Abstract : A novel framework for fault diagnosis is proposed. New rules are presented to enhance decisions making under a Bayesian latent mixture model. The proposed decision rules improve Fisher discriminant analysis based scheme. These rules allow it to deal with known and not acknowledged faults. Performance evaluation using the Tennessee Eastman Process data is presented and shows a good performance of the proposed approach compared to existing methods.
Mohamed Amine Atoui, Vincent Cocquempot. New Decision rules for Fisher discriminant analysis : applied to fault diagnosis. ECC 2021, Jun 2021, Online, Netherlands. ⟨hal-03175810⟩