Half way seminar Constantin Cronrath

​Reinforcement learning through global off-line optimization and smart local on-line exploration

Join the seminar from PC, Mac, Linux, iOS or Android via Zoom​
The seminar can be accessed through Zoom, and it will open shortly before 14:00. We would kindly ask you to keep the video off and mute the microphone during the seminar. At the end of the session there will be an opportunity to ask questions through Zoom. In case there will be any updates about the event, these will be posted on this website.

Constantin Cronrath is a PhD student in the research group Automation, Division of Systems and Control
Examiner is Professor Bengt Lennartson

The emergence of the internet of things, big data, and cloud computing has given rise to the concept of the digital twin in manufacturing. The digital twin concept essentially refers to an ultra-realistic digital model of a product or system, coupled to it by a bidirectional automated data exchange, used for simulation, optimization, and control. Digital twins are regarded as enablers of smart and autonomous manufacturing systems. When used for optimization and control of the manufacturing system, the high-dimensionality and black-box characteristic of the digital twin are a challenge. Furthermore, slight model or data uncertainties will remain, although digital twins strive to represent their physical counterpart as accurately as possible. However, the digital twin may be used offline for global optimization. Under the assumption of sufficiently small uncertainties, only a local re-optimization is then required on-line. We are thus interested in smart exploration methods capable of searching efficiently (in terms of data samples) for the true local optima of the physical counterpart. An algorithm is presented to compensate for those residual uncertainties through reinforcement learning and data fed back from the manufacturing system. When learning, the digital twin acts as teacher and safety policy to ensure minimal performance. In addition, simultaneous perturbations stochastic approximation is shown to be efficient in exploring the static system locally.  We test the algorithm in a sheet metal assembly context, in which locators of the fixture are optimally adjusted for individual assemblies.   

Category Seminar
Location: online
Starts: 25 February, 2021, 14:00
Ends: 25 February, 2021, 15:00

Page manager Published: Mon 15 Feb 2021.