ABSTRACT

The assessment and prediction of reliability of space mechanisms are very critical challenges, due to multi-factor, multi-level and high-dynamic (J.Q. Hua, 2011). The reliability problem of space mechanisms have become increasingly complex, dynamical, cascading characters of and cannot match the assumptions of static state, independence and large samples in reliability theory (Joseph Homer Saleh, 2011). Traditionally, the reliability of space mechanisms has been evaluated by testing individual systems and verifying their performance is within some acceptable limits. In order to collect sufficient data, sometimes some limited testing have to be implemented for components or subsystems when full scale testing is not feasible for system under actual work environment. Simulation modeling methods are often available for whole system while the full scale testing for the environment is not feasible. The data of component level is easier and cheaper to obtain compared with the system data, so it is beneficial if establishing individual component and/or subsystem level models and incorporating them into a system level model is possible. Therefore the uncertainty in the system model is relate with the component and/or subsystem level data (Angel Urbina, 2009). Moreover

of these components must be taken to obtain an estimate of the variability for quantify the uncertainty of the model. But the interactions of these components were never tested. No information on the coupling of components could be available. It may add an extra uncertainty. In addition, the interactions of components could have been tested at excitation levels that arenot comparable to those of the full system. In Ref. (Urbina, A. 2006), the information available at the different levels was not used to augment the knowledge about the model of a component in a different configuration. This data was merely used to assess the model’s predictive capability for the intended work environment (U.S. Department of Energy, 2000, AIAA, 1998) (i.e. model validation). By using Bayesian updating techniques and Bayesian networks, it is possible to incorporate the available data, updating the model parameters and the model predictions consequently, to reflect the new information that was previously not available for any individual level of complexity. This methodology has been explored in Ref. (Rebba, R. 2005,) for a system where one is the application domain of interest and for one component. Our paper extends that work to dynamic coupling and cascading of complexity and to a multicomponent problem and it adds the additional complications of working with actual experimental data and complex dynamics simulation models.