Dust Explosions in the Process Industries Second Edition phần 8 docx

Sizing of dust explosion vents 453

Similar cyclone explosion experiments were conducted in Japan more recently by

Hayashi and Matsuda (1988). Their apparatus is illustrated in Figure 6.13.

The volume of the cyclone vessel was 0.32 m3, its total height 1.8 m and the diameter of

the upper cylindrical part 0.6 m. Dust clouds were blown into the cyclone through a

150 mm diameter duct. The desired dust concentration was acquired by independent

control of the air flow through the duct (suction fan at downstream end of system), and the

dust feeding rate into the air flow. The dust trapped in the cyclone dropped into a 0.15 m3

dust collecting chamber bolted to the bottom outlet. The exhaust duct of 0.032 m2 cross

section and 3 m length ended in a 0.73 m3 cubical quenching box fitted with two vents of

0.3 m2 and 0.1 m2 respectively. The venting of the cyclone itself was through the 0.032 m2

Figure 6.1 3

industrial conditions (From Hayashi and Matsuda, 1988)

Experimental cyclone plant for studying dust explosion development under realistic

exhaust duct and the almost 10 m long 0.008 m2 dust feeding duct. During explosion

experiments two water spraying nozzles for flame quenching were in operation in the

exhaust duct in order to protect the fan just outside the quenching box. The ignition source

was a 5 kJ chemical igniter located in the dust feeding duct about 2 m upstream of the

cyclone. Two different polymer dusts were used in the experiments, namely an ABS resin

dust of median particle size 180 pm, and an ethylene-vinyl acetate copolymer dust (EVA)

of median particle size 40 pm.

In addition to the realistic ‘dynamic’ explosion experiments, Hayashi and Matsuda

(1988) conducted a series of experiments with the same two dusts, using an artificial ‘static’

454 Dust Explosions in the Process industries

dust cloud generation method, very similar to that used in the experiments being the basis

of the VDI 3673 (1979 edition). As illustrated in Figure 6.14, the dust feeding duct was

then blocked at the entrance to the cyclone, which reduced the effective vent area slightly,

to 0.032 m2.

Figure 6.1 4 0.32 m3 cyclone modified for gen￾eration of dust clouds by high-pressure injection

through perforated dust dispersion tubes (From

Hayashi and Matsuda, 1988)

A system of two pressurized dust reservoirs and perforated tube dispersion nozzles were

employed for generating the dust clouds. The 5 kJ ignition source was located inside the

cyclone, half way up on the axis (indicated by X2). The ignition source was activated about

100 ms after onset of dust dispersion.

Envelopes embracing the results of both series of experiments are given in Figure 6.15.

As can be seen, the artificial ‘static’ method of dust dispersion gave considerably higher

maximum explosion pressures in the cyclone, than the realistic ‘dynamic’ method. This is

in accordance with the results of the earlier realistic cyclone experiments of Tonkin and

Berlemont (1972). It is of interest to compare the ‘static’ results in Figure 6.15 with

predictions by VDI 3673 (1979 edition). A slight extrapolation of the nomographs to

0.32 m2 vent area, assuming St 1 dusts, gives an expected maximum overpressure of about

2.5 bar(g), which is of the same order as the highest pressures of 1.5 bar(g) measured for

Sizing of dust explosion vents 455

Figure 6.1 5 Results from vented dust explosions in a

0.32 rn3 cyclone using two different polymer dusts

and two different methods of dust cloud generation.

0.03-0.04m2 open vents with ducts. Data from

Hayashi and Matsuda (I 988) (From Eckhoff, 1990)

the artificial ‘static’ dust dispersion method, and much higher than the pressures measured

in the realistic experiments.

The NFPA 68 (1988 edition) includes an alternative nomograph which covers all St 1

dusts that do not yield higher P,, in standard closed bomb tests than 9 bar(g). This

nomograph gives much lower Pred values than the standard nomograph, in particular for

small volumes. In the case of the 0.32 m3 cyclone with a 0.032 m2 vent, the alternative

nomograph gives Pred equal to 0.50 bar(g), which in fact is close to the realistic

experimental values. This alternative nomograph originates from Bartknecht (1987), and

represents a considerable liberalization, by a factor of two or so, of the vent area

requirements for most St 1 and St 2 dusts. However, the scientific and technical basis for

this liberalization does not seem to have been fully disclosed in the open literature.

6.2.5

REALISTIC EXPERIMENTS IN BAG FILTERS

6.2.5.1

Vented explosions in a 6.7 m3 industrial bag filter unit in UK

Lunn and Cairns (1985) reported on a series of dust explosion experiments in a 6.7 m3

industrial bag filter unit. The experiments were conducted during normal operation of the

filter, which was of the pulsed-air, self-cleaning type. Four different dusts were used, and

their Ks, values were determined according to IS0 (1985) (see Chapter 7). The ignition

source was located in the hopper below the filter bag section. In the experiments of main

interest here, the vent was in the roof of the filter housing. Hence, in order to get to the

456 Dust Explosions in the Process Industries

vent, the flame had to propagate all the way up from the hopper and through the

congested filter bag section. The results from the experiments are summarized in Figure

6.16, together with the corresponding VDI 3673 (1979 edition) predictions.

Figure 6.16 first shows that the Pred in the actual filter explosions were mostly

considerably lower than the corresponding VDI 3673 predictions and close to the

theoretical minimum value 0.1 bar(g) at which the vent cover ruptured. Secondly, there is

no sensible correlation between the VDI 3673 ranking of expected pressures according to

the Ks, values, and the ranking actually found.

Figure 6.16 Maximum explosion pressures Prd

measured in dust explosions in an industrial 6.7 m3

bag filter unit in normal operation. P,,, = 0.1 bar(g).

Data from Lunn and Cairns (1985). Comparison

with VDI 3673 (7 979 edition) (From Eckhoff, 1990)

Lunn and Cairns (1985) also reported on a series of dust explosion experiments in a

generously vented 8.6 m3 empty horizontal cylindrical vessel of LID = 6. The same dusts

were used as in the filter experiments, but the dust clouds were generated ‘artificially’ by

injection from pressurized reservoirs as in the standard VDI 3673 method. In spite of the

similarity between the dust dispersion method used and the VDI 3673 dispersion method,

there was no correlation between Pred and Ksr.

6.2.5.2

Vented explosions in a 5.8 m3 bag filter in Norway

These experiments were reported in detail by Eckhoff, Alfert and Fuhre (1989). A

perspective drawing of the experimental filter is shown in Figure 6.17 and a photograph of

a vented maize starch explosion in the filter in Figure 6.18.

Dust explosions were initiated in the filter during normal operation. A practical

worst-case situation was realized by blowing dust suspensions of the most explosible

concentration into the filter at 35 m/s and igniting the cloud in the filter during injection.

Four dusts were used, namely, maize starch and peat dust, both having Ksr = 115 bar m/s,

and polypropylene and silicon dusts, both having Kst = 125 bar m/s. Considerable effort

was made to identify worst-case conditions of dust concentration, and ignition-timing. At

these conditions, experimental correlations of vent area and Pred were determined for

each dust.

Sizing of dust explosion vents 457

Figure 6.17 5.8 m3 experimental bag filter in Norway (from Eckhoff, Alfert and Fuhre, 19891

Figure 6.18 Maize starch explosion in 5.8 m3 experimental bag filter unit in Norway. Vent area

0.16 m2. Static opening pressure of vent cover 0.10 bar(@. Maximum explosion pressure 0.15 bar@).

for a much clearer picture see colour plate 8

458 Dust Explosions in the Process Industries

As shown in Figure 6.19, the peat dust gave significantly lower explosion pressures than

those predicted by VDI 3673 (1979), even if the predictions were based on the volume of

the dusty filter section (3.8 m3) only.

Figure 6.19 Results from vented peat dust

explosions in a 5.8 m3 filter at P,,, = 0.1 bar(@.

Comparison with VDI 3673 (I 979 edition) and

vent sizing method used in Norway (Eckhoff

(1 988)). Injected dust concentration 600 g/m3.

e = dusty section of filter, 0 = clean section of

filter (From Eckhoff, 1990)

Figure 6.20 summarizes the results for all the four dusts. As can be seen, the explosion

pressures measured were generally considerably lower than those predicted by VDI 3673

(1979 edition) for all the four dusts as long as the ignition source was a nitrocellulose

flame. However, the singular result obtained for silicon dust ignited by a silicon dust flame

emphasizes the different nature of initiation and propagation of metal dust flames, as

compared with flames of organic dusts. (See discussion by Eckhoff, Alfert and Fuhre

(1989), and Chapter 4.)

As illustrated by Figure 6.19, Pred scattered considerably, even when the nominal

experimental conditions were identical. This again illustrates the risk-analytical aspect of

the vent sizing problem (see Section 6.6). Figure 6.19 suggests that VDI 3673 is quite

conservative, whereas the method used in Norway is quite liberal, in agreement with the

picture in Figure 6.3.

In Figures 6.20 and 6.21 the 5.8 rn3 filter results for all four dusts are plotted as functions

of Ksr from 1 m3 IS0 standard tests, and (dPldt),,, from Hartmann bomb tests. (See

Chapter 7.)

Predictions by various vent sizing methods have also been included for comparison. The

data in Figure 6.20 show poor correlation between the maximum explosion pressures

measured in the filter at a given vent area, and the maximum rates of pressure rise

determined in standard laboratory tests. Although the Kst values of the four dusts were

very similar, ranging from 115 to 125 bar ds, the Pred (nitrocellulose flame ignition) for

the four dusts varied by a factor of two to three.

In the case of the Hartmann bomb Figure 6.21 indicates a weak positive correlation

between Pred and (dPldt),,, for nitrocellulose ignition, but it is by no means convincing.

Figure 6.21 also gives the corresponding correlations predicted by three different vent

sizing methods based on Hartmann bomb tests. Both the Swedish and the Norwegian

methods are quite liberal. The Rust method oversizes the vents for the organic dusts

excessively for (dPldt),,, > 150 bark There is, however, fair agreement with the data for

silicon dust ignited by a silicon dust flame.