Time-reversed Impact Dataset
- Citation Author(s):
- Submitted by:
- Camilo Hernandez
- Last updated:
- Fri, 11/04/2022 - 15:01
- Data Format:
- Link to Paper:
Touch-screens are the basic and convenient human-computer interface. They are extensively used in digital musical applications, where a complex action-perception loop is involved. Therefore, it is crucial to establish a rich vibrotacticle feedback to improve the quality of the user's interaction. This paper explores the capacity of Generative Adversarial Networks (GANs) to generate time-reversed signals that can achieve localized vibrotactile feedback on a rigid surface.
We adopt the Generative Adversarial Networks (GANs) framework for its indicate and flexible training capacity. The experimental setup for data generation and verification consists of an aluminum plate with a piezoelectric actuator bonded to its surface and a compact laser vibrometer. The given signals are sent into the experimental setup and a vibration scan is carried out. A localized peak generated with a signal synthesized by the trained GAN model is observed and studied. Later, different metrics are proposed to evaluate the quality of the generated samples and the obtained localized peak. Finally, the feasibility of this approach to generate localized vibrations in the range of 200 - 300 Hz for touch-screen applications is verified.
The time-reversed impacts were acquired using a piezoelectric transducer (a 7 x 7x 0.2 mm^3 piezo-ceramic transducer (Steminc - SMPL7W8T02412WL) ) bonded to a 250 x 16 x 2 mm^3 (l x w x h) aluminum beam (Aluminum AW-6082) using epoxy glue.
The impacts are created with a pneumatic piston.
There are 178 unique locations in the center of the beam. Starting at X_start = 63 mm until X_end = 240 mm with a spatial resolution of 1mm.
At each location 30 repetitions are acquired.
The impact signals are acquired with a sampling rate of 250 kHz. Then, they are time-reversed (flipped In the time axis). Finally, they are cropped to a length of 16384 samples.
More details on the impact acquisition process are displayed on our previous publication:
C. H. Mejia, J. Chavanne, P. Germano, and Y. Perriard, “Effect of the Impact Contact Duration on Machine Learning Models for Impact Position Detection,” in 2020 23rd International Conference on Electrical Machines and Systems (ICEMS), Nov. 2020, pp. 2063–2068, iSSN: 2642-5513