Sprinkler systems allow a considerable reduction of fire risk in buildings. Unfortunately, sprinklers are not fail-safe technical systems. Relatively high rates of sprinkler failures evoke the problem of reliability. A solution to this problem is considered from several viewpoints. The diversity of sprinklers' failure modes is the first challenge for estimating reliability (failure probability). It is found that the use of the available data for estimation is problematic. The second challenge is that the published data is insufficiently described to allow a verification of its relevance to the specific case of failure probability estimation. It is suggested to apply the published data with partial relevance to Bayesian inference about failure probabilities. The data is used for developing prior distributions of the unknown values of the probabilities. Bayesian inference is carried out on the basis of binomial distribution used to model the operation of sprinklers on demand basis. A problem of aging and a possible increase in failure probability in the course of sprinkler service is shortly discussed. 

Article in English