[3Rin4] Interactive 1

[3Rin4-63] Slime Detection Using Machine Learning with Soft Labels and Conventional Technique

〇Junki Yamaguchi¹, Sohei Arisaka¹, Yuki Tamagawa¹, Kojiro Takesue¹ (1.Kajima Corporation)

Keywords:Machine Learning, Time Series Classification, Soft Labels, Construction

During pile construction, it is necessary to confirm that there is no slime at the bottom of a pile. Detection currently depends on skilled workers, so it is difficult to quantify and reproduce results. To solve this, we are developing a new method for slime detection based on measured tension data. In our previous research, we proposed a method using machine learning, and verified its applicability. In this paper, we proceeded with the following two considerations. First, we added data of which even skilled workers were unsure whether slime existed or not. Second, we regarded multiple measurements as one sample. Considering these, we propose a method using machine learning that is close to the conventional method for slime detection.