Combining maximum likelihood estimation and LSTM neural network to forecast reliability distributions: a study based on real data from the sugar-energy sector

Abstract

The recent advances in reliability engineering and assets management allow industries to use advanced mathematical and statistical techniques to improve their maintenance strategies. Conversely, many of these methods requires the previous knowledge of the failure history for the assets, which are not always available. This study shows that is possible to support maintenance management decisions providing the distributions for assets with known and unknown failures history. The approach combines maximum likelihood estimation with a LSTM neural network to find and predict optimal reliability and failure distributions for data actual assets data and failures history. It is shown as an efficient and probably useful approach for practical business situations in the results section.