An adaptive empirical model for real-time condition monitoring of nuclear power plant components

Ahmed, Ibrahim Ahmed; Zio, Enrico

Permalink

https://hdl.handle.net/2142/121792 Copy

Description

Title

An adaptive empirical model for real-time condition monitoring of nuclear power plant components

Author(s)

• Ahmed, Ibrahim Ahmed
• Zio, Enrico

Issue Date

2023

Keyword(s)

• Auto associative kernel regression
• Auto associative bilateral kernel regression
• Signal reconstruction
• Real-time condition monitoring
• Fault detection
• Nuclear power plant

Abstract

Advancements in data analysis and computational efficiency are motivating the nuclear and other industries to apply condition-based maintenance (CBM) for early failure prevention and mitigation, minimizing unplanned shutdown, increasing safety and reducing maintenance costs. In CBM, Fault detection systems monitor the health state of the components and aid the operator/maintenance engineer to decide whether a maintenance intervention is required. Typically, a fault detection system is a decision tool based on the model that reconstructs the values of on-line signals expected in normal conditions and the residual calculator that analyses the differences between the measured on-line signal values and the reconstructed values where an alarm is triggered if the residuals are statistically deviated from the allowable range. While several data-driven models have been developed for condition monitoring in NPP, there are several challenges that need to be addressed such as adaptability of the model to evolving environments. Generally, the reconstruction model is trained using a fixed number of observations of historical data collected over time during normal operating conditions. Instead of fixing the data, it must be desirable to update (appending and/or thinning) it by selecting ‘better’ normal operation data to provide better results from the fault detection system. This need is apparent due to the following reasons: a) the range of normal operation can be extended further than that of initial memory data; b) too much data in a narrow range can be noisy; and c) normal conditions may change depending on ageing or human intervention, such as load following mode operations of a nuclear power plant. Addressing this challenge would be beneficial not only for the current operational NPPs but also for the advanced and small modular reactor designs, particularly, in the aspect of load following mode of operations which are anticipated in the modern/future reactor design concepts. To alleviate this challenge, an adaptive empirical model is developed. The proposed model consists of: 1) a signal reconstruction model based on the ensemble of Auto Associative Kernel Regression (AAKR) and Auto Associative Bilateral Kernel Regression (AABKR) for reconstructing the values of the signal expected in normal condition during steady-state and transition from one normal operating condition to another, respectively; 2) a state transition detector based on the derivatives for automatically detecting the transition from one normal operating condition to another; 3) an adaptive loss function with a forgetting factor formulated for real-time automatic updating of the reconstruction models; and 4) a Luus–Jaakola (LJ) optimization procedure employed for the determination of model parameters in an efficient and less computational manner. The performance of the proposed method is, first, evaluated on a synthetic data generated from a numerical process that emulates different evolution patterns of normal operating conditions of a component and, then, applied to real-time simulation data collected from a pressurized water reactor (PWR) NPP. The results demonstrate the effectiveness and applicability of the proposed method.

Type of Resource

text

Language

eng

Handle URL

https://hdl.handle.net/2142/121792

Owning Collections