Ensemble Learning for Predictive Maintenance
|Högskolan i Halmstad - CAISR - Center for Applied Intelligent Systems Research
|Funding from Vinnova
|SEK 115 834
|November 2018 - February 2019
|Personal mobility between societal sectors
|Funding for staff exchange and artificial intelligence (AI)
Purpose and goal
The work has resulted in improved models for predictive maintenance, implemented through integration of several different techniques and using multiple data sources. The ensemble model created in this project, which combines classifiers from different data domains, leads to improved accuracy and recall for several different components. Combining the complementary features allows us to decompose a complex, difficult learning problem into easier sub-problems.
Expected results and effects
The existing collaboration between Halmstad University and Volvo Group, in particular Volvo Trucks Aftermarket, has been strengthened significantly through this mobility grant. The knowledge exchange in both directions has lead to improved technical solutions which now allow for more precise estimation of the failure risk of various components.
Planned approach and implementation
The project consisted of efforts attempting to answer two research questions. First, what are the constraints and challenges of integrating ensemble approach into existing model? Second, how diverse are individual models based on each data domain, and how could we improve this diversity among models? Both questions have proven valuable and provided important insights into the final developed solution.