Back to Search
Start Over
A digital image flow meter for granular flows with a comparison of direct regression and neural network computational methods.
- Source :
-
Flow Measurement & Instrumentation . Apr2019, Vol. 66, p18-27. 10p. - Publication Year :
- 2019
-
Abstract
- Abstract Effective measurement of dense granular flow rates is essential for ensuring optimal performance of a wide variety of industrial processes with digital imaging processing techniques being developed and implemented in many manufacturing control applications. This paper presents a digital image flow meter system that utilizes sequential image pairs to determine granular mass flow rates with a comparison of two different computational strategies: Direct Regression (DR) of the displacements and Neural Network (NN) modeling. Results show DR is a robust method that can accurately predict flow rates with an average relative error of 7.56% without calibration despite its simplicity. Both methods can have the relative error reduced below 3% by time-averaging over a series of measurements. NN models were found to predict flow rates from image pairs faster than DR, but the NN predictions had a higher variance and lower accuracy. The proposed granular flow metering strategy has the potential of utilizing inexpensive hardware to effectively estimate flow rates and can be easily implemented in hardware platforms where there is a visible granular flow. Highlights • A robust, yet inexpensive, flow meter system for granular flows was developed. • Computation times are fast enough to be run in real time on a standard desktop computer or smart device. • Analytical methods (Direct Regression and Neural Networks) were compared and discussed. • Errors were found to be dramatically reduced by time-averaging. [ABSTRACT FROM AUTHOR]
Details
- Language :
- English
- ISSN :
- 09555986
- Volume :
- 66
- Database :
- Academic Search Index
- Journal :
- Flow Measurement & Instrumentation
- Publication Type :
- Academic Journal
- Accession number :
- 135625066
- Full Text :
- https://doi.org/10.1016/j.flowmeasinst.2019.01.014