Smith, James R. (1987, ASQC) Tennessee Technological University, Cookeville,TN
The reliability of a system depends on the reliabilities of the components that comprise the system, along with the component configuration. Analytical models have been developed which permit computation of system reliability for some basic designs, usually assuming an exponential distribution of component lives. When either the system design or component life distributions are too complex for analytical models, there are two alternatives for estimating system reliability. One method is life testing the entire system. Often this method is too time consuming and expensive. The other alternative is simulation.This paper illustrates the use of simulation to compare alternative system designs with cost as the measure of effectiveness. A generic system design as far as the component functions is assumed. The question is how many redundancies for each function should be used to minimize overall cost. Given is the cost of system failure and the component cost. Also, given are the distributions of component lives.The basic system to be discussed has three functions that must be performed. Distributions of component lives are normal for function 1, exponential for function 2, and lognormal for function 3.A flow chart for the generic system design will be given. Generation of random variates from normal, exponential, and lognormal distributions will be discussed. For each alternative system design, a table of system reliability vs. time will be established by simulation. The reliability at each time value will be estimated by using 1000 replications. Given the reliability plot and the cost of failure for any given time, an expected cost is developed. From the expected costs for the different system designs, the optimum design is selected from the given set of alternatives.