Generative AI based methodology for Road Profile Mix Creation

The acquisition of road profile data is crucial for various automotive testing applications, including vehicle dynamics analysis, chassis endurance tests, and simulation of vehicle-road interactions. This is necessary for conducting virtual tests to accelerate research and development processes and can significantly reduce testing costs. However, most of the on-road measurements lack comprehensive and relevant road profile data. Conducting on-road trials to acquire this data is a laborious and time-consuming process, often impeded by logistical and environmental challenges. This research proposes a generative AI-based methodology for creating diverse and realistic road profile mixes from the existing on-road dataset of front axle displacement and road profile measured with a laser sensor. By leveraging advanced machine learning techniques, the proposed approach seeks to generate synthetic road profiles that accurately reflect real-world conditions, thereby reducing the dependency on extensive on-road measurements. The methodology involves classifying the existing road profile data into respective road classes (as per ISO 8608:2016) and training a random forest classification model based on the data. This model is in turn used as a base to train a conditional Generative Adversarial Network (cGAN) on the same dataset. The GAN learns to produce synthetic road profile data with the same statistical properties and variability of the training data. This generated data is then validated against real world data for ensuring accuracy and robustness of the model. This approach not only addresses the physical constraints associated with on-road data collection but also offers a promising solution to enhance the ability to simulate and analyse vehicle-road interactions under diverse conditions.