Session: 02-05-01 Data-driven and AI-based Models

Submission Number: 181411

Research on Anomaly Detection Methods for Marine Structure Health Monitoring Data Based on Attention-Enhanced LSTM 

As a critical component of modern transportation infrastructure, the operational safety of ships is vital to public safety and economic stability. With vessels' service lives continually extending, structural aging has become increasingly prominent, drawing widespread attention to ship structural safety and reliability. Ship Structural Health Monitoring (SHM) systems have emerged as an essential technological means to ensure operational safety, extend structural lifespan, and enhance emergency response capabilities. Within data-driven SHM frameworks, high-quality data is fundamental for accurately assessing structural health and formulating effective maintenance plans. However, due to complex operational environments, sensor failures, and communication interruptions, monitoring data often contains anomalies, which compromise the accuracy of data analysis and structural safety evaluation. Thus, effective anomaly diagnosis and data repair are of significant importance.

 

Current approaches for anomaly handling predominantly rely on fixed signal thresholds for filtering or simple noise removal, often requiring manual data segment selection, leading to inefficiency, high missed detections, and false alarms. To address these challenges, this research systematically investigates anomaly diagnosis and data repair methods based on practical monitoring data processing needs. The main contributions include:

(1) Summarizing and categorizing common monitoring data formats from real-world engineering structures, proposing a data format standardization approach, and visualizing massive standardized monitoring data by converting it into images, thereby establishing a classification criterion for monitoring data images;

(2) Developing an efficient and accurate anomaly diagnosis model for monitoring data using deep learning techniques;

(3) Repairing identified anomalous data through preprocessing operations such as outlier removal, missing value imputation, data smoothing, and standardization to improve data quality and usability.

Presenting Author: Anwen Sun Huazhong University Of Science And Technology

Presenting Author Biography: Sun Anwen, a graduate student from Huazhong University of Science and Technology, focuses on data processing in structural health monitoring.

Authors:

Anwen Sun Huazhong University Of Science And Technology
Bingquan Yang Huazhong University of Science and Technology
Weixin Zhou Huazhong University of Science and Technology
Xiansheng Zhou Wuhan Second Ship Design and Research Institute
Minghe Wang Wuhan Second Ship Design and Research Institute
Jingxi Liu Huazhong University of Science and Technology