Work

Building a prototype from irregularly sampled and distorted data for outlier time series detection

Year: 2023

Type: conference-paper

Abstract: This paper presents a framework for unsupervised time-series outlier detection in the context of distorted and irregularly sampled data. The framework includes the design of a prototype, computed as a barycenter, that synthesizes a set of unlabeled time series. The barycenter computation is allowed by an efficient formulation of the well-known Dynamic Time Warping (DTW) algorithm, namely Soft-DTW. Then, a score based on the distance DTW is used to assess the similarity between the prototype and new time series, latter labelled via a normality threshold. The framework is evaluated on movement data collected from industrial robots operating in a plant of the automotive manufacturer Vitesco Technologies.

Source: HAL

Authors: Lacoquelle, Charlotte, Pucel, Xavier, Travé-Massuyès, Louise, Merle, Christophe, Raymonet, Axel, Enaux, Benoît

Institutions: Laboratoire d'Analyse et d'Architecture des Systèmes, Office National d'Études et de Recherches Aérospatiales, Université de Toulouse, Communauté d'universités et établissements de Toulouse, Actia Group (France)

Language: English

Topic: Time Series Analysis and Forecasting

Subfield: Signal Processing

Field: Computer Science

Domain: Physical Sciences

Open Access status: green

Funder: Agence Nationale de la Recherche

Awards: G3010527862, G3879958799, G602161992

OpenAlex ID: W7135540282

FWCI:

Cites: 3

Cited by:

Related to: 10