Neural network realization of Markov model of TMR systems with compensating failures

Reliability synthesis as a reverse of reliability analysis has the same importance as reliability estimation. One important issue in reliability synthesis is to find failure rates and repair rates of units, given the desired reliability requirement. Compensation of failures can be very significant for fail-fast TMR systems. This paper presents a new model of TMR systems in which fail-fast rates and compensating failures are considered, and employs a forward neural network to compute the failure rates and repair rates that meet the desired reliability requirement and the preassumed fail-fast rates. The failure rates and repair rates are obtained after the neural network becomes stable. Simulation results show that the model presented in this paper can allow higher failure rates of units to achieve the same reliability than the classical model. So we can design a redundant digital system with a much lower cost