Artificial Intelligence based vehicle fault analysis

In the rapidly evolving automotive industry, especially within the super sports cars segment, integrating new technologies is essential for maintaining a competitive edge. Modern vehicles, increasingly equipped with electronic components, generate vast amounts of data during prototype testing, necessitating advanced data analysis techniques. This thesis explores developing and implementing a Variational Autoencoder, a novel neural network model, to automate fault analysis during new vehicle testing. The project addresses the complexity of data interpretation by leveraging artificial intelligence (AI) to enhance precision and efficiency in fault detection. Rigorous testing of various anomaly detection methodologies revealed the Variational Autoencoder as a robust solution for dimensionality reduction and anomaly detection, demonstrating impressive performance in identifying faults through a reconstruction approach. A key aspect of this initiative is the model's ability to automatically correlate signals with reported faults, significantly streamlining the diagnostic process. By employing AI, the model provides real-time data processing capabilities, crucial for timely decision-making during live testing phases. This AI-enhanced workflow not only improves fault identification accuracy but also facilitates the generation of comprehensive reports for each detected anomaly. Additionally, this thesis outlines the sophisticated communication buses within vehicle architectures, data transformation processes, and the implementation of AI models in real-world testing scenarios. The results underscore AI's indispensable role in refining vehicle performance, ensuring safety, and driving innovation in super sports car development. Ultimately, this project illustrates how AI can revolutionize fault analysis in the automotive industry, enhancing efficiency and reliability.