Truck freight is one of three major resource transportation methods used around the world. The failure of a truck carrying critical cargo an bottleneck processes and cause a signi ficant follow-on fi nancial impact. The cost of field failures is exacerbated when the failure occurs in a remote location. Maintenance equipment must be transported to site, or the vehicle itself must be towed back to a servicing facility. The aim of this project is to optimise the preventive maintenance schedule for a fleet of freight vehicles that operate in remote environments through simulations and genetic algorithms
(GA). The project was conducted in collaboration with an industry partner who owns a fleet of fi ve prime movers, which conduct journeys between 30 km and 1,300 km long in remote locations.
The fleet experiences approximately one fi eld failure each month. A `one-size- ts-all’ solution is currently employed for preventive maintenance scheduling, where a common maintenance strategy is
recommended by manufacturers for all prime mover makes and models. A simulation is developed to model the maintenance constraints and operation logistics of a fleet of prime movers owned by an industry project collaborator. Maintenance data of the fleet is analysed to identify reliability distributions for components in the vehicles. These distributions are used to model stochastic failures in the simulation. Multiple tests using the simulation and genetic algorithm successfully converged to the same, optimal maintenance schedule. This supported the viability of using simulations and GAs as an optimisation technique.
Given its success in the case study, the optimisation technique was tested in a variety of operational contexts to investigate whether it could also be used for trucks travelling different journey lengths or
other types of freight. Multiple maintenance models were also tested. The GA optimised the schedule for these different test cases. Comparative trials were also conducted to compare the performance of the GA against hill climbing and random search algorithms. The GA converged in a smaller number of trials than the other algorithms and the top scoring schedule it found had a fitness score 21.6% higher
than those found by other methods. The results of these tests were used to develop recommendations for fleet owners intending to apply this technique in the future.