Data-driven predictive maintenance policy based on multi-objective optimization approaches for the component repairing problem
Ornella Pisacane, Domenico Potena, Sara Antomarioni, Maurizio Bevilacqua, Filippo Emanuele Ciarapica, Claudia Diamantini
Abstract
In systems with many components that are required to be constantly active, such as refineries, predicting the components that will break in a time interval after a stoppage may significantly increase their reliability. However, predicting the set of components to be repaired is a challenging task, especially when several conditions (e.g. breakage probability, repair time and cost) have to be considered simultaneously. A data-driven predictive maintenance policy is proposed for maximizing the system reliability and minimizing the maximum repair time, considering both budget and human resources constraints. Therefore, a data-driven algorithm is designed for extracting component breakage probabilities. Then, two bi-objective optimization approaches are proposed for determining the set of components to repair. The former is based on the formulation of a bi-objective mixed integer linear programming model solved through the AUGMEnted ε-CONstraint (AUGMECON) method. The latter implements a bi-objective large neighbourhood search, outperforming the first approach.