DEPENDABLE, INTELLIGENT VOTING FOR REAL-TIME CONTROL SOFTWARE

An intelligent and dependable voting mechanism for use in real-time co ntrol applications is presented. Strategies proposed by current safety standards advocate N-version software to minimize the effects of unde tected software design faults (bugs). This requires diversify in desig n but presents a problem in that truly diverse code produces diverse r esults; that is, differences in output values, timeliness and reliabil ity. Reaching a consensus requires an intelligent voter, especially wh en non-stop operation is demanded, e.g. in aerospace applications. Thi s paper, therefore, firstly considers the applicable safety standards and the requirements for an intelligent voter service. The use of repl icated voters to improve reliability is examined and a mechanism to en sure non-stop operation is presented. The formal mathematical analysis used to verify the crucial behavioural properties of the voting servi ce design is detailed. Finally, the use of neural nets and genetic alg orithms to create N-version redundant voters, is considered.