Automatic classification of water drop impact characteristics using audio information

This paper presents a novel approach to measuring drop characteristics based on audio recordings and machine-learning-oriented signal processing. The study examines the possibilities of classifying and identifying drop characteristics, namely, volume and velocity, based on audio recordings of the drop impact sequences. A microphone was positioned above a water surface, and an audio recording database of drops falling onto the water surface was obtained to reflect practical application scenarios. Two sets of drop volumes were used, with drops free-falling from cylinders of two different diameters. The drops were released from various heights to achieve different impact velocities. Preliminary analysis demonstrates the time-frequency behavior of drop impact events using both synchronized image sequences and spectrograms of the audio recordings. Various classification approaches were evaluated to assess the scope of drop-impact-velocity and drop-volume categorization; these included classical feature extraction (FE), frequency-based FE, and recent general-purpose classification methods. The resulting detection rates were 99.5% for drop height and 94.4% for drop volume detection. • Audio-based machine learning classifies drop impact velocity and volume. • High classification accuracy achieved (99.5% velocity, 94.4% volume). • The method enables non-intrusive, vision-free monitoring of drop characteristics.