Joachim Grimstad ist derzeit Doktorand an der Universität Stuttgart, Institut für Automatisierungs- und Softwaretechnik. Er hat einen B.Eng. - Subsea Technology, Maintenance and Operations von der Western Norway University of Applied Sciences und einen M.Sc.Eng - RAMS (Reliability, Availability, Maintainability, Safety) von der Norwegian University of Science and Technology.
Seine Hauptforschungsinteressen liegen in den Bereichen Instandhaltung, Zuverlässigkeit und Sicherheit, insbesondere funktionale Sicherheit sowie Sicherheit und Zuverlässigkeit von sicherheitskritischen Systemen.
Forschungsschwerpunkt: - Risk Management and Adversarial Neural Networks
Beschreibung: - With the increase in popularity and adaptation of Cyber-Physical Systems (CPS), the role of control room operators has changed significantly. As systems become larger both in terms of extent and complexity and increasingly comprise elements from multiple domains, the requirements of the control room operators in turn have increased. History has shown that control room operators with insufficient knowledge of the system can pose a real threat to safe operation, this especially holds for uncommon operations or during the development of undesired events. Perhaps the two most well-known such incidents are the Three Mile Island accident (1979) and Chornobyl (1986), where operator errors exacerbated design faults and caused otherwise avoidable major accidents. In this research, the ambition is to contribute to a solution to this issue, by utilizing adversarial neural networks. The key concept is to have two neural networks competing for the safety and required functionality of physical models of the system. One neural network (red) attempts to influence the system in such a way that safety and functionality are compromised. The competing (blue) network attempts to alleviate any negative interactions and ensure the system is safe and, in a state where the required function can be performed, or at the very least ensure the system is ushered into a safe state. This research aims to glean knowledge that might otherwise have been overlooked. The red network may uncover combinations of faults, conditions, or actions that may lead to undesired incidents and demonstrate potential design faults. The blue network may offer operators support during operations, propose corrective actions, or suggest new and innovative operational procedures to ensure safe operation or bring the system to a safe state.