Optimising the reliability that can be claimed for a software-based system based on failure-free tests of its components

TL;DR

This paper introduces a numerical linear programming method to optimize the confidence bounds on system reliability based on component tests, offering a more efficient alternative to previous integer programming approaches.

Contribution

It presents a novel linear programming-based optimization technique for reliability testing that is more efficient and reusable across different testing budgets.

Findings

Linear programming outperforms integer programming in efficiency.

The method can derive optimal test plans for arbitrary system structures.

The solution can be reused for different testing constraints.

Abstract

This short paper describes a numerical method for optimising the conservative confidence bound on the reliability of a system based on tests of its individual components. This is an alternative to the algorithmic approaches identified in Bishop and Povyakalo (RESS 2020). For a given maximum number of component tests, the numerical method can derive an optimal test plan for any arbitrary system structure. The optimisation method is based on linear programming which is more efficient that the alternative integer programming. In addition, the optimisation process need only be performed once for any given system structure as the solution can be re-used to compute an optimal integer test plan for a different maximum number of component tests. This approach might have broader application to other optimisation problems that are normally implemented using integer programming methods.