Vibration control system for structures

Abstract

To improve vibration control performance.SOLUTION: A vibration control system of a structure is represented by a mass model comprising: a support end; a first mass point aligned with the support end in a predetermined direction; a first spring element connecting the support end and the first mass point, the first spring element generating elastic reaction force that acts along an orthogonal direction orthogonal to the predetermined direction; a second mass point that can be displaced relative to the first mass point in the orthogonal direction; and an exciter that is connected to the first mass point, and excites the second mass point in the orthogonal direction. In the vibration control system, a control rule for controlling exciting force of the exciter is determined using reinforcement learning.SELECTED DRAWING: Figure 4

Description

[Technical Field]

[0001] The present invention relates to a vibration damping system for structures. [Background Art]

[0002] As a technique for reducing vibration of structures such as horizontal vibration of buildings and vertical vibration of floors caused by various disturbances, AMD (Active Mass Damper), which performs vibration damping by actively moving a mass, is known (see, for example, Patent Document 1). In AMD, control is generally performed in accordance with a control law based on control engineering. [Prior Art Literature] [Patent Literature]

[0003] [Patent Document 1] Japanese Unexamined Patent Publication No. Hei 1-275867 [Summary of the Invention] [Problem to be Solved by the Invention]

[0004] A control law based on control engineering requires control theory, and it has been difficult to maximize the performance of AMD (that is, to optimally move the mass).

[0005] The present invention has been made in view of such a problem, and an object of the present invention is to improve vibration damping performance. [Means for Solving the Problem]

[0006] The main invention for achieving the above objective is a vibration damping system for a structure represented by a mass model, comprising: a support end; a first mass point aligned with the support end in a predetermined direction; a first spring element connecting the support end and the first mass point, which generates an elastic reaction force acting along an orthogonal direction perpendicular to the predetermined direction; a second mass point that is relatively displaceable with respect to the first mass point in the orthogonal direction; and an exciter connected to the first mass point that excites the second mass point in the orthogonal direction, wherein a control law for controlling the excitation force of the exciter is determined using reinforcement learning, the second mass point is the mass of a mass damper that suppresses the vibration of the first mass point, and the reinforcement learning is a framework for an agent placed in a certain environment to acquire optimal behavioral rules through interaction with the environment, and takes a learning wave as input. The first mass point and the second mass point Simulation of a point mass model Under the same configuration Repeated multiple times i. Obtain a control law that uses the excitation force of the aforementioned vibrator as the output. It is characterized by the following: Other features of the present invention will be made clearer by description in this specification and the accompanying drawings. [Effects of the Invention]

[0007] According to the present invention, vibration damping performance can be improved. [Brief explanation of the drawing]

[0008] [Figure 1] Figure 1A is a diagram showing the vibration damping system configuration of a structure according to the first embodiment. Figure 1B is a model of Figure 1A. Figure 1C is an equivalent diagram of Figure 1B. [Figure 2] This is a block diagram showing the configuration of the control section of the actuator 40. [Figure 3] This is a flowchart illustrating the construction of an AMD control system in a comparative example. [Figure 4] This is a flowchart of the AMD control system construction in this embodiment. [Figure 5] Figure 5A shows the vibration damping system configuration of the structure according to the second embodiment. Figure 5B is a model of Figure 5A. [Figure 6]Figure 6A shows the vibration damping system configuration of the structure according to the third embodiment. Figure 6B is a model of Figure 6A. [Figure 7] This figure shows a modified example of the third embodiment. [Figure 8] This figure shows another modified example of the third embodiment. [Figure 9] Figure 9A shows the vibration damping system configuration of the structure according to the fourth embodiment. Figure 9B is a model of Figure 9A. [Figure 10] This figure shows a modified example of the fourth embodiment. [Figure 11] This figure shows another modified example of the fourth embodiment. [Modes for carrying out the invention]

[0009] The following matters become clear from this specification and the accompanying drawings:

[0010] A vibration damping system for a structure represented by a mass model is revealed, comprising: a support end; a first mass point aligned with the support end in a predetermined direction; a first spring element connecting the support end and the first mass point, which generates an elastic reaction force acting along an orthogonal direction perpendicular to the predetermined direction; a second mass point that can be displaced relative to the first mass point in the orthogonal direction; and an exciter connected to the first mass point that excites the second mass point in the orthogonal direction, wherein the control law for controlling the excitation force of the exciter is determined using reinforcement learning. This type of vibration damping system for structures can achieve a control law superior to control laws based on control theory (e.g., optimal control). This enhances the ability of vibration damping by exciting a second mass point with an exciter (AMD), thereby improving vibration damping performance.

[0011] In a vibration damping system for such a structure, it is desirable that the reward in the reinforcement learning is determined based on the conserved quantity of the first point mass. Such a vibration damping system for a structure allows for the acquisition of an excellent control law.

[0012] In the vibration damping system for such a structure, it is preferable that the stored amount is a competitive type. According to such a vibration damping system for a structure, an excellent control law can be obtained.

[0013] In the vibration damping system for such a structure, it is preferable that the stored amount is energy. According to such a vibration damping system for a structure, a more excellent control law can be obtained.

[0014] In the vibration damping system for such a structure, it is preferable that the energy includes at least kinetic energy. According to such a vibration damping system for a structure, vibration damping performance can be improved.

[0015] In the vibration damping system for such a structure, it is preferable that the energy further includes potential energy of a spring system. According to such a vibration damping system for a structure, vibration damping performance can be further improved.

[0016] The vibration damping system for such a structure may further comprise a second spring element provided in parallel with the vibrator, whose spring direction is the orthogonal direction. According to such a vibration damping system for a structure, high vibration damping performance can be achieved even when the second spring element is provided.

[0017] In the vibration damping system for such a structure, the predetermined direction is a vertical direction and the orthogonal direction is a horizontal direction, and a rolling bearing that rollably supports the second mass point in the horizontal direction may be provided on the first mass point. According to such a vibration damping system for a structure, vibration of the first mass point in the horizontal direction can be damped by vibrating (rolling) the second mass point in the horizontal direction on the first mass point.

[0018] The vibration damping system for such a structure is such that the predetermined direction is vertical, the orthogonal direction is horizontal, and the second spring element may be a laminated rubber that supports the second mass on the first mass. With this type of vibration damping system for structures, the horizontal vibration of the first mass can be damped by exciting the second mass, which is supported by laminated rubber on top of the first mass, in the horizontal direction.

[0019] The vibration damping system for such a structure may be such that the predetermined direction is horizontal and the orthogonal direction is vertical. Such a vibration control system for structures can suppress vertical vibrations, such as those in bridges.

[0020] ===First Embodiment=== <<About the vibration control system configuration>> Figure 1A is a diagram showing the vibration damping system configuration of the structure according to the first embodiment, Figure 1B is a model of Figure 1A, and Figure 1C is an equivalent diagram of Figure 1B. Figure 2 is a block diagram showing the configuration of the control section of the actuator 40.

[0021] As shown in Figure 1A, the structure to be vibration-damped in the first embodiment is a bridge 10 that is fixed at both ends (support ends 12) in the longitudinal direction and spanned horizontally. A mass 20 is provided in the longitudinal center of this bridge 10 via a spring 30 and an actuator 40. Furthermore, as shown in Figure 2, a group of sensors 60 (for example, an acceleration sensor to detect acceleration, a displacement sensor to detect displacement (relative displacement, etc.), etc.) that detects the state of the bridge 10 and the mass 20 is provided, and a controller 70 that calculates state quantities and controls the excitation of the actuator 40 based on the output of each sensor in the group of sensors 60 is provided.

[0022] When Figure 1A is modeled as a point mass model, it becomes the two-point mass model shown in Figure 1B, and further simplification results in Figure 1C. In Figures 1B and 1C, the bridge 10 and mass 20 are each represented as point masses. Of these, the point mass representing the bridge 10 corresponds to the first point mass, and the point mass representing mass 20 corresponds to the second point mass.

[0023] In the point mass models in Figures 1B and 1C, the bridge 10 is aligned horizontally with the support end 12. The spring element 14 (corresponding to the first spring element) shown in the point mass models in Figures 1B and 1C corresponds to the stiffness of the bridge 10. The spring element 14 connects the support end 12 and the bridge 10 (first point mass), and generates an elastic reaction force acting along a direction perpendicular to the direction in which the bridge 10 and the support end 12 are aligned (in this case, the horizontal direction) (in this case, the vertical direction).

[0024] Mass 20 is a weight (mass body) for AMD control and is mounted on the bridge 10 via a spring 30 (aligned vertically with the bridge 10). This allows mass 20 to be displaced vertically relative to the bridge 10.

[0025] The spring 30 (corresponding to the second spring element) is positioned between the bridge 10 and the mass 20 (parallel to the actuator 40) so that its spring direction is vertical, and supports the mass 20.

[0026] The actuator 40 (vibrator) is connected to the bridge 10 and vibrates the mass 20 vertically based on control by the controller 70.

[0027] By adjusting (controlling) the excitation force (control force) of this actuator 40, the vertical swaying (up and down vibration) of the bridge 10 caused by people walking on it can be suppressed.

[0028] As shown in Figure 1C, the vibration damping system of the structure in this embodiment is a two-mass system model, and its behavior can be simulated by solving the following equation (1) by numerical analysis. JPEG0007917038000001.jpg6170

[0029] Furthermore, each term of equation (1) is shown in equations (2) to (5) below. JPEG0007917038000002.jpg90170 Here, JPEG0007917038000003.jpg163170

[0030] <<AMD Control (Comparative Example)>> Figure 3 is a flow chart of constructing an AMD control system in a comparative example. Here, optimal control, which is known as an example of active control, is used.

[0031] First, each parameter is set (S001). Here, for the purpose of minimizing the absolute values of the [displacement, velocity] and control force of [mass 20, bridge 10] as much as possible in each step, the evaluation function J is defined as the weighted sum of the squares of the [displacement, velocity] and control force of [mass 20, bridge 10], as shown in the following equation (6). JPEG0007917038000004.jpg8170

[0032] In equation (6), x(t) is the aforementioned equation (4), u(t) is the control force, and ε is the weight applied to the control force. Further, Q is the weight applied to x(t), and JPEG0007917038000005.jpg20170 .

[0033] JPEG0007917038000006.jpg28170

[0034] JPEG0007917038000007.jpg39170

[0035] Next, the feedback gain of the control law is calculated (S002). When the feedback gain of x(t) that minimizes the evaluation function J is theoretically obtained, the control force u(t) is expressed by the following equation (9). JPEG0007917038000008.jpg8170 Here, P is the solution to the Riccati equation expressed by the following equation (10). JPEG0007917038000009.jpg8170

[0036] JPEG0007917038000010.jpg27170 By solving this formula, the control law can be obtained.

[0037] Next, the effect of the obtained control law is confirmed by conducting a simulation in which walking excitation force is input (S003).

[0038] If the control law verification result is unsatisfactory (No in S004), the process returns to step S001 and the parameters are reset. On the other hand, if the control law verification result is satisfactory (Yes in S004), implementation is performed (S005). Specifically, the control force u(t) is calculated based on x(t) acquired and calculated in real time from each sensor (such as acceleration sensors and displacement sensors) in the sensor group 60 and applied to the actuator 40.

[0039] In this example, optimal control (including optimal feedback control, LQ control, LQR control, and optimal regulator control) was explained, but other active control methods (control laws) such as LQG control, bilinear optimal control, instantaneous optimal control, suboptimal control, suboptimal merged control, pole placement control, skyhook control, H∞ control, PID control, sliding mode control, On / Off control, EF (Energy Function) control, fuzzy control, predictive control, hysteresis control, direct output feedback control, direct velocity feedback control (DVFB control), disturbance cancellation control, μ synthesis control, and gain scheduled control are known.

[0040] However, even when using the control law as described above, there is a risk that control that maximizes the performance of AMD cannot be performed. Specifically, since the aforementioned optimal control is premised on a control force expressed as a linear combination of state quantities, it is necessary to carry out design by allowing for a safety factor in response to various constraints, which may result in an inefficient control law. In addition, to satisfy various constraints that differ for each structure, a designer was required to adjust parameters through trial and error via multiple simulations. Therefore, the design relies heavily on the designer's experience, an optimal solution is not necessarily obtained, and design requires time and labor. Furthermore, since it is necessary to obtain a theoretical solution, the degree of freedom of the evaluation function is low, and it is difficult to perform optimization for a specific disturbance. In addition, it has been difficult to apply linear theory to structures that have non-linear restoring force, primarily such as concrete structures.

[0041] Therefore, in the present embodiment, vibration damping performance is improved by using artificial intelligence (hereinafter referred to as AI) for AMD control. Machine learning using AI is classified into three categories: supervised learning, unsupervised learning, and reinforcement learning, and the present embodiment uses reinforcement learning among these.

[0042] <<AMD control (present embodiment)>> Fig. 4 is a flow diagram for constructing an AMD control system according to the present embodiment. In the present embodiment, reinforcement learning by AI is used as described above. Reinforcement learning is a method that learns actions that maximize value through trial and error even without teacher data. In addition, Q-Learning and DQN (Deep Q-Learning) are used as reinforcement learning algorithms.

[0043] <Parameter setting> First, each parameter (state st, reward Rt, action at) is set (S101).

[0044] [Action at] Action at is the control force output to actuator 40 at time t. The discretization number and maximum control force are set for the control force of actuator 40. In this embodiment, the maximum control force is set to 100 (N) and divided into 5 parts as follows. at={-100,-50,0,50,100} ····(12)

[0045] [State st] State st is the state quantity of bridge 10 and mass 20 at time t, and sets what input determines action at. Ideally, we would like to reward actions that reduce the state quantity, but in the case of vibrations caused by springs, it is necessary to distinguish whether the reduction in the state quantity is due to a natural decrease in vibration or a decrease due to the control of actuator 40. In particular, in the case of a competitive reward setting (described later) as in this embodiment, if the vibration frequencies of the two are different, the effect of vibration becomes significant due to the phase difference in their responses.

[0046] To clarify this, we consider a conserved quantity S(t) defined by equation (13) below as a state variable that does not depend on the phase during the response, while minimizing the effects of vibration. JPEG0007917038000011.jpg58170

[0047] JPEG0007917038000012.jpg28170

[0048] JPEG0007917038000013.jpg47170

[0049] [Reward rt] JPEG0007917038000014.jpg38170

[0050] JPEG0007917038000015.jpg6170 By expanding the reward set in this embodiment, it can be made into a competitive type as represented by the following equation (16). JPEG0007917038000016.jpg47170

[0051] The opponent's S(t) can be calculated based on prior response analysis for various scenarios, including no mass, passive control (with or without attenuation), existing active control, and the opponent's own active control (self-play), or it can be a function F(t) with respect to time t set appropriately (including 0 and constants). • If the learning subject is semi-active, the opponent may also be semi-active, or the opponent may be as described above. While ε=δ is the standard, setting the value of ε to be smaller than δ allows you to set a stronger opponent. • Targeting opponents that are too strong from the start will prevent the learning process from converging. It's better to gradually increase the difficulty of your opponents.

[0052] Other examples of immediate rewards (SRTs) include the following (combinations are optional): Let S(t) be a linear combination of the mean, sum, and maximum values ​​of the absolute values, raw values, reciprocals, and absolute values ​​of the reciprocals of the displacement, velocity, and acceleration of the [bridge 10 and mass 20] for the most recent n time steps. The reward is the difference, double difference, derivative, second derivative, and reciprocal of S(t) (in this case, there is no opponent). • You may also include the term for cell 20 in S(t). You may give a negative reward at each step until convergence occurs. • If the clearance is exceeded, a negative reward may be given. • You may clip the rewards.

[0053] JPEG0007917038000017.jpg6170 · Rewards are given at regular intervals (including only at the end) according to a special index that represents the magnitude of shaking over time. The special index includes values ​​such as the residential performance evaluation rank, measured seismic intensity, and values ​​obtained by simulating the indoor environment and racks using Unity, etc., to evaluate the damage situation and cargo fall situation. JPEG0007917038000018.jpg17170

[0054] <Executing the learning process> Next, learning is performed (S102). As mentioned above, reinforcement learning is performed in this embodiment. Reinforcement learning is a framework for an agent placed in a certain environment to acquire optimal behavioral rules through interaction with the environment. Here, the simulation of a point mass model with a learning wave as input is repeated multiple times, and learning is performed according to the following. JPEG0007917038000019.jpg38170

[0055] [Q Learning] Q-learning is a representative reinforcement learning method that uses an action-value function (Q-function) to acquire an optimal policy. The Q-function is a table that defines the expected value of the cumulative reward to be obtained in the future if a certain action is continuously selected in a given state, for all combinations of states and actions. By repeatedly updating the Q-function according to the following equation, the optimal policy that maximizes the cumulative reward is acquired. JPEG0007917038000020.jpg7170 Here, α is the learning rate (0 < α < 1), a coefficient that adjusts the update speed of the Q function, and γ is the discount rate (0 < γ < 1), a coefficient that determines how much to discount future Q values ​​(rewards) when considering them.

[0056] JPEG0007917038000021.jpg16170

[0057] [DQN] JPEG0007917038000022.jpg82170

[0058] <Confirmation of control laws> JPEG0007917038000023.jpg60170

[0059] Then, the system determines whether the result is good or bad (S104). If the result is bad (NO in S104), the system returns to step S101 and sets the parameters again.

[0060] If the results are favorable (YES in S104), implementation is performed (S105). That is, based on st acquired and calculated in real time from each sensor (accelerometer, displacement sensor, etc.) of the sensor group 60, a control force u(t) calculated using equation (19) or equation (20) is applied to the actuator 40.

[0061] According to this embodiment (AI control), a higher vibration damping effect was obtained compared to the comparative example (optimal control) in the simulation of step S103 and the verification in the implementation (experiment) of step S105. In particular, a high vibration damping effect was obtained when DQN was used as the reinforcement learning algorithm.

[0062] The image should be something like JPEG0007917038000024.jpg38170 (let's call this Condition 3). When comparing this Condition 3 with the aforementioned Conditions 1 and 2, it was confirmed that vibrations can be appropriately damped in the order of Condition 2 (mechanical energy + damping term), Condition 1 (mechanical energy), and Condition 3. In this way, by determining the reward in reinforcement learning based on the mechanical energy of Bridge 10, an excellent control law can be obtained, thereby improving vibration damping performance.

[0063] As explained above, in this embodiment, the control law for AMD control (the control law for controlling the excitation force of the actuator 40) is acquired through AI-based reinforcement learning. This makes it possible to maximize the capabilities of AMD and improve vibration damping performance.

[0064] Furthermore, in AI control, various constraints such as maximum control force and clearance can be defined in the form of states and rewards, and nonlinear control using neural networks is possible, thus enabling the acquisition of a control law that maximizes the control effect within various constraints.

[0065] Furthermore, by defining the settings for states and rewards, as well as various hyperparameters, trial and error for each property can be minimized. As a result, a stable control law can be obtained, and the time and effort required for design are reduced.

[0066] Furthermore, because the optimal control law is sequentially updated through repeated simulations (without the need to obtain a theoretical solution), it is possible to design a control law with a high degree of freedom. For example, considering a competitive reward system, it is possible to further enhance the control effect by using a superior control as an opponent, or to customize the control effect while taking into account the constraints of the device. It is also possible to optimize the control for free vibration by making the disturbance an impulse, or to optimize the control for disturbances that are likely to occur in structures, for example. As a result, it is possible to obtain a control law with better vibration damping effect than the comparative example (optimal control), or a control law that is more specialized for the specific environment.

[0067] ===Second Embodiment=== Figure 5A shows the vibration damping system configuration of the structure according to the second embodiment. Figure 5B is a model of Figure 5A.

[0068] As shown in Figure 5A, the structure to be vibration-damped in the second embodiment is a building 100, and a mass 200 is arranged on the top of the building 100 so as to be able to roll horizontally (rolling top AMD). Also, similar to the first embodiment, sensors (not shown) for detecting the state of the building 100 and the mass 200, and a controller (not shown) for controlling the actuator 400 based on the output of the sensors are provided (the same applies to the third and fourth embodiments).

[0069] When Figure 5A is modeled as a point mass model, it becomes the two-point mass model shown in Figure 5B. In Figure 5B, building 100 and mass 200 are each represented as point masses. Of these, the point mass representing building 100 corresponds to the first point mass, and the point mass representing mass 200 corresponds to the second point mass.

[0070] In the point mass model in Figure 5B, the building 100 is aligned vertically (corresponding to a predetermined direction) with the support end 102 (in this case, the ground). Also, the spring element 104 in Figure 5B corresponds to the stiffness of the building 100. The spring element 104 connects the support end 102 and the building 100 (first point mass), and generates an elastic reaction force acting along a direction perpendicular to the direction in which the building 100 and the support end 102 are aligned (in this case, the vertical direction) (in this case, the horizontal direction).

[0071] Mass 200 is a weight (mass body) for AMD control and is located at the top of building 100. A rolling support 500 (roller, etc.) is also provided above building 100 (between building 100 and mass 200) to support mass 200 so that it can roll horizontally. This allows mass 200 to be displaced horizontally relative to building 100.

[0072] The actuator 400 (vibrator) is connected to the building 100 and vibrates the mass 200 in the horizontal direction. By adjusting (controlling) the excitation force (control force) of this actuator 400, the lateral swaying (horizontal vibration) of the building 100 can be suppressed.

[0073] In this second embodiment as well, similar to the first embodiment, vibration damping performance can be improved by controlling the actuator 400 with an AI-based control law.

[0074] While this example uses a rolling bearing AMD, it is not limited to this type; similar vibration damping can be achieved with a sliding bearing AMD as well.

[0075] ===Third Embodiment=== Figure 6A shows the vibration damping system configuration of the structure according to the third embodiment. Figure 6B is a model of Figure 6A. Note that parts identical to those in the second embodiment (Figures 5A and 5B) are denoted by the same reference numerals and their descriptions are omitted.

[0076] In the third embodiment, a laminated rubber 301 is provided between the building 100 and the mass 200. The laminated rubber 301 supports the mass 200 so that it can be displaced in the horizontal direction. The laminated rubber 301 also restores the mass 200 to its original position when it is displaced in the horizontal direction. As a result, as shown in Figure 6B, a spring element 300, which is the horizontal rigidity of the laminated rubber 301, is provided in parallel with the actuator 400 between the building 100 and the mass 200. Since Figure 6B is the same point mass model as Figure 1C of the first embodiment (although the direction is different), in this case as well, vibration damping performance can be improved by controlling the actuator 400 with an AI-based control law.

[0077] <Variation> Figure 7 shows a modified example of the third embodiment. In Figure 7, a mass 200 is suspended from the top of the building 100 by a pendulum 302. This allows the mass 200 to be displaced horizontally relative to the building 100. The horizontal rigidity of the pendulum 302 corresponds to the spring element 300.

[0078] Figure 8 shows another modified example of the third embodiment. In this modified example, a base 303 is provided on top of the building 100. The upper surface of the base 303 is configured in a mortar shape with a gentle slope. A mass 200, which is supported so as to be able to roll by a rolling bearing 500, and an actuator 400 are provided on the base 303. This structure generates a restoring force (rigidity: spring element 300) that returns the mass 200 to the center when it is displaced horizontally.

[0079] In the case of these modified examples (Figures 7 and 8), the model is the same as that of Figure 6B. Therefore, by controlling the actuator 400 with an AI-based control law, the vibration damping performance can be improved.

[0080] ===Fourth Embodiment=== Figure 9A shows the vibration damping system configuration of the structure according to the fourth embodiment. Figure 9B is a model of Figure 9A. Note that parts identical to those in the third embodiment (Figures 6A and 6B) are denoted by the same reference numerals and their descriptions are omitted.

[0081] In the fourth embodiment, a mass 150 is provided between the building 100 and the mass 200. The mass 150 is provided on the top of the building 100 via laminated rubber 161 and functions as a TMD (Tuned Mass Damper) that synchronizes with the shaking of the building 100 and suppresses the vibration of the building 100.

[0082] Furthermore, a mass 200 supported by a rolling bearing 500 and an actuator 400 are provided on top of the mass 150. The mass 200 is supported by the rolling bearing 500 so as to be able to roll (horizontally displace) horizontally relative to the mass 150, and the actuator 400 is connected to the mass 150 and excites the mass 200 in the horizontal direction (AMD).

[0083] When Figure 9A is modeled as a point mass model, it becomes the three-point mass model shown in Figure 9B. In Figure 5B, the building 100, mass 150, and mass 200 are each represented as point masses. Horizontal rigidity (spring element 160) of laminated rubber 161 is provided between mass 150 and building 100. Even in this case, since building 100 and mass 150 move in sync, they can be considered as a single point mass (corresponding to the first point mass). Also, spring element 104 and spring element 160 can be considered as a single spring element (corresponding to the first spring element). Therefore, as with the previously described embodiment, in this fourth embodiment as well, vibration damping performance can be improved by controlling the actuator 400 with an AI-based control law.

[0084] <Variation> Figure 10 shows a modified example of the fourth embodiment. In this example, a mass 150 is suspended from the top of the building 100 by a pendulum 162 (spring element 160). Above the mass 150, a mass 200 is supported so as to be able to roll by a rolling bearing 500, and an actuator 400 is provided. The mass 200 is able to be displaced horizontally relative to the mass 150, and the actuator 400 excites the mass 200 in the horizontal direction.

[0085] Figure 11 shows another modified example of the fourth embodiment. In this example, a bowl-shaped base 163 (spring element 160) with a gentle slope on its upper surface is placed on top of the building 100, and a mass 150 is placed on top of it, which is supported so as to be able to roll by a rolling bearing 600. Furthermore, a mass 200 supported by a rolling bearing 500 and an actuator 400 are placed on top of the mass 150. The mass 200 is able to be displaced horizontally relative to the mass 150, and the actuator 400 vibrates the mass 200 in the horizontal direction.

[0086] In the case of these modified examples (Figures 10 and 11), the model is the same as that in Figure 9B. Therefore, by controlling the actuator 400 with an AI-based control law, the vibration damping performance can be improved.

[0087] ===Other Examples=== The embodiments of the present invention have been described above, but these embodiments are intended to facilitate understanding of the present invention and are not intended to limit its interpretation. Furthermore, the present invention can be modified or improved without departing from its spirit, and it goes without saying that the present invention includes equivalents thereof. For example, the following modifications are possible.

[0088] In the embodiments described above, Q-learning and DQN (Deep Q-Learning) were used as reinforcement learning algorithms, but the invention is not limited to these, and other reinforcement learning algorithms (e.g., Sarsa, Monte Carlo method, etc.) may also be used.

[0089] In the embodiments described above, a bridge 10 and a building 100 were described as the structure (first point mass) for ease of understanding, but it goes without saying that they may also be composed of columns, beams, walls, floors, etc. [Explanation of Symbols]

[0090] 10 bridge, 12 support end, 14 spring element (first spring element), 20 squares, 30 springs (second spring element), 40 actuators, 60 sensor groups, 70 controllers, 100 Building, 102 Support end, 104 Spring element (first spring element), 150 squares, 160 spring elements, 161 Laminated rubber, 162 Pendulum, 163 Unit, 200 squares, 300 Spring element (second spring element), 301 Laminated rubber, 302 pendulums, 303 units, 400 actuators, 500 rolling bearing, 600 rolling bearing,

Claims

1. Support end and The support end and the first point mass aligned in a predetermined direction, A first spring element connecting the support end and the first point of mass, the first spring element generating an elastic reaction force acting along an orthogonal direction perpendicular to the predetermined direction, A second mass point that is relatively displaceable in the direction perpendicular to the first mass point, A vibrator connected to the first mass point and which excites the second mass point in the orthogonal direction, A vibration control system for a structure represented by a point mass model having, A control law for controlling the excitation force of the aforementioned vibrator is determined using reinforcement learning. The second mass point is used as the mass of a mass damper that suppresses the vibration of the first mass point. The aforementioned reinforcement learning is a framework for an agent placed in a certain environment to acquire optimal behavioral rules through interaction with the environment. It involves repeatedly simulating the point mass model, consisting of the first and second point masses, with a learned wave as input, under the same configuration conditions, thereby acquiring a control law with the excitation force of the vibrator as the output. A vibration control system for structures characterized by the following features.

2. A vibration damping system for a structure according to claim 1, The learning target of the aforementioned reinforcement learning is semi-active. A vibration control system for structures characterized by the following features.

Images