Vibration control system and vibration control method

Abstract

To improve habitability against wind.SOLUTION: A vibration control system for suppressing vibration in a structure, comprises: a semi-active mass damper arranged in the structure; and a control unit that controls the semi-active mass damper when an external force acts on the structure. The control unit controls the semi-active mass damper using a control law that reinforcement-learns the damping force to improve habitability against wind.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a vibration control system and a vibration control method for a structure. [Background technology]

[0002] Known methods for using a mass to reduce vibrations of a structure caused by various disturbances include tuned mass dampers (TMDs), which tune the mass to the vibration, and active mass dampers (AMDs; see Patent Document 1, for example), which actively move the mass. TMDs have a simple structure and do not use electricity, making them inexpensive and highly reliable, but they have inferior vibration-damping capabilities compared to AMDs. AMDs have high vibration-damping capabilities, but they tend to have a complex structure because they require control to move the mass, and they are expensive and generally consume a lot of energy.

[0003] Semi-active mass dampers (SAMDs) are also known, which allow for the adjustment of the stiffness of the springs connecting the structure and the mass, and the damping force of the dampers. SAMDs are intermediate in performance and cost between TMDs and AMDs, and are attracting attention as they combine vibration control capabilities with reliability. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Publication No. 1-275867 Summary of the Invention [Problem to be solved by the invention]

[0005] Generally, SAMD is controlled based on control laws (control theory) based on control engineering, and each parameter of the control theory must be adjusted by experienced engineers through trial and error. This makes it difficult to achieve optimal vibration control performance to counter wind-induced building sway (wind sway), and there is a risk that livability will not be improved.

[0006] The present invention has been made in view of the above-mentioned problems, and an object of the present invention is to improve habitability against wind. [Means for solving the problem]

[0007] The main invention for achieving the above object is a vibration control system for suppressing vibrations in a structure, comprising a semi-active mass damper placed in the structure and a control unit that controls the semi-active mass damper when an external force acts on the structure, wherein the control unit controls the semi-active mass damper using a control law that has been reinforced learned to determine a damping force for improving livability against wind.

[0008] Other features of the present invention will become apparent from the description of this specification and the accompanying drawings. [Effects of the Invention]

[0009] According to the present invention, it is possible to improve habitability against wind. [Brief explanation of the drawings]

[0010] [Figure 1] 1 is a diagram illustrating a configuration of a vibration control system according to an embodiment of the present invention. [Figure 2] This is a model of Figure 1. [Figure 3] FIG. 1 is a flow diagram illustrating general SAMD control. [Figure 4] FIG. 2 is a flow diagram illustrating SAMD control according to the present embodiment. [Figure 5] This is a diagram of the model in Figure 2 simplified to two mass points. [Figure 6] FIG. 10 is a diagram showing a learning wave of a comparative example. [Figure 7] FIG. 10 is a diagram showing a learning wave of an embodiment. [Figure 8] This is a diagram showing the wind force (actual wind force) observed in a wind tunnel experiment. [Figure 9] 10A and 10B are diagrams showing the results of numerical analysis of a comparative example and an example. [Figure 10] FIG. 10 is an explanatory diagram of a first modified example. [Figure 11] FIG. 10 is an explanatory diagram of a second modified example. [Figure 12] FIG. 10 is an explanatory diagram of a third modified example. [Figure 13] FIG. 10 is an explanatory diagram of a fourth modified example. [Figure 14] FIG. 10 is an explanatory diagram of a fifth modified example. DETAILED DESCRIPTION OF THE INVENTION

[0011] At least the following matters will become clear from the description of this specification and the accompanying drawings.

[0012] (Aspect 1) A vibration control system for suppressing vibrations of a structure, comprising: a semi-active mass damper disposed in the structure; and a control unit that controls the semi-active mass damper when an external force acts on the structure, wherein the control unit controls the semi-active mass damper using a control law that has been subjected to reinforcement learning of a damping force for improving habitability against wind.

[0013] According to the vibration control system of the first aspect, it is possible to obtain an optimum control law for wind-induced vibration of a structure through reinforcement learning, thereby improving habitability against wind.

[0014] (Aspect 2) In the vibration control system according to aspect 1, it is preferable that the learning wave of the control law is a wave that simulates the wind.

[0015] According to the vibration control system of aspect 2, a control law that is effective against wind can be obtained by reinforcement learning, thereby improving livability.

[0016] (Aspect 3) In the vibration control system according to the first or second aspect, it is desirable that the excitation force on the structure due to the learning wave of the control law fluctuates only within either a positive or negative range.

[0017] According to the vibration damping system of aspect 3, it is possible to improve the vibration damping performance against wind.

[0018] (Aspect 4) In the vibration control system according to aspect 1 or 2, it is preferable that the control law is learned using a composite wave obtained by combining a sine wave with a DC component wave.

[0019] According to the vibration damping system of the fourth aspect, it is possible to improve the vibration damping performance against wind.

[0020] (Aspect 5) In the vibration damping system according to aspect 4, the DC component wave may include a step-like component wave.

[0021] According to the vibration damping system of the fifth aspect, it is possible to improve the vibration damping performance against wind.

[0022] (Aspect 6) In the vibration control system according to any one of the first to fifth aspects, it is preferable that the structure is a building, and the semi-active mass damper is disposed at the top of the building.

[0023] According to the vibration control system of the sixth aspect, horizontal shaking (horizontal vibration) of a building can be efficiently suppressed.

[0024] (Aspect 7) A vibration control method for controlling vibrations of a structure equipped with a semi-active mass damper, characterized in that the semi-active mass damper is controlled using a control law that has undergone reinforcement learning to determine the damping force for improving habitability against wind.

[0025] According to the vibration damping method of the seventh aspect, it is possible to improve the habitability against wind.

[0026] === Implementation form === <<About vibration control systems>> FIG. 1 is a diagram showing the configuration of a vibration control system according to this embodiment.

[0027] The vibration control system of this embodiment shown in FIG. 1 is a system for suppressing horizontal shaking (horizontal vibration) of a building 10, and includes a semi-active mass damper (SAMD) 20, a sensor 60, and a controller .

[0028] The building 10 is an n-story (n-layer) structure to be subjected to vibration control, and the lower end of the building 10 is fixed to the ground. In addition, a support 12 for fixing the SAMD 20 is provided on the top (rooftop) of the building 10.

[0029] The SAMD 20 is provided at the top of the building 10. The SAMD 20 of this embodiment includes a weight 30, a spring 40, and a semi-active damper (variable damper) 50.

[0030] The weight 30 is a mass for controlling vibration. The weight 30 in this embodiment has a rolling bearing and is disposed at the top of the building 10 so as to be displaceable in the horizontal direction relative to the building 10. By utilizing the movement of this weight 30 as a vibration control force, the shaking of the building 10 can be suppressed.

[0031] The spring 40 and the semi-active damper 50 connect the weight 30 and the building 10 (here, the support 12 of the building 10). That is, one end of each of the spring 40 and the semi-active damper 50 is connected to the weight 30, and the other end of each is connected to the building 10 (support 12).

[0032] When weight 30 is displaced in the horizontal direction, spring 40 generates an elastic force (elastic reaction force) in the direction opposite to the displacement, restoring weight 30 to its original position.

[0033] The semi-active damper 50 is a damper that can vary the damping force, and generates a damping force according to the relative displacement between the weight 30 and the building 10. The semi-active damper 50 generates a damping force only in the direction opposite to the relative velocity of the weight 30 with respect to the top floor of the building 10 (i.e., it does not generate a damping force in the direction of the relative velocity).

[0034] The method for changing the damping force of the damper is not particularly limited, but examples include adjusting the diameter of an orifice inside the damper and using a liquid (magnetic fluid) whose resistance changes depending on the application of voltage. The stiffness of the spring 40 may also be made variable. One method for changing the stiffness of the spring 40 is to use an air spring and externally control the air pressure.

[0035] The sensors 60 include, for example, an acceleration sensor that measures (detects) the acceleration of an object, a displacement sensor that detects displacement, etc. The sensors 60 are provided on the top of the weight 30 and the building 10. The sensors 60 can detect the positions (relative positions), speeds, and accelerations of the building 10 and the weight 30.

[0036] When an external force (for example, wind load or earthquake force) acts on the building 10, the controller 70 (corresponding to a control unit) controls the SAMD 20 (specifically, the damping force of the semi-active damper 50) based on the detection result of the sensor 60. Note that the controller 70 of this embodiment controls the damping force of the semi-active damper 50 so as to reduce horizontal vibration of the building 10 caused by wind (so-called wind sway). For this reason, in this embodiment, as will be described later, reinforcement learning is performed on the damping force to improve habitability against wind. More specifically, reinforcement learning is performed using a learning wave that simulates wind. The controller 70 then controls the SAMD 20 using the control law obtained by the reinforcement learning (details will be described later).

[0037] 1, the controller 70 is provided at the top (rooftop) of the building 10, but is not limited thereto. For example, the controller 70 may be provided at another location, such as the top floor (nth floor) of the building 10. The controller 70 may then control the SAMD 20 at the top of the building 10 from that location based on the detection results of the sensor 60.

[0038] <About the point mass model> FIG. 2 is a diagram that models FIG. 1. When FIG. 1 is modeled as a mass point model, it becomes a model of a system of n+1 mass points (n floors + mass) shown in FIG. 2. In FIG. 2, each floor of the building 10 and the weight 30 (mass) are shown as mass points. In FIG. 2, the spring components (k1 to k n ) indicates the stiffness of each floor of the building 10 (for example, the horizontal stiffness of the columns). n ) indicates the damping coefficient of each floor of the building 10.

[0039] The equation of motion for the model in Figure 2 can be expressed by the following equation (1): As mentioned above, the control damping force u(t) by the semi-active damper 50 acts only in the direction opposite to the relative velocity of the weight 30 (mass) with respect to the top floor of the building 10 (the damping force cannot act in the direction of the relative velocity). JPEG2025137008000002.jpg11161 where: JPEG2025137008000003.jpg22161 JPEG2025137008000004.jpg29150 JPEG2025137008000005.jpg31150 JPEG2025137008000006.jpg38150 x j (t) (j=1, 2, ... n): Displacement of the jth floor of the building (m) x mass (t): Displacement of mass (m) JPEG2025137008000007.jpg6150 JPEG2025137008000008.jpg11161 m j : Mass of the jth floor of the building (kg) m mass : Mass of the mass (kg) k j : Stiffness of the jth floor of the building (N / m) k mass : Mass stiffness (N / m) c j : damping coefficient of the jth floor of the building u(t): Control damping force (N) (where u(t) ≥ 0) f j (t): Wind load acting on the jth floor of the building (N)

[0040] By solving the equation of motion (1) through numerical analysis, the behavior of the n+1 mass point diameter can be simulated.

[0041] <<General SAMD Control>> Figure 3 is a flow diagram explaining general SAMD control. Here, we will explain the case where optimal control is used as an example of control, but the control law is not limited to optimal control, and the designer can select from skyhook control, H∞ control, etc. as appropriate.

[0042] First, each parameter is set (S11).

[0043] The objective of optimal control is to minimize the evaluation function J, which consists of multiple state quantities. Here, the evaluation function J is defined as shown below, with the objectives of quickly converging the [displacement, velocity] and control force of the [mass, top floor] to zero as well as minimizing the absolute values ​​of the [displacement, velocity] and control force of the [mass, top floor] at each step as much as possible. In other words, the evaluation function J is defined as the weighted sum of the squares of the [displacement, velocity] and control force of the [mass, top floor]. JPEG2025137008000009.jpg15150In addition, JPEG2025137008000010.jpg11150q A(t): Mass displacement (m) q B (t): Displacement of the top floor of the building (m) JPEG2025137008000011.jpg10150 JPEG2025137008000012.jpg12150u(t):Control force JPEG2025137008000013.jpg21150ε: Weight applied to control force

[0044] Here, the weights of q1 to q4 and ε are, for example, the displacement q B If you want to make (t) smaller, you can increase the coefficient q4 that applies to it, and adjust it empirically and by trial and error within the overall balance.

[0045] Using optimal control feedback theory, when ω satisfies the following optimal tuning condition equation, q1=q2=q3=0 and only the weights of q4 and ε need to be adjusted. JPEG2025137008000014.jpg17150

[0046] Next, the feedback gain of the control law is calculated (S12). 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 (4). JPEG2025137008000015.jpg16150

[0047] Here, P is the solution of the Riccati equation expressed by the following equation (5). JPEG2025137008000016.jpg17150

[0048] The control force u(t) can be rewritten as a linear combination of the [displacement, velocity] of the [mass, top floor of the building] using gains G1 to G4 as follows: JPEG2025137008000017.jpg11150 where: JPEG2025137008000018.jpg11150 JPEG2025137008000019.jpg11150 The control law can be obtained by solving the formula (6).

[0049] Next, the effectiveness of the control law obtained from equation (6) is confirmed by a simulation in which the actual wind force, etc. is input into equation (6) (S13).

[0050] If the control law check result is unsatisfactory (NO in S14), the process returns to step S11 and the parameters are reset.

[0051] On the other hand, if the confirmation result of the control law is good (YES in S14), implementation is performed (S15). That is, the controller 70 applies the control damping force u(t) calculated by equation (6) to the semi-active damper 50 based on x(t) acquired and calculated in real time from the sensors 60 installed at various locations (here, the weight 30 and the building 10).

[0052] However, even if the control law obtained as described above is used, there is a risk that the SAMD20's capabilities may not be fully utilized. For example, the optimal control described above assumes a control force expressed as a linear combination of state variables, which necessitates the design of a safety factor for various constraints, potentially resulting in an inefficient control law. Furthermore, to satisfy the various constraints that vary from property to property, engineers must adjust parameters through trial and error using multiple simulations. This relies heavily on the engineer's experience, does not necessarily result in an optimal solution, and requires time and effort for the design. Furthermore, it is difficult to apply linear theory to structures with nonlinear restoring forces, such as concrete structures. Furthermore, because a theoretical solution must be found, the degree of freedom of the evaluation function is limited, making it difficult to optimize for specific disturbances (e.g., wind).

[0053] Therefore, in this embodiment, artificial intelligence (hereinafter referred to as AI) is used to control the SAMD 20 to improve vibration control performance. More specifically, the vibration control performance against wind is improved (improving livability). Note that machine learning using AI is classified into three types: supervised learning, unsupervised learning, and reinforcement learning. Of these, reinforcement learning is used in this embodiment.

[0054] <<<SAMD Control of This Embodiment>>> FIG. 4 is a flow diagram illustrating the SAMD control of this embodiment.

[0055] First, for simplification, as shown in Figure 5, the n mass points of building 10 are aggregated into one equivalent mass point, and a two-mass model is created in which SAMD 20 is installed at this aggregated mass point (S1 in Figure 4). Note that Figure 5 is a simplified diagram of the model in Figure 2, simplified to two mass points (one mass point + mass). In building 10 in Figure 5, m represents the building mass, c represents the building damping coefficient, and k represents the building stiffness.

[0056] Using this model, we perform AI reinforcement learning, as described below. Reinforcement learning is a method of learning behaviors that maximize value through trial and error, even without training data. Examples of such reinforcement learning algorithms include Actor-Critic, Q-Learning, and DQN (Deep Q-Learning).

[0057] <<Parameter settings>> Next, each parameter (state s t , reward r t , action a t ) is set (S2 in Figure 4).

[0058] [Action a t 〕 action a tis the control force output to the semi-active damper 50 at time t. Here, the maximum damping coefficient (c(t)) or maximum damping force (u(t)) is set for the semi-active damper 50. For example, if the reinforcement learning algorithm is Actor-Critic and the maximum damping coefficient is 0.068 kNs / m, the settings are as follows: JPEG2025137008000020.jpg10150 In the case of Q-Learning, the data is discretized into five divisions, etc.

[0059] [State s t 〕 Status t is the state quantity of the building 10 and the weight 30 (mass) at time t, and what is the input to act a t For example, set the displacement and velocity of the top floor and mass of the building as follows: JPEG2025137008000021.jpg11150 In the case of Q-Learning, each element is further discretized into three divisions, etc.

[0060] [Reward r t 〕 Future rewards t In this embodiment, a reward was given according to a comparison with the optimal control result, with the goal of exceeding the response of the top floor of the building during the aforementioned optimal control. JPEG2025137008000022.jpg11150 where: E(t): Energy of the top floor of the building (energy = kinetic energy + potential energy of the spring) E T (t): Energy consumption at the top of the building under optimal control

[0061] <<Running the training>> Next, the simulation of the mass point model with the learning wave as the input is repeated multiple times, and learning (reinforcement learning) is performed according to the following (S3 in FIG. 4). In this embodiment, for the purpose of reducing horizontal vibration (improving habitability) caused by wind excitation, a wave simulating wind is used as the learning wave. Specifically, a composite wave of a sine wave and a DC component wave is used (see the examples described later).

[0062] Here, for a certain state s t when a certain action a t is continuously selected, the total expected value G t of the reward r t obtained is defined as Q(s t , a t ).

[0063] <In the case of Q-Learning> According to the following formula (10), the update of Q(s t , a t ) is repeated for each step. JPEG2025137008000023.jpg16150 Note that α: learning rate (0 < α < 1) Q: coefficient for adjusting the update speed of the function γ: discount rate (0 < γ < 1)

[0064] <In the case of Actor-Critic> Similar to Q-Learning, the update of Q(s t , a t ) is repeated, but the policy π(a t ) for selecting a certain action a t in a certain state s t | s t ) is also represented by a neural network and is updated to maximize Q(s t , a t ).

[0065] <In the case of DQN> The parameters θ of the neural network are sequentially updated so as to minimize the loss function L(θ t ) defined by the following formula (11). JPEG2025137008000024.jpg17150

[0066] <<Verification of Control Law>> Next, by performing a simulation in which actual wind waves, etc. are input into the model (n-th order + SAMD20) of FIG. 2, the effect of the obtained control law is verified (S4 in FIG. 4). As shown in the following equation, G t The action a at which t G becomes maximum is selected as the damping coefficient (or damping force) u(t).

[0067] <In the case of Q-Learning and Actor-Critic> JPEG2025137008000025.jpg15150

[0068] <In the case of DQN> JPEG2025137008000026.jpg11150

[0069] Here, the difference between the damping coefficient and the damping force will be explained.

[0070] (When controlling the damping coefficient c(t)) The damping force u(t) generated by the semi-active damper 50 is obtained by the product of the velocity of the piston of the damper and the damping coefficient c(t). That is, the damping force u(t) can be expressed by the following equation (14) (where c(t) ≥ 0). JPEG2025137008000027.jpg11150

[0071] Substituting Equation (14) into the vector u of Equation (1), the vector u can be expressed by the following equation (15). JPEG2025137008000028.jpg27150

[0072] (When controlling the damping force u´(t)) Assuming that the damping force u'(t) generated by the semi-active damper 50 can be adjusted to any magnitude regardless of the speed of the damper piston, the damping force u(t) can be expressed by the following equation (16). u(t)=u´(t) (16)

[0073] The maximum acceleration on the top floor (highest floor) of the building was used as an index for the analysis results.

[0074] If the result of the check is not good (NO in S5), the process returns to step S1 and the parameters are set again.

[0075] <<Implementation>> On the other hand, if the result in step S5 is good (YES in S5), implementation is performed (S6). That is, the controller 70 receives and calculates the state s t Based on this, the damping coefficient (or damping force) u(t) calculated in step S04 is applied to the semi-active damper 50.

[0076] In this way, by controlling the SAMD20 with AI-based control laws, it is possible to achieve control laws that are not based on control theory. Furthermore, nonlinear control is possible using neural networks, making it possible to obtain control laws that are specialized for specific environments. Furthermore, because the AI ​​obtains the optimal control laws, the effort required for tuning by engineers for each property is reduced.

[0077] In this embodiment, a learning wave simulating wind (see Examples) is used as a learning wave used in reinforcement learning to obtain a control law, which makes it possible to suppress swaying due to wind (wind-induced swaying) (enhancing wind-induced vibration control performance) and improve livability.

[0078] ===Example=== In this example, a model (28 mass point model) in which a SAMD 20 is installed at the top of a 27-story building 10 was used to evaluate the case in which a wind load acts on the building 10. The conditions for the 2 mass point model shown in Figure 5 are as follows: Building first natural period: 4.02(s) Building primary effective mass (m): 12.2 (kton) Building damping coefficient (c): 1.0 (%) Mass weight (m mass ) :100(ton) Mass ratio (mass / building): 0.82 (%)

[0079] <Learning wave of comparative example> 6 is a diagram showing the learning wave of the comparative example. The horizontal axis of the diagram represents elapsed time (s), and the vertical axis represents the excitation force (kN). The learning wave of the comparative example (FIG. 6) can be expressed by the following equation (17). f(t)=Asinω1t (17) This is a sine wave with the first natural frequency ω1 of the 28 mass point model consisting of Building 10 (27 floors) and SAMD20.

[0080] <Learning wave in this example> 7 is a diagram showing the learning wave of the example. As in the comparative example, the horizontal axis of the figure represents elapsed time (s) and the vertical axis represents excitation force (kN). The learning wave of the example simulates wind and can be expressed by the following equation (18). f(t)=Asinω1t+B (18)

[0081] This learning wave is a composite wave obtained by adding (combining) a DC component wave (B) to equation (17) of the learning wave (sine wave) of the comparative example. The sine wave has a half amplitude of 20 kN and a period equal to the natural period of the building. By adding the DC component wave, the excitation force is biased toward the positive side, as shown in FIG. 7. Here, it is only positive and fluctuates within a positive range only. Depending on the value (sign) of B in equation (18), the excitation force may be biased toward the negative side (for example, it may fluctuate within a negative range only).

[0082] <Evaluation> In the cases where the learning waves of the comparative example (Fig. 6) and the working example (Fig. 7) were used, AI-based reinforcement learning was performed to determine the control law for each. The Actor-Critic algorithm was used for the reinforcement learning. Then, a numerical analysis was performed when actual wind force was applied to the model of Fig. 2 and control was performed using the control law obtained by reinforcement learning, and the maximum acceleration of the top floor of the building was evaluated (simulated). Fig. 8 shows the wind force (actual wind force) observed in a wind tunnel experiment. As can be seen from the figure, the actual wind force has a waveform similar to that of the learning wave of the working example (Fig. 7).

[0083] <Evaluation results> FIG. 9 shows the results of the numerical analysis of the comparative example and the working example. The horizontal axis of the figure shows the maximum acceleration (mm / s 2 ) and the vertical axis indicates maximum mass displacement (mm). Note that maximum mass displacement refers to the maximum relative displacement between the top of building 10 and weight 30. In the figure, the further to the left the plot is (i.e., the smaller the maximum acceleration value), the more vibration has been reduced (the greater the vibration control effect).

[0084] In Figure 9, the black circles (●) indicate the analysis results of the comparative example (when the learning wave in Figure 6 was used), and the black squares (■) indicate the analysis results of the example (when the learning wave in Figure 7 was used). The figure also shows the results when a TMD was used under the same conditions (i.e., when the damping force was not controlled). The numbers in each plot for the TMD are the values ​​of the damping coefficient.

[0085] 9, it can be seen that the maximum acceleration value is smaller in the example than in the comparative example (vibration can be reduced). Furthermore, while the maximum acceleration in the comparative example is approximately the same as that of the TMD, the maximum acceleration in the example is smaller than that of the TMD.

[0086] This confirmed that by using learning waves that simulate wind for reinforcement learning, it is possible to improve livability against wind (improved vibration control performance) compared to the comparative example and TMD.

[0087] In this embodiment, a constant value (B) is added to the AC component (sine wave in equation (17)) as the DC component wave, but the DC component wave does not have to be constant. For example, a waveform that changes stepwise may be added to the sine wave as the DC component wave. Even in this case, an effective control law for wind can be obtained, and occupant comfort can be improved.

[0088] === Variations === Modified examples of the vibration control system will be described below. Note that in the following modified examples, illustrations and descriptions of the sensor 60, controller 70, etc. will be omitted. Also, parts with the same configuration will be given the same reference numerals, and descriptions thereof will be omitted.

[0089] <First Modification> 10 is an explanatory diagram of a first modified example. In the first modified example, two supports (support 12a and support 12b) are provided at the top of a building 10, spaced apart in the horizontal direction. A SAMD 21 is provided between support 12a and support 12b. The SAMD 21 includes a spring 41 in addition to the components of the SAMD 20 (weight 30, spring 40, and semi-active damper 50).

[0090] A spring 40 and a semi-active damper 50 are provided between the weight 30 and the support portion 12a, and a spring 41 is provided between the weight 30 and the support portion 12b.

[0091] In this first variant, too, wind-induced vibration can be suppressed and livability improved by controlling the damping force of the semi-active damper 50 using a control law that is reinforced learned using a learning wave that simulates wind.

[0092] <Second Modification> 11 is an explanatory diagram of a second modified example. In this second modified example, a SAMD 22 is provided at the top of a building 10. The SAMD 22 includes a weight 32, a laminated rubber bearing 42, and a semi-active damper 52.

[0093] The weight 32 is a mass similar to the weight 30. However, the weight 32 does not have a function of displacement relative to the building 10 (such as a rolling bearing).

[0094] The laminated rubber bearing 42 is provided between the top of the building 10 and the weight 32. The laminated rubber bearing 42 supports the weight 32 so that it can be displaced horizontally relative to the building 10. Furthermore, if the weight 32 is displaced horizontally, the laminated rubber bearing 42 returns it to its original position. As a result, a spring element (similar to the spring 40 in the above-described embodiment) that is the horizontal rigidity of the laminated rubber bearing 42 is provided between the building 10 and the weight 32.

[0095] In addition, one end of the semi-active damper 52 is connected to the underside of the weight 32 and the other end is connected to the top of the building 10.

[0096] In this second variant, too, wind-induced vibration can be suppressed and livability improved by controlling the damping force of the semi-active damper 52 using a control law that is reinforced learned using a learning wave that simulates wind.

[0097] <Third Modification> 12 is an explanatory diagram of a third modified example. In this third modified example, a SAMD 23 is provided at the top of a building 10. The SAMD 23 includes a weight 33, a laminated rubber bearing 43, and a semi-active damper 53.

[0098] The weight 33 is a mass body with a very large planar area, and in addition to its function as a weight, it also functions as a place to install a heliport or equipment, for example.

[0099] The laminated rubber bearings 43 are similar to the laminated rubber bearings 42 of the second modified example, and therefore a description thereof will be omitted. However, because the planar area of ​​the weight 33 is large, the number of laminated rubber bearings 43 provided is greater than the number of laminated rubber bearings 42 of the second modified example.

[0100] One end of the semi-active damper 53 is connected to the underside of the weight 33 and the other end is connected to the top of the building 10 .

[0101] In this third variant, too, wind-induced vibration can be suppressed and livability improved by controlling the damping force of the semi-active damper 53 using a control law that has been reinforced learned using a learning wave that simulates wind.

[0102] <Fourth Modification> 13 is an explanatory diagram of the fourth modified example. In the fourth modified example, supports 12a, 12b and a SAMD 21 (weights 30, springs 40, 41, and a semi-active damper 50) are provided on the top floor of a building 10. These components are similar to those in the first modified example, and therefore their description will be omitted. In this way, the location of the semi-active mass damper is not limited to the top (rooftop) of the building, but may also be, for example, the top floor of the building.

[0103] <Fifth Modification> 14 is an explanatory diagram of the fifth modified example. In the fifth modified example, support parts 12a and 12b and a SAMD 21 (weight 30, springs 40 and 41, semi-active damper 50) are provided under the ceiling of the top floor of a building 10.

[0104] The above-described embodiments are intended to facilitate understanding of the present invention and are not intended to limit the present invention. The present invention may be modified or improved without departing from the spirit thereof, and the present invention naturally includes equivalents thereof.

[0105] In the above embodiment, the case where wind acts as an external force has been described, but the present invention is not limited to this. For example, the present invention can also be applied to a weak earthquake. [Explanation of symbols]

[0106] 10 Buildings (structures) 12,12a,12b Support part 20~23 Semi-active mass damper (SAMD) 30, 32, 33 Weights 40,41 Spring 42,43 Laminated rubber bearing 50, 52, 53 Semi-active damper 60 sensors 70 Controller (control unit)

Claims

1. A vibration control system for suppressing vibration of a structure, a semi-active mass damper disposed in the structure; a control unit that controls the semi-active mass damper when an external force acts on the structure; Equipped with the control unit controls the semi-active mass damper using a control rule that has undergone reinforcement learning to determine a damping force for improving habitability against wind. A vibration control system characterized by:

2. 2. The vibration damping system of claim 1, The learning wave of the control law simulates the wind. A vibration control system characterized by:

3. 3. The vibration damping system according to claim 1 or 2, The excitation force on the structure due to the learning wave of the control law fluctuates only within either a positive or negative range. A vibration control system characterized by:

4. 3. The vibration damping system according to claim 1 or 2, The control law is learned using a composite wave obtained by combining a sine wave with a DC component wave. A vibration control system characterized by:

5. 5. The vibration damping system of claim 4, The DC component wave includes a step-like component wave. A vibration control system characterized by:

6. 3. The vibration damping system according to claim 1 or 2, the structure is a building, the semi-active mass damper is located at the top of the building; A vibration control system characterized by:

7. A vibration control method for a structure equipped with a semi-active mass damper, comprising: The semi-active mass damper is controlled using a control rule that has undergone reinforcement learning to determine the damping force for improving habitability against wind. A vibration damping method characterized by:

Images