Vibration analysis apparatus and vibration analysis method

Abstract

Significantly reduces the time required for vibration analysis. [Solution] The vibration analysis device for a system including rotational force transmission elements comprises a machine learning model unit 51 that surrogate models the rotational force transmission elements whose torque changes with respect to changes in rotation angle have hysteresis, and a CAE model unit 50 that constructs a CAE model to which the output parameters output from the machine learning model unit 51 are applied, and performs vibration analysis of the system based on the CAE model.

Description

Technical Field

[0001] The present disclosure relates to a vibration analysis device and a vibration analysis method.

Background Art

[0002] For example, Patent Document 1 discloses a data prediction device that outputs continuous values applicable to a CAE model, which includes a machine learning model section constructed by machine learning and a physical model section constructed based on known natural laws. Patent Document 1 shows a method for deriving the load of an engine mount having the characteristics of a spring and a damper using the data prediction device. Specifically, the machine learning model section of the data prediction device derives each of the spring stiffness and the damping coefficient as parameters from variables consisting of the deformation amount and the deformation speed of the engine mount, and inputs them to the physical model section. The physical model section derives the load based on the input spring stiffness and damping coefficient and the variables consisting of the deformation amount and the deformation speed of the engine mount, and inputs it to the CAE model section.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] By the way, for example, as one of the rotational force transmission system elements of a vehicle, there is a dual mass flywheel. The dual mass flywheel is disposed between the engine and the transmission, and by including a spring inside, it absorbs torque fluctuations and rotational vibrations of the engine, and contributes to reducing so-called booming noise.

[0005] It should be noted that in the original text, there are some tags like which seem to be incomplete or have incorrect numbering in the context. Also, the translation of "

先行技術文献

Prior Art Documents

特許文献

Patent Documents

[0006] A typical surrogate model uses CAE simulation results as training data to train a machine learning model, and the trained machine learning model then calculates the simulation results. For example, by using a deep learning model trained on analysis results for the entire drive system, it is thought that vibration analysis can be performed much faster than with CAE.

[0007] However, during the vehicle development process, system configuration changes and specification changes often occur within the scope of surrogation (replacement of CAE with machine learning methods). When system configuration changes or specification changes occur within the surrogate scope, it becomes necessary to accumulate training data by performing vibration analysis of the modified system using CAE, and to retrain the deep learning model using the accumulated training data. However, it takes a long time to secure a sufficient amount of training data, and even if a sufficient amount of training data is accumulated, retraining a deep learning model using a large amount of training data is time-consuming. Therefore, when the entire system, which is subject to repeated changes, is designated as the surrogate scope, the development team faces the problem of not being able to fully realize the reduction in man-hours required for vibration analysis. This problem is not limited to systems including dual-mass flywheels, but can occur in other systems as well.

[0008] This disclosure is made in view of the above points, and its purpose is to significantly reduce the time required for vibration analysis. [Means for solving the problem]

[0009] To achieve the above objectives, the technology disclosed herein does not define the surrogate scope as the entire system, but rather defines the surrogate scope as the elements that cause the CAE analysis time to be long, and excludes the elements that allow the CAE analysis time to be short from the surrogate scope.

[0010] One aspect of this disclosure may be a vibration analysis device for a system including a rotational force transmission element. The vibration analysis device comprises a machine learning model unit that surrogate models the rotational force transmission element having hysteresis in torque change with respect to change in rotation angle, and a CAE model unit that constructs a CAE model to which the output parameters output from the machine learning model unit are applied, and performs vibration analysis of the system based on the CAE model.

[0011] In other words, a rotational force transmission system element whose torque change in response to a change in rotation angle has hysteresis becomes a nonlinear model during analysis. Analyzing such a nonlinear model using CAE, for example, would require an enormous amount of time. However, in this embodiment, such a rotational force transmission system element with nonlinear characteristics is surrogate-modeled, thus reducing analysis time. In addition, in this embodiment, the entire system is not surrogate-modeled; elements other than the rotational force transmission system element with nonlinear characteristics are analyzed by CAE in the CAE model section. Therefore, even if the system configuration or specifications are changed multiple times during the system development process, the accumulation and retraining of training data is unnecessary, further reducing analysis time.

[0012] The rotational force transmission element may be a component that houses a biasing member. Specifically, the rotational force transmission element may be a component in which hysteresis occurs due to frictional force generated between the biasing member and the wall surface of the component housing the biasing member, caused by the centrifugal force acting on the biasing member during rotation. For example, as the rotational speed of the rotational force transmission element increases, the centrifugal force acting on the biasing member increases, and as a result, the frictional force generated between the biasing member and the wall surface of the component housing the biasing member also increases. Alternatively, the rotational force transmission element may be a component in which hysteresis occurs due to frictional force generated between the biasing member and the wall surface of the component housing the biasing member, caused by the compressive force acting on the biasing member during rotational force transmission. When the compressive force acting on the biasing member during rotational force transmission increases, the frictional force generated between the biasing member and the wall surface of the component housing the biasing member may also increase. The hysteresis caused by this frictional force can sometimes lead to nonlinear torque changes. In such cases, surrogate modeling of the rotational force transmission system elements can significantly reduce analysis time.

[0013] The rotational force transmission element may be a dual-mass flywheel disposed between the engine and transmission of a vehicle. That is, a dual-mass flywheel has a structure in which a spring is interposed between the engine-side flywheel and the transmission-side flywheel. In such a structure, hysteresis is easily generated by fluctuations in rotational speed and rotational force, so using the vibration analysis device according to this embodiment will result in a significant reduction in analysis time.

[0014] The rotational force transmission system element may be surrogate modeled when the hysteresis exceeds a predetermined threshold determined from the spring constant of the biasing member and the frictional force, while the rotational force transmission system element may not be surrogate modeled when the hysteresis is below the predetermined threshold. For example, if the hysteresis is small enough to be negligible in vibration analysis, conventional vibration analysis methods are acceptable, but if the hysteresis is large, the vibration analysis method according to this embodiment can be used to obtain highly accurate analysis results in a short time.

[0015] The machine learning model unit can also be configured with a model trained on training data in which data where the torque change including hysteresis falls outside a predetermined range determined from the spring constant of the biasing member and the frictional force is discarded as noise. Alternatively, the machine learning model unit can be configured with a model trained on training data in which data in the region where the biasing member undergoes plastic deformation is discarded as noise. This can suppress divergence during analysis.

[0016] Another aspect of this disclosure may also be based on a vibration analysis method for a system including a rotational force transmission element. In this vibration analysis method, a machine learning model is constructed that surrogates the rotational force transmission element, which has hysteresis in torque changes with respect to changes in rotation angle. A CAE model is constructed to which the output parameters output from the machine learning model are applied, and a vibration analysis of the system is performed based on the CAE model. [Effects of the Invention]

[0017] As described above, the technology of this disclosure allows for vibration analysis of a vehicle powertrain system based on a CAE model to which output parameters output from a machine learning model unit that surrogate-models rotational force transmission system elements having hysteresis in response to changes in rotation angle are applied. Therefore, even if the system configuration or specifications change, the time required for vibration analysis can be significantly reduced. [Brief explanation of the drawing]

[0018] [Figure 1] Figure 1 shows a vibration analysis device according to one embodiment of the present invention. [Figure 2] Figure 2 is a schematic diagram of a vehicle powertrain system whose vibrations are analyzed by a vibration analysis device. [Figure 3] Figure 3 is a graph showing the vibration characteristics of a dual-mass flywheel. [Figure 4]FIG. 4 is a diagram schematically showing the relationship between the spring and the wall surface. [Figure 5] FIG. 5 is a diagram showing an example of a physical model of a dual mass flywheel. [Figure 6] FIG. 6 is a diagram showing an example of the hardware configuration of a vibration analysis device. [Figure 7] FIG. 7 is a diagram showing a configuration example of a model for executing vibration analysis processing. [Figure 8] FIG. 8 is a diagram showing a configuration example of a DMF single body excitation model. [Figure 9] FIG. 9 is a diagram showing the waveform of the excitation torque. [Figure 10] FIG. 10 is a diagram showing examples of the first to third teacher data. [Figure 11] FIG. 11 is a diagram showing the shape of the hysteresis loop for each rotational speed. [Figure 12] FIG. 12 is a diagram showing an example of the conditions for data cleansing. [Figure 13] FIG. 13 is a graph showing the analysis processing results. [Figure 14] FIG. 14 is a graph showing the relationship between the results of vibration analysis and the target.

Embodiments for Carrying Out the Invention

[0019] Hereinafter, embodiments of the present invention will be described in detail based on the drawings. Note that the following description of the preferred embodiments is merely illustrative in nature and is not intended to limit the present invention, its applications, or its uses. For example, the relative sizes and positional relationships of the components shown in the figures are for the purpose of explaining one embodiment and do not limit the present invention.

[0020] Figure 1 shows a vibration analysis device 1 according to one embodiment of the present invention. The vibration analysis device 1 according to this embodiment is a device for performing vibration analysis of a vehicle powertrain system 100 as shown in Figure 2. By using this vibration analysis device 1, a vibration analysis method for the vehicle powertrain system 100 can be performed. Although not shown, the vibration analysis device 1 can also be applied to vibration analysis of systems other than the vehicle powertrain system 100. For example, the present invention can be applied to systems that include elements that cannot be modeled with linear springs, such as systems that include centrifugal pendulum type dynamic vibration absorbers and systems that include torsional dampers in the motor mounting section.

[0021] The vehicle powertrain system 100 includes an engine 101, a transmission 102, a propeller shaft 103, a differential gear 104, and left and right drive shafts 105. The engine 101 may be a reciprocating engine or a rotary engine. Figure 2 illustrates the case where the vehicle powertrain system 100 is installed in a front-engine, rear-wheel-drive (FR) vehicle. However, the present invention can be used not only for vibration analysis of the vehicle powertrain system 100 installed in an FR vehicle, but also for vibration analysis of vehicle powertrain systems installed in front-engine, front-wheel-drive (FF) vehicles and four-wheel-drive vehicles.

[0022] In Figure 2, since it is a front-engine, rear-wheel-drive (FR) vehicle, the crankshaft 101a of the engine 101 is positioned to extend in the longitudinal direction of the vehicle. The transmission 102 is fastened to the rear end of the engine 101. The transmission 102 is a manual transmission, but it may also be an automatic transmission. Automatic transmissions include automatic transmissions with torque converters as well as dual-clutch transmissions. The propeller shaft 103 extends in the longitudinal direction of the vehicle, and its front end is connected to the output shaft (not shown) of the transmission 102. The differential gear 104 is connected to the rear end of the propeller shaft 103. Drive shafts 105 are connected to both the left and right sides of the differential gear 104. Wheels 106 are attached to each drive shaft 105.

[0023] A dual-mass flywheel (hereinafter referred to as "DMF") 107 is positioned between the engine 101 and the transmission 102. The DMF 107 is a component included in the vehicle powertrain system 100 and is an example of a rotational force transmission element. The DMF 107 itself is a conventionally known component, so detailed illustrations are omitted, but it comprises a primary flywheel fixed to the crankshaft 101a of the engine 101, a secondary flywheel fixed to the input shaft 102a of the transmission 102, and a metal spring 110 (schematically shown in Figure 4). The primary flywheel and the secondary flywheel are connected to each other in such a way that relative rotation is permitted within a predetermined angular range. By providing a stopper that restricts the rotation of the secondary flywheel, the relative rotational angular range between the primary flywheel and the secondary flywheel can be set to a predetermined angular range.

[0024] The spring 110 has an arc shape centered on the rotation axis C of the DMF 107 (shown in Figure 4) and is housed in the housing member 111 (its shape is schematically shown in Figure 4). The housing member 111 is formed in an arc shape centered on the rotation axis C. In other words, the spring 110 is housed in the housing member 111 in such a way that its arc shape is maintained. Thus, the housing member 111 is an example of a component that houses the biasing member, the spring 110. The spring 110 is also immersed in oil.

[0025] The spring 110 constantly biases one of the primary and secondary flywheels circumferentially relative to the other, and by expanding and contracting in response to torque fluctuations, it has the effect of absorbing gear noises from the transmission 102, booming noises and vibrations from the vehicle's powertrain system 100. For example, booming noise is an unpleasant sound heard in the passenger compartment, caused by torque fluctuations of the engine 101 acting as an excitation force, which is transmitted from the vehicle's powertrain system 100 to the suspension (not shown). This can be reduced by absorbing the torque fluctuations with the spring 110. In the case of an automatic transmission equipped with a torque converter, a lock-up damper (not shown) is disposed between the engine 101 and the transmission 102 instead of the DMF 107. This lock-up damper is also a conventionally known component and has a spring as a biasing member. In this case, the lock-up damper becomes a rotational force transmission element.

[0026] Figure 3 is a graph showing the vibration characteristics of the DMF107. The horizontal axis of the graph in Figure 3 represents the rotation angle of one of the primary flywheel and secondary flywheel relative to the other. The range is not particularly limited, but for example, it is considered to be in the range of 10 degrees to 15 degrees, and the primary flywheel and secondary flywheel can only rotate relative to each other within this angular range. The rotation angle increases as you move to the right on the horizontal axis.

[0027] The vertical axis of the graph in Figure 3 represents the torque output from the secondary flywheel when the primary flywheel is rotated relative to the secondary flywheel. Torque increases as you move upwards on the vertical axis. In Figure 3, the dashed line shows the torque characteristics of a linear spring (F=kx).

[0028] The vibration characteristics of DMF107 are not those of a linear spring, but are represented by line L1, which is drawn above the dashed line, and line L2, which is drawn below the dashed line. Lines L1 and L2 form a hysteresis loop of DMF107. When the primary flywheel is rotated relative to the secondary flywheel in the direction that compresses the spring 110, DMF107 exhibits vibration characteristics as shown by line L1. When the primary flywheel is rotated relative to the secondary flywheel in the direction that extends the spring 110 from its compressed state, DMF107 exhibits vibration characteristics as shown by line L2. The distance A between lines L1 and L2 is the hysteresis, and DMF107 is a component in which the torque change with respect to the change in rotation angle has hysteresis. In other words, DMF107 has nonlinearity.

[0029] The reason the DMF107 exhibits nonlinear characteristics is that, as shown in Figure 4, frictional force is generated between the spring 110 and the wall surface of the spring housing member 111 when the spring 110 expands and contracts. That is, since the spring 110 is held in an arc shape while housed in the housing member 111, centrifugal force acts on the spring 110 when the DMF107 rotates. When centrifugal force acts on the spring 110, the spring 110 is pressed against the wall surface of the housing member 111, so frictional force is generated between the spring 110 and the wall surface of the housing member 111. The higher the rotational speed of the DMF107 (the rotational speed of the crankshaft 101a), the greater the centrifugal force acting on the spring 110, and consequently, the greater the frictional force generated between the spring 110 and the wall surface of the housing member 111. In short, the DMF107 is a component in which hysteresis occurs due to the frictional force generated between the spring 110 and the wall surface of the housing member 111, caused by the centrifugal force acting on the spring 110 during rotation.

[0030] Furthermore, when the DMF107 transmits the rotational force of the crankshaft 101a of the engine 101, the compressive force acting on the spring 110 changes due to the change in the relative angle between the primary flywheel and the secondary flywheel. When the compressive force acting on the spring 110 increases, because the spring 110 is arc-shaped, it is pressed more strongly against the wall surface of the housing member 111, and consequently, the frictional force generated between the spring 110 and the wall surface of the housing member 111 also increases. Conversely, when the compressive force acting on the spring 110 decreases, the force pressing the spring 110 against the wall surface of the housing member 111 decreases, and consequently, the frictional force generated between the spring 110 and the wall surface of the housing member 111 also decreases. In short, the spring 110 is a component in which hysteresis occurs due to the frictional force generated between the spring 110 and the wall surface of the housing member 111 caused by the compressive force acting on the spring 110 during rotational force transmission.

[0031] As described above, the characteristics of the spring 110 housed in the housing member 111 of the DMF107 are nonlinear, making it difficult to simplify the physical model of the DMF107. Therefore, Figure 5 schematically shows an example of a physical model that precisely reproduces the DMF107 (detailed DMF model). In the physical model shown in Figure 5, the spring 110 is divided into multiple parts, resulting in multiple spring models 110A. Each spring model 110A has a compression stiffness corresponding to the actual spring 110. Between the multiple spring models 110A, elements 110B corresponding to the bending stiffness and mass of the actual spring 110 are set.

[0032] The model of the housing member 111 that houses the spring 110 includes an arc-shaped outer circumferential wall surface 111A and an inner circumferential wall surface 111B, and multiple spring models 110A are connected in series between the outer circumferential wall surface 111A and the inner circumferential wall surface 111B. The outer circumferential wall surface 111A and the inner circumferential wall surface 111B are contact elements that come into contact with the spring models 110A.

[0033] The primary flywheel model of the DMF107 is represented as a retainer 107A in Figure 5, and this retainer 107A is coupled with the outer peripheral wall surface 111A and the inner peripheral wall surface 111B to form a single unit. On the other hand, the secondary flywheel model of the DMF107 is represented as an output unit 107B in Figure 5, and this output unit 107B has degrees of freedom that allow it to rotate around the rotation axis C (shown in Figure 4) of the DMF107 relative to the retainer 107A. Thus, the physical model of the DMF107 is a model that divides a single spring 110 into multiple parts and precisely reproduces the actual behavior not only with torsional rigidity but also with numerous contact elements and bending rigidity, and it has dozens of degrees of freedom.

[0034] Since the DMF107 includes a spring 110 whose torque changes with respect to changes in rotation angle exhibit hysteresis, a precise modeling of the nonlinearity is necessary to make the DMF107 a model that can be analyzed by CAE, as shown in Figure 5. Consequently, the CAE analysis time for the DMF107 becomes lengthy.

[0035] Therefore, a method known as surrogation, which involves replacing CAE with machine learning-based analysis, can be considered. Generally, surrogation can significantly reduce analysis time, but in the case of vibration analysis of a vehicle powertrain system 100, surrogating the entire vehicle powertrain system 100 may not yield sufficient reductions in the amount of work required for vibration analysis.

[0036] In other words, during the vehicle development process, the type (gasoline engine, diesel engine, rotary engine, etc.), mounting position, and orientation (longitudinal or transverse) of the engine 101 may change; the number of cylinders and displacement of the engine 101 may change; the type, number of gears, and mounting position of the transmission 102 may change; the length and strength requirements of the propeller shaft 103 may change; the mounting position of the differential gear 104 may change; and the length and strength requirements of the drive shaft 105 may change. In addition, the vehicle may become a hybrid vehicle in which the engine 101 and an electric motor are combined. In short, during the vehicle development process, changes to the configuration and specifications of the vehicle powertrain system 100 often occur.

[0037] If the entire vehicle powertrain system 100 is surrogate-based, then, for example, if the type of engine 101 is changed, it becomes necessary to accumulate training data by performing vibration analysis of the vehicle powertrain system 100 after the engine 101 has been changed using CAE, and to retrain the deep learning model using the accumulated training data. However, it takes a long time to secure enough training data to enable retraining, and even if a sufficient amount of training data can be accumulated, retraining a deep learning model using a large amount of training data is time-consuming, and it is also necessary to examine the validity of the retrained model, which requires considerable effort. Even after retraining is completed, if further system changes or specification changes are made afterward, it becomes necessary to accumulate training data, retrain, and examine the validity of the retrained model for the changed system, leading to a further increase in the amount of work required.

[0038] In response to this, the vibration analysis device 1 of this embodiment does not define the surrogate range as the entire vehicle powertrain system 100, but rather defines the DMF107, which is an element that causes the CAE analysis time to be long, as the surrogate range, while excluding elements that allow for a short CAE analysis time (elements other than DMF107) from the surrogate range.

[0039] Specifically, the vibration analysis device 1 can have a hardware configuration as shown in Figures 1 and 6. The vibration analysis device 1 includes, for example, a computer 10, a display unit 20, and an operation unit 30. The computer 10 may be a desktop general-purpose computer, a notebook general-purpose computer, or a dedicated computer with a configuration specialized for vibration analysis. As shown in Figure 6, the computer 10 has a control unit 11, a storage unit 12, and a communication module 13. The control unit 11 has, for example, a CPU (Central Processing Unit) 11a, a ROM (Read Only Memory) 11b, and a RAM (Random Access Memory) 11c. In addition to the above, the control unit 11 may also have a GPU (Graphics Processing Unit), etc.

[0040] The storage unit 12 is composed of a read / write storage device such as a solid-state drive (SSD) or a hard disk drive (HDD), and is connected to the control unit 11, enabling data transmission and reception between the two. The storage unit 12 stores the operating system of the computer 10 and a program for operating the computer 10 as a vibration analysis device 1. The control unit 11 operates according to the program stored in the storage unit 12 and executes the vibration analysis process described later. The program for operating the computer 10 as a vibration analysis device 1 does not have to be stored in the storage unit 12; for example, it may be stored in the server 40 described later, or recorded on a storage medium such as a CD-ROM or DVD. As for how to provide the program for operating the computer 10 as a vibration analysis device 1 to the user, it may be provided stored on a storage medium, or it may be stored on a server (not shown) and provided in a form where the user accesses the server and downloads it.

[0041] The communication module 13 is connected to a network such as an intranet or the internet, and enables the transmission and reception of data between the computer 10 and the server 40. The server 40 may be provided as needed, and the computer 10 may include the server 40.

[0042] The display unit 20 is composed of, for example, a liquid crystal display panel or an organic EL (Electro-Luminescence) panel. The display unit 20 is connected to the control unit 11 and controlled by the control unit 11, and is capable of displaying various user interfaces, analysis models, analysis results, etc.

[0043] The operation unit 30 consists of equipment for the user to operate the computer 10. The operation unit 30 includes, for example, a keyboard 30a and a mouse 30b, but is not limited to these, and may also include various pointing devices such as a touch panel that detects touch operations by the user.

[0044] Figure 7 shows an example of the configuration of a model in which the vibration analysis device 1 performs vibration analysis processing. As shown in this figure, by executing a program to operate the computer 10 as the vibration analysis device 1, a CAE model unit 50 and a machine learning model unit 51 are constructed in the control unit 11 of the computer 10. The machine learning model unit 51 is constructed by pre-training with training data and outputs results for the input data. In this embodiment, the machine learning model unit 51 is a surrogate model of the DMF 107 (surrogate model). The CAE model unit 50 constructs a CAE model to which the output parameters output from the machine learning model unit 51 are applied, and performs vibration analysis of the vehicle powertrain system 100 based on the CAE model.

[0045] The CAE model unit 50 performs a CAE analysis at time step t and inputs the analysis results into the machine learning model unit 51. The machine learning model unit 51, which receives the analysis results from the CAE model unit 50, also receives explanatory variables to ensure generality. When the machine learning model unit 51 receives the analysis results and explanatory variables from the CAE model unit 50 as input data, it performs inference processing based on the input data and inputs the output parameters obtained by the inference processing into the CAE model unit 50 as time step t+1. The CAE model unit 50 applies the output parameters output from the machine learning model unit 51 to the CAE model and performs vibration analysis of the vehicle powertrain system 100 at time step t+1 based on the CAE model. By continuing this process, vibration analysis for a predetermined time can be performed continuously.

[0046] When constructing the machine learning model unit 51, the input and output of the surrogate model are designed first. Examples of input parameters to the surrogate model include the rotational speed (rpm) of the primary flywheel fixed to the crankshaft 101a of the engine 101, the relative twist angle (deg) between the primary flywheel and the secondary flywheel, the relative angular velocity (deg / s) between the primary flywheel and the secondary flywheel, the stiffness (Nm / deg) of the spring 110, and the velocity-dependent damping coefficient (s) of the spring 110. The rotational speed (rpm) of the primary flywheel, the relative twist angle (deg) between the primary flywheel and the secondary flywheel, and the relative angular velocity (deg / s) between the primary flywheel and the secondary flywheel are parameters input to ensure the accuracy of the vibration analysis and correspond to the convergence result (t[step]) of the CAE. The stiffness (Nm / deg) and velocity-dependent damping coefficient (s) are parameters input to ensure generality and are constants during the vibration analysis.

[0047] The rotational speed (rpm) of the primary flywheel is input to represent the frictional force (frictional force between spring 110 and the wall surface of housing member 111) that depends on the centrifugal force of spring 110. The relative twist angle (deg) between the primary flywheel and the secondary flywheel is input to represent the stiffness term in the equation of motion and the frictional force (frictional force between spring 110 and the wall surface of housing member 111) that depends on the load (twist angle). The relative angular velocity (deg / s) between the primary flywheel and the secondary flywheel is input to represent the damping term in the equation of motion.

[0048] Stiffness (Nm / deg) is frequently changed during the development of the vehicle powertrain system 100 and has a significant impact on the accuracy of vibration analysis. This stiffness (Nm / deg) is a design parameter. Similarly, the velocity-dependent damping coefficient (s) is frequently changed during the development of the vehicle powertrain system 100 and has a significant impact on the accuracy of vibration analysis. This velocity-dependent damping coefficient (s) is an identification parameter.

[0049] An example of an output parameter from a surrogate model is the total generated torque (N·m) between the primary and secondary flywheels. This total generated torque (N·m) is a parameter that is sequentially fed back to the CAE model as an external force term (t+1[step]), and is also a parameter that ensures the accuracy of the vibration analysis. As described above, when designing the input and output of a surrogate model, the surrogate range, which becomes a black box, is kept to a minimum in order to ensure the versatility of the model, and the input and output parameters are selected from the viewpoint of reproduction characteristics and versatility.

[0050] Furthermore, after designing the input / output for the surrogate model, training data is generated for learning. The training data can be obtained by performing excitation analysis on the DMF107 alone, which is within the surrogate range. To ensure versatility, excitation analysis is performed under approximately 100,000 conditions or more, and the resulting analysis data is used as training data. Data processing and learning processes for the training data can be performed using, for example, a GPU. The training data may be stored on, for example, server 40 or storage unit 12. During learning, learning may be performed using computer 10 or another computer (not shown).

[0051] When performing excitation analysis of the DMF107 unit alone, a DMF unit excitation model is prepared. This DMF unit excitation model is a forced torque excitation model constructed to obtain comprehensive vibration characteristics of the DMF107, and the model shown in Figure 8 can be cited as an example. In Figure 8, the detailed DMF model is the model shown in Figure 5, and the stiffness of spring 110 and the velocity-dependent damping coefficient are the analysis condition parameters. The excitation torque when forcing torque excitation of the input mass is also an analysis condition parameter.

[0052] The rotational speed of the input mass corresponds to the rotational speed of the primary flywheel. The torque generated between the dummy mass and the output mass corresponds to the total generated torque (N·m) between the primary flywheel and the secondary flywheel. The relative twist angle between the input mass and the output mass corresponds to the relative twist angle between the primary flywheel and the secondary flywheel. The relative angular velocity between the input mass and the output mass corresponds to the relative angular velocity between the primary flywheel and the secondary flywheel. The output mass has a maximum inertia, and the excitation analysis at a specific rotational speed is performed using the inertial energy of the maximum inertia.

[0053] By changing the analysis condition parameters, excitation analysis can be performed with varying excitation force and rotational speed using only the DMF107. For example, the excitation torque can be input to the input mass in the form of a waveform as shown in Figure 9. The input waveform is generated by changing the amplitude and wavelength, and the excitation torque of a magnitude based on the generated input waveform is input to the input mass. In Figure 9, the amplitude and T of the excitation torque are shown. DC This can be determined by the output range of engine 101. Amplitude and T DC These are excitation torque conditions, and multiple settings are set in the low-rigidity region, the high-rigidity region, and the partial region (intermediate region), respectively. In the low-rigidity region, the secondary flywheel often hits the stopper, so the conditions are set more closely than in the high-rigidity region. Also, in Figure 9, "f" can be determined based on the number of cylinders of engine 101, and can be obtained, for example, by converting to the basic order for 4-cylinder and 6-cylinder engines. "f" is the excitation frequency, and multiple settings are set in the low-rigidity region, the high-rigidity region, and the partial region.

[0054] The stiffness of spring 110 can be determined by the output range of engine 101. The speed-dependent damping coefficient of spring 110 includes the viscosity range of the oil in which spring 110 is immersed, and is the damping coefficient of the entire DMF 107. The stiffness of spring 110 and the speed-dependent damping coefficient of spring 110 are versatile parameters, and multiple settings are provided for the low-stiffness range, the high-stiffness range, and the partial range, respectively.

[0055] Figure 10 shows examples of the first to third training data obtained by excitation analysis of a DMF-only excitation model. The graphs shown in Figure 10 can be displayed on the display unit 20 by the user operating the operation unit 30, and can be viewed by the user. The first training data is training data in the region where the DMF 107 is sliding, and the waveform is similar to the waveform shown in Figure 3. The second training data is training data in the region where the DMF 107 is partially fixed. The third training data is training data in the region where the DMF 107 is fixed. In this way, by performing vibration analysis under multiple conditions using a model like the one shown in Figure 8, the behavior of sliding, partial fixing, and fixing can be comprehensively obtained.

[0056] Figure 11 shows the training data for the hysteresis loop of the DMF107. The graph shown in Figure 11 was also obtained by performing vibration analysis under multiple conditions using the model shown in Figure 8, and can be displayed on the display unit 20 by the user operating the control unit 30, allowing the user to verify it. As shown in Figure 11, changing the rotational speed of the engine 101 results in hysteresis loops with different shapes. By using hysteresis loops with different shapes for each rotational speed as training data, it is possible to reproduce the DMF107, which exhibits nonlinear characteristics due to frictional force dependent on the torsion angle and frictional force dependent on the rotational speed. This allows for the acquisition of comprehensive torsional characteristics.

[0057] The training data used does not have to be the entire dataset; for example, a stationary time interval (stationary interval) can be extracted, and only the analysis results of the stationary interval can be used as training data. Parameters to ensure generality are added as constants to the analysis results.

[0058] While a large amount of training data can be obtained by setting numerous conditions, this data may contain disturbances (noise). In a machine learning model 51 trained on training data containing disturbances, the analysis may diverge during the accuracy verification stage after training. Therefore, data cleansing is performed to construct training data that does not contain disturbances. Specifically, the machine learning model 51 is trained on training data that does not contain data where the torque change, including hysteresis, falls outside a predetermined range determined from the spring constant of the spring 110 and the frictional force (the frictional force between the spring 110 and the wall surface of the housing member 111). Figure 12 shows an example of data cleansing conditions and indicates the cleansing threshold for each stiffness condition. The cleansing threshold is a value that defines the outer edge of a predetermined range determined from the spring constant of the spring 110 and the frictional force, and in Figure 12, the grayed-out area corresponds to the predetermined range.

[0059] The point at which the secondary flywheel strikes the stopper can be defined as the starting point of plastic deformation of the spring 110. In Figure 12, the characteristics of each specification during the excitation analysis of the DMF single-unit excitation model are shown with solid lines, and the encompassing line for the stopper strike point of each specification is shown with a dashed line. In the DMF 107, relative rotation does not occur beyond the stopper strike point, so the training data in the region beyond the encompassing line is discarded. In other words, the DMF 107 has different plastic deformation start angles for the spring 110 for each specification, and by utilizing this characteristic to perform data cleansing and using the machine learning model 51 trained on the data-cleaned training data, the analysis does not diverge at the stage of accuracy verification after training. Thus, the machine learning model 51 is trained on training data that does not include data in the region where the spring 110 undergoes plastic deformation.

[0060] Furthermore, if the stiffness of the spring 110 is high, the hysteresis will be smaller compared to the case where the stiffness is low. When the stiffness of the spring 110 is sufficiently high, the hysteresis becomes small enough to be almost negligible in vibration analysis, so in such cases, it becomes possible to perform a highly accurate analysis in a short time without using a surrogate model. Therefore, the vibration analysis device 1 may be configured to surrogate model the DMF 107 when the hysteresis exceeds a predetermined threshold obtained from the spring constant of the spring 110 and the friction force (friction force between the spring 110 and the wall surface of the housing member 111), but not when the hysteresis is below the predetermined threshold.

[0061] In the vibration analysis method according to this embodiment, a machine learning model is constructed by surrogating the DMF107, which has hysteresis in the torque change in response to changes in rotation angle, as described above. A CAE model is also constructed to which the output parameters output from the machine learning model are applied. Then, a vibration analysis of the vehicle powertrain system 100 is performed based on the CAE model. The results of the vibration analysis can be displayed on the display unit 20 by operating the operation unit 30.

[0062] Figure 13 is a graph showing the analysis results, illustrating the relationship between engine speed and the vibration response of the drive shaft 105. Figure 14 is a graph showing the relationship between the vibration analysis results and the target. By performing order tracking processing on the analysis results shown in Figure 13, a graph of the analysis results shown in Figure 14 can be obtained. The user can display the graph shown in Figure 14 on the display unit 20 to compare the analysis results with the target and distinguish between areas where the analysis results meet the target and areas where they do not. If the target is not met, the user can change the specifications, perform the vibration analysis again, and then compare it with the target.

[0063] (Effects of the embodiment) As described above, in this embodiment, when performing vibration analysis of a system including a DMF107 with nonlinear characteristics, such as a vehicle powertrain system 100, a CAE model is constructed to which output parameters output from a machine learning model unit 51 that has modeled the DMF107 as a surrogate model are applied, and vibration analysis of the vehicle powertrain system 100 can be performed based on the CAE model.

[0064] By surrogate modeling the DMF107, which has nonlinear characteristics, the time required for vibration analysis is reduced. Moreover, since the entire vehicle powertrain system 100 is not surrogate modeled, and elements other than the DMF107 are analyzed by CAE in the CAE model unit 50, even if the configuration or specifications are changed multiple times during the development process of the vehicle powertrain system 100, the accumulation and retraining of training data is unnecessary, and further reductions in analysis time are possible.

[0065] The embodiments described above are merely illustrative in all respects and should not be interpreted restrictively. Furthermore, any modifications or changes that fall within the equivalent scope of the claims are all within the scope of the present invention. [Industrial applicability]

[0066] As explained above, the technology relating to this disclosure can be used, for example, for vibration analysis of systems including rotational force transmission elements. [Explanation of symbols]

[0067] 1 Vibration analysis device 50 CAE Model Section 51 Machine Learning Models Department 107 Dual Mass Flywheel (DMF) 110 spring

Claims

1. A vibration analysis apparatus for a system including rotational force transmission elements, A machine learning model unit that surrogates the rotational force transmission system element having hysteresis in torque change with respect to change in rotation angle, A vibration analysis apparatus comprising: a CAE model unit that constructs a CAE model to which output parameters output from the machine learning model unit are applied, and a CAE model unit that performs vibration analysis of the system based on the CAE model.

2. In the vibration analysis apparatus according to claim 1, The aforementioned rotational force transmission system element is a component that houses a biasing member, in a vibration analysis device.

3. In the vibration analysis apparatus according to claim 2, The rotational force transmission element is a component in which hysteresis occurs due to a frictional force generated between the biasing member and the wall surface of the member housing the biasing member, caused by the centrifugal force acting on the biasing member during rotation.

4. In the vibration analysis apparatus according to claim 2, The rotational force transmission system element is a component in which hysteresis occurs due to a frictional force generated between the biasing member and the wall surface of the member housing the biasing member, caused by a compressive force acting on the biasing member during rotational force transmission.

5. In the vibration analysis apparatus according to claim 2, The aforementioned rotational force transmission element is a dual-mass flywheel disposed between the engine and transmission of a vehicle, in a vibration analysis device.

6. In the vibration analysis apparatus according to claim 3 or 4, The biasing member is made of a spring, A vibration analysis device that surrogate models the rotational force transmission system elements when the hysteresis exceeds a predetermined threshold determined from the spring constant of the biasing member and the frictional force, while not surrogate models the rotational force transmission system elements when the hysteresis is below the predetermined threshold.

7. In the vibration analysis apparatus according to claim 3 or 4, The biasing member is made of a spring, The vibration analysis device is characterized in that the machine learning model unit is trained using training data that does not include data in which the torque change, including the hysteresis, falls outside a predetermined range determined from the spring constant of the biasing member and the frictional force.

8. In the vibration analysis apparatus according to claim 3 or 4, The biasing member is made of a spring, The vibration analysis device is characterized in that the machine learning model unit is trained using training data that does not include data from the region in which the biasing member undergoes plastic deformation.

9. A method for analyzing the vibration of a system including rotational force transmission elements, A machine learning model was constructed that surrogates the rotational force transmission system element, which has hysteresis in torque change in response to changes in rotation angle. A CAE model is constructed to which the output parameters output from the aforementioned machine learning model are applied. A vibration analysis method that performs vibration analysis of the system based on the CAE model.

Images