Artificial Neural Networks for Impact Position Detection in Haptic Surfaces

Recently, the interest in haptic feedback is growing thanks to its ability to enhance the interaction with Human Machine Interfaces (HMIs). This research project is exploring the potential of machine learning combined with piezoelectric actuators to generate localized vibrational feedback over a thin rigid surface. With this goal in mind, this paper studied the potential of neural networks and machine learning algorithms to extract the position, where an impact has occurred. A data-set with 5310 stress signals labeled with the position at which the impact has occurred, was obtained using an automated Linear Impact Generator (LIG). Each signal was transformed into a spectrogram using the Fast Fourier Transform. During the study, different neural networks and machine learning algorithms were implemented and a supervised training process was carried out. At the end of the paper, the results of the different models are compared. The best model has an error (Validation MAE) of 4% and (Test MAE) of 8% in the impact position detection over an aluminum thin plate.