JP7634838B2 - Apparatus, method and program for analyzing vibration damping device characteristics using machine learning - Google Patents

Description

The present invention relates to a device, method, and program for analyzing the characteristics of a vibration damping device using machine learning.

Vibration damping devices that are installed in vehicles and the like to dampen vibrations are known. For example, Patent Document 1 describes how machine learning is used to identify data related to up-and-down vibrations when the vehicle is running normally.

JP 2019-69651 A

There is room for improvement in using machine learning to properly analyze the characteristics of vibration damping devices.

The present invention has been made in consideration of the above, and aims to provide a vibration damping device characteristic analysis device, characteristic analysis method, and program that use machine learning to appropriately analyze the characteristics of a vibration damping device.

In order to solve the above-mentioned problems and achieve the object, the vibration damping device characteristic analysis device using machine learning according to the present disclosure includes an AI model acquisition unit that acquires an AI model that has been machine-learned to learn the correspondence between a spectrogram of acceleration on the unsprung side of a first vibration damping device provided between the body and wheels of a vehicle and a spectrogram of acceleration on the sprung side of the first vibration damping device; a measurement acquisition unit that acquires an input measurement value that is a measurement value of acceleration on one of the unsprung side and the sprung side of a second vibration damping device, and an output measurement value that is a measurement value of acceleration on the other of the unsprung side and the sprung side of the second vibration damping device; a predicted value acquisition unit that inputs the input measurement value as a spectrogram into the AI model and acquires a spectrogram of an output predicted value that is a predicted value of acceleration on the other of the unsprung side and the sprung side of the second vibration damping device; and a characteristic information generation unit that generates characteristic information indicating the difference in the spectrogram of the acceleration of the second vibration damping device relative to the first vibration damping device based on the difference between the spectrogram of the output predicted value and the spectrogram of the output measurement value.

In order to solve the above-mentioned problems and achieve the objective, the characteristic analysis method according to the present disclosure includes the steps of: acquiring an AI model that uses machine learning to learn the correspondence between a spectrogram of acceleration on the unsprung side of a first vibration damping device provided between the body and the wheels of a vehicle and a spectrogram of acceleration on the sprung side of the first vibration damping device; acquiring an input measurement value that is a measurement value of acceleration on one of the unsprung side and the sprung side of a second vibration damping device, and an output measurement value that is a measurement value of acceleration on the other of the unsprung side and the sprung side of the second vibration damping device; inputting the input measurement value into the AI model as a spectrogram to acquire a spectrogram of an output predicted value that is a predicted value of acceleration on the other of the unsprung side and the sprung side of the second vibration damping device; and generating characteristic information that indicates the difference in the spectrogram of the acceleration of the second vibration damping device relative to the first vibration damping device based on the difference between the spectrogram of the output predicted value and the spectrogram of the output measurement value.

In order to solve the above-mentioned problems and achieve the object, the program according to the present disclosure causes a computer to execute the steps of: acquiring an AI model that uses machine learning to learn the correspondence between a spectrogram of acceleration on the unsprung side of a first vibration damping device provided between the body and the wheels of a vehicle and a spectrogram of acceleration on the sprung side of the first vibration damping device; acquiring an input measurement value that is a measurement value of acceleration on one of the unsprung side and the sprung side of a second vibration damping device, and an output measurement value that is a measurement value of acceleration on the other of the unsprung side and the sprung side of the second vibration damping device; inputting the input measurement value into the AI model as a spectrogram to acquire a spectrogram of an output predicted value that is a predicted value of acceleration on the other of the unsprung side and the sprung side of the second vibration damping device; and generating characteristic information indicating the difference in the spectrogram of the acceleration of the second vibration damping device relative to the first vibration damping device based on the difference between the spectrogram of the output predicted value and the spectrogram of the output measurement value.

The present invention allows the characteristics of a vibration damping device to be properly analyzed.

FIG. 1 is a schematic diagram of a vibration damping device according to the present embodiment. FIG. 2 is a schematic cross-sectional view of the vibration damping device according to the present embodiment. FIG. 3 is a schematic block diagram of the characteristic analysis device according to this embodiment. FIG. 4 is a schematic diagram for explaining image data for an AI model. FIG. 5 is a conceptual schematic diagram of an AI model. FIG. 6 is a diagram illustrating an example of the characteristic information. FIG. 7 is a flowchart illustrating a process flow of the characteristic analysis device.

Below, a preferred embodiment of the present invention will be described in detail with reference to the drawings. Note that the present invention is not limited to the embodiment described below.

(Vibration damping device)
The characteristic analysis device 100 according to the present embodiment analyzes the characteristics of the vibration damping device using machine learning. An example of the configuration of the vibration damping device will be described below.

Figure 1 is a schematic diagram of a vibration damping device according to this embodiment, and Figure 2 is a schematic cross-sectional view of the vibration damping device according to this embodiment. As shown in Figure 1, the vibration damping device 1 according to this embodiment is a device that is provided between the body BD and the wheels TR (axles connected to the wheels TR) of a vehicle V, and generates a damping force to suppress vibration of the body BD. However, the vibration damping device 1 is not limited to being used for vehicles, and may be used in any mechanism.

As shown in FIG. 2, the vibration damping device 1 according to this embodiment includes a cylinder 10, a piston 20 (valve piston) slidably inserted into the cylinder 10, a piston rod 30 inserted into the cylinder 10 so as to be able to move forward and backward, and a base valve 40 provided within the cylinder 10. The piston rod 30 is connected to the piston 20 and extends to the outside of the cylinder 10.

The cylinder 10 is filled with a working fluid O. In this embodiment, the working fluid O is oil, but it may be any fluid such as water. The cylinder 10 has an inner tube 12, an outer tube 14, a bearing 16, and an oil seal 18. The inner tube 12 is a cylindrical member. The outer tube 14 is a cylindrical member that is provided to surround the outer circumferential surface (side surface) of the inner tube 12. The bearing 16 is a guide bush that slidably supports the piston rod 30. The oil seal 18 slidably supports the piston rod 30 while sealing the working fluid O and the like.

The above is the vibration damping device 1 according to this embodiment, but the configuration of the vibration damping device 1 is not limited to the above example. For example, in the above description, a so-called twin-tube shock absorber was used, but the present invention is not limited to this and may be a so-called monotube shock absorber.

(Characteristics Analysis Device)
3 is a schematic block diagram of the characteristic analysis device according to the present embodiment. The characteristic analysis device 100 can be said to be a computer, and includes an input unit 70, an output unit 72, a communication unit 74, a storage unit 76, and a control unit 78. The input unit 70 is a user interface that accepts user operations (inputs), and may be, for example, a mouse or a keyboard. The output unit 72 is a device that outputs information, and may be, for example, a display. The communication unit 74 is a communication module that communicates with an external device, and may be, for example, an antenna. The characteristic analysis device 100 communicates with an external device by wireless communication, but may also be wired communication, and any communication method may be used.

The storage unit 76 is a memory that stores various information such as the calculation contents and programs of the control unit 78, and includes at least one of a main storage device such as a RAM (Random Access Memory), a ROM (Read Only Memory), and an external storage device such as a HDD (Hard Disk Drive). The program for the control unit 78 stored in the storage unit 76 may be stored in a recording medium readable by the characteristic analysis device 100. The storage unit 76 also stores an AI model N, which is a learning model in AI (Artificial Intelligence).

The control unit 78 is a calculation device and includes a calculation circuit such as a CPU (Central Processing Unit). The control unit 78 includes a learning unit 80, an AI model acquisition unit 82, a measurement value acquisition unit 84, a prediction value acquisition unit 86, and a characteristic information generation unit 88. The control unit 78 realizes the learning unit 80, the AI model acquisition unit 82, the measurement value acquisition unit 84, the prediction value acquisition unit 86, and the characteristic information generation unit 88 by reading and executing a program (software) from the storage unit 76, and executes the processing. The control unit 78 may execute these processes using one CPU, or may be provided with multiple CPUs and execute the processing using the multiple CPUs. Also, at least a part of the processing of the learning unit 80, the AI model acquisition unit 82, the measurement value acquisition unit 84, the prediction value acquisition unit 86, and the characteristic information generation unit 88 may be realized by a hardware circuit.

(Learning Department)
The learning unit 80 performs machine learning on the AI model N before learning to generate a trained AI model N. The learning unit 80 trains the AI model N to learn the correspondence between the vibration degree on the vibration source side of the first vibration damping device and the vibration degree on the opposite side of the vibration source of the first vibration damping device. Here, the first vibration damping device is the vibration damping device 1 that is the reference for machine learning on which the AI model N is trained. In addition, the vibration source side of the first vibration damping device refers to a position closer to the vibration source of the first vibration damping device (the wheel TR in this embodiment), and can be said to be a position (unsprung position) closer to the vibration source side than the mechanism that damps the vibration of the first vibration damping device. In this embodiment, the vibration source side of the first vibration damping device may refer to the cylinder 10 of the first vibration damping device. In addition, the vibration degree may be any parameter as long as it is a parameter indicating the degree of vibration, but in this embodiment, it is acceleration. In other words, the vibration degree on the vibration source side of the first vibration damping device can be said to refer to the unsprung acceleration of the first vibration damping device. On the other hand, the side opposite the vibration source of the first vibration damping device refers to the side farther from the vibration source of the first vibration damping device (the wheel TR in this embodiment), and can be said to be the position opposite the vibration source (the position on the spring) with respect to the mechanism that damps the vibration of the first vibration damping device. In this embodiment, the side opposite the vibration source of the first vibration damping device may refer to the piston rod 30 of the first vibration damping device. In other words, the degree of vibration on the side opposite the vibration source of the first vibration damping device can be said to refer to the acceleration on the spring of the first vibration damping device. Hereinafter, the side opposite the vibration source of the first vibration damping device will be appropriately referred to as the damping side of the first vibration damping device.

The learning unit 80 acquires the measured value of the vibration degree (here, acceleration) on the vibration source side of the first vibration damping device as teacher data for input to the untrained AI model, and acquires the measured value of the vibration degree (here, acceleration) on the damping side of the first vibration damping device as teacher data for output that becomes a solution to teach the output data from the untrained AI model. The teacher data will be explained in more detail below.

In this embodiment, a first sensor S1 that detects the degree of vibration (here, acceleration) on the vibration source side of the first vibration damping device and a second sensor S2 that detects the degree of vibration (here, acceleration) on the damping side of the first vibration damping device are attached to a vehicle V equipped with a first vibration damping device, and the vehicle V is actually driven. Then, the first sensor S1 detects the acceleration at a predetermined time on the vibration source side of the first vibration damping device while the vehicle V is driving, and the second sensor S2 detects the acceleration at a predetermined time on the damping side of the first vibration damping device while the vehicle V is driving. The learning unit 80 acquires the detection results of the first sensor S1 and the second sensor S2 from a device (data logger) that records the detection results of the first sensor S1 and the second sensor S2 via the communication unit 74, and uses them as teacher data. The first sensor S1 may be attached at any position as long as it can detect the degree of vibration on the vibration source side of the first vibration damping device, but for example, it is attached at a position closer to the wheel than the mechanism that damps the vibration of the first vibration damping device (for example, cylinder 10). The second sensor S2 may be attached at any position as long as it can detect the degree of vibration on the damping side of the first vibration damping device, but for example, it is attached at a position on the opposite side of the wheel than the mechanism that damps the vibration of the first vibration damping device (for example, piston rod 30). The driving conditions of the vehicle V when the first sensor S1 and the second sensor S2 are made to perform detection may be set arbitrarily, for example, to drive a test course at a predetermined speed.

Figure 4 is a schematic diagram for explaining image data for an AI model. In this embodiment, the learning unit 80 converts the waveform data of the acceleration at each predetermined time detected by the first sensor S1 and the second sensor S2 into image data (spectrogram) indicating the acceleration at each time and frequency, and uses it as teacher data. That is, the acceleration waveform can be converted into data indicating the acceleration at each frequency by FFT (Fast Fourier Transform). Therefore, as shown in the example of Figure 4, the waveform data of the acceleration at each predetermined time detected by the first sensor S1 and the second sensor S2 can be converted into data indicating the acceleration at each time and frequency by dividing the acceleration data at each time by frequency. That is, the learning unit 80 divides the acceleration data at each predetermined time detected by the first sensor S1 by FFT processing to generate image data indicating the acceleration at each time and frequency on the vibration source side of the first vibration damping device. That is, the learning unit 80 generates image data showing the acceleration for each time and frequency on the vibration source side of the first vibration damping device by setting the acceleration for each time and frequency on the vibration source side of the first vibration damping device to a gradation value for each coordinate (pixel) in a two-dimensional coordinate system with time as the X-axis and frequency as the Y-axis. Similarly, the learning unit 80 divides the acceleration data for each time by frequency using FFT processing for the waveform data of the acceleration for each predetermined time detected by the second sensor S2, and generates image data showing the acceleration for each time and frequency on the damping side of the first vibration damping device. That is, the learning unit 80 generates image data showing the acceleration for each time and frequency on the damping side of the first vibration damping device by setting the acceleration for each time and frequency on the damping side of the first vibration damping device to a gradation value for each coordinate (pixel) in a two-dimensional coordinate system with time as the X-axis and frequency as the Y-axis. Note that FIG. 4 shows an example of image data, which is a two-dimensional image (spectrogram) in which acceleration is expressed by a gradation value (for example, brightness).

The learning unit 80 uses image data showing the acceleration for each time and frequency on the vibration source side of the first vibration damping device as teacher data for input to the unlearned AI model, and uses image data showing the acceleration for each time and frequency on the damping side of the first vibration damping device as teacher data for output to the unlearned AI model N. The learning unit 80 inputs the teacher data for input and the teacher data for output to the unlearned AI model, and causes the AI model N to machine-learn the correspondence between the degree of vibration on the vibration source side of the first vibration damping device and the degree of vibration on the damping side of the first vibration damping device. More specifically, the learning unit 80 causes the AI model N to machine-learn the correspondence between the acceleration for each time and frequency on the vibration source side of the first vibration damping device and the acceleration for each time and frequency on the damping side of the first vibration damping device.

FIG. 5 is a conceptual schematic diagram of the AI model N. The learning unit 80 generates the AI model N that has been machine-learned by setting weight coefficients and bias values in the AI model using teacher data. In other words, the AI model N can be said to be a model in which weight coefficients and bias values have been learned by the learning unit 80. The AI model N is a learning model learned by deep learning, and is composed of a model (neural network) that reconstructs a sprung spectrogram from an unsprung spectrogram, learned by deep learning, and variables. Deep learning is one of the methods of machine learning, and in the narrow sense, is composed of, for example, a neural network with four or more layers. In the example of this embodiment, the AI model N is a CNN (Conventional Neural Network) model, and as shown in FIG. 5, for example, includes an encoder layer MN1 and a decoder layer MN2 including multiple convolution layers and multiple pooling layers. When input data IN is input, the AI model N performs calculations in the encoder layer MN1 and the decoder layer MN2 and outputs output data ON.

The learning unit 80 trains the untrained AI model in the above manner to generate a trained AI model N. The trained AI model N is a model that has been machine-learned to obtain the correspondence between the degree of vibration on the vibration source side of the first vibration damping device and the degree of vibration on the damping side of the first vibration damping device. More specifically, the trained AI model N is a model that has been machine-learned to obtain the correspondence between the acceleration per time and frequency on the vibration source side of the first vibration damping device and the acceleration per time and frequency on the damping side of the first vibration damping device.

In this embodiment, the AI model N is trained using image data suitable for CNN, and the image data is also input when the AI model N is used, thereby improving the prediction accuracy of the AI model N. However, in this embodiment, the AI model N is not limited to using image data indicating acceleration for each time and frequency as training data for the AI model N, and any data indicating the degree of vibration on the vibration source side and the damping side of the first vibration damping device may be used as training data for the AI model N. Furthermore, the AI model N is not limited to a CNN model, and may be an AI model of any type.

(Regarding the second vibration damping device)
Since the AI model N has learned the correspondence relationship between the vibration degree on the vibration source side of the first vibration damping device and the vibration degree on the damping side, when the vibration degree on the vibration source side of the first vibration damping device is input, it can predict the unknown vibration degree on the damping side of the first vibration damping device. In addition, the characteristic analysis device 100 of this embodiment uses the AI model N that has machine-learned about the first vibration damping device to analyze the characteristics of a second vibration damping device that is different from the first vibration damping device. The second vibration damping device is the vibration damping device 1 that is the subject of the characteristic analysis, and has different specifications from the first vibration damping device. In this embodiment, the second vibration damping device has different specifications of the working fluid O from the first vibration damping device. However, the difference in specifications between the second vibration damping device and the first vibration damping device is not limited to the working fluid O and may be any, but it is preferable that the second vibration damping device and the first vibration damping device have different specifications that affect the friction force in the cylinder 10. For example, the second vibration damping device and the first vibration damping device preferably differ in at least one of the specifications of the working fluid O, the specifications of the oil seal 18, the specifications of the bearing 16, the specifications of the piston 20, and the specifications of the valve (base valve 40). Also, it is preferable that the difference in the specifications between the second vibration damping device and the first vibration damping device is one type. Note that the specification here may refer to the physical properties of the working fluid O, or may refer to at least one of the shape, dimensions, and material of the parts such as the oil seal 18, the bearing 16, the piston 20, and the valve.

The analysis of the characteristics of the second vibration damping device using the characteristic analysis device 100 is described in detail below.

The AI model acquisition unit 82 acquires the trained AI model N that has been trained by the learning unit 80. For example, the AI model acquisition unit 82 reads out the trained AI model N from the memory unit 76. Note that, for example, if the AI model N has been trained by another device, the AI model acquisition unit 82 may acquire the trained AI model N from that device via the communication unit 74.

(Measurement value acquisition unit)
The measurement value acquisition unit 84 acquires a measurement value of the vibration degree (here, acceleration) on the vibration source side of the second vibration damping device and a measurement value of the vibration degree (here, acceleration) on the damping side (opposite side to the vibration source) of the second vibration damping device. The vibration source side of the second vibration damping device refers to a position closer to the vibration source (wheel in this embodiment) of the second vibration damping device, and can be said to be a position closer to the vibration source side than the mechanism that damps the vibration of the second vibration damping device (unsprung position). In this embodiment, the vibration source side of the second vibration damping device may refer to the cylinder 10 of the second vibration damping device. Moreover, the damping side (opposite side to the vibration source) of the second vibration damping device refers to a side farther from the vibration source (wheel in this embodiment) of the second vibration damping device, and can be said to be a position on the opposite side to the vibration source with respect to the mechanism that damps the vibration of the second vibration damping device (spring position). In this embodiment, the damping side of the second vibration damping device may refer to the piston rod 30 of the first vibration damping device.

In this embodiment, a first sensor S1 for detecting the degree of vibration (here, acceleration) on the vibration source side of the second vibration damping device and a second sensor S2 for detecting the degree of vibration (here, acceleration) on the damping side of the second vibration damping device are attached to a vehicle V equipped with a second vibration damping device, and the vehicle V is actually driven. Then, the first sensor S1 detects the acceleration at a predetermined time on the vibration source side of the second vibration damping device while the vehicle V is moving, and the second sensor S2 detects the acceleration at a predetermined time on the damping side of the second vibration damping device while the vehicle V is moving. The measurement value acquisition unit 84 acquires the detection results of the first sensor S1 and the second sensor S2 via the communication unit 74 from a device (data logger) in which the detection results of the first sensor S1 and the second sensor S2 are recorded. Note that the first sensor S1 may be attached at any position as long as it can detect the degree of vibration on the vibration source side of the second vibration damping device, but for example, it is attached at a position closer to the wheel side than the mechanism that damps the vibration of the second vibration damping device (for example, the cylinder 10). The second sensor S2 may be attached at any position as long as it can detect the degree of vibration on the damping side of the second vibration damping device. For example, the second sensor S2 is attached at a position on the opposite side of the wheel from the mechanism that damps the vibration of the second vibration damping device (for example, the piston rod 30). The running conditions of the vehicle V when the first sensor S1 and the second sensor S2 perform detection may be set arbitrarily, such as running at a predetermined speed on a test course. However, it is preferable that the mounting positions of the first sensor S1 and the second sensor S2, the specifications of the vehicle V, and the running conditions of the vehicle V when detecting the degree of vibration of the first vibration damping device are the same as when the degree of vibration of the first vibration damping device is detected.

Hereinafter, the measurement value of the vibration degree on the vibration source side of the second vibration damping device acquired by the measurement value acquisition unit 84 will be referred to as the input measurement value, and the measurement value of the vibration degree on the damping side of the second vibration damping device acquired by the measurement value acquisition unit 84 will be referred to as the output measurement value, as appropriate.

(Prediction value acquisition unit)
The predicted value acquisition unit 86 inputs the input measurement values acquired by the measurement value acquisition unit 84 to the trained AI model N. In the AI model N, the input measurement values are input as input data and a calculation is performed. As a result, the AI model N outputs an output predicted value, which is a predicted value of the degree of vibration on the damping side (the side opposite to the vibration source) of the second vibration damping device, as output data. In this way, the predicted value acquisition unit 86 acquires the output predicted value. In this way, the AI model N can be said to be a trained program that outputs an output predicted value when an input measurement value is input.

More specifically, in this embodiment, the predicted value acquisition unit 86 generates input data to be input to the AI model N based on the input measurement values acquired by the measurement value acquisition unit 84. Specifically, the predicted value acquisition unit 86 divides the acceleration waveform data (input measurement values for each time) detected by the first sensor S1 for each predetermined time by FFT processing, and generates input image data (spectrogram of input measurement values) indicating the acceleration measurement values for each time and frequency on the vibration source side of the second vibration damping device. That is, the predicted value acquisition unit 86 generates input image data by setting the acceleration measurement values for each time and frequency on the vibration source side of the second vibration damping device as gradation values for each coordinate (pixel) in a two-dimensional coordinate system with time as the X-axis and frequency as the Y-axis.

The predicted value acquisition unit 86 inputs the input image data to the AI model N as input data. In the AI model N, a calculation is performed, and output image data (spectrogram of output predicted value) indicating the predicted value of acceleration for each time and frequency on the damping side of the second vibration damping device is output as output data. In this way, the predicted value acquisition unit 86 acquires the output image data. Note that the output image data refers to data in which the predicted value (output predicted value) of acceleration for each time and frequency on the damping side of the second vibration damping device is expressed as a gradation value for each coordinate (pixel) in a two-dimensional coordinate system with time as the X-axis and frequency as the Y-axis.

(Characteristics Information Generation Unit)
The characteristic information generating unit 88 generates characteristic information based on the difference between the output predicted value (predicted value of acceleration on the damping side of the second vibration damping device) and the output measured value (measured value of acceleration on the damping side of the second vibration damping device). The characteristic information is information indicating the difference in the degree of vibration of the second vibration damping device with respect to the first vibration damping device. That is, since the AI model N has learned the degree of vibration of the first vibration damping device, it is assumed that the output predicted value of the degree of vibration of the second vibration damping device calculated by the AI model N will be close to the output measured value when the characteristics of the first vibration damping device and the second vibration damping device are similar. Therefore, it can be said that the difference between the output predicted value and the output measured value is an index indicating the difference in the degree of vibration of the second vibration damping device with respect to the first vibration damping device. Therefore, the characteristic information generating unit 88 can generate characteristic information indicating the difference in the degree of vibration of the second vibration damping device with respect to the first vibration damping device from the difference between the output predicted value and the output measured value. In this way, by generating characteristic information from the difference between the output prediction value and the output measurement value, the influence of the difference in specifications between the second vibration damping device and the first vibration damping device on the vibration degree can be appropriately extracted, and the characteristics of the second vibration damping device can be appropriately analyzed. This can assist, for example, in the development of a vibration damping device. Note that, although it is possible to estimate the influence of the difference in specifications on the vibration degree by analyzing the measured values on the damping side of the first vibration damping device and the second vibration damping device, there are cases in which it is not possible to appropriately extract only the influence of the difference in specifications on the vibration degree due to the influence of errors in the conditions at the time of measurement, etc. In contrast, in this embodiment, by using the output prediction value, it is possible to absorb errors in the conditions at the time of measurement, etc., and appropriately extract the influence of the difference in specifications on the vibration degree.

More specifically, in this embodiment, the characteristic information generating unit 88 acquires the output image data calculated by the AI model N. In addition, the characteristic information generating unit 88 divides the acceleration waveform data (output measurement value per time) detected by the second sensor S2 for each predetermined time by FFT processing, and generates measurement image data (spectrogram of output measurement value) showing the acceleration measurement value for each time and frequency on the damping side of the second vibration damping device. The measurement image data refers to data in which the acceleration measurement value (output measurement value) for each time and frequency on the damping side of the second vibration damping device is expressed as a gradation value for each coordinate (pixel) in a two-dimensional coordinate system with time as the X-axis and frequency as the Y-axis. The characteristic information generating unit 88 generates characteristic information based on the difference between the output image data and the measurement image data. More specifically, the characteristic information generating unit 88 calculates the difference value between the output prediction value in the output image data and the output measurement value in the measurement image data for each coordinate. The characteristic information generating unit 88 generates characteristic information based on the difference between the predicted output value and the measured output value for each coordinate. Since the coordinates here are divided by time and frequency, the characteristic information can be said to be data indicating the difference between the predicted output value and the measured output value for each frequency. In this way, by generating characteristic information from the difference between the predicted output value and the measured output value for each frequency, it is possible to extract frequency bands where the difference in vibration degree becomes large due to differences in the specifications of the second vibration damping device and the first vibration damping device. In addition, it is also possible to appropriately extract the difference in vibration degree in high frequency bands that were previously difficult to detect.

FIG. 6 is a diagram showing an example of characteristic information. The characteristic information may be data itself indicating the difference between the output prediction value and the output measurement value, or may be data generated based on the difference between the output prediction value and the output measurement value. For example, the characteristic information generating unit 88 may compare data of the predicted value of the vibration source side of the first vibration damping device obtained by inputting the measured value of the vibration source side of the first vibration damping device into the AI model N with data of the predicted value of the vibration source side of the second vibration damping device obtained by inputting the measured value of the vibration source side of the second vibration damping device into the AI model N as characteristic information. In this case, for example, the characteristic information generating unit 88 inputs image data indicating the measured value of the acceleration for each time and frequency on the vibration source side of the first vibration damping device into the AI model N to obtain image data indicating the predicted value of the acceleration for each time and frequency on the damping side of the first vibration damping device. The characteristic information generating unit 88 calculates the difference between the predicted value of acceleration for each time and frequency on the damping side of the first vibration damping device and the measured value of acceleration for each time and frequency on the damping side of the first vibration damping device, and generates data indicating the difference between the predicted value and the measured value on the damping side of the first vibration damping device for each frequency as the first characteristic information based on the calculated difference. The characteristic information generating unit 88 then generates data indicating the difference between the output predicted value and the output measured value for each frequency as the second characteristic information, and information including the first characteristic information and the second characteristic information as the characteristic information. In the example of FIG. 6, the line segment L1 shows an example of the first characteristic information, and the line segment L2 shows an example of the second characteristic information. By including the first characteristic information and the second characteristic information in this way, for example, it is possible to more appropriately extract a frequency band in which the difference in vibration degree becomes large due to the difference in specifications between the second vibration damping device and the first vibration damping device.

The characteristic information generating unit 88 may output the generated characteristic information to the output unit 72, or may output (transmit) it to another device.

(Processing flow)
The flow of the process of the characteristic analysis device 100 described above will be described. FIG. 7 is a flowchart illustrating the process flow of the characteristic analysis device. As shown in FIG. 7, the characteristic analysis device 100 acquires a trained AI model N by the AI model acquisition unit 82 (step S10), and acquires an input measurement value (a measurement value of the degree of vibration on the vibration source side of the second vibration damping device) and an output measurement value (a measurement value of the degree of vibration on the damping side of the second vibration damping device) by the measurement value acquisition unit 84 (step S12). The characteristic analysis device 100 generates input image data (measurement values of acceleration for each time and frequency on the vibration source side of the second vibration damping device) by the predicted value acquisition unit 86, inputs the input image data to the AI model N (step S14), and acquires output image data (output predicted values for each time and frequency on the damping side of the second vibration damping device) (step S16). The characteristic analysis device 100 generates, by the characteristic information generating unit 88, characteristic information indicating the difference in the degree of vibration of the second vibration damping device relative to the first vibration damping device, based on the difference between the predicted output value and the measured output value (step S18).

In the above explanation, the vibration source side (unsprung side) is the input data of the AI model N, and the damping side (upsprung side) is the output data from the AI model N. However, this is not limited to this. Conversely, the damping side (upsprung side) may be the input data of the AI model N, and the vibration source side (unsprung side) may be the output data from the AI model N. In this case, the AI model N is trained using teacher data in which the vibration degree of the damping side of the first vibration damping device is the input data and the vibration degree of the vibration source side of the first vibration damping device is the output data. Then, the vibration degree of the damping side of the second vibration damping device is input to the trained AI model N, and a predicted value of the vibration degree of the vibration source side of the second vibration damping device is obtained as output data.

(effect)
As described above, the characteristic analysis device 100 according to the present embodiment analyzes the characteristics of the vibration damping device using machine learning, and includes an AI model acquisition unit 82, a measurement value acquisition unit 84, a prediction value acquisition unit 86, and a characteristic information generation unit 88. The AI model acquisition unit 82 acquires an AI model N that has been machine-learned to learn the correspondence between a spectrogram of acceleration on the unsprung side of the first vibration damping device provided between the vehicle body BD and the wheels TR of the vehicle V and a spectrogram of acceleration on the sprung side of the first vibration damping device. The measurement value acquisition unit 84 acquires an input measurement value that is a measurement value of acceleration on one side of the unsprung side and the sprung side of the second vibration damping device, and an output measurement value that is a measurement value of acceleration on the other side of the unsprung side and the sprung side of the second vibration damping device. The prediction value acquisition unit 86 inputs the input measurement value as a spectrogram to the AI model N, and acquires a spectrogram of an output prediction value that is a prediction value of acceleration on the other side of the unsprung side and the sprung side of the second vibration damping device. The characteristic information generating unit 88 generates characteristic information indicating the difference in the spectrogram of the acceleration of the second vibration damping device relative to the first vibration damping device, based on the difference between the spectrogram of the output predicted value and the spectrogram of the output measured value.

The characteristic analysis device 100 obtains the output prediction value of the second vibration damping device using the AI model N that has been machine-learned using the data of the first vibration damping device. The characteristic analysis device 100 then generates characteristic information from the difference between the output prediction value of the second vibration damping device and the output measurement value of the second vibration damping device. Therefore, this characteristic information becomes information indicating the effect of the difference in specifications between the second vibration damping device and the first vibration damping device on the degree of vibration, making it possible to properly analyze the characteristics of the second vibration damping device. This can assist, for example, in the development of a vibration damping device. Furthermore, by using the output prediction value, it is possible to absorb errors in the conditions at the time of measurement and properly extract the effect of the difference in specifications on the degree of vibration. For example, a user may compare a vehicle equipped with the first vibration damping device and a vehicle equipped with the second vibration damping device and select a vibration damping device based on the difference in ride comfort. In such a case, by presenting the characteristic information generated in this embodiment to the user, the difference in ride comfort can be made objectively clear and the selection of a vibration damping device can be supported. In addition, even in the past, it was possible to provide the user with data comparing the measured value of the vibration degree of the first vibration damping device with the measured value of the vibration degree of the second vibration damping device, but vibrations can occur accidentally due to the driver's driving habits (for example, vibrations occurring when turning the steering wheel), the driver's weight, temperature, wind, and other external environmental factors, resulting in data containing noise elements and a risk of low accuracy. In contrast, in this embodiment, by generating characteristic information using AI model N, it is possible to extract only characteristic data from a large amount of data and eliminate uncharacteristic data, thereby reducing the influence of noise elements and providing highly accurate data.

In addition, it is preferable that the AI model N is a model that has been machine-learned to learn the correspondence between image data indicating the degree of vibration on the vibration source side of the first vibration damping device for each time and frequency, and image data indicating the degree of vibration on the damping side of the first vibration damping device for each time and frequency. It is preferable that the predicted value acquisition unit 86 inputs input image data indicating the input measurement value for each time and frequency to the AI model N, and acquires output image data indicating the output measurement prediction value for each time and frequency. By using image data as data for the AI model N in this way, it is possible to improve the prediction accuracy of the output prediction value for each frequency and appropriately analyze the characteristics of the second vibration damping device.

Furthermore, it is preferable that the characteristic information acquisition unit 88 generates characteristic information based on the difference between the predicted output value for each time and frequency in the output image data and the measured output value for each time and frequency, and includes in the characteristic information information information indicating the difference in the degree of vibration of the second vibration damping device for each frequency relative to the first vibration damping device. By including the difference in the degree of vibration for each frequency in the characteristic information in this manner, it is possible to appropriately extract frequency bands where the difference in the degree of vibration becomes large due to differences in the specifications between the second vibration damping device and the first vibration damping device.

The degree of vibration is preferably acceleration. By using acceleration as the degree of vibration, it becomes possible to properly analyze the characteristics of the second vibration damping device.

It is also preferable that the second vibration damping device has specifications that affect frictional force differently from the first vibration damping device. Frictional force affects ride comfort, such as the steering ability and ground feel of the vibration damping device, but it is difficult to analyze how frictional force affects the characteristics of the vibration damping device and, as a result, the ride comfort. In contrast, by generating characteristic information using AI model N, as in this embodiment, it is possible to properly extract the effect that differences in specifications have on differences in the degree of vibration, and properly analyze the effect of frictional force on the characteristics of the vibration damping device.

It is also preferable that the second vibration damping device differs from the first vibration damping device in at least one of the working fluid O, the oil seal 18, the bearing 16, the piston 20, and the valve (base valve 40). According to this embodiment, it becomes possible to appropriately analyze the effect of these differences in specifications on the characteristics of the vibration damping device.

Also, AI model N uses the measured value of the vibration degree on the vibration source side of the first vibration damping device and the measured value of the vibration degree on the damping side (opposite the vibration source) of the first vibration damping device as training data to machine-learn the correspondence. By using the measured values as training data for the AI model, the prediction accuracy of the AI model can be improved.

Although the embodiments and examples of the present invention have been described above, the embodiments are not limited to the contents of these embodiments. The above-mentioned components include those that a person skilled in the art can easily imagine, those that are substantially the same, and those that are within the so-called equivalent range. Furthermore, the above-mentioned components can be combined as appropriate. Furthermore, various omissions, substitutions, or modifications of the components can be made without departing from the spirit of the above-mentioned embodiments.

REFERENCE SIGNS LIST 1 Vibration damping device 80 Learning unit 82 AI model acquisition unit 84 Measurement value acquisition unit 86 Prediction value acquisition unit 88 Characteristics information generation unit 100 Characteristics analysis device N AI model

Claims (3)

an AI model acquisition unit that acquires an AI model that has been machine-learned to learn a correspondence relationship between a spectrogram of acceleration on an unsprung side of a first vibration damping device provided between a body and a wheel of a vehicle and a spectrogram of acceleration on an upsprung side of the first vibration damping device;
a measurement acquisition unit that acquires an input measurement value that is a measurement value of acceleration on one of the unsprung side and the sprung side of the second vibration damping device, and an output measurement value that is a measurement value of acceleration on the other of the unsprung side and the sprung side of the second vibration damping device;
a predicted value acquisition unit that inputs the input measurement value into the AI model as a spectrogram and acquires a spectrogram of an output predicted value, which is a predicted value of acceleration on the other of the unsprung side and the sprung side of the second vibration damping device;
a characteristic information generating unit that generates characteristic information indicating a difference between a spectrogram of the acceleration of the second vibration damping device and the spectrogram of the first vibration damping device based on a difference between a spectrogram of the output prediction value and a spectrogram of the output measurement value;
Including,
A device for analyzing the characteristics of vibration damping devices using machine learning. A step of acquiring an AI model by machine learning a correspondence relationship between a spectrogram of acceleration on an unsprung side of a first vibration damping device provided between a body and a wheel of a vehicle and a spectrogram of acceleration on an upsprung side of the first vibration damping device;
obtaining an input measurement, which is a measurement of acceleration at one of the unsprung and sprung sides of a second vibration damping device, and an output measurement, which is a measurement of acceleration at the other of the unsprung and sprung sides of the second vibration damping device;
inputting the input measurement values into the AI model as spectrograms to obtain a spectrogram of an output predicted value, which is a predicted value of acceleration on the other of the unsprung side and the sprung side of the second vibration damping device;
generating characteristic information indicating a difference between a spectrogram of the acceleration of the second vibration damping device and the spectrogram of the first vibration damping device based on a difference between a spectrogram of the output prediction value and a spectrogram of the output measurement value;
Including,
Characterization methods. A step of acquiring an AI model by machine learning a correspondence relationship between a spectrogram of acceleration on an unsprung side of a first vibration damping device provided between a body and a wheel of a vehicle and a spectrogram of acceleration on an upsprung side of the first vibration damping device;
obtaining an input measurement, which is a measurement of acceleration at one of the unsprung and sprung sides of a second vibration damping device, and an output measurement, which is a measurement of acceleration at the other of the unsprung and sprung sides of the second vibration damping device;
inputting the input measurement values into the AI model as spectrograms to obtain a spectrogram of an output predicted value, which is a predicted value of acceleration on the other of the unsprung side and the sprung side of the second vibration damping device;
generating characteristic information indicating a difference between a spectrogram of the acceleration of the second vibration damping device and the spectrogram of the first vibration damping device based on a difference between a spectrogram of the output prediction value and a spectrogram of the output measurement value;
The computer executes the
program.

Images