Fault-tolerant control using transputers

Considers parallel processing as a means of introducing fault tolerance into a real-time design, not for the purpose of increasing the overall response time of the system. A novel method of error detection and recovery is introduced to increase the reliability of the system. The complete system is implemented on a small network of parallel processing devices (transputers). All code is written in occam. The fault tolerant structure described in this paper uses both voting and acceptance tests to mask and detect errors in the system and uses reconfiguration (software) to provide error recovery and thus an increase in the reliability of the system. By using a network of transputers this error detection and recovery can be achieved in real-time. It is shown that transputers and occam have many features that are not only suitable for the construction of safety critical systems, but actually improve the design by the constraints they put on the designer. The synchronous communications and the occam model of parallelism help to improve the reasoning about the systems operation, and thus the systems reliability. The techniques introduced in the paper are demonstrated by application to an inverted pendulum control problem. The results show that the techniques proposed in this paper provide an increase in system reliability, whilst still providing the required real-time response to control the system.