This study proposes a scheme for fault detection and identification (FDI) of a class of non-linear hybrid systems. For fault detection, the proposed method uses a self-switched unscented Kalman filter (UKF) where a component filter with the appropriate dynamic model is chosen to suit the current mode of the hybrid plant. The mode of the hybrid plant, usually defined by system states, is deduced based on the state estimates. The statistical tests of measured datasets are used to detect and identify the fault parameters. For fault identification, a bank of such switched UKF filters is used. A three-tank system has been used to demonstrate the effectiveness of the scheme. Different types of faults, that is, leakage, clogging and abrupt changes in inflow (actuator fault) have been considered. t-Statistical test has been performed on the residuals for FDI and threshold calculation purposes. It is shown that all the types of faults can be successfully detected and identified by the proposed method. The performance of the proposed system is compared with the same for an extended Kalman filter-based FDI system. It has been shown that the UKF-based scheme has lower latency and higher range coverage.

An improved estimation method for a class of nonlinear hybrid systems has been proposed in this paper using a self-switched R-adaptive Extended Kalman Filter. The term 'estimation of a hybrid system' implies state estimation as well as mode estimation of a plant. In hybrid systems, where modes are determined by the output variables, the mode determination may become erroneous due to inaccurately known or wrongly initialized measurement noise covariances. Innovation and residual based R-adaptive Extended Kalman filters are employed here successfully for adapting true sensor noise covariance. A three tank system has been used in this work to demonstrate the effectiveness of the scheme. Simulation results show the effects on the performance of the estimator due to inaccurately or wrongly assigned magnitudes of sensor noise covariance using innovation based and residual based adaptive extended Kalman filter. A comparison with an EKF based self-switched filter shows that the proposed A-EKF based self-switched filter is more robust.

An improved fault detection scheme for a non-linear hybrid system with delayed measurement by using a modified non-linear adaptive state estimator is proposed. The proposed estimator performs acceptably even when the (i) covariance of the measurement noise is unknown and also (ii) when the measurements are delayed. The algorithm for the proposed estimator called time-delayed R-adaptive central difference Kalman filter (TD-RACDKF) uses a modified R-adaptive 2nd order Central Difference estimator, also called Central Difference Kalman Filter (CDKF) in some literature. Algorithm for the proposed TD-RACDKF has been presented and its performance evaluated as well as characterised with Monte Carlo simulation on two standard non-linear systems with delayed measurements. The characterisation includes comparison with existing estimators. Having demonstrated the improved performance of the proposed state estimator, its use for fault detection of non-linear hybrid systems was investigated. The performance of the fault detection scheme has been illustrated with the help of a benchmark non-linear hybrid system, namely 'three tank system' and by comparison with a previously available extended KF-based estimator.

No abstract

The main objective of this research is to propose a novel machine learning based fault detection algorithm for a nonlinear switched continuous system (NSCS) with time delayed measurements. A simulation model has been established through the analysis of the relevant constraints such as estimation errors derived from self-switched time- delayed (TD) Unscented Kalman Filter (UKF). Machine learning techniques, specifically probabilistic principal components analysis (PPCA) and the support vector machine (SVM) have been successfully adopted on the database of the proposed model to detect and isolate the faults. ‘Three tank system’, the commonly used bench mark problem, has been re-employed here to prove the efficiency of the proposed estimation and fault detection algorithm. In this work, it is conjectured that the available measurements are time delayed. Through this study, the efficacy of the proposed time- delayed Unscented Kalman Filter (TD-UKF) and Machine Learning based fault detection have been demonstrated on the afore mentioned benchmark problem. The results indicates that Machine learning based algorithm can detect the fault with more accuracy and less false alarm than other statistical methods. The finding of this research can be used for any type of industrial system to detect the faults with better potential and higher range coverage.