Overview
A novel framework has been developed for generating representative datasets tailored for AI-based failure analysis in the real-time validation of automotive software systems. This approach integrates Hardware-in-the-Loop (HIL) simulation with real-time Fault Injection (FI) to produce diverse data samples, encompassing both sequential and textual formats, under various fault conditions.
Research Context
The data-driven approach has seen extensive application across different stages of the system development life cycle, primarily due to its capacity to extract knowledge from historical datasets. In the domain of real-time system validation, significant progress has been made in intelligent failure analysis leveraging historical data. This data-driven methodology is noted for its superiority over conventional alternatives, such as model-based and signal-based approaches. However, a persistent challenge in this field is the availability of representative datasets. This necessitates the exploration of new methods for generating representative faulty data in various forms for different engineering applications.
Approach
The proposed framework utilizes Hardware-in-the-Loop (HIL) simulation as a core component. HIL simulation provides a platform for emulating the real-world environment of automotive software systems. Complementing the HIL setup is a real-time Fault Injection (FI) method. This method is designed to introduce single and simultaneous faults into the simulated system. The integration of HIL simulation and real-time FI allows for the generation and collection of data samples that reflect system behavior under these fault conditions. The framework is engineered to generate two primary types of data: sequential data, which captures time-series information, and textual data, specifically fault logs.
Findings
The framework successfully simulates and captures system behavior under various fault conditions within system components. This capability includes the generation of data corresponding to both single faults and scenarios where multiple faults occur simultaneously. The output includes sequential data streams and textual fault logs. The results indicate the applicability of the developed framework for generating the required data for Machine Learning applications during system validation phases.
Why This Matters
The availability of representative datasets is a significant challenge in AI-based failure analysis for real-time system validation. This framework addresses this by providing a method to generate such data, which is crucial for the development and training of Machine Learning models used in intelligent failure analysis.
Potential Applications
- Generating data for Machine Learning applications during automotive system validation phases.
- Simulating and capturing system behavior under single faults.
- Simulating and capturing system behavior under simultaneous faults.