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.