Developing Theoretical and Empirical Definitions of Safety Problems in Driving Suitable for Active Safety Function Evaluation

Licentiate thesis, 2009

In passive safety, the requirement specifications used for evaluation of protective functions are both standardised and specified at a high level of detail regarding evaluation scenario definition, performance metrics and pass/fail criteria. For active safety, while several propositions for evaluation scenarios have been made, neither these, nor performance metrics and pass/fail criteria have yet reached a similar level of detail and standardisation. The objective of this thesis is to address two underlying reasons for this difference. One is theoretical in nature. On a general level, a set of principles and concepts which capture the fundamental ideas of a field of science can be called a conceptual framework. For active safety function evaluation, such a framework is currently lacking. To address this issue, a conceptual framework called Situational control was developed. The framework integrates fundamental ideas relevant for active safety function evaluation into a holistic and practically applicable picture. Its applicability was demonstrated by applying it in the context of writing and implementing requirement specifications for active safety function evaluation. The second reason is of empirical character. To evaluate the extent to which active safety functions prevent and/or mitigate crashes, it is essential to characterize the sequence of events which leads to collisions in a way which includes information on causal factors. To do this, data from official databases (macroscopic data), and in-depth case studies is often used. Macroscopic data is usually statistically representative but has limited information on why crashes happen, while the opposite is true of case studies. Using the two in combination would therefore seem ideal. However, the principles for connecting them are far from clear and current approaches suffer inherent weaknesses. To address this issue, a generalization methodology which links information in case studies to macroscopic crash types, in a way which covers not only context but also causation similarity, was developed. The feasibility of the methodology was tested through application on three sets of intersection crash data. Results indicate that the methodology was sufficiently successful to warrant further exploration with larger data sets.