Details

Autor(en) / Beteiligte

• Bektaş, Nurullah
• Kegyes-Brassai, Orsolya

Titel

Enhancing seismic assessment and risk management of buildings: A neural network-based rapid visual screening method development

Ist Teil von

• Engineering structures, 2024-04, Vol.304, p.117606, Article 117606

Ort / Verlag

Elsevier Ltd

Erscheinungsjahr

2024

Quelle

Alma/SFX Local Collection

Beschreibungen/Notizen

• Some of the existing buildings are designed based on lower design standards or even without considering seismic design standards. Recent earthquakes have further highlighted the vulnerability of these buildings when subjected to severe seismic activity. Consequently, it has become imperative to conduct seismic vulnerability assessments of the existing building stock. Therefore, the assessment of the existing building stock is required through the utilization of Rapid Visual Screening (RVS) methods. However, the existing conventional RVS methods used in seismic building assessments have shown limited accuracy. Furthermore, because these methods were developed based on expert opinions and/or due to access limitations to detailed assessment-based generated data used for their development, further enhancing them is challenging. To address these limitations, a new RVS method, which leverages Neural Networks (NN) and building-specific parameters, for reinforced concrete, adobe mud, bamboo, brick, stone, and timber buildings has been proposed in this study. Unlike conventional methods that rely on site seismicity class, the developed data-driven approach incorporates building-specific parameters such as the fundamental structural period and building spectral acceleration. The developed RVS method is specifically tailored to analyze diverse types of buildings in regions with varying seismicity risks, all in preparation for an impending earthquake. In this study, the developed RVS method demonstrated a promising 68% test accuracy, effectively representing the building performance against earthquakes. These findings illustrate the potential of the developed NN based RVS method in assessing existing buildings, thereby mitigating potential loss of life and property during imminent earthquake and alleviating the associated economic burden. Furthermore, this study introduces a new RVS method that can pave the way for future advancements in the field of seismic vulnerability assessment of existing buildings. •Developed a novel Rapid Visual Screening (RVS) method that leverages Neural Networks (NN) algorithms to address limitations in conventional RVS methods.•Achieved a significant test accuracy rate of 68% with the developed RVS method, showcasing its effectiveness in predicting building damage states.•Utilized a substantial dataset comprising over 700,000 buildings, highlighting the limitations of existing RVS methods developed with limited building data, which may lead to overfitting.•Demonstrated that RVS methods developed using the same data in the literature are not suitable for general use, underscoring the need for a more robust and comprehensive approach to building vulnerability assessment.•Emphasized the importance of feature engineering and the incorporation of additional parameters, such as fundamental building period, spectral acceleration, distance to the source, and height-area ratio, to enhance the model's performance and comprehensiveness.