論文

近年、日本では自然災害が頻発化・激甚化しているが、従来の専門家から住民への一方向的な防災教育では、住民の主体的な参加意識が十分に育まれず、災害情報が全ての市民に届かない。また、受け取っても避難行動に移せないことが課題である。本研究では、防災に「市民科学」と「アート」を組み合わせた新たな防災活動の枠組みを構築し、地域の災害レジリエンス向上を目指す。提案する防災活動の枠組みは「ワークショップの実施」「作品の展示」「作品の鑑賞」の3つのステップを循環させる構造を持つ。参加者はハザードマップを読み解き、災害リスクと防災行動をアートで表現し、作品を地理情報にリンクした防災行動マップ上で展示・公開する。4回のワークショップ検証を通じて、参加者の防災情報の主体性構築、理解促進、班員との協力関係構築が確認された。さらに、各セクションをWebアプリケーション化した支援ツールを開発し、UnityとArcGISを統合して円滑な運営を実現した。デジタル化により知識習得と作品制作の一括化を図り、参加者の創造性と防災意識の向上を促進した。一方で、高齢者の操作性に課題が見られ、多世代対応のインターフェース開発の必要性が示された。今後は、ユーザビリティの向上と地域コミュニティでの長期的検証が課題である。

The nonlinear energy sink (NES) is a vibration control device that utilizes the nonlinear resonance between an auxiliary mass and the primary structure. By exploiting nonlinear restoring forces, NES can maintain its vibration reduction performance even when the natural period of a building deviates from design value. Among various types of NES proposed, challenges remain in ensuring sufficient restoring force under small displacements and enabling control in two horizontal directions. We proposed a bidirectional NES with pretensioned springs (PB-NES) that addresses these challenges by employing pretensioned tensile springs to secure restoring force under small displacements, and by introducing a dual-carriage mechanism to achieve simultaneous bidirectional control. Numerical analyses and shaking table tests using scaled models confirm that the proposed PB-NES provides effective control in two horizontal directions and exhibits robustness against variations in the natural period of the structure.

繰り返し地震動が入力し木造建物が倒壊して人的被害が発生することを防ぐため，構造ヘルスモニタリングシステムによる迅速な損傷評価が期待されている．ニューラルネットワークは建物応答から倒壊の危険性を判別する有力な手法であるが，未学習の損傷パターンの判別を苦手としている．本研究では，ニューラルネットワークを用いた木造建物の損傷判別器の構築において，学習データを作成する際に対象建物モデルの応答解析の入力にパルス性模擬地震動を用いることで損傷パターンの多様性を考慮する手法を提案し，判別精度を比較することでその有効性を検証した．

建物の地震応答データによって描かれる復元力－層間変形角関係曲線（履歴ループ）は，建物の健全度合いによって形状が変化する．本研究では構造ヘルスモニタリングにこの性質を活用し，畳み込みニューラルネットワーク（CNN）を用いて履歴ループ画像をもとに将来の地震動による倒壊の危険性を判別する方法を提案した．提案手法において座標軸の範囲や解像度の設定を検討し，各ケースにおける正解率及び危険モデルの再現率を比較した．入力地震波の最大入力加速度の下限値を設定し，これを調節することで危険モデルの再現率を90%程度とすることができた．よってこの最大入力加速度の下限値はCNNによる損傷判別において影響力のあるパラメータの一つであるということができる．

Super-high-rise buildings commonly utilize tuned mass dampers (TMDs) for vibration control. However, buildings with different natural periods in two horizontal directions typically require multiple TMDs, increasing weight, and costs. In this study, a bidirectional TMD (BTMD) was developed to control vibrations simultaneously in both horizontal directions of high-rise buildings with different natural periods by independently tuning natural frequencies in both directions through rotational swing and orthogonal linear oscillation of the mass component. Numerical analyses and shaking table experiments on reduced-scale building models confirmed its effectiveness in suppressing structural responses of peak and root-mean-square acceleration and displacement. The proposed BTMD offers a cost-effective and lightweight solution for enhancing the seismic resilience of high-rise buildings with different dynamic properties in two directions.

Q–Δ resonance is the phenomenon in which the large displacement of a vertical member or structure with different stiffnesses in the two horizontal directions induces a torsional mode response owing to geometric nonlinearity and the related internal resonance. This phenomenon may have a substantial impact on super-high-rise buildings, whose horizontal displacement response can be remarkably large due to long-period ground motion. In this study, the responses of two non-eccentric super-high-rise buildings, which satisfied the Q–Δ resonance conditions, to long-period design ground motions were investigated using 3D frame models. One was a 320-m-tall, center-core, super-high-rise building model with outriggers, and the other was a 200-m-tall, center-core, super-high-rise building model without outriggers, in which the bending deformation became relatively large. Time-history response analysis was conducted considering geometric nonlinearity, and the results suggested that the torsional response induced by the Q–Δ resonance could significantly increase the horizontal acceleration response. The countermeasures against long-period ground motion in super-high-rise buildings should consider the torsional response even in non-eccentric buildings.

Structural health monitoring is an effective way to quickly and efficiently detect earthquake damage to a building. However, its applications to wooden buildings are still limited. One of the challenges in predicting the seismic response of a wooden structure is that the effect of variance in its material and structural properties is relatively significant. To handle this problem, we propose a method to update the damage classifier using Bayesian system identification. The damage classifier is constructed using neural network (NN) and machine learning techniques. The method utilizes the weighted loss function for the NN to reflect the updated probability distribution of the structural parameters. The NN is tuned using backpropagation, where the training data are consequently weighed by the weighed loss function, directly using the value of the posterior probability, according to the data obtained from the numerical model of a target structure with the corresponding parameters. Using the result of the Bayesian updating, the tuned NN damage classifier achieved a higher accuracy than before tuning by reflecting the posterior probability distribution of the structural parameters.

When large displacements occur in buildings with different stiffnesses in the two horizontal directions, torsional moments are generated owing to geometric nonlinearities. Consequently, when the sum or difference of the translational mode natural frequencies in the two horizontal directions is close to the torsional mode natural frequency, the torsional mode is excited. This is known as Q-? resonance. Therefore, we created a three-dimensional frame model of a center-core high-rise building with a biaxially symmetrical plane that satisfied the Q-? resonance condition and examined whether the torsional response was excited by long-period design waves. A time-history response analysis was conducted assuming elastic response and considering geometric nonlinearities. It was confirmed that the torsional mode could be induced.

Structural health monitoring (SHM), which evaluates building damage based on seismic response data, is useful for rapidly assessing damage due to earthquakes. If a reliable damage assessment system for houses can be developed, the loss of human lives caused by the aftershocks of major earthquakes can be reduced. When a target building is damaged by strong seismic ground motions, the shape of the restoring force–story drift angle curve (hysteresis loop) changes. Therefore, based on the hysteresis loop images of wooden buildings, this study proposes a method to classify the risk of collapse because of future seismic ground motions using a convolutional neural network (CNN). To develop training data, structural response analysis is conducted using “wallstat”, which is a numerical analysis software for wooden structures. SHM has more severe consequences when it misclassifies an unsafe condition as safe (miss) than when it misclassifies a safe one as unsafe (false alarm). Therefore, we propose the usage of a dataset that does not contain safe model data near the threshold value of major damage for training to decrease the possibility of misses, and we verify the effect of this. In creating the hysteresis loop images, we examine the range of the axes using two methods: fixing it at specific values and matching it to the range of the minimum and maximum values. The investigation results suggest that the trained CNN learns the changes in structural characteristics, such as stiffness reduction, damping increase, and residual displacement.

To evaluate Q–Δ resonance, which is a torsional resonance phenomenon caused by geometric nonlinearity, a simplified model for a multistory non-eccentric building with a shear-type structure was proposed. A four-story small-scale specimen was designed and fabricated, assuming a non-eccentric steel structure with a height of 324 m, and conducted shaking table tests of the specimen. Q–Δ resonance was confirmed as predicted by the simplified model. In addition, the simplified model reproduced the resonance amplitude calculated using a finite element model of the specimen with constrained out-of-plane slab rotation. The simplified model can be used to predict the Q–Δ resonance response when the bending deformation of an entire building is very small.

To reduce volcanic disasters, it is important for residents to practice mutual support under the leadership of voluntary disaster reduction groups. However, many communities lack the human resource for their future leaders. To address this issue, we proposed a digital tool for residents to develop leaders by themselves. The system requirements were determined to enable residents to conduct autonomous and effective tabletop exercise in both in-person and remote formats. The usefulness of the tool was confirmed through workshops with residents of the areas at the foot of Mt. Fuji.

This study proposes a new warm standby method for computer servers based on seismic response. The proposed method triggers a warm standby in the case that the exceedance probability of the safety limit of seismic response predicted by the last two-second observation record exceeds a preset threshold. Numerical simulations showed that the proposed method reduced the loss cost by 1.64% compared to conventional methods. To validate this method, we conducted shaking table tests. When a simulated earthquake motion was input, the loss cost was reduced by 60.6%. However, when observed records were input, the loss cost was increased by 3.89%. The results suggest that it is necessary improve the input ground motion model.

When the torsional moment owing to geometric nonlinearity synchronizes with the torsional mode response, an internal resonance phenomenon occurs in a building structure system. The author’s research group calls this phenomenon Q–Δ resonance. The probability distribution of the torsional response was evaluated using Monte Carlo simulations on a 60-story building model, the input wave being simulated using nonstationary colored-noise ground motion. The results showed that the torsional response greatly increased the horizontal acceleration response and the response amplification ratio between the center and corner of the top floor could exceed 1.2 when the model satisfied the Q–Δ resonance condition that the natural frequency of the torsional second mode matched the sum of the first natural frequencies in the x- and y-directions. It was also evident that closer proximity to the resonance condition increased the amplification ratio. These results suggest that Q–Δ resonance may need to be considered in seismic design.

Torsional vibration of buildings subjected to earthquake motion has been studied for many years, but the torsional vibration of high-rise buildings has not yet been clarified. Many previous studies have focused on eccentricity, the mismatch between the center of gravity and the rigid center, as the cause of torsional vibration. However, in buildings with no eccentricity, torsional vibrations could be observed. The authors' research group focused on the “Q-Delta effect,” a phenomenon in which torsional torque is generated due to geometric nonlinearity and “Q-Delta resonance,” a phenomenon in which resonance occurs when the period of torsional torque generation coincides with the natural period of the torsional mode. In order to analyze the characteristics of Q-Delta resonance, a small-scale specimen was designed and fabricated from an 81-story high-rise building with biaxially symmetrical plans, which was reduced to four stories. As a result, the torsional response of the experimental and simulation results generally agreed, confirming that the torsional response increases when the Q-Delta resonance condition is satisfied. In addition, a comparison of the torsional response with a specimen with a smaller tower ratio and less bending deformation was conducted, and it was confirmed that the peak of the torsional response was steeper in the present study case.

To evaluate Q–Δ resonance, which is a torsional resonance phenomenon caused by geometric nonlinearity, a simplified model for a multistory non-eccentric building with a shear-type structure was proposed. A four-story small-scale specimen was designed and fabricated, assuming a non-eccentric steel structure with a height of 324 m, and conducted shaking table tests of the specimen. Q–Δ resonance was confirmed as predicted by the simplified model. In addition, the simplified model reproduced the resonance amplitude calculated using a finite element model of the specimen with constrained out-of-plane slab rotation. The simplified model can be used to predict the Q–Δ resonance response when the bending deformation of an entire building is very small.

Torsional moments can occur in a column owing to the geometric nonlinearity when the top of the column undergoes a large displacement in two horizontal directions. Periodically generated torsional moments can excite the torsional mode of a building owing to the internal resonance. For high-rise buildings, there is a risk that long-period seismic motions can cause a non-negligible torsional response due to this phenomenon. To investigate this risk, this study focuses on non-eccentric single-story and two-story shear-type structural models with an aggregate column, which is assumed to be an elastic Timoshenko beam. We derived the equations of motion for the models under horizontal ground motions and confirmed that the equations were identical to those of the models with an aggregate column of an elastic Euler beam. We then performed time-history response analyses of the models satisfying the internal resonance conditions under the input ground motions of harmonic waves, Gaussian white noise acceleration, acceleration records of previous earthquakes, and a long-period design wave. The results showed that the maximum horizontal acceleration at the corner of the top floor slab could be increased by more than 60% of the acceleration at the center of the slab in the worst case.

In previous studies of torsional response, researchers considered that torsional vibration of building is caused by eccentric distribution of stiffness, damping, and mass of a structure, or spatially non-uniform ground motion input to a long or large base mat of a structure. However, we have discovered that the torque generated by horizontal displacement and perpendicular inertial force, which we call the Q–Delta effect, can cause torsional vibration. In order to investigate this effect, we considered a single finite-size mass-linear elastic shear and torsion spring model and derived the resonance condition with respect to a torsional response from the equation of motion. In this study, we discuss probability distribution of the induced torque and torsional displacement response to uncorrelated Gaussian white noise input ground acceleration in two translational directions. In the frequency domain, the power spectral density function of the torque induced by the Q–Delta effect is proportional to a convolution of two frequency response functions in two translational directions. In the time domain, the probability density function of the induced torque is the modified Bessel function of the second kind. Time-history response analysis was conducted to verify these theoretical results and to investigate the distribution of torsional response to Gaussian white noise input ground acceleration.

We propose a quantitative phase model to predict element waves of adjacent element faults to simulate ground motion using the stochastic Green's function method. We paired seismic records under the specified conditions of near-identical magnitude and hypocenter depth. To model the acceleration waveform similarities of the seismic record pair, a relationship between the mean value of Fourier amplitudes and an absolute value of Fourier phase difference was represented using a beta distribution. We then derived a regression formula of beta-distribution shape parameters with respect to the distance between the two hypocenters of each pair. Based on the proposed model, a seismic wave was synthesized, along with a calculated RMS error in acceleration response spectra between the synthesized wave and the observed record of the 2016 Kumamoto earthquake. It was confirmed that the proposed model decreased the error successfully.

Corporate social responsibility (CSR) is now considered to be one of the most important activities for companies as it greatly affects both companies and local communities. This study analyzes the effects of corporate social responsibility activities on the life recovery of employees. A questionnaire survey on the Great East Japan Earthquake Disaster in Iwaki City (Fukushima, Japan) was conducted. Iwaki City was among the areas most severely affected by the disaster. The effect of CSR activities on the life recovery of employees was analyzed by structural equation modeling. Life recovery largely depended on health and human relationships. CSR activities related to these two factors, such as work–life balance and local community activities, increased the life recovery of employees. Companies have large resources for improving local community resilience and local communities can recover from a disaster in a timely and effective manner when companies provide appropriate assistance. This study reveals how companies can contribute to the recovery of local communities through their CSR activities.

We are developing an artificial intelligence system for structural health monitoring that can detect local damage in a building structure by using the E-Simulator numerical simulation system that is being developed by the Japanese National Research Institute for Earth Science and Disaster Resilience. In this study, we confirmed the applicability of a multiclass classifier using a deep neural network to address the problem of identifying damage patterns in braces installed in a steel frame. Experimental data obtained from shaking table tests were used for training and testing. Cross-validation tests were conducted for several cases with different numbers of sensors, sensor degrees of freedom, and nodes in the hidden layers of the network. The results demonstrated that the accuracy of the damage pattern detection from the constructed classifier exceeded 77% when the appropriate hidden layers were selected and reached 87.9% for the best case.