Performance monitoring and fault prediction using a linear predictive coding algorithm

Feed-forward neural networks for smooth operation of the high-voltage power transmission network

“…It is also proved that neural network is a very powerful abstraction to learn patterns [ 42,43,47,48 ]. In our case it means that we may consider facility downtimes also as a pattern and use the neural network for fault prediction [ [70][71][72][73][74][75][76][77][78][79][80] ].…”

A robust fault diagnosis scheme based on signal modal estimation

“…Knowledge or expert system-based fault diagnosis schemes have also been investigated (Singh et al 1987, Doraiswami andJiang 1989 Recently, a real-time performance monitoring scheme for dynamic systems was developed by Doraiswami (1993) based on advanced signal processing techniques. The scheme developed is capable of monitoring the behaviour of dynamic systems, such as stability margins and robust performance, by accurately estimating the underlying mathematical model of the system.…”

“…rather than model coefficients as in parameter estimation-based approaches; the effect of system disturbance is minimized because the modal estimation algorithm will treat the disturbance as additional dynamics which will then be eliminated in the model validation stage using model reduction techniques (see Doraiswami 1993); it is sufficient for the algorithm to use only one measurement signal from the dynamic system to carry out the entire fault diagnosis since any signal within the control loop contains all the necessary modal information for fault diagnosis; and faults of various magnitudes or severities can easily be accommodated by proper selection of the range of the physical system parameter variations when preparing the root locus.…”

Influence matrix approach to fault diagnosis and controller tuning

“…This matrix will have constant components due to the assumption of multihearity. By analyzing the geometric alignment of the feature vector with respect to the column vectors of the influence matnx, the source of fault can be isolated and the seventy of the fault estimated in case only one of the physical parameters is faulty [2]. An experimental verification on a physical system is in order: the model of a physical system is highly complex, unknown and often nonlinear..…”

“…Instead of a single identification experiment (as is done conventionally), a number of simple identification experiments are performed by subjecting each physical parameter to some known perturbations. For each case, the system response to a rich input signal is measured, and the model parameters (feature vector) are estimated using an SVD-based batch least squared algorithm [2]. This is done assuming different model mctures (orders) for the system.…”