Scholarlink

<p>Detecting changes or shifts in data over time is known as change-point detection. This phenomenon is crucial for identifying the deterioration of industrial components and preventing costly breakdowns or failures. There are several supervised and unsupervised approaches used in change-point detection, which involve evaluating the difference between the sampling distributions of two-time windows. The accurate detection of change-points is a critical challenge addressed by Industry 4.0 and can enable timely action to avoid costly failures in industrial elements. This paper discusses the use of distance-based common spatial patterns (DB-CSPs) as an offline change-point detection technique in multivariate time series data. DB-CSP is a supervised approach that projects the data into a subspace to identify the most relevant features that differentiate between two-time windows. Afterward, a classification algorithm is used to effectively detect changes in the data. We demonstrate the adequacy of LynxSight using public datasets and apply the new approach to a real industrial use case, achieving better results than some state-of-the-art techniques.</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

Lynxsight: change-point detection through different distance-based common spatial patterns

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年5月13日

查看全文
  • Arantzazu Florez,
  • Itsaso Rodríguez-Moreno,
  • Ane M. Florez-Tapia,
  • Arkaitz Artetxe,
  • Basilio Sierra

摘要

Detecting changes or shifts in data over time is known as change-point detection. This phenomenon is crucial for identifying the deterioration of industrial components and preventing costly breakdowns or failures. There are several supervised and unsupervised approaches used in change-point detection, which involve evaluating the difference between the sampling distributions of two-time windows. The accurate detection of change-points is a critical challenge addressed by Industry 4.0 and can enable timely action to avoid costly failures in industrial elements. This paper discusses the use of distance-based common spatial patterns (DB-CSPs) as an offline change-point detection technique in multivariate time series data. DB-CSP is a supervised approach that projects the data into a subspace to identify the most relevant features that differentiate between two-time windows. Afterward, a classification algorithm is used to effectively detect changes in the data. We demonstrate the adequacy of LynxSight using public datasets and apply the new approach to a real industrial use case, achieving better results than some state-of-the-art techniques.