Concepts, Methods, and Systems for Machine Data Analysis

Christopher Josef Rothschedl

Research output: ThesisDoctoral Thesis

79 Downloads (Pure)


This thesis is concerned with the means of acquiring data from cyber physical systems, including the required infrastructure, and the use of both classical and novel approaches to derive insights from time series data. The systems discussed are associated with industries which require the establishment of complete data management frameworks to integrate essential domain expertise to gain knowledge. In a generic manner, the development of a holistic, secure and flexible concept for data life cycle management is presented, for heterogeneous fleets of mobile machines. Apart from providing robustness, such a concept must be capable of adopting changes, such as extensions with subsystems or replacements of components resulting from obsolescence of technology. After a study of the fundamental requirements for hardware, machine interfaces, data handling and storage, as well as data provisioning, an implementation is shown for different machines used in the mining and materials handling industry. To take the insights from these implementation scenarios into account, the concept for a new system for machines of the geotechnical engineering sector is developed and implemented. The framework for a qualitative data flow is presented, which allows experts to interact with the data of their machines; a step necessary to create added value from time series. It consists of multiple levels of data preparation and presentation methods, to identify elements of interest. Outliers can be highlighted for further investigations, based on rules applied to key performance indicators. Furthermore, it is shown that true knowledge discovery can be supported significantly by mimicking the mechanisms of the emergence of natural language. This step takes the specific nature of the data into account, since the time series emanate from physical systems and, hence, must abide by the laws of physics. An exemplary evaluation performed in this manner reveals implicit hierarchical structure in the operational data. Only an initial set of language elements are defined as input for a subsequent iterative process. A hierarchy of compounded frequent elements is yielded, the top layer of which reveals the existence of two major sequences that correlate with the two main operation modes. It is shown that the interpretation of the results by domain experts is indispensable for knowledge gain. This is emphasised by the metaphorical capacity exhibited by language-affine evaluation approaches, which are discussed in detail. A model for the emergence of language, based on phenomenological aspects, is proposed to combine the factors of relevance for knowledge discovery.
Translated title of the contributionKonzepte, Methoden und Systeme für die Analyse von Maschinendaten
Original languageEnglish
Awarding Institution
  • Montanuniversität
  • Lee, Peter, Assessor B (external)
  • O'Leary, Paul, Assessor A (internal)
Publication statusPublished - 2020

Bibliographical note

embargoed until null


  • Cyber Physical Systems
  • Mining and Materials Handling
  • Geotechnical Engineering
  • Domain Expertise
  • Data Science
  • Knowledge Discovery
  • Symbolic Time Series Analysis
  • Natural Language
  • Hierarchical Structure
  • Implicit Structure
  • Phenomenology

Cite this