- Email: [email protected]
Scaling of dust explosion violence from laboratory scale to full industrial scale – A challenging case history from the past
Journal of Loss Prevention in the Process Industries xxx (2015) 1e10
Contents lists available at ScienceDirect
Journal of Loss Prevention in the Process Industries journal homepage: www.elsevier.com/locate/jlp
Scaling of dust explosion violence from laboratory scale to full industrial scale e A challenging case history from the past Rolf K. Eckhoff University of Bergen, Bergen, Norway
a r t i c l e i n f o
a b s t r a c t
Article history: Received 13 August 2014 Received in revised form 22 December 2014 Accepted 29 December 2014 Available online xxx
The standardized KSt parameter still seems to be widely used as a universal criterion for ranking explosion violence to be expected from various dusts in given industrial situations. However, this may not be a generally valid approach. In the case of dust explosion venting, the maximum pressure Pmax generated in a given vented industrial enclosure is not only influenced by inherent dust parameters (dust chemistry including moisture, and sizes and shapes of individual dust particles). Process-related parameters (degree of dust dispersion, cloud turbulence, and dust concentration) also play key roles. This view seems to be confirmed by some results from a series of large scale vented dust explosion experiments in a 500 m3 silo conducted in Norway by CMI, (now GexCon AS) during 1980e1982. Therefore, these results have been brought forward again in the present paper. The original purpose of the 500 m3 silo experiments was to obtain correlations between Pmax in the vented silo and the vent area in the silo top surface, for two different dusts, viz. a wheat grain dust collected in a Norwegian grain import silo facility, and a soya meal used for production of fish farming food. Both dusts were tested in the standard 20-L-sphere in two independent laboratories, and also in the Hartmann bomb in two independent laboratories. Pmax and (dP/dt)max were significantly lower for the soya meal than for the wheat grain dust in all laboratory tests. Because the available amount of wheat grain dust was much larger than the quite limited amount of available soya meal, a complete series of 16 vented silo experiments was first performed with the wheat grain dust, starting with the largest vent area and ending with the smallest one. Then, to avoid unnecessary laborious changes of vent areas, the first experiment with soya dust was performed with the smallest area. The dust cloud in the silo was produced in exactly the same way as with the wheat grain dust. However, contrary to expectations based on the laboratory-scale tests, the soya meal exploded more violently in the large silo than the wheat grain dust, and the silo was blown apart in the very first experiment with this material. The probable reason is that the two dusts responded differently to the dust cloud formation process in the silo on the one hand and in the laboratory-scale apparatuses on the other. This re-confirms that a differentiated philosophy for design of dust explosion vents is indeed needed. Appropriate attention must be paid to the influence of the actual dust cloud generation process on the required vent area. The location and type of the ignition source also play important roles. It may seem that tailored design has to become the future solution for tackling this complex reality, not least for large storage silos. It is the view of the present author that the ongoing development of CFD-based computer codes offers the most promising line of attack. This also applies to design of systems for dust explosion isolation and suppression. © 2014 Elsevier Ltd. All rights reserved.
Keywords: Dust explosions Pressure development Explosion venting Silo cells Wheat grain dust Soya meal
1. Background The research presented in this paper was first presented in the report by Eckhoff et al. (1982). A condensed version was presented by Eckhoff and Fuhre (1984). The project was a joint venture
E-mail addresses: [email protected], [email protected].
between and sponsored by the following agencies: -
Fire research Station, UK Health and Safety Executive, UK National Ports Council, UK Labour Protection Foundation (Arbetarskyddsfonden), Sweden The State Grain Company (Statens Kornforretning), Norway Vaksdal Milling Company, Norway
http://dx.doi.org/10.1016/j.jlp.2014.12.020 0950-4230/© 2014 Elsevier Ltd. All rights reserved.
Please cite this article in press as: Eckhoff, R.K., Scaling of dust explosion violence from laboratory scale to full industrial scale e A challenging case history from the past, Journal of Loss Prevention in the Process Industries (2015), http://dx.doi.org/10.1016/j.jlp.2014.12.020
2
R.K. Eckhoff / Journal of Loss Prevention in the Process Industries xxx (2015) 1e10
- Royal Norwegian Council for Scientific and Industrial Research - Norwegian Fire Insurance Company (Norges Brannkasse) - Chr. Michelsen Institute, Norway The technical/scientific justification for conducting this kind of quite demanding large-scale work was given by Eckhoff (1982). 2. Nature of vented dust explosions 2.1. Basic features The overpressure development with time P(t) in a vented enclosure in which a dust cloud deflagration takes place, is the net result of two simultaneous competing processes: Heating of the dust cloud due to the burning of the dust, causing the pressure to increase. Flow of unburnt and burning dust cloud, and combustion products through the vent opening, causing the pressure to decrease. When considering the two competing processes, predicting the rate of heat generation in the enclosure is by far the most demanding task. Appreciation of this fact was the basic motivation for performing the large-scale silo experiments reported in the present paper. During the 33 years that have elapsed since these experiments were performed, much valuable work has been carried out to develop comprehensive computational tools for predicting the course of dust explosion venting processes. But much work still remains to be done. 2.2. Factors influencing the heat generation rate When trying to assess the instantaneous rate of heat production during a dust explosion in a vented enclosure, several factors play a role: a) chemical composition of the dust, including moisture b) distributions of particle sizes and shapes in the dust, determining the specific surface of the dust in the fully dispersed state c) degree of dust dispersion/agglomeration of the dust particles, determining the effective specific surface relevant to combustion in the dust cloud in the actual industrial situation d) distribution of the dust concentration in the actual cloud e) distribution of the initial turbulence in the actual cloud f) possibility of generation of explosion-induced turbulence in the still unburnt part of the cloud (which also depends on the location of the ignition source).
main material to be stored in the silo is in itself sufficiently fine to give explosible clouds in air, such clouds are most likely to be generated somewhere in the silo whenever new material is discharged into it, whether pneumatically or mechanically. If the main material is coarse, such as grain, explosible clouds may be generated by unburnt dust being blown into the silo by preceding explosions elsewhere in the plant. Dust clouds could, for example, be injected through the various openings close to the silo top. Injection through the hopper exit at the bottom seems a more unlikely scenario. Another process of dust cloud generation could be that dust layers, which have accumulated on the inside of the silo wall and roof, are disturbed and dispersed into a cloud by air blasts or mechanical vibrations induced, for example, by preceding explosions elsewhere in the plant. The identification of possible ignition sources and their likely locations in silo cells is another central problem. Dust flames from preceding explosions entering the silo through various openings are one possibility. Dispersion of smouldering dust deposits in the silo itself is another. The possible roles of electrical and mechanical sparks remain a topic of discussion. The influence of the location of the ignition source in the silo cell on the explosion development in the cell was discussed specifically by Eckhoff (1987). 4. The dusts used in the silo explosion experiments Two different dusts were used. The first was a wheat grain dust collected in the bag filters of the largest Norwegian grain import silo, in Stavanger. The second was a soya meal supplied by another Norwegian company and used for production of fish farming food. Measurements of the moisture content of the wheat grain dust showed an average of 10.5% by weight, and of the soya meal 9%. A number of samples of the two dusts were taken from a representative number of bags on site at Boge and transferred to the test laboratory of Chr. Michelsen Institute in Bergen (presently GexCon AS) in sealed containers for determination of particle size distribution, moisture content, and Pmax and (dP/dt)max (1.2 L Hartmann bomb). Samples were also sent to the Fire Research Station, UK and CibaeGeigy AG, Switzerland, for independent, parallel determinations of Pmax and (dP/dt)max in both the Hartmann bomb
Whereas factors (a) and (b) can be assessed accurately in laboratory tests using representative dust samples, factors (c) to (f) are determined entirely by the actual industrial dust cloud generation process, the internal geometry of the enclosure, and the location of the ignition source. The influence of the latter factors cannot be easily assessed by current laboratory tests. However, as discussed by Eckhoff (1984), laboratory tests have demonstrated the importance of these factors in semi-quantitative terms. 3. How can explosible dusts clouds be generated and accidentally ignited in large silo cells in practice in industry? As discussed by Eckhoff (1987) this question certainly has numerous answers, depending on the actual circumstances. If the
Fig. 1. Particle size distributions of the two test dusts obtained in three different laboratories. From Eckhoff et al. (1982).
Please cite this article in press as: Eckhoff, R.K., Scaling of dust explosion violence from laboratory scale to full industrial scale e A challenging case history from the past, Journal of Loss Prevention in the Process Industries (2015), http://dx.doi.org/10.1016/j.jlp.2014.12.020
R.K. Eckhoff / Journal of Loss Prevention in the Process Industries xxx (2015) 1e10
3
and the 20 1iter sphere. Fig. 1 gives the results from determination of the particle size distributions of the two dusts in three independent laboratories (Fire Research Station, UK, CibaeGeigy, Switzerland, and CMI), all using the Alpine air jet sieve. As can be seen, there was reasonable agreement between the results from all three laboratories, confirming that the median particle size by mass was about 20 mm for the wheat grain dust, and about 50 mm for the soya meal. The results from the explosibility tests in the 20 L sphere at both FRS and CibaeGeigy are given in Fig. 2. Fig. 3 shows the corresponding results from tests in the 1.2 L Hartmann bomb at FRS, using two different types of ignition source, viz. an electric spark and a glowing coil of electrically heated resistance wire. As can be seen, the explosion violence, in terms of (dP/dt)max, was consistently considerably higher for the wheat grain dust than for the soya meal in both explosion bombs. Also the maximum explosion pressure was higher for the wheat grain dust than for the soya meal. The agreement between the results obtained in the various laboratories was quite satisfactory. A summary of the results is given in Table 1. The Hartmann bomb data in the table are those obtained with electric spark ignition. The ratio of the (dP/dt)max values found for the two dusts in the 20 L sphere was about 2.0, and
Fig. 3. Hartmann bomb tests at FRS of the explosibility of the two dusts using two different ignition sources. From Eckhoff et al. (1982).
in Hartmann bomb about 4.5. This means that both apparatuses predicted a significantly lower combustion rate for the soya meal than for the wheat grain dust. The corresponding ratios for Pmax were 1.1 and 1.3. 5. The experimental 500 m3 silo installation
Fig. 2. Results from 20-L-bomb explosibility tests of the two dusts. From Eckhoff et al. (1982).
The experiments discussed in this paper were conducted during the first of two consecutive silo dust explosion programs (1980e1982). A battery of 8 identical semi-condemned 500 m3 steel silo cells at Boge, 45 km east of Bergen, were made available for the two programs. The battery consisted of two parallel cell rows of 4 cells each. Each silo cell had a diameter of about 5.5 m and height of about 21 m, corresponding to a length-to-diameter ratio (L/D) of about 4. All the experiments discussed in the present paper were conducted in one of these cells, as indicated in Fig. 4. An overview of the total experimental set-up with the dust injection facilities is shown in Fig. 5. A cross section of the experimental silo cell with ignition point and diagnostic instrumentation is shown in Fig. 6. An external winding staircase extending to the top of the experimental silo cell not only provided access to the silo top for arranging the required vent opening, but it also permitted easy mounting and inspection of diagnostic instruments at various desired levels above the silo bottom. As illustrated in Fig. 6 four different types of measurements
Please cite this article in press as: Eckhoff, R.K., Scaling of dust explosion violence from laboratory scale to full industrial scale e A challenging case history from the past, Journal of Loss Prevention in the Process Industries (2015), http://dx.doi.org/10.1016/j.jlp.2014.12.020
4
R.K. Eckhoff / Journal of Loss Prevention in the Process Industries xxx (2015) 1e10
Table 1 Summary of explosion properties of the two dusts used in the 500 m3 silo experiments. Dust type
Median particle size by mass (mm)
Moisture content (% by mass)
(dP/dt)max 1.2 L Hartm. bomb (bar/s)
(dP/dt)max 20 L sphere (bar/s)
Pmax Hartmann bomb (bar(g))
Pmax 20 L sphere (bar(g))
Wheat grain Soya meal
20 50
12e15 9
125 27
350 175
5.8 4.3
7.3 6.6
were performed during an explosion in the silo: a) Eight specially constructed dust cloud extraction probes, distributed throughout the silo volume as indicated in Fig. 6, were used in an attempt at measuring the dust concentration distribution in the silo just prior to ignition. These probes were used only in the wheat grain dust experiments. Unfortunately, for aerodynamic reasons, they were only able to collect the finest particles in the dust cloud.
b) Three piezoelectric Kistler No. 412 pressure transducers were used for measuring the development with time of the explosion pressure inside the silo cell. In some tests only the top and bottom transducers were in use, as indicated in Fig. 6. c) Three narrow-angle optical photo-diode probes were used for detecting the flame front arrival times inside the silo cell, as also indicated in Fig. 6.
Fig. 4. Overview of the experimental test site at Boge, Vaksdal, outside Bergen, Norway. From Eckhoff et al. (1982).
Please cite this article in press as: Eckhoff, R.K., Scaling of dust explosion violence from laboratory scale to full industrial scale e A challenging case history from the past, Journal of Loss Prevention in the Process Industries (2015), http://dx.doi.org/10.1016/j.jlp.2014.12.020
R.K. Eckhoff / Journal of Loss Prevention in the Process Industries xxx (2015) 1e10
5
Fig. 5. Overview of the experimental set-up with dust injection facilities. From Eckhoff et al. (1982).
d) Sixteen mm, 24 frame/s movie recordings of the explosions were taken from a convenient position on the hillside about 100 m from the silo top, as indicated in Fig. 4. In order to enable conduction of experiments with different vent areas, a strong steel grid was constructed across the entire top surface of the experimental silo cell, permitting any desired part of it to be blocked by bolting the required number of square 0.5 m 0.5 m steel plates to the grid. The range of vent openings used is illustrated in Fig. 7. 6. Generation of experimental dust clouds in the 500 m3 silo 6.1. The method For practical/financial reasons it was decided to generate the dust clouds in the silo by blowing into it a known quantity of dust by means of a standard 120 Hp “VacuVator” Model PTA 114144 (Dunbar Kapple Inc., Illinois, USA), which the project was allowed to borrow from one of the sponsors without any costs. As indicated in Fig. 6, the dust jet entered the silo vertically upwards approximately at the silo centre. It was estimated that typical air velocities during dust injection into the silo were of the order of 10e15 m/s. To prevent the dust from escaping through the vent opening at the silo top during the injection period, the vent opening was always sealed with a thin sheet of plastic. In this way the dust cloud formation process was also kept independent of the vent area. In all the experiments the dust clouds were practically quiescent at the moment of ignition. This was achieved as follows: During the dust injection period the blower was operated at a steady, high speed until a fairly distinct change of its sound indicated that most of the dust had been blown into the silo. At this point the blower power was reduced to “no load”, and the automatic system for the sequential dust concentration sampling, start of Ampex tape
recorder, and ignition source firing was triggered manually by the blower operator. The duration of this sequence was about 5 s. 6.2. Structures of the experimental dust clouds in the silo Although the same dust injection method was used for both dusts, the resulting dust cloud structures in the silo were most probably quite dissimilar, because of differences in the dispersibility and particle size distributions. In the case of the wheat grain dust, which is difficult to disperse because of the high content of fibrous material, considerable quantities of unburnt dust (up to 100e150 kg) were normally found deposited on the silo bottom after the explosion. This clearly means that a significant fraction of the dust being blown into the silo cell was in the form of large agglomerates, able to settle out of suspension before the dust cloud was ignited. However, this settling process apparently had a significant homogenizing effect on the concentration distribution of the remaining cloud of finely dispersed dust, inasmuch as the automatic dust concentration measurement system revealed a fairly even distribution of the fine dust fractions throughout the entire silo volume, both horizontally and vertically. Contrary to the wheat grain dust, the soya meal was very easy to disperse. However, because of the total collapse of the experimental silo in the very first experiment with this material, no data for the dust concentration distribution at the moment of ignition was obtained. 7. Outline of the dust explosion experiments in the 500 m3 silo Because the quantity of wheat grain dust being available for the experiments was much larger than the available quantity of soya meal, it was decided to start the experiments with the wheat grain dust. Hence, all the first 16 silo explosion experiments were with
Please cite this article in press as: Eckhoff, R.K., Scaling of dust explosion violence from laboratory scale to full industrial scale e A challenging case history from the past, Journal of Loss Prevention in the Process Industries (2015), http://dx.doi.org/10.1016/j.jlp.2014.12.020
6
R.K. Eckhoff / Journal of Loss Prevention in the Process Industries xxx (2015) 1e10
injection. The ignition point was close to the silo bottom, as indicated in Fig. 6. For safety reasons the first of the 16 experiments with the wheat grain dust were conducted with the largest attainable vent opening of 14.2 m2 (see Fig. 7). In the successive experiments the vent opening was reduced in steps, as also illustrated in Fig. 7. After each explosion the silo was inspected for any sign of damage or deformation. The changing of the vent area between experiments was a quite laborious process and it was decided to conduct the experiments with the soya meal by starting with the smallest opening (2.1 m2), with which the experiments with the wheat dust ended, and then increase the vent area stepwise in successive experiments. This decision was based on the results from the explosion violence tests in the 20-L-sphere and the Hartmann bomb given in Figs. 2 and 3 and Table 1. Relying on the commonly accepted stateof-the-art in 1980 it was then just assumed that also the explosion violence to be expected from the soya meal in the big silo would be less severe than that found in the silo with the wheat grain dust.
8. Results from the 500 m3 silo experiments 8.1. Explosion pressure development Two typical traces of explosion pressure versus time during wheat grain dust explosions in the 500 m3 silo are given in Figs. 8 and 9. The frequency of the oscillating traces in Fig. 8 of 3e4 Hz agrees well with the first harmonic of the standing sound wave in the gas column in the silo (silo top open, i.e. 1/4 l ¼ 21 m, and hence l ¼ 84 m, where l is the wave length). Apart from the oscillatory pressure wave pattern obtained with the largest vent area of 14.2 m2, the pressure time histories were generally one single distinct bell-shaped pulse of about 1 s duration, as shown in Fig. 9. Fig. 10 gives the pressure trace from the single destructive soya meal explosion. The shape of this trace may indicate that the maximum explosion pressure that would have resulted, had the silo been sufficiently strong, would perhaps not have been much higher than the 0.6 bar(g) at which the silo ruptured.
8.2. Vertical flame front propagation velocities in the 500 m3 silo
Fig. 6. Cross section of experimental silo cell with ignition source location, point of dust injection and diagnostics. From Eckhoff et al. (1982).
this dust. In all these experiments, and also in the single experiment with the soya meal, about 330 kg of the dust was blown into the silo in the form of a free jet extending upwards in the upper half of the silo volume, as already explained and indicated in Figs. 5 and 6. Downwards migration of the dust cloud below the exit of the dust injection pipe occurred by rapid settling of coarse particles and particle agglomerates. Based on the amount of dust injected, the nominal average dust concentration was about 660 g/m3 both with wheat grain dust and with soya meal. However, before ignition of the dust clouds was activated a significant part of the dust had settled out at the silo bottom, and hence the real average dust concentration in the exploding cloud was somewhat lower than this value. Ignition was effectuated 3e5 s after termination of dust
The upwards vertical flame front speeds in the upper half of the silo were estimated on the basis of the flame-arrival-time data provided by the three flame detectors located as indicated in Fig. 6. A typical trend for the wheat grain dust explosions would be a few m/s halfway down in the silo, and subsequent acceleration to about 40e50 m/s just below the vent opening. It was observed that the duration of the flame signal at a given photo diode station was considerably longer than the time required for the flame to travel from the bottom of the silo to its top. This suggests that the combustion process in the silo was to a large extent volumetric, i.e. the flame was very thick. This opposes idealized models of a thin flame sweeping through the cloud, which was sometimes assumed in the past. Very long burn-out times mean that the maximum rate of heat production in the silo will occur at the moment when the flame reaches the vent, because at this moment the quantity of dust that is burning simultaneously inside the silo is at the maximum. On this basis one would expect a systematic coincidence between the moment of flame arrival at the silo top and the occurrence of the peak pressure in the silo. This was in fact also observed experimentally.
Please cite this article in press as: Eckhoff, R.K., Scaling of dust explosion violence from laboratory scale to full industrial scale e A challenging case history from the past, Journal of Loss Prevention in the Process Industries (2015), http://dx.doi.org/10.1016/j.jlp.2014.12.020
R.K. Eckhoff / Journal of Loss Prevention in the Process Industries xxx (2015) 1e10
7
Fig. 7. Range of vent openings used in the 500 m3 silo experiments. From Eckhoff et al. (1982).
8.3. Dust concentration distributions in the 500 m3 silo immediately prior to ignition An attempt was made at estimating the dust concentration distribution throughout the 500 m3 silo immediately before ignition, by simultaneous measurements at 8 different points. Fig. 6 shows the locations of the 8 dust sampling probes. The technique used was fast extraction of a sample of about 4 L of dust cloud through a small filter. The initial assumption was that all the dust in the sampled cloud volume would be sucked into the probe and hence would be available for subsequent weighing. This would work if the particles were sufficiently small to follow the air stream
into the sampler. However, the wheat grain dust contained a substantial fraction of agglomerates behaving as large single particle units, and hence this condition was not satisfied. Therefore, only a fraction of the dust that was initially in the air volume flowing through the dust sampler, would actually enter the filter and be collected there. This was confirmed in controlled laboratory tests where the ‘yield’ never exceeded 50%. Because of poorer dust dispersion in the silo than in the laboratory tests, the ‘yields’ in the silo tests were probably even less than 50%. Much of the dust that should ideally have entered the filter probe, instead accumulated on the outside of the probe, around its entrance, and thus could not in any reliable way be considered part of the sampled dust. Also, the
Please cite this article in press as: Eckhoff, R.K., Scaling of dust explosion violence from laboratory scale to full industrial scale e A challenging case history from the past, Journal of Loss Prevention in the Process Industries (2015), http://dx.doi.org/10.1016/j.jlp.2014.12.020
8
R.K. Eckhoff / Journal of Loss Prevention in the Process Industries xxx (2015) 1e10
and hence some of the sampled dust occasionally got burnt. In spite of these shortcomings, the dust concentration sampling system did provide an indication of the evenness of the concentration of the fine dust fraction throughout the silo volume, just prior to ignition. Two examples of measured distributions are given in Fig. 11. Systematic radial concentration gradients were not typical, although such gradients did occur in some tests, as shown close to silo bottom in the upper part of Fig. 11. In some experiments even the vertical concentration distribution of the fine dust fraction appeared to be quite homogeneous (lower part of Fig. 11). However, some vertical concentration gradients were observed in most cases, with the maximum concentration occurring close to the silo top
Fig. 8. Typical explosion pressure development in vented 500 m3 silo with wheat grain dust and 14.2 m2vent opening. Pressure-versus-time records from pressure probe 1 (lower trace) and pressure probe 2 (upper trace). For positions of pressure probes see Fig. 6. From Eckhoff et al. (1982).
Fig. 9. Typical explosion pressure development in vented 500 m3 silo with wheat grain dust and 8.6 m2vent opening. Pressure-versus-time records from pressure probe 1 (lower trace) and pressure probe 2 (upper trace). For positions of pressure probes see Fig. 6. From Eckhoff et al. (1982).
Fig. 10. Pressure development in the destructive vented explosion in the 500 m3 silo with soya meal and 2.1 m2 vent opening. Pressure-versus-time record from pressure probe 2. For position of pressure probe see Fig. 6. From Eckhoff et al. (1982).
dust sample probes did not shield the sampled dust inside sufficiently from the heat of the explosion and subsequent fire outside,
Fig. 11. Two examples of distributions the concentration of the finest wheat grain dust fraction in the 500 m3 silo as measured by the dust sampling probes just prior to ignition. From Eckhoff et al. (1982).
Please cite this article in press as: Eckhoff, R.K., Scaling of dust explosion violence from laboratory scale to full industrial scale e A challenging case history from the past, Journal of Loss Prevention in the Process Industries (2015), http://dx.doi.org/10.1016/j.jlp.2014.12.020
R.K. Eckhoff / Journal of Loss Prevention in the Process Industries xxx (2015) 1e10
and/or close to the silo bottom (upper part of Fig. 11). 9. Discussion More than 70 years ago the great American pioneers of dust explosion research, Hartmann et al. (1943) wrote: “Over the past 30 years, various investigators have worked on means of producing uniform dust clouds; comparison of results indicates that none of them has been wholly successful. The mechanism to produce such a cloud, of sufficient volume to be usable in test work, remains to be perfected”. This statement is still valid today, with a slight adjustment: “Over the past 100 years …”. However, during these 100 years another aspect has become even more challenging: If it were at all possible to produce a “perfect” experimental dust cloud, would this help us predicting the course of real industrial dust explosions? Eckhoff (1984) challenged the relevance of using existing standard test methods for predicting the potential explosion violence of real dust clouds in the process industries. Dust clouds encountered in industry are not “perfect” but possess varying structures of dust concentration, degree of dust dispersion and degree of turbulence, which are all results of the prevailing process conditions. This fact has been the main concern of the present paper. As shown above, the big silo installation ruptured and collapsed in the very first experiment with the soya meal. The central question then becomes: Why did the soya meal explode more violently in the silo than the wheat grain dust, when both the Hartmann bomb tests and the 20-L tests beyond any doubt had predicted the opposite trend? The particle size analyses of the two powders, determined by three independent laboratories (see Fig. 1), showed that the wheat grain dust was considerably finer than the soya meal, which gave further a priori support to the validity of ranking the expected explosibilities of the two dusts in a real process situation on the basis of the results in Figs. 2 and 3 and Table 1. The point is that the wheat grain dust and the soya meal undoubtedly responded quite differently to the rather gentle dust cloud formation/dispersion processes in the silo on the one hand, and the much more intense dispersion processes in both the 20-L test bomb and the Alpine air jet sieve, on the other. Because the soya meal mostly consists of easy-to-disperse equi-dimensional particles, it was most probably well dispersed even in the silo experiments. The wheat grain dust, on the other hand, containing an appreciable fraction of both fibres and flakes, most probably was much more difficult to disperse in the silo than the soya meal. In the 20-L apparatus even the wheat grain dust most probably got completely dispersed. It is even quite likely that some of the fibres and flakes were crushed into much smaller particles in this violent dispersion process. The results from the Hartmann bomb tests indicate that similar effects occurred even in this apparatus. The ability of the intense dust dispersion process of the 20-L apparatus to crush the primary particles of some dusts was conclusively demonstrated by Kalejaiye et al. (2006). Some particle crushing would also be expected to occur in the Alpine air jet sieve, in particular with dusts containing fragile fibres and flakes, as wheat grain dust. However, because this was not the case in the silo, a significant proportion of the wheat grain dust particles appeared there as large un-dispersed agglomerates. Therefore the combustion rate of the wheat grain dust clouds in the silo was lower than that of the clouds of the soya meal, contrary to what was predicted by the Hartmann bomb and the 20-L bomb tests. The dispersibility of a dust is a decisive inherent dust property in relation to formation of dust cloud structures and resulting dust cloud burning rates. Therefore, adequate experimental methods for assessing dust dispersibility are required. In Chapter 7 of Eckhoff
9
(2003) some published methods are described. The problem of dust dispersibility in relation to dust cloud formation at large is discussed in Chapter 3 of Eckhoff (2003). It is clear, however, that more research is needed in this area before comprehensive models of dust dispersion processes can be developed. The possibility of developing standard test methods for assessing the dispersibility (“dustiness”) of dusts is currently discussed internationally. There is a continued need for discussing what dust cloud formation and ignition processes and associated combustion rates should, under various circumstances in industry, be regarded as the “worst cases” that the explosion protection, including venting, should be able to accommodate. Fig. 12 shows the marked influence of the mode of dust cloud generation on the maximum pressure Pred in a dust explosion in a vented industrial enclosure. In the VDI (1979) method the dust was injected into the silo at high rate from a number of pre-pressurized bottles mounted along the silo wall from its bottom to its top. Fig. 12 illustrates the need for adopting a differentiated approach to sizing of dust explosion vents. The basic understanding of flame propagation processes inside and outside vented enclosures is still unsatisfactory. Neither the processes by which dust clouds in vented enclosures are generated, nor the way in which the clouds burn, are sufficiently well understood. Consequently, adequate venting theories do not exist, and one must rely on experimental evidence. Recent European standardization does in principle to some extent open up for a differentiated approach to vent sizing, accounting for the variations in dust cloud structures encountered in practice in industry. In most practical cases this will result in more liberal vent area requirements than those put forward in some very rigorous standards. However, it is believed that the ultimate long term solution for design of explosion venting arrangements will be comprehensive computer models. Such models will not only be useful for design of explosion venting systems, but also of systems for dust explosion suppression and isolation. Skjold (2007, 2010, 2014a, 2014b) and Skjold et al. (2014) have summarized the state-of-the-art in the promising effort to develop the CFD-based code DESC aimed at meeting this need.
Fig. 12. Maximum pressures Pred in vented maize starch explosions in a 20 m3 silo of L/ D ¼ 6. From Eckhoff (2003), based on experiments by W. Bartknecht and S. Radandt et al.
Please cite this article in press as: Eckhoff, R.K., Scaling of dust explosion violence from laboratory scale to full industrial scale e A challenging case history from the past, Journal of Loss Prevention in the Process Industries (2015), http://dx.doi.org/10.1016/j.jlp.2014.12.020
10
R.K. Eckhoff / Journal of Loss Prevention in the Process Industries xxx (2015) 1e10
10. Conclusions 1. The standardized KSt parameter still seems to be widely used as a universal criterion for ranking the explosion violence to be expected from various dusts in given industrial situations. However, this may not be a generally valid approach. In the case of dust explosion venting the maximum pressure Pmax generated in a given vented industrial enclosure will not only be a result of inherent dust parameters (dust chemistry including moisture, and sizes and shapes of individual dust particles). Process-related parameters (degree of dust dispersion, cloud turbulence, and dust concentration) also play key roles. 2. In the research conducted during 1980e82, which is reconsidered in the present paper, two different dusts were used, viz. a wheat grain dust and a soya meal dust. Both dusts were tested in the standard 20-L bomb in two independent laboratories (in UK and Switzerland), and also in the Hartmann bomb (in UK and Norway). In all laboratory tests both Pmax and (dP/dt)max were significantly lower for the soya meal than for the wheat grain dust. 3. However, in the 500 m3 vented silo cell the soya meal in fact produced more violent explosions than the wheat grain dust. 4. This gives further support to the view that a differentiated approach to design of dust explosion vents is indeed needed. To some, but insufficient, extent attempts have been made at meeting this need in more recent dust explosion venting codes. As regards very large silos in particular, one still has to rely on rather questionable extrapolation. 5. In future the most likely solution to this complex and challenging reality will be tailored design by means of CFD-based computer codes. This not only applies to design of explosion venting systems, but also to systems for explosion suppression and isolation. In the case of venting, this will be particularly relevant and useful for very large silos of volumes of several thousand m3, for which no experimental data exist. Acknowledgement The author wishes to gratefully acknowledge the sponsoring organizations from both UK, Sweden and Norway, and their enthusiastic representatives, for having provided the support without which the large-scale silo experiments described in this paper would never have become a reality. The author is particularly indebted to his close co-worker during 17 years, Mr. K. Fuhre, who
carried a major responsibility both in the construction of the largescale experimental facility and in the planning and running of the experiments. Sincere thanks are also expressed to the local population of Boge outside Bergen, Norway, for their continued understanding and forbearance in connection with the rather unusual activity that took place in the early 1980es in their otherwise very quiet and pleasant local community. References Eckhoff, R.K., 1982. Current dust explosion research at the CMI. In: Lee, J.H.S., Guirao, C.M. (Eds.), Proceedings of the International Conference on FueleAir Explosions Held at McGill University, Montreal, Canada, 4e6 November 1981. University of Waterloo Press, pp. 657e678. Eckhoff, R.K., Fuhre, K., Henery, M.J., Parker, S.J., Thorsen, H.G., 1982. Dust Explosion Experiments in a Vented 500 m3 Silo Cell. CMI-Report (Chr. Michelsen Institute, Bergen, Norway, presently: GexCon AS, Bergen, Norway) No. 813307e1. Eckhoff, R.K., 1984. Relevance of using (dP/dt)max data from laboratory-scale tests for predicting explosion rates in practical industrial situations. VDI-Berichte 494. In: Proceedings of the International VDI Colloquium “Safe Handling of Flammable Dusts” in Nürnberg, Germany October 26e28, 1983. Verein Deutscher Ingenieure, Düsseldorf, Germany, pp. 207e217. Eckhoff, R.K., Fuhre, K., 1984. Dust explosion experiments in a vented 500 m3 silo cell. J. Occup. Accid. 6, 229e240. Eckhoff, R.K., 1987. A differentiated approach to sizing of dust explosion vents: influence of ignition source location with particular reference to large, slender silos. In: Proceedings of Symposium on Industrial Dust Explosions. ASTM, Philadelphia, PA, USA, pp. 265e280. ASTM Publication Code Number 04958000-31. Eckhoff, R.K., 2003. Dust Explosions in the Process Industries, third ed. Gulf Professional Publishing, an imprint of Elsevier Science, Amsterdam. Hartmann, I., Nagy, J., Brown, H.R., 1943. Inflammability and Explosibility of Metal Powders. United States Department of the Interior, Bureau of Mines. Report of Investigations No. 3722. Kalejaiye, O., Amyotte, P.R., Pegg, M.J., 2006. Effectiveness of dust dispersion in the 20-l Siwek chamber. In: Proceedings of 6th International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions, Aug. 27eSept. 1, vol. 1. Dalhousie University, Halifax, NS, Canada, pp. 253e278. Skjold, T., 2007. Review of the DESC project. J. Loss Prev. Process Indust. 20, 291e302. Skjold, T., 2010. Flame propagation in dust clouds: challenges for model validation. In: Proceedings Eighth International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions (ISHPMIE), Yokohama, p. 11. Skjold, T., 2014a. Simulating vented maize starch explosions in a 236 m3 silo. In: Eleventh International Symposium on Fire Safety Science, University of Canterbury, New Zealand, 10e14 February 2014. International Association for Fire Safety Science (IAFSS). Skjold, T., 2014b. Flame Propagation in Dust Clouds. Numerical and Experimental Investigation (Ph.D. thesis). University of Bergen, Norway. Skjold, T., Castellanos, D., Lien, K.O., Eckhoff, R.K., 2014. Experimental and numerical investigations of constant volume dust and gas explosions in a 3.6 metre flame acceleration tube. J. Loss Prev. Process Indust. 30, 164e176. VDI, 1979. Druckentlastung von Staubexplosionen. VDI-Richtlinie No. 3673. Verein Deutscher Ingenieure, Düsseldorf, Germany.
Please cite this article in press as: Eckhoff, R.K., Scaling of dust explosion violence from laboratory scale to full industrial scale e A challenging case history from the past, Journal of Loss Prevention in the Process Industries (2015), http://dx.doi.org/10.1016/j.jlp.2014.12.020
Contents lists available at ScienceDirect
Journal of Loss Prevention in the Process Industries journal homepage: www.elsevier.com/locate/jlp
Scaling of dust explosion violence from laboratory scale to full industrial scale e A challenging case history from the past Rolf K. Eckhoff University of Bergen, Bergen, Norway
a r t i c l e i n f o
a b s t r a c t
Article history: Received 13 August 2014 Received in revised form 22 December 2014 Accepted 29 December 2014 Available online xxx
The standardized KSt parameter still seems to be widely used as a universal criterion for ranking explosion violence to be expected from various dusts in given industrial situations. However, this may not be a generally valid approach. In the case of dust explosion venting, the maximum pressure Pmax generated in a given vented industrial enclosure is not only influenced by inherent dust parameters (dust chemistry including moisture, and sizes and shapes of individual dust particles). Process-related parameters (degree of dust dispersion, cloud turbulence, and dust concentration) also play key roles. This view seems to be confirmed by some results from a series of large scale vented dust explosion experiments in a 500 m3 silo conducted in Norway by CMI, (now GexCon AS) during 1980e1982. Therefore, these results have been brought forward again in the present paper. The original purpose of the 500 m3 silo experiments was to obtain correlations between Pmax in the vented silo and the vent area in the silo top surface, for two different dusts, viz. a wheat grain dust collected in a Norwegian grain import silo facility, and a soya meal used for production of fish farming food. Both dusts were tested in the standard 20-L-sphere in two independent laboratories, and also in the Hartmann bomb in two independent laboratories. Pmax and (dP/dt)max were significantly lower for the soya meal than for the wheat grain dust in all laboratory tests. Because the available amount of wheat grain dust was much larger than the quite limited amount of available soya meal, a complete series of 16 vented silo experiments was first performed with the wheat grain dust, starting with the largest vent area and ending with the smallest one. Then, to avoid unnecessary laborious changes of vent areas, the first experiment with soya dust was performed with the smallest area. The dust cloud in the silo was produced in exactly the same way as with the wheat grain dust. However, contrary to expectations based on the laboratory-scale tests, the soya meal exploded more violently in the large silo than the wheat grain dust, and the silo was blown apart in the very first experiment with this material. The probable reason is that the two dusts responded differently to the dust cloud formation process in the silo on the one hand and in the laboratory-scale apparatuses on the other. This re-confirms that a differentiated philosophy for design of dust explosion vents is indeed needed. Appropriate attention must be paid to the influence of the actual dust cloud generation process on the required vent area. The location and type of the ignition source also play important roles. It may seem that tailored design has to become the future solution for tackling this complex reality, not least for large storage silos. It is the view of the present author that the ongoing development of CFD-based computer codes offers the most promising line of attack. This also applies to design of systems for dust explosion isolation and suppression. © 2014 Elsevier Ltd. All rights reserved.
Keywords: Dust explosions Pressure development Explosion venting Silo cells Wheat grain dust Soya meal
1. Background The research presented in this paper was first presented in the report by Eckhoff et al. (1982). A condensed version was presented by Eckhoff and Fuhre (1984). The project was a joint venture
E-mail addresses: [email protected], [email protected].
between and sponsored by the following agencies: -
Fire research Station, UK Health and Safety Executive, UK National Ports Council, UK Labour Protection Foundation (Arbetarskyddsfonden), Sweden The State Grain Company (Statens Kornforretning), Norway Vaksdal Milling Company, Norway
http://dx.doi.org/10.1016/j.jlp.2014.12.020 0950-4230/© 2014 Elsevier Ltd. All rights reserved.
Please cite this article in press as: Eckhoff, R.K., Scaling of dust explosion violence from laboratory scale to full industrial scale e A challenging case history from the past, Journal of Loss Prevention in the Process Industries (2015), http://dx.doi.org/10.1016/j.jlp.2014.12.020
2
R.K. Eckhoff / Journal of Loss Prevention in the Process Industries xxx (2015) 1e10
- Royal Norwegian Council for Scientific and Industrial Research - Norwegian Fire Insurance Company (Norges Brannkasse) - Chr. Michelsen Institute, Norway The technical/scientific justification for conducting this kind of quite demanding large-scale work was given by Eckhoff (1982). 2. Nature of vented dust explosions 2.1. Basic features The overpressure development with time P(t) in a vented enclosure in which a dust cloud deflagration takes place, is the net result of two simultaneous competing processes: Heating of the dust cloud due to the burning of the dust, causing the pressure to increase. Flow of unburnt and burning dust cloud, and combustion products through the vent opening, causing the pressure to decrease. When considering the two competing processes, predicting the rate of heat generation in the enclosure is by far the most demanding task. Appreciation of this fact was the basic motivation for performing the large-scale silo experiments reported in the present paper. During the 33 years that have elapsed since these experiments were performed, much valuable work has been carried out to develop comprehensive computational tools for predicting the course of dust explosion venting processes. But much work still remains to be done. 2.2. Factors influencing the heat generation rate When trying to assess the instantaneous rate of heat production during a dust explosion in a vented enclosure, several factors play a role: a) chemical composition of the dust, including moisture b) distributions of particle sizes and shapes in the dust, determining the specific surface of the dust in the fully dispersed state c) degree of dust dispersion/agglomeration of the dust particles, determining the effective specific surface relevant to combustion in the dust cloud in the actual industrial situation d) distribution of the dust concentration in the actual cloud e) distribution of the initial turbulence in the actual cloud f) possibility of generation of explosion-induced turbulence in the still unburnt part of the cloud (which also depends on the location of the ignition source).
main material to be stored in the silo is in itself sufficiently fine to give explosible clouds in air, such clouds are most likely to be generated somewhere in the silo whenever new material is discharged into it, whether pneumatically or mechanically. If the main material is coarse, such as grain, explosible clouds may be generated by unburnt dust being blown into the silo by preceding explosions elsewhere in the plant. Dust clouds could, for example, be injected through the various openings close to the silo top. Injection through the hopper exit at the bottom seems a more unlikely scenario. Another process of dust cloud generation could be that dust layers, which have accumulated on the inside of the silo wall and roof, are disturbed and dispersed into a cloud by air blasts or mechanical vibrations induced, for example, by preceding explosions elsewhere in the plant. The identification of possible ignition sources and their likely locations in silo cells is another central problem. Dust flames from preceding explosions entering the silo through various openings are one possibility. Dispersion of smouldering dust deposits in the silo itself is another. The possible roles of electrical and mechanical sparks remain a topic of discussion. The influence of the location of the ignition source in the silo cell on the explosion development in the cell was discussed specifically by Eckhoff (1987). 4. The dusts used in the silo explosion experiments Two different dusts were used. The first was a wheat grain dust collected in the bag filters of the largest Norwegian grain import silo, in Stavanger. The second was a soya meal supplied by another Norwegian company and used for production of fish farming food. Measurements of the moisture content of the wheat grain dust showed an average of 10.5% by weight, and of the soya meal 9%. A number of samples of the two dusts were taken from a representative number of bags on site at Boge and transferred to the test laboratory of Chr. Michelsen Institute in Bergen (presently GexCon AS) in sealed containers for determination of particle size distribution, moisture content, and Pmax and (dP/dt)max (1.2 L Hartmann bomb). Samples were also sent to the Fire Research Station, UK and CibaeGeigy AG, Switzerland, for independent, parallel determinations of Pmax and (dP/dt)max in both the Hartmann bomb
Whereas factors (a) and (b) can be assessed accurately in laboratory tests using representative dust samples, factors (c) to (f) are determined entirely by the actual industrial dust cloud generation process, the internal geometry of the enclosure, and the location of the ignition source. The influence of the latter factors cannot be easily assessed by current laboratory tests. However, as discussed by Eckhoff (1984), laboratory tests have demonstrated the importance of these factors in semi-quantitative terms. 3. How can explosible dusts clouds be generated and accidentally ignited in large silo cells in practice in industry? As discussed by Eckhoff (1987) this question certainly has numerous answers, depending on the actual circumstances. If the
Fig. 1. Particle size distributions of the two test dusts obtained in three different laboratories. From Eckhoff et al. (1982).
Please cite this article in press as: Eckhoff, R.K., Scaling of dust explosion violence from laboratory scale to full industrial scale e A challenging case history from the past, Journal of Loss Prevention in the Process Industries (2015), http://dx.doi.org/10.1016/j.jlp.2014.12.020
R.K. Eckhoff / Journal of Loss Prevention in the Process Industries xxx (2015) 1e10
3
and the 20 1iter sphere. Fig. 1 gives the results from determination of the particle size distributions of the two dusts in three independent laboratories (Fire Research Station, UK, CibaeGeigy, Switzerland, and CMI), all using the Alpine air jet sieve. As can be seen, there was reasonable agreement between the results from all three laboratories, confirming that the median particle size by mass was about 20 mm for the wheat grain dust, and about 50 mm for the soya meal. The results from the explosibility tests in the 20 L sphere at both FRS and CibaeGeigy are given in Fig. 2. Fig. 3 shows the corresponding results from tests in the 1.2 L Hartmann bomb at FRS, using two different types of ignition source, viz. an electric spark and a glowing coil of electrically heated resistance wire. As can be seen, the explosion violence, in terms of (dP/dt)max, was consistently considerably higher for the wheat grain dust than for the soya meal in both explosion bombs. Also the maximum explosion pressure was higher for the wheat grain dust than for the soya meal. The agreement between the results obtained in the various laboratories was quite satisfactory. A summary of the results is given in Table 1. The Hartmann bomb data in the table are those obtained with electric spark ignition. The ratio of the (dP/dt)max values found for the two dusts in the 20 L sphere was about 2.0, and
Fig. 3. Hartmann bomb tests at FRS of the explosibility of the two dusts using two different ignition sources. From Eckhoff et al. (1982).
in Hartmann bomb about 4.5. This means that both apparatuses predicted a significantly lower combustion rate for the soya meal than for the wheat grain dust. The corresponding ratios for Pmax were 1.1 and 1.3. 5. The experimental 500 m3 silo installation
Fig. 2. Results from 20-L-bomb explosibility tests of the two dusts. From Eckhoff et al. (1982).
The experiments discussed in this paper were conducted during the first of two consecutive silo dust explosion programs (1980e1982). A battery of 8 identical semi-condemned 500 m3 steel silo cells at Boge, 45 km east of Bergen, were made available for the two programs. The battery consisted of two parallel cell rows of 4 cells each. Each silo cell had a diameter of about 5.5 m and height of about 21 m, corresponding to a length-to-diameter ratio (L/D) of about 4. All the experiments discussed in the present paper were conducted in one of these cells, as indicated in Fig. 4. An overview of the total experimental set-up with the dust injection facilities is shown in Fig. 5. A cross section of the experimental silo cell with ignition point and diagnostic instrumentation is shown in Fig. 6. An external winding staircase extending to the top of the experimental silo cell not only provided access to the silo top for arranging the required vent opening, but it also permitted easy mounting and inspection of diagnostic instruments at various desired levels above the silo bottom. As illustrated in Fig. 6 four different types of measurements
Please cite this article in press as: Eckhoff, R.K., Scaling of dust explosion violence from laboratory scale to full industrial scale e A challenging case history from the past, Journal of Loss Prevention in the Process Industries (2015), http://dx.doi.org/10.1016/j.jlp.2014.12.020
4
R.K. Eckhoff / Journal of Loss Prevention in the Process Industries xxx (2015) 1e10
Table 1 Summary of explosion properties of the two dusts used in the 500 m3 silo experiments. Dust type
Median particle size by mass (mm)
Moisture content (% by mass)
(dP/dt)max 1.2 L Hartm. bomb (bar/s)
(dP/dt)max 20 L sphere (bar/s)
Pmax Hartmann bomb (bar(g))
Pmax 20 L sphere (bar(g))
Wheat grain Soya meal
20 50
12e15 9
125 27
350 175
5.8 4.3
7.3 6.6
were performed during an explosion in the silo: a) Eight specially constructed dust cloud extraction probes, distributed throughout the silo volume as indicated in Fig. 6, were used in an attempt at measuring the dust concentration distribution in the silo just prior to ignition. These probes were used only in the wheat grain dust experiments. Unfortunately, for aerodynamic reasons, they were only able to collect the finest particles in the dust cloud.
b) Three piezoelectric Kistler No. 412 pressure transducers were used for measuring the development with time of the explosion pressure inside the silo cell. In some tests only the top and bottom transducers were in use, as indicated in Fig. 6. c) Three narrow-angle optical photo-diode probes were used for detecting the flame front arrival times inside the silo cell, as also indicated in Fig. 6.
Fig. 4. Overview of the experimental test site at Boge, Vaksdal, outside Bergen, Norway. From Eckhoff et al. (1982).
Please cite this article in press as: Eckhoff, R.K., Scaling of dust explosion violence from laboratory scale to full industrial scale e A challenging case history from the past, Journal of Loss Prevention in the Process Industries (2015), http://dx.doi.org/10.1016/j.jlp.2014.12.020
R.K. Eckhoff / Journal of Loss Prevention in the Process Industries xxx (2015) 1e10
5
Fig. 5. Overview of the experimental set-up with dust injection facilities. From Eckhoff et al. (1982).
d) Sixteen mm, 24 frame/s movie recordings of the explosions were taken from a convenient position on the hillside about 100 m from the silo top, as indicated in Fig. 4. In order to enable conduction of experiments with different vent areas, a strong steel grid was constructed across the entire top surface of the experimental silo cell, permitting any desired part of it to be blocked by bolting the required number of square 0.5 m 0.5 m steel plates to the grid. The range of vent openings used is illustrated in Fig. 7. 6. Generation of experimental dust clouds in the 500 m3 silo 6.1. The method For practical/financial reasons it was decided to generate the dust clouds in the silo by blowing into it a known quantity of dust by means of a standard 120 Hp “VacuVator” Model PTA 114144 (Dunbar Kapple Inc., Illinois, USA), which the project was allowed to borrow from one of the sponsors without any costs. As indicated in Fig. 6, the dust jet entered the silo vertically upwards approximately at the silo centre. It was estimated that typical air velocities during dust injection into the silo were of the order of 10e15 m/s. To prevent the dust from escaping through the vent opening at the silo top during the injection period, the vent opening was always sealed with a thin sheet of plastic. In this way the dust cloud formation process was also kept independent of the vent area. In all the experiments the dust clouds were practically quiescent at the moment of ignition. This was achieved as follows: During the dust injection period the blower was operated at a steady, high speed until a fairly distinct change of its sound indicated that most of the dust had been blown into the silo. At this point the blower power was reduced to “no load”, and the automatic system for the sequential dust concentration sampling, start of Ampex tape
recorder, and ignition source firing was triggered manually by the blower operator. The duration of this sequence was about 5 s. 6.2. Structures of the experimental dust clouds in the silo Although the same dust injection method was used for both dusts, the resulting dust cloud structures in the silo were most probably quite dissimilar, because of differences in the dispersibility and particle size distributions. In the case of the wheat grain dust, which is difficult to disperse because of the high content of fibrous material, considerable quantities of unburnt dust (up to 100e150 kg) were normally found deposited on the silo bottom after the explosion. This clearly means that a significant fraction of the dust being blown into the silo cell was in the form of large agglomerates, able to settle out of suspension before the dust cloud was ignited. However, this settling process apparently had a significant homogenizing effect on the concentration distribution of the remaining cloud of finely dispersed dust, inasmuch as the automatic dust concentration measurement system revealed a fairly even distribution of the fine dust fractions throughout the entire silo volume, both horizontally and vertically. Contrary to the wheat grain dust, the soya meal was very easy to disperse. However, because of the total collapse of the experimental silo in the very first experiment with this material, no data for the dust concentration distribution at the moment of ignition was obtained. 7. Outline of the dust explosion experiments in the 500 m3 silo Because the quantity of wheat grain dust being available for the experiments was much larger than the available quantity of soya meal, it was decided to start the experiments with the wheat grain dust. Hence, all the first 16 silo explosion experiments were with
Please cite this article in press as: Eckhoff, R.K., Scaling of dust explosion violence from laboratory scale to full industrial scale e A challenging case history from the past, Journal of Loss Prevention in the Process Industries (2015), http://dx.doi.org/10.1016/j.jlp.2014.12.020
6
R.K. Eckhoff / Journal of Loss Prevention in the Process Industries xxx (2015) 1e10
injection. The ignition point was close to the silo bottom, as indicated in Fig. 6. For safety reasons the first of the 16 experiments with the wheat grain dust were conducted with the largest attainable vent opening of 14.2 m2 (see Fig. 7). In the successive experiments the vent opening was reduced in steps, as also illustrated in Fig. 7. After each explosion the silo was inspected for any sign of damage or deformation. The changing of the vent area between experiments was a quite laborious process and it was decided to conduct the experiments with the soya meal by starting with the smallest opening (2.1 m2), with which the experiments with the wheat dust ended, and then increase the vent area stepwise in successive experiments. This decision was based on the results from the explosion violence tests in the 20-L-sphere and the Hartmann bomb given in Figs. 2 and 3 and Table 1. Relying on the commonly accepted stateof-the-art in 1980 it was then just assumed that also the explosion violence to be expected from the soya meal in the big silo would be less severe than that found in the silo with the wheat grain dust.
8. Results from the 500 m3 silo experiments 8.1. Explosion pressure development Two typical traces of explosion pressure versus time during wheat grain dust explosions in the 500 m3 silo are given in Figs. 8 and 9. The frequency of the oscillating traces in Fig. 8 of 3e4 Hz agrees well with the first harmonic of the standing sound wave in the gas column in the silo (silo top open, i.e. 1/4 l ¼ 21 m, and hence l ¼ 84 m, where l is the wave length). Apart from the oscillatory pressure wave pattern obtained with the largest vent area of 14.2 m2, the pressure time histories were generally one single distinct bell-shaped pulse of about 1 s duration, as shown in Fig. 9. Fig. 10 gives the pressure trace from the single destructive soya meal explosion. The shape of this trace may indicate that the maximum explosion pressure that would have resulted, had the silo been sufficiently strong, would perhaps not have been much higher than the 0.6 bar(g) at which the silo ruptured.
8.2. Vertical flame front propagation velocities in the 500 m3 silo
Fig. 6. Cross section of experimental silo cell with ignition source location, point of dust injection and diagnostics. From Eckhoff et al. (1982).
this dust. In all these experiments, and also in the single experiment with the soya meal, about 330 kg of the dust was blown into the silo in the form of a free jet extending upwards in the upper half of the silo volume, as already explained and indicated in Figs. 5 and 6. Downwards migration of the dust cloud below the exit of the dust injection pipe occurred by rapid settling of coarse particles and particle agglomerates. Based on the amount of dust injected, the nominal average dust concentration was about 660 g/m3 both with wheat grain dust and with soya meal. However, before ignition of the dust clouds was activated a significant part of the dust had settled out at the silo bottom, and hence the real average dust concentration in the exploding cloud was somewhat lower than this value. Ignition was effectuated 3e5 s after termination of dust
The upwards vertical flame front speeds in the upper half of the silo were estimated on the basis of the flame-arrival-time data provided by the three flame detectors located as indicated in Fig. 6. A typical trend for the wheat grain dust explosions would be a few m/s halfway down in the silo, and subsequent acceleration to about 40e50 m/s just below the vent opening. It was observed that the duration of the flame signal at a given photo diode station was considerably longer than the time required for the flame to travel from the bottom of the silo to its top. This suggests that the combustion process in the silo was to a large extent volumetric, i.e. the flame was very thick. This opposes idealized models of a thin flame sweeping through the cloud, which was sometimes assumed in the past. Very long burn-out times mean that the maximum rate of heat production in the silo will occur at the moment when the flame reaches the vent, because at this moment the quantity of dust that is burning simultaneously inside the silo is at the maximum. On this basis one would expect a systematic coincidence between the moment of flame arrival at the silo top and the occurrence of the peak pressure in the silo. This was in fact also observed experimentally.
Please cite this article in press as: Eckhoff, R.K., Scaling of dust explosion violence from laboratory scale to full industrial scale e A challenging case history from the past, Journal of Loss Prevention in the Process Industries (2015), http://dx.doi.org/10.1016/j.jlp.2014.12.020
R.K. Eckhoff / Journal of Loss Prevention in the Process Industries xxx (2015) 1e10
7
Fig. 7. Range of vent openings used in the 500 m3 silo experiments. From Eckhoff et al. (1982).
8.3. Dust concentration distributions in the 500 m3 silo immediately prior to ignition An attempt was made at estimating the dust concentration distribution throughout the 500 m3 silo immediately before ignition, by simultaneous measurements at 8 different points. Fig. 6 shows the locations of the 8 dust sampling probes. The technique used was fast extraction of a sample of about 4 L of dust cloud through a small filter. The initial assumption was that all the dust in the sampled cloud volume would be sucked into the probe and hence would be available for subsequent weighing. This would work if the particles were sufficiently small to follow the air stream
into the sampler. However, the wheat grain dust contained a substantial fraction of agglomerates behaving as large single particle units, and hence this condition was not satisfied. Therefore, only a fraction of the dust that was initially in the air volume flowing through the dust sampler, would actually enter the filter and be collected there. This was confirmed in controlled laboratory tests where the ‘yield’ never exceeded 50%. Because of poorer dust dispersion in the silo than in the laboratory tests, the ‘yields’ in the silo tests were probably even less than 50%. Much of the dust that should ideally have entered the filter probe, instead accumulated on the outside of the probe, around its entrance, and thus could not in any reliable way be considered part of the sampled dust. Also, the
Please cite this article in press as: Eckhoff, R.K., Scaling of dust explosion violence from laboratory scale to full industrial scale e A challenging case history from the past, Journal of Loss Prevention in the Process Industries (2015), http://dx.doi.org/10.1016/j.jlp.2014.12.020
8
R.K. Eckhoff / Journal of Loss Prevention in the Process Industries xxx (2015) 1e10
and hence some of the sampled dust occasionally got burnt. In spite of these shortcomings, the dust concentration sampling system did provide an indication of the evenness of the concentration of the fine dust fraction throughout the silo volume, just prior to ignition. Two examples of measured distributions are given in Fig. 11. Systematic radial concentration gradients were not typical, although such gradients did occur in some tests, as shown close to silo bottom in the upper part of Fig. 11. In some experiments even the vertical concentration distribution of the fine dust fraction appeared to be quite homogeneous (lower part of Fig. 11). However, some vertical concentration gradients were observed in most cases, with the maximum concentration occurring close to the silo top
Fig. 8. Typical explosion pressure development in vented 500 m3 silo with wheat grain dust and 14.2 m2vent opening. Pressure-versus-time records from pressure probe 1 (lower trace) and pressure probe 2 (upper trace). For positions of pressure probes see Fig. 6. From Eckhoff et al. (1982).
Fig. 9. Typical explosion pressure development in vented 500 m3 silo with wheat grain dust and 8.6 m2vent opening. Pressure-versus-time records from pressure probe 1 (lower trace) and pressure probe 2 (upper trace). For positions of pressure probes see Fig. 6. From Eckhoff et al. (1982).
Fig. 10. Pressure development in the destructive vented explosion in the 500 m3 silo with soya meal and 2.1 m2 vent opening. Pressure-versus-time record from pressure probe 2. For position of pressure probe see Fig. 6. From Eckhoff et al. (1982).
dust sample probes did not shield the sampled dust inside sufficiently from the heat of the explosion and subsequent fire outside,
Fig. 11. Two examples of distributions the concentration of the finest wheat grain dust fraction in the 500 m3 silo as measured by the dust sampling probes just prior to ignition. From Eckhoff et al. (1982).
Please cite this article in press as: Eckhoff, R.K., Scaling of dust explosion violence from laboratory scale to full industrial scale e A challenging case history from the past, Journal of Loss Prevention in the Process Industries (2015), http://dx.doi.org/10.1016/j.jlp.2014.12.020
R.K. Eckhoff / Journal of Loss Prevention in the Process Industries xxx (2015) 1e10
and/or close to the silo bottom (upper part of Fig. 11). 9. Discussion More than 70 years ago the great American pioneers of dust explosion research, Hartmann et al. (1943) wrote: “Over the past 30 years, various investigators have worked on means of producing uniform dust clouds; comparison of results indicates that none of them has been wholly successful. The mechanism to produce such a cloud, of sufficient volume to be usable in test work, remains to be perfected”. This statement is still valid today, with a slight adjustment: “Over the past 100 years …”. However, during these 100 years another aspect has become even more challenging: If it were at all possible to produce a “perfect” experimental dust cloud, would this help us predicting the course of real industrial dust explosions? Eckhoff (1984) challenged the relevance of using existing standard test methods for predicting the potential explosion violence of real dust clouds in the process industries. Dust clouds encountered in industry are not “perfect” but possess varying structures of dust concentration, degree of dust dispersion and degree of turbulence, which are all results of the prevailing process conditions. This fact has been the main concern of the present paper. As shown above, the big silo installation ruptured and collapsed in the very first experiment with the soya meal. The central question then becomes: Why did the soya meal explode more violently in the silo than the wheat grain dust, when both the Hartmann bomb tests and the 20-L tests beyond any doubt had predicted the opposite trend? The particle size analyses of the two powders, determined by three independent laboratories (see Fig. 1), showed that the wheat grain dust was considerably finer than the soya meal, which gave further a priori support to the validity of ranking the expected explosibilities of the two dusts in a real process situation on the basis of the results in Figs. 2 and 3 and Table 1. The point is that the wheat grain dust and the soya meal undoubtedly responded quite differently to the rather gentle dust cloud formation/dispersion processes in the silo on the one hand, and the much more intense dispersion processes in both the 20-L test bomb and the Alpine air jet sieve, on the other. Because the soya meal mostly consists of easy-to-disperse equi-dimensional particles, it was most probably well dispersed even in the silo experiments. The wheat grain dust, on the other hand, containing an appreciable fraction of both fibres and flakes, most probably was much more difficult to disperse in the silo than the soya meal. In the 20-L apparatus even the wheat grain dust most probably got completely dispersed. It is even quite likely that some of the fibres and flakes were crushed into much smaller particles in this violent dispersion process. The results from the Hartmann bomb tests indicate that similar effects occurred even in this apparatus. The ability of the intense dust dispersion process of the 20-L apparatus to crush the primary particles of some dusts was conclusively demonstrated by Kalejaiye et al. (2006). Some particle crushing would also be expected to occur in the Alpine air jet sieve, in particular with dusts containing fragile fibres and flakes, as wheat grain dust. However, because this was not the case in the silo, a significant proportion of the wheat grain dust particles appeared there as large un-dispersed agglomerates. Therefore the combustion rate of the wheat grain dust clouds in the silo was lower than that of the clouds of the soya meal, contrary to what was predicted by the Hartmann bomb and the 20-L bomb tests. The dispersibility of a dust is a decisive inherent dust property in relation to formation of dust cloud structures and resulting dust cloud burning rates. Therefore, adequate experimental methods for assessing dust dispersibility are required. In Chapter 7 of Eckhoff
9
(2003) some published methods are described. The problem of dust dispersibility in relation to dust cloud formation at large is discussed in Chapter 3 of Eckhoff (2003). It is clear, however, that more research is needed in this area before comprehensive models of dust dispersion processes can be developed. The possibility of developing standard test methods for assessing the dispersibility (“dustiness”) of dusts is currently discussed internationally. There is a continued need for discussing what dust cloud formation and ignition processes and associated combustion rates should, under various circumstances in industry, be regarded as the “worst cases” that the explosion protection, including venting, should be able to accommodate. Fig. 12 shows the marked influence of the mode of dust cloud generation on the maximum pressure Pred in a dust explosion in a vented industrial enclosure. In the VDI (1979) method the dust was injected into the silo at high rate from a number of pre-pressurized bottles mounted along the silo wall from its bottom to its top. Fig. 12 illustrates the need for adopting a differentiated approach to sizing of dust explosion vents. The basic understanding of flame propagation processes inside and outside vented enclosures is still unsatisfactory. Neither the processes by which dust clouds in vented enclosures are generated, nor the way in which the clouds burn, are sufficiently well understood. Consequently, adequate venting theories do not exist, and one must rely on experimental evidence. Recent European standardization does in principle to some extent open up for a differentiated approach to vent sizing, accounting for the variations in dust cloud structures encountered in practice in industry. In most practical cases this will result in more liberal vent area requirements than those put forward in some very rigorous standards. However, it is believed that the ultimate long term solution for design of explosion venting arrangements will be comprehensive computer models. Such models will not only be useful for design of explosion venting systems, but also of systems for dust explosion suppression and isolation. Skjold (2007, 2010, 2014a, 2014b) and Skjold et al. (2014) have summarized the state-of-the-art in the promising effort to develop the CFD-based code DESC aimed at meeting this need.
Fig. 12. Maximum pressures Pred in vented maize starch explosions in a 20 m3 silo of L/ D ¼ 6. From Eckhoff (2003), based on experiments by W. Bartknecht and S. Radandt et al.
Please cite this article in press as: Eckhoff, R.K., Scaling of dust explosion violence from laboratory scale to full industrial scale e A challenging case history from the past, Journal of Loss Prevention in the Process Industries (2015), http://dx.doi.org/10.1016/j.jlp.2014.12.020
10
R.K. Eckhoff / Journal of Loss Prevention in the Process Industries xxx (2015) 1e10
10. Conclusions 1. The standardized KSt parameter still seems to be widely used as a universal criterion for ranking the explosion violence to be expected from various dusts in given industrial situations. However, this may not be a generally valid approach. In the case of dust explosion venting the maximum pressure Pmax generated in a given vented industrial enclosure will not only be a result of inherent dust parameters (dust chemistry including moisture, and sizes and shapes of individual dust particles). Process-related parameters (degree of dust dispersion, cloud turbulence, and dust concentration) also play key roles. 2. In the research conducted during 1980e82, which is reconsidered in the present paper, two different dusts were used, viz. a wheat grain dust and a soya meal dust. Both dusts were tested in the standard 20-L bomb in two independent laboratories (in UK and Switzerland), and also in the Hartmann bomb (in UK and Norway). In all laboratory tests both Pmax and (dP/dt)max were significantly lower for the soya meal than for the wheat grain dust. 3. However, in the 500 m3 vented silo cell the soya meal in fact produced more violent explosions than the wheat grain dust. 4. This gives further support to the view that a differentiated approach to design of dust explosion vents is indeed needed. To some, but insufficient, extent attempts have been made at meeting this need in more recent dust explosion venting codes. As regards very large silos in particular, one still has to rely on rather questionable extrapolation. 5. In future the most likely solution to this complex and challenging reality will be tailored design by means of CFD-based computer codes. This not only applies to design of explosion venting systems, but also to systems for explosion suppression and isolation. In the case of venting, this will be particularly relevant and useful for very large silos of volumes of several thousand m3, for which no experimental data exist. Acknowledgement The author wishes to gratefully acknowledge the sponsoring organizations from both UK, Sweden and Norway, and their enthusiastic representatives, for having provided the support without which the large-scale silo experiments described in this paper would never have become a reality. The author is particularly indebted to his close co-worker during 17 years, Mr. K. Fuhre, who
carried a major responsibility both in the construction of the largescale experimental facility and in the planning and running of the experiments. Sincere thanks are also expressed to the local population of Boge outside Bergen, Norway, for their continued understanding and forbearance in connection with the rather unusual activity that took place in the early 1980es in their otherwise very quiet and pleasant local community. References Eckhoff, R.K., 1982. Current dust explosion research at the CMI. In: Lee, J.H.S., Guirao, C.M. (Eds.), Proceedings of the International Conference on FueleAir Explosions Held at McGill University, Montreal, Canada, 4e6 November 1981. University of Waterloo Press, pp. 657e678. Eckhoff, R.K., Fuhre, K., Henery, M.J., Parker, S.J., Thorsen, H.G., 1982. Dust Explosion Experiments in a Vented 500 m3 Silo Cell. CMI-Report (Chr. Michelsen Institute, Bergen, Norway, presently: GexCon AS, Bergen, Norway) No. 813307e1. Eckhoff, R.K., 1984. Relevance of using (dP/dt)max data from laboratory-scale tests for predicting explosion rates in practical industrial situations. VDI-Berichte 494. In: Proceedings of the International VDI Colloquium “Safe Handling of Flammable Dusts” in Nürnberg, Germany October 26e28, 1983. Verein Deutscher Ingenieure, Düsseldorf, Germany, pp. 207e217. Eckhoff, R.K., Fuhre, K., 1984. Dust explosion experiments in a vented 500 m3 silo cell. J. Occup. Accid. 6, 229e240. Eckhoff, R.K., 1987. A differentiated approach to sizing of dust explosion vents: influence of ignition source location with particular reference to large, slender silos. In: Proceedings of Symposium on Industrial Dust Explosions. ASTM, Philadelphia, PA, USA, pp. 265e280. ASTM Publication Code Number 04958000-31. Eckhoff, R.K., 2003. Dust Explosions in the Process Industries, third ed. Gulf Professional Publishing, an imprint of Elsevier Science, Amsterdam. Hartmann, I., Nagy, J., Brown, H.R., 1943. Inflammability and Explosibility of Metal Powders. United States Department of the Interior, Bureau of Mines. Report of Investigations No. 3722. Kalejaiye, O., Amyotte, P.R., Pegg, M.J., 2006. Effectiveness of dust dispersion in the 20-l Siwek chamber. In: Proceedings of 6th International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions, Aug. 27eSept. 1, vol. 1. Dalhousie University, Halifax, NS, Canada, pp. 253e278. Skjold, T., 2007. Review of the DESC project. J. Loss Prev. Process Indust. 20, 291e302. Skjold, T., 2010. Flame propagation in dust clouds: challenges for model validation. In: Proceedings Eighth International Symposium on Hazards, Prevention and Mitigation of Industrial Explosions (ISHPMIE), Yokohama, p. 11. Skjold, T., 2014a. Simulating vented maize starch explosions in a 236 m3 silo. In: Eleventh International Symposium on Fire Safety Science, University of Canterbury, New Zealand, 10e14 February 2014. International Association for Fire Safety Science (IAFSS). Skjold, T., 2014b. Flame Propagation in Dust Clouds. Numerical and Experimental Investigation (Ph.D. thesis). University of Bergen, Norway. Skjold, T., Castellanos, D., Lien, K.O., Eckhoff, R.K., 2014. Experimental and numerical investigations of constant volume dust and gas explosions in a 3.6 metre flame acceleration tube. J. Loss Prev. Process Indust. 30, 164e176. VDI, 1979. Druckentlastung von Staubexplosionen. VDI-Richtlinie No. 3673. Verein Deutscher Ingenieure, Düsseldorf, Germany.
Please cite this article in press as: Eckhoff, R.K., Scaling of dust explosion violence from laboratory scale to full industrial scale e A challenging case history from the past, Journal of Loss Prevention in the Process Industries (2015), http://dx.doi.org/10.1016/j.jlp.2014.12.020