Accuracy Improvement of Fault Diagnosis Methods for Small Modular Pressurized Water Reactors Based on Machine Learning Methods - Slides

Aug 6, 2026

Upload by Himbo

The paper presents a fault diagnosis approach for small modular pressurized water reactors (SMPWR), addressing the difficulty of capturing SMPWR-specific complex fault characteristics in conventional methods. Feature selection and feature extraction can cause interference, and SMPWR makes reliable feature determination challenging. The work proposes advanced deep learning to automatically learn fault features, reduce human-driven feature engineering, and construct more accurate and reasonable intelligent diagnostic models for complex system faults.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages3

Reading Time8 min

AccessFree to download

Introduction

Motivation for SMPWR fault diagnosis

Literature Review

Existing fault diagnosis approaches

Research Methods

Virtual simulation data generation

Transfer learning with CNN feature extraction

CNN input design and classification output

Accuracy Improvement of Fault Diagnosis Methods for Small Modular Pressurized Water Reactors Based on Machine Learning Methods Yong Liu Xi’an Jiaotong University, Xi'an 710049, China. Abstract: The paper proposes a fault diagnosis method for small modular pressurized water reactors (SMPWR) . Due to the compactness and complexity of SMPWR, for general fault diagnosis methods, it is difficult to extract the corresponding special complex fault characteristics, resulting in a great increase in the difficulty of diagnosis. Therefore, this paper intends to propose a more applicable diagnosis method in close combination with the characteristics of SMPWR. Traditional machine learning methods based on feature selection and feature extraction work well in other research areas, but when it comes to SMPWR, it is difficult to determine the features that should be extracted, and different feature selection can cause great interference in the diagnosis results. Therefore, this project proposes to use advanced deep learning algorithms to automatically learn fault features of complex systems, reduce human interference, and build up more reasonable and accurate intelligent diagnostic models. Keywords: Fault Diagnosis; Machine Learning; Small Modular Pressurized Water Reactors Introduction Under the circumstances of global warming, with the progress of science and technology and productivity further released, demands for energy, especially the low carbon energy will be more and more vigorous. Compared with other new energy sources, nuclear energy starts earlier and develops longer, has the advantages of long design service life, low comprehensive cost, and small land occupation space, and has become an important part of the global energy source. Since 2000, the United States and Russia have all carried out researches on small modular reactors (SMR) with innovative technological characteristics. Compared with traditional nuclear water reactors, the small modular reactor (SMR) has the characteristics of high inherent safety, diverse applications, flexible deployment, low investment, short construction period, etc. According to statistics from the International Atomic Energy Agency (IAEA), there are currently more than 50 types of small modular reactors at different research and development stages worldwide. [1] While in China, the small modular pressurized water reactor named as NHR200-Ⅱ adopts the design concept of integrated layout, full power natural circulation, self stabilization, passive safety, and is equipped with an intermediate isolation circuit, which can actually eliminate large-scale radioactive release. [2] But it is a pity that due to the nuclear accident in Fukushima Daiichi, researches on nuclear power have been stagnant all over the world. From my perspectives, what we should learn from nuclear accidents is that we ought to import more automation application and reduce human intervention, when in emergency, instead of overthrowing the whole nuclear industry. On the other hand, in order to make the progress of nuclear industry sustainable, we should pay attention to fault diagnosis. From my perspectives, designing an automation system, which detects faults accurately and on time and even proposes some workable methods for operators to act, are significantly important to research on in nuclear industry. Volume 10 Issue 5-107- Literature Review Fault diagnosis technology can monitor the system operation condition, determine whether faults occur, and also determine the time, location, size and type of the fault, complete fault detection, separation and estimation, which plays an important role in improving the operation reliability of complex industrial systems and reduce the system operation risks. [3] In present, the fault diagnosis methods for closed-loop systems mainly contain model-based methods, signal processing-based methods and knowledge-based methods. For complex systems such as the Small Modular Pre

FAQs

Why are conventional fault diagnosis methods less effective for small modular pressurized water reactors (SMPWR)?

SMPWR’s compactness and complexity make it hard to extract SMPWR-specific complex fault characteristics, which increases diagnostic difficulty.

What problem does the paper identify with traditional machine learning feature engineering for SMPWR?

When features are determined through different feature selection strategies, it can significantly interfere with diagnosis results, and it is difficult to decide which features should be extracted.

How does the proposed method improve accuracy for SMPWR fault diagnosis?

It uses deep learning, including transfer learning with a convolutional neural network, to automatically learn fault representations from original data and reduce human interference in feature extraction.

Aug 6, 2026

Upload by Himbo