Reliability estimation from appropriate testing of plant protection software

Plant protection software may be realistically tested using inputs from a plant model before its initial use, or when it is not feasible to take the plant into certain fault conditions. If statistical estimation of software reliability is to be performed using the test results, it is not sufficient for the plant model to produce inputs which are simply correct in the sense that the plant could have produced them. In addition, the operational distribution of the input space must be simulated. This paper illustrates how to perform such a simulation, by developing an example in which an existing non-random plant model is randomised to simulate the operational distribution of the software. In addition, two methods of estimating the probability of failure on demand (PFD) for a program are reported. Both methods estimate a pfd given results from dynamic testing, during which the program is exercised according to its operational distribution. The first method is standard and has been used previously in the context of software testing. The second estimation method has been developed recently within a program of Nuclear Electric for research into software reliability testing. The distinguishing foundational assumptions of the two methods are discussed