Abstract: One common problem is a small sample of uncertain parameters in the reliability analysis of the space mechanism. In this case, the boundary of the sample data is normally smaller than the boundary of the parameter. It is questionable to conduct non-probabilistic reliability analysis by establishing interval or ellipsoid convex set models with sample data and adopting uniform distribution to simply quantify convex set variables. To address the weakness of non-probabilistic reliability methods, an imprecise probabilistic reliability analysis method based on ellipsoid model is put forward. A high dimension construction method of ellipsoid model based on sample features is established. The ellipsoid model is standardized into a sphere model. The uncertainty is quantified by the principle of indifference reduction, and the joint probability density function of variables is derived. An important sampling method based on ellipsoid model is used to solve the imprecise probability reliability. The accuracy and feasibility of the method has been verified by a numerical example and a project example. Achieved good balance between robustness and accuracy in reliability analysis, it can be used as a useful supplement to probabilistic reliability method.

Key words: ellipsoid model, imprecise probabilistic reliability, non-probabilistic reliability, uncertainty quantification