We study, in particular, the optimization of maintenance operations in the offshore wind turbine industry and propose an extension of \cite{yildirim2017} to the case of mixed sources of condition monitoring information. However, although researchers recently made significant progress in POMDP solvers, enabling them to solve more realistic problems with larger state and action spaces \cite{papakonstantinou2014}, the obtained large-scale POMDP remains way too large to be solved efficiently. Such complexity is due to the combinatorial explosion in the state and action spaces when having multiple units in the system. Our main contribution is to propose a novel and efficient hybrid heuristic, mixing dynamic programming (DP) and integer linear programming (ILP), to overcome the curse of dimensionality. We also analyze the impact of imperfect monitoring on optimal maintenance policies, extending a previous work \cite{roux-esrel2022}.