Towards an Improved Safety Benefit Assessment for Heavy Trucks - Introduction of a framework for the combination of different data sources

Licentiate thesis, 2020

Although heavy goods vehicles (HGVs) were only involved in 4.4% out of more than 1 million crashes that occurred on European roads in 2017, their share in crashes with fatal outcome was almost three times larger (12%). Advanced Driver Assistance Systems (ADAS) have the potential to mitigate the consequences of these crashes or avoid them altogether. In order to prioritise the most promising system, several types of safety benefit assessment are performed separately and independently of each other. These assessments miss however a combination into a common output, i.e. they are not able to provide a holistic overview but only show compartmentalised results.
The first objective of this thesis is to provide a framework that can incorporate multiple data sources and combine their results into one common safety benefit output. The proposed framework within this thesis is based on Bayesian modelling and can update prior information (e.g. simulation results of a new ADAS) with new observations (e.g. test track results of the ADAS). The framework can incorporate additional information such as user acceptance and market penetration of the ADAS for an improved benefit assessment. The output of the framework can easily be incorporated as prior knowledge in new safety benefit assessments, e.g. when new data is available.
The second objective is to prepare the application of the framework for the assessment of the safety benefit associated to the introduction of new ADAS for long-haul trucks. In order to specify the most critical crash scenarios for HGVs in Europe, a detailed, three-level analysis of crashes involving long-haul trucks was performed, starting on a general European level and going to in-depth crash data. The identified target scenarios are (a) rear-end crashes with the truck as the striking vehicle, (b) crashes between a right-turning truck and adjacent cyclist and (c) crashes between a truck and a pedestrian crossing in front of the truck. These three scenarios should be the basis for ADAS development and further addressed by driver behaviour modelling in the future.
Future work will focus on improving simulation results by incorporating more accurate driver models, that are better able to represent truck driver behaviour, e.g. brake or steering reactions. These models will help to obtain more valid simulation results, and thereby increase the output quality of the framework.