Seminário Dr. Prof. Danilo Marcondes Filho

Title : Multivariate Time Series Clustering using U-Statistic-based

method: An Approach for Batch Process Monitoring

 

Abstract: Time series clustering is a consolidated research area. One of the important applications of time series clustering methods is fault detection. Following this aim, this paper presents a preliminary study about the clusterization of multivariate time series in the field of Statistical Process Control (SPC). In particular, we are focused on batch processes, a peculiar process that generates multivariate time series data for each batch sample. Here we extend a study of Cintra et al. (2023), which proposed a control approach based on U-Statistic theory focused on univariate time series. They present an approach to evaluate new batch samples by measuring U-statistics-based distances between Autocorrelation Functions (ACF) of reference and new batches, one variable at a time. This work describes an approach to evaluate batches considering additionally the cross-correlation between variables in each multivariate time series. The control charts considering the Cross-Correlation Function (CCF) and ACF of each variable jointly are built to evaluate a new batch. Through preliminary simulations, our method signalizes changes in the process affecting its dynamics, considering both serial and cross-correlation.

Palestrante : Prof. Dr. Danilo Marcondes Filho
Dados: 23/07/2024
Horário : 13h30min
Local:  Sala A101 do Instituto de Matemática e Estatística da UFRGS (Av. Bento Gonçalves, 9500 Prédio 43-111)
Scroll to Top