A low-error calibration model for an electrostatic gas-solid flow sensor fusion obtained via machine learning techniques with experimental data

Sensor fusion is the use of software that intelligently combines data from multiple sensors in
order to improve overall system performance. This technique can be applied to measurement
of mass flow rate of solids in a pipeline with non-intrusive electrostatic techniques. Data fusion
from multiple heterogeneous/homogenous sensors can overcome limitations of an individual
sensor and measured variable. It is shown that the output voltage of a ring-shaped electrode is
predominantly a function of solids mass flow rate, air-solids ratio and particle velocity. By
additionally incorporating measured flow velocity in a proposed mathematical model (obtained
via machine learning), meter output voltage could be predicted/calculated with superior
accuracy, for a range of different flow parameters from numerous experiments with a pneumatic
conveying system. A transposed model utilised in software enables accurate mass flow
measurement with velocity compensation. Accurate mass flow measurement facilitates
enhanced monitoring and controllability of blast furnaces, power stations, chemical reactors
etc. where there is a flow of solid fuel/reactant in pipelines. Optimisation of highly materially
consumptive and energy intensive processes can yield significant reductions in waste and
emissions (CO2, NOx) and increased efficiencies in global production of energy and materials.
Keywords: sensor fusion, machine learning, electrostatic flow measurement, gas-solid flow,
pneumatic conveying, non-linear regression