DEEP LEARNING FRAMEWORK FOR POST-HAZARD CONDITION MONITORING OF NUCLEAR SAFETY SYSTEMS

Abstract

A novel approach is presented to conduct data-driven condition assessment in nuclear safety systems with the aid of deep learning. With the resurgence of nuclear energy due to the ever-increasing demand for electricity and carbon free power generation, ensuring safe operations at nuclear facilities is important. Nuclear safety systems, such as equipment-piping, undergo aging and subsequent degradation due to flow-accelerated erosion and corrosion. Conventional non-destructive techniques implemented during plant outages can take weeks and months to scan all the systems in their entirety. Continuous condition monitoring of such systems would result in lowering the maintenance costs along with extending the operating lifetime for a nuclear power plant. Additionally, the proposed framework should be able to detect minor degradation caused due to aging of nuclear facilities. Uncertainty in the degradation severity levels is also incorporated in the design of the condition assessment methodology. In this paper, the use of artificial intelligence (AI) algorithms as well as vibration-based health monitoring for degradation detection has been demonstrated. A simple equipment-piping system subjected to an external hazard, such as an earthquake, is selected as an application case study. A proof-of-concept is presented wherein the proposed framework utilizes the data collected from sensors to generate a machine learning data repository, demonstrates pattern recognition and feature extraction, explores the design of an artificial neural network (ANN), and develops a sensor placement strategy. The effectiveness of the proposed framework is demonstrated on a realistic primary safety system of a two-loop reactor plant. It is shown that the proposed post-hazard condition monitoring framework is able to detect degraded locations along with the severity levels with high degree of accuracy.