Power Consumption in Shaking Flasks

on Rotary Shaking Machines: I. Power

Consumption Measurement in Unbaffled
Flasks at Low Liquid Viscosity

Jochen Büchs,1 Ulrike Maier,1 Claudia Milbradt,1 Bernd Zoels2

1
Chair of Biochemical Engineering, Sammelbau, Aachen University of
Technology, D-52074 Aachen, Germany, telephone: +49-241-805546;
fax: +49-241-8888-265; e-mail: buechs@biovt.rwth-aachen.de
2
ZET/ZT-A15, BASF AG, Ludwigshafen, Germany
Received 7 September 1999; accepted 14 February 2000

Abstract: In this first article of a series a new method is perimental results may be readily reproduced in other labo-
introduced that enables the accurate determination of ratories.
the power consumption in a shaking flask. The method is
based on torque measurements in the drive and appro- The power consumption (per unit volume) in fermenta-
priate compensation of the friction losses. The results for tion broth plays an essential role in fermentation of aerobic
unbaffled shaking flasks at low viscosities are presented microbes in shaking flasks and agitated tank bioreactors. It
after varying shaking frequency, flask size, filling volume, is regarded as one of the crucial values for characterizing
shaking diameter, and surface quality (hydrophilic and
and upscaling cultures (Sumino et al., 1972). The math-
hydrophobic) of the inner flask walls. The order of mag-
nitude of the values of power consumption in shaking ematical estimation of the power consumption in agitated
flasks is equal to, or even higher than, the values typical tank bioreactors of different geometries is no longer a prob-
for agitated tank bioreactors. A physically based model lem due to much research in this area. The nondimensional
equation for shaking flasks is derived that introduces a description of power as the power number (or Newton num-
modified power number and a resulting constant as the
only fitting parameter. With this equation, the measured ber) as a function of the Reynolds number is well-
results are correlated with sufficient accuracy. For the established. Yet, regarding shaking flasks, little is known
first time, comprehensive data for the power consump- about how much power to put into the fermentation broth
tion in unbaffled shaking flasks at low viscosity is avail- and which relating variables are of importance.
able, giving a detailed picture of the influences of the
different variables. © 2000 John Wiley & Sons, Inc. Biotech-
The first published attempt at determining the power con-
nol Bioeng 68: 589–593, 2000. sumption is based on temperature increase in heat-insulated
Keywords: power consumption; measuring method; flasks (Sumino et al., 1972). The actual temperature in-
shaking flasks; modeling, Reynolds number; power num- crease, however, was quite low. Interpretation of these re-
ber sults is complicated by the heat capacity of the wetted glass
walls and also by the undefined thermal conduction to the
INTRODUCTION unwetted glass material (upper part of the flask). Another
method for determining the power consumption in fermen-
Shaking flasks have proven to be an invaluable standard tation broth is to record the total electric power consumption
tool in practical biotechnological work—one reason being, of the whole shaking machine. This method was used suc-
among others, their ease of handling (Calam, 1986; Mc-
cessfully to evaluate the power consumption in cylindrical
Daniel et al., 1965; van Suijdam et al., 1978). Many experi-
vessels with relatively large liquid volumes of 2.1 to 6.9 L
ments can be carried out simultaneously with minimal ma-
(Kato et al., 1995). Vessels with a liquid height equal to the
terial expense and practically no supervision. Shaking flasks
vessel diameter were used for the regeneration of very sen-
are often used in the first screening stages of biotechnologi-
sitive hairy root cells (Kato et al., 1997). However, these
cal development work. Because very decisive selection and
conditions are very specific for this particular application.
direction is undertaken using shaking flasks, it must be en-
When employing our commercially available shaking ma-
sured that their use does not take place in unsuitable (i.e.,
chines under conditions typical for cultivating microorgan-
nonconsistent) experimental conditions (Clark et al., 1995).
An accurate definition of the factors evaluating the perfor- isms (volumes of <0.2 L) the registration of the electric
mance of shaking flasks is therefore invaluable so that ex- power consumption turned out to be highly inaccurate. The
efficiency of the electrical drive is difficult to determine and
varies considerably with the load. Easily reproducible re-
Correspondence to: J. Büchs sults, however, were obtained with a measurement device,

© 2000 John Wiley & Sons, Inc.

including a mechanical torque indicator placed in the exter- f(Re)⳱constant] succeeds by the definition of a modified
nal drive of the agitator. The advantage of this method is power number, Ne⬘, already used by Clark and Vermeulen
that the power consumption of up to 63 shaking flasks can (1964):
be measured simultaneously, and therefore there is much
d P
greater accuracy. Ne⬘ = Ne ⭈ = = C1 ⭈ ␲4 ⭈ f 共Re兲 (7)
The aim this study is to introduce a new method for h ␳ ⭈ n3 ⭈ d 4 ⭈ h
reliable determination of the power consumption in shaking Analogous to the description of power consumption in agi-
flasks. For the first time, the influence of relevant param- tated tank bioreactors a model for shaking flasks is derived
eters on power consumption (per unit volume) can be ac- in what follows. The required friction area (A) is the inter-
curately determined. An investigation of the influence of facial area between rotating liquid and flask walls. The de-
elevated viscosity and of baffles will be presented in sub- tailed shape of the circulating liquid in the flask varies with
sequent articles in this series. the speed, filling volume, and shaking diameter of the shak-
ing machine. If viscosity is low, the surface can be itera-
MATHEMATICAL DESCRIPTION OF tively calculated (Büchs et al., 1995). However, this in-
TEST RESULTS volves a great deal of calculation effort and the result can no
longer be described as a simple applicable explicit equation.
Power (P) is generally considered as the product of torque
In contrast to the exact shape, the total integral value of
(M) and angular velocity (␻). On the other hand, torque can
the friction area (A) varies only little with the speed and
be described as the product of force (F) and lever arm (r).
shaking diameter of the shaking machine (Büchs et al.,
Force is obtained by multiplying shear stress (␶) and the
1995). Therefore, the simplified quantity, h ≅ V1/3
L , describes
corresponding friction area (A). Consequently:
a characteristic length scale for the friction area between
P ⳱ M䡠␻ ⳱ F䡠r䡠␻ ⳱ ␶䡠A䡠r䡠␻ (1) liquid and flask walls. This length scale is in form of a
simple applicable equation and is sufficient for our pur-
If the agitated tank is considered first, then the friction area
poses.
(A) can be expressed as a cylindrical plane circumscribed by
Under the simplifying assumption that turbulent flow pre-
the tips of the stirrer blades. In this case, we obtain:
vails in the large and small flasks, the correlation, f(Re) ⳱
A ⳱ 2䡠␲䡠r䡠h (2) C2 䡠 Re−0.2, for turbulent flow conditions in undisturbed hy-
drodynamic systems (Schlichting, 1979) may be applied.
In accordance with basic hydrodynamic knowledge, shear The equation for a nondimensional power number in un-
stress is obtained as follows (Schlichting, 1979): baffled flasks is then analogous to Eq. (7):
␶ ⳱ C1 䡠 ␳ 䡠 r2 䡠 ␻2 䡠 f(Re) (3) P
Ne⬘ = = C1 ⭈ ␲4 ⭈ C2 ⭈ Re−0.2 = C3 ⭈ Re− 0.2
with ␳ ⭈ n ⭈ d 4 ⭈ V L1 Ⲑ 3
3

(8)
␳⭈n⭈d2
Re = (4) As shown in Eq. (4), the Reynolds number (Re) is defined
␩
using the maximum inside flask diameter. This length scale
With ␻ ⳱ 2 䡠 ␲ 䡠 n and d ⳱ 2 䡠 r, Eqs. (1)–(3) can be com- characterizes the actual flow of the circulating liquid inside
bined as: the flask. In contrast, Kato et al. (1995) used the shaking
diameter as a length scale.
P ⳱ C1 䡠 ␲4 䡠 f(Re) 䡠 ␳ 䡠 n3 䡠 d4 䡠 h (5)
For direct description of the power consumption in shak-
This equation can be transferred to: ing flasks, Eq. (8) can be transformed to:
P h P n3 ⭈ d 4 n3 ⭈ d 4
Ne = = C1 ⭈ ␲4 ⭈ f 共Re兲⭈ (6) = Ne⬘ ⭈ ␳⭈ 2 Ⲑ 3 = C3 ⭈ ␳⭈ 2 Ⲑ 3 ⭈ Re− 0.2 (9)
␳⭈n ⭈d
3 5 d VL VL VL
The power number (Ne) therefore consists of a constant Eq. (9) implies that the specific power consumption is de-
product (C1 䡠 ␲4), a function influenced by the Reynolds pendent on the shaking frequency according to P ∼ n2.8.
number [f(Re)], and the dimensionless group (h/d). There- Such a correlation is typically found in unbaffled agitated
fore, it is no surprise that most models mentioned in the tank bioreactors. The model includes C3 as the only fitting
literature for description of the power number, which is parameter for description of all results.
influenced by geometrical stirrer dimensions, contain the
h/d group with an exponent close to 1 ( Bates et al., 1963;
MATERIALS AND METHODS
Bittins and Zehner, 1994; Braginski and Begachev, 1972;
Clark and Vermeulen, 1964; Lee et al., 1994; O’Connell The rotary shaking machine is firmly fixed to a frame. The
and Mack, 1964; Sano and Usui, 1985, 1987). A global shaking table is powered by a drive with a torque sensor (0
description of the power consumption of geometric dissimi- to 1 Nm). Torque and speed data are recorded and saved on
lar agitators in fully baffled turbulent tank reactors [⇒ a PC via an A/D converter. From the shaking frequency and

590 BIOTECHNOLOGY AND BIOENGINEERING, VOL. 68, NO. 6, JUNE 20, 2000
the measured torque the power consumption of the shaker drive with an integrated torque sensor of 0 to 1 Nm (Type
drive can be calculated with the first part of Eq. (1). The MRD1, Janke & Kunkel, Staufen, Germany) and a unit for
effect on measurement is increased by mounting as many speed control and speed measurement (Type RE162E,
flasks on the shaking table as possible. Mechanical friction Janke & Kunkel) was recalibrated at the beginning of the
loss in the drive and wind resistance of the flasks are com- test series. The device was turned horizontally and a defined
pensated by performing referential measurements with so- lever arm (pulley wheel) was mounted to the torque sensor
lidified agar of the same weight. Agar prevents the move- shaft. A thread was attached to the wheel and connected to
ment of the liquid and therefore power consumption. From a weight, standing on a balance. While varying the thread
the difference between measurements with normal liquid tension by vertically moving the balance, the corresponding
and agar-hardened liquid the power consumption (per unit weight and indicator values of the torque measurement de-
volume) can be calculated by means of vice were recorded and correspondingly converted.
P 共M1 − M2兲⭈2⭈␲⭈n The shaking flasks were filled with demineralized water
= (10) and closed with aluminum foil to avoid evaporation of water
VL z⭈VL during the test series. The power consumption in the de-
Not taken into account in this compensation method is that, mineralized-water-filled flasks was measured. The shaking
with normally rotating liquid, the center of gravity of the frequency was increased in steps and at each stage the mean
contents moves around at a certain distance from the middle of torque and speed signals was taken for at least 7 min. For
of the flask. This results in another tilting momentum being referential measurement, 3 wt% agar (Type 0138-01-4,
transmitted to the shaking machine bearings, possibly lead- Difco, Detroit, MI) was added to the shaking flask, subse-
ing to additional mechanical friction losses not registered by quently heated in an autoclave to 121°C for 1 min, and then
referential measurements with agar-hardened liquid. To ex- cooled down. The difference in weight due to heating was
amine the influence of the aforementioned tilting moment, minimal. Afterward, the referential measurements were
the flasks were each filled with a steel ball of the same mass, taken with agar-hardened liquid.
producing almost the same movement of its center of grav- For examination, narrow-necked unbaffled glass flasks of
ity around the middle of the flask. It is assumed that the steel 100-, 250-, 300-, 500-, 1000-, and 2000-mL (German stan-
ball rotates frictionless in the flask, thus producing the same dard, DIN 12380) volume were measured. Figure 3 shows
friction loss in the bearings as the rotating liquid. the maximum inside diameters of the different flasks.
However, the examinations showed that referential mea- To obtain a definite hydrophilic glass surface, the flasks
surements with agar-hardened liquid and steel balls led to were filled with 10% HNO3 and heated in an autoclave to
the same results as with our mechanical set-up. The com- 120°C for 20 min. To obtain hydrophobic glass walls, the
pensation method with agar was therefore used for most inside surface of the flasks was wetted (5 min) with dichlo-
measurements. Similar compensation methods were devel- rodimethylsilane (Sigma, 5 wt% in toluene).
oped by Kato et al. (1995). Their cylindrical vessels were For this work measurements with water were performed
filled with plasticine of the same weight as the liquid, or the with variation of shaking frequency (80 to 380 min−1), flask
liquid rotation was prevented by placing a lid on the liquid size (100 to 2000 mL), filling volume (4% to 20% of nomi-
surface. nal volume), and surface quality of the flask walls on rotary
The torque measurement device, containing an agitation shaking machines with 2.5-cm and 5-cm shaking diameters.
All measurements were taken at a temperature of 37°C. The
measurement error using this method depends on the filling
volume but was always less than ±5% of the measuring
value. Figure 2 shows the reproduced measurement series
for a filling volume of 10 mL, taken 3 months after the first
series.

RESULTS AND DISCUSSION

To examine the influence of the filling volume and the
shaking frequency, 250-mL flasks were filled with different
volumes and measured at different speeds. Figure 2 shows
the results of the investigations.
It is noteworthy that, in general, in shaking flasks rela-
tively high-power consumption values are obtained, which
are at least of the same order of magnitude as the usual
values in agitated tank bioreactors. Power consumption in-
Figure 1. Measuring device for determination of the power consumption
creases with higher shaking frequency as expected. With
in shaking flasks: (1) drive with integrated torque sensor; (2) free wheel;
(3) flexible coupling; (4) rotary table shaker; (5) speed measurement and increasing filling volume, the power consumption de-
control; (6) torque measurement; (7) A/D converter; and (8) data acquisi- creases, because the friction area between the glass walls
tion. and liquid does not rise in directly proportion to the filling

BÜCHS ET AL.: POWER CONSUMPTION IN SHAKING FLASKS ON ROTARY SHAKING MACHINES. I 591
 Figure 3. Power consumption in hydrophilic shaking flasks, each filled
Figure 2. Power consumption in hydrophilic 250-mL flasks at different
with 1/10 of the nominal flask volume: (—) 2.5-cm shaking diameter; (---)
shaking frequencies and filling volumes, and 2.5-cm shaking diameter. (䊉)
5-cm shaking diameter.
Reproduced measurement series for 10-mL filling volume, taken 3 months
after the first series.
equation for power consumption proposed by Kato et al.
(1995) cannot be applied to our results. This is possibly due
volume. The order of magnitude of the power consumption to the completely different shaking conditions examined. In
and its qualitative dependency on shaking frequency and the shaking frequency and volume range investigated by
filling volume verify the earlier results of Sumino et al. Kato et al. the bottom of the shaking vessel was still com-
(1972). A quantitative comparison is not possible, however, pletely wetted. Therefore, with increasing shaking fre-
because they did not specify the detailed conditions of the quency, the friction area between the vessel wall and the
five measured points they obtained from the unbaffled shak- rotating liquid will also increase. This could be a reason
ing flasks on the rotary shaking machines (see Fig. 7 of why Kato et al. found an unusually strong dependency of
Sumino et al., 1972). the shaking frequency on the power consumption (P ∼
Figure 3 shows results after varying the flask size at two n5.75). For the shaking conditions investigated in our work,
different shaking diameters. Obviously, at low liquid vis- the middle of the flask bottom usually ran dry, resulting in
cosity, the shaking diameter generally has no significant a dependency of P ∼ n2.8.
effect on power consumption. The only minor exception As Figure 4 shows, a larger error of calculated power
appears to be in the 2000-mL flask. The reason for this consumption occurs in the range of 0.01 to 0.2 kW/m3. This
deviation is presently being investigated in our laboratory. can be physically explained by the fact that these values
Another noticeable effect, which will be presented in sub- were taken at low shaking frequency (80 to 120 min−1) and,
sequent articles from this series, occurs at higher viscosities therefore, at low Reynolds numbers (Re ≈ 5000 to 50,000).
and in baffled flasks (Büchs et al., 1999). At equal shaking Presumably, in this area, the transition between laminar and
frequency, the power consumption increases with increas- turbulent flow can be found. This is supported by the com-
ing flask size, because the relative velocity between rotating parison with the critical Reynolds numbers of the flow
liquid and flask walls rises with the flask diameter. The caused, for example, by a rotating disk in a resting fluid
results from the 250-mL and 300-mL flasks are almost iden-
tical, because the geometrical dimensions of these flasks
scarcely differ.
The surface quality of the glass walls of the flasks (hy-
drophilic and hydrophobic) shows no significant effect on
the power consumption (results not shown). This is of much
interest because we discovered a considerable difference in
oxygen transfer between flasks with hydrophilic and hydro-
phobic surfaces in earlier works (Büchs et al., 1993).
In Figure 4, measured values of power consumption are
plotted against values calculated by Eq. (9). A value of 1.94
was obtained for the constant C3 by least-squares nonlinear
fitting. It should be emphasized that the constant C3 is the
only parameter to be fitted for description of the measured
data obtained at different operating conditions. At power
consumption rates of higher than about 0.5 kW/m3, which is Figure 4. Measured and calculated [Eq. (9)] specific power consumption
typical for microbiological experiments, almost all mea- in shaking flasks: shaking frequency (n) 80 to 380 min−1; filling volume
sured results lie within the range of ±30% tolerance. The (VL) 4% to 20% of nominal flask volume; 2.5 to 5 cm shaking diameter.

592 BIOTECHNOLOGY AND BIOENGINEERING, VOL. 68, NO. 6, JUNE 20, 2000
(Recrit ≈ 50,000; Schlichting, 1979) or the flow between two References
concentrically rotating cylinders (inner cylinder rests, outer
Bates RL, Fondy PL, Corpstein RR. 1963. An examination of some geo-
cylinder rotates, Recrit ≈ 66,000; Schlichting, 1979). Our metric parameters of impeller power. Ind Eng Chem Proc Des Dev
model was not derived for laminar flow conditions. Further- 2:310–314.
more, the main liquid–flask interaction at low shaking fre- Bittins K, Zehner P. 1994. Power and discharge numbers of radial-flow
quency takes place at the flask base, which our model does impellers. Fluid-dynamic interactions between impeller and baffles.
not account for. Therefore, it is no surprise that, in this Chem Eng Prog 33:295–301.
Braginski LN, Begachev VI. 1972. Interrelation between the circumferen-
region, our model does not fit the measured data adequately. tial liquid velocity and the power in mixing. Teor Osn Khim Tekhnol
It can be concluded that a reliable new method has been 6:260–268.
developed to measure power consumption in shaking flasks. Büchs J, Haarde W, Zehner P. 1993. Sauerstoffversorgung von Mikroor-
In this study, a physically based model is presented that ganismen in Schüttelkolben. Paper presented at the Fachausschussta-
satisfactorily fits the measured data of practical relevance gung “Bioverfahrenstechnik,” Potsdam, Germany.
Büchs J, Kramm H, Mayer B, Zoels B. 1995. Messung und Modellierung
for unbaffled flasks of ⱕ1 L nominal volume at low liquid des Leistungseintrages in Schüttelkolben. Paper presented at GVC-
viscosity. An extended approach will be presented in a sub- Fachausschusstagung “Mischvorgänge,” Münster, Germany.
sequent study to determine the power consumption in shak- Büchs J, Maier U, Milbradt C, Zoels B. 2000. Power consumption in
ing flasks at elevated viscosities of up to 200 mPa 䡠 s. shaking flasks on rotary shaking machines: II. Nondimensional de-
scription of the specific power consumption and flow regimes in un-
These experiments were performed at the life science laboratory baffled flasks at elevated liquid viscosity. Biotechnol Bioeng 68:
of BASF AG, Ludwigshafen, Germany. The authors thank B. 594–601.
Mayer and the staff members of the biological pilot plant (ZHV/ Calam CT. 1986. Shake-flask fermentations. In: Demain AL, Solomon NA,
TB) for their technical assistance. The support of D. Nicholls editors. Manual of industrial microbiology and biotechnology Wash-
during preparation of this article is gratefully acknowledged. ington, DC: American Society for Microbiology. p 59–65.
Clark MW, Vermeulen T. 1964. Incipient vortex formation in baffled
agitated vessels. AIChE J 10:420–422.
NOMENCLATURE Clark GJ, Langley D, Bushell ME. 1995. Oxygen limitation can induce
microbial secondary metabolite formation: investigations with minia-
A friction area (m2) ture electrodes in shaker and bioreactor culture. Microbiology 141:
C1, C2 constants (−) 663–669.
C3 constant C3 ⳱ C1 䡠 ␲4 䡠 C2 (−) Kato Y, Hiraoka S, Tada Y, Sato K, Ohishi T. 1997. Measurement of mass
d stirrer diameter, maximum inner diameter of flask (m) transfer rate from free surface in shaking vessel type bioreactor. J
F force (N) Chem Eng Jpn 30:362–365.
f(Re) function, depending on flow regime (−) Kato Y, Hiraoka S, Tada Y, Shirai S, Ue T, Koh S-T, Yamaguchi T. 1995.
h height of stirrer blades, vertical length scale for the friction area Power consumption of horizontally shaking vessel with circulating
between liquid and flask walls (m) motion. Kagaku-Kogaku-Ronbunshu 21:365–371.
M torque (Nm) Lee Y-S, Lee S-K, Kim J-S. 1995. Estimation of power input in a baffled
M1 measured torque with normal liquid (Nm) agitated vessel. J Korean Inst Chem Eng 33:46–55.
M2 measured torque with agar-hardened liquid (Nm) McDaniel LE, Bailey EG, Zimmerli A. 1965. Effect of oxygen supply rates
n shaking frequency, number of revolutions (s−1) on growth of Escherichia coli. I. Studies in unbaffled and baffled
Ne power number, Ne ⳱ P/␳ 䡠 n3 䡠 d5 (−) shake flasks. Appl Microbiol 13:109.
Ne⬘ modified power number, Ne⬘ ⳱ P/␳ 䡠 n3 䡠 d4 䡠 V1/3 L (−) O’Connell FP, Mack DE. 1950. Simple turbines in fully baffled tanks.
P power (W) Chem Eng Progr 46:358–362.
r lever arm of the power-input device (m) Sano Y, Usui H. 1985. Interrelations among mixing time, power number
Re Reynolds number, Re ⳱ n 䡠 d2/␩ (−) and discharge flow rate number in baffled mixing vessels. J Chem Eng
Recrit critical Reynolds number (−) Jpn 18:47–52.
VL filling volume of one flask (m3) Sano Y, Usui H. 1987. Effects of paddle dimensions and baffle conditions
z number of shaking flasks per shaking table (−) on the interrelations among discharge flow rate, mixing power and
mixing time in mixing vessels. J Chem Eng Jpn 20:399–404.
Greek symbols Schlichting H. 1979. Boundary layer theory, 7th ed. New York: McGraw
Hill.
␩ dynamic viscosity of fluid (Pa 䡠 s) Sumino Y, Akiyama S, Fukada H. 1972. Performance of the shaking flask.
␳ liquid density (kg/m3) I. Power consumption. J Ferment Technol 50:203–208.
␻ angular velocity (s−1) van Suijdam JC, Kossen NWF, Joha AC. 1978. Model for oxygen transfer
␶ shear stress (N/m2) in a shake flask. Biotechnol Bioeng 20:1695–1709.

BÜCHS ET AL.: POWER CONSUMPTION IN SHAKING FLASKS ON ROTARY SHAKING MACHINES. I 593