Synthetic Data Generation for the Enrichment of Civil Engineering Machine Data

Artificial Intelligence (AI) is one of the most auspicious technologies in the mobile machine domain. It promises to optimize the machine operation to reduce energy consumption or provide an assistant function to support the operator in challenging machine movements. A large amount of machine data is required to train and build AI models. These data sets are often not available due to missing or faulty sensors in the machine. However, construction machines are partly equipped with temporary sensors for data collection so that small data sets are available. Nevertheless, these data sets are very small and must be extended with more realistic data. Generating synthetic data to enrich real data is a promising approach to overcome the obstacle of small data sets. This paper presents a data generator to produce synthetic, physically-informed data for the pendulum trajectory of a flexible attachment tool on a construction machine. The data generator calculates a reference trajectory based on a physical model of the machine. This reference trajectory is generated by solving an optimization problem to cover the machine movement that an experienced machine operator would drive. Reasonable deviations of these trajectories are generated by varying machine characteristics and adding external forces to the physical model to simulate rough environmental conditions. The data generator is implemented for the grab system movement of a civil engineering machine.