Inferring the shape of a solid inside a draining tank from its liquid level dynamics

TL;DR

This paper presents a method to reconstruct the shape of a solid inside a draining tank by analyzing liquid level dynamics, combining modeling, Bayesian inference, and experiments, achieving accurate shape estimation.

Contribution

It introduces a novel approach that infers the shape of an internal solid from liquid level data using Bayesian methods and experimental validation.

Findings

Achieved less than 10% mean error in shape reconstruction

Successfully inferred the cross-sectional area of a bottle-shaped solid

Demonstrated practical non-destructive shape characterization

Abstract

We aim to reconstruct the shape of an exogenous, heavy solid contained in a tank from measurements of the liquid level in the tank as it drains (driven by gravity) through a small orifice in its side. (Because the solid displaces liquid, the rate of decrease of the liquid level provides information about the cross-sectional area of the solid at that height; as the liquid level drops, it "scans" the area of the solid as a function of height.) We combine mathematical modeling, Bayesian statistical inversion, Monte Carlo simulation, and wet experiments of a tank draining of water to demonstrate and test our ability to infer the cross-sectional area of the exogenous solid as a function of height. In our experiment, the posterior distribution over the [held-out] shape of the solid (a bottle) agreed reasonably well with our length-measurements (<10% mean reconstruction error on its radius). Our approach may be practically useful to non-destructively characterize the geometry of an unknown solid, or a packed bed of solid particles, contained in an opaque tank.