Optimal Part Flow in Maintenance Service Contracts of Gas Turbines

Abstract

IGT manufacturers offer maintenance service contracts that guarantee specific production rates. At every maintenance action (i.e., corrective or scheduled), capital parts are removed from the GTs and repaired at workshop, unless they are scrapped because they have reached their pre-fixed maximum number of working hours. The repaired parts are put back at the warehouse for future use. The parts removed from the GTs are replaced by parts newly purchased or taken from the warehouse. This maintenance policy entails a part flow, which is managed through decisions on both the removed parts (repair or scrap?) and the parts to be installed on the GT (parts new or taken from the warehouse?). Such decisions strongly impact the profitability of the maintenance service contract and depend on many variables, e.g. remaining time up to the end of the contract, availability of spares, costs related to the repair and purchase actions, etc. Furthermore, in the dynamic of the part flow, every decision conditions the successive ones, as it modifies the warehouse composition. We formalize the part flow problem as a Sequential Decision Problem and solve it by both integer linear programming framework and reinforcement learning, taking as reference a scaled-down case study derived from industrial practice. Final considerations are drawn about both approaches.