Rouse, James Paul
(2014)
Computational component analysis techniques for high temperature power plant applications.
PhD thesis, University of Nottingham.
Abstract
There is a trend in the power industry for high temperature components (such as steam pipe work) to be operated in an increasingly arduous fashion. This would involve the use of elevated steam temperatures/pressures and a greater frequency of start up/shut down cycles. Such generation strategies are being adopted due to the need for thermally efficient power supply that can match fluctuating market demands. If these generation strategies are to be implemented safely it is critical that careful analysis of the system components is conducted in order to ensure that premature failure does not occur. The advanced material models and techniques that are used in academia to simulate these components are often out of reach of the engineers working in industry.
The present work describes the development of an analysis “toolbox” that takes several advanced material models (which can accommodate complex loading conditions) and applies them in numerical (finite element analysis, FEA) and approximate life estimation methods. The toolbox comprises several modules, each of which relates to a specific aspect of component analysis. In this thesis, the fundamental procedures behind these modules are developed in novel ways in addition to the development of the toolbox as a whole. The toolbox modules may be roughly divided into the definition of a component’s material, geometry and loading condition, followed by some form of analysis procedure and a report of the key results.
A material’s behaviour is commonly determined from mechanical tests. For in service components, scoop sampling is an exciting new method to extract small amounts of material which may then be tested using several novel small specimen techniques. An investigation has been conduced in the present work that verifies the safety of this method and allows the localised stress behaviour around an excavation to be estimated. Material constants in material behaviour models are usually determined by fitting the outputs of the model to experimental data in an optimisation procedure. A great deal of work has been completed on this topic using the complex Chaboche unified visco-plasticity model. This has led to the formation of the combined parallel optimisation strategy and the development of data cleaning for the determination of material constants in any model.
Due to the high temperature conditions power plant components operate in, creep is a major concern. Several damage material models have been compared which can represent failure due to creep. Generally, these models can be divided into power law and hyperbolic sine functions. Through a comparative investigation using multiple component geometries, it has been found that the hyperbolic sine function creep law gives lower predictions of failure time than the power law models at realistic stress levels. Hyperbolic sine function failure lives were also more representative of reality. It is therefore critical when performing component analysis to consider the form of a material model as well as the loading range its material constants are applicable to. The Chaboche unified visco-plasticity model has also been discussed. Using this model, both hardening due to the accumulation of plastic strain and viscous effects (such as creep stress relaxation) may be described. Models like this will play an important role in the analysis of high temperature components as they experience fluctuations in both load and temperature.
Although it appears simple, the geometry of a high temperature pipe bend in a power plant is actually complex due to the manufacturing process employed (a straight pipe section is heated through induction coils and bent using a fixed radius arm). The pipe’s wall thickness not only varies circumferentially around the pipe’s cross section but also around the bend itself. Through the analysis of industrial data (collected by ultrasonic measurement of components during outage inspections) several novel geometry factors have been developed that quantify this dimension variation. A new method to analyse such pipe bends has also been created that interpolates the stress states between two dimensional (2D) models that represent the cross section of a pipe bend at several key locations.
Once a geometry, loading condition and material has been defined, an analysis procedure may be employed in order to assess the condition of the component. As creep is a key concern under high temperature conditions, most of the analysis procedures discussed in the present work are focused on the prediction of peak rupture stresses (δR) which may be used to estimate failure lives due to creep. Several approximate (errors are typically less than 5%) parametric relationships have been developed that allow peak rupture stresses to be determined based on, for example, pipe bend geometry factors. In addition, to aid in bespoke FEA analyses, a collection of routines with a graphical user interface (GUI) have been created that can write input files for a commercial FEA code (ABAQUS), run the job and post process the results. This can save a great amount of user effort when attempting to analyse components. Finally, an original neural network (that uses a partially connected, multiple input node architecture) has been proposed that predicts δR in pipe bends operating under steady-state creep conditions. Both internal pressure and system loads have been incorporated as inputs for this neural network. This has required the definition of several new load factors that describe the system loads acting on a component.
Recommendations for future developments based on this research have also been given. Future developments may look to include fatigue effects in parametric equations, as well as considering the effect of varying loading conditions (possibly through a damage fraction approach). The Chaboche model (or similar unified model) may be modified to include temperature dependency and damage effects (allowing for a wider application to component analysis). The effect of geometry variation may be included in the neural network, again extending its applicability, and stresses due to temperature distributions in the piping components may be incorporated (at present, these have not been considered, however system loads may be thermally driven).
The work presented in this thesis addresses a complete analysis procedure, from collecting material information from a component through scoop sampling, to determining material constants for this material by an optimisation procedure and analysing the component using either numerical or approximate methods. Although pipe bends have been considered for the significant part of this work due to the relatively small amount of research reported in literature, similar methodologies may be applied to other power plant components of interest, such as welds, steam headers or branch pipes.
Actions (Archive Staff Only)
 |
Edit View |
Tools
Tools
