Published papers

The Q–Δ effect is a phenomenon that a column with different lateral stiffness generates a restoring torsional moment owing to geometric nonlinearity when its top displaces in two horizontal directions simultaneously. The Q–Δ resonance that is resonance of a torsional mode induced by the periodic torsional moment due to the Q–Δ effect could cause considerable torsional vibration even in non-eccentric buildings. However, the Q–Δ resonance in higher-order modes has not been investigated well. We first constructed a model of a two-story non-eccentricity structure and derived the conditions of the Q–Δ resonance considering higher-order modes. Then, we conducted shaking table experiments and finite element analysis to reproduce the experiments. As a result, it was confirmed that the second mode in the torsional direction was excited as predicted by the derived Q–Δ resonance condition. Moreover, the finite element analysis could reproduce the experimental result with high accuracy.

We developed a computer vision system to control the seismic response of medical equipment with casters in a building subjected to earthquake excitation. The controller employs a neural network and the reinforced learning was conducted based on dynamic simulations of a controlled equipment under building floor motion. In the dynamic simulations, we used Open Dynamics Engine, which is a simulator library of three-dimensional rigid body dynamics, and we introduced a friction model to simulate the dynamic behavior of the casters. To validate the proposed system, we conducted shaking table tests and it was confirmed that the proposed control could successfully suppress the response of the equipment.

The biaxially symmetric column induces torsional moment owing to the bending–torsion interaction when the top of the column is given horizontal displacement in the two directions of the principal axes of the cross section and the rotation of the column ends around the axes is constrained. Hence, horizontal ground motions in two directions can cause non-negligible torsional vibration in a high-rise building owing to autoparametric resonance. To investigate this phenomenon, in this study, we focus on a single-story structure with an aggregated column, which is a model for columns or core structures in tall buildings and is assumed to be an elastic Euler beam; subsequently we derive the equations of motion under horizontal ground motions in two directions. The resonance condition in the torsional direction is also investigated. Time-series response analyses are performed for input ground motions of harmonic waves, white noise, acceleration records of prior earthquakes, and a long-period design wave, and the relationship between the ratio of two translational natural frequencies and the increase in maximum responses is explained. The results demonstrate that the absolute acceleration response to the long-period design wave could increase by more than 50% in the worst-case scenario.

It is well-known that semi-active solution can achieve building isolation with much less energy requirements than active solutions. Also, it has been shown in previous studies that compared to linear damping, non-linear damping performs better for building isolation under sinusoidal ground motions. The present study is concerned with the extension of the application of the semi-actively implemented non-linear damping to building isolation under seismic loadings. A two-degree-of-freedom (2-DOF) scaled building model is used for simulation studies. Experimental tests on a physical building model have been used to validate the effectiveness of the 2-DOF scaled building model in representing the behaviors of a physical building structure. The optimal design of the semi-actively implemented non-linear damping for building isolation under design seismic motions is then carried out using the 2-DOF scaled building model based on simulation studies. The results show that an optimal design of semi-actively implemented non-linear damping can improve the performance of building isolation under design seismic motions in terms of both absolute acceleration and inter-story drift.

Deep neural networks (DNNs) are a powerful tool for structural health monitoring because they can automatically identify features that are useful for classifying and recognizing damage patterns of a target structure with high accuracy. However, it can misclassify input data of an unlearned damage pattern as any of the learned damage patterns. To address this shortcoming, this paper presents a method to detect unlearned damage patterns by using the collective decision of support vector machines (SVMs). SVMs are constructed using feature vectors from training data, which are stored in the output layer of a DNN. To validate the proposed method, we used two different datasets, one containing experimental data of a steel frame structure and the other containing simulated and experimental data of a wooden house. In both cases, it correctly identified data of both learned and unlearned damage patterns. The proposed method can enhance the effectiveness of structural health monitoring (SHM). In addition, because it does not employ SHM-specific characteristics, it can be used in various pattern recognition applications, such as image and audio processing.

The volcanic hazard map is one of the effective proactive measures for volcanic disasters. However, the general public has little interest in the hazard maps because it is not easy to understand them. To address this problem, this study develops a volcanic hazard map using augmented reality (AR). The effectiveness of the proposed AR hazard map was verified through the workshops held in Fujiyoshida City, in which participants actually used the prototype map. As a result, it was clarified that the AR hazard map could effectively supplement a printed hazard map and it was also found that the display switching function of the hazards was particularly effective for promoting understanding.

In order to reduce volcanic disasters, it is important for residents to practice mutual assistance, which can be led by a voluntary disaster reduction group. However, many communities lack the human resources for their future leaders. To address this issue, we created a framework of disaster imagination game to develop next-generation leaders. The effectiveness of the game was verified through workshops held in Fujiyoshida City. The framework allows residents to train independently by using a web application software.

In recent disasters, there have been many cases that people failed to escape despite evacuation advisories or did not make appropriate preparations in advance. One of the possible causes is the influence of cognitive bias. However, countermeasures against cognitive bias have not been developed in community disaster management. In this study, we developed a training tool in the form of web application software to overcome cognitive bias by acquiring knowledge about possible cognitive bias in response to volcanic disasters and discussing countermeasures with community members. The effectiveness of the tool was verified in workshops held in Fujiyoshida City. This training tool allows residents to conduct training on their own using a video conferencing system.

Because of geometric nonlinearities, vibrations in two horizontal directions can generate a torsional moment that induces a torsional response, even in a structure with no eccentricity; we term this phenomenon Q–Δ effect. In this study, the torsional response caused by the Q–Δ resonance was investigated by performing a shaking table test that involves a single-layer symmetric specimen. The specimen was designed and fabricated by focusing on one of two resonance conditions. Its moment of inertia was adjustable by changing locations of weights, and the natural frequency in the torsional mode could be modified. We developed a numerical model of the specimen, in which the columns connected to the slab were integrated into an elastic Euler beam. It was confirmed that the torsional response increased near the predicted Q–Δ resonance point. In addition, the acceleration at the corner of the slab was significantly increased. The formulated equations of motion provided a better prediction of the actual phenomena. Because the specimen corresponds to a stiff and slender high-rise building when converted into a full scale, the result suggests a need to be aware of the risk of torsional response increase due to the Q–Δ resonance.

Eccentric distribution of stiffness, damping, and mass of a structure, and spatially non-uniform ground motion input to a long or large base mat of a structure are well-known causes of torsional response. We have discovered that the torque generated by horizontal displacement and perpendicular inertial force, which we call the Q-Δ effect, can be a cause of torsional response. We formulated the equation of motion of a single finite-size mass-linear elastic shear and torsion spring model and clarified the resonance condition of the torsional response to sinusoidal ground acceleration. Time-history response analysis verified that the torsional response forms beat and the maximum torsional response of the simulation result agrees with that theoretically predicted. Further time-history response analysis conducted of white noise ground acceleration showed that even one-directional white noise ground acceleration can induce torsional response in a linear elastic system without any structural eccentricity.

Many buildings in Japan currently have a base-isolation system with a low stiffness that is designed to shift the natural frequency of the building below the frequencies of the ground motion due to earthquakes. However, the ground motion observed during the 2011 Tohoku earthquake contained strong long-period waves that lasted for a record length of three minutes. To provide a novel and better solution against the long-period waves while maintaining the performance of the standard isolation range, the exploitation of the characteristics of nonlinear damping is proposed in this paper. This is motivated by previous studies of the authors, which have demonstrated that nonlinear damping can achieve desired performance over both low and high frequency regions and the optimal nonlinear damping force can be realized by closed loop controlled semi-active dampers. Simulation results have shown strong vibration isolation performance on a building model with identified parameters and have indicated that nonlinear damping can achieve low acceleration transmissibilities round the structural natural frequency as well as the higher ground motion frequencies that have been frequently observed during most earthquakes in Japan. In addition, physical building model based laboratory experiments are also conducted, The results demonstrate the advantages of the proposed nonlinear damping technologies over both traditional linear damping and more advanced Linear-Quadratic Gaussian (LQG) feedback control which have been used in practice to address building isolation system design and implementation problems. In comparison with the tuned-mass damper and other active control methods, the proposed solution offers a more pragmatic, low-cost, robust and effective alternative that can be readily installed into the base-isolation system of most buildings.

Deep neural network has been applied to damage classification in structural health monitoring. However, it has a drawback that it classifies data of an unlearned pattern as any of learned patterns. We propose a method to detect unlearned-damage-pattern data based on k-nearest neighbors algorithm and Welch’s t test. As input data of neural network, we used acceleration records acquired in shaking table test using steel frame specimen, which is a 1/3-scale model of a four-story steel structure and imitates structural damage with relaxed braces. The records were divided into one-second-long samples and used for training and verifying neural network. We extracted feature data in an output layer of the network and clarified that the data do not follow to a Gaussian distribution. Therefore, Anomaly index based on k-nearest neighbors algorithm is introduced to measure the non-similarity to learned pattern data. Using data of shaking table test of a steel frame structure, we verified that unlearned-damage-pattern data can be detected by Welch's t test.

The Sosokan building of Keio University located in Yokohama, Japan, is a base‐isolated building with a semi‐active control system. The semi‐active dampers’ damping coefficient is changed to best reproduce the optimal control force in linear‐quadratic‐gaussian control. However, the implemented control gain was calculated on the basis of the structural parameters applied at the design stage. In this study, we identified the building's actual structural parameters using observational records from the 2011 Tohoku earthquake. Using the updated structural model and 1000 simulated ground motions, we conducted a dynamic response analysis, which resulted in a fragility curve. Then, we evaluated the damage probability using the fragility curve and a seismic hazard curve. Finally, we determined the optimal control parameters to minimize the damage probability and used these parameters to implement the semi‐active control system.

Although the magnitudes of inland earthquakes are smaller than those of earthquakes in subduction zones, they have caused catastrophic disasters in the past. Seismic damage to structures strongly depends on the geometric relationship between a fault and the structure itself, especially near a fault. In recent years, the locations and activities of inland active faults have been identified in Japan. However, the position of a rupture-starting point and rupture extent are uncertain. Together with this uncertainty, we have to consider the difference in the seismic performances of structures, upon which the extent of damage also depends. In this study, seismic ground motions near a fault were simulated with consideration of the above-mentioned uncertainty, and the geometric distribution of the damage probability of wooden structures near a fault was investigated. The effect of seismic performance was also investigated. First, the ground motion with directivity on an engineering bedrock was simulated with a non-uniform slip model of an earthquake-source fault using the stochastic Green’s function. Element seismic waves overtook the phases of neighboring waves. Then, a rupture-starting point was decided stochastically and we assumed that the location of the earthquake-source fault followed a uniform distribution in the possible range. Ground motion on the surface was created by inputting the ground motion into a surface-layer model composed of loam using the one-dimensional equivalent linear analysis. If the maximum shear strain was over 5% in the response analysis result, we trimmed the input waves to maintain their reliability. Wooden structures were modeled considering the restoring force characteristics in accordance with the criteria of three seismic grades. Time-history response analysis was carried out by inputting the ground motion to the building model. Binomial discrimination as to whether or not damage occurred was conducted on the basis of the maximum story drift angle of the building models. Using the results of binomial discrimination, fragility curves assuming a lognormal distribution were constructed for each site using a maximum-likelihood method. Here the peak ground velocity was used as a ground-motion intensity parameter. A stochastic model of earthquake occurrence time was defined in consideration of the uncertainty of the earthquake scale, assuming a stochastic model for the variation of the slip quantity of the earthquake-source fault. We used a Brownian passage time distribution for making these models. Hazard curves of the ground motion were evaluated by this probability model. Finally, the earthquake damage probability of wooden structures was calculated for each site based on the fragility curve and the hazard curve mentioned above. As a result, the influence of the geometric relationship between a fault and the wooden structures and the seismic performance on the earthquake damage probability was obtained. The damage probability of wooden structures was significantly higher in the vicinity of the center of the possible range of the earthquake-source fault. In the vicinity of the fault, earthquake damage probability was reduced more effectively by changing the seismic grade from 1 to 2 than from 0 to 1.

Recently, disasters due to concomitant hazards of earthquake and snow have been increasing in Japan. When engineers verify the structural safety of buildings in a heavy-snow region by means of structural calculation, a coefficient value of 0.35 is used for snow weight under load combination with earthquake with reference to Building Standard Law of Japan. This paper investigates the necessity of load combination of moderate earthquake and moderate snow. The fragility surface of a large span steel frame under seismic excitation and snow load is developed based on dynamic analysis. Two kinds of damage-related responses, the maximum stress of members and the maximum response acceleration of ceilings are examined. With respect to the maximum stress, the evaluated fragility surface and joint hazard surface of earthquake and snow suggest that additional consideration is required for load combination of moderate earthquake and moderate snow.

This study formulates the seismic fragility function for ceilings based on the results of shaking table tests obtained from 19 studies. The maximum likelihood method is employed to fit a three-parameter log-normal function. As an application example, damage occurrence probability is evaluated for a vibration-controlled 15-story building.

Time history response analysis was conducted to evaluate torsional response of wooden houses with eccentrically-located mass, such as balconies and solar power panels, under strong ground motion. Five kinds of two-story models were examined and the following findings were observed: (1) In the cases of models with a longer width in the east-west direction, a model with a living room on the first floor showed larger torsional response than that on the second floor because of difference in the amount of walls in the first story. (2) A model with a longer width in the north-south direction showed larger torsional response than that in the east-west direction because of larger eccentric distance. (3) A model with a double-height ceiling room had twice larger response than that without it. Thus, it is necessary to design houses with a double-height ceiling room with careful consideration of torsional response.

Road blockage caused by collapsed buildings hinders evacuation, firefighting, and lifesaving activities. It also expands the damage of fire spreading. We propose a method to evaluate the disaster reduction performance of a district with three indices of fire extinguishing probability, ratio of non-burned buildings, and evacuation achievement rate. In the evaluation of these indices, we use the Monte Carlo method based on building collapse rate, fire outbreak probability, etc. As a result of evaluating the proposed indices for two virtual districts: one with latticed road and the other with the radial-ring road, we find that the proposed indices reflect the characteristics of the road network and can express the disaster reduction performance of the districts. In addition, they can facilitate the priority of seismic retrofitting plans considering the characteristics of the district.

The Sosokan university building has base isolation and a semi-active control system. The semi-active dampers’ damping coefficient is changed so as to best reproduce the optimal control force in linear-quadratic-Gaussian control. However, the implemented control gain was calculated based on the structural parameters at the design stage. In this study, we identify the building’s actual structural parameters by using observation records of the 2011 Tohoku earthquake. Dynamic response analysis is conducted by using the updated structural model and 1000 simulated ground motions. Then, the damage probability is evaluated using a fragility curve derived from the result of dynamic response analysis as well as a seismic hazard curve. Finally, the optimal control parameters are obtained to minimize the damage probability, and these parameters are implemented in the semi-active control system.

An inter-story tuned mass damper (ISTMD) is a variant form of a tuned mass damper, which is placed between consecutive floors to suppress the resonant response of a structure. We derived design parameters of an ISTMD under ground excitation that minimize story deformation on the basis of the fixed-points theory. Then, a shaking table tests were conducted to investigate the performance of an ISTMD with derived parameters. The result showed that the ISTMD worked well to suppress the response around the resonance point and it reduced the displacement and maximum acceleration responses more effectively than an ordinary damper when the El Centro wave was input to the specimen.