An Empirical Investigation of Using Models During Requirements Engineering in the Automotive Industry

Abstract: Context: The automotive industry is undergoing a major transformation from a manufacturing industry towards an industry that relies heavily on software. As one of the main factors for project success, requirements engineering (RE) plays a major role in this transition. Similar to other areas of automotive engineering, the use of models during RE has been suggested to increase productivity and tackle increasing complexity by means of abstraction. Existing modelling frameworks often prescribe a variety of different, formal models for RE, trying to maximise the benefit obtained from model-based engineering (MBE). However, these frameworks are typically based on assumptions from anecdotal evidence and experience, without empirical data supporting these assumptions. Objective: The overall aim of our research is to investigate the potential benefits and drawbacks of using model-based RE in an automotive environment based on empirical evidence. To do so, we present an investigation of the current industrial practice of MBE in the automotive industry, existing challenges in automotive RE, and potential use cases for model-based RE. Furthermore, we explore two use cases for model-based RE, namely the creation of behavioural requirements models for validation and verification purposes and the use of existing trace models to support communication. Method: We address the aims of this thesis using three empirical strategies: case study, design science and survey. We collected quantitative and qualitative data using interviews as well as questionnaires. Results: Our results show that using models during automotive RE can be beneficial, if restricted to certain aspects of RE. In particular, models supporting communication and stakeholder interaction are promising. We show that the use of abstract models of behavioural requirements are considered beneficial for system testing purposes, even though they abstract from the detailed functional requirements. Furthermore, we demonstrate that existing data can be understood as a model to uncover dependencies between stakeholders. Conclusions: Our results question the feasibility to construct and maintain large amounts of formal models for RE. Instead, models during RE should be used for a few, important use cases. Additionally, MBE can be used as a means to understand existing problems in software engineering.