Notice of Retraction A novel weighted voting algorithm based on neural networks for fault-tolerant systems

Voting algorithms are used in a wide area of control systems from real-time and safety-critical control systems to pattern recognition, image processing and human organization systems in order to arbitrating among redundant results of processing in redundant hardware modules or software versions. From a point of view, voting algorithms can be categorized to agreement-based voters like plurality and majority or some voters which produce output regardless to agreement existence among the results of redundant variants. In some applications it is necessary to use second type voters including median and weighted average. Although both of median and weighted average voters are the choicest voters for highly available application, weighted average voting is often more trustable than median. Meanwhile median voter simply selects the mid-value of results; weighted average voter assigns weight to each input, based on their pre-determined priority or their differences, so that the share of more trustable inputs will increase rather than the inputs with low probable correctness. This paper introduces a novel weighted average voting algorithm based on neural networks that is capable of improving the rate of system reliability. Our experimental results showed that the neural weighted average voter has increases the reliability 116.63% in general and 309.82%, 130.27% and 9.37% respectively for large, medium and small errors in comparison with weighted average, and 73.87% in general and 160.44%, 83.59% and 7.52% respectively for large, medium and small errors in comparison with median voter.