Abnormal sound inspection device for vehicles

Abstract

The present invention discloses an abnormal sound inspection device for a vehicle. A neural network is stored in a storage device of the abnormal sound inspection device. The neural network is machine-learned in a manner that takes an original sound characteristic value representing the characteristics of a sound emitted by a transmission and an evaluation sound characteristic value representing the characteristics of a sound reaching a vehicle cabin as inputs and takes a path component characteristic value representing the characteristics of a vibration transmission of a vehicle component located on a vibration transmission path from the transmission to the vehicle cabin as outputs. An execution device of the abnormal sound inspection device calculates the output of the neural network that takes the measured values ​​of the original sound characteristic value and the evaluation sound characteristic value as inputs as an estimated value of the path component characteristic value.

Description

Technical Field

[0001] The present invention relates to an abnormal sound inspection device for a vehicle. Background Art

[0002] Japanese Patent Application Laid-Open No. 2008-151538 describes a device for determining the presence of abnormality of a transmission or the like using a neural network that performs machine learning to take frequency components of sounds in a vehicle interior as input and outputs a determination result of the presence of abnormality. Summary of the invention

[0003] The sound generated by the transmission due to the rattling of the gears is sometimes heard as an uncomfortable abnormal sound in the vehicle cabin. The sound generated by the transmission reaches the vehicle cabin through multiple paths. The sound reaching the vehicle cabin changes according to the vibration transmission characteristics of the vehicle components located in its transmission path. Therefore, even if the sound generated by the transmission itself is within the range that can be allowed by design, it may sometimes be heard by passengers as an uncomfortable abnormal sound according to the vibration transmission characteristics of the vehicle components located in the transmission path of the sound and vibration until it reaches the vehicle cabin. In this way, multiple vehicle components are related to abnormal sounds.

[0004] When using conventional analytical methods to perform precise abnormal sound inspections for determining vehicle components that are the cause of abnormal sound, predicting the presence or absence of abnormal sound generation, etc., it is necessary to measure the vibration transfer characteristics of each vehicle component related to the abnormal sound. However, as described above, there are many vehicle components related to abnormal sound, and some of them include components whose vibration transfer characteristics cannot be measured unless they are removed from the vehicle. Therefore, it is not easy to perform precise abnormal sound inspections.

[0005] Here, a vehicle component that is a source of abnormal sound is referred to as a sound source component, a vehicle component located in a vibration transmission path from the sound source component to an evaluation position of the abnormal sound is referred to as a path component, a value indicating a vibration transmission characteristic of the path component is referred to as a path component characteristic value, a value indicating a characteristic of a sound emitted by the sound source component is referred to as an original sound characteristic value, a value indicating a characteristic of a sound reaching the evaluation position of the abnormal sound is referred to as an evaluation sound characteristic value, one of the path component characteristic value and the evaluation sound characteristic value is referred to as a first characteristic value, and the other is referred to as a second characteristic value. In this case, the abnormal sound inspection device for a vehicle includes: a storage device storing a neural network that takes the original sound characteristic value and the first characteristic value as input and outputs the second characteristic value, the neural network being learned by using the measured values ​​of the original sound characteristic value, the first characteristic value, and the second characteristic value as training data; and an execution device that performs an estimation process of calculating the output of the neural network that takes the measured values ​​of the original sound characteristic value and the first characteristic value as input as an estimated value of the second characteristic value.

[0006] In the neural network stored in the storage device of the abnormal sound inspection device, the relationship between the original sound characteristic value, the evaluation sound characteristic value, and the path component characteristic value is learned. Therefore, if the neural network is used, it is possible to calculate the estimated value of the path component characteristic value based on the measured values ​​of the original sound characteristic value and the evaluation sound characteristic value, or to calculate the estimated value of the evaluation sound characteristic value based on the measured values ​​of the original sound characteristic value and the path component characteristic value. As a result, even if the path component is not removed from the vehicle for measurement, its vibration transfer characteristics can be estimated, or even if the path component is not installed on the vehicle, it can be estimated whether an abnormal sound is generated when the path component is installed on the vehicle. Therefore, it is possible to simply and accurately perform abnormal sound inspection.

[0007] The input of the neural network may include a driving state quantity representing the driving state of the vehicle, and the training data may include a measured value of the driving state quantity when the evaluation sound characteristic value is measured. In this case, the learning of the neural network and the estimation of the second characteristic value can be performed in a form reflecting the changes of the original sound characteristic value and the evaluation sound characteristic value based on the driving state of the vehicle.

[0008] In addition, as the path component characteristic value, for example, the transmission coefficient and phase of the path component vibration can be used. In addition, as the original sound characteristic value and the evaluation sound characteristic value, either the waveform data or the spectrum of the sound can be used.

[0009] Furthermore, when the neural network is configured so that the path component characteristic value is set as the second characteristic value, the execution device is configured to perform a determination process in which it is determined whether the path component is the cause of the abnormal sound based on the calculated value of the path component characteristic value in the estimation process in a state where the generation of the abnormal sound is confirmed. In this case, the path component that is the cause of the abnormal sound can be identified based on the determination result of the determination process. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Features, advantages, and technical and industrial significance of exemplary embodiments of the present invention will be described below with reference to the accompanying drawings, in which like symbols represent like elements, and in which:

[0011] Figure 1 It is a diagram schematically showing the structure of a first embodiment of the abnormal sound inspection device for a vehicle.

[0012] Figure 2 : is a diagram schematically showing the structure of a neural network used in the abnormal sound inspection device.

[0013] Figure 3 It is a flowchart of a determination routine executed by the execution device of the abnormal sound inspection device.

[0014] Figure 4 This is a diagram schematically showing a configuration of a neural network used in the second embodiment of the abnormal sound inspection device for a vehicle.

[0015] Figure 5 This is a diagram schematically showing a configuration of a neural network used in the third embodiment of the abnormal sound inspection device for a vehicle. DETAILED DESCRIPTION

[0016] (First embodiment)

[0017] Below, refer to Figure 1 to Figure 3 , a first embodiment of the abnormal sound inspection device for a vehicle is described in detail.

[0018] <About abnormal sound of the vehicle to be inspected>

[0019] First, refer to Figure 1 , the structure of the drive system of the vehicle 10 to be inspected by the abnormal sound inspection device of this embodiment is described. In the vehicle 10, an engine 11 is mounted as a drive source. The output of the engine 11 is connected to the wheel shaft 15 via the transmission 12, the propeller shaft 13, and the differential gear mechanism 14. The engine 11 and the transmission 12 are suspended on the vehicle body 18 via the engine mount 16 and the transmission mount 17. The drive system of the vehicle 10 includes these engine 11, transmission 12, propeller shaft 13, differential gear mechanism 14, wheel shaft 15, etc. There is a sound insulation member 21 between the structural components of these drive systems and the vehicle cabin 20 where passengers ride.

[0020] In addition, an electronic control unit 19 is mounted in the vehicle 10. A driving state quantity, which is a state quantity indicating the driving state of the vehicle 10, is input to the electronic control unit 19 from sensors provided in various parts of the vehicle 10. The driving state quantity includes quantities indicating the operating state of the engine 11, such as the vehicle speed V, the output speed NE of the engine 11, the output torque TE, and the cooling water temperature. In addition, quantities indicating the operating state of the transmission 12, such as the output speed NO, the output torque TO, the gear, and the working oil temperature of the transmission 12, are also included in the driving state quantity. Then, the electronic control unit 19 controls the driving state of the vehicle 10, such as controlling the operating state of the engine 11 and switching the gears of the transmission 12, based on the input driving state quantity.

[0021] During the travel of such a vehicle 10, the gear rattling sound is sometimes generated in the transmission 12. Moreover, the gear rattling sound is sometimes transmitted to the cabin 20 as an abnormal sound that makes passengers feel uncomfortable. The vibration generated in the transmission 12 reaches the cabin 20 via various paths. The sound heard by the passengers in the cabin 20 is the sound formed by the combination of the vibrations transmitted via various paths. In addition, as the transmission path of the vibration from the transmission 12 in the vehicle 10 to the cabin 20, there are, for example, the following paths (I) to (V). There are many paths such as the following: (I) A path from the transmission 12 to the cabin 20 via the sound insulation 21. (II) A path from the transmission 12 to the vehicle body 18 via the transmission mount 17, and from the vehicle body 18 to the cabin 20. (III) A path from the transmission 12 to the vehicle body 18 via the engine 11 and the engine mount 16, and from the vehicle body 18 to the cabin 20. (IV) A path from the transmission 12 to the propeller shaft 13, and from the propeller shaft 13 to the cabin 20 via the vehicle body 18. (V) A path from the transmission 12 to the differential gear mechanism 14 via the propeller shaft 13, and from the differential gear mechanism 14 to the cabin 20 via the vehicle body 18, etc. How the tooth rattle sound generated in the transmission 12 is heard by the passengers in the cabin 20 is greatly influenced by the vibration transmission characteristics of the vehicle components located in such a vibration transmission path.

[0022] In the following description, vehicle components located in each vibration transmission path from the transmission 12 as the source of abnormal sound to the vehicle interior 20 as the evaluation position of abnormal sound are described as path components. The vibration transmission characteristics of each path component have deviations during manufacturing and changes over time. In addition, the cause of abnormal sound is sometimes not the transmission 12 as the source, but the path component. The abnormal sound inspection device 30 of this embodiment is configured as a device for inspecting the path component that is the cause when an abnormal sound caused by a path component is generated.

[0023] <Structure of abnormal sound inspection device>

[0024] like Figure 1 As shown, the abnormal sound inspection device 30 of this embodiment is constituted as an electronic computer including an execution device 31 that executes a process related to the abnormal sound inspection and a storage device 32 that stores programs and data used in the process. The storage device 32 stores a neural network 33 for abnormal sound inspection.

[0025] The measuring device 34 is connected to the abnormal sound inspection device 30. A microphone 35, a pulse hammer 36 for an excitation test, and an acceleration sensor 37 are connected to the measuring device 34. Such a measuring device 34 is used to measure the original sound characteristic value and the evaluation sound characteristic value used in the abnormal sound inspection for determining the path component that is the cause of the abnormal sound. The original sound characteristic value is a value indicating the characteristic of the sound emitted by the vehicle component that is the source of the abnormal sound. In the following description, the vehicle component that is the source of the abnormal sound is recorded as the sound source component. In the present embodiment, the transmission 12 is the sound source component. The evaluation sound characteristic value is a value indicating the characteristic of the sound reaching the evaluation position of the abnormal sound from the sound source component. In the present embodiment, a predetermined position in the vehicle cabin 20 is set as the evaluation position.

[0026] <Measurement of original sound characteristic values ​​and evaluation sound characteristic values>

[0027] Next, the measurement of the original sound characteristic value and the evaluation sound characteristic value will be described. In the present embodiment, the engine 11 of the vehicle 10 to be inspected for abnormal sound is stopped, and the original sound characteristic value and the evaluation sound characteristic value are measured in the state where the vehicle 10 is stopped. In addition, during the measurement, the microphone 35 is set at a predetermined position in the vehicle cabin 20 set as the evaluation position, and the acceleration sensor 37 is installed on the transmission 12. Then, in this state, the transmission 12 is pulsed excited by the pulse hammer 36. After the pulse excitation, the measuring device 34 acquires the outputs of the acceleration sensor 37 and the microphone 35 at predetermined sampling periods. In the present embodiment, the time series data of the output of the acceleration sensor 37 acquired by the measuring device 34 at this time is used as the measurement value of the original sound characteristic value. In addition, in the present embodiment, the time series data of the output of the microphone 35 acquired by the measuring device 34 at this time is used as the measurement value of the evaluation sound characteristic value. That is, in the present embodiment, the waveform data of the vibration of the transmission 12 generated by the pulse excitation is measured as the original sound characteristic value. In addition, in the present embodiment, waveform data of the sound in the vehicle interior 20 when measuring the original sound characteristic value is measured as the evaluation sound characteristic value.

[0028] <Structure of Neural Network>

[0029] During the abnormal sound inspection, the abnormal sound inspection device 30 acquires the measured values ​​of the original sound characteristic value and the evaluation sound characteristic value in the vehicle 10 to be inspected from the measuring device 34. Then, the abnormal sound inspection device 30 performs the abnormal sound inspection using the acquired measured values ​​and the neural network 33 stored in the storage device 32. The structure of the neural network 33 used for such abnormal sound inspection is described below.

[0030] like Figure 2As shown, the neural network 33 has an input layer having "n" nodes, an intermediate layer having "m" nodes, and an output layer having "p" nodes. In addition, in the following description, "i" represents an arbitrary integer greater than 1 and less than n, "j" represents an arbitrary integer greater than 1 and less than m, and "k" represents an arbitrary integer greater than 1 and less than p.

[0031] exist Figure 2 , the input values ​​to each node of the input layer are represented as X1, X2, ..., Xn. The measured values ​​of the original sound characteristic values ​​are input to X1 to Xa in the input values, and the measured values ​​of the evaluation sound characteristic values ​​are input to Xa+1 to Xn. Specifically, the time series data of the output of the acceleration sensor 37 measured in the above-mentioned excitation test is input to X1 to Xa. In addition, the time series data of the output of the microphone 35 measured in the excitation test is input to Xa+1 to Xn.

[0032] In addition, Figure 2 In the example, the input values ​​to each node in the middle layer are represented as U1, U2, ..., Um, and the output values ​​of each node in the middle layer are represented as Z1, Z2, ..., Zm. The input value Uj of each node in the middle layer is calculated as the sum of the values ​​obtained by multiplying the input values ​​X1, X2, ..., Xn of the input layer by the weight Wij. The output value Zj of each node in the middle layer is calculated as the return value of the activation function F with the input value Uj of the node as the independent variable. In the present embodiment, the Sigmoid function is used as the activation function F.

[0033] Furthermore, in Figure 2 , the input values ​​to each node of the output layer are represented as Y1, Y2, ..., Yp. The sum of the values ​​obtained by multiplying the output value Zj of each node of the intermediate layer by the weight Vjk is input to Y1~Yp. Then, Y1~Yp, which are the input values ​​of each node of the output layer, are the outputs of the neural network 33. In the neural network 33, Y1~Yp represent values ​​showing the vibration transfer characteristics of one of the path components, such as the propeller shaft 13. Specifically, Y1~Yb respectively represent the vibration transfer coefficient, that is, the gain, of each frequency of the vibration transfer function of the path component. In addition, Yb+1~Yp respectively represent the phase of each frequency of the vibration transfer function of the path component.

[0034] As described above, the neural network 33 is configured to receive the original sound characteristic value and the evaluation sound characteristic value as input, and to receive the vibration transfer characteristic value of the path component as output. In this embodiment, the evaluation sound characteristic value corresponds to the first characteristic value, and the vibration transfer characteristic value of the path component corresponds to the second characteristic value.

[0035] Incidentally, a plurality of neural networks 33 corresponding to different path components are stored in the storage device 32. As the path component for setting the neural network 33, a component with a high possibility of being the cause of the abnormal sound is selected. That is, a path component with a large change in vibration when passing through the component, individual difference in vibration transfer characteristics, and large change over time. In addition, the input value of each node of the output layer of the neural network 33 may include the vibration transfer coefficient and phase of a plurality of inspection target path components, and the neural networks 33 of the plurality of path components may be aggregated into one.

[0036] <Learning of Neural Networks>

[0037] Next, the method of generating such a neural network 33, that is, the learning of the neural network 33, will be described. The learning of the neural network 33 is performed by a learning computer. The storage device 32 of the abnormal sound inspection device 30 stores the neural network 33 that has been learned by the learning computer, that is, the learned network.

[0038] When learning the neural network 33, the original sound characteristic value and the evaluation sound characteristic value are measured by vibration test in a plurality of vehicles 10 of the same model. In addition, the vibration transfer characteristic value of the path component is measured in each vehicle 10. The vibration transfer characteristic value of the path component is measured, for example, by vibration test in a path component alone. Then, for each vehicle 10, a data group is created that summarizes the original sound characteristic value, the evaluation sound characteristic value, and the vibration transfer characteristic value of the path component.

[0039] The learning of the neural network 33 is performed using the training data containing multiple data groups produced in this way. Specifically, first, the values ​​of the original sound characteristic values ​​and the evaluation sound characteristic values ​​in the data group are input into the input layer of the neural network 33 as the values ​​of X1 to Xn. Then, the error back propagation method is used to correct the values ​​of each weight Wij, Vjk in such a way that the error between the values ​​of Y1 to Yp output by the neural network 33 for these inputs and the values ​​of the vibration transfer characteristic values ​​of the path components in the data group becomes smaller. This correction process of the weights Wij, Vjk is repeated until the above error becomes less than a predetermined value. Then, when the above error becomes less than a predetermined value, it is determined that the learning of the neural network 33 is completed.

[0040] <Abnormal sound check>

[0041] Next, an embodiment of the abnormal sound inspection performed by the abnormal sound inspection device 30 of the present embodiment will be described. The abnormal sound inspection here is performed to determine the cause of the abnormal sound when a user takes a vehicle 10 that generates an abnormal sound to a dealer or the like. In the case where the cause of the abnormal sound is the transmission 12 that is the source of the abnormal sound, that is, when the transmission 12 generates a loud sound, it is easy to find out that the cause is the transmission 12 through human hearing. However, in the case where the cause of the abnormal sound is a path component, it is difficult to determine the cause only by human hearing. The abnormal sound inspection here is an inspection performed when it is confirmed that the cause of the abnormal sound is not the transmission 12 but is suspected to be a certain component in the path component.

[0042] In the abnormal sound inspection, first, the acceleration sensor 37 is installed in the transmission 12 of the vehicle 10 to be inspected, and the microphone 35 is installed in the vehicle interior 20. Then, the original sound characteristic value and the evaluation sound characteristic value are measured in the above-mentioned manner using the measuring device 34. After the measurement, the execution device 31 executes a determination process for determining whether each path component of the path component is the cause of the abnormal sound.

[0043] Figure 3 The following is a processing sequence of the determination routine executed by the execution device 31 for the above determination. This routine is executed individually for each path component provided with the neural network 33. In the following description, the path component to be the object of the determination routine is described as the inspection target component.

[0044] When this routine is started, the execution device 31 first obtains the measured values ​​of the original sound characteristic value and the evaluation sound characteristic value from the measuring device 34 in step S100. Then, the execution device 31 calculates the output of the neural network 33 using these measured values ​​as input in step S110. The output of the neural network 33 calculated here becomes the vibration transfer characteristic value of the inspection target component of the vehicle 10 as the inspection target, that is, the estimated value of the transmission coefficient and phase of the vibration of the component.

[0045] In addition, the storage device 32 stores in advance the standard values ​​of the transmission coefficient and phase of the vibration of the inspection object. Then, the execution device 31 compares the standard value with the calculated value of the transmission coefficient and phase of the vibration in step S110 in step S120 to determine whether the inspection object is the cause of the abnormal sound. That is, when the execution device 31 determines that the calculated value of the transmission coefficient and the phase deviates greatly from the standard value (S130: Yes), the execution device 31 outputs the determination result that the inspection object is the cause of the abnormal sound (S140). In contrast, when the execution device 31 determines that the above deviation is not large (S130: No), the execution device 31 outputs the determination result that the inspection object is not the cause of the abnormal sound (S150).

[0046] <Functions and Effects of the First Embodiment>

[0047] In the present embodiment, the characteristic value of the vibration transmission of the path member is estimated using the neural network 33. That is, the neural network 33 receives as input the original sound characteristic value indicating the characteristic of the sound emitted by the transmission 12 as the source of the abnormal sound and the evaluation sound characteristic value indicating the characteristic of the sound transmitted to the vehicle interior 20 as the evaluation position of the abnormal sound, and outputs the vibration transmission characteristic value of the path member. Then, the neural network 33 learns using the measured values ​​of the original sound characteristic value, the evaluation sound characteristic value, and the characteristic value of the vibration transmission of the path member as training data.

[0048] The waveform of the sound transmitted from the transmission 12 to the cabin 20 reflects the vibration transfer characteristics of each of the vehicle components, i.e., the path components, located in the vibration transfer path from the transmission 12 to the cabin 20. If the number of path components existing in the vibration transfer path from the transmission 12 to the cabin 20 is small, the characteristic values ​​of the vibration transfer of the path components can be obtained based on the original sound characteristic values ​​and the evaluation sound characteristic values ​​even by conventional methods such as waveform analysis. However, in the actual vehicle 10, there are multiple path components in the vibration transfer path from the transmission 12 to the cabin 20, and it is difficult to obtain the vibration transfer characteristics of each path component by conventional methods. In contrast, in the above-mentioned neural network 33, even if the relationship between the original sound characteristic values ​​and the evaluation sound characteristic values ​​and the characteristic values ​​of the vibration transfer of the path components is unclear, their relationship can be learned.

[0049] According to the abnormal sound inspection device for a vehicle according to the present embodiment described above, the following effects can be achieved.

[0050] (1) By using the neural network 33 configured as described above, it is possible to accurately estimate the characteristic values ​​of vibration transmission of the path member based on the measurement results of the original sound characteristic values ​​and the evaluation sound characteristic values ​​by the vibration test.

[0051] (2) By using the estimation result of the vibration transfer characteristic value of each path member, the path member that causes the abnormal sound can be identified.

[0052] (3) Since the original sound characteristic value and the evaluation sound characteristic value are measured by the vibration test of the transmission 12, it is possible to easily perform an inspection for identifying the cause of the abnormal sound.

[0053] (Second embodiment)

[0054] Next, refer to Figure 4, a second embodiment of the abnormal sound inspection device for a vehicle will be described in detail. In addition, in this embodiment, the same reference numerals are attached to the same structures as those in the above-mentioned embodiment, and the detailed description thereof will be omitted.

[0055] In the first embodiment, the original sound characteristic value and the evaluation sound characteristic value are measured for abnormal sound inspection and learning of the neural network 33 while the vehicle 10 is not moving. However, the sound generated by the transmission 12 and the vibration transfer characteristics of each path component change according to the running state of the vehicle 10. Therefore, in the present embodiment, the original sound characteristic value and the evaluation sound characteristic value are measured while the vehicle 10 is running.

[0056] Figure 4 FIG. 3 shows the structure of the neural network 33 used in this embodiment. Figure 4 In the example, the measured values ​​of the original sound characteristic value are input to the nodes X1 to Xa in the input layer, the measured values ​​of the evaluation sound characteristic value are input to the nodes Xa+1 to Xc, and the measured values ​​of the driving state quantity of the vehicle 10 are input to the nodes Xc+1 to Xn. The driving state quantity of the vehicle 10 includes the vehicle speed V, the output speed NE and output torque TE of the engine 11, and the output speed NO and output torque TO of the transmission 12. In addition, Figure 4 In the case of Figure 3 The same structure is true for the case.

[0057] When measuring the original sound characteristic value and the evaluation sound characteristic value in the present embodiment, the microphone 35 is set in the vehicle cabin 20, and the acceleration sensor 37 is installed in the transmission 12. Then, the vehicle 10 is driven on the stand. The measuring device 34 obtains the waveform data of the sound in the transmission 12 and the vehicle cabin 20 as the measurement value of the original sound characteristic value and the evaluation sound characteristic value based on the output of the microphone 35 and the acceleration sensor 37 at this time. In addition, the measuring device 34 obtains the driving state quantity of the vehicle 10 from the electronic control unit 19 of the vehicle 10. In this way, in the present embodiment, the original sound characteristic value and the evaluation sound characteristic value are measured in a state where the vehicle 10 is actually driven, instead of through an excitation test based on a pulse hammer 36.

[0058] In the present embodiment, a data set in which the measured values ​​of the original sound characteristic value, the evaluation sound characteristic value, the running state quantity of the vehicle 10, and the vibration transfer characteristic value of the path component are collected into one is used as training data to perform learning of the neural network 33. The training data includes a plurality of data sets including the original sound characteristic value, the evaluation sound characteristic value, the running state quantity of the vehicle 10, and the vibration transfer characteristic value of the path component measured in various running states in a plurality of vehicles 10 having different usage periods or the like.

[0059] The execution device 31 in the abnormal sound inspection device 30 of the present embodiment calculates the output of the neural network 33 which takes the original sound characteristic value, the evaluation sound characteristic value and the measured value of the driving state quantity of the vehicle 10 as the estimated value of the vibration transfer characteristic value of the path component, and performs the abnormal sound inspection. That is, the execution device 31 of the present embodiment executes the judgment routine as follows. That is, the execution device 31 of the present embodiment executes the judgment routine as follows. Figure 3 In step S100, the execution device 31 obtains the measured values ​​of the driving state quantity of the vehicle 10 during these measurements together with the measured values ​​of the original sound characteristic value and the evaluation sound characteristic value. Next, in step S110, the execution device 31 calculates the value of the driving state quantity of the vehicle 10 using the measured values ​​obtained in step S100 as input. Figure 4 The output of the neural network 33 is used as the vibration transfer characteristic value of the path component. Then, the execution device 31 of this embodiment performs the abnormal sound inspection by performing the same processing as in the case of the first embodiment after step S120.

[0060] As described above, in the first embodiment, the excitation test of the transmission 12 using the pulse hammer 36 is performed to measure the original sound characteristic value and the evaluation sound characteristic value. Then, the learning of the neural network 33 is performed based on the measured values. However, the generation condition of abnormal sound varies depending on the driving state of the vehicle 10. Therefore, based only on the measured values ​​during the excitation test, the relationship between the original sound characteristic value and the evaluation sound characteristic value and the vibration transfer characteristic value of the path component may not be properly learned.

[0061] In contrast, in the present embodiment, the neural network 33 is learned based on the measured values ​​of the original sound characteristic values ​​and the evaluation sound characteristic values ​​during the running of the vehicle 10. Therefore, the neural network 33 can be learned in a form that reflects the generation conditions of abnormal sounds during the running of the actual vehicle 10. In addition, the running state quantity of the vehicle 10 is included in the input of the neural network 33 and the training data for learning the neural network 33. Therefore, the neural network 33 can be learned in a form that reflects the changes in the generation conditions of abnormal sounds caused by the running state of the vehicle 10.

[0062] (Third embodiment)

[0063] Next, refer to Figure 5 , a third embodiment of the abnormal sound inspection device for a vehicle will be described in detail. In addition, in this embodiment, the same reference numerals are attached to the same structures as those in the above-mentioned embodiments, and their detailed descriptions are omitted.

[0064] Sometimes, a new vehicle model is developed based on an existing vehicle model, using common components except for some components. When the component changed from the basic vehicle model is a component equivalent to a path component, an uncomfortable abnormal sound may be generated in conjunction with the change. Therefore, sometimes, when developing such a vehicle model, it is evaluated whether an uncomfortable abnormal sound is generated in conjunction with the component change. The abnormal sound inspection device of this embodiment is configured as a device for performing an abnormal sound inspection for such an evaluation.

[0065] Figure 5 FIG. 3 shows the structure of the neural network 33 used in this embodiment. Figure 5 In the example, the measured values ​​of the original sound characteristic values ​​are input to X1 to Xa in each node of the input layer, the path component characteristic values ​​indicating the characteristics of the vibration transmission of the path component are input to Xa+1 to Xd, and the measured values ​​of the driving state quantity of the vehicle 10 are input to Xd+1 to Xn. On the other hand, Y1 to Yp, which are the values ​​of each node of the output layer, are the evaluation sound characteristic values. In addition, the path component characteristic values ​​of a plurality of path components are included in Xa+1 to Xd, which are the input values ​​of the input layer.

[0066] In the present embodiment, a data set including the original sound characteristic value, the evaluation sound characteristic value, the measured values ​​of the running state quantity of the vehicle 10, and the path component characteristic value measured in the same manner as in the case of the second embodiment is used as training data for the neural network 33. However, in the case of the present embodiment, the measured values ​​of the original sound characteristic value, the running state quantity of the vehicle 10, and the path component characteristic value in the data set are input as input values ​​when the neural network 33 is learned. Then, the values ​​of each weight Wij, Vjk are corrected in such a way that the error between the values ​​Y1 to Yp output by the neural network 33 and the measured values ​​of the evaluation sound characteristic value in the data set becomes small in response to these inputs, thereby advancing the learning of the neural network 33.

[0067] In the abnormal sound inspection for evaluating the presence or absence of abnormal sound associated with component changes, the vibration transmission characteristic value of the changed path component is obtained through the excitation test of the prototype, simulation, etc. In addition, the characteristic value of the sound emitted by the transmission 12 in each driving state of the vehicle 10 is measured through the driving test, simulation, etc. in the prototype of the vehicle 10. In addition, the measured value of the original sound characteristic value can also be used as the measured value of the basic model. Then, the execution device 31 calculates the output of the neural network 33 that takes the path component characteristic value of each path component including the changed component, the measured value of the original sound characteristic value and the driving state quantity of the vehicle 10 when measured as an estimated value of the evaluation sound characteristic value. Thus, the evaluation sound characteristic value of the model under development, that is, the waveform data of the sound heard by the passengers in the vehicle cabin 20, is estimated. Then, based on the estimation result, an evaluation is performed to determine whether an uncomfortable abnormal sound is generated. In such an abnormal sound inspection device 30 of the present embodiment, even if an actual vehicle is not prototyped or a precise physical model of vibration transmission of an actual vehicle is not made, it is possible to confirm the presence or absence of abnormal sound associated with component changes.

[0068] The above-mentioned embodiments can be implemented by changing the following aspects. The present embodiment and the following modified examples can be implemented in combination with each other within the scope of no technical contradiction.

[0069] In the above embodiment, the original sound characteristic value is measured using the acceleration sensor 37 attached to the transmission 12 , but the original sound characteristic value may be measured using a microphone provided near the transmission 12 .

[0070] In the above embodiment, the interior of the vehicle 20 is set as the evaluation position of the abnormal sound. Sometimes the rattling sound of the gears in the transmission 12 leaks outside the vehicle, and the sound is recognized as an uncomfortable abnormal sound by people outside the vehicle. When checking such abnormal sound outside the vehicle, an evaluation position can be set outside the vehicle 10, and the evaluation sound characteristic value can be measured at the evaluation position.

[0071] The abnormal sound inspection device 30 may be configured as a device for inspecting abnormal sounds caused by sounds emitted by vehicle components other than the transmission 12. For example, the engine 11, propeller shaft 13, and differential gear mechanism 14 may also be sound source components of vehicle components that emit sounds that cause abnormal sounds.

[0072] In the above embodiment, the waveform data of the sound is used as the original sound characteristic value and the evaluation sound characteristic value, but the data of the spectrum of the sound may also be used as the original sound characteristic value and the evaluation sound characteristic value. In addition, the spectrum is a spectrum that represents the magnitude of each frequency component of the sound as a function of the frequency.

[0073] In the above embodiment, the vibration transfer coefficient and phase of the vibration transfer function of the path component are used as the path component characteristic value. Instead of or in addition to the vibration transfer coefficient and phase, the value of the coherence function may be used as the path component characteristic value.

[0074] In the above embodiment, the vehicle 10 using the engine 11 as a driving source is the object of the abnormal sound inspection. However, vehicles using other driving methods such as electric vehicles and hybrid vehicles may also be the objects of the abnormal sound inspection.

[0075] In the above embodiment, the intermediate layer of the neural network 33 is set to one layer, but a plurality of intermediate layers may be provided.

Claims

1. An abnormal sound inspection device for a vehicle, for determining the cause of an abnormal sound, the abnormal sound inspection device comprising: storage device and execution device, When a vehicle component that is a source of an abnormal sound is set as a sound source component, a vehicle component located in a vibration transmission path from the sound source component to an evaluation position of the abnormal sound is set as a path component, a value indicating a vibration transmission characteristic of the path component is set as a path component characteristic value, a value indicating a characteristic of a sound emitted by the sound source component is set as an original sound characteristic value, a value indicating a characteristic of a sound reaching the evaluation position of the abnormal sound is set as an evaluation sound characteristic value, the evaluation sound characteristic value is set as a first characteristic value, and the path component characteristic value is set as a second characteristic value, The storage device stores a neural network that takes the original sound characteristic value and the first characteristic value as input and outputs the second characteristic value, and the neural network is learned by using the measured values ​​of the original sound characteristic value, the first characteristic value, and the second characteristic value as training data, The execution device performs estimation processing and determination processing on each of the path components on the transmission path of the vibration as an inspection target component. The storage device also stores in advance standard values ​​of the transmission coefficient and phase of the vibration of the inspection object component. In the estimation process, the output of the neural network, which takes the original sound characteristic value and the measured value of the first characteristic value as input, is calculated as an estimated value of the second characteristic value, and the estimated value of the second characteristic value is an estimated value of the transmission coefficient and phase of the vibration of the inspection object component. In the determination process, the standard values ​​of the vibration transmission coefficient and phase of the inspection object component are compared with the estimated values ​​of the vibration transmission coefficient and phase of the inspection object component obtained in the estimation process to determine whether the inspection object component is the cause of the abnormal sound.

2. The abnormal sound inspection device for a vehicle according to claim 1, in, The input of the neural network includes a running state quantity which is a quantity indicating the running state of the vehicle, and the training data includes a measured value of the running state quantity when the evaluation sound characteristic value is measured.

3. The abnormal sound inspection device for a vehicle according to claim 1 or 2, in, Either the waveform data or the frequency spectrum of the sound is used as the original sound characteristic value and the evaluation sound characteristic value.

4. An abnormal sound inspection device for a vehicle, for evaluating the presence or absence of an abnormal sound generated in association with a change in a vehicle component, the abnormal sound inspection device comprising: storage device and execution device, When a vehicle component that is a source of an abnormal sound is set as a sound source component, a vehicle component located in a vibration transmission path from the sound source component to an evaluation position of the abnormal sound is set as a path component, a value indicating a vibration transmission characteristic of the path component is set as a path component characteristic value, a value indicating a characteristic of a sound emitted by the sound source component is set as an original sound characteristic value, a value indicating a characteristic of a sound reaching the evaluation position of the abnormal sound is set as an evaluation sound characteristic value, the path component characteristic value is set as a first characteristic value, and the evaluation sound characteristic value is set as a second characteristic value, The storage device stores a neural network that takes the original sound characteristic value and the first characteristic value as input and outputs the second characteristic value, and the neural network is learned by using the measured values ​​of the original sound characteristic value, the first characteristic value, and the second characteristic value as training data, When a path component on the transmission path of the vibration is changed in the vehicle, a path component characteristic value of the changed path component is measured, and the execution device performs an inference process. In the inference process, the output of the neural network, which takes the original sound characteristic value and the measured value of the path component characteristic value of the changed path component as input, is calculated as an inferred value of the second characteristic value, and the presence or absence of abnormal sound associated with the change of the path component is confirmed based on the inferred value of the second characteristic value.

5. The abnormal sound inspection device for a vehicle according to claim 4, in, The input of the neural network includes a running state quantity which is a quantity indicating the running state of the vehicle, and the training data includes a measured value of the running state quantity when the evaluation sound characteristic value is measured.

6. The abnormal sound inspection device for a vehicle according to claim 4 or 5, in, The transmission coefficient and phase of the vibration of the path member are used as the path member characteristic values.

7. The abnormal sound inspection device for a vehicle according to claim 4 or 5, in, Either the waveform data or the frequency spectrum of the sound is used as the original sound characteristic value and the evaluation sound characteristic value.

Images