Ph.D. Project: Prediction of the Cylinder Condition in Marine EnginesUsing Neural Networks

  • Sørensen, John Aasted (Project Manager)
  • Hansen, Lars Kai (Project Participant)
  • Fog, Torben L. (Project Participant)
  • Peteren, Peter Sunn (Project Participant)
  • Lautrup, Benny (Project Participant)

    Project Details


    Marine engine monitoring is an active research area with a long history.
    Successful monitoring is vital for marine traffic safety and significant economic factors can be involved e.g., in the form
    of transport delay costs and additional use of spare parts. At present, only quite simple electronic methods exists for monitoring the cylinder condition in marine engines. Certain mechanical systems have been constructed, although robust, they do not provide adequate information about the specific fault conditions. Development of new and better methods for signal analysis in fault diagnosing is therefore of great interest.
    The goal of the project is to develop a detailed and reliable system for monitoring the cylinder condition in marine engines. The cylinder condition will be monitored by use of sensors which either directly or indirectly can measure important parameters of the cylinder condition (temperature, cylinder pressure and sound/vibrations).
    This demands integration of information from sources with different signal characteristics and signal-to-noise ratios in a comprehensive evaluation of the cylinder condition (signal/sensor fusion). Also, design of performance criteria by use of for instance Bayesian analysis and integration of specific expert knowledge
    (prior information) will be considered. One specific form of prior information is the so-called wavelet representation for sound/vibration signals. In this case the network input could be represented as sound/vibration ``images'' describing time dependent development of the signal's frequency spectrum. Such representation can be useful for detection of anomalies and non-stationarity.
    The diagnosing tool will be a neural network and therefore a detailed study of
    neural network architectures and performance optimization methods will be necessary.
    Especially methods for analyzing multivariate time series (simultaneous prediction of several parameters) will be in focus.
    Effective start/end date01/12/199431/05/1998

    Collaborative partners


    • Unknown


    Explore the research topics touched on by this project. These labels are generated based on the underlying awards/grants. Together they form a unique fingerprint.