AI-in-the-Loop Framework for Functional Testing Automation in the Automotive Industry

The increasing system complexity of modern vehicles has significantly raised the demands on verification and validation processes within the automotive industry. Conventional functional testing, which relies on rigid, scripted procedures, lacks the flexibility and time efficiency to keep pace with accelerated development cycles. In this context, the introduction of artificial intelligence (AI) into the testing workflow offers promising opportunities to bring adaptability and enhance operator support. This thesis investigates the application of AI methods to the automation of functional testing processes in the automotive domain. In particular, it explores how this technology can complement conventional Hardware-in-the-Loop (HiL) setups by enabling more adaptive and intelligent testing strategies. Several representative use cases are developed and implemented as proof-of-concepts: (i) anomaly detection on the driver’s panel, both in real time during test execution and in post-processing of recorded data, enabling faster and more reliable fault identification; (ii) icon and popup recognition and classification using a Siamese neural network, supporting automated monitoring of visual outputs and interface validation; (iii) seat movement detection through the tracking of visual markers, allowing automated verification of actuator behavior and mobile parts; and (iv) test list correction and automatic routine generation with the assistance of large language models (LLMs), which minimize the manual effort involved in revising tests, verifying their alignment with requirements and vehicle configurations, and converting natural language descriptions into structured, executable test procedures. Each of these use cases is quantitatively evaluated through repeatable metrics, demonstrating measurable benefits in terms of flexibility, fault detection, and overall testing efficiency. The results highlight how AI-driven solutions can not only reduce reliance on manual supervision but also provide dynamic adaptability compared to static test scripts. Ultimately, the thesis contributes to the ongoing discussion on integrating AI technologies into automotive testing, emphasizing their potential as enablers of more robust, automated, and time-efficient validation processes.