Abstract
To achieve the safety of intelligent systems for real-world applications, not only processor faults but also real-world errors such as sensor errors, failure of the controlled object and operation errors must be covered. From this point of view, an intelligent fault-tolerant system for real-world applications is proposed based on triple-modular redundancy. The system consists of a master processor that performs the actual control operations and two redundant modules. In the redundant modules, real-world process is simulated together with the control operations using knowledge-based inference strategy. To keep the independency between the triplicated modules, the simulation for error detection and recovery is performed without actual external sensor signals used in the master processor. It is made clear that the safety operation can be assured for a very wide ranges of faults and errors such as sensor errors, failure of the controlled object and operation errors.
