Scholarlink

<p>Modern industry requires the convergence of Operational Technologies (OT) with Information Technologies (IT). Industrial communications networks have been exhibiting characteristics of both domains. However, inherent problems from the IT world are now being considered on the factory floor. Once communication protocols based on Real-Time Ethernet are susceptible to intrusions, this work proposes a method for detecting attacks in PROFINET networks. The methodology employed feature extraction from data traffic using a sliding window technique, dimensionality reduction with an autoencoder, and classification using a One-Class Support Vector Machine (OCSVM). The algorithm is an unsupervised machine learning classifier, which reduces computational effort. This technique is well-established and meets the defined requirements, contributing to its implementation in other approaches. A total of six distinct scenarios are analyzed. It is highlighted that two of them involve data acquired in real industry applications. The results show accuracy in the range of 97.3–100%, depending on the scenario.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Unsupervised Machine Learning-Based Intrusion Detection in PROFINET Networks

  • 期刊论文

  • 发表日期：

    2025年6月14日

查看全文
  • Afonso Celso Turcato,
  • Guilherme Serpa Sestito,
  • Andre Luis Dias,
  • Rogerio Andrade Flauzino

摘要

Modern industry requires the convergence of Operational Technologies (OT) with Information Technologies (IT). Industrial communications networks have been exhibiting characteristics of both domains. However, inherent problems from the IT world are now being considered on the factory floor. Once communication protocols based on Real-Time Ethernet are susceptible to intrusions, this work proposes a method for detecting attacks in PROFINET networks. The methodology employed feature extraction from data traffic using a sliding window technique, dimensionality reduction with an autoencoder, and classification using a One-Class Support Vector Machine (OCSVM). The algorithm is an unsupervised machine learning classifier, which reduces computational effort. This technique is well-established and meets the defined requirements, contributing to its implementation in other approaches. A total of six distinct scenarios are analyzed. It is highlighted that two of them involve data acquired in real industry applications. The results show accuracy in the range of 97.3–100%, depending on the scenario.