Experimental investigation of nanoparticles precipitation in a rotating packed bed

Experimental investigation of nanoparticles precipitation in a rotating packed bed

Particuology 8 (2010) 372–378 Contents lists available at ScienceDirect Particuology journal homepage: www.elsevier.com/locate/partic Experimental ...

516KB Sizes 0 Downloads 39 Views

Particuology 8 (2010) 372–378

Contents lists available at ScienceDirect

Particuology journal homepage: www.elsevier.com/locate/partic

Experimental investigation of nanoparticles precipitation in a rotating packed bed Yang Xiang, Guangwen Chu, Lixiong Wen, Kuang Yang, Guangting Xiao, Jianfeng Chen ∗ Key Laboratory for Nanomaterials, Ministry of Education, and Research Center of the Ministry of Education for High Gravity Engineering and Technology, Beijing University of Chemical Technology, Beijing 100029, China

a r t i c l e

i n f o

Article history: Received 25 August 2009 Accepted 19 May 2010 Keywords: Rotating packing bed BaSO4 Nanoparticle Precipitation Mixing

a b s t r a c t Precipitation of BaSO4 nanoparticles was studied for the first time in a specially designed rotating packed bed (RPB), which allowed sampling at different radial positions to provide better insight of the mechanism of precipitation in RPB. Particle size and morphology were characterized by TEM, while the quality of synthesized BaSO4 powders was analyzed by XRD and BET, and compared with those prepared in a stirred-tank reactor. The important role of the inlet region of the RPB in the whole precipitation process was experimentally confirmed, as a significant essence for the design of industrial RPB for the precipitation of sparingly soluble materials. The effects of different operating conditions on particle size were also investigated, showing that particle size decreases with increasing rotational speed and liquid flow rate, due to the enhancement of micromixing in the RPB. © 2010 Chinese Society of Particuology and Institute of Process Engineering, Chinese Academy of Sciences. Published by Elsevier B.V. All rights reserved.

1. Introduction Precipitation of sparingly soluble materials has been widely used in various industries, e.g. magnetic materials, pigments, catalysts and drugs. The precipitation process is usually very fast, and mixing, especially micromixing (mixing at the molecular scale), can strongly influence particle size distribution (PSD) and product morphology. Since Danckwerts (1958) claimed the importance of mixing on homogeneous reactions, such as polymerization and precipitation, reactors including continuous and semi-batch stirred tanks (Aslund & Rasmuson, 1992; Baldyga, Podgorska, & Pohorecki, 1995; Chen, Zheng, & Chen, 1996; David & Marcant, 1994; Garside & Tavare, 1985; Tavare, 1995; Torbacke & Rasmuson, 2001; Zauner & Jones, 2002), pipe reactors (Falk & Schaer, 2001; Marchisio, Barresi, & Garbero, 2002), and Couette-type reactors (Barresi, Marchisio, & Baldi, 1999), have been used to investigate the influence of mixing on precipitation. Nanoparticles can be produced by applying very high primary nucleation rates, which require very strong supersaturation, such as can be achieved by high reactant concentration but low solubility, as well as rapid micromixing at the reactant feed point for fast chemical reaction (Mersmann, 1999). Obviously, conventional reactor types are less suitable. Recent work reported on new reactors for nanoparticles preparation capa-

∗ Corresponding author. Tel.: +86 10 64446466; fax: +86 10 64434784. E-mail address: [email protected] (J. Chen).

ble of greatly intensified mixing, such as microchannel reactors (Kockmann, Kastner, & Woias, 2008; Schwarzer & Peukert, 2004; Sue, Kimura, & Aral, 2004; Takagi, Maki, Miyahara, & Mae, 2004; Wang, Nakamura, Uehara, Miyazaki, & Maeda, 2002), membrane reactors (Chen, Luo, Xu, & Wang, 2004; Jia & Liu, 2002) and rotating packed beds (Chen & Shao, 2003; Chen, Wang, Guo, Wang, & Zheng, 2000; Chen et al., 2004b). Rotating packed bed, which uses centrifugal force to intensify mass transfer and mixing-limited processes, is a novel highefficiency multiphase contactor. Since its emergence, RPB has been applied in absorption (Lin, Liu, & Tan, 2003), desorption (Fowler & Khan, 1987) and distillation (Kelleher & Fair, 1996), etc. The use of the rotating packed bed (RPB) as a precipitation or crystallization reactor in synthesizing organic or inorganic nanoparticles has recently received extensive attention. For instance, the first commercial scale production of CaCO3 up to an annual capacity of 10,000 tonnes was successfully realized in 2000 (Chen & Shao, 2003). The production of nanoparticles with uniform PSD by precipitation requires rapid mixing with the characteristic micromixing time much less than the characteristic time of nucleation and crystal growth. However, the effect of mixing on precipitation has not been well understood, and but few works about optimum operating conditions have been conducted in the RPB, thus retarding the design of RPB for industrial nanoparticles production. In this work, an RPB with special configuration was designed to facilitate radial sampling for better insight on the effect of micromixing on precipitation of BaSO4 , typically the effects of operational conditions (e.g. rotation speed, liquid flow rate, super-

1674-2001/$ – see front matter © 2010 Chinese Society of Particuology and Institute of Process Engineering, Chinese Academy of Sciences. Published by Elsevier B.V. All rights reserved.


X. Yang et al. / Particuology 8 (2010) 372–378


Nomenclature a CA0 , CB0 CP0 CV d G KSP L S Sf Si,max VA , VB

volumetric ratio feed concentration of A and B, respectively (mol/L) concentration of product P (mol/L) dimensionless variance mean particle size (nm) crystal growth rate (m/s) solubility product (mol2 /m6 ) crystal size (m) supersaturation (S = (CA CB /KSP )0.5 ) final supersaturation after complete mixing initial maximum supersaturation volume flow rate of A and B, respectively (L/h)

Subscripts 0 initial A species of BaCl2 B species of Na2 SO4 P species of BaSO4

saturation and volumetric ratio) on PSD. The BaSO4 nanoparticles synthesized by RPB are compared to particles prepared by stirredtank reactor. The nucleation induction time in RPB, estimated by a theoretical equation, is compared to the characteristic micromixing time reported in the literature. 2. Experimental Guo et al. (2000) and Yang, Chu, Zhang, Shen, and Chen (2005) indicated that the inlet region of packing in a RPB plays a crucial role on mass transfer and mixing. Thus, a special RPB with 11 sampling tubes ( 2 mm) at different radial positions was built to study the effects of radial thickness of packing on liquid–liquid precipitation, as shown in Fig. 1. The inner and outer diameters of packing were 60 mm and 446 mm, respectively, and the axial depth of packing

Fig. 1. Sketch of RPB with radial sampling tubes (1 shell; 2 electrical motor; 3 rotor; 4 packing; 5 sampling tubes; 6 liquid inlets; 7 liquid outlet).

Fig. 2. Flowchart of experimental set-up (1 RPB; 2 pumps; 3 flowmeters; 4 tanks).

was 50 mm. The first sampling tube was located at R = 10 mm (the radial length from the inner edge of packing), and other sampling tubes were installed every 17 mm along the radial direction (Yang et al., 2005). The liquid distributor consisted of two pipes (10 mm outside diameter and 1.5 mm wall thickness) with one having three  2 mm nozzles and the other four  3 mm nozzles. The two reactants were injected side by side from these nozzles, in a 2 mm gap between the packing and the nozzle tips. The packing consisted of wire mesh with a specific area of 400 m2 /m3 and a free voidage of 95%. The precipitation of BaSO4 in aqueous solution was chosen as the test reaction, using BaCl2 and Na2 SO4 as feeds, and experimenting at 25 ± 1 ◦ C. The initial aqueous feed solutions of BaCl2 (AR grade, Beijing Chemical Reagent Co.) and Na2 SO4 (AR grade, Beijing Chemical Plant) were prepared with deionized water and then filtered with 0.22 ␮m millipore filters to remove impurities. Fig. 2 shows the experimental set-up. The two reactants were delivered into RPB by centrifugal pumps from storage tanks, and mixed to produce nanoparticles in the packing. After the operation reached steady state under some liquid flow rate and rotation speed, samples were collected from the outlet and sampling tubes along different radial positions of RPB. To terminate the reaction, nucleation and further crystal growth after flowing out the reactor, samples were immediately diluted with a large amount of saturated solution of barium sulfate in breakers. We checked the effect of the amount of saturated solution of barium sulfate in breakers on particle size and shape, and confirmed that the saturated solution of barium sulfate could effectively inhibit particle aggregation without however influencing particle size and morphology. In all runs, the experimental conditions were as follows: rotation speed (N), 400–1200 rpm; volume flow rate of reagent A (VA ), 180–420 L/h, and that of reagent B (VB ), 18–84 L/h; volume ratio of the two reagents (a = VA /VB ), 5 or 10; initial concentrations of reagent A (CA0 ) were 0.055, 0.06 and 0.12 mol/L. Stoichiometry is achieved if CB0 = aCA0 , corresponding to the concentrations of reagent B (CB0 ) of 0.55, 0.3 and 0.6 mol/L. After samples were treated ultrasonically, the particle size and morphology of BaSO4 were observed by using a transmission electron microscope (Hitachi H-800, Japan), and the particle size and distribution were determined by Image-Pro Plus software (release 5.0, MediaCybernetics, USA) based on the TEM photomicrographs obtained. Mean particle size (d) and dimensionless variance (CV)


X. Yang et al. / Particuology 8 (2010) 372–378

are calculated as 1 di , m m




(1/m) CV =

m i=1

(di − d)





where di is the size of some particle and m is the number of particles required. X-ray diffraction analysis was performed to detect any change in the physical characteristics and crystallinity of the nanopaticle BaSO4 powder by using a model XRD-6000 diffractometer (Shimadzu Inc., Japan). The measuring unit consisted of a rotating anode in transmission technique with a specification that Cu K 1 radiation was generated at 30 mA and 40 kV. Sample powder was carefully grounded and placed in an aluminum sample holder. The scanning speed was 5◦ /min from 5◦ to 50◦ with a step size of 0.05◦ . The specific surface areas of BaSO4 power prepared by RPB and those by stirred-tank reactor were measured using N2 adsorption method. In this method, the calculation was implemented by Surface Area Analyzer ASAP 2010-M (Micromeritics Instrument Corporation, USA) based on the BET equation. Before analyzing, approximately 200 mg of powder was loaded into a sample cell and degassed for at least 4 h. 3. Results and discussion 3.1. PSD at different radial positions Fig. 3 shows the experimental results of mean particle size (d) and dimensionless variance (CV) along different radial positions of packing in the RPB, indicating that d and CV decrease sharply in the inlet region of packing toward almost constancy. Samples at different radial positions of packing reflect the state of mixing of the two reactants: for perfect mixing, the particle size of the precipitate is small and PSD, narrow. It can be seen that mixing is finished at the peripheral packing under our experimental conditions, though the inlet region of packing in the inner rotator has a remarkable influence on liquid–liquid precipitation. This result is highly significant for the design of industrial RPB for the precipitation of sparingly soluble materials, that is, the optimum radial thickness of packing (e.g. 40–50 mm in our experiments), to minimize the total charge including equipment and operating cost upon decreasing rotator size and RPB volume.

Fig. 3. (a) Mean particle size d and (b) dimensionless variance CV at different radial positions R.

3.2. Effect of supersaturation on PSD Fig. 4 presents the TEM photographs of BaSO4 nanoparticles obtained by RPB at different initial concentrations of CP0 , showing that mean particle size d at CP0 = 0.1 mol/L is smaller than that at CP0 = 0.05 mol/L, that is, decreasing with increasing reactant concentrations. Nielsen (1964) indicated that the dependence of primary nucleation rate on supersaturation is an exponential function, with an exponent of 5–18. With increasing supersaturation, nucleation is strongly accelerated to bring about more nuclei, thus inhibit-

Fig. 4. TEM images of BaSO4 particles (a = 5, VA = 420 L/h, N = 800 rpm, outlet sample): (a) CP0 = 0.05 mol/L; (b) CP0 = 0.1 mol/L.

X. Yang et al. / Particuology 8 (2010) 372–378


Fig. 5. TEM images of BaSO4 particles (CP0 = 0.05 mol/L, VA = 420 L/h, N = 800 rpm, outlet sample): (a) a = 5; (b) a = 10.

ing crystal growth, and so decreasing the particle size of BaSO4 . 3.3. Effect of volumetric ratio on PSD Fig. 5 presents TEM photographs of BaSO4 nanoparticles prepared by RPB, to show increase of mean particle size of BaSO4 with increasing volumetric ratio a. The highest initial supersaturation is, when an eddy of reactant A with concentration CA0 first contacts an eddy of reactant B with concentration CB0 , Si,max =


3.5. Effect of rotation speed on PSD Fig. 7 shows the effect of rotation speed on mean particle size and dimensionless variance measured at the outlet of RPB. The mean particle size of precipitated BaSO4 decreases with increasing rotation speed, to reach a relatively steady value when rotation speed exceeds 1000 rpm. The dimensionless variance decreases slightly with increasing rotation speed. Similar to liquid flow rate, higher rotation speed causes larger centrifugal force, to result in higher relative velocity between liquid from nozzles and the rotating packing to enhance micromixing efficiency of the inlet region


After mixing but without consumption of any reagents the supersaturation is Sf =

CA0 CB0 , (1 + a)(1 + 1/a)KSP


The supersaturation ratio Si,max /Sf between the first contacting reagents and the final mixing condition is an indicator of the broadness of the range of supersaturation encountered. David (2001) indicated that unbalanced flow rates increase sensitivity to mixing effects, i.e. the mixing more strongly influences the precipitation process when the supersaturation ratio increases with increasing volumetric ratio. Consequently, uniformity of supersaturation distribution worsens, resulting in the increase of mean particle size and variance. 3.4. Effect of liquid flow rate on PSD Fig. 6 shows the effects of liquid flow rate on mean particle size and dimensionless variance measured at the outlet of RPB, demonstrating that d and CV of the precipitated BaSO4 decrease initially with increasing liquid flow rate, and then levels out when the flow rate (VA ) of reactant A exceeds 300 L/h. Higher liquid flow rate brings about greater initial radial velocity of droplets from nozzles, which causes higher relative velocity between liquid elements and the rotating packing, hence intensifying the micromixing in the inlet region of the packing. In addition, the residence time of liquid within RPB decreases with increasing liquid flow rate (Guo et al., 2000), thus increasing the coalescence–redispersion frequency between droplets, and also leading to intensified mixing. Such intensified mixing causes more homogeneous supersaturation in the radial direction, to result in increase of nuclei number and decrease of particle size. However, such intensifying effects will become negligible when the liquid flow rate is high, due to the change of rate-controlling mechanism from mixing rate to crystallization kinetics.

Fig. 6. Mean particle size and dimensionless variance at different liquid flow rates (sampling at the outlet): (a) d vs. VA ; (b) CV vs. VA .


X. Yang et al. / Particuology 8 (2010) 372–378

Fig. 8. XRD pattern for BaSO4 nanoparticles.

The precipitation process usually includes reaction, nucleation and crystal growth. Comparison of the characteristic time constants for precipitation yields the rate-controlling mechanism. For nth-order chemical kinetics the characteristic time constant for chemical reaction is tR = (kr C A0

m−1 −1




This type of reaction (e.g. reaction between BaCl2 and Na2 SO4 ) occurs instantaneously. The reaction rate constant kr is thus close to infinity, and the characteristic time tR approaches zero. The induction time of primary nucleation, which has frequently been used to characterize the rate of nucleation process, was estimated theoretically by Dirken and Ring (1991) Fig. 7. Mean particle size and dimensionless variance at different rotation speeds (sampling at the outlet): (a) d vs. N; (b) CV vs. N.

tind =

2 n 6dm , Di lnS


of the packing. Meanwhile, a higher rotation speed also decreases the residence time of the liquid (Guo et al., 2000). The combination of these effects induces more homogeneous supersaturation to eventually produce precipitates with smaller particle size at high rotation speeds. But this effect becomes negligible at higher rotation speeds.

where dm is the molecular diameter, n is the number of ions required to form a critical nucleus and Di is the diffusion coefficient of the solute. In practice the induction period is often considered to be inversely proportional to the rate of nucleation, that is, the characteristic time necessary to form a certain number of nuclei per unit volume,

3.6. Powder X-ray diffraction study

tind =

X-ray powder diffraction was performed to determine the physical state of the BaSO4 nanoparticles prepared by RPB, and the corresponding pattern was displayed in Fig. 8. The crystalline peaks were found in the diffraction pattern of the as-prepared BaSO4 as well as those of JCDPS standard atlas. In addition, X-ray diffraction peaks were quite sharp, clearly suggesting high crystallinity of the BaSO4 nanoparticles obtained by reactive precipitation in RPB.

¯ (the average where rN is the nucleation rate for fresh feed and N concentration of crystals in the bulk) can be estimated as

3.7. Comparison with conventional reactor Fig. 9 compares the TEM photographs and PSD of BaSO4 particles obtained by RPB against those obtained in a stirred-tank reactor, showing that the RPB BaSO4 nonaparticles are near-spherical, with narrow PSD with a mean size of about 47 nm. On the other hand, the stirred-tank-reactor particles possess assorted shapes with wide PSD and a mean size of 6–8 times larger. In addition, the BET of the RPB BaSO4 nanoparticles was significantly higher, 25.6235 m2 /g, as compared to 6.1234 m2 /g for the stirred-tank-reactor.

¯ = N

¯ N , rN

C¯ A0 MP , P L3



and the average size of crystals, L, must be found experimentally. For the case of BaSO4 precipitation, the order of magnitude of tind in RPB is estimated to be 10−4 s using the crystallization kinetics by Nielsen (1964), and coupled with our experimental conditions and results. Yang et al. (2005) reported that the micromixing time of a RPB is in the order of 10−5 to 10−4 s, indicating that micromixing in RPB is shorter than the nucleation induction time. Nevertheless, the characteristic time of micromixing in stirred-tank reactor is 1–200 ms (Guichardon & Falk, 2000), which is far larger than that in RPB. Therefore, it is concluded that the RPB, with the aid of centrifugal force, could have improved micromixing efficiency, thus widening its application in the chemical industry.

X. Yang et al. / Particuology 8 (2010) 372–378


Fig. 9. TEM images of BaSO4 nanoparticles (CP0 = 0.05 mol/L, a = 5): (a) RPB, N = 800 rpm, VA = 420 L/h; (b) stirred tank, N = 400 rpm.

4. Conclusions Experiments of BaSO4 precipitation within a RPB having sampling tubes along different radial positions verified that the inlet region is vital for the liquid–liquid precipitation process. This result is significantly favorable for the packing design of industrial RPB (e.g. 40–50 mm of optimum radial packing in our experiments). The product particle size decreases with increasing rotation speed, liquid flow rate and reactant concentrations, and increase with increasing volumetric ratio. Moreover, XRD and BET results confirmed that the BaSO4 nanoparticles prepared by RPB are of higher quality as compared to particles prepared by stirred-tank reactor, and the estimated induction time of nucleation was comparable to the characteristic time of micromixing in RPB. Acknowledgements This work was supported by the National Natural Science Foundation of China (Nos. 20821004, 20990221), the Beijing Municipal Commission of Education (No. JD100100403), and the Innovation Team Program of Ministry of Education of China. References Aslund, B., & Rasmuson, A. C. (1992). Semibatch reaction crystallization of benzoic acid. AIChE Journal, 38, 328–342. Baldyga, J., Podgorska, W., & Pohorecki, R. (1995). Mixing-precipitation model with application to double feed semibatch precipitation. Chemical Engineering Science, 50, 1281–1300.

Barresi, A. A., Marchisio, D., & Baldi, G. (1999). On the role of micro- and mesomixing in a continuous Couette-type precipitator. Chemical Engineering Science, 54, 2339–2349. Chen, J. F., & Shao, L. (2003). Mass production of nanoparticles by high gravity reactive precipitation technology with low cost. China Particuology, 1, 64–69. Chen, J. F., Zheng, C., & Chen, G. T. (1996). Interaction of macro- and micromixing on particle size distribution in reactive precipitation. Chemical Engineering Science, 51, 1957–1966. Chen, J. F., Wang, Y. H., Guo, F., Wang, X. M., & Zheng, C. (2000). Synthesis of nanoparticles with novel technology: High-gravity reactive precipitation. Industrial & Engineering Chemistry Research, 39, 948–954. Chen, G. G., Luo, G. S., Xu, J. H., & Wang, J. D. (2004). Membrane dispersion precipitation method to prepare nanoparticles. Powder Technology, 139, 180– 185. Chen, J. F., Zhou, M. Y., Shao, L., Wang, Y. Y., Yun, J., Chew, N. Y. K., et al. (2004). Feasibility of preparing nanodrugs by high-gravity reactive precipitation. International Journal of Pharmaceutics, 269, 267–274. Danckwerts, P. V. (1958). The effect of incomplete mixing on homogeneous reactions. Chemical Engineering Science, 8, 93–102. David, R. (2001). General rules for prediction of the intensity of micromixing effects on precipitations. Powder Technology, 121, 2–8. David, R., & Marcant, B. (1994). Prediction of micromixing effects in precipitation: Case of double-jet precipitators. AIChE Journal, 40, 424–432. Dirken, J. A., & Ring, T. A. (1991). Fundamentals of crystallization: Kinetic effects on particle size distributions and morphology. Chemical Engineering Science, 46, 2389–2427. Falk, L., & Schaer, E. (2001). A PDF modelling of precipitation reactors. Chemical Engineering Science, 56, 2445–2457. Fowler, R., & Khan, A. S. (November 1987). VOC removal with a rotary air stripper. In AIChE annual meeting New York. Garside, J., & Tavare, N. S. (1985). Mixing, reaction and precipitation: Limits of micromixing in an MSMPR crystallizer. Chemical Engineering Science, 40, 1485–1493. Guichardon, P., & Falk, L. (2000). Characterization of micromixing efficiency by the iodide–iodate reaction system. Part I. Experimental procedure. Chemical Engineering Science, 55, 4233–4243.


X. Yang et al. / Particuology 8 (2010) 372–378

Guo, K., Guo, F., Feng, Y. D., Chen, J. F., Zheng, C., & Gardner, N. C. (2000). Synchronous visual and RTD study on liquid flow in rotating packed-bed contactor. Chemical Engineering Science, 55, 1699–1706. Jia, Z. Q., & Liu, Z. Z. (2002). Synthesis of nanosized BaSO4 particles with a membrane reactor: Effects of operating parameters on particles. Journal of Membrane Science, 209, 153–161. Kelleher, T., & Fair, J. R. (1996). Distillation studies in a high-gravity contactor. Industrial & Engineering Chemistry Research, 35, 4646–4655. Kockmann, N., Kastner, J., & Woias, P. (2008). Reactive particle precipitation in liquid microchannel flow. Chemical Engineering Journal, 135, S110–S116. Lin, C. C., Liu, W. T., & Tan, C. S. (2003). Removal of carbon dioxide by absorption in a rotating packed bed. Industrial & Engineering Chemistry Research, 42, 2381–2386. Marchisio, D. L., Barresi, A. A., & Garbero, M. (2002). Nucleation, growth and agglomeration in barium sulfate turbulent precipitation. AIChE Journal, 48, 2039–2050. Mersmann, A. (1999). Crystallization and precipitation. Chemical Engineering & Processing, 38, 345–353. Nielsen, A. E. (1964). Kinetics of precipitation. New York: Macmillan Co. Schwarzer, H. C., & Peukert, W. (2004). Combined experimental/numerical study on the precipitation of nanoparticles. AIChE Journal, 50, 3234–3247.

Sue, K., Kimura, K., & Aral, K. (2004). Hydrothermal syntheesisi of ZnO nanocrystals using microreactor. Materials Letters, 58, 3229–3231. Takagi, M., Maki, T., Miyahara, M., & Mae, K. (2004). Production of titania nanoparticles by using a new microreactor assembled with same axle dual pipe. Chemical Engineering Journal, 101, 269–276. Tavare, N. S. (1995). Mixing, reaction and precipitation: Interaction by exchange with mean micromixing models. AIChE Journal, 41, 2537–2548. Torbacke, M., & Rasmuson, A. C. (2001). Influence of different scales of mixing in reaction crystallization. Chemical Engineering Science, 56, 2459–2473. Wang, H. Z., Nakamura, H., Uehara, M., Miyazaki, M., & Maeda, H. (2002). Preparation of titania particles utilizing the insoluble phase interface in a microchannel reactor. Chemical Communications, 14, 1462–1463. Yang, H. J., Chu, G. W., Zhang, J. W., Shen, Z. G., & Chen, J. F. (2005). Micromixing efficiency in a rotating packed bed: Experiments and simulation. Industrial & Engineering Chemistry Research, 44, 7730–7737. Zauner, R., & Jones, A. G. (2002). On the influence of mixing on crystal precipitation processes: Application of the segregated feed model. Chemical Engineering Science, 57, 821–831.