666 J. Seville et al the pressure difference generated across the candle wall may be high enough for cake detachment near the closed end, but not high enough near the open end (Fig. 6). In addition, if the pulse tube is wrongly placed, secondary entrainment may occur through the candle wall near the open end, in the filtration direction, thus ensuring no cake detachment in that region. Properly designed venturis can reduce the requirement of high pressure pulse gas for a particular cleaning duty; n effect, they act as pumps for the pulse gas [19]. However, they can make little difference to the distribution of the cleaning pressure [20]. What is required is a way of modifying this distribution by better candle design. Chuah et al. [21] propose one such modification; Grannell [22] has investigated others. Maldistribution will obviously be more of a problem for candles with more extreme length-to-diameter ratios; Clift et al. [18] give some guidance on this a Pulse cleaning is more complex than filtration from the point of view of fluid chanics, partly because the fow velocities are often larger, but also because th components which give rise to the internal axial pressure drop are in opposition. This internal pressure drop occurs partially because of the momentum changes in the gas and partially because of wall frictional effects, as in conventional pipe flow In filtration, the momentum change In pulse cleaning, however, the two effects is the more significant factor [18, but both contributions act in the same direction act in the opposite sense, so that a minimum can occur in the internal pressure if high reverse flows are reached [15]. It is reassuring to note that all of these effects are predictable using both simple one-dimensional models and more sophisticated computational fluid dynamics(CFD)codes [21] Clearly, if resistance to flow of the medium is not uniform along the candle length, this can also lead to flow maldistribution. Little has been published in this area on granular candles, but low-density candles can show considerable changes in resistance along their length and in opposite senses, according to whether they ar made with the former on the outside or the inside. If the variation is in the right ense,it can help to counteract the maldistribution problems discussed above [1 3. 2. Cake detachmen Particle mechanics as a subject is at a much earlier stage of development than fluid mechanics(see, e.g. [23). Although there are in principle some possible approaches to predicting the failure stress for a compact of identical spherical particles, even this problem has not been resolved satisfactorily [24] and in practice it is necessary to fall back on experiment. The basic validity of the coupon test has been shown by Koch et al. [17] who compared the stresses necessary to remove filter cakes by reverse flow and acceleration in a centrifuge, with comparable results. Figure 7 shows how the results of such a determination might be used in practice On the left-hand side is a set of cake detachment curves; on the right, an imaginary axial distribution of cleaning pressure. If the measured cake detachment stress curve is'a', then most of the cake will be removed by the pulse; if it is 'c', then very little will. This simplistic approach begs lots of questions, however. What level of666 J. Seville et al. the pressure difference generated across the candle wall may be high enough for cake detachment near the closed end, but not high enough near the open end (Fig. 6). In addition, if the pulse tube is wrongly placed, secondary entrainment may occur through the candle wall near the open end, in the ltration direction, thus ensuring no cake detachment in that region. Properly designed venturis can reduce the requirement of high pressure pulse gas for a particular cleaning duty; in effect, they act as pumps for the pulse gas [19]. However, they can make little difference to the distribution of the cleaning pressure [20]. What is required is a way of modifying this distribution by better candle design. Chuah et al. [21] propose one such modication; Grannell [22] has investigated others. Maldistribution will obviously be more of a problem for candles with more extreme length-to-diameter ratios; Clift et al. [18] give some guidance on this. Pulse cleaning is more complex than ltration from the point of view of uid mechanics, partly because the ow velocities are often larger, but also because the components which give rise to the internal axial pressure drop are in opposition. This internal pressure drop occurs partially because of the momentum changes in the gas and partially because of wall frictional effects, as in conventional pipe ow. In ltration, the momentum change is the more signicant factor [18], but both contributions act in the same direction. In pulse cleaning, however, the two effects act in the opposite sense, so that a minimum can occur in the internal pressure if high reverse ows are reached [15]. It is reassuring to note that all of these effects are predictable using both simple one-dimensional models and more sophisticated computational uid dynamics (CFD) codes [21]. Clearly, if resistance to ow of the medium is not uniform along the candle length, this can also lead to ow maldistribution. Little has been published in this area on granular candles, but low-density candles can show considerable changes in resistance along their length and in opposite senses, according to whether they are made with the former on the outside or the inside. If the variation is in the right sense, it can help to counteract the maldistribution problems discussed above [15]. 3.2. Cake detachment Particle mechanics as a subject is at a much earlier stage of development than uid mechanics (see, e.g. [23]). Although there are in principle some possible approaches to predicting the failure stress for a compact of identical spherical particles, even this problem has not been resolved satisfactorily [24] and in practice it is necessary to fall back on experiment. The basic validity of the coupon test has been shown by Koch et al. [17] who compared the stresses necessary to remove lter cakes by reverse ow and acceleration in a centrifuge, with comparable results. Figure 7 shows how the results of such a determination might be used in practice. On the left-hand side is a set of cake detachment curves; on the right, an imaginary axial distribution of cleaning pressure. If the measured cake detachment stress curve is ‘a’, then most of the cake will be removed by the pulse; if it is ‘c’, then very little will. This simplistic approach begs lots of questions, however. What level of