[2J01-06] Reactor Maintenance, Inspection and Preservation Technology

[2J01] Development of Abnormal Sign Detection System using AI for Nuclear Power Plant

(1)Study of detection algorithm using autoencoder

*Isaku Nagura¹, Shinya Tominaga¹, Miyake Ryota¹, Toshio Aoki¹, Susumu Naito², Yasunori Taguchi², Yuichi Kato², Kouta Nakata² (1. Toshiba ESS, 2. Toshiba)

Keywords:Abnormal sign detection, Deep learning, Autoencoder

We develop the abnormal sign detection function which collects a lot of process data and analyzes using AI. This function can support daily operation and soundness check activities and it may be able to contribute to the improvement in operating rate, maintainability and safety of nuclear power plant. However, to put this function into practical use, it is important that the user can handle it easily, user can apply it all systems or all plant states, and it can detect abnormal sign on process data with high performance including unknown events. In order to satisfy these requirements, this research has applied an auto-encoder as a deep learning method and has been developing its algorithm so that it can learn a large amount of process data including time series changes. In this research, we have confirmed the effective detection performance of this algorithm using plant data generated by a full scope simulator and it is going to be reported.