Details of a Researcher - Kohiyama, Masayuki

Presentations -

Division display 21 - 40 of about 161 ／ All the affair displays >>

Evaluation of Torsional Response of Super High-Rise Buildings Induced by Q–Δ Resonance Using Single- and Double-Layer Models

MAKI Shiori, KOHIYAMA Masayuki

Annual Meeting of Architectural Institute of Japan (Online) ,

2021.09

Oral presentation (general), Architectural Institute of Japan

LQR Control Design Scheme for a Seismic Response Control Device for Improving the Resilience of a Medical Facility

KOHIYAMA Masayuki, WAKABAYASHI Kento, TAKAHASHI Kento

Annual Meeting of Architectural Institute of Japan (Online) ,

2021.09

Oral presentation (general), Architectural Institute of Japan

Fundamental Study on Q-Δ Resonance of Super High-Rise Building with Twin Towers

ANAMIZU Ryosuke, KOHIYAMA Masayuki,

Annual Meeting of Architectural Institute of Japan (Online) ,

2021.09

Other, Architectural Institute of Japan

Fundamental Study on Transfer Learning for Deep Neural Network to Classify Damage Patterns of a Wooden Building

KUMAGAI Takuma, MOGI Chisa, KOHIYAMA Masayuki, YAMASHITA Takuzo

Annual Meeting of Architectural Institute of Japan (Online) ,

2021.09

Oral presentation (general), Architectural Institute of Japan

「レジリエント建築」で実現する持続可能な都市と社会

小檜山雅之

第50回信頼性・保全性・安全性シンポジウム (オンライン開催) ,

2021.06

Oral presentation (invited, special), 一般財団法人日本科学技術連盟

ねじれ２次モードのＱ—Δ共振の検証実験と有限要素モデルによる再現解析

穴水亮輔，水鳥文哉，小檜山雅之

日本地震工学会・大会－2020 (オンライン開催) ,

2020.12

Oral presentation (general)

Interpretation of Deep Neural Network for Damage Pattern Classification Using Phase Plane

Kumagai, T., Kohiyama, M., Yamashita, T.

7th Asian-Pacific Symposium on Structural Reliability and Its Applications (APSSRA2020) (The University of Tokyo (online conference)) ,

2020.10

Oral presentation (general)

The author’s research group has proposed a damage pattern classification method using a deep neural network (DNN) for monitoring the structural health of a building. We aimed to make the DNN for damage pattern classification, more accountable using the “interpretation” and “explanation” methods. We proposed generating input data that maximize the classification probability of each damage pattern, called a prototype, and drew a trajectory on a two-dimensional phase plane of appropriately selected state variables with color information on the degree of influence on the classification. We applied the proposed method to DNNs for damage classification of a wooden structure and a steel frame. The acceleration–velocity plane provided useful information about the mechanical characteristics of the DNN used in the damage pattern classification, and we thus demonstrated the effectiveness of the proposed method.

Wireless Seismic Response Control of an Instrument with Casters Based on Reinforced Learning Considering Stochastic Models of Computer Vision

Wakabayashi, K., Kohiyama, M., Eguchi, R., Takahashi, M.

7th Asian-Pacific Symposium on Structural Reliability and Its Applications (APSSRA2020) (The University of Tokyo (online conference)) ,

2020.10

Oral presentation (general)

We developed a computer-vision-based system to control the response of medical instruments with casters in a building subjected to earthquake excitation. The controller employed a neural network, and reinforced learning based on the dynamic simulations of a controlled instrument under building floor motion was conducted. In the dynamic simulations, we used the Open Dynamics Engine, which is a simulator library of three-dimensional rigid body dynamics and introduced a friction model to simulate the dynamic behavior of the casters. To accurately simulate the dynamic behavior of the controlled instrument, we considered the control delay of the proposed system and introduced stochastic models of errors to identify the displacement of the instrument using computer vision. To validate the proposed system, we performed dynamic simulations of a controlled instrument under building floor motion. It was revealed that the instrument response was significantly suppressed by introducing the stochastic model of errors in the training of NN. This result suggests the usefulness of stochastic models in enhancing the performance of reinforced learning based on dynamic simulations.

Experiment on Torsional Response of Symmetric Structure Induced by Q–Δ Resonance,

Mizutori, F., Yokoyama, H., Kohiyama, M.

17th World Conference on Earthquake Engineering (17WCEE) (Sendai, Japan) ,

2020.09

Oral presentation (general)

Study on Visual Interpretation and Explanation Method for Deep Neural Network to Classify Building Damage Patterns Based on Acceleration Response

KUMAGAI Takuma, KOHIYAMA Masayuki, YAMASHITA Takuzo

Summaries of Technical Papers of Annual Meeting, Architectural Institute of Japan,

2020.07

Other, Architectural Institute of Japan

Validation Experiment of Torsional Response Induced by Q-Δ Resonance of a Single-Story Structure under White Noise Ground Motion

MIZUTORI Fumiya, KOHIYAMA Masayuki

Summaries of Technical Papers of Annual Meeting, Architectural Institute of Japan,

2020.07

Other, Architectural Institute of Japan

Simulation of Horizontal Bidirectional Ground Motion Using Stochastic Green’s Function Method Considering Phase Similarity in Element Waves with Circular Distribution

KOYAMA Yuki, YOKOYAMA Haruka, KOHIYAMA Masayuki

Summaries of Technical Papers of Annual Meeting, Architectural Institute of Japan,

2020.07

Other, Architectural Institute of Japan

Fundamental Study on Q-Δ Resonance Considering Second Mode in a Non-Eccentric Structure

ANAMIZU Ryosuke, MIZUTORI Fumiya, KOHIYAMA Masayuki

Summaries of Technical Papers of Annual Meeting, Architectural Institute of Japan,

2020.07

Other, Architectural Institute of Japan

Finite element analysis of the Q‒Δ resonance in a two-story structure

Kohiyama, M., Anamizu, R., Mizutori, F.

3rd International Conference on Computational Engineering and Science for Safety and Environmental Problems (COMPSAFE2020),

2020.01

Oral presentation (general)

水平変位に直交する慣性力が誘発するねじれ応答の共振現象「Q-Δ共振」

小檜山雅之

日本建築学会近畿支部 (大阪科学技術センター) ,

2019.12

Public lecture, seminar, tutorial, course, or other speech, 日本建築学会近畿支部構造力学部会

建築構造シミュレーションと深層学習を用いた構造ヘルスモニタリング

小檜山雅之

日本地震工学会・大会－2019 (京都大学吉田キャンパス（京都府京都市）) ,

2019.09

Oral presentation (invited, special), Japan Association for Earthquake Engineering

国立研究開発法人防災科学技術研究所では開発中のシミュレーションシステム「数値震動台」を建築物の構造ヘルスモニタリングに活用するための研究を行っている．対象建物について複数の損傷パターンを想定し，様々な入力地震動に対して地震応答解析を行うことで，深層ニューラルネットワークの訓練データを多量に生成することができる．本報告では鋼構造骨組の振動台実験結果を用いて学習を行い，深層ニューラルネットワークの損傷パターン判別精度を検討した結果を紹介する．また，未学習の損傷パターンに対する誤判定を防ぐため，新たに開発した未学習パターンデータ検知法も紹介する．

統計的グリーン関数法における要素地震波の巻き込み正規分布を用いた位相類似性モデル

横山春果，小檜山雅之

日本地震工学会・大会－2019 (京都大学吉田キャンパス（京都府京都市）) ,

2019.09

Poster presentation, Japan Association for Earthquake Engineering

2016年熊本地震の余震記録を用いて震源が近接したペアを作成し、フーリエ振幅平均値とフーリエ位相差の関係を巻き込み正規分布を用いてモデル化した。次にフーリエ振幅平均値から巻き込み正規分布のパラメータを予測する回帰式を求め、その回帰係数と震源間距離と震源距離の回帰式を導き、要素地震の位相の設定方法を提案した。提案した位相モデル、ランダム位相モデル、コヒーレント位相モデルの3つの模擬地震動と実観測記録の加速度応答スペクトルのRMS誤差を比較し、提案モデルが他の2つモデルに比べ誤差が小さいことを確認した。

Seismic Response Control of Instrument with Casters Using Computer Vision Based on Dynamic Analysis Considering Friction

Yousuke KAWANISHI, Kento WAKABAYASHI, and Masayuki KOHIYAMA

Summaries of Technical Papers of Annual Meeting, Architectural Institute of Japan (金沢工業大学扇が丘キャンパス（石川県野々市市）) ,

2019.09

Oral presentation (general), Architectural Institute of Japan

Shaking Table Experiment of Torsional Response Induced by Inertia Force Perpendicular to Horizontal Displacement

Fumiya MIZUTORI, Hiroki YOKOYAMA, and Masayuki KOHIYAMA

Summaries of Technical Papers of Annual Meeting, Architectural Institute of Japan (金沢工業大学扇が丘キャンパス（石川県野々市市）) ,

2019.09

Oral presentation (general), Architectural Institute of Japan

Detection of Unlearned Pattern Using Support Vector Machine in Classification of Damage Patterns Based on Deep Neural Network

Masayuki KOHIYAMA, Kazuya OKA, and Takuzo YAMASHITA

Summaries of Technical Papers of Annual Meeting, Architectural Institute of Japan (金沢工業大学扇が丘キャンパス（石川県野々市市）) ,

2019.09

Oral presentation (general), Architectural Institute of Japan

　 Previous page - Next page 　