Automated Validation of Test Cases Using Generative AI: Development of program for generating test cases utilizing generative AI for system requirements in automotive industry

Abstract

This project was provided by Volvo and was intended to help automate their current system of creating validation test cases using generative artificial intelligence (AI). Automation of this system reduces development time and costs while enhancing efficiency. This project covered theories regarding AI, Large Language Model (LLM), prompt engineering, test cases and system requirements. Following an evaluation of available open-source LLMs, the model QWQ-32B was selected. By applying prompt engineering techniques, the model was able to generate not only validation test cases but also executable test code. The program was able to generate adequate results according to Volvo staff. However, the results needs some adjustments in order to be viable for a variety of test cases. Potential improvements, such as the integration of Retrieval-Augmented Generation (RAG), are discussed in this report as future directions to address current limitations. As a result of this project, Volvo has been provided with a solid foundation for automating generation of validation test cases and executable test code.