Functional Decomposition Approach - Reducing the Safety Validation Effort for Highly Automated Driving

Abstract

This dissertation is concerned with the application of functional decomposition - which is known from other fields, for example mathematics or computer science - for the validation of automated driving functions. The approach aims to reduce the number of required test cases compared to a scenario-based black box system test. The first part of this thesis analyzes the state of the art for the verification and validation of automated driving functions. A validation of highly automated driving functions with exist-ing methods is not feasible due to the “open world” with a multitude of influence parameters that leads to a high number of required tests. This challenge indicates the need to develop new approaches to reduce the validation effort. The development of a method for the application of functional decomposition for the defi-nition of so-called particulate test cases - i.e. test cases that serve to verify and ultimately validate one or more functional layers of an automated driving function - is therefore defined as the goal of this work. Subsequently, requirements for the development of a validation method are defined and research questions are derived. The main part of the thesis focuses on these research questions and the development of the following substeps of the methodology: - Definition of independent functional layers and their interfaces - Definition of criteria for evaluating the functional layers - Allocation of influence parameters - Generation of particulate test cases Furthermore, the potential to reduce the validation effort by functional decomposition is quantified. This shows that the absolute test effort as well as the possible reduction of the test effort by functional decomposition strongly depend on the required test coverage and the discretization of the influence parameters. Depending on the required test coverage, the amount of required test cases can be reduced by up to two orders of magnitude by the intro-duced approach. In the final part of this thesis, the goals set at the beginning are compared with the actual gain of knowledge and remaining as well as newly added questions for further research pro-jects are presented. These concern above all a further detailing and automation of the method for an application in practice, as well as further measures for the reduction of the validation effort beyond the application of the functional decomposition as for example the definition of the minimum necessary test space coverage and the discretization of influence parameters with a continuous value range.