JP2022036594A - Noise detection device - Google Patents

Abstract

To provide a noise detection device with which it is possible to easily construct a system capable of detecting noise occurrence with good accuracy.SOLUTION: Storage devices 34, 44 store mapping data 37, 47 having been trained by machine learning using training data 48 in which the presence of noise occurrence is added, as correct answer labels, to a plurality of vehicle parameters related to noise occurrence. This device accepts as input the plurality of vehicle parameters detected by various types of sensors 21-27 and outputs, using the mapping data 37 stored in the storage device 34, an output variable that indicates the probability of noise occurrence. The result data that indicates the result of assessment of noise occurrence by occupants is acquired using an external apparatus 51. The mapping data 37, 47 stored in the storage devices 34, 44 are trained using, as new training data 48A, data in which the result of the assessment is added as correct-answer labels to the plurality of vehicle parameters detected when the presence of noise occurrence is assessed by occupants.SELECTED DRAWING: Figure 1

Description

The present invention relates to a noise detection device that detects the generation of noise in a vehicle.

In a vehicle such as an automobile, various abnormal noises are generated as the vehicle is driven.
Conventionally, a noise detection device for detecting the generation of such noise has been proposed (for example, Patent Document 1). The noise detection device described in Patent Document 1 detects the generation of tapping noise generated in the power split mechanism (planetary gear) of a hybrid vehicle. In this device, vehicle parameters related to the generation of tapping noise (for example, the twist angle of the crankshaft damper and the torque difference between cylinders) are detected. Then, the vehicle parameter and the determination value are compared, and the occurrence of tapping noise is detected based on the comparison result.

Japanese Unexamined Patent Publication No. 2014-172595

Noise generated by driving a vehicle may reduce the ride quality of the vehicle if it gets into the ears of the occupants. Whether or not such noise gets into the ears of the occupants is determined by various factors such as the driving environment (congestion status, weather) of the vehicle and the open / closed state of the passenger compartment window. From this, in order to detect the generation of noise at a level that reduces the ride quality of the vehicle, a detection system is provided in which judgment values are set for all of the various factors and noise generation is detected based on those judgment values. It is conceivable to build. In this case, there is a problem that the man-hours required for adapting the detection system become large and the construction of the detection system becomes complicated.

The noise detection device for solving the above-mentioned problems is a noise detection device for detecting the generation of noise in a vehicle, in which a detection unit for detecting a plurality of vehicle parameters related to the generation of the noise and the presence / absence of the generation of the noise are detected. Stores a neural network machine-learned using a data acquisition unit that acquires result data indicating the results determined by the occupants and teacher data that indicates the presence or absence of noise generation as correct answer labels for the plurality of vehicle parameters. It is equipped with a storage unit. The noise detection device includes an output unit that inputs a plurality of vehicle parameters detected by the detection unit and outputs an output variable indicating the noise generation probability by using a neural network stored in the storage unit. .. The noise detection device assigns the result of the occupant's determination of the presence or absence of the noise to the plurality of vehicle parameters detected by the detection unit when the occupant determines the presence or absence of the noise as a correct answer label. A learning unit for learning a neural network stored in the storage unit is provided by using the data as teacher data.

According to the above configuration, teacher data in which a plurality of vehicle parameters related to noise generation and a correct answer label are set is prepared, and a neural network is trained using the teacher data to generate noise. It is possible to easily construct a system capable of accurately detecting the generation of noise in consideration of such various factors (vehicle parameters). Moreover, the result of the occupant actually determining the presence or absence of noise generation can be reflected in the neural network used for detecting the noise generation. Therefore, the noise generation detection standard can be set according to the user's standard.

The schematic block diagram of the noise detection apparatus of one Embodiment. A flowchart showing a vehicle control process. A flowchart showing the work by personnel. A flowchart showing the update process.

Hereinafter, an embodiment of the noise detection device will be described.
First, the structure of the vehicle to which the noise detection device of the present embodiment is applied will be described.
As shown in FIG. 1, the vehicle 10 includes an internal combustion engine 11, a power split mechanism 12, an automatic transmission 13, a drive wheel 14, a first motor generator 15, and a second motor generator 16.

The internal combustion engine 11 has four cylinders. A crankshaft damper 18 is provided on the crankshaft 17 which is the output shaft of the internal combustion engine 11. A power split mechanism 12 is connected to the crankshaft 17 of the internal combustion engine 11. The power split mechanism 12 is a planetary gear mechanism having a sun gear S, a ring gear R, and a carrier C. A crankshaft 17 is connected to the carrier C of the power split mechanism 12. The rotating shaft 15A of the first motor generator 15 is connected to the sun gear S. The rotation shaft 16A of the second motor generator 16 is connected to the ring gear shaft RA, which is the output shaft of the ring gear R. Further, the input shaft of the automatic transmission 13 is connected to the ring gear shaft RA. The left and right drive wheels 14 are connected to the output shaft of the automatic transmission 13 via a differential gear.

When the internal combustion engine 11 is driven and torque is input from the crankshaft 17 to the carrier C of the power split mechanism 12, the torque is split into the sun gear S side and the ring gear R side. When the first motor generator 15 operates as a motor and torque is input to the sun gear S of the power split mechanism 12, the torque is split into the carrier C side and the ring gear R side. When the second motor generator 16 operates as a motor and torque is input to the ring gear shaft RA, the torque is transmitted to the automatic transmission 13. Further, when the torque from the drive wheel 14 side is input to the second motor generator 16 via the ring gear shaft RA, the second motor generator 16 functions as a generator and generates a regenerative braking force in the vehicle 10. Is possible.

As shown in FIG. 1, the vehicle 10 has various sensors for detecting the state. As various sensors, for example, the crank angle sensor 21 for detecting the rotation speed (engine rotation speed NE) and the rotation phase (crank angle CA) of the crankshaft 17, and the operation amount of the accelerator pedal 19 (accelerator operation amount ACC) are used. An accelerator position sensor 22 for detection is provided. Further, the vehicle 10 includes a speed sensor 23 for detecting the rotation speed of the rotation shaft 15A of the first motor generator 15 (motor rotation speed NM1) and the rotation speed of the rotation shaft 16A of the second motor generator 16 (motor rotation speed). A speed sensor 24 for detecting NM2) is provided. In addition, the vehicle 10 is also provided with a vehicle speed sensor 25 for detecting the traveling speed (vehicle speed SPD), a window switch 26 for detecting whether or not the window of the vehicle interior is open, and the like. The vehicle 10 is also provided with a navigation system 27 and the like. The navigation system 27 is connected to a communication network 50, and acquires vehicle information such as weather information (rainfall amount) and road information (tunnel, congestion status) via the communication network 50. In this embodiment, various sensors correspond to detection units that detect a plurality of vehicle parameters related to the generation of noise.

The vehicle 10 includes a control device 30. Output signals of various sensors are input to the control device 30. The control device 30 includes a CPU 31, a peripheral circuit 32, a ROM 33, a storage device 34 as a storage unit, and a communication device 35. The CPU 31, the peripheral circuit 32, the ROM 33, the storage device 34, and the communication device 35 are communicably connected by the bus 36. Various programs are stored in the ROM 33 in advance for the CPU 31 to execute various controls. The mapping data 37 is stored in the storage device 34. The mapping M1 defined by the mapping data 37 outputs an output variable y indicating the probability of occurrence of tapping noise when the input variable x is input. A specific description of the map M1 will be described later. The peripheral circuit 32 includes a circuit that generates a clock signal that defines the internal operation, a power supply circuit, a reset circuit, and the like. The communication device 35 is a device for communicating with the center 40 via the communication network 50 outside the vehicle 10. In the present embodiment, the control device 30 corresponds to the output unit.

The CPU 31 controls the internal combustion engine 11, the first motor generator 15, the second motor generator 16, and the like by executing various programs stored in the ROM 33. Specifically, the CPU 31 calculates a vehicle required output, which is a required value of the output required for the vehicle 10 to travel, based on the accelerator operation amount ACC and the vehicle speed SPD. The CPU 31 determines the torque distribution of the internal combustion engine 11, the first motor generator 15, and the second motor generator 16 based on the vehicle required output. The CPU 31 controls the output of the internal combustion engine 11 and the power running and regeneration of the first motor generator 15 and the second motor generator 16 based on this torque distribution.

Further, the CPU 31 uses the mapping data 37 stored in the storage device 34 to execute various programs stored in the ROM 33, thereby causing noise in the vehicle 10 (specifically, tapping generated in the power dividing mechanism 12). It is determined whether or not noise) may occur. A specific description of the process for determining the generation of noise (noise determination process) will be described later.

Then, when the CPU 31 determines that noise may be generated through the noise determination process, the CPU 31 executes various programs stored in the ROM 33 to control (noise) to avoid the generation of noise in the vehicle 10. Avoidance control) is executed.

Specifically, the noise avoidance control is executed as follows. In the noise avoidance control, the engine rotation speed NE is set to be higher than that in the normal control (that is, when the noise avoidance control is not executed) for a predetermined period T after it is determined that noise may occur. The operation control of the internal combustion engine 11 is executed. Further, in the predetermined period T, the torque distribution of the first motor generator 15 and the second motor generator 16 is determined in a manner that satisfies the vehicle required output according to the output of the internal combustion engine 11 at this time, and the first motor generators thereof. The power running and regeneration of the 15 and the second motor generator 16 are controlled. By executing such noise avoidance control, the operating state of the internal combustion engine 11 is different from the operating state in which noise is likely to occur (for example, the operating state in which the crankshaft 17 resonates due to a misfire). Be made. As a result, the generation of noise can be suppressed.

An example of the mapping M1 defined by the mapping data 37 is a function approximation device, which is a fully connected forward propagation type neural network having one intermediate layer. Specifically, in the mapping M1 defined by the mapping data 37, the input variables x (1) to x (8) and the input variables x (0) as bias parameters have coefficients wFjk (j = 1 to m, k). By substituting each of the "m" values converted by the linear map defined by = 0 to 8) into the activation function f, the values of the nodes in the middle layer are determined. Further, the output variable y (1) is determined by substituting each of the values obtained by converting the values of the nodes of the intermediate layer into the activation function g by the linear mapping defined by the coefficient wSij (i = 1). The output variable y (1) is a variable indicating the probability that noise will occur.

The above neural network includes a softmax function. As a result, in the present embodiment, the output variable y (1) is set as a continuous value in a predetermined region in which the magnitude of the plausibility that noise actually occurs is larger than "0" and smaller than "1". It will be quantified. In the present embodiment, the larger the output variable y (1), the higher the probability of noise generation.

The map M1 defined by the map data 37 is generated as follows, for example. That is, first, various experiments and simulations are performed to detect a plurality of vehicle parameters related to the generation of tapping noise, and it is determined whether or not the tapping noise is generated. Then, a plurality of data in which the determination result of the presence or absence of noise generation is added as a correct answer label to the detection data of a plurality of vehicle parameters (specifically, the input variables x (1) to x (8)) are prepared, and these data are prepared. Is used as the teacher data 48 for machine learning to generate a trained mapping M1.

In this embodiment, the following values are used as the input variables x (1) to x (8).
The twist angle θ of the crankshaft damper 18 is substituted into the input variable x (1).
This twist angle θ is obtained as follows. First, based on the motor rotation speeds NM1 and NM2 and the gear ratio RG of the power splitting mechanism 12, the rotation speed NC of the carrier C of the power splitting mechanism 12 is derived from the relational expression "NC = (RG × NM1 + NM2) / (1 + RG)". Is calculated. Then, as shown in the relational expression "θ = 2π × ∫ (NE-NC) dt", the value obtained by subtracting the rotation speed NC from the engine rotation speed NE is the integrated value of the twist angle θ of the crankshaft damper 18. Is calculated as. In the vehicle 10, the larger the twist angle θ, the easier it is for tapping noise to occur, and the stronger the tapping noise when tapping noise occurs.

The difference in generated torque between the cylinders of the internal combustion engine 11 (torque difference between cylinders ΔTR) is substituted into the input variable x (2).
The torque difference between cylinders ΔTR is obtained as follows. First, the crank angle CA, the engine rotation speed NE, the moment of inertia around the output shaft of the internal combustion engine 11, the motor rotation speed NM1, the moment of inertia around the rotation shaft of the first motor generator 15, the output torque of the first motor generator 15, and Based on the gear ratio RG of the power split mechanism 12, the generated torques TR1, TR2, TR3, TR4 in each cylinder of the internal combustion engine 11 are calculated. Then, the value obtained by subtracting the minimum value MIN from the maximum value MAX among the generated torques TR1, TR2, TR3, and TR4 is calculated as the inter-cylinder torque difference ΔTR (= MAX-MIN). In the vehicle 10, the larger the torque difference ΔTR between cylinders is, the more likely it is that tapping noise is generated, and the stronger the tapping noise is when tapping noise is generated.

The vehicle speed SPD is substituted into the input variable x (3).
In the vehicle 10, the higher the vehicle speed SPD, the larger the background noise. Therefore, even if the tapping noise is generated, the tapping noise is less likely to be heard by the occupant.

A numerical value of the operating state of the window switch 26 is assigned to the input variable x (4).
When the window in the passenger compartment is open, the background noise is louder than when the window is closed, so even if tapping noise occurs, it is difficult for the occupant to hear the tapping noise.

The input variable x (5) is substituted with a numerical value of the amount of rainfall at the place where the vehicle 10 is traveling.
Information about rainfall is obtained via the navigation system 27 and the communication network 50. The more rainfall there is, the louder the noise inside the vehicle due to rainfall, so even if tapping noise occurs, it is difficult for the occupants to hear the tapping noise.

The input variable x (6) is substituted with a numerical value of information (tunnel, unpaved road, etc.) about the travel path of the vehicle 10.
Information about the travel path of the vehicle 10 is acquired via the navigation system 27 and the communication network 50. When the vehicle 10 is traveling on a tunnel or an unpaved road, the road noise becomes large, so that when the tapping noise is generated, the tapping noise is less likely to be heard by the occupant.

The input variable x (7) is substituted with a numerical value of the congestion status on the traveling path of the vehicle 10.
The congestion status on the travel path of the vehicle 10 is acquired via the communication network 50. In the vehicle 10, when the road is congested, the background noise becomes louder than when it is not, so that the tapping noise is hard to be heard by the occupant even when the tapping noise is generated.

A value indicating the volume of the navigation system 27 (including the audio system) is assigned to the input variable x (8).
The louder the volume of the navigation system 27, the louder the background noise. Therefore, when the tapping noise is generated, the tapping noise is less likely to be heard by the occupant.

The noise detection device of the present embodiment has a center 40 that acquires and manages various information of the vehicle 10.
The center 40 includes a CPU 41, a peripheral circuit 42, a ROM 43, a storage device 44 as a storage unit, and a communication device 45. The CPU 41, the peripheral circuit 42, the ROM 43, the storage device 44, and the communication device 45 are communicably connected by the bus 46. Various programs are stored in the ROM 43 in advance for the CPU 41 to execute various controls. The storage device 44 stores the mapping data 47 and the teacher data 48 used for learning the mapping data 47. The CPU 41 executes machine learning using the teacher data 48 by executing various programs stored in the ROM 43 to generate a learned map (specifically, the map data 47). In this embodiment, the center 40 corresponds to the learning unit.

The communication device 45 is a device for communicating with the vehicle 10 via the external communication network 50. The communication device 45 enables data transmission / reception between the center 40 and the vehicle 10. The CPU 41 of the center 40 transmits the learned mapping data 47 to the control device 30 of the vehicle 10 when machine learning based on the teacher data 48 is performed by executing various programs stored in the ROM 43. Then, when the control device 30 of the vehicle 10 receives the mapping data 47 transmitted from the center 40, the mapping data 47 updates the mapping data 37 stored in the storage device 34.

The noise detection device of the present embodiment has an external device 51 that can be connected to the control device 30 of the vehicle 10 and the center 40 via the communication network 50. The external device 51 is located at a store or a maintenance shop. An example of the external device 51 is a personal computer. Through the operation of the external device 51, it is possible to transmit arbitrary information such as information interviewed and investigated by the user to the center 40 together with the vehicle information. In the present embodiment, the external device 51 corresponds to a data acquisition unit that acquires result data indicating a result in which the occupant determines whether or not noise is generated.

Hereinafter, processing related to vehicle control (vehicle control processing) including noise determination processing and noise avoidance control will be specifically described.
FIG. 2 shows an execution procedure of the vehicle control process. The series of processes shown in the flowchart of the figure is executed by the control device 30 of the vehicle 10 as interrupt processes at predetermined intervals.

As shown in FIG. 2, in this process, first, various values for generating the input variables x (1) to x (8) of the map M1 are acquired (step S11), and various values are used. The input variables x (1) to x (8) of the map M1 are generated in (step S12).

Specifically, in the process of step S12, the twist angle θ of the crankshaft damper 18 is calculated and substituted into the input variable x (1), and the torque difference between cylinders ΔTR is calculated and substituted into the input variable x (2). Then, the vehicle speed SPD is detected and assigned to the input variable x (3). Further, a numerical value of the operating state of the window switch 26 is calculated and substituted into the input variable x (4), a numerical value of the rainfall amount is calculated and substituted into the input variable x (5), and the vehicle 10 is used. A numerical value of information (tunnel, unpaved road, etc.) about the traveling road of No. 1 is calculated and assigned to the input variable x (6). Further, a numerical value of the congestion status on the traveling road of the vehicle 10 is calculated and substituted into the input variable x (7), and a value indicating the volume of the navigation system 27 is detected and substituted into the input variable x (8). To.

After that, the input variables x (1) to the input variables x (8) generated in the process of step S12 and the input variables x (0) as bias parameters are input to the mapping M1 defined by the mapping data 37. An output variable y, which is a variable indicating the probability that tapping noise will occur (noise generation probability), is calculated (step S13). In this embodiment, the noise generation probability is output from the map M1 defined by the map data 37 in this way.

Then, when the noise generation probability is 50% or more (step S14: YES), specifically, when the output variable y is “0.50” or more, there is a high possibility that tapping noise will occur. The noise avoidance control described above is executed for a predetermined period T (step S15). In this case, the generation of tapping noise is suppressed through the execution of noise avoidance control. Further, in this case, the input variables x (1) to the input variables x (8) are stored in the storage device 34 together with the history of execution of the noise avoidance control (history of execution) (step S16). After the process of step S16, this process is terminated.

On the other hand, when the noise generation probability is less than 50% and 45% or more (step S14: NO and step S17: YES), specifically, the output variable y is less than "0.50" and "0. If it is 45 ”or higher, the noise avoidance control is not executed (the process of step S15 is jumped). However, in this case, although the noise generation probability is not high to the level that causes tapping noise, it is assumed that the input variable x (1) to the input variable x (8) at this time are noise avoidance control. It is stored in the storage device 34 together with the history not executed (history without execution) (step S18). After the process of step S18, this process is terminated.

On the other hand, when the noise generation probability is less than 45% (step S14: NO and step S17: NO), the process for executing the noise avoidance control and the input variables x (1) to the input variables x (8) are stored. This process is terminated without executing the process to be performed.

In this embodiment, as described above, a teacher in which detection data of a plurality of vehicle parameters related to the generation of tapping noise (specifically, input variables x (1) to x (8)) and a correct label are set. Data 48 is prepared. Then, by learning a neural network (specifically, the mapping M1 defined by the mapping data 37) using the teacher data 48, noise is generated in consideration of various factors (vehicle parameters) related to the generation of noise. It is possible to construct a system that can accurately detect. According to the present embodiment, the detection system is optimized as compared with the case where judgment values are set for all of a plurality of vehicle parameters and a detection system for detecting noise generation based on the judgment values is constructed. Since the time and effort required for this can be significantly reduced, a detection system can be easily constructed.

In the present embodiment, the data stored in the storage device 34 in the vehicle control process (FIG. 2), specifically, the input variables x (1) to the input variables x (8) and the history with execution or the history without execution are set as a set. A new teacher data 48A is created based on the new data, and the mapping data 37 and 47 are learned using the teacher data 48A.

Hereinafter, the details of the process for learning the map data 37 and 47 will be described with reference to FIGS. 3 and 4. FIG. 3 shows a procedure for executing work performed by personnel of a dealer or a maintenance shop when the vehicle 10 is brought to the dealer or the maintenance shop. FIG. 4 shows an execution procedure of a process (update process) for updating the map data 37 based on the data input to the center 40. The series of processes shown in FIG. 4 is a process executed by the center 40 as an interrupt process for each predetermined cycle.

First, when the vehicle 10 is brought to a store or a maintenance shop, the control device 30 of the vehicle 10 is connected to the center 40 and the external device 51 via the communication network 50.
In addition, the personnel of the dealer or the maintenance shop conduct an interview survey on the usability of the vehicle 10 including information on the presence or absence of tapping noise for the user as a occupant of the vehicle 10.

Then, as shown in FIG. 3, when there is data stored in the control device 30 of the vehicle 10 in the vehicle control process (step S21: YES), the data, specifically, input variables x (1) to input variables. The data in which x (8) and the history with execution or the history without execution are set is associated with the result data indicating the result of determining the presence or absence of tapping noise by the user who is the occupant of the vehicle 10. Then, it is transmitted to the center 40 (step S22). The work of transmitting data to the center 40 is performed through the operation of the external device 51 by the personnel of the store or the maintenance shop. The work of associating the data (including the input variable x (1) to the input variable x (8)) stored in the control device 30 of the vehicle 10 with the result data in the vehicle control process is performed by the input variable x (1). It can be performed by comparing the date and time when the input variable x (8) was calculated with the date and time when the tapping noise was generated obtained in the interview survey.

If there is no data stored in the control device 30 of the vehicle 10 in the vehicle control process (step S21: NO), the process of transmitting the data to the center 40 is not performed (jumping the process of step S22). , This process is terminated.

The center 40 learns the mapping data 37 and 47 as follows based on the data received from the control device 30 of the vehicle 10 and the external device 51.
As shown in FIG. 4, when there is data received by the center 40 (step S31: YES), the received data includes a history of execution (step S32: YES), and the user determines that noise has occurred. When the result data indicating that the noise has been generated is included (step S33: YES), the teacher data 48A indicating that noise has occurred is created and stored in the storage device 44 of the center 40 (step S34). Specifically, assuming that the data is when the user actually recognizes the generation of noise (noise generation abnormality), the input variables x (1) to x (8) of the data received at this time and the correct answer label (noise generation). Teacher data 48A is formed as a set with (yes).

On the other hand, when the data received by the center 40 includes a history of execution (step S32: YES) and includes result data indicating that the user has not determined that noise has occurred (step S33:). NO), teacher data 48A indicating that noise is not generated is created and stored in the storage device 44 of the center 40 (step S35). Specifically, assuming that the data is in a situation where the user does not recognize the noise generation abnormality, the input variables x (1) to x (8) of the data received at this time and the correct answer label (no noise generation) are set. The teacher data 48A is formed.

On the other hand, when the data received by the center 40 includes a history of no execution (step S32: NO) and includes result data indicating that the user has determined that noise has occurred (step S36: YES). , Teacher data 48A indicating that noise has occurred is created and stored in the storage device 44 of the center 40 (step S37). Specifically, assuming that the data is when the user actually recognizes the noise generation abnormality, the input variables x (1) to x (8) of the data received at this time and the correct label (with noise generation) are set. The teacher data 48A is formed.

When the new teacher data 48A is created in this way (steps S34, S35, S37), the neural network stored in the storage device 44 of the center 40 is stored by executing machine learning using the teacher data 48A. The network (specifically, mapping data 47) is updated (step S38). In the process of step S38, the map data 47 is learned and updated by using the backpropagation method.

The map data 47 updated at this time is transmitted to the control device 30 in order to update the map data 37 stored in the control device 30 of the vehicle 10 (step S39). As a result, the mapping data 47 transmitted from the center 40 and received by the control device 30 of the vehicle 10 is replaced with the mapping data 37 stored in the control device 30 of the vehicle 10 at this time. The stored mapping data 37 will be updated. After that, the noise determination process is executed by using the map M1 defined by the updated map data 37. At this time, in the mapping data 37 transmitted from the center 40 and received by the control device 30 of the vehicle 10, in addition to the learning update by the data of the user's vehicle 10, the learning update by the data of the vehicle 10 of another user is added. Also includes minutes.

When the data received by the center 40 includes a history of no execution (step S32: NO) and includes result data indicating that the user has not determined that noise has occurred (step S36:). NO), new teacher data 48A is not created (the process of step S37 is jumped). Then, also in this case, the mapping data 47 stored in the storage device 44 of the center 40 is transmitted to the control device 30 for use in updating the mapping data 37 stored in the control device 30 of the vehicle 10. (Step S39). Further, even when there is no data newly received by the center 40 (step S31: NO), the new teacher data 48A is not created, and the mapping data 47 stored in the storage device 44 of the center 40 is the vehicle. It is transmitted to the control device 30 for use in updating the mapping data 37 stored in the control device 30 of the 10 (step S39). Through such processing, the mapping data 37 stored in the control device 30 of the user's vehicle 10 is replaced with the latest mapping data 47 updated by the data of the other user's vehicle 10 and updated.

Here, when tapping noise occurs in the vehicle 10, there are individual differences in whether or not the tapping noise is unpleasant. Therefore, it can be said that it is difficult to set the standard for determining the occurrence of tapping noise to a standard level that everyone can understand.

In the present embodiment, the teacher data 48 is created with the result of the occupant determining whether or not noise is generated in the vehicle 10 with respect to the detection data of the plurality of vehicle parameters as the correct label. Then, using the teacher data 48, the mapping data 37 and 47 stored in the center 40 (and the control device 30 of the vehicle 10) are learned and updated.

As a result, the result of the occupant actually determining whether or not noise is generated in the vehicle 10 can be reflected in the mapping data 37 used for detecting the generation of noise. Therefore, the noise generation detection standard in the vehicle 10 can be set according to the user's standard.

Therefore, according to the present embodiment, it is possible to accurately determine that the vehicle 10 is likely to generate noise by using the mapping data 37, and the noise avoidance control is executed based on the determination result. Can be done. As a result, the generation of tapping noise can be accurately suppressed, and the deterioration of the ride quality of the vehicle 10 due to the generation of tapping noise can be suitably suppressed. Moreover, since it is possible to suppress the unnecessary execution of the noise avoidance control, it is also possible to suppress the deterioration of the fuel efficiency performance due to the execution of the noise avoidance control.

In the present embodiment, the mapping data 47 stored in the storage device 44 of the center 40 is learned and updated based on the data transmitted from the control device 30 of the vehicle 10 and input to the center 40. Then, the map data 37 stored in the control device 30 is updated, such that the updated map data 47 is overwritten and replaced with the map data 37 stored in the control device 30 of the vehicle 10. .. This makes it possible to learn the map M1 stored in the control device 30 of each vehicle 10 based on the actual data collected from a large number of vehicles 10 on the market. Therefore, the noise generation detection standard in the vehicle 10 can be suitably set according to the standard user standard.

As described above, according to the present embodiment, the effects described below can be obtained.
(1) A teacher data 48 in which a plurality of vehicle parameters related to the generation of noise and a correct label are set is prepared, and a neural network (mapping data 37) is trained using the teacher data 48. .. Therefore, it is possible to easily construct a system capable of accurately detecting the generation of noise in consideration of various factors related to the generation of noise.

(2) New teacher data 48A is created by assigning the result of the occupant's determination as to whether or not noise is generated to the detection data of a plurality of vehicle parameters as a correct answer label. Then, using the teacher data 48A, the mapping data 37 and 47 stored in the center 40 (and the control device 30 of the vehicle 10) are learned and updated. As a result, the result of the occupant actually determining the presence or absence of noise generation can be reflected in the mapping data 37 and 47 used for detecting the noise generation, so that the noise generation detection standard is adjusted to the user's standard. Can be determined.

The above embodiment can be modified and implemented as follows. The above embodiment and the following modified examples can be implemented in combination with each other within a technically consistent range.

As the vehicle parameter used as the input variable x of the mapping M1, any value can be adopted as long as it is a vehicle parameter related to the generation of tapping noise.
-As the noise avoidance control, any vehicle control can be adopted as long as the generation of noise in the vehicle 10 can be suppressed. As the noise avoidance control, for example, when the operation position of the shift lever is the parking position, the operation control of the second motor generator 16 may be executed so as to generate torque in the second motor generator 16. .. According to such control, the gear teeth in the meshed state can be pressed inside the power split mechanism 12, so that the generation of tapping noise can be suppressed. In addition, as noise avoidance control, it is also possible to execute the operation control of the internal combustion engine 11 so that the output torque of the internal combustion engine 11 becomes smaller than that in the normal control. According to such control, the torque intermittently generated in each cylinder of the internal combustion engine 11 can be suppressed to be small, so that the generation of tapping noise due to the torque can be suppressed.

The process of step S17 and the process of step S18 in the vehicle control process (FIG. 2) may be omitted. Further, the process of step S36 and the process of step S37 of the update process (FIG. 4) can be omitted.

The noise determination process (processes of steps S11 to S13 in FIG. 2) may be executed by the center 40. In this case, the transmission / reception of a plurality of vehicle parameters and the transmission / reception of the output variable y between the control device 30 of the vehicle 10 and the center 40 may be performed through the communication network 50. In this case, the CPU 41 of the center 40 uses the mapping data 47 stored in the storage device 44 to execute various programs stored in the ROM 43, thereby indicating an output variable y indicating the probability of noise generation. Can be output. In such a configuration, the mapping data 37 stored in the storage device 34 of the control device 30 of the vehicle 10 is omitted, and the execution program of the noise determination process stored in the ROM 33 of the control device 30 is omitted. Can be done.

-The switch may be displayed on the touch panel type display device, and the occupant may operate the switch when the occupant determines that noise has occurred. According to the same configuration, the operation information of the switch can be used as the result data showing the result of the occupant determining whether or not the noise is generated. In the above configuration, the input variable x (1) to the input variable x (8), the history with execution (or the history without execution), and the result data (switch) showing the result of the occupant determining whether or not noise is generated. The data combined with the operation information) can be transmitted to the center 40 through the communication network 50.

The detection data of a plurality of vehicle parameters (in the above embodiment, the input variables x (1) to x (8)) are not limited to being stored in the control device 30 of the vehicle 10, but the control device 30 of the vehicle 10 is not limited to being stored. May be transmitted to the center 40 and stored in the storage device 44 of the center 40. In the same configuration, the data transmitted to and stored in the center 40 is the data in which the detection data of a plurality of vehicle parameters is associated with the information for identifying the vehicle 10 (vehicle body number, user information, etc.) and the information indicating the detection date and time. Is preferable. According to the same configuration, a process of creating teacher data 48A (steps S32 to S36 in FIG. 4) is executed based on the detection data of a plurality of vehicle parameters stored in the storage device 44 of the center 40. Can be done.

In the vehicle control process (FIG. 2), the process of creating the teacher data 48A based on the data stored in the storage device 34 of the vehicle 10 and the process of learning the mapping data 37 using the teacher data 48A are performed by the vehicle 10. It may be executed by the control device 30. In the same configuration, the center 40 can be omitted.

-In the above embodiment, as the neural network, a neural network having one intermediate layer is exemplified, but the number of intermediate layers may be two or more.
-In the above embodiment, the fully connected forward propagation type neural network is exemplified as the neural network, but the neural network is not limited to this. For example, as the neural network, a regression coupling type neural network may be adopted.

-The above embodiment is also used for a device that detects noise other than tapping noise, such as noise generated by an in-vehicle inverter and noise generated by an internal combustion engine 11, and a device that executes noise avoidance control for avoiding the generation of the noise. The noise detection device for the internal combustion engine can be applied.

10 ... Vehicle 21 ... Crank angle sensor 22 ... Accelerator position sensor 23 ... Speed sensor 24 ... Speed sensor 25 ... Vehicle speed sensor 26 ... Window switch 27 ... Navigation system 30 ... Control device 34 ... Storage device 37 ... Mapping data 40 ... Center 44 ... Storage device 47 ... Mapping data 48, 48A ... Teacher data 50 ... Communication network 51 ... External device

Claims (1)

In a noise detection device that detects the generation of noise in a vehicle
A detector that detects a plurality of vehicle parameters related to the generation of noise,
A data acquisition unit that acquires result data indicating the result of the occupant's determination of the presence or absence of noise.
A storage unit that stores a neural network machine-learned using teacher data in which the presence or absence of noise is given as a correct label for the plurality of vehicle parameters.
An output unit that inputs a plurality of vehicle parameters detected by the detection unit and outputs an output variable indicating the noise generation probability by using a neural network stored in the storage unit.
When the occupant determines whether or not the noise is generated, the data obtained by assigning the result of the occupant's determination as the correct answer label to the plurality of vehicle parameters detected by the detection unit as the teacher data is used as the teacher data. Using the learning unit to learn the neural network stored in the storage unit,
A noise detection device characterized by being equipped with.

Images