Detection of shorted-turns in the field winding of turbine-generator rotors using novelty detectors-development and field test

A hybrid neural network model for rule generation and its application to process fault detection and diagnosis

“…In the area of power systems engineering, a number of successful FDD applications have been reported, e.g. for diagnosing stator and/or rotor faults as well as bearing defects in induction motor (Awadallah and Morcos, 2003), for identifying faulty conditions related to gases, such as overheating, corona, arching, that can lead to breakdown of insulating materials in power transformer (Zhang et al, 1996;De and Chatterjee, 2002;De, 2004;Castro and Miranda, 2005); for detecting defective circuit breakers based on vibration signatures (Yang et al, 1994;Lee et al, 2003;Negnevitsky and Pavlovsky, 2005), and for detecting shorted windings in operational turbine generator (Streifel et al, 1996). Various types of ANN models have been deployed in the aforementioned applications.…”

“…In the domain of machine learning, ANNs (Zhang et al, 1996;Streifel et al, 1996;Awadallah and Morcos, 2003;Negnevitsky and Pavlovsky, 2005), rule-based expert systems (Lee and Park, 1996;Tan et al, 2002), and decision trees (DTs) (Sheng and Rovnyak, 2004;Gao et al, 1998;Togami et al, 1995;Umanol et al, 1994) are among the popular techniques that have been employed extensively for undertaking FDD problems. Rule-based expert systems have been criticized for the requirement of an exhaustive collection of expert knowledge and an enormous effort in maintaining the knowledge base.…”

A New Robust Method To Detect Rotor Faults in Salient-Pole Synchronous Machines Using Structural Asymmetries

“…Many researchers have proposed different techniques to detect the inter-turn faults in the field winding [3][4][5]. A detailed analysis has been presented on brushless synchronous generators using the 30, 90, and 150 Hz armature circulating currents as indicators of field winding deterioration [4].…”

Time-based multi-layer perceptron for novelty detection in sensor networks

“…A predetermined threshold can be used to detect novel, previously unknown data distribution patterns. Streifel et al [9] suggest that the threshold can be applied in the form of a hypersphere placed around data of known distributions. However, this approach is often inefficient as data have different spreads along feature axes and the threshold can be estimated reliably by using additional data presented to the trained network [5].…”