Time-trend analysis of offshore fire incidents using nonhomogeneous Poisson process through Bayesian inference

Abstract

Incident trend analysis has been important in the past for understanding how a system has been performing over time. The system can refer to a particular equipment, facility or organization and its performance can be monitored in terms of numbers or rates of failures, incidents or events over time. A good trend analysis leads to better projections into the future, enabling a more accurate prediction of future incidents or failures. In most cases however, it is generally assumed that incident or failure rates remain constant over time and the same value of the rate is used in all estimations. This study uses data of past offshore fire incidents in the Gulf of Mexico to predict future incidents and shows that such an assumption can fail to provide accurate predictions. The data is normalized to account for the year-to-year variation in operation and shows how using a nonhomogeneous Poisson process (NHPP) assumption, where failure rate is a function of time, enables a better understanding of performance, and can be used to predict future incidents more accurately. This will help regulatory bodies to understand whether operation in the Gulf of Mexico has been improving or not and to take proactive measures before the next fire incident occurs.