<p>Tribology plays a crucial role in the design of mechanical systems relying on friction to operate. As an example of those systems, ultrasonic ceramic actuators (UCMs) are a type of piezoelectric actuators that leverage the non-linear relationship between friction and relative velocity to drive objects over millimetre ranges using ultrasonic microscopic oscillations of their crystal. The contact dynamics of those actuators are characterised by small relative displacements at high frequency, leading to acceleration up to 50,000&#xa0;ms⁻². Although mentions are made about the implications of high accelerations in the field of tribology, state-of-the-art analytical models fail to grasp the full range of dynamics. This study aims at exploring the friction force between surfaces having relative motion matching the one occurring inside UCMs. To do so, a custom sliding test bench generates displacement at various high accelerations using impacts from an air piston. The experimental findings show the evolution of tangential friction force versus relative velocity curves as acceleration increases when it comes to values up to 1&#xa0;ms⁻¹ and 11,000&#xa0;ms⁻². Analytical and empirical models fail to maintain accuracy across the full range of measured accelerations. Using the dataset that we created, we propose a deep learning approach outperforming the two analytical models used for comparison.</p>