Scheduling of Aircraft Heavy Maintenance

Aircraft require regular inspection and maintenance, based on calendar age, hours flown, and cycles of takeoff-landing. Engines, landing gear, and other major systems have limits on their life in service before overhaul. There is a stream of modifications coming from the manufacturers. All this work needs to be packaged up into a series of maintenance checks, cycling through work that can be done in a single shift without a hangar, to a work package requiring several weeks in a hangar. Efficient packaging and scheduling of this work plays an important part in maximising the availability of the fleet for flying.

The planner is faced with a mixture of work which can be planned well in advance, changing out the landing gear, and work which needs to be squeezed in as soon as possible, the manufacturer has advised of the wrong type of O-ring in a pump. A tool to assist in preparing schedules has to be fast, efficient in its long range planning, flexible to handle the ad hoc work, and complete in its handling of all of the work to be scheduled.

As much as possible of the workings of the scheduler should be able to be modified by the planners, rather than computer programmers. The inputs, operation, and outputs of the scheduler need to be auditable, to ensure that no airworthiness regulations are breached. The more transparent and intuitive the method used, the more confidence and reliance planners will have in using it. A "black box" approach to aircraft maintenance scheduling would have problems of modifiability, auditability and reliability. OASIS is not just a "white box", where the planners can see the workings inside. To a large extent, the planners build the workings of the scheduling machine, and assume responsibility for the quality of the schedules produced.

OASIS is built on top of Orion, a knowledge-based system, and uses a knowledge network to create and schedule Limit based maintenance events as well as independent events. The OASIS scheduler will attempt to place the Limit based events as close to their limits as possible whilst ensuring that they do not use more than the available resources at any point in time. Hints within the network will attempt to book resources initially back from the hard limit by a planning pad, but eat up the planning pad if there is no alternative. The resources being booked can be Facilities such as Hangars or Human resources, or consumable spares, perhaps with refurbishment requirements like undercarriage, with the refurbishment process built into the model.

Representation Of Multiple Bases

Where a maintenance check can occur at several bases, the resource availability can be pooled, or by using a positive to negative range for Resource Intensity, bases can trade availability among themselves, the calculations for doing so all embedded in the network.

Swallowing Of Activities

A larger check, brought forward, should swallow any minor checks within range. This is easily accomplished by making the duration of the smaller check include zero as a possibility, and if a larger check comes close enough, forcing the duration of the smaller check to zero. Additional modelling allows some particular work in the smaller check to move to the larger one, while other work, which would also occur in the larger check, disappears.

Piggybacking Of Events

You will sometimes want to attach some packet of work to one of the checks that must occur on the plane, but may be unsure as to which check it should be. The starting dates of the checks are tentative, and there may be constraints on the event to be piggybacked which conflict with parts of those ranges. You can build logic into the network so the appropriate check will be chosen after the starting dates are known to more accuracy.

The Knowledge Modelling Approach

Modelling the knowledge about the application directly leads to the construction of the "scheduling machine". Most of the logic in the model is generated by planners, as they describe the various checks to be carried out, in terms of activities and resource use and interrelation with other checks. This information is combined with plane histories and expected flying time to create the network model.