Operational expenditures dominate the cost of long field-life systems, with maintenance comprising a significant proportion for many systems. However, engineers lack the tools to assess maintenance during conceptual design. Familiar systems mitigate this problem by providing historical maintenance data from which empirical models can be derived. Emerging technologies, like marine renewables, lack operational maintenance data. As a result, engineers must make decisions with no historical data and minimal, if any, operational experience. The high operations cost incurred from basic maintenance tasks, and the loss of energy production during maintenance, further highlight maintenance as a critical cost driver. This paper develops a data-driven model to estimate maintenance intervals of long field-life systems during conceptual design. The model links the elementary functions of a component to maintenance requirements. Relative maintenance considerations were determined by mining function and maintenance data from manuals of long field-life systems. Machine learning was applied to generate a function-maintenance model from the maintenance data. The model consisted of functions grouped into buckets of increasing maintenance demand. The machine learning model was applied to an exemplary long field-life system, a wave energy converter, to explore possible redesigns to reduce maintenance costs. This paper shows that maintenance costs, actions and intervals can be confidently accounted. The function-maintenance model offers two beneficial impacts: it reduces life cycle cost uncertainty, and allows engineers to make informed decisions during conceptual design when redesign costs are at their lowest.