A1087
Title: Detecting covariance shifts in multichannel profiles
Authors: Davide Forcina - University of Naples Federico II (Italy) [presenting]
Christian Capezza - University of Naples Federico II (Italy)
Antonio Lepore - Universita di Napoli Federico II (Italy)
Biagio Palumbo - University of Naples Federico II (Italy)
Abstract: Modern industrial systems generate multichannel profile data continuously, requiring effective real-time monitoring and fault diagnosis. While many existing methods prioritize detecting shifts in the process mean, changes in the covariance structure are just as important, as they reflect the dynamic interdependencies among multiple variables. A functional graphical modeling framework is introduced to represent conditional dependencies in multichannel profile data, addressing challenges posed by high dimensionality and sparsity. The method leverages penalized likelihood ratio tests with adaptive penalty terms to detect a wide range of covariance structure changes. To enhance interpretability, a diagnostic procedure based on change-point detection is used to pinpoint the specific relationships that have changed. Simulation studies and a case study on multichannel temperature profile monitoring demonstrate the superior performance of the proposed approach compared to existing methods.
