Efficient Elimination Orders for the Elimination Problem in Diagnosis

A consistency relation is a constraint on the time evolution of known variables (and their time derivatives) that is fulfilled if the known variables are consistent with a model. Such relations are useful in diagnosis and can be derived using elimination theory. Unfortunately, even apparently small elimination problems proves impossible to compute on standard computers. An approach to lessen the computational burden is to divide the complete elimination problem into a set of smaller elimination problems. This is done by analysing the structure of the model equations using graph theoretical algorithms from the field of sparse factorization of symmetric matrices. The algorithms are implemented in Mathematica and exemplified on a fluid-flow system where the original elimination problem does not terminate. Applying the proposed algorithms give an elimination strategy that terminates with a solution in just a few seconds.