In a series of previous I/ITSEC papers, an emerging technological approach to artificial intelligence was introduced and applications to supervised learning and unsupervised learning were demonstrated.

This paper demonstrates how this emerging technology’s model-based reinforcement learning capability is used to choose the optimal aviation maintenance policy when considering both immediate and subsequent costs.  At a high level, the goal is to show how prescriptive analytics can aid in decision-making for aviation maintenance.  The architecture may be any mathematical and/or logical instantiation and does not require neural networks – although neural networks are wholly realizable within this framework.  A benefit of this approach is a fully transparent and explainable model which allows us to open the “black box” of mainstream AI modes.

Aircraft states are defined according to equipment operational capability (EOC) codes which classify the degradation of equipment mission capability. Based on the EOC codes, actions are selected from a set of possible choices.  The consequence of an action is realized immediately as a nonlinear availability or financial cost.  Each decision determines the transitional probabilities for the next aircraft state.  Each action prescribes a policy for stochastic decision process – which impose constraints on the model.  The optimal policy is realized by minimizing the expected average cost in the long run.

The paper also explains how this high-level model may be extended to be a more effective decision-making/prescriptive analytics tool for aviation maintainers and fleet management.  Aircraft states could be expanded by individual EOC code for higher fidelity and cost metrics could be obtained to address airframe-centric maintenance requirements.  Further extensions include capital equipment in general, whether military, industrial, or commercial.