Raza Saeed graduated from the University of Salahaddin, Iraq, with a BSc and MSc and has passed his PhD viva with a thesis entitled ‘Condition Monitoring of Hydraulic Turbine Rotors’. The external examiner was Prof Luiz Wrobel from Brunel University, UK and the internal was Prof Carlos Brebbia.
Development of fatigue cracks in hydraulic turbine runners has become one of the major factors contributing to the economic and safety of the hydro power station. During a routine inspection of Unit 2 in Derbendikhan power station, in February 2002, it was found that multiple fatigue cracks had developed mainly situated at the transition between the blades and crown/band. In order to investigate the causes of the cracks in the runner, the field data from start up to runaway between 1997 and 2002 were investigated. The total operation time for these five years of operation was more than 15000 hours and the operation conditions were recorded every hour.
In this study, based on the available data obtained from on site measurements, very large and complex model geometries of flow domain and whole turbine runner were created. Due to the complexity of these models, stress analysis of the Francis turbine runner was performed by numerical methods. Therefore, three-dimensional (3D) Computational Fluid Dynamics (CFD) simulation of the flow through the whole turbine runner was performed to calculate pressure distributions on the runner. The results obtained from the CFD analysis for different boundary conditions were incorporated into a FEM model to calculate stress distributions in the runner. The results indicated that the maximum stresses were situated at the transition between the blades and the crown on the trailing edge, which explained the appearance of fatigue cracks in these areas.
Since the computational domain has a very complicated geometry, the analysis of the fluid flow model and turbine runner models at each operational condition required large computer memory and computational time (more than eight hours). As a result, in order to analyze real life operational data (which consisted of more than 15,000 operation point), a simplified simulation technique was employed. In this study a developed simplified 3D model of the turbine runner blade was used to calculate the maximum stresses in the runner blades during operational periods. The results revealed that the level of maximum stresses was below the strength limit, which therefore, confirmed that the existence of fatigue cracks observed in the blades could not be attributed to the stresses alone, but could be produced by a combination of different factors. To evaluate the accuracy of the simplified model, the results were compared with those from the whole turbine runner model at eight different operating conditions. These results demonstrated that the simplified model was suitable for modelling the turbine blade, due to small discrepancies between two models.
With the absence of fault detection and monitoring condition system, fatigue cracks might cause severe damage and the turbine runner could be completely destroyed. For complex systems where analytical models can not be used, Artificial Intelligent (AI) techniques have been widely reported for fault detection and monitoring turbine conditions. Therefore, in this study, an intelligent system was developed for fault detection and monitoring.
AI techniques such as, Artificial Neural Network (ANN) and Adaptive Neuro-Fuzzy Interface System (ANFIS) were performed based on the vibration data obtained from the numerical analysis. In order to examine the capability of the AI techniques, as a first stage, these techniques were used to identify the cracks in the simplified model. A study was conducted for the identification of cracks based on natural frequencies and Frequency Response Functions (FRFs). At the next stage, the damage detection and condition monitoring of the whole turbine were investigated based on vibration data. Integrating CFD simulation of the flow domain with the FEM simulation of the whole turbine runner was used to obtain vibration data of both the intact and damaged turbine runner under different load conditions and different damage cases. Those vibration signals were used for predicting crack length in the runner and estimating the turbine operating conditions.
In order to provide the high performance automated diagnostic, multiple ANN and ANFIS models were proposed. The results showed that the multiple ANN and ANFIS models were capable of identifying faults better than using them individually. Finally, in order to improve AI techniques generalisation and robustness, different types and different levels of noise were added to the input vibration data.
As a result of his research, both examiners recommended the award of Doctor of Philosophy.