[SIML] Visualization of Time Series Signals in a Web Application

Art der Arbeit:

Bachelor Thesis

Betreuer:

Melodia, Luciano
Lehrstuhl für Informatik 6 (Datenmanagement)
E-Mail: luciano.melodia@fau.de

Lenz, Richard
Lehrstuhl für Informatik 6 (Datenmanagement)
Telefon +49.9131.85.27899, E-Mail: richard.lenz@fau.de

Beschreibung der Arbeit:

Time series data are collected extensively in industry and often inspected by hand. However, currently market-ready tools do not implement the functionality needed for the case of demand. Such a demand case prevails in power plants. Power plants are monitored extensively due to safety and efficiency aspects. In particular, the thermodynamic processes in the turbines are monitored. Of great interest are periodicities in the behaviour of power plant turbines which occur naturally, although not immediately apparent, in the data. These are to be visualised in a suitable way for existing data in an application in order to provide the domain expert with the best possible insight into his time series.

Given:

• In about 120 time series with 70,000 measured values.

• The time series are provided with meta information, such as unit, type of measurement, interval and location of measurement.

Task:

A graphical tool for processing the data is to be built, consisting of three components. Your task is to implement at least one of these components.

The first component should be a web interface that visualises the location of the data. Furthermore, it should be possible to examine a certain signal. That means, the engineer shall get first information by elementary statistics like mean value and standard deviation etc. The engineer should be able to select a number of measurement variables which he can inspect. Secondly, the entire signal should be visualised. Finally, elementary tools for topological data analysis should be implemented, which reveal periodicities in the data.

A second component is a classification interface. The user shall have the possibility to load pre-trained models of neural networks and use them as a classifier for each individual signal. The corresponding signal identifier should be classified. The user should be able to choose from several models. For the corresponding label the user should be able to choose one of the machine generated proposals. Furthermore, manual input should be possible. The label provided should be saved appropriately.

The third component consists of an interface with which the user can specify and train models of neural networks. Particular attention has to be paid that training progress is displayed accordingly and that the user cannot make invalid, if nonsensical, specifications for the choice of hyper parameters.

Bearbeitungszustand:

Die Arbeit ist bereits abgeschlossen.