device 70 controls the internal combustion engine 10 and controls a throttle valve 14, a fuel injection valve 20, an ignition device 22, an EGR valve 34, an intake side The variable valve timing device 40 and the variable exhaust valve timing device 44 are operated. FIG. 1 shows operation signals MS1 to MS6 of the throttle valve 14, the fuel injection valve 20, the ignition device 22, the EGR valve 34, the intake-side valve timing variable device 40, and the exhaust-side valve timing variable device 44, respectively. ing.

  In controlling the control amount, the control device 70 outputs an output signal Scr of the crank angle sensor 80 that outputs a pulse at every 10 ° CA except for the angular interval between the tooth portions 52, that is, excluding the toothless portion 54, and an air flow meter 82. And the upstream detection value Afu which is a detection value of the air-fuel ratio sensor 83 provided on the upstream side of the catalyst 30. The control device 70 detects the coolant temperature THW, which is the temperature of the cooling water of the internal combustion engine 10 detected by the coolant temperature sensor 84, the shift position Sft of the transmission 64 detected by the shift position sensor 86, and the acceleration sensor 88. The vertical acceleration Dacc of the vehicle VC is referred to.

  The control device 70 includes a CPU 72, a ROM 74, a storage device 76, which is an electrically rewritable nonvolatile memory, and a peripheral circuit 77, which can be communicated by a local network 78. Note that the peripheral circuit 77 includes a circuit that generates a clock signal that defines an internal operation, a power supply circuit, a reset circuit, and the like.

  The control device 70 controls the control amount by causing the CPU 72 to execute a program stored in the ROM 74. Further, the control device 70 executes a process of determining whether or not the internal combustion engine 10 has misfired. That is, in the present embodiment, the operating state in which a misfire has occurred is the default operating state, and the control device 70 that detects the presence or absence of such a misfire corresponds to a state detection system.

  FIG. 2 shows a procedure of a misfire detection process for detecting the presence or absence of a misfire. The misfire detection process shown in FIG. 2 is realized by the CPU 72 repeatedly executing a misfire detection program, which is the program 74a stored in the ROM 74 shown in FIG. 1, at a predetermined cycle, for example. In the following, a step number of each process is represented by a number preceded by “S”.

  In the series of processes shown in FIG. 2, the CPU 72 first acquires the minute rotation times T30 (1), T30 (2),... T30 (24) (S10). The minute rotation time T30 is calculated by the CPU 72 measuring the time required for the crankshaft 24 to rotate by 30 ° CA based on the output signal Scr of the crank angle sensor 80. Here, when the numbers in parentheses such as the minute rotation times T30 (1) and T30 (2) are different, it indicates that the rotation angle intervals are different within 720 ° CA which is one combustion cycle. That is, the minute rotation times T30 (1) to T30 (24) indicate the rotation time at each angular interval obtained by equally dividing the rotation angle area of 720 ° CA by 30 ° CA.

  More specifically, the CPU 72 counts the time of rotation by 30 ° CA based on the output signal Scr, and sets the time as the pre-filter processing time NF30. Next, the CPU 72 calculates a post-filter processing time AF30 by performing digital filter processing using the pre-filter processing time NF30 as an input. Then, the CPU 72 normalizes the post-filtering time AF30 such that the difference between the maximum value and the minimum value of the post-filtering time AF30 at 720 ° CA, for example, at 720 ° CA becomes “1”, thereby obtaining the minute rotation time T30. Is calculated.

  Next, the CPU 72 acquires the rotation speed NE and the charging efficiency η, the ignition timing aig, the target value Af * of the air-fuel ratio, the EGR rate Regr, the water temperature THW, the shift position Sft, and the road surface state variable SR (S12a).

  Here, the rotation speed NE is calculated by the CPU 72 based on the output signal Scr of the crank angle sensor 80, and the charging efficiency η is calculated by the CPU 72 based on the rotation speed NE and the intake air amount Ga. The rotation speed NE is the average value of the rotation speed of the crankshaft 24 when the compression top dead center appears, that is, in the present embodiment, the crankshaft 24 rotates by an angle interval larger than 180 ° CA. It is desirable that the rotation speed NE be an average value of the rotation speed when the crankshaft 24 rotates by a rotation angle equal to or more than one rotation of the crankshaft 24. Here, the average value is not limited to the simple average, and may be, for example, an exponential moving average process, and may be a plurality of sampling values such as a minute rotation time T30 when the crankshaft 24 rotates by one or more rotation angles. Shall be calculated by The charging efficiency η is a parameter that determines the amount of air charged into the combustion chamber 18. The EGR rate Regr is a ratio of the fluid flowing into the intake passage 12 from the EGR passage 32 to the fluid flowing into the intake passage 12. The road surface state variable SR is a parameter calculated by the CPU 72 based on the acceleration Dacc. For example, a variable that can take two different values depending on whether the road on which the vehicle VC is traveling has a lot of undulations and a small amount of undulations. And it is sufficient.

  In the present embodiment, when the road surface state variable SR is “0”, it indicates that the road surface on which the vehicle VC is traveling is a paved road, and when the road surface state variable SR is “1”, The road surface on which the vehicle VC is traveling is an unpaved road. Incidentally, the calculation processing of the road surface state variable SR is performed through the processing shown in FIG. The process illustrated in FIG. 3 is realized by the CPU 72 repeatedly executing the program stored in the ROM 74 at a predetermined cycle, for example.

  In the series of processes shown in FIG. 3, the CPU 72 first obtains the acceleration Dacc (S30). Next, the CPU 72 determines whether or not the road surface state variable SR is “1” (S32). When determining that the road surface state variable SR is “0” (S32: NO), the CPU 72 determines whether or not the state in which the absolute value of the variation ΔDacc of the acceleration Dacc is equal to or more than the predetermined value ΔDH has continued for a predetermined time. (S34). Here, the change amount ΔDacc may be a difference between the previous value and the current value of the acceleration Dacc. This process is a process for determining whether or not the road on which the vehicle VC is traveling is a paved road. When determining that the predetermined time has elapsed (S34: YES), the CPU 72 substitutes "1" for the road surface state variable SR and updates the road surface state variable SR (S36). Then, the CPU 72 once ends this series of processing.

  On the other hand, when the CPU 72 determines that the road surface state variable SR is “1” (S32: YES), the state in which the absolute value of the change amount ΔDacc of the acceleration Dacc becomes equal to or smaller than the specified value ΔDL (<ΔDH) for a predetermined time period It is determined whether or not it has been continued (S40). Then, when determining that the predetermined time has been maintained (S40: YES), the CPU 72 updates the road surface state variable SR by substituting “0” into the road surface state variable SR (S42). Then, the CPU 72 once ends this series of processing.

  When the CPU 72 makes a negative determination in the processing of S34 and S40 (S34: NO, S40: NO), the series of processing is temporarily ended without updating the road surface state variable SR.

  In S12a, when the rotation speed NE, the charging efficiency η, the ignition timing aig, the target value Af *, the EGR rate Regr, the water temperature THW, the shift position Sft, and the road surface state variable SR are acquired, as shown in FIG. The values obtained by the processing of S10 and S12a are substituted into input variables x (1) to x (32) of the mapping for calculating the probability of occurrence of (S14a). More specifically, the CPU 72 sets “s = 1 to 24” and substitutes the minute rotation time T30 (s) for the input variable x (s). That is, the input variables x (1) to x (24) are time-series data of the minute rotation time T30. In the present embodiment, the time series data of the minute rotation time T30 corresponds to a rotation waveform variable. Further, the CPU 72 substitutes the rotation speed NE into the input variable x (25) and substitutes the charging efficiency η into the input variable x (26). Also, the ignition timing aig, the target value Af *, the EGR rate Regr, the water temperature THW, the shift position Sft, and the road surface state variable SR are substituted for the input variables x (27) to x (32).

  Here, the ignition timing aig, the target value Af *, and the EGR rate Regr are adjustment variables for adjusting the combustion speed of the air-fuel mixture in the combustion chamber 18 by operating the operation unit of the internal combustion engine 10. That is, the ignition timing aig is an operation amount of the ignition device 22 as an operation unit, and is an adjustment variable for adjusting the combustion speed by operating the ignition device 22. The target value Af * is an adjustment variable for adjusting the combustion speed by operating the fuel injection valve 20 as an operation unit. The EGR rate Regr is an adjustment variable for adjusting the combustion speed by operating the EGR valve 34 as an operation unit. The reason why these adjustment variables are set as the input variables x is that the behavior of the crankshaft 24 changes depending on the combustion speed of the air-fuel mixture. This is in view of the fact that it tends to change depending on the combustion speed of the fuel.

  On the other hand, the water temperature THW is a state variable of the internal combustion engine 10, and particularly a parameter that changes the friction of the sliding portion, such as the friction between the piston and the cylinder. In view of the fact that the rotation behavior of the crankshaft 24 changes depending on the water temperature THW, the water temperature THW is set as the input variable x in the present embodiment.

  The shift position Sft is a state variable of the drive system connected to the crankshaft 24. Since the shift positions Sft differ from each other, since the inertia constants from the crankshaft 24 to the drive wheels 69 are different from each other, the shift behavior of the crankshaft 24 is different, so that the shift position Sft is set to the input variable x. I do.

  Also, if the road surface has irregularities, vibration occurs in the vehicle VC, and this vibration is transmitted to the crankshaft 24. For this reason, since the rotation behavior of the crankshaft 24 differs depending on the road surface state, the road surface state variable SR is set as the input variable x in the present embodiment.

  Next, the CPU 72 inputs the input variables x (1) to x (32) into the mapping defined by the mapping data 76a stored in the storage device 76 shown in FIG. The probability P (i) of misfire in 4) is calculated as a combustion state variable (S16a). The mapping data 76a defines a mapping that can output the probability P (i) of the misfire occurring in the cylinder #i during the period corresponding to the minute rotation times T30 (1) to T30 (24) acquired by the process of S10. Data. Here, the probability P (i) quantifies the plausibility of the fact that a misfire has actually occurred based on the input variables x (1) to x (32). However, in the present embodiment, the maximum value of the probability P (i) that misfire has occurred in the cylinder #i is smaller than “1”, and the minimum value is larger than “0”. That is, in the present embodiment, the probability P (i) is quantified as a continuous value within a predetermined area larger than “0” and smaller than “1”, indicating the degree of plausibility that an actual misfire has occurred. It was done.

  In the present embodiment, the mapping is such that the sum of the probabilities P (1) to P (4) at which a misfire has occurred is “1” by normalizing the output of the neural network and the neural network having one hidden layer. And a softmax function for The above-mentioned neural network performs an input-side conversion on each of an input-side coefficient wFjk (j = 0 to n, k = 0 to 32) and an output of an input-side linear mapping which is a linear mapping defined by the input-side coefficient wFjk. It includes the activation function h (x) as a non-linear mapping. In the present embodiment, a hyperbolic tangent “tanh (x)” is exemplified as the activation function h (x). Further, the neural network nonlinearly converts each of the output-side coefficient wSij (i = 1 to 4, j = 0 to n) and the output of the output-side linear mapping which is a linear mapping defined by the output-side coefficient wSij. An activation function f (x) as an output-side nonlinear mapping is included. In the present embodiment, the hyperbolic tangent “tanh (x)” is exemplified as the activation function f (x). The value n indicates the dimension of the intermediate layer. In the present embodiment, the value n is smaller than the dimension of the input variable x (here, 32 dimensions). The input-side coefficient wFj0 is a bias parameter, and is a coefficient of the input variable x (0) by defining the input variable x (0) as “1”. The output-side coefficient wSi0 is a bias parameter, which is multiplied by “1”. This can be realized, for example, by defining “wF00 · x (0) + wF01 · x (1) +.

  Specifically, the CPU 72 calculates a probability prototype y (i) which is an output of the neural network defined by the input coefficient wFjk, the output coefficient wSij, and the activation functions h (x) and f (x). The probability prototype y (i) is a parameter having a positive correlation with the probability of misfire in cylinder #i. Then, the CPU 72 calculates the probability P (i) that a misfire has occurred in the cylinder #i based on the output of the softmax function having the probability prototypes y (1) to y (4) as inputs. In the following description, such a mapping is referred to as a detection mapping.

  Next, the CPU 72 determines whether or not the maximum value P (m) among the probabilities P (1) to P (4) at which misfire has occurred is equal to or larger than the threshold value Pth (S18). Here, the variable m takes one of values from 1 to 4, and the threshold value Pth is set to a value equal to or more than “１ ／”. Then, when determining that the probability is equal to or greater than the threshold value Pth (S18: YES), the CPU 72 increments the number of misfires N (m) of the cylinder #m having the highest probability (S20). Then, the CPU 72 determines whether or not any of the numbers N (1) to N (4) exceeds a predetermined number Nth (S22). If the CPU 72 determines that there is an engine whose number is equal to or more than the predetermined number Nth (S22: YES), misfires with a frequency exceeding the allowable range in the specific cylinder #q (q is one of 1 to 4) are performed. It is determined that the error has occurred, and “1” is substituted for the fail flag F (S24). At this time, the CPU 72 stores the information of the cylinder #q in which the misfire has occurred in the storage device 76 and holds the information until at least the misfire is eliminated in the cylinder #q.

  On the other hand, when determining that the maximum value P (m) is less than the threshold value Pth (S18: NO), the CPU 72 determines whether or not a predetermined period has elapsed since the processing of S24 or the processing of S28 described below is performed. Is determined (S26). Here, the predetermined period is longer than the period of one combustion cycle, and desirably has a length of 10 times or more of one combustion cycle.

When determining that the predetermined period has elapsed (S26: YES), the CPU 72 initializes the number of times N (1) to N (4) and initializes the fail flag F (S28).
Note that the CPU 72 ends the series of processes shown in FIG. 2 once when the processes in S24 and S28 are completed or when a negative determination is made in the processes in S22 and S26.

  The control device 70 issues a warning shown in FIG. 1 in order to urge the user to deal with the abnormality when the fail flag F is “1” as a handling process for dealing with the default operation state. A notification process for operating the light 90 is executed.

  Incidentally, the mapping data 76a is generated as follows, for example. That is, the internal combustion engine 10 is operated with the dynamometer connected to the crankshaft 24 on the test bench, and the timing at which fuel required in each of the cylinders # 1 to # 4 should be injected is selected at random. Stop fuel injection. In a cylinder in which fuel injection is stopped, data in which the value of the probability P (i), which is a combustion state variable, is set to "1" is used as teacher data. In a cylinder in which fuel injection is not stopped, a combustion state variable is set. The data in which the value of the probability P (i) is “0” is included in the teacher data. Then, by using the training data that is a set of values of the input variables x (1) to x (32) generated by the processing of S14a, the probability P (i) that is the combustion state variable is obtained by the same processing as the processing of S16a. Is calculated. The values of the input-side coefficient wFjk and the output-side coefficient wSij are learned so as to reduce the difference between the value of the probability P (i) thus calculated and the teacher data. Specifically, for example, the values of the input side coefficient wFjk and the output side coefficient wSij may be learned so as to minimize the tolerance entropy.

  In the present embodiment, the training data includes all the parameters acquired in S12a, executes a process corresponding to S14a, and uses a 32-dimensional mapping of the input variable x. However, here, the input variable x (0) is not included in the number of dimensions. Also, the training data corresponding to the case where the road surface state variable SR indicates an unpaved road having a lot of undulations on the road surface simulates, for example, a load torque applied to the crankshaft 24 when vibration is applied to the vehicle by a dynamometer. And generate it. In addition, for example, the internal combustion engine 10 or a dynamometer is mounted on a device capable of generating vibration, and the vibration is generated to generate training data by simulating a load torque applied to the crankshaft 24 when vibration is applied to the vehicle. May be.

  As described above, in the present embodiment, the adjustment variable for adjusting the combustion speed, the state variable of the internal combustion engine 10, the state variable of the driving system, and the road surface state variable SR are added to the input variable x. These are parameters that affect the rotational behavior of the crankshaft 24, but it is difficult to take into account the conventional method of determining the presence or absence of misfire based on the difference between the times required for rotation at adjacent angular intervals. It is a thing. That is, when these parameters are taken into consideration, the number of adaptation steps is enormous and the feasibility is poor. On the other hand, in the present embodiment, by using machine learning, it is possible to construct a logic for judging misfire by taking these factors into consideration within the range of practical adaptation man-hours.

Here, the operation and effect of the present embodiment will be described.
When the internal combustion engine 10 is operating, the CPU 72 sequentially calculates the minute rotation time T30, and inputs the minute rotation time T30 for one combustion cycle to the detection map defined by the mapping data 76a as a rotation waveform variable, thereby obtaining a cylinder. The probabilities P (1) to P (4) of misfires are calculated in each of # 1 to # 4. Here, the minute rotation time T30 is a parameter indicating the rotation speed of the crankshaft 24 at an angular interval smaller than 180 ° CA, which is the interval at which the compression top dead center appears. In addition, a minute rotation time T30 for each 30 ° CA in one combustion cycle is input to the detection map. Further, the values of the input-side coefficient wFjk and the output-side coefficient wSij used for the operation performed on the minute rotation time T30 are values that have been learned by machine learning. Therefore, the probability P (i) of misfire can be calculated based on the rotation behavior of the crankshaft 24 on a minute time scale. For this reason, when compared with the case where the presence or absence of a misfire is determined based on the difference between adjacent angular intervals with respect to the time required for rotation about the appearance interval of the compression top dead center, the rotation behavior of the crankshaft 24 is more detailed. Since the presence / absence of a misfire can be determined based on appropriate information, the accuracy of the misfire determination can be easily increased.

According to the embodiment described above, the following functions and effects can be obtained.
(1) In the internal combustion engine 10 mounted on the vehicle VC, when the road surface on which the vehicle VC is traveling has irregularities, the vehicle VC generates vibration, and the vibration is transmitted to the crankshaft 24. Thus, the state of the road surface affects the rotational behavior of the crankshaft 24. On the other hand, in the control device 70, the road surface state variable SR relating to the road surface state is also included in the input of the detection mapping. Then, in the control device 70, the probability P (i, which is a combustion state variable calculated by a joint operation of T30 (1) to T30 (24), which are rotation waveform variables based on parameters learned by machine learning, and the road surface state variable SR, is calculated. ), It is determined whether or not the misfire has occurred. Therefore, the calculation of the combustion state variable can be performed in consideration of the influence of the state of the road surface on which the vehicle VC is traveling. Therefore, the accuracy of detecting misfire in the internal combustion engine 10 mounted on the vehicle VC can be improved.

  (2) The rotational speed NE and the charging efficiency η as operating point variables that define the operating point of the internal combustion engine 10 were input to the detection map. The operation amount of the operation unit of the internal combustion engine 10 such as the fuel injection valve 20 and the ignition device 22 tends to be determined based on the operating point of the internal combustion engine 10. Therefore, the operating point variable is a variable including information regarding the operation amount of each operation unit. Therefore, by using the operating point variable as the input of the detection mapping, the value of the combustion state variable can be calculated based on the information regarding the operation amount of each operation unit, and thus the change in the rotational behavior of the crankshaft 24 due to the operation amount. Thus, the value of the combustion state variable can be calculated with higher accuracy.

  Further, by using the operating point variable as an input variable, the value of the combustion state variable is calculated by a joint operation between the rotation waveform variable and the operating point variable using an input coefficient wFjk which is a parameter learned by machine learning. Therefore, it is not necessary to adapt the adaptation value for each operating point variable.

(2nd Embodiment)
Next, a second embodiment will be described with reference to the drawings, focusing on differences from the first embodiment. In the present embodiment, the misfire determination processing is performed outside the vehicle.

FIG. 4 shows a state detection system according to the present embodiment. In FIG. 4, members corresponding to those shown in FIG. 1 are denoted by the same reference numerals for convenience.
The control device 70 in the vehicle VC shown in FIG. The communication device 79 is a device for communicating with the center 120 via the network 110 outside the vehicle VC.

  The center 120 analyzes data transmitted from a plurality of vehicles VC. The center 120 includes a CPU 122, a ROM 124, a storage device 126, a peripheral circuit 127, and a communication device 129, which can communicate with each other via a local network 128. The main program 124a is stored in the ROM 124, and the mapping data 126a is stored in the storage device 126. The main program 124a in the present embodiment is a misfire detection main program. The mapping data 126a is the same as the mapping data 76a in the first embodiment.

  FIG. 5 shows a procedure of a misfire detection process according to the present embodiment. The processing shown in FIG. 5A is realized by the CPU 72 executing the subprogram 74c stored in the ROM 74 shown in FIG. Note that the subprogram 74c in the present embodiment is a misfire detection subprogram. 5B is realized by the CPU 122 executing the main program 124a stored in the ROM 124. In FIG. 5, processes corresponding to the processes shown in FIG. 2 are denoted by the same step numbers for convenience. Hereinafter, the processing illustrated in FIG. 5 will be described along the time sequence of the misfire detection processing.

  In the vehicle VC, the CPU 72 executes the process of S10 shown in FIG. Next, the CPU 72 acquires the rotation speed NE, the charging efficiency η, and the road surface state variable SR (S12). In S12a of the misfire detection processing according to the first embodiment, in addition to the rotational speed NE, the charging efficiency η, and the road surface state variable SR, the ignition timing aig, the target value Af *, the EGR rate Regr, the water temperature THW, the shift position Sft. In S12, the rotation speed NE, the charging efficiency η, and the road surface state variable SR are obtained.

  Then, by operating the communication device 79, the CPU 72 transmits the data acquired in the processing of S10 and S12 to the center 120 together with the identification information of the vehicle VC (S150).

  On the other hand, as shown in FIG. 5B, the CPU 122 of the center 120 receives the transmitted data (S160), and repeatedly executes the processing of S14 to S20 using the received data.

  In the process of S14, the CPU 122 substitutes the values acquired by the processes of S10 and S12 into the input variables x (1) to x (27) of the mapping for detection for calculating the probability of misfire. Here, the CPU 122 sets “s = 1 to 24” and substitutes the minute rotation time T30 (s) for the input variable x (s). Further, the CPU 122 substitutes the rotation speed NE into the input variable x (25) and substitutes the charging efficiency η into the input variable x (26). Then, the CPU 122 substitutes the road surface state variable SR for the input variable x (27).

  In the process of S16, the CPU 122 inputs the input variables x (1) to x (27) to the detection mapping defined by the mapping data 126a stored in the storage device 126 shown in FIG. The probability P (i) of misfire occurring at (i = 1 to 4) is calculated as a combustion state variable.

  Then, the CPU 122 repeatedly performs the processing of S14 to S20 to determine that the number of misfires of the specific cylinder has become equal to or greater than the predetermined number Nth within the predetermined period in the same vehicle specified by the identification information (S22: YES), it is determined that a misfire has occurred (S162). On the other hand, when the processes of S14 to S20 are repeatedly performed for a predetermined period, if the number of misfires of the specific cylinder is less than the predetermined number Nth (S26: YES), the normality is determined and the number N (1) to N (4) is initialized (S164). Note that the predetermined period here is based on the point in time when the processing in S162 or S164 is performed.

  When the processing of S162 and S164 is completed, the CPU 122 operates the communication device 129 based on the identification information and transmits a signal regarding the determination result to the vehicle VC to which the data targeted for the processing of S14 to S20 has been transmitted. (S166). Here, in the case of a determination result indicating that a misfire has occurred, the information regarding the determination result includes information regarding the cylinder in which the misfire has occurred. When completing the process of S166 or making a negative determination in the processes of S22 and S26, the CPU 122 temporarily ends the series of processes illustrated in FIG.

  On the other hand, as shown in FIG. 5A, when CPU 72 in vehicle VC receives a signal related to the determination result transmitted from center 120 (S152: YES), the determination result indicates that misfire has occurred. It is determined whether or not (S154). If it is determined that a misfire has occurred (S154: YES), the CPU 72 proceeds to the process of S24, while if it is a result of no misfire (S154: NO), the fail flag F is initialized. (S28b).

Note that the CPU 72 temporarily ends the processing illustrated in FIG. 5A when the processing of S24 and S28b is completed or when a negative determination is made in the processing of S152.
As described above, in the present embodiment, the processing load of the control device 70 can be reduced by executing the misfire determination processing in the center 120.

(Third embodiment)
Hereinafter, a third embodiment of a state detection system for an internal combustion engine will be described with reference to the drawings, focusing on differences from the first embodiment. The state detection system of the present embodiment has the same configuration as that of the first embodiment, except for the contents of the program 74a stored in the ROM 74 and the mapping data 76a stored in the storage device 76.

  FIG. 6 shows the procedure of the misfire detection process according to the present embodiment. The process shown in FIG. 6 is realized by the CPU 72 repeatedly executing the misfire detection program, which is the program 74a stored in the ROM 74, for example, at a predetermined cycle.

  In the series of processes shown in FIG. 6, the CPU 72 first obtains the minute rotation time T30 (S210). The minute rotation time T30 is calculated by the CPU 72 measuring the time required for the crankshaft 24 to rotate by 30 ° CA based on the output signal Scr of the crank angle sensor 80. Next, the CPU 72 sets the latest minute rotation time T30 acquired in the processing of S210 as the minute rotation time T30 (0), and sets the variable “m” of the minute rotation time T30 (m) to a larger value as the value becomes earlier. (S212). That is, assuming that “m = 1, 2, 3,...”, The minute rotation time T30 (m−1) immediately before the processing of S212 is performed is set as the minute rotation time T30 (m). Thereby, for example, the minute rotation time T30 acquired by the processing of S210 when the processing of FIG. 6 was executed last time becomes the minute rotation time T30 (1).

  Next, the CPU 72 determines that the minute rotation time T30 acquired in the process of S210 is required for rotation of any of the cylinders # 1 to # 4 at an angular interval from 30 ° CA before the compression top dead center to the compression top dead center. It is determined whether it is time (S214). When determining that it is the time required for rotation at the angular interval up to the compression top dead center (S214: YES), the CPU 72 determines whether or not there is a misfire in the cylinder whose compression top dead center was reached by 360 ° CA before. First, the value of the rotation waveform variable to be input to the process of determining the presence or absence of misfire is calculated.

  That is, the CPU 72 first calculates, as the inter-cylinder variable ΔTa, the difference between the values of the minute rotation time T30 related to the angular interval from 30 ° CA before the compression top dead center to the compression top dead center, which are 180 ° apart from each other (S216). ). Specifically, the CPU 72 sets “m = 1, 2, 3,...” And sets the inter-cylinder variable ΔTa (m−1) to “T30 (6m−6) −T30 (6m)”.

  FIG. 7 illustrates the inter-cylinder variable ΔTa. In the present embodiment, an example is shown in which the compression top dead center appears in the order of cylinder # 1, cylinder # 3, cylinder # 4, and cylinder # 2, and the combustion stroke occurs in that order. FIG. 7 shows the detection target of the presence or absence of misfire by acquiring the minute rotation time T30 (0) of the angular interval from 30 ° CA before the compression top dead center of cylinder # 4 to the compression top dead center of the cylinder # 4 in the process of S210. Shows an example in which is the cylinder # 1. In this case, the inter-cylinder variable ΔTa (0) is a minute rotation time corresponding to each of the compression top dead center of the cylinder # 4 and the compression top dead center of the cylinder # 3 which was the compression top dead center immediately before. It is the difference between T30s. In FIG. 7, the inter-cylinder variable ΔTa (2) corresponds to the minute rotation time T30 (12) corresponding to the compression top dead center of the cylinder # 1 to be subjected to misfire detection and the compression top dead center of the cylinder # 2. Is described as a difference from the minute rotation time T30 (18).

  Referring back to FIG. 6, the CPU 72 calculates an inter-cylinder variable ΔTb which is a difference between values of the inter-cylinder variables ΔTa (0), ΔTa (1), ΔTa (2),. (S218). More specifically, the CPU 72 sets “m = 1, 2, 3,...” And sets the inter-cylinder variable ΔTb (m−1) to “ΔTa (m−1) −ΔTa (m + 3)”.

FIG. 7 illustrates the inter-cylinder variable ΔTb. FIG. 7 describes that the inter-cylinder variable ΔTb (2) is “ΔTa (2) −Ta (6)”.
Next, the CPU 72 calculates a variation pattern variable FL indicating a relative magnitude relationship between the inter-cylinder variable ΔTb corresponding to the cylinder whose misfire is to be detected and the inter-cylinder variable ΔTb corresponding to the other cylinders. (S220). In the present embodiment, the variation pattern variables FL [02], FL [12], and FL [32] are calculated.

  Here, the variation pattern variable FL [02] is defined by “ΔTb (0) / ΔTb (2)”. That is, using the example of FIG. 7, the variation pattern variable FL [02] is an inter-cylinder variable ΔTb (2) corresponding to the cylinder # 1 for which misfire is to be detected. This is a value obtained by dividing the inter-cylinder variable ΔTb (0) corresponding to the cylinder # 4. Further, the variation pattern variable FL [12] is defined by “ΔTb (1) / ΔTb (2)”. That is, using the example of FIG. 7, the variation pattern variable FL [12] is an inter-cylinder variable ΔTb (2) corresponding to the cylinder # 1 for which misfire is to be detected, and the next cylinder to be the compression top dead center This is a value obtained by dividing the inter-cylinder variable ΔTb (1) corresponding to # 3. Further, the variation pattern variable FL [32] is defined by “ΔTb (3) / ΔTb (2)”. That is, using the example of FIG. 7, the variation pattern variable FL [32] is an inter-cylinder variable ΔTb (2) corresponding to the cylinder # 1 for which misfire is to be detected. This is a value obtained by dividing the inter-cylinder variable ΔTb (3) corresponding to the cylinder # 2 that has been changed.

  Next, the CPU 72 acquires the rotation speed NE, the charging efficiency η, and the road surface state variable SR (S222). The method of acquiring the rotation speed NE, the charging efficiency η, and the road surface state variable SR is the same as in the first embodiment.

  Then, the CPU 72 obtains the input variables x (1) to x (7) of the detection mapping for outputting the misfire variable PR which is a variable relating to the probability of the occurrence of misfire in the cylinder to be detected by the processing of S218 and S220. The value of the rotation waveform variable thus obtained and the value of the variable acquired by the process of S222 are substituted (S224). That is, the CPU 72 substitutes the inter-cylinder variable ΔTb (2) for the input variable x (1), substitutes the variation pattern variable FL [02] for the input variable x (2), and substitutes the variation pattern for the input variable x (3). The variable FL [12] is substituted, and the variation pattern variable FL [32] is substituted for the input variable x (4). The CPU 72 substitutes the rotational speed NE for the input variable x (5), substitutes the charging efficiency η for the input variable x (6), and substitutes the road surface state variable SR for the input variable x (7).

  Next, the CPU 72 inputs the input variables x (1) to x (7) to the detection mapping defined by the mapping data 76a stored in the storage device 76 shown in FIG. Is calculated (S226). That is, in this embodiment, the misfire variable PR is calculated as the combustion state variable.

  In the present embodiment, the mapping for detection is constituted by a neural network having a single intermediate layer. The neural network includes an input side for nonlinearly transforming an input-side coefficient wFjk (j = 0 to n, k = 0 to 7) and an output of an input-side linear mapping which is a linear mapping defined by the input-side coefficient wFjk. It includes the activation function h (x) as a non-linear mapping. In the present embodiment, ReLU is exemplified as the activation function h (x). ReLU is a function that outputs the smaller of the input and “0”. Incidentally, wFj0 and the like are bias parameters, and the input variable x (0) is defined as “1”.

  In addition, the neural network includes an output-side coefficient wSij (i = 1 to 2, j = 0 to n) and a prototype variable yR (output of an output-side linear mapping that is a linear mapping defined by the output-side coefficient wSij). 1) and yR (2) as inputs and include a softmax function that outputs a misfire variable PR. Thus, in the present embodiment, the misfire variable PR quantifies the degree of plausibility that an actual misfire has occurred as a continuous value within a predetermined region larger than “0” and smaller than “1”. It will be.

  Next, the CPU 72 determines whether or not the value of the misfire variable PR is equal to or greater than the determination value PRth (S228). When determining that the value is equal to or greater than the determination value PRth (S228: YES), the CPU 72 increments the counter CR (S230). Then, the CPU 72 determines whether or not a predetermined period has elapsed from the time when the process of S228 is first executed or the time when the process of S236 described later is performed (S232). Here, the predetermined period is longer than the period of one combustion cycle, and desirably has a length of 10 times or more of one combustion cycle.

  When determining that the predetermined period has elapsed (S232: YES), the CPU 72 determines whether or not the counter CR is equal to or greater than a threshold value CRth (S234). This process is a process for determining whether or not a misfire has occurred with a frequency exceeding the allowable range. When determining that the value is less than the threshold value CRth (S234: NO), the CPU 72 initializes the counter CR (S236). On the other hand, when determining that the difference is equal to or larger than the threshold value CRth (S234: YES), the CPU 72 executes a notification process of operating the warning light 90 shown in FIG. 1 to urge the user to deal with the abnormality (S238). ).

  Note that the CPU 72 temporarily ends the series of processes illustrated in FIG. 6 when the processes of S236 and S238 are completed or when a negative determination is made in the processes of S214, S228, and S232.

  Incidentally, similarly to the mapping data 76a exemplified in the first embodiment, the mapping data 76a of the present embodiment was learned by operating the internal combustion engine 10 with the dynamometer connected to the crankshaft 24 on the test bench. Things.

Here, the operation and effect of the present embodiment will be described.
The CPU 72 determines the presence or absence of a misfire by calculating the value of the misfire variable PR based on the rotation waveform variable. Therefore, effects similar to those of the first embodiment can be obtained.

According to the third embodiment, the following operation and effect can be further obtained.
(3) A rotation waveform variable serving as an input variable x is generated by selectively using a value near the compression top dead center in the minute rotation time T30. The most significant difference between the presence and absence of a misfire is in the value near the compression top dead center in the minute rotation time T30. Therefore, by selectively using a value in the vicinity of the compression top dead center of the minute rotation time T30, while suppressing the dimension of the input variable x from increasing, information necessary for determining the presence or absence of a misfire is minimized. Can be captured.

  (4) The rotation waveform variables include the inter-cylinder variable ΔTb (2). The inter-cylinder variable ΔTb (2) is obtained by quantifying the difference between the minute rotation times T30 near the compression top dead center of the cylinder to be detected for misfire and the cylinder adjacent thereto in one dimension in advance. Therefore, information necessary for determining the presence or absence of a misfire can be efficiently captured using a variable having a small number of dimensions.

  (5) The rotation waveform variables include not only the inter-cylinder variable ΔTb (2) but also the fluctuation pattern variable FL. Since vibrations from the road surface and the like are superimposed on the crankshaft 24, there is a concern that erroneous determination may occur when the rotation waveform variable is only the inter-cylinder variable ΔTb (2). On the other hand, in the present embodiment, when the value of the misfire variable PR is calculated using the fluctuation pattern variable FL in addition to the inter-cylinder variable ΔTb (2), the calculation is performed only from the inter-cylinder variable ΔTb (2). As compared with, the value of the misfire variable PR can be a value indicating the degree of plausibility that misfire has occurred, that is, the value indicating the probability with higher accuracy.

  Moreover, in the present embodiment, the value of the misfire variable PR is calculated by a joint operation of the inter-cylinder variable ΔTb (2) and the variation pattern variable FL based on the input coefficient wFjk, which is a parameter learned by machine learning. Therefore, the comparison between the inter-cylinder variable ΔTb (2) and the determination value and the determination of the presence or absence of a misfire based on the comparison between the fluctuation pattern variable FL and the determination value are compared with each other. The presence or absence of a misfire can be determined based on a more detailed relationship between the variation pattern variable FL and the misfire.

(Fourth embodiment)
Hereinafter, the fourth embodiment will be described with reference to the drawings, focusing on differences from the third embodiment. In the present embodiment, the process of calculating the misfire variable PR is performed outside the vehicle.

  The state detection system of the present embodiment differs from the state detection system shown in FIG. 4 in the contents other than the contents of the subprogram 74c stored in the ROM 74, the main program 124a stored in the ROM 124, and the mapping data 126a stored in the storage device 126. The configuration is the same as that of the embodiment. Other similar configurations are denoted by the same reference numerals for convenience.

  FIG. 8 shows a procedure of a misfire detection process according to the present embodiment. The process illustrated in FIG. 8A is realized by the CPU 72 executing the subprogram 74c stored in the ROM 74 illustrated in FIG. Although the contents are different from those of the second embodiment, the subprogram 74c in the present embodiment is also a misfire detection subprogram for detecting misfire. 8B is realized by the CPU 122 executing the main program 124a stored in the ROM 124. Although the contents are different from those of the second embodiment, the main program 124a in this embodiment is also a misfire detection main program for detecting misfire. In FIG. 8, processes corresponding to the processes shown in FIG. 6 are given the same step numbers for convenience. Hereinafter, the processing illustrated in FIG. 8 will be described along the time sequence of the misfire detection processing.

  In the vehicle VC, when the CPU 72 makes an affirmative determination in the process of S214 shown in FIG. 8A, the minute rotation times T30 (0), T30 (6), T30 (12), T30 (18), T30 (24), T30 (30), T30 (36), T30 (42), and T30 (48) are acquired (S250). The minute rotation time T30 constitutes a rotation waveform variable that is a variable including information on a difference between the minute rotation times T30 at each of the different angular intervals. In particular, the minute rotation time T30 is a time required for rotation at an angular interval from 30 ° CA before the compression top dead center to the compression top dead center, and is a value for nine times of the appearance timing of the compression top dead center. . Therefore, the set data of the minute rotation times T30 is a variable indicating information on the difference between the minute rotation times T30 corresponding to the respective different compression top dead centers. The nine minute rotation times T30 are the minute rotation times T30 used when calculating the inter-cylinder variable ΔTb (2) and the variation pattern variables FL [02], FL [12], and FL [32]. It is all.

  Next, after executing the process of S222, the CPU 72 operates the communication device 79 to transmit the data acquired in the processes of S250 and S222 to the center 120 together with the identification information of the vehicle VC (vehicle ID) ( S252).

  On the other hand, the CPU 122 of the center 120 receives the transmitted data as shown in FIG. 8B (S260). Then, the CPU 122 substitutes the values of the variables obtained by the processing of S260 into the input variables x (1) to x (12) (S262). That is, the CPU 122 assigns the minute rotation time T30 (0) to the input variable x (1), assigns the minute rotation time T30 (6) to the input variable x (2), and assigns the minute rotation to the input variable x (3). The time T30 (12) is substituted, and the minute rotation time T30 (18) is substituted for the input variable x (4). Further, the CPU 122 substitutes the minute rotation time T30 (24) for the input variable x (5), substitutes the minute rotation time T30 (30) for the input variable x (6), and substitutes the minute rotation time T30 for the input variable x (7). T30 (36) is substituted. Further, the CPU 122 substitutes the minute rotation time T30 (42) into the input variable x (8) and substitutes the minute rotation time T30 (48) into the input variable x (9). Further, the CPU 122 substitutes the rotation speed NE into the input variable x (10), substitutes the charging efficiency η into the input variable x (11), and substitutes the road surface state variable SR into the input variable x (12).

  Next, the CPU 122 inputs the input variables x (1) to x (12) to the detection mapping defined by the mapping data 126a stored in the storage device 126 shown in FIG. Is calculated (S264).

  In the present embodiment, in this detection mapping, the number of intermediate layers is “α”, the activation functions h1 to hα of each intermediate layer are ReLU, and the activation function of the output layer is a softmax function. It is composed of a certain neural network. For example, the value of each node in the first intermediate layer is obtained by converting the input variables x (1) to x () into a linear mapping defined by coefficients w (1) ji (j = 0 to n1, i = 0 to 12). 12) is generated by inputting the output at the time of input to the activation function h1. That is, assuming that m = 1, 2,..., Α, the value of each node of the m-th intermediate layer is generated by inputting the output of the linear mapping defined by the coefficient w (m) to the activation function hm. Is done. In FIG. 8, n1, n2,..., Nα are the numbers of nodes in the first, second,. Incidentally, w (1) j0 and the like are bias parameters, and the input variable x (0) is defined as “1”.

  Next, by operating the communication device 129, the CPU 122 transmits a signal indicating the value of the misfire variable PR to the vehicle VC to which the data received in the process of S260 has been transmitted (S266), and FIG. Is terminated once. On the other hand, as shown in FIG. 8A, the CPU 72 receives the value of the misfire variable PR (S254), and executes the processes of S228 to S238.

As described above, in the present embodiment, the processing of S264 is executed in the center 120, so that the calculation load of the CPU 72 can be reduced.
(Fifth embodiment)
Next, a fifth embodiment will be described with reference to the drawings, focusing on differences from the above embodiments.

  The state detection system of each of the above embodiments is configured as a system that detects a state in which a misfire has occurred in the internal combustion engine 10 based on the rotation fluctuation of the crankshaft 24. Even when the air-fuel ratio varies among the cylinders or when the air-fuel ratio imbalance among the cylinders occurs, the variation in the combustion state between the cylinders occurs, and the rotational fluctuation of the crankshaft 24 increases. The state detection system of the present embodiment is configured as a system that detects such an air-fuel ratio imbalance between cylinders. Note that the configuration of the state detection system of the present embodiment is basically the same as the configuration shown in FIG. However, a program for detecting the air-fuel ratio imbalance between cylinders is stored in the ROM 74 of the state detection system of this embodiment as the program 74a instead of the misfire detection program.

  FIG. 9 shows a procedure of the imbalance detection processing according to the present embodiment. The process shown in FIG. 9 is realized by the CPU 72 repeatedly executing the imbalance detection program, which is the program 74a stored in the ROM 74, at a predetermined control cycle.

  In the series of processes shown in FIG. 9, first, the CPU 72 determines whether or not the execution condition of the imbalance detection process is satisfied (S310). The execution condition includes that the purge of the fuel vapor with respect to the intake air of the internal combustion engine 10 and the recirculation of the exhaust gas are not performed.

  Next, in the processing of S312a, the CPU 72 determines the minute rotation time T30 (1), T30 (2),..., T30 (24), the upstream average value Afuave (1), Afuave (2),. ), The rotation speed NE, the charging efficiency η, and the 0.5th-order amplitude Ampf / 2 are acquired. The minute rotation time T30 is calculated by the CPU 72 measuring the time required for the crankshaft 24 to rotate by 30 ° CA based on the output signal Scr of the crank angle sensor 80. Here, when the numbers in parentheses such as the minute rotation times T30 (1) and T30 (2) are different, it indicates that the rotation angle intervals are different within 720 ° CA which is one combustion cycle. That is, the minute rotation times T30 (1) to T30 (24) indicate the rotation time at each angular interval obtained by equally dividing the rotation angle area of 720 ° CA by 30 ° CA.

  More specifically, the CPU 72 measures the time during which the crankshaft 24 rotates by 30 ° CA based on the output signal Scr, and sets the time as the pre-filtering time NF30. Next, the CPU 72 calculates a post-filter processing time AF30 by performing digital filter processing using the pre-filter processing time NF30 as an input. Then, the CPU 72 normalizes the post-filtering time AF30 so that the difference between the maximum value and the minimum value of the post-filtering time AF30 in a predetermined period (for example, 720 ° CA) becomes “1”, thereby obtaining the minute rotation time T30. Is calculated.

  If m = 1 to 24, the upstream average value Afuave (m) is the average value of the upstream detection values Afu at the same 30 ° CA angular interval as the minute rotation times T30 (m). The 0.5th-order amplitude Ampf / 2 is the intensity of the 0.5th-order component of the rotation frequency of the crankshaft 24, and is calculated by the CPU 72 by the Fourier transform of the time-series data of the minute rotation time T30.

  In addition, in the process of S312a, the CPU 72 also acquires the ignition timing average value aigave, the EGR rate average value Regrave, the water temperature THW, the shift position Sft, and the road surface state variable SR. Here, the average ignition timing aigave and the average EGR rate Regrave are the average values of the ignition timing aig and the EGR rate Regr in the cycle of the process of FIG. The method of obtaining the road surface state variable SR is the same as in the first embodiment.

  Next, the CPU 72 substitutes the value obtained by the processing of S312a into the input variables x (1) to x (56) of the detection mapping that outputs the imbalance rate Riv (S314a). Specifically, the CPU 72 sets the value of the minute rotation time T30 (m) to the input variable x (m) and sets the value of the upstream average value Afuave (m) to the input variable x (24 + m), assuming that “m = 1 to 24”. , The value of the rotational speed NE as the input variable x (49), the value of the charging efficiency η as the input variable x (50), and the value of the 0.5-order amplitude Ampf / 2 as the input variable x (51). substitute. Further, the CPU 72 sets the input variable x (52) to the value of the ignition timing average value aigave, the input variable x (53) to the value of the EGR rate average value Regrave, the input variable x (54) to the value of the water temperature THW, The value of the shift position Sft is substituted for the input variable x (55), and the value of the road surface state variable SR is substituted for the input variable x (56).

  In the present embodiment, the imbalance rate Riv is set to “0” in the cylinder in which the target injection amount of fuel is being injected, and becomes a positive value when the actual injection amount is larger than the target injection amount, When the number is small, the value is negative.

  Next, the CPU 72 inputs the values of the input variables x (1) to x (56) into the detection mapping defined by the mapping data 76a stored in the storage device 76 shown in FIG. The values of the imbalance rates Riv (1) to Riv (4) for i = 1 to 4) are calculated (S316a).

  In the present embodiment, the mapping for detection is constituted by a neural network having a single intermediate layer. The above-mentioned neural network performs an input-side conversion on each of an input-side coefficient wFjk (j = 0 to n, k = 0 to 56) and an output of an input-side linear mapping which is a linear mapping defined by the input-side coefficient wFjk. It includes the activation function h (x) as a non-linear mapping. In the present embodiment, a hyperbolic tangent “tanh (x)” is exemplified as the activation function h (x). Further, the neural network nonlinearly converts each of the output-side coefficient wSij (i = 1 to 4, j = 0 to n) and the output of the output-side linear mapping which is a linear mapping defined by the output-side coefficient wSij. An activation function f (x) as an output-side nonlinear mapping is included. In the present embodiment, the hyperbolic tangent “tanh (x)” is exemplified as the activation function f (x). The value n indicates the dimension of the intermediate layer.

Note that the CPU 72 temporarily ends the series of processes illustrated in FIG. 9 when the process of S316a is completed or when a negative determination is made in the process of S310.
FIG. 10 shows a procedure of a process using the imbalance rate Riv (i). The process illustrated in FIG. 10 is realized by the CPU 72 repeatedly executing the program stored in the ROM 74 illustrated in FIG. 1 every time the imbalance ratio Riv (i) is calculated, for example.

  In the series of processes shown in FIG. 10, the CPU 72 first performs the imbalance learning value Lib (i) by an exponential moving average process that receives the value of the imbalance ratio Riv (i) newly calculated by the process of FIG. ) Is updated (S320). Specifically, the CPU 72 calculates a value obtained by multiplying the imbalance learning value Lib (i) stored in the storage device 76 by the coefficient α and a value obtained by multiplying the imbalance rate Riv (i) by “1−α”. The imbalance learning value Lib is updated by the sum of the values (S320). Note that “0 <α <1”.

  Next, the CPU 72 determines whether or not the imbalance learning value Lib (i) is equal to or greater than the lean allowable limit LL and equal to or less than the rich allowable limit LH (S322). When determining that the imbalance learning value Lib (i) is less than the lean-side allowable limit LL or determining that the imbalance learning value Lib (i) is larger than the rich-side allowable limit (S322: NO), the CPU 72 repairs the user. Is operated by operating the warning light 90 (S324).

  On the other hand, when the CPU 72 determines that the value is equal to or more than the lean-side allowable limit value LL and equal to or less than the rich-side allowable limit value LH (S322: YES), or when the process of S324 is completed, the CPU 72 The required injection amount Qd (#i) is corrected (S326). That is, the CPU 72 adds the correction amount ΔQd (Live (i)) corresponding to the imbalance learning value Lib (i) to the required injection amount Qd (#i) of each cylinder, thereby obtaining the required injection amount Qd (#i). ) Is corrected. Here, the correction amount ΔQd (Live (i)) has a negative value when the imbalance learning value Lib (i) is larger than zero, and has a positive value when the imbalance learning value Lib (i) is smaller than zero. When the imbalance learning value Lib (i) is zero, the correction amount ΔQd (Live (i)) is also zero.

  When completing the process of S326, the CPU 72 temporarily ends the series of processes illustrated in FIG. By the way, in the present embodiment, when the affirmative determination is made in the processing of S310 and the processing of S312a is executed, the processing of S326 is temporarily stopped.

Incidentally, the mapping data 76a in the present embodiment is generated as follows, for example.
First, a plurality of fuel injection valves 20 having various values different from zero in imbalance ratio and three fuel injection valves 20 in which imbalance ratio is zero are prepared in advance by single measurement. Then, the internal combustion engine 10 connected to the torque converter 60 and equipped with three fuel injection valves 20 having an imbalance rate of zero and one fuel injection valve 20 having an imbalance rate different from zero is mounted on a test bench. The torque converter 60 is operated with the dynamometer connected to the output shaft. Note that the imbalance ratio Rivt of each of the mounted fuel injection valves 20 is used as teacher data.

  Then, the value of the imbalance rate Rivt is calculated by the same processing as the processing of S314a and S316a, using the values of the rotational waveform variables and the variables acquired by the processing of S312a. The values of the input-side coefficient wFjk and the output-side coefficient wSij are learned so as to reduce the difference between the value of the imbalance ratio Rivt thus calculated and the teacher data. Specifically, for example, the values of the input side coefficient wFjk and the output side coefficient wSij may be learned so as to minimize the tolerance entropy.

  In the present embodiment, the road surface state variable SR is included in the input to the detection mapping. Therefore, the imbalance rate can be calculated based on the influence of the state of the road on which the vehicle is traveling. Therefore, the detection accuracy of the inter-cylinder air-fuel ratio imbalance in the internal combustion engine 10 mounted on the vehicle VC can be improved.

According to the embodiment described above, the following functions and effects can be obtained.
(1) The rotational speed NE and the charging efficiency η as operating point variables that define the operating point of the internal combustion engine 10 were input to the detection map. The operation amounts of the operation units of the internal combustion engine 10, such as the ignition device 22, the EGR valve 34, and the variable intake valve timing device 40, tend to be determined based on the operating point of the internal combustion engine 10. Therefore, the operating point variable is a variable including information regarding the operation amount of each operation unit. Therefore, by using the operating point variable as the input of the detection mapping, the imbalance rate Riv (i) can be calculated based on the information regarding the operation amount of each operation unit, and the imbalance rate Riv (i) can be further increased. It can be calculated with high accuracy.

  (2) The upstream average value Afuave was included in the input of the detection mapping. Thereby, as compared with the case of using the upstream detection value Afu for each time interval of the time series data, a more accurate oxygen and unburned fuel flowing into the catalyst 30 can be obtained without increasing the number of data of the time series data. Information can be obtained, and the imbalance rate Riv (i) can be calculated with higher accuracy.

(Sixth embodiment)
Next, a sixth embodiment will be described with reference to the drawings, focusing on differences from the fifth embodiment. In the present embodiment, the process of calculating the imbalance rate Riv (i) is performed outside the vehicle.

  The state detection system of the present embodiment differs from the state detection system shown in FIG. 4 in the contents other than the contents of the subprogram 74c stored in the ROM 74, the main program 124a stored in the ROM 124, and the mapping data 126a stored in the storage device 126. The configuration is the same as that of the embodiment. Other similar configurations are denoted by the same reference numerals for convenience.

  FIG. 11 shows a procedure of the imbalance detection process according to the present embodiment. The processing shown in FIG. 11A is realized by the CPU 72 executing the subprogram 74c stored in the ROM 74 shown in FIG. Note that the subprogram 74c in the present embodiment is an imbalance detection subprogram for detecting imbalance. 11B is realized by the CPU 122 executing the main program 124a stored in the ROM 124. Note that the main program 124a in the present embodiment is an imbalance detection main program for detecting imbalance. In FIG. 11, processes corresponding to the processes shown in FIG. 9 are given the same step numbers for convenience. Hereinafter, the processing illustrated in FIG. 11 will be described along the time sequence of the imbalance detection processing.

  As shown in FIG. 11A, in the vehicle VC, when the processing of S312a is completed, the CPU 72 operates the communication device 79 to convert the data acquired in the processing of S312a into the vehicle ID which is the identification information of the vehicle VC. Is transmitted to the center 120 (S332).

  On the other hand, as shown in FIG. 11B, the CPU 122 of the center 120 receives the transmitted data (S340) and executes the processes of S314a and S316a. Then, by operating the communication device 129, the CPU 122 transmits a signal relating to the imbalance rate Riv (i) to the vehicle VC to which the data received in the process of S340 has been transmitted (S342), and is illustrated in FIG. A series of processing ends once. On the other hand, as shown in FIG. 11A, the CPU 72 receives a signal related to the imbalance rate Riv (i) (S334), and once ends a series of processes shown in FIG. 11A.

As described above, in the present embodiment, the center 120 calculates the imbalance rate Riv (i), so that the calculation load on the CPU 72 can be reduced.
(Correspondence)
The correspondence between the items in the above embodiment and the items described in the section of “Means for Solving the Problem” is as follows. The storage device corresponds to the storage device 76. In the first, second, fifth, and sixth embodiments, the minute rotation times T30 (1) to T30 (24) correspond to rotation waveform variables. In the third and fourth embodiments, the inter-cylinder variable ΔTb (2) and the fluctuation pattern variables FL [02], FL [12], FL [32] correspond to the rotation waveform variables. The first execution device corresponds to the CPU 72 and the ROM 74. The second execution device corresponds to the CPU 122 and the ROM 124. The acquisition processing corresponds to the processing of S10, S12a, S12, S218, S222, S250, and S312a. The output value calculation processing corresponds to the processing of S16a, S16, S226, S264, and S316a. The external transmission processing corresponds to the processing of S166, S266, and S342. The vehicle-side reception processing corresponds to the processing of S152, S254, and S334. The determination processing corresponds to the processing of S18, S20, S22, S24, S162, S228, S230, S232, S234, and S322. The data analysis device corresponds to the center 120.

  Further, in the first and second embodiments, the probability P (i) of misfire corresponds to the combustion state variable, and in the third and fourth embodiments, the misfire variable PR corresponds to the combustion state variable. In the fifth and sixth embodiments, the imbalance rate Riv (i) corresponds to the combustion state variable.

  In the first embodiment, the second embodiment, the third embodiment, and the fourth embodiment, the state in which a misfire has occurred corresponds to a predetermined operation state, and the fifth embodiment and the sixth embodiment have a cylinder-to-cylinder state. The state in which the air-fuel ratio varies corresponds to the predetermined operating state.

(Other embodiments)
Each of the above embodiments can be modified and implemented as follows. The above embodiments and the following modified examples can be implemented in combination with each other within a technically consistent range.

-"Sampling interval of minute rotation time T30"
The sampling interval of the minute rotation time T30 as an input to the mapping is not limited to 30 ° CA. For example, the angle interval may be smaller than 30 ° CA, such as 10 ° CA. However, the angle interval is not limited to 30 ° CA or less, but may be 45 ° CA or the like.

"About the sampling interval of the upstream average value Afuave"
The sampling interval of the upstream average value Afuave which is input to the mapping is not limited to 30 ° CA. For example, the angle interval may be smaller than 30 ° CA, such as 10 ° CA. However, the angle interval is not limited to 30 ° CA or less, but may be 45 ° CA or the like.

・ "About time series data"
The time series data of the upstream average value Afuave and the time series data of the minute rotation time T30, which are input of the mapping, are not limited to the time series data at the rotation angle interval of 720 °. For example, time-series data at a rotation angle interval wider than 720 ° CA may be used. Time-series data at an angle interval smaller than 720 ° CA may be used.

・ "Rotation waveform variable as input to detection mapping"
In the first embodiment and the second embodiment, the minute rotation time T30 at each of a plurality of intervals obtained by dividing the rotation angle interval of 720 ° CA, which is one combustion cycle, is input to the detection map. Not limited to For example, 0 to 720 ° CA is defined as a first interval, and 0 to 20, 40 to 60, 80 to 100, 120 to 140, 160 to 180,... As the interval, the time required for the rotation may be input to the detection mapping.

  The variable constituting the rotation waveform variable is not limited to the minute rotation time which is the time required for the rotation at the second interval. For example, a value obtained by dividing the second interval by the minute rotation time may be used. Note that it is not essential that the variables constituting the rotation waveform variables have been subjected to normalization processing in which the difference between the maximum value and the minimum value is a fixed value. The filter processing as preprocessing for inputting a map is not limited to the above-described processing. For example, the crankshaft 24 is rotated by the input shaft 66 based on the minute rotation time of the input shaft 66 of the transmission 64. The processing for removing the influence component may be performed. However, it is not essential that the filtering process is performed.

  In the process of S226 in the third embodiment, the rotation waveform variable is configured by the inter-cylinder variable ΔTb (2) and the variation pattern variables FL [02], FL [12], and FL [32], but is not limited thereto. For example, the variation pattern variables constituting the rotation waveform variable may be any one or two of the variation pattern variables FL [02], FL [12], and FL [32]. Further, for example, four or more fluctuation pattern variables such as the fluctuation pattern variables FL [02], FL [12], FL [32], and FL [42] may be included.

  In the processing of S264 in the fourth embodiment, the rotation waveform variable is configured by the minute rotation time T30 corresponding to each of the nine timings at which the appearance timing of the compression top dead center is different from each other, but is not limited thereto. For example, the rotation waveform is obtained by the minute rotation time T30 in each of the sections at least twice the angular interval at which the compression top dead center appears, divided at intervals of 30 ° CA, with the compression top dead center of the cylinder whose misfire is to be detected as the center. Variables may be configured. In the above description, it is not essential that the compression top dead center of the cylinder whose misfire is to be detected be set at the center. Furthermore, the minute rotation time here is not limited to the time required for rotation at an interval of 30 ° CA. Instead of the minute rotation time, an instantaneous rotation speed obtained by dividing the predetermined angular interval by the time required for rotation at the predetermined angular interval may be used.

・ "About inter-cylinder variables"
As the inter-cylinder variable ΔTb, the appearance timing of the compression top dead center is the difference between the values of the minute rotation times T30 corresponding to the respective compression top dead centers of a pair of cylinders adjacent to each other separated by 720 ° CA. Not exclusively. For example, even when the appearance timing of the compression top dead center is 360 ° CA apart from each other, the difference between the minute rotation times T30 corresponding to the respective compression top dead centers of cylinders separated by 720 ° CA is the difference between the values separated by 720 ° CA. Good. In this case, the inter-cylinder variable ΔTb (2) is “T30 (12) −T30 (24) − {T30 (36) −T30 (48)}”.

  Further, the present invention is not limited to the difference between the minute rotation times T30 corresponding to the respective compression top dead centers of the pair of cylinders, which are separated by 720 ° CA. May be the difference between the minute rotation times T30 corresponding to the respective compression top dead centers.

Further, for example, the inter-cylinder variable may be a ratio between the minute rotation times T30 corresponding to the respective compression top dead centers of the pair of cylinders.
The minute rotation time when defining the inter-cylinder variable ΔTb is not limited to the time required for rotation at 30 ° CA, but may be, for example, the time required for rotation at 45 ° CA. At this time, it is desirable that the minute rotation time is a time required for rotation at an angular interval equal to or less than an appearance interval of the compression top dead center.

Further, in the above, an instantaneous rotation speed obtained by dividing the predetermined angular interval by the time required for rotation at the predetermined angular interval may be used instead of the minute rotation time.
・ “About fluctuation pattern variables”
The definition of the variation pattern variable is not limited to the one exemplified in the above embodiment. For example, the definition of the variation pattern variable may be changed by changing the inter-cylinder variable ΔTb to one exemplified in the section “About the inter-cylinder variable”.

  Further, it is not essential to define the variation pattern variable as a ratio between the inter-cylinder variables ΔTb corresponding to the appearance timings of the different compression top dead centers, and a difference may be used instead of the ratio. Even in this case, by including the operating point variable of the internal combustion engine 10 in the input, the value of the misfire variable PR can be calculated reflecting that the magnitude of the variation pattern variable changes according to the operating point.

・ "About the state variable of the drive train as an input to the mapping for detection"
2, 9, and 11, the shift position Sft is used as the state variable of the drive system connected to the crankshaft 24, but the present invention is not limited to this. For example, the lock-up clutch 62 may be in the state. This state may be, for example, a binary state of a fastening state or a releasing state, or may be a state of identifying a fastening state, a releasing state, and a slipping state, respectively.

Further, for example, the oil temperature of the transmission 64 may be used. Further, for example, the rotation speed of the drive wheel 69 may be used. It should be noted that the number of state variables of the drive system device that is the input of the mapping is not limited to one.
・ "About the air-fuel ratio detection variable as an input to the detection mapping"
In the above embodiment, the upstream average value Afuave at each of a plurality of intervals obtained by dividing the rotation angle interval of 720 ° CA, which is one combustion cycle, is input to the detection mapping, but the invention is not limited to this. For example, the upstream average value Afuave in the second interval may be input to the detection mapping.

  In the above embodiment, the upstream average value Afuave is an average value of a plurality of sampling values in one second interval, but is not limited to this. For example, the upstream average value Afuave may be calculated by sampling the upstream detection value Afu in each of the second intervals over a plurality of combustion cycles and averaging them at each second interval.

The air-fuel ratio detection variable is not limited to the upstream average value Afuave, but may be the upstream detection value Afu.
-"Adjustment variables for adjusting the combustion rate of the air-fuel mixture as the input of the detection map"
In the process of S316a, the ignition timing average aigave and the EGR rate average Regreg are used as adjustment variables that are input of the detection mapping and adjust the combustion speed. You may use only.

  The ignition variable as the variable indicating the ignition timing is not limited to the ignition timing average value aigave, but may be the ignition timing aig. The variable indicating the EGR rate is not limited to the average value of the EGR rate Regrave, and may be the EGR rate Regr.

Further, for example, when the present invention is applied to a compression ignition type internal combustion engine as described in the section “About the internal combustion engine”, the injection timing and the average value thereof may be used instead of the ignition variable.
The adjustment variables that become the input of the detection map and adjust the combustion speed are not limited to the above variables, and include, for example, the valve opening timing and the lift amount of the intake valve 16 and the valve opening timing and the lift amount of the exhaust valve 26. You may. Further, for example, during the cylinder deactivation control in which the combustion control of the air-fuel mixture is stopped in some cylinders, the information may be information on the cylinders in which the combustion control is stopped. Further, for example, in the case of the internal combustion engine 10 including the supercharger and the wastegate valve, the opening degree of the wastegate valve may be used.

・ “Road surface variables as input of detection mapping”
The road surface state variable SR is not limited to a binary variable corresponding to the size of the undulation, but may be a variable having three or more values corresponding to the size of the undulation. The method for generating the road surface state variable SR based on the acceleration Dacc is not limited to the method illustrated in FIG. 3, and for example, an average value of the absolute value of the acceleration Dacc per predetermined time may be used. According to this, the road surface state variable SR is a continuous quantity.

  The road surface state variable SR is not limited to the one quantified based on the detected value of the vertical acceleration of the vehicle VC. For example, it may be quantified by the lateral acceleration or the longitudinal acceleration of the vehicle. Furthermore, the input of the mapping is quantified using any two or all of three accelerations of a vertical acceleration of the vehicle, a lateral acceleration, and a longitudinal acceleration. Is also good.

  The road surface state variable SR is quantified by the acceleration Dacc detected by the acceleration sensor 88, but is not limited thereto. For example, data obtained by analyzing road surface image data obtained by a camera or data related to radar reflected waves may be used.

  For example, when analyzing road surface image data obtained by a camera, a convolutional neural network may be used. The convolutional neural network includes, for example, a feature extraction part in which several convolutional layers and a pooling layer are alternately stacked, and an identification part including a fully connected layer for performing classification using the softmax method. By using such a convolutional neural network, it is possible to obtain imaging data as input and output classified according to the magnitude of the undulation.

  Alternatively, for example, a road surface state at the current location may be acquired using map data having road surface state information and GPS, and this may be used. Alternatively, the air pressure of each wheel may be used. That is, the difference between the air pressures of the wheels includes information on the deflection of the wheels due to the inclination or unevenness of the road surface. Therefore, the information indicating the road surface state is included in the difference between the air pressures of the respective wheels. Further, for example, a detection value of a sensor such as a fuel gauge or an oil gauge that detects the level of the fluid level in the vehicle may be used. That is, the fluctuation of the liquid level includes information such as unevenness of the road surface. Alternatively, a detection value of a sensor that detects the hydraulic pressure of the suspension may be used. That is, since the suspension operates under the influence of the road surface unevenness, the hydraulic pressure includes information such as road surface unevenness.

  The road surface state variable is not limited to a quantified road surface unevenness or the like. For example, it may be a value that quantifies the magnitude of friction, such as the presence or absence of freezing, or a variable that identifies both the undulation and the magnitude of friction. In addition, as an input for quantifying the magnitude of friction, a difference in wheel speed of each wheel or the like can be used. Further, as input for determining whether or not the road surface is frozen, information on the outside air temperature and image data obtained by a camera can be used.

・ "About operating point variables as input to the mapping for detection"
The operating point variables are not limited to the rotation speed NE and the charging efficiency η. For example, the intake air amount Ga and the rotation speed NE may be used. When a compression ignition type internal combustion engine is used, for example, as described in the section “About the internal combustion engine” below, the injection amount and the rotational speed NE may be used. Note that it is not essential that the operating point variable is input as the detection mapping. For example, in the case where the invention is applied to an internal combustion engine mounted on a series hybrid vehicle described in the section "About the vehicle", when the internal combustion engine is operated only at a specific operating point, an operating point variable is input. The value of the combustion state variable can be calculated with high accuracy even if it is not included in the variable.

・ "Input to the mapping for detection"
The input to the detection mapping that is input in addition to the instantaneous speed parameter is not limited to those illustrated in the above-described embodiment and the above-described modification. For example, a parameter related to the environment where the internal combustion engine 10 is placed may be used. Specifically, for example, atmospheric pressure, which is a parameter affecting the proportion of oxygen in the intake air, may be used. Further, for example, an intake air temperature which is a parameter affecting the combustion speed of the air-fuel mixture in the combustion chamber 18 may be used. Further, for example, humidity, which is a parameter affecting the combustion speed of the air-fuel mixture in the combustion chamber 18, may be used. It should be noted that a humidity sensor may be used to grasp the humidity, but a state of the wiper or a detection value of a sensor that detects raindrops may be used. Alternatively, the data may be data on the state of an auxiliary machine mechanically connected to the crankshaft 24.

  It is not essential that the input to the detection map include the operating point of the internal combustion engine 10. For example, when the internal combustion engine is mounted on a series hybrid vehicle and its operating point is assumed to be controlled in a narrow range as described in the section "About the internal combustion engine" below, the operating point is set to It is not necessary to include it.

  Further, the input to the neural network, the input to the regression equation described in the column of “Machine Learning Algorithm” below, and the like are not limited to those in which each dimension is composed of a single physical quantity or variation pattern variable FL. For example, in the above-described embodiment and the like, a plurality of types of physical quantities and some of the variation pattern variables FL input to the detection mapping are obtained by principal component analysis instead of being directly input to a neural network or a regression equation. Some principal components may be input directly to a neural network or regression equation. However, when the principal component is used as the input of the neural network or the regression equation, it is not essential that only a part of the input to the neural network or the regression equation is the principal component, and the whole may be the principal component. When the principal component is used as the input to the detection mapping, the mapping data 76a and 126a include data defining the detection mapping that defines the principal component.

・ "Response processing"
In the above embodiment, by operating the warning light 90, the fact that a misfire has occurred is notified through visual information, but the present invention is not limited to this. For example, by operating a speaker, the fact that a misfire has occurred may be reported through auditory information. Further, for example, the control device 70 illustrated in FIG. 1 may include the communication device 79, and the communication device 79 may be operated to transmit a signal indicating that a misfire has occurred to the user's portable terminal. This can be realized by installing an application program for executing the notification process in the user's mobile terminal.

  The coping process is not limited to the notification process. For example, an operation process for operating an operation unit for controlling the combustion of the air-fuel mixture in the combustion chamber 18 of the internal combustion engine 10 in accordance with information indicating that misfire has occurred may be used. Specifically, for example, the ignition timing of a cylinder in which misfire has occurred may be advanced by using the operation unit as the ignition device 22. Further, for example, the fuel injection valve 20 may be used as the fuel injection valve 20 to increase the fuel injection amount of the cylinder in which misfire has occurred.

・ About the execution device
The execution device is not limited to one that includes the CPU 72 and the ROM 74 or the CPU 122 and the ROM 124 and executes software processing. For example, a dedicated hardware circuit (for example, an ASIC or the like) that performs hardware processing on at least a part of the software processing in the above embodiment may be provided. That is, the execution device may have any of the following configurations (a) to (c). (A) A processing device that executes all of the above processing according to a program and a program storage device such as a ROM that stores the program are provided. (B) A processing device and a program storage device that execute a part of the above-described processing according to a program, and a dedicated hardware circuit that executes the remaining processing. (C) A dedicated hardware circuit for executing all of the above processing is provided. Here, there may be a plurality of software execution devices including a processing device and a program storage device, and a plurality of dedicated hardware circuits.

・ "About storage device"
In the above embodiment, the storage device that stores the mapping data 76a and 126a is a storage device that stores the misfire detection program, the misfire detection main program, and the imbalance detection main program, that is, a storage device that is different from the ROM 74 and the ROM 124. However, it is not limited to this.

・ About the data analyzer
For example, in the second embodiment, instead of the processing of S162 and S164, the processing of S24 and S28b may be executed by the center 120. Similarly, in the fourth embodiment and the sixth embodiment, the notification process may be performed at the center 120 and the process of transmitting a command to execute the notification process may be performed.

  In the second embodiment, the processing in FIG. 5B may be executed by, for example, a mobile terminal owned by the user. This can be realized by installing an application program for executing the processing of FIG. 5B in the portable terminal. At this time, the transmission / reception processing of the vehicle ID may be deleted, for example, by setting such that the effective distance of data transmission in the processing of S162 is about the length of the vehicle. The same applies to the fourth and sixth embodiments.

-"Mapping data of the first embodiment and the second embodiment"
It is not essential that the activation function h (x) of the intermediate layer be the same function in all dimensions.

  The neural network constituting the mapping that outputs the probability of misfire as an output is not limited to a neural network having a single hidden layer. For example, the intermediate layer may be a neural network having two or more layers. When the number of intermediate layers is large, the use of the system shown in FIG. 4 is particularly effective in reducing the calculation load of the control device 70.

  The activation function h (x) is not limited to the one using the hyperbolic tangent. For example, it may be a logistic jigmoid function. The activation function f (x) in the processing of FIGS. 2 and 5 is not limited to the one using the hyperbolic tangent. For example, it may be a logistic jigmoid function. In this case, the output of the activation function f (x) may be used as the probability P without using the softmax function. In this case, the probabilities P (1) to P (4) of misfires in the cylinders # 1 to # 4 take values of "0" to "1" independently of each other.

  The mapping for outputting the probability of misfire is not limited to the one using a neural network. For example, a coefficient that defines a parameter obtained by reducing the dimension of data acquired for inputting a mapping, such as the minute rotation time T30, by principal component analysis may be replaced with the input-side coefficient wFij.

"About the mapping data of the third embodiment"
The mapping data that defines the mapping used for the calculation performed in the vehicle may be the data that defines the mapping exemplified in the process of S264.

For example, according to the description of FIG. 8, the number of intermediate layers of the neural network is expressed in more than two layers, but is not limited thereto.
In this embodiment, the activation functions h, h1, h2,... Hα are ReLU, and the output activation function is a softmax function. However, the present invention is not limited to this. For example, the activation functions h, h1, h2,... Hα may be hyperbolic tangents. Also, for example, the activation functions h, h1, h2,... Hα may be logistic jigmoid functions.

  Further, for example, the output activation function may be a logistic jigmoid function. In this case, for example, the number of nodes in the output layer may be one, and the output variable may be the misfire variable PR. In that case, if the value of the output variable is equal to or more than a predetermined value, it is determined that there is an abnormality.

"About the mapping data of the fifth embodiment and the sixth embodiment"
For example, in S316a, the number of intermediate layers in the neural network may be two or more.

  In S316a, the number of intermediate layers in the neural network is expressed in more than two layers, but is not limited thereto. In particular, from the viewpoint of reducing the calculation load of the control device 70, it is desirable to reduce the number of intermediate layers of the neural network to one or two layers.

  In this embodiment, the activation functions h, f, h1, h2,... Hα are hyperbolic tangents, but are not limited thereto. For example, some or all of the activation functions h, h1, h2,... Hα may be ReLU, or may be, for example, a logistic jigmoid function.

  The detection mapping that outputs the imbalance ratio Riv is not limited to one that has been learned targeting a phenomenon in which the fuel injection amount deviates from the assumed one for only one cylinder. For example, it may be learned for a phenomenon in which the fuel injection amount in two cylinders deviates from the assumed value. In this case, the intensity of the primary component of the rotation frequency of the crankshaft 24 may be included in the input to the mapping.

・ "Machine learning algorithm"
Machine learning algorithms are not limited to those using neural networks. For example, a regression equation may be used. This corresponds to the above-mentioned neural network having no intermediate layer. Further, for example, a support vector machine may be used. In this case, the magnitude of the output value itself has no meaning. For example, whether or not a misfire has occurred is expressed according to whether or not the value is positive. In other words, this is different from the case where the value of the combustion state variable has three or more values and the magnitude of those values represents the magnitude of the probability of misfire.

・ About "Generation of mapping data"
In the above embodiment, data obtained when the internal combustion engine 10 is operated with the dynamometer connected to the crankshaft 24 via the torque converter 60 and the transmission 64 is used as training data, but the present invention is not limited to this. For example, data when the internal combustion engine 10 is driven while mounted on the vehicle VC may be used as training data.

・ "External transmission process"
In the process of S266, the value of the misfire variable PR is transmitted, but is not limited thereto. For example, the values of prototype variables yR (1) and yR (2), which are input of a softmax function as an output activation function, may be transmitted. Further, for example, the processing of S228 to S236 may be performed in the center 120, and the determination result of whether there is an abnormality may be transmitted.

・ "About the default operation state"
If the operation state of the internal combustion engine 10 involves variation in the combustion state between the cylinders, an operation state other than misfire or an inter-cylinder air-fuel ratio imbalance may be set as the detection mode of the state detection system. For example, even if so-called compression loss occurs in a specific cylinder due to insufficient compression of the intake air in the cylinder due to sticking of an intake valve or an exhaust valve, the combustion state between the cylinders varies. The rotation fluctuation of the crankshaft 24 increases. Therefore, if such compression loss detection is performed using the above-mentioned detection mapping using the rotational waveform variables and the road surface state variables as inputs, the compression loss can be detected with high accuracy based on the influence of the road surface condition. Can be detected.

・ "About internal combustion engine"
In the above-described embodiment, the in-cylinder injection valve that injects fuel into the combustion chamber 18 has been exemplified as the fuel injection valve, but is not limited thereto. For example, a port injection valve that injects fuel into the intake passage 12 may be used. Further, for example, both a port injection valve and an in-cylinder injection valve may be provided.

The internal combustion engine is not limited to a spark ignition type internal combustion engine, but may be, for example, a compression ignition type internal combustion engine using light oil or the like as fuel.
It is not essential that the internal combustion engine constitutes the drive system. For example, it may be mounted on a so-called series hybrid vehicle in which the crankshaft is mechanically connected to the on-vehicle generator and the power transmission from the drive wheel 69 is cut off.

・ About the vehicle
The vehicle is not limited to a vehicle in which the device that generates the propulsion force of the vehicle is only an internal combustion engine. For example, in addition to the series hybrid vehicle described in the section “About the internal combustion engine”, a parallel hybrid vehicle or a series-parallel hybrid It may be a car.

·"others"
The drive system device interposed between the crankshaft and the drive wheels is not limited to a stepped transmission, and may be, for example, a continuously variable transmission.

  Reference Signs List 10: internal combustion engine, 12: intake passage, 14: throttle valve, 16: intake valve, 18: combustion chamber, 20: fuel injection valve, 22: ignition device, 24: crankshaft, 26: exhaust valve, 28: exhaust passage , 30 ... catalyst, 32 ... EGR passage, 34 ... EGR valve, 40 ... variable intake valve timing, 42 ... variable intake camshaft, 44 ... variable exhaust valve timing, 46 ... exhaust camshaft, 50 ... crank Rotor, 52: tooth portion, 54: missing tooth portion, 60: torque converter, 62: lock-up clutch, 64: transmission, 66: input shaft, 68: output shaft, 69: drive wheel, 70: control device, 72 .. CPU, 74 ROM, 74a program, 74c subprogram, 76 storage device, 76a mapping data, 77 peripheral circuit, 78 local network 79: communication device, 80: crank angle sensor, 82: air flow meter, 83: air-fuel ratio sensor, 84: water temperature sensor, 86: shift position sensor, 88: acceleration sensor, 90: warning light, 110: network, 120: center , 122 CPU, 124 ROM, 124a main program, 126 storage device, 126a mapping data, 127 peripheral circuit, 128 local network, 129 communication device.

Claims (8)

A state detection system for an internal combustion engine that is applied to an internal combustion engine mounted on a vehicle and detects a predetermined operating state of the internal combustion engine with variation in combustion state between cylinders,
A rotation waveform variable which is a variable including information on a cylinder-to-cylinder difference in a rotation speed of a crankshaft of the internal combustion engine during a generation period of a combustion torque of each cylinder, and a road surface condition which is a variable relating to a road surface state on which the vehicle is traveling A mapping in which a variable is input and a combustion state variable that is a variable relating to the degree of variation in combustion state between cylinders is output, and a joint calculation of the rotation waveform variable and the road surface state variable based on parameters learned by machine learning. When a mapping for outputting the value of the combustion state variable is a detection mapping,
A storage device that stores mapping data that is data defining the detection mapping,
Acquisition processing for acquiring the value of the rotation waveform variable based on the output of a sensor that detects the rotational behavior of the crankshaft, and acquiring the value of the road surface state variable based on the output of a sensor that detects the road surface state, and the acquisition A determination process of determining whether or not the internal combustion engine is in the predetermined operating state based on an output value of the detection mapping to which the respective values of the rotation waveform variable and the road surface state variable obtained through the process are input; An execution device for executing,
An internal combustion engine state detection system comprising:   2. The state detection system for an internal combustion engine according to claim 1, wherein the predetermined operation state is a state in which a misfire has occurred.   2. The internal combustion engine state detection system according to claim 1, wherein the predetermined operation state is a state in which a variation in air-fuel ratio between cylinders has occurred.   The state detection system for an internal combustion engine according to any one of claims 1 to 3, wherein the road surface state variable is a variable indicating whether a road surface on which the vehicle is traveling is a paved road or an unpaved road. .   The execution device, when it is determined that the internal combustion engine is in the predetermined operation state by the determination processing, performs execution processing for operating the hardware to cope with the predetermined operation state. The state detection system for an internal combustion engine according to any one of claims 1 to 4. The determination process includes an output value calculation process of calculating the combustion state variable, which is an output value of the detection map, to which the values of the rotational waveform variable and the road surface state variable acquired through the acquisition process are input.
The execution device includes a first execution device mounted on the vehicle and a second execution device mounted on a separate device from the vehicle,
The first execution device executes the acquisition process and a vehicle-side reception process that receives a signal based on a calculation result of the output value calculation process,
The said 2nd execution device performs the said output value calculation process and the external side transmission process which transmits the signal based on the calculation result of the said output value calculation process to the said vehicle. 2. The state detection system for an internal combustion engine according to claim 1.   A data analysis device comprising the second execution device according to claim 6 and the storage device.   A vehicle comprising the first execution device according to claim 6.