Bi-objective predictive maintenance optimization for aero-engines: Mathematical models and metaheuristic algorithms

This work presents two bi-objective predictive maintenance optimizations for aero-engines incorporating remaining useful life (RUL) prediction and maintenance scheduling. An effective hybrid deep learning model is first designed for assessing the aero-engines RUL. According to aero-engines estimated RUL, we develop two novel bi-objective mixed integer linear programming (MILP) models to address aero-engines predictive maintenance problem, aiming to simultaneously minimize the maximum maintenance completion time and total maintenance costs of all aero-engines. To address these two bi-objective problems, we first develop an iterative ϵ-constraint (IEC) and weighted-sum (WS) methods, which are exactly solved by translating the bi-objective MILP into many single-objective MILPs, thus losing computational efficiency in practical-scale instances. Meanwhile, we design a tailored non-dominated sorting genetic algorithm II with an embedded variable neighborhood search to obtain approximate optimal solutions for large-scale maintenance problems. Experimental results on the NASA aero-engine dataset demonstrate that the designed bi-objective predictive maintenance optimization methods can flexibly provide accurate RUL evaluation and effective maintenance scheduling to decrease maintenance time and costs.