PSI - Issue 38

4

Florian Grober et al. / Procedia Structural Integrity 38 (2022) 352–361 Grober, Janßen, Küçükay / Structu al Integrity Procedia 00 (2021) 00 – 000

355

Fig. 1. Conversion of the road network into a mathematical graph.

The test route can be understood as a path along the network graph. In order to determine it, an estimation of the expected loads at each edge is required. Since an a priori calculation of the expected loads is too complex, an experience database is created instead. For this purpose, the driver passes as far as possible every street of the test area at the beginning of the durability road test. Using a map matching procedure based on Newson and Krumm (2009), each GPS position sample of the vehicle is assigned to a road section. On the basis of the associated timestamps at the start and the end of each segment, the corresponding load signal fragments can be extracted from the measurement and stored in the database. During the following test drive the database is continuously updated after each travelled edge of the graph. By operating a separate database for each driver, the driver's influence on the expected loads can be captured and taken into account. On closer inspection, it becomes obvious that the loads do not depend solely on the driving behavior on the road sections, but in particular also on the turning maneuvers carried out at intersections. For example, for a left turn the driver firstly decelerates, steers through the curve and then accelerates again on the next section. The resulting loads are harder as if he had driven straight-on. Therefore, the links of the road sections along the route are also relevant for the expected loads. To take this aspect into account, Pfeiffer et al. (2014) suggests that the nodes of the graph should be equipped with turn-dependent weights, which represent the loads and are taken into account by addition. Since this procedure remains rather imprecise, an alternative method is proposed here, which is based on a new way of modelling the road network: As the example of the left turn maneuver has shown, the turn-dependent loads vary especially near to the intersections. Therefore, the road network graph is modelled in such a way that the nodes are no longer positioned in the centers of the intersections, but instead in the middles of the road sections. At these points, a constant driving style is most likely to be expected, which allows a transition between two adjacent edges without discontinuities. The graph is created by connecting all nodes that are directly accessible from each other with edges. Directly accessible means that no other node has to be passed. It is recommended to omit the nodes in the centers of very short road sections.

Fig. 2. Conventional versus alternative modeling of the road network graph.

Fig. 2 shows that the new modeling creates more edges and thus increases the effort required to fill the experience database of expected loads. However, this effort is justified by the increased precision.