Korte titel Opdracht
Learning Equipment Vibration Patterns for Predictive Maintenance
Computer Science, Mathematics, Physics
Affiniteit met/Wat verwacht Thales van de stagiair/afstudeerder?
We are looking for a student with a strong affinity with machine learning techniques. Particularly, affinity with statistical modeling techniques will be important for this assignment. Experience with programming languages such as Java, Python, C++, etc. and/or statistical software packages (e.g. Matlab) are considered a plus. Additionally, we expect from the student to have a high level of initiative, self-confidence and the ability to work independently.
Achtergrond van de afdeling, wie zijn wij?
Thales Research & Technology Netherlands is a research lab in Delft located near the TU Delft. At the research lab, decisions support systems are developed to support decision making in critical domains such as: predictive maintenance, crisis management, search and rescue, cyber security, mission management, etc. Research is focused on developing comprehensive methods and tools for supporting a wide range of decision-making problems.
Time-based preventive maintenance of in-service equipment is cost-ineffective, labor intensive and results in unnecessary process interruptions. Additionally, such a maintenance strategy cannot avoid problems that are developing between timed servicing tasks. Ideally, servicing of equipment should be performed when necessary such that maintenance costs, the number of planned stops and labor can be reduced and the availability of the equipment increased. Through periodic or continuous condition monitoring of equipment predictive maintenance tasks can be performed when it is most cost-effective and before system performance drops below a minimum.
One important technique for equipment monitoring is vibration analysis. In-service equipment components generate specific vibration patterns that reveal information about the state of the system. From the vibration patterns information can be deduced about fault models, remaining useful life and other information required to implement effective predictive maintenance strategies. The assignment is about learning probabilistic models to detect the possible fault modes of the system (diagnosis) and make predictions about the remaining useful life of components (prognostics). In order to investigate the performance of the learned model different probabilistic models will be investigated. The models are learned based on features extracted from raw vibration pattern data that will be collected from an accelerometer.
Thales Research & Technology NL
Patrick de Oude