Deutsche Gesellschaft für Zerstörungsfreie Prüfung, May 26 - 28, 2014, Potsdam, Germany
Programme: Di.2.C - Bauwesen II (Di.2.C.1 - 10:40)
Unsupervised Fusion of Scattered Data Collected by a Multi-Sensor Robot on Concrete
P. Cotic1, E. Niederleithinger2, M. Stoppel2
1Institute of Mathematics, Physics and Mechanics, Ljubljana, Slowenien
2BAM, Berlin
Kurzfassung:
A large part of parking garages and bridge decks suffer from severe corrosion of the reinforcement, which is reflected in cracking, spalling and losses of the concrete cross section. In order to evaluate the extent of degradation adequately and to divide the investigated structure into zones with defined damage classes, investigation of surfaces of some thousand square metres with a dense grid is necessary. To solve this problem, a self-navigating mobile robot system has been developed within the BetoScan project (www.betoscan.de). With the BetoScan system, potential mapping, as well as the distribution of concrete cover and moisture can be assessed simultaneously by an automated multi-sensor system. Due to practical limitations such as rough surface, the need to avoid obstacles and the sensor arrangement on the robot, measurements are performed at scattered points, which are not necessarily coincident for different sensors. Sparse grids are used in order to provide fast inspection of large areas. Consequently, for an efficient evaluation of results, data have to be interpolated on a regular and dense grid using methods such as nearest-neighbour interpolation, spline functions, inverse distance weighting or kriging. In the post-processing step, data analysis is currently performed manually by direct comparison of the results from several used NDT methods. Thus, to promote an efficient data evaluation framework, which could speed up and simplify the evaluation of large data sets, an unsupervised data fusion is of major interest. A case study involving a BetoScan data set acquired from a reinforced concrete floor of a parking garage in Germany is presented. The data set includes potential mapping, cover metering with eddy current, as well as microwave moisture measurements. A number of methods for interpolation of scattered data are examined. In the post-processing step, the investigated structure is segmented into zones using clustering based data fusion methods. The challenge to accurately define the damage associated to these zones is discussed.