Centrifugal Pump-Based Predictive Models for Kraft Black Liquor Viscosity: An Artificial Neural Network Approach

Abstract

Previous investigators have shown that the Newtonian viscosity of black liquor (BL), a byproduct of kraft pulping process, can be estimated online from the performance parameters of an installed centrifugal pump (CP). Unfortunately, the existing models from which such estimates can be obtained lack the necessary robustness for process control applications and/or would require a substantial amount of data for periodic updates. This study developed a generalized artificial neural network (ANN)-based model which directly accounts for the effect of aging on the pump performance (hence the model). Simulation results show that ANN predicts BL viscosity better than the existing linear models as the former gives accurate and robust predictions at all practical operating points of the pump. Moreover, the ANN model requires just a single data point for its periodic recalibration as the pump ages significantly. The methodologies presented here can easily be adapted for use in any process industry where Newtonian process fluids are transferred by a CP.

Performance prediction of kinetic and screw pumps delivering slurry

“…The two parameter model for non-Newtonian homogenous slurry with yield stress can be modeled by Bingham plastic model. 22–38 …”

“…20,21 The ANN model is used to predict the performance of CFP with semi-open impeller blades using flow rate and splitter blade lengths, head reduction of centrifugal slurry pump, effect of aging on the pump performance, performance of helicon-axial multiphase pump, flow rate of photovoltaic water pumping system and performance of heat pumps for hot water heater. 22–27 The ANN models have been used in other engineering areas for optimizations and designs. 28–30…”

Neural Network-Based Model for Joint Prediction of the Newtonian and NonNewtonian Viscosities of Black Liquor

“…Prior to the network's training, the input and the output variables were pre-processed to fall in the range [-1, 1] (see [26]). The solids concentration was expressed in a fractional form (0-1) while temperature was expressed in Kelvin.…”

“…The shear rate (s -1 ) and the viscosity (mPa.s) were logtransformed prior to being scaled to within [-1, 1]. The steps described in [26] for training ANN via Bayesian regularization algorithm available in the MATLAB (R) neural network toolbox, were followed to arrive at the networks reported in this paper. The typical criteria for the evaluation of the goodness of a Bayesian-trained network include Marquardt adjustment parameter (MU), the effective number of parameters per total number of tunable parameters (#), sum of squared weights (SSW), and the network prediction error.…”

Artificial Neural Network Modeling of Surface Tension for Pure Organic Compounds

“…One of the applications of neural network models is to map an input space to an output space and function as a lookup table. Thus, in recent years, artificial neural networks have been applied to formulation of chemical and physical properties. − …”