Outline

Today’s powertrain control development relies on traditional map-based and model-based control approaches. Due to growing system complexity and real-world performance requirements, these expert-intensive and time-consuming approaches will lead to challenges of unacceptable development time and costs. Consequently, automotive control development is facing a turning point in the near future.
Machine Learning (ML) is a disruptive technology, which offers powerful features to address these challenges. To significantly reduce the time, cost, and effort required for powertrain control calibration, ML methods offer opportunities to i) create dynamic models needed in predictive control approaches such as model predictive control (MPC), ii) design of supervisory learning-based controllers, iii) efficiently parametrize control models, and iv) design real-world powertrain load cycles for automated testing in the laboratory.
Utilizing ML along with cloud computing and vehicle to infrastructure (V2I) communications enables even further reductions as well as improved real-world performance. For example, fleet data can be used to enhance vehicle individual as well as fleet performance. Also, cloud-based system architectures open opportunities to solve complex and computational demanding optimization problems.
Currently, limited results are published on ML-based methods for automotive powertrain and vehicle applications. There is clearly a need to demonstrate the potential, limitations and challenges of these promising methods. This will concretize the anticipated ML benefits and accelerate industry adaption of Machine Learning. We target high quality publications that i) introduce new ML-based powertrain applications; ii) clearly specify the performance benefits and impact on development effort; and iii) deal with implementation challenges in real-world.