Fault Prediction of Lifting Systems on Offshore Drilling Platforms Through a Random Forest Approach

The lifting system is one of the key components of many offshore drilling platforms. Due to the severity of operating environment, occurrence rates of various failures related to the lifting system are relatively high; meanwhile, because most drilling platforms are far away from any continents, the vacant cost out of repairing is unbearable. In this manner, the solution of applying spare parts on premises is widely adopted; however, the capacity of spare parts is subject to the considerable manufacturing cost and maintenance overhead. Therefore, how to predict the number of failures for the next cycle based on fault history is a key challenge for improving the efficiency of the failure recovery and reducing the vacant time of drilling platforms. Based on the random forest model, we train a regression model according to an annual history dataset of inverter and gearbox failures of several lifting systems. Simulative experiments show that this model can predict various failures with attractive performance, assist platform operators in making spare parts decisions and avoid potential loss of idling offshore drilling platforms.