Machine learning-based fault estimation of nonlinear descriptor systems

This article focuses on the problem of fault estimation of nonlinear descriptor systems (NLDS) using intelligent approaches. Firstly, an extended Kalman filter for descriptor systems is employed for state estimation. Then, the residuals are generated and mapped to detect and confirm the fault. Finally, machine learning approach and neural network models are used to estimate faults. For machine learning approach, Gaussian process regression is employed to estimate fault magnitude. Additionally, a back propagation neural network is also applied for fault estimation. The efficacy of the proposed methods are demonstrated with the help of benchmark chemical mixing tank descriptor system (Yeu et al., 2008) and two-phase reactor and condenser with recycle (Kumar and Daoutidis, 1996). It is observed that the Gaussian process approach outperforms neural network-based approach for fault estimation.