In this paper, we present a novel maintenance concept based on condition monitoring and dynamic maintenance packaging, by showing how to connect the information flow from low-level sensors to high-level operations and planning under uncertainty. Today, condition-based maintenance systems are focused on data collection and custom-made rule based systems for data analysis. In many cases, the focus is on measuring "everything" without considering how to use the measurements. In addition, the measurements are often noisy and the future is unpredictable which adds a lot of uncertainty. As a consequence, maintenance is often planned in advance and not replanned when new condition data is available. This often reduces the benefits of condition monitoring. The concept is based on the combination of robust, dynamically adapted maintenance optimization and statistical data analysis where the uncertainty is considered. This approach ties together low-level data acquisition and high-level planning and optimization. The concept has been illustrated in a context of rail vehicle maintenance, where measurements of brake pad and pantograph contact strip wear is used to predict the near future condition, and plan the maintenance activities.