Fault estimation for multi-rate descriptor systems using bi-directional long short-term memory neural network

Abstract

Fault estimation in multi-rate descriptor systems, which involve both differential and algebraic states, is particularly challenging due to the complexity introduced by multi-rate measurements. This paper proposes a novel fault estimation approach that combines a differential-algebraic equation based extended Kalman filter (DAE-EKF) with a bi-directional long short-term memory (bi-LSTM) neural network. The DAE-EKF is used to generate multi-rate residuals, which serve as inputs to neural networks to estimate faults. bi-LSTM networks improve upon LSTMs by processing data in both forward and backward directions, using past and future information. This bidirectional approach enhances temporal dependency capture, making bi-LSTMs ideal for accurate fault estimation. The efficacy of the proposed method is demonstrated using simulation studies on a two-phase reactor-condenser system with recycle and a reactive distillation system. The proposed approach has shown superior fault estimation capability for multi-rate descriptor systems compared to DAE-EKF with conventional feedforward neural networks and DAE-EKF with LSTM.