Masters Degrees (Industrial Engineering)

Permanent URI for this collection

https://scholar.sun.ac.za/handle/10019.1/776

Browse

Browsing Masters Degrees (Industrial Engineering) by browse.metadata.advisor "Braun, Prof. Dr.-Ing. Anja"

Now showing 1 - 1 of 1
  • Thumbnail Image
    Item
    Development of a procedure model to forecast machine health based on long-term power consumption data
    (Stellenbosch : Stellenbosch University, 2022-08-31) Holzapfel, Tim; Von Leipzig, Konrad; Braun, Prof. Dr.-Ing. Anja; Stellenbosch University. Faculty of Engineering. Dept. of Industrial Engineering.
    ENGLISH ABSTRACT: Energy consumption data is already being studied for a variety of analyses. Still, there are fields of application in which energy consumption analysis is hardly used. One such application is the degradation of a machine. In this research work, a procedure model is developed based on a use case to infer the health of the machine. Through wear and tear of the machine, energy gets lost for instance in the form of friction or heat. This energy loss must be additionally supplied to the machine by electrical energy. The degradation of the machine can thus be represented by long-term trend modelling of the power consumption. A process model describes the individual steps of the data analysis and provides information about the necessary inputs and the output of the model. In the second step, the process model is implemented in a backend software segment. The software segment includes pre-processing of the data, analysis of the data, real-time data processing as well as visual presentation of the results, and alerting in case of critical conditions. The analysis of the data includes trend modelling on the one hand and predicting the power consumption using machine learning approaches on the other. The verification of the trend model yields logical results. Due to the short period of the data basis, the validity of the results is limited. For the prediction of power consumption, the quality of the forecast was checked using error metrics. The best result was achieved by the random forest regressor with a coefficient of determination of 0.79.