Power Hop Detection in High-Torque Vehicles: A Machine Learning Approach with Focus on Data Coverage and Model Generalization

Power hop is a vibration phenomenon that occurs during high accelerations from low speed. In severe cases it can lead to component damage or deformation. Therefore, the affected vehicles must be safeguarded against these vibrations by a safe design of the components and by additional software-based functions. Conventional software-based solutions, such as Traction Control Systems (TCS), often perform delayed interventions and apply harsh torque adjustments that reduce driving comfort. Motivated by these challenges, this paper proposes a novel approach for power hop detection in a high-torque vehicle based on Long Short-Term-Memory Network (LSTM) and real-time measurements. Unlike conventional methods, our LSTM precisely detects the start of power hop, enabling proactive torque adjustments. Due to its impact on vehicle stability, the model must achieve a high level of reliability and robustness. Given the importance of data quality in Machine Learning (ML), we consider data-related principles outlined in ISO/PAS 8800. First, the data acquisition through multiple driving tests is described. Second, two datasets are extracted and analyzed for representativeness and variability using k-Nearest Neighbors (kNN) and Dynamic Time Warping (DTW) to ensure broad coverage. Third, to evaluate the impact of the dataset variability on model generalization, these datasets are used to train two LSTMs. The results show that a dataset with higher variability in time series improves model generalization on unseen data.