Linear systems with large uncertainties, with applications to truss structures (Q868690)

The authors present a new approach to provide realistic bounds for the solution set \(\Sigma\) of linear systems \(Ax= b\), where \(A\) and \(b\) are allowed to vary within finite bounds \(\underline A\), \(\overline A\), and \(\underline b\), \(\overline b\), respectively. The method is applied to large linear systems of the form \((K+ BDA)u= a+ Fb\) with uncertainties in the diagonal matrix \(D\) and in the vector \(b\) only. These uncertainties may be large and arise in finite element calculations for structures with uncertain geometry, material properties, or loads. The method involves a computable overestimation factor which is frequently small. Moreover, it is based on classical vector inequalities, the construction of an initial enclosure and an iterative improvement using interval arithmetic and centered forms. For the latter ones tools for assessing the quality of overestimation are given, limitations of their application for larger uncertainties are discussed. Alternatives for initial enclosures are presented, and inner enclosures of \(\Sigma\) are used in order to bound the amount of overestimation on \(\Sigma\). All bounds are implemented for a class of mechanical finite element problems (truss structures with fixed topology and geometry, but uncertain elasticity, cross sections and forces). ANSYS and the MATLAB package INTLAB are used for the computations which illustrate the efficiency of the method.