This final thesis presents the development of a prototype touch-sensitive interface. For determining the location and force amplitude of the touch, the interface does not rely on conventional technologies, but rather on the analysis of the dynamic response of a mechanical structure. For this purpose, an experimental system with a 3D-printed plate and piezoresistive sensors was designed. The collected dynamic response signals were used to train a convolutional neural network with a multi-task architecture. A comparative analysis showed that the classification approach, which divides the surface into discrete areas, is more robust and achieves higher reliability if compared to the regression approach. During evaluation on the test set, the classification model achieved 96% accuracy in determining the location and 88% accuracy in classifying the impact force amplitude. The work demonstrates that by analyzing structural dynamics and using data-driven approaches, it is possible to develop an effective alternative touch interface.
|
