| This research develops concepts and methodologies in system self-assessment of performance reliability with multivariate time series modeling. System self-assessment of performance reliability is implemented by monitoring and predicting physical performance measures in real-time. The conditional performance reliability over a period of future time is then deduced from the forecast results. This research focuses on system performance reliability assessment with multi-input and multi-output (multivariate performance measures and multiple failure modes). In the research, multiple performance variables, sampled across time, are treated as a multivariate time series. The prediction of performance measures is conducted based on the multivariate time series model.; This research is an extension of univariate performance reliability assessment in real-time. This research explores applicable forecasting methodologies that can be adapted in real-time system performance forecasting. Different forecasting methods, including state-space modeling supported by Kalman filtering, multivariate adapted exponential smoothing, univariate linear trend exponential smoothing, and univariate adaptive exponential smoothing are studied, compared and evaluated. The numerical connection between multivariate performance measures and multiple failure modes are developed in the research. In order to cope with the qualitative statements, vagueness, or subjectivity of opinion that exist in reliability assessment, the concept and applications of fuzzy reliability are also explored. A prototype software package for real-time system self-assessment of survival is also developed for implementation. This performance reliability model mainly contains three aspects: real-time data acquisition from selected performance measures, performance modeling and forecasting, and projection of forecasting results into performance reliability. |
