A Study on Seismic Response Prediction Method Based on RC Non-Structural Wall Damage Photographs Using Deep Learning

2023 Fiscal Year Final Research Report

• Project/Area Number: 21K04354
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Review Section: Basic Section 23010:Building structures and materials-related
• Research Institution: Kyushu University
• Principal Investigator: Yoshioka Tomokazu 九州大学, 芸術工学研究院, 教授 (40304852)
• Project Period (FY): 2021-04-01 – 2024-03-31
• Keywords: 地震被害 / 深層学習 / RC方立壁

Outline of Final Research Achievements

This study aimed to establish a method for estimating the maximum member angles experienced by condominium residents by measuring the amount of damage from photographs of earthquake-damaged RC vertical walls taken by them. As a part of this research, horizontal loading tests were conducted on full-scale RC vertical walls with shear span ratios of 0.5 to 0.67, assuming shear failure, and an equation relating the area of concrete loss and member angle was derived from the test results. Using photographs of damage to RC vertical walls taken at a damaged condominium building as training data, we proposed a method for estimating the experienced member angle from the measured concrete loss area by generating a detector to extract cracks and concrete loss separately by fine-tuning a pre-trained deep learning model. The proposed method is to estimate the experienced member angle from the measured concrete delamination loss area.

Free Research Field

建築構造

Academic Significance and Societal Importance of the Research Achievements

本研究は、区分所有の共同住宅（所謂、分譲マンション）の地震被害情報を、震災時の意思決定に大きな困難を抱える区分所有者に速やかに提供することで、復旧工事の実施、継続使用に向けての早期の合意形成を支援することを目的の一つとしている。また、大都市のマンション化率は30％前後であり、都市に所在する膨大な数の共同住宅の地震被害情報を、デジタルデータとして行政等の組織が一元的に収集、管理することで、より効率的な復旧支援、復旧計画の立案が可能となり、本研究の成果は震災に対する都市の強靭な回復力（レジリエンス）の向上を図るための基礎技術となり得る。