Clive Grainger


Advanced Fluid Dynamics Laboratory

CSIRO Division of Building, Construction and Engineering

Graham Road, Highett

Victoria, 3190

Tony McCray


Queensland Alumina Ltd

Parsons Point, Gladstone

Queensland, 4680

Colin Paulson

Manager - Coal Utilisation

CSIRO Division of Energy Technology

Delhi road, North Ryde

NSW, 2113


Increased dust precipitation from several existing Electrostatic Precipitators (ESPs) has been obtained by modifying the ductwork and screens leading into the ESPs.

The principle applied was that improving the uniformity of the gas flow entering an ESP leads to improved dust collection. This principle is supported by an analysis based upon a modified form of the Deutsch Equation which applies to the collection of dust in ESPs.

Laser Doppler Velocimeter (LDV) measurements of water flow in accurate physical scale models of the full-scale ESPs have been used to quantify the non-uniformities in the gas flow in the existing ESPs and to assess modifications. The LDV technique represents a significant advance over the earlier flow visualisation approach reported at the last Alumina Quality Workshop.

Dust output levels in modified ESPs can be reduced by half or better.


dust, precipitation, ESP


Clive Grainger, Tony McCray, Colin Paulson


In the alumina industry, electrostatic precipitators (ESPs) are widely used to remove dust from gases exhausting from calciners and boilers. To maximise production (which increases emissions) whilst simultaneously meeting reducing environmental emission targets requires that ESP dust collection efficiencies should be the maximum possible.

Welsh (1996) reported at the last Alumina Quality Workshop how the performance of an ESP had been improved substantially through small-scale physical modelling of the flow in the ESP and the ductwork leading up to it. This was achieved primarily through flow visualisation using dye in water for the case where an ESP had major flow mal-distributions. Since then, performance improvements have been achieved in several more ESPs after modifications developed in small-scale physical models of the ESPs at the CSIRO Advanced Fluid Dynamics Laboratory (AFDL) were applied to the full scale units. This new work has been made possible through the use of a Laser Doppler Velocimeter (LDV) to measure velocities in the model ESPs which is an improved technique compared to flow visualisation alone.

The basis for the improvement in each ESP has been to make the velocity distribution more uniform at the entry to the main ESP chamber. Figure.1 illustrates the situation in many ESPs, where a dust laden gas stream enters at A-A, then passes through a short, wide-angled diffuser B-C and into the ESP proper after C-C. The jet of gas passing through B-B does not follow the walls B-C ; rather its inertia carries it into the centre of the ESP to create a velocity distribution as shown. Such a velocity distribution causes a poor ESP performance (as will be further discussed below). Therefore, ESP manufacturers have inserted screens into the diffusers (eg. across C-C ) in an attempt to make the velocity distributions more uniform.

Figure 1

Side elevation of a typical wide-angled diffuser at entry to an ESP chamber

The approach taken to improving the ESPs was to first measure the velocity distribution at the entry of the ESP with the existing configuration then to modify the vanes and screens within the ductwork upstream of the ESP to produce a measurably better velocity distribution. As outlined below, a method was developed to estimate the likely improvement in dust collection efficiency corresponding to the improved velocity distribution.


To investigate the flow of hot gases in the full scale ESP, it was convenient to construct a relatively small transparent acrylic model, and to examine the flow of water inside that model. It was convenient because

  • the model was much smaller (thus cheaper, easier to handle) using water rather than a gas,
  • it is easier to seed water than gas for the Laser Doppler Velocimeter, and
  • flow visualisation is easier with dye in water than with a gas.

However, it was essential to first establish that such modelling was valid. This required that two criteria were satisfied:

Geometric similarity. Each model was made an exact geometric model of the full scale unit, except for screens, which were made with the same proportion of open area, but relatively bigger holes (for reasons explained below). Surface roughness was small in the full scale units, and was simulated by smooth acrylic.

Kinematic similarity. It was required that ratios of velocities in each model were the same at corresponding points as in the full scale ESP - that is, we required kinematic similarity between the model and the full scale unit. This was achieved by ensuring that the Reynolds number was sufficiently high in the model to give turbulent flow everywhere that the full scale unit had turbulent flow. Further, the Reynolds numbers in each model were made high enough to give pressure loss coefficients similar to the full scale unit.

The Reynolds number is the ratio of inertia forces to viscous forces for the fluid (water or gas). For example, for a pipe flow, the Reynolds number, ReD, is given by


where U = fluid average velocity (m.s-1),

D= pipe diameter (m), and

n = fluid kinematic viscosity (m2.s-1).

At low enough fluid speeds, Reynolds numbers are low and viscous forces dominate inertia forces, so that the flow is laminar, but the flow becomes turbulent above some critical fluid speed, which depends upon the geometry. For a pipe flow, the critical Reynolds number is approximately 2200.

In the full scale units and the models, Reynolds numbers were calculated for inlet pipes, diffusers and holes in screens to ensure that the models were kinematically similar to the full scale units and that loss coefficients were similar. Reynolds numbers in the model were of the same order of magnitude as in full scale even though the models were much smaller than full scale because the viscosity of the working fluid (water) was approximately 40 times smaller than the viscosity of the hot gases in the full scale units. Minimum desired Reynolds numbers for components in the models were taken from Miller (1990). Some of these are listed in Table 1.

Table 1

Typical Reynolds numbers in ESPs

Flow feature

Typical full scale Reynolds number

Minimum desired Reynolds number

Inlet pipe bend

600 000

200 000

Wide angle diffuser

600 000

10 000

Screen holes

2 200

1 000

Model screen hole diameters (for thin screens) were increased where necessary to achieve Reynolds numbers (based on the hole diameter) of at least 1000, whilst the fraction of open area was kept the same as in the full scale units. This was to ensure that the screen pressure loss coefficients in the models were equal to those in the full scale units.


The geometry of an ESP which was modelled at AFDL is shown schematically in Figure.2 below. Upstream of the ESP, there was a multi-clone which led into the pipe shown, then the flow traversed a bend, a circular-to-rectangular transition (which also expanded in cross-sectional area), and a short wide-angled diffuser containing a partial screen and a full screen.

Figure 2

Schematic views of the ductwork upstream of an Electrostatic Precipitator

The ESP Height:Width:Length was approximately 1:1:0.7. An LDV was used to measure the velocity of the water flowing through the model ESP chamber (without collector electrodes). Measurements were taken on an array of 144 grid points covering the plane parallel to the screen at entry to the ESP and just downstream of it (where the collector electrodes would begin). A contour map of velocity (scaled to full scale) in the ESP in its original configuration is shown in Figure.3 below. Also, the flow was visualised by injecting blue dye into the water at appropriate points. This assisted in understanding where flow separations occurred, and thus modifications could be focussed on improving the flow at those points.

Figure 3

Velocity distribution in the ESP chamber in the original configuration


Dust is collected by the collecting plates in an ESP more efficiently when the gas velocity through the ESP is low, rather than high. This assertion is based on an equation developed at Flakt by Matts (1963), where the fractional collection efficiency, e , is related to the collecting area and the through flow volume rate by the equation


where SCA = the specific collecting area of the dust collecting plates, m2.(m3.s-1)-1

= surface area of collecting plates / gas volume rate passing them,

wK = the migration velocity of particles towards the collecting plates m.s-1,

e = fractional efficiency = 1- mOUT / mIN ,

mIN = mass rate of dust entering the ESP, and

mOUT = mass rate of dust discharged by the ESP.

The migration velocity, wK , has been found to be constant over a wide range of values of SCA, and thus it follows from (2) that as the gas volume rate reduces and the SCA increases, then the collection efficiency increases.

Therefore, for a given gas volume rate through an ESP, the best dust collection efficiency will be obtained with a uniform velocity distribution.


First, the migration velocity, wK , for the full scale ESP in its original configuration was calculated by an iterative technique based on applying (2) to the measured velocity distribution in the model in the original configuration. At each step in the iteration, the dust collection efficiency for the full scale ESP was estimated for a given migration velocity as detailed in the basic calculation method below. Then, by a simple root-finding method, the migration velocity was found for which the calculated collection efficiency matched the measured full scale value.

Next, that migration velocity was used together with velocity distributions measured in the model for alternate geometric configurations to estimate the full scale ESP dust collection efficiency for those alternate configurations.

5.1 Basic Calculation Method

The sequence of steps detailed below was followed in order to obtain a value for the fractional collection efficiency, e , for the whole ESP, corresponding to a given value of migration velocity, wk , and for a given model velocity distribution.

Measurements of the velocity distribution into the ESP in the model were made at a grid of hundreds of points covering the cross section at the inlet to the ESP. The grid of measurements is shown in Figure 4 below. For the purpose of this analysis, the flow into the ESP was therefore considered as consisting of an array of streamtubes, each centred upon the location of a velocity measurement.

Each streamtube had a known cross-sectional area and a known through velocity. Therefore, for each streamtube, the gas volume rate passing through it was calculated. Further, by assuming that the collecting plates were uniformly spread over the entire cross-sectional area, then each streamtube was allocated a proportion of the total collecting plate area corresponding to its proportion of the cross-sectional area. Thus, it was then possible to calculate the specific collecting area for each streamtube.

Figure 4

Mesh of measurement locations defining streamtubes for the dust collection analysis

For each streamtube, for a given migration velocity, wk , equation (2) allowed the fractional collection efficiency, e , for that streamtube to be calculated from the known specific collection area of that streamtube. This is permissible because the migration velocity is considered constant over a wide range of values of SCA.

By assuming that the inlet dust load (g.m-3) for each streamtube was the same, and equal to the average for the whole ESP, then the outlet dust load for each streamtube was calculated using its value of e . For each streamtube, using the known gas volume rate through it, the outlet dust rate (g.s-1) through it was then calculated. For streamtubes with negative velocity, the dust collection was considered equal to zero.

Adding the outlet dust rates (g.s-1) of each streamtube gave the total outlet dust rate for the ESP. Dividing that figure by the gas volume rate for the whole ESP gave the total outlet dust load (g.m-3), from which the fractional collection efficiency for the whole ESP was also calculated.

5.2 ESP Performance for Uniform Velocity Distribution

The collection efficiency for an ideal uniform velocity distribution was calculated from equation (2) using the migration velocity for the ESP in its original configuration as calculated by the method described above. For the specific collecting area (SCA), the total surface area of collecting electrode plates (CE plates) was divided by the total gas volume rate through the ESP.


As shown in Figure.3, the velocity distribution of the gas entering the ESP chamber was far from uniform. Indeed, there were reverse flow regions at the bottom on each side, and jets in the centre of the chamber at top and bottom. By applying the analysis described above to that velocity distribution, a migration velocity of 0.76 m.s-1 was calculated using the quantities listed in Table 1 for the full scale ESP. Also included in that table is the collection efficiency for an ideal velocity distribution, which was calculated to rise from the original 99.2 - 99.4 % to 99.89 % - that is, the dust escaping would reduce from 0.6 - 0.8 % to 0.11 %.

Table 2

Full scale quantities for the ESP in its original configuration




Full scale gas rate

39.3 m3.s-1

Temperature 260 C

Total CE plate area

2406 m2

Collecting electrode plates

Original collection efficiency

99.2 - 99.4 %

Say 99.3 %

Calculated migration velocity

0.76 m.s-1

At 99.3 % efficiency

Ideal collection efficiency

99.89 %

With uniform velocity


With the model ESP, various different screens and vanes were inserted into the ductwork upstream of the ESP chamber in an attempt to improve the velocity distribution. This work included flow visualisation with dye, to identify details of the flow, together with LDV traverses to measure the velocity distribution at the entry to the ESP chamber and application of the theory of screen performance. Eventually an improved distribution was obtained with a horizontal splitter vane in the inlet bend, together with three full height screens in the wide-angled diffuser.

The velocity distribution measured in the model with the improved ESP configuration is shown below in Figure.5. This shows that the negative velocity zones were removed, and the velocity peaks were clipped. This velocity distribution yielded a prediction of dust collection efficiency of 99.74 % (ie. 0.26 % escapes compared with the original 0.7 % ).

After the model testing had been completed, the modifications were implemented in the full scale ESP at Queensland Alumina at Gladstone. Measurements of the full scale dust collection efficiency for the modified ESP showed an improvement from the original 99.3 % to 99.65 % - ie. the dust escaping had been reduced by 50 %. This was close to the prediction of 99.74 % made by the method described above on the basis of the velocity distribution measured in the model.

Figure 5

Velocity distribution in the ESP for the improved configuration


Other factors which may have influenced the change of efficiency of the ESP, include dust picked up in the hoppers under the collecting plates, dust re-entrained when the sheets of dust are rapped off the plates, and dust laden gas by-passing below and above the plates.

Conclusions drawn from this work are :

  • Velocity distributions entering electrostatic precipitators are quite non-uniform, and may include regions of reversed flow because exceedingly wide-angled diffusers are used in the upstream ductwork.
  • Both theoretically and in practice, modifying the ductwork upstream of the ESP to reduce those non-uniformities in velocity leads to improved dust collection.
  • Small scale physical models of the ESPs are a valid and convenient means to develop improvements in velocity distributions by changes to the ductwork..
  • Improvements in full scale dust collection have been accurately predicted on the basis of velocity distributions measured in small scale physical models coupled with the analysis described in this paper.


Matts, S., and Ohnfeldt, P.O., (1963-1964). Efficient gas cleaning with electrostatic precipitators. Flakten/SF Review (Second edition).

Miller, D.S., (1990). Internal Flow Systems (Second Edition) , BHRA (Information Services).

Welsh, M.C., Pullum, L., Downie, R.J., Cooper, P.I. and Blackburn, H. (1996). The role of small-scale physical modelling of fluid dynamic processes in mineral processing. Fourth International Alumina Quality Workshop, Darwin 2-7 June 1996.