5層鉄骨試験体の振動台実験に基づく機械学習による損傷検出手法の検証

摘要

In this paper, the damage detection method based on machine learning on a five-story steel frame which has simulated damage is shown through shaking table tests. Supervised learning and unsupervised learning are utilized and damage detection method by using response analysis data is applied. The following results are shown:.1) By supervised learning in case of participation function as attribute, damaged story can be estimated. By unsupervised learning in case of participation function as attribute, damaged cases can be estimated as anomaly.2) Acceleration data from response analysis as the training data and shaking table test data as the verification data are utilized for supervised learning, and damaged story can be estimated.%擬似的損傷を有する5層鉄骨造試験体の振動台実験データに、機械学習による損傷検出手法を適用した。教師あり学習と教師なし学習の両方の方法や時刻歴応答解析データも活用した損傷層の推定を行った。本研究の範囲內で以下に示す知見が得られた。1)固有振動数と刺激関数は、1層に擬似的損傷がある時には1次モードが変化し、5層に擬似的損傷がある時には高次モードが変化する傾向があった。