Hackathon 2019 #TeamPink – Optimize inspection planning
By Andreas Kristensen, engineer, Rambøll, Silja Nielsen, Engineer, Rambøll, Niels Liljedahl Christensen, Computer Scientist, Alexandra Institute and Nicolas Frandsen, Engineer, Rambøll
The main challenge, as we see it, is to make qualified statements about the maintainability of the system based on data which reflect a healthy system (i.e. where all structural elements work as intended without errors).
Our solution is as follows:
We seek to implement a real-time monitoring algorithm (using e.g. machine learning models such as random forest, neural network etc.) that is able to predict the dynamic behavior (oscillations) of the vertical hangers as a function of weather data and estimates of weights from traffic data. The idea is to make a statistical comparison of the predicted value of e.g. D-max (maximum amplitude of the oscillations) with a measured value of the same parameter. This comparison will in real-time yield a binary output corresponding to “Raise alarm” or “Do not raise alarm” depending on a chosen statistical significance level. Further development could include the implementation of the above algorithm to monitor several structural elements at once.
Our initial studies show that it is important to give thoughts to which parameters might correlate with each other. As an example, we achieved higher correlation coefficients when using the north/south- and east/west-components of the wind vector compared to using values of wind speed and wind direction. By a little data handling you might get some intel on your parameter of interest which does not show explicitly from the data at hand.
See the ppt presentation here.
Read more about the other solutions here: