Data augmentation of time series data

Type of thesis: Bachelor- und Masterarbeit / location: Leipzig / Status of thesis: Open theses

Data augmentation is a method of artificially creating new data from an existing data set to increase the total amount of data. The technique is used as a preparatory step in the field of machine learning. In the experimental field, ultrasonic data were recorded during drilling in metal. The data was kindly provided by the Institute for Microelectronic and Mechatronic Systems in Ilmenau. Ultrasonic testing is a common quality assurance technique in the metal industry to investigate structural integrity and potential defects. The ultrasonic data is used to train a classifier for the early detection of drill a bit of wear.

This bachelor/master thesis focuses on the application and evaluation of data augmentation methods to augment real recorded ultrasonic data in metal drilling. The use of machine learning algorithms to analyze such data usually requires a large amount of annotated examples. The main objective of this work is to extend the dataset using data augmentation techniques to improve the performance of machine learning models for defect detection.

The task of this work includes the following steps:

In this work, the existing data will be artificially augmented. For this purpose, the state of the kind of data augmentation methods has to be described in a comprehensive literature review. At least two suitable methods will be applied to the existing data. The classifier will be trained on the artificially augmented data, and the goodness of the classifier will be evaluated with an appropriate metric. The performed steps, experiments, and results have to be documented.


Dr. Thomas Burghardt

Leipzig University

Service and Transfer Center, Living Lab