Modelling of maintenance performance indicators

All data generated from the project will be processed and developed into a model that can produce Maintenance Performance Indicators (MPIs) for condition of the S&Cs. Issues that have to be addressed include the abundance of fast streaming and complex data sets generated from conventional measurement vehicles and the state-of-the-art sensors implemented in selected S&Cs, as well as the misalignment of data streams as this often necessitates the development of dynamic time warping techniques before analysis. One of the main challenges for data that arrive from a variety of sources at high volume is to lower the dimensionality by taking advantage of the correlation structure among the variables to enable extracting important features. New methods will be needed to address the issues such as: Statistical Process Control and multivariate feature extraction and variable reduction. Methods in line with Multi-way Principal Component Analysis and Sparse Principal Component Analysis inevitably will have to be explored with “Big Data” issues in mind. The expected main results are as follows:

  • Preparation of data from existing measurement vehicles.
  • Development of a model which summarises inputs from other work packages and is able to deliver one or more MPI´s for the condition of individual S&Cs.
  • Correlation of sensor data to measuring vehicle data, so that vehicle data can be used reliably as input for the MPI model without requiring implementation of expensive sensors.



Bjarne Kjær Ersbøll
Professor, Head of section
DTU Compute
+45 45 25 34 13