RAPIDS - Reliable Adaptive Predictive maintenance and Intelligent Decision Support

Purpose and goal:

The project revolves around developing machine learning models based on increased availability of streamed log data from vehicles and integration of these models in the decision-making processes for maintenance. It specifically deals with how uncertainty in predictions can be estimated and weighed in order to make robust individual-based decisions. Central is also how new data can be fed back to the models in order to improve performance and predictive power over time.

Expected results and effects:

The project will help to strengthen the research fields of machine learning and forecasting. The project contributes with findings for how uncertainty in predictions should be estimated and weighed in order to make robust and individual-based decisions and how new information can be fed back to prediction models to improve performance and predictability over time.

Methodology for how to optimally use streaming monitoring signals for maintenance planning contributes to better interaction between forecast models on different time scales and on board and in the cloud. Approach and implementation:

The project develops theory and generally applicable methods which are then tested and demonstrated on real and interesting user cases. The intention is to have an iterative and incremental work process. The project consists of 5 work packages (WP).

WP 1 performs administration and management of the project. WP 2 and 3 focus on predictive models for forecasting based on streamed data and on how uncertainty information can be quantified and weighed in for decisions. WP 4 and 5 work with efficient handling and feedback of data.