Reconstruction-Free EIT for Injection-Pattern Classification and Superficial Gas Velocity Regression as Proxies for Local Gas Holdup in Bubble Columns

Abstract Electrical impedance tomography (EIT) as a noninvasive tomographic technique is increasingly applied to multiphase reactor monitoring; however, conventional image reconstruction is ill-posed and regularization-dependent and may be redundant in applications where the primary objective is operating-state identification rather than explicit spatial conductivity field reconstruction. Here, we present a reconstruction-free, measurement-domain framework for bubble-column monitoring that maps raw complex boundary impedance data directly to two reactor-relevant inference tasks: (i) gas injection pattern classification and (ii) superficial gas velocities regression. Together, these two quantities ─ the spatial injection distribution and the total volumetric flow ─ constitute the primary process-state information from which gas holdup can subsequently be inferred and are therefore reported as proxies for local gas holdup monitoring. Experiments were conducted in an acrylic bubble column (600 mm height, 104 mm inner diameter) equipped with a 256-electrode array distributed over eight axial rings and operated at four excitation frequencies (1 kHz-1 MHz). Experiments covered gas flow rates between 1.0 and 6.5 L min–1 (Ug = 1.96 → 12.75 mm s–1), within which near-perfect gas injection pattern classification was achieved with accuracies of 93–100% for excitation frequencies between 1 and 100 kHz using the full 256-electrode configuration. For quantitative superficial gas velocity estimation, increasing calibration density along Ug reduced the mean absolute error from 0.388 to 0.105 L min–1 (MAE[Ug] = 0.76 → 0.205 mm s–1, i.e. 7.0% → 1.9% of the operating range) for localized injection and from 0.298 to 0.157 L min–1 (MAE[Ug] = 0.585 → 0.307 mm s–1, i.e. 5.4% → 2.8% of the operating range) for distributed injection conditions. These results demonstrate that direct inference from raw EIT boundary measurements enables accurate, real-time monitoring of bubble-column operation without tomographic reconstruction and provide quantitative guidance on excitation frequency selection, axial sensing placement, and calibration resolution.