Vi-liquid: unknown liquid identification with your smartphone vibration

Traditional liquid-analysis instruments are expensive, invasive, and rarely available to end users. This paper studies whether a commodity smartphone can identify an unknown liquid without external sensing hardware. Our key observation is that, under fixed mechanical excitation, liquid viscosity induces both a measurable boundary shear load and a dissipation pattern on the container wall. Based on this observation, we build a physicsgrounded vibration-viscosity model that links steady-state amplitude and free-decay attenuation to viscosity, and we realize the model on a smartphone using only the built-in vibro-motor and accelerometer. A practical deployment must overcome three system constraints: severe under-sampling at the mobile operating-system API, strong straight-path self-interference from the motor to the inertial sensor, and volume-dependent changes in the coupled wall-liquid resonance. Vi-Liquid addresses them with phase-shift-based supersampling rate reconstruction, OMP-based sparse recovery, spectral subtraction of a calibrated straight-path template, and volume compensation in the frequency domain. Across 30 liquids, Vi-Liquid achieves a mean relative viscosity error of 2.9% and 95.47% identification accuracy, and it also supports proof-of-concept screening for water contamination, changes in urine composition, and alcohol concentration. These results show that liquid identification by physics-based active vibration sensing is feasible on commodity smartphones without specialized sensor add-ons or liquid-specific training.