将非概率凸模型理论与摄动理论相结合，通过有界不确定参数结构的特征值问题，对凸模型理论的一次近似算法作出一种改进。改进后的算法由于在计算中不用特征值导数，与Elishakoff的算法相比，不仅拓广了凸模型理论的应用范围，而且还可提高算法的计算效率。

将非概率凸模型理论与摄动理论相结合，通过有界不确定参数结构的特征值问题，对凸模型理论的一次近似算法作出一种改进。改进后的算法由于在计算中不用特征值导数，与Elishakoff的算法相比，不仅拓广了凸模型理论的应用范围，而且还可提高算法的计算效率。

This paper is concerned with the problem of comparison of two non-probabilistic set-theoretical models for dynamic response and buckling failure measures of bars with unknown-but-bounded initial imperfections. Two kinds of non-probabilistic set-theoretical models are convex models and interval analysis models. In convex models and interval analysis models, the uncertain quantities are considered to be unknown except that they belong to given sets in an appropriate vector space. In this case, all information about the dynamic response and buckling failure measures of bars is provided by the set of dynamic responses and buckling failure measures consistent with the constraints on the uncertain quantities. The dynamic response estimate is actually a set in appropriate response space rather than a single vector. The set estimate is the smallest calculable set which contains the uncertain dynamic response, but it is usually impractical to calculate this set. Two set estimate methods are developed which can calculate the time varying box or hyperrectangle, i.e. interval vector in the response space that contains the true dynamic response. Comparison between convex models and interval analysis models in mathematical proofs and numerical calculations shows that, under the condition of the outer enclosed ellipsoid from a hyperrectangle or an interval vector, the set dynamic response predicted by interval analysis models is smaller than that yielded by convex models; under the condition of the outer enclosed hyperrectangle or an interval vector from an ellipsoid, the dynamic response set calculated by convex models is smaller than that obtained by interval analysis models.

By a counter example, we show that there seem to be some problems in Ben-Haim's theory of robust reliability of dynamical systems based on convex models. We still point out that the property of the expansion of convex models is just the addition of a convex model and a real vector, and the property of the translation of convex models is just the scalar multiplication convex models. By means of the partial-order relation of the superscribed hyperrectangle or interval vectors of convex models, we present a correct criterion of reliability of the dynamical system with bounded uncertainty. Based on them, we propose the expansion function which is different from the one of Ben-Haim. Following Ben-Haim's thoughts, based on the new expansion function, we again define the input, failure, and overall reliability indices. By Ben-Haim's example, we obtain some results different from his. The conclusion and results may be thought of as to the further development of Ben-Haim's robust reliability.

In this study, a new, deterministic method is discussed for estimating the maximum, or least favorable frequency, and the minimum, or best favorable frequency, of structures with uncertain but non-random parameters. The favorable bound estimate is actually a set in eigenvalue space rather than a single vector. The obtained optimum estimate is the smallest calculable set which contains the uncertain system eigenvaluse. This kind of engenvalue problem involves uncertain but non-random interval stiffness and mass matrices. If one views the deviation amplitude of the interval matrix as a perturbation around the nominal value of the interval matrix, one can solve the generalized eigenvalue problem of the uncertain but non-random interval matrices. By applying the interval extension matrix perturbation formulation, the interval perturbation approximating formula is presented for evaluating interval eigenvalues of uncertain but non-random interval stiffness and mass matrices. A perturbation method is developed which allows one to calculate eigenvalues of an uncertain but non-random interval matrix pair that always contains the system's true stiffness and mass matrices. Inextensive computational effort is a characteristic of the presented method. A numerical example illustrates the application of the proposed method. Copyright.