Synthetic On-Board Diagnostics Data Generation and Evaluation for Vehicle Diagnostic Testing

The generation of data plays a vital role in machine learning (ML) techniques by providing the foundation for training and improvement of forecast models. As one application area for these models, in-vehicle systems, like vehicle diagnostics, have the potential to enhance the reliability and durability of vehicles by utilizing ML models in the testing phases. However, acquiring a high volume of quality onboard diagnostics (OBD) data is time-consuming and poses challenges like the risk of exposing sensitive information. To address this issue, synthetic data generation offers a promising alternative that is already in use in other domains. Thereby, synthetic data allows the exploitation of knowledge found in original data, ensuring the privacy of sensitive data, with less time costs of data acquisition. The application of such synthetically generated data could be found in predictive maintenance, predictive diagnostics, anomaly detection, and others. For this purpose, the research presented in this contribution investigates the use of statistical and ML-based synthetic OBD data generation methods. The models are evaluated with the custom-developed evaluation method that fits the attributes of the OBD data used. Finally, an important result is the successful generation of synthetic OBD data that can be used to enhance the SAE J1699 OBD compliance test, together with tools and insights for models and evaluation.