The optimization of maintenance schedules and spare inventory for industrial components has been traditionally studied separately. Maintenance models assume that spare parts are always available in stock and so make decisions based solely on lifetime distributions. Spare inventory models, on the other hand, usually rely on time-series methodologies to predict the demand for spare parts without considering the impact on maintenance schedule. Nevertheless, optimizing these two problems in isolation or in a sequential fashion can lead to sub-optimal solutions due to the tight coupling between spare inventory and maintenance decisions. This research line seeks to formulate new joint optimization problems to coordinate condition-based maintenance and spares provisioning while using insights from degradation data. Focus applications have been in wind farms and deep space habitats. 

It is estimated that Operations and Maintenance (O&M) activities account for up to 30% of the running costs of wind farm operations. Condition-based maintenance (CBM) has received significant attention for its potential to reduce the operational costs of wind farms. CBM leverages condition monitoring data from sensors to optimize maintenance activities. These strategies combine predictive analytics and optimization methods to adapt maintenance decisions according to the observed degradation of wind turbines. Existing CBM strategies for wind farms rely on the assumption that predictive analytics can accurately estimate the remaining lifetime distribution (RLD) of wind turbines, allowing for the direct implementation of stochastic programming or threshold-based policies. However, estimated RLDs can be inaccurate due to noisy sensors or limited training data. In this line of work, we investigate how inaccurate estimated RLDs can be integrated into optimization models to coordinate the maintenance schedule of wind turbines.

This research line focuses on the use of Machine Learning (ML) techniques for fault diagnostic and failure prognostic purposes, specifically in the domain of batteries and intermittent faults. We aim to develop advanced ML algorithms that can effectively analyze battery data to detect early signs of degradation, identify faults, and accurately predict the remaining useful life of batteries. Additionally, we will focus on addressing the challenges posed by intermittent faults, which are often difficult to detect and diagnose. By leveraging ML techniques, we aim to capture subtle patterns and variations in intermittent fault data to enhance fault detection and diagnosis accuracy.