Abstract
Long field-life systems are those for which operational, as opposed to capital, expenditures dominate the life time cost of the system. In established product categories, operational experience and historical maintenance data allows engineers to make informed decisions influencing maintenance during the early design stages when design changes are at their cheapest and few costs have been committed. Design engineers of novel products lack the same maintenance assessment resources afforded to designers working on established products. There may be little to no operational experience and historical maintenance data may not exist for components and assemblies used in novel designs. Prior research has shown that knowledge of functions can assist in making informed design decision early in the process. This research examines the circumstances under which functions can be used to predict the maintenance intervals of long field-life systems. A model to predict maintenance intervals was built using machine learning over maintenance data mined from an established engineered system, a civil aviation jet and its engines. Through functional analysis, the essential functions of the system's parts were determined and associated with the parts' maintenance intervals. A supervised machine learning model was used to learn the required maintenance, resulting in a function-maintenance model. This function-maintenance model was evaluated in three ways. First, the model was calibrated using classical machine learning metrics. Then, through grounding by applying the model to a large design repository containing known devices and their associated functions. Finally, to evaluate the model's application to novel systems, it was tested on a prototype renewable energy system and compared against the assessment of an expert designer with experience in the field. Through these evaluations, I conclude that there is an identifiable relationship between functionality and maintenance intervals. Furthermore, this relationship can be leveraged to inform early design decisions influencing the maintenance of novel systems. In developing a model to assess maintenance intervals of possible designs and the understanding to inform design decisions, this research contributes to methods to reduce design costs through earlier decision making and lifetime costs through improved system maintainability.