Development of fault diagnosis for nuclear power plant using deep learning and infrared sensor equipped UAV

Abstract

Fault component detection is necessary for safety and maintenance in large-scale industrial fields including nuclear power plants. Therefore, this study proposes a method for diagnosing a power plant composed of numerous components based on deep learning using a UAV with an IR sensor and a camera. The proposed method could diagnose the components and recognize the fault component in real time. In this study, a ther-mal-hydraulic integral effect test facility, which is a scaled-down nuclear power plant, is utilized considering the nuclear power plant. The database for the application of deep learning was performed by combining an IR in-tensity map and general image to enhance the performance of component classification and fault detection. Deep learning was applied using object detection and classification methods based on convolutional neural networks (CNNs) that are effective for image processing. As a result, this technology can diagnose the multi-component by a single measurement instrument. The optimal performance of component classification and fault detection was 55.9 ms per 16 batches, demonstrating a mean average precision (mAP) of 0.9913. This technology could be applied to various industries as a comprehensive component condition monitoring method for operating effi-ciency and safety.