Doctoral Dissertations

Date of Award

12-1994

Degree Type

Dissertation

Degree Name

Doctor of Philosophy

Major

Nuclear Engineering

Major Professor

Belle R. Upadhyaya

Committee Members

R. E. Uhrig, R. B. Perez, J. D. Landes, A. E. Ruggles

Abstract

The development of an accurate technique for the service life prediction of steam generator tubes is influenced by two major problems in practice. These are: (1) availability of limited information regarding the in-service conditions of the tubes and (2) effects of material aging and degradation resulting from service exposure. In an attempt to solve the above problems, this dissertation research focused on the development of a model-based methodology for the life assessment of steam generator tubes subjected to a certain degradation process. The methodology combines engineering analysis of the degradation process under study with the analysis of process field data and information to establish semi-empirical models to forecast the future trend in the degradation. The projection of this trend to a pre-defined allowable degradation level was used to determine the life expectancy of the component. The salient feature of this methodology is in its capacity to recognize the process nonlinearities and to identify the correct process trends which cannot be detected by simple applications of traditional forecasting techniques. This capacity greatly reduces the amount of required field data for a good forecast. The semi-empirical prediction models were developed using both parametric and non-parametric techniques. The functional forms of the parametric models were determined based on both the approximated physics of the degradation process and the availability of field data for a reliable estimation of the model parameters. To improve the model prediction, the unknown process dynamics were enhanced by the incorporation of an empirical dynamic formulation of field data. The non-parametric models were established based on artificial neural network techniques. The neural network prediction (NNP) models were trained with a wide range of possible degradation paths generated by the process model. The trained NNP models perform an intelligent data mapping to match the trends in the input data to one of the learned process paths and thereby forecasting the future trends of the process within a specified time-frame. The proposed life assessment approach was used to predict the wear process of a Once-Through Steam Generator (OTSG) tube within its 15th tube support as a complex application of trend forecasting. To implement the methodology, a tube wear process model was developed to simulate the process trend over time with regard to aging and degradation mechanisms resulting from service exposure. The simulated wear data were used to establish various semi-empirical prediction models to determine suitable models for process trend prediction with different block sizes of data. The results show the degradation of friction coefficient and material aging could markedly increase the rate of tube wear with subsequent tube life reduction. They also indicated that the power-exponential model [atbexp(ct)], developed based on this methodology, provided a more accurate prediction of the tube life than the power-law model (atb) suggested in the literature. A significant improvement in the prediction of the power-law model was obtained by including a dynamic formulation of the historical data. The results also demonstrated that NNP models perform extremely well for both trend recognition and prediction, even with a limited amount of data. An attractive feature of NNP models is that they are far less noise-sensitive than parametric models. The estimation of residual life of tubing was demonstrated using both parametric and NNP models. The methodology developed in this research is generic in nature and can be applied to the life assessment of both stationary components and rotating machinery.

Download
Files over 3MB may be slow to open. For best results, right-click and select "save as..."

Share

COinS