A Novel Health Monitoring System for Synchronous Generators using Magnetic Signatures

Abstract

The Ph.D. research work presented in this thesis deals with the health monitoring of synchronous generators utilized in hydropower plants. Although various methods are available for fault detection in synchronous generators, the applicability of the available tools in the real world encounters numerous difficulties. A novel health monitoring system that can address the challenges in the field is proposed, consisting of a tailor-made sensor, signal processors, pattern recognition algorithm, and artificial intelligence. The proposed health monitoring system utilizes a stray magnetic field that can be measured on the stator backside of hydro generators. The proposed non-invasive health monitoring system is able to detect the fault type, estimate the severity, and find the location of the fault in the hydro generator. Moreover, the proposed health monitoring system has a high sensitivity that can detect a fault with low severity. In addition, the proposed pattern recognition algorithm is able to detect the fault without any need for prior knowledge about the reference generator. Finite element modeling of five synchronous generators is performed to investigate the existence of the stray magnetic field on the stator back side. In addition, the impact of topology, power rating, and design specification on the amplitude and pattern of the stray magnetic field are investigated. Although the amplitude of the stray magnetic field can be changed, the hidden pattern in the calculated signal for the faulty operation of the generators is identical. Several faults are investigated, including inter-turn short circuit fault in the rotor field winding, static eccentricity, dynamic eccentricity, mixed eccentricity, and broken damper bar faults. The impact of a fault on the stray magnetic field is also investigated using finite element modeling. Pattern recognition is the key part of this Ph.D. work. Several signal processing tools are used to extract the hidden patterns in the stray magnetic field signal. Fast Fourier transform, short-time Fourier transform, discrete wavelet transform, and continuous wavelet transform are used for feature extraction. Three unique patterns for the inter-turn short circuit fault are introduced that can detect even 1 shorted turn in the rotor field winding. A comprehensive algorithm is also proposed to detect eccentricity faults (static, dynamic, and mixed). The proposed algorithm is able to detect the location of the static eccentricity fault. Finally, the application of wavelet entropy on a stray magnetic field based on a broken damper bar fault is proposed that can detect the broken damper bar during both transient and steady-state operations of the generators. Reducing human error and reducing the cost of educating technicians to evaluate the patterns of the health monitoring system is achieved using an artificial intelligence system. Various classifiers are trained and their performance is assessed based on several meticulous evaluation functions. Among the proposed classifiers, a stacking classifier with logistic regression as a meta-classifier is selected due to its high performance and low computational complexity. The proposed method is tested using hand-out data and shows that the method can detect a fault with 92.74% precision. Extensive experimental tests are conducted on a tailor-made 100 kVA synchronous generator to verify the theoretical hypothesis. Various types of fault, including the inter-turn short circuit fault, static eccentricity fault, misalignment, and broken damper bar fault, with different severity can be applied to the setup. The tests are performed in both the no-load and on-load operations while the generator is connected to the local load and power grid. A custom-made sensor is designed to capture the stray magnetic field on the stator backside. Finally, two field tests are conducted in two hydro power plants in Norway to validate the proposed health monitoring system in reality.