Reliability of sprinkler systems: Exploration and analysis of data from nuclear and non-nuclear installations

Sprinkler systems are an important part of fire safety of nuclear installations. As a part of effort to make fire-PSA of our utilities more quantitative a literature survey from open sources worldwide of available reliability data on sprinkler systems was carried out. Since the result of the survey was rather poor quantitatively, it was decided to mine available original Finnish nuclear and non-nuclear data, since nuclear power plants present a rather small device population. Sprinklers are becoming a key element for the fire safety in modern, open non-nuclear buildings. Therefore, the study included both nuclear power plants and non-nuclear buildings protected by sprinkler installations. Data needed for estimating of reliability of sprinkler systems were collected from available sources in Finnish nuclear and non-nuclear installations. Population sizes on sprinkler system installations and components therein as well as covered floor areas were counted individually from Finnish nuclear power plants. From non-nuclear installations corresponding data were estimated by counting relevant things from drawings of 102 buildings, and plotting from that sample needed probability distributions. The total populations of sprinkler systems and components were compiled based on available direct data and these distributions. From nuclear power plants electronic maintenance reports were obtained, observed failures and other reliability relevant data were selected, classified according to failure severity, and stored on spreadsheets for further analysis. A short summary of failures was made, which was hampered by a small sample size. From non-nuclear buildings inspection statistics from years 1985-1997 were surveyed, and observed failures were classified and stored on spreadsheets. Finally, a reliability model is proposed based on earlier formal work, and failure frequencies obtained by preliminary data analysis of this work. For a model utilising available information in the non-nuclear data body, it has to be analysed more comprehensively, than was possible in these studies.