Evaluation the performance of voting algorithms used in fault tolerant control systems

Abstract Redundant application algorithm “variants” are used to minimise errors in fault-tolerant real-time control systems. Voting algorithms arbitrate between variant results to generate a single output. Both variants and voting algorithms can be implemented in either hardware or software. The behaviour of voting algorithms in multiple error scenarios within a Tiple Modular Redundancy (TMR) system is considered in this paper. Complete disagreement is defined as those cases where no two variant results are the same. A special experimental harness to test a range of voters is introduced and the testing method is explained. Several performance measures to evaluate the behaviour of voters are evaluated. The meaning and the degree of usefulness of these performance criteria from the viewpoint of application engineer is explained. Finally various presentation methods from which the strength and weakness points of each voter can be studied are suggested.