Testing of safety systems is essential to ensure that their probability of failure on demand (PFD) is maintained below acceptable limits throughout their lifetime. It is commonly assumed that more frequent testing and a shorter interval between tests imply a lower PFD, because test results provide information about the functional status of components. Nevertheless, it is also well known that tests may induce stresses in the tested components. In fact, some tests may induce levels of stress that over time will accumulate and lead to significant damage to the tested components. This means that reducing the intervals between tests may actually increase the average PFD of the safety system within a large operational cycle. The traditional models for evaluation of PFD with constant failure rates do not capture the damaging effect of the tests.
One case of interest is the requirement by the US Bureau of Safety and Environmental Enforcement (BSEE) that pressure tests should be performed on some blowout preventer (BOP) components every two weeks, a requirement that is also followed by regulators in several other countries. Considering the damage induced by the tests, is this optimal, or are companies testing their BOPs too much?
In this paper, we derive optimal testing policies under two different scheduling schemes: (i) periodic testing with equal intervals between tests, and; (ii) adaptive scheduling, where test intervals are allowed to vary throughout the operational cycle. We show that adaptive test scheduling with long initial test intervals that gradually decrease as the component accumulates damage give a lower average PFD than a periodic testing with constant intervals for the same total number of test. We observe that, for constant interval testing, the impact of tests may outweigh the value of the information the tests provide, and depending on the mechanism of damage, there may exist a maximum number of tests after which further tests will increase the PFD. However, using the optimal adaptive test schedule, the average PFD always decreases when more tests are added. Our results indicate that equal test intervals and frequent testing are generally not optimal, and that better reliability can be achieved if the degradation mechanisms are understood and accounted for in test planning.