direkt zum Inhalt springen

direkt zum Hauptnavigationsmenü

Sie sind hier

TU Berlin

Inhalt des Dokuments

MA: Application of Neural Networks in Unsupervised Anomaly Detection from Time and Frequency Data of Rotating Machines

Rotating machines are ubiquitous in a variety of mechanical systems which are used for many applications. The rotating elements have characteristic periodic oscillations which can be measured by vibration sensors. These vibrations originate from e.g. electric engines or bearings. Changes in the characteristic vibration of rotating machines can mean that a failure of the rotating machine occurred or will occur in near future, these can include all parts of the machine which affect the vibration. An early detection of such failures is of urgent importance for many industries as this can reduce the downtime of the machine, resulting in faster repair times and less monetary loss.
Recent advances in neural networks and generative models enables the learning of patterns in a variety of data. Thus, it is possible to use neural networks for unsupervised anomaly detection as divergent patterns can be recognized. The Fast Fourier Transformation (FFT) is able to convert the signal of such sensors into the frequency domain, where the intensity of frequencies composing the original signal can be seen. The combination of vibration and FFT data can be used to learn the pattern of such rotating machines utilizing neural networks.
In this thesis, the applicability of neural networks on time and frequency data of rotating machines shall be examined. Therefore, several datasets of normal and abnormal data are present. An extensive evaluation of the approach developed during the thesis shall be conducted.
If this sounds interesting to you, please send me an email with a little bit of background information on yourself.

Requirements:
Good Knowledge in Programming, Machine Learning
Of Advantage: Knowledge and Experience with neural networks as well as basic knowledge of physics 

Zusatzinformationen / Extras

Quick Access:

Schnellnavigation zur Seite über Nummerneingabe

Auxiliary Functions

Contact

Kevin Styp-Rekowski
+49 30 314-27494
Room TEL 1210