Machine Learning for Model-based Diagnosis of Cyberphysical systems
Type: Bachelor CS
If you are interested please contact :
Model-based reasoning uses experts knowledge of a problem domain to build diagnostics models. Based on a description or design of a piece of equipement (an ASML lithographic machine, printer, etc.), a compositional model is constructed with the aim to simulate the behaviour of a system in order to diagnose faults in the actual system. By faults we mean that particular components of the physical or cyberphysical system are no longer properly working. There is an increasing interest of using model-based reasoning by companies, such as ASML and CANON.
Despite being quite a powerful techniques, a big challenge is into the translation of measurements that are continuous timeseries into abstract domain. Current techniques use thresholds to define different regions of utilization. However, for complex systems, such a technique fails to capture all the utilization states. For example, a heater usually functions based on thermal cycles. The thermal cycles are not periodic due to influence of the usage of the system on the temperature. Furthermore, when an anomaly occurs there is a time delay between the moment that the anomaly occurred and the moment that the thresholds is met. One idea is to use machine learning techniques to allow us to automatically define the utilization states and, to label a given period of time based on these suggested states. The technique will be applied to a data set of more than 100 sensor data from a complex cyberphysical system. This project is in collaboration with ESI/TNO.