[3Q3-J-13] AI application: analysis of physical behaviors in artifacts

[3Q3-J-13-02] Slime detection during pile construction using machine learning

〇Sohei Arisaka¹, Yuki Tamagawa¹, Kojiro Takesue¹ (1. Kajima Corporation)

Keywords:Machine Learning, Time Series Classification, 1-dimensional Convolutional Neural Network, Construction

During pile construction, an inspection is needed to check absence of bottom slime which leads to settlement and inclination of structures. A conventional method for slime detection is dependent on individual judgement known by a sense of a hand. Therefore, there are some problems in terms of reproducibility and quantification. In order to solve these problems, we are studying a new method for slime detection using measured tension data. In this paper, we applied machine learning to judge whether slime exists or not from the tension data. Among 6 algorithms we compared, 1-dimendional Convolutional Neural Network achieved the best performance at 93% accuracy. According to this result, we verified that machine learning is effective for the slime detection.