International Dairy Journal 51 (2015) 84e91

Contents lists available at ScienceDirect

International Dairy Journal

journal homepage: www.elsevier.com/locate/idairyj

Steam-frothing of milk for coffee: Evaluation for foam properties using

video analysis and feature extraction
Morten Münchow a, c, d, e, Leif Jørgensen b, Jose Manuel Amigo a, Klavs Sørensen a,
Richard Ipsen a, *
a
Department of Food Science, Faculty of Science, University of Copenhagen, Rolighedsvej 30, DK-1958 Frederiksberg C, Denmark
b
Naturmælk Amba, Gerrebækvej 24, DK-6360 Tinglev, Denmark
c
CoffeeMind, Hansstedvej 35, 2500 Valby, Denmark
d
Europa 1989, Amagertorv 1, DK-1160 Copenhagen K, Denmark
e
The Speciality Coffee Association of Europe, Bicknacre, Chelmsford, Essex CM3 4HF, United Kingdom

a r t i c l e i n f o a b s t r a c t

Article history: A method for evaluation of the foaming properties of steam-frothed milk, based on image analysis
Received 13 March 2015 (feature extraction) carried out on a video taken immediately after foam formation, was developed. The
Received in revised form method was shown to be able to analyse steam-frothed milk made using a conventional espresso ma-
21 July 2015
chine, such as commonly used by baristas. Samples of milk, processed in a commercial dairy plant, were
Accepted 22 July 2015
made with varying fat (0.5%, 0.9%, 1.5%, 2.6%, 3.5%) or protein (3.0%, 3.4%. 3.8%) content and analysed
Available online 4 August 2015
using the developed method. Increased protein content was shown to cause a delay in the formation of a
distinct foam layer and also a slightly increased foam volume. Increased fat content was found to
decrease foam stability but also resulted in less drainage.
© 2015 Elsevier Ltd. All rights reserved.

1. Introduction Goh, Kravchuk, and Deeth (2009) compared steam frothing of

milk with mechanical agitation and air bubbling for foaming of
Steam injection or steam frothing of milk refers to milk being various milk types. They found little difference in the stability of the
foamed by incorporation of steam, blown into the milk through a foams but showed that steam frothing produced significantly
nozzle (Huppertz, 2010). This process will cause a considerable stronger foams compared with air bubbling. Huppertz (2010) noted
increase in the temperature of the milk, which in practice, however, that, with respect to the effects of composition and processing on
does not exceed 65
C and rapidly decreases to <50
C within 4 min the foaming properties of skim milk, foams prepared by different
s, milk is usually added to coffee as a
(Silva et al., 2008). In cafe methods (mechanical agitation or stream injection) could be
steam-frothed foam to produce products such as cafe latte and considered similar.
cappuccino. The foam is produced by the cafe  staff (also known as A main factor known to affect the quality of steam frothed milk
baristas) in an artisanal manner and requires skills to maximise the is the degree of lipolysis, which depresses the foaming ability and
quality of the final coffee. Steam-frothing puts specific demands on the foam stability, and increases the coarseness of foams (Deeth &
the milk used to ensure a consistent high quality foam. This, in turn, Smith, 1983; Kamath, Wulandewi, & Deeth, 2008b). In fact, there
has generated an increasing interest from dairy companies in appears to be a critical concentration of free fatty acids above which
marketing milk specifically suited for steam-frothing (e.g., mar- milk foams made using air injection are extremely unstable, irre-
keted as barista milk or cafe  milk). There is also, from our experi- spective of the temperature of foaming (Kamath et al., 2008b).
ence, considerable and increasing interest from producers of coffee Addition of citrate to milk concentrate during powder manufacture
makers and vending machines. or to milk reconstituted from milk powders has also been shown to
enhance the steam-frothing properties of milk (Augustin & Clarke,
2008). Ward, Goddard, Augustin, and McKinnon (1997) found that
addition of EDTA increased foamability and foam stability of
* Corresponding author. Tel.: þ45 35333225.
reconstituted skim milk powder when whipped, and attributed this
E-mail address: ri@food.ku.dk (R. Ipsen).
URL: http://coffee-mind.com to the release of caseins into the serum phase. In foams made using

http://dx.doi.org/10.1016/j.idairyj.2015.07.009
0958-6946/© 2015 Elsevier Ltd. All rights reserved.
 M. Münchow et al. / International Dairy Journal 51 (2015) 84e91 85

air-injection, however, addition of chelating agents has been shown Italy). The espresso machine was turned on 2 h before the first
to decrease the foam stability (Deeth & Lewis, 2015), highlighting sample was run to ensure an even heat distribution and steam
that the method (as well as the temperature) of foaming has a generation. The steam wand has three holes on the tip that creates a
major influence on obtained results. jet stream of steam. Each of the holes is at 120
to each other and
Little is known in detail about the variations in raw milk they are at the same distance with reference to the centre of the
composition affecting the quality of manually steam-frothed milk nozzle. Furthermore, the holes have a 45
grade with reference to
foam, but baristas, in our experience, often complain about day-to- the vertical axis of the steam wand and a 1.5 mm diameter.
day variations. Hence, to provide novel insight into the mechanisms A standard steaming procedure aimed at mimicking the pro-
behind stabilisation of steam-frothed milk and the factors influ- fessional performance of a barista was developed in a separate pilot
encing it, we have developed a method to evaluate the foam study. The jug was placed on an elevation table, ensuring the steam
properties. nozzle of the espresso machine was fixed precisely at the surface of
This method is simple to execute, can be set up in cafe s and the milk and thus able to create an overrun similar to a barista
barista training centres/schools, and is based on recording and making a cappuccino (approximately 25e30%). The overrun is
measuring the foam development evolution. The evolution of the influenced by the positioning of the steam nozzle; the higher the
initial foam formation and subsequent drainage dynamics is nozzle is placed above the surface of the milk, the higher the ach-
recorded in image frames from the transmitted light over time and ieved overrun. The obtained overrun was measured and was within
subsequently treated by image analysis techniques. The method the indicated target interval at all times (results not shown). The
has been utilised to explore how variations in fat and protein pressure in the boiler of the espresso machine was regulated to be
content in milk affect the stability of foam made by steam-frothing between 132 and 152 kPa using the built-in manometer. Prior to
using a conventional espresso machine. placing the milk jug on the elevation table, the machine was shut
off and the steam tank was drained for steam through the steam
2. Materials and methods nozzle until the pressure dropped to 112 kPa. This was done to
avoid variance caused by random states of on/off settings of the
2.1. Milk boiler. The machine was then turned back on and when the pres-
sure reached 132 kPa the steaming procedure was initiated,
Raw bulk milk was processed in a commercial dairy plant ensuring the boiler was always on during the steaming procedure.
(Naturmælk amba, Tinglev, Denmark) and was initially stand- The temperature of the milk in the jug was measured just before
ardised with respect to fat (0.5%, 0.9%, 1.5%, 2.6%, 3.5%) and protein the steaming procedure started and was typically between 4 and
(3.0%, 3.4%. 3.8%) content using a milk separator (MSE 8501177; 5
C after being transferred to the steel jug. Temperature was
GEA Westfalia, Oelde, Germany) with a capacity of 10,000 L h1 and recorded during steaming and the steaming process was termi-
a UF plant (DSS, Silkeborg, Denmark) fitted with spiral poly- nated when the temperature reached 65
C above the initial
propylene membranes (GR70PP 6338/30, Alfa Laval, Lund, Swe- measured temperature just prior to steaming (i.e., 70
C ± 1
C).
den). Following standardisation, the milk was subjected to Immediately after completion of the steaming procedure the
conventional pasteurisation (72
C for 15 s) in a plate heat 400 mL of steamed milk was transferred into a specially made glass
exchanger (Type N35, SPX Flow Technology, Kolding, Denmark, vessel with internal dimensions of 140  135  20 mm.
capacity 10,000 L h1). The resulting fifteen samples of milk (1 kg
each) were delivered to the university following each day of 2.3. Measurement of foam properties
manufacture and stored at 5
C for 48 h until analysis. Note that the
milk used was not homogenised since Naturmælk does not A custom-made measuring apparatus (Fig. 1) was used, con-
homogenise milk for any of their products. Also, most specialty sisting of a box (280  205  205 mm; built from wooden plates of
coffee companies in Denmark do, in fact, use non-homogenised 7 mm thickness) painted black on the inside. The glass vessel was
milk since this is perceived as a more natural/less processed placed at one end of the box and a video camera (Guppy F-080;
product. Three independent repetitions were done and resulting Allied Vision Technologies GmbH, Stadtroda, Germany), fitted with
samples were systematically steam-frothed and analysed three a Cosmicar 6 mm 1:1.2 lens (Ricoh, Tokyo, Japan) was fitted into a
days after initial processing. hole (Ø 40 mm) placed exactly in centre of the plate opposite the
glass vessel. A light source (Photolamp Kaiser RB 260 digital (Kaiser
2.2. Steam-frothing Fototechnik GmbH, Buchen, Germany) with an Osram Dulux S
11W/11-860 lamp (Osram GmbH, Munich, Germany)) with a dif-
Milk at 4
C (400 g) was transferred into a 1000 mL stainless fusor in front was placed 100 mm from the glass vessel and
steel jug and steam-frothed on a conventional espresso machine pointing directly towards it. The camera, the wooden box and the
(Dalla Corte Super Mini, Version 2009, Dalla Corte s.r.l., Baranzate, light source were placed in a dark room and fixed on a steel bar. The

Fig. 1. Schematic diagram of the specially designed equipment used for the experiments.
86 M. Münchow et al. / International Dairy Journal 51 (2015) 84e91

setup was designed so that transmitted light was recorded by the reflects the ageing of the milk/foam layer and is a function of the
video camera with a frame speed of 1 fps (frame per second) over a disproportionation and/or bubble coalescence occurring in the
period of 600 s, making a total of 600 frames for every sample. The foam layer.
amount of pixels recorded was 640  480 and the equivalent focus All features in the replicates were normalised and aligned to
area of the sample was 140  105 mm, giving a final resolution of equalise the time and the height between samples. The results can
48 mm2 per pixel. be seen in Fig. 3. In this figure, the height has been normalised from
0 to 1 in every set of curves for a better visual comparison and for
2.4. Image treatment and feature extraction further analysis. Moreover, the x-axis denotes the time in seconds
for each curve. Differences between samples were also quantita-
Each video was converted into an image using the following tively assessed from the relative distance between the single points
procedure. The vertical column of pixels at the midpoint of each extracted from the images.
frame (i.e., each second) was taken and combined with the columns Foam stability was analysed by taking the value of the upper (%
extracted from all other frames of the same video. The final result foam lost) and lower (% increase in drainage) power curves after 2
(Fig. 2A) provides an overview of how the steam-frothed milk foam and 5 min as follows:
develops over time, with the y-axis representing the height and the
x-axis representing the time (1 fps; i.e., one column of the picture % foam lost ¼ abs ðIntensity2min
was generated every second). From these images, the three main  Intensity5min Þ100=Intensity2min (1)
parts forming the sample (air, foam and milk) can easily be
distinguished. Images were then locally thresholded by using the
% increase in drainage ¼ abs ðIntensity5min
grey-scale intensity values and the corresponding histograms
shown in Fig. 2A to extract features used in further analysis  Intensity2min Þ100=Intensity5min
(Fig. 2B), namely, the upper and lower power curves, the sigmoidal, (2)
and the upper and lower linear curves. These features define the
behaviour of the foam/liquid system with time: the upper and where intensity represents the normalised height of the upper (%
lower power curves define the contact between the different in- foam lost) and lower (% increase in drainage) curves at the corre-
terfaces, foam/air and foam/milk, respectively, and how the steam sponding time (see Fig. 3 for further information). The times were
frothed milk gradually collapses. selected accordingly to the release time that baristas consider when
These curves are very similar in general shape to what can be they are serving a cappuccino: 2 min is the estimated time for
observed from a visual quantification of the volume of drained preparation and serving, and 5 min is the average time to drink the
liquid and foam as seen by Huppertz (2010) for foams from preparation. To establish these reference time points we had an
reconstituted skim milk. The time resolution is better, it is easier to observer placed at Cafe Europa 1989 on a random day to observe
separate the various phases and, in addition, more information can serving time after preparation and first interaction time after
be obtained. The upper and lower linear curves give the quantita- serving for every customer ordering cappuccino or lattes that day.
tive information of the speed of stabilisation of the system (i.e., the This is not considered to be representative for all cafes as such but
transformation from wet to dry foam, and the lower linear indicates just meant as a qualified indication for the time reference points in
the initial formation of a drained milk layer). The sigmoidal curve our calculations to provide some practical relevance.

Fig. 2. Panel A: conversion of the video recorded into a single grey-scale image; extracted features indicated. T1, start time start for pouring of steam-frothed milk into glass
container; T2, end of pouring; T3 (triple point), time of completion of initial separation into a milk layer with a distinct foam layer on top. The height of the foam is measured at the
triple point (maximal foam height). Panel B: features (i.e., specific curves) extracted from images after applying thresholding.
 M. Münchow et al. / International Dairy Journal 51 (2015) 84e91 87

Fig. 3. Specific feature curves extracted for all the samples investigated. The sigmoidal is shown in blue, lower power in red, upper power in green, lower linear in pink and upper
linear in light blue.

2.5. Correlation analysis These two sub-matrices are the scores T (I,F) and the loadings
P(J,F), where F (f ¼ 1 … F) denotes the amount of principal com-
Apart from these features, three different singular points were ponents (PCs) selected. The super-index in the loadings denotes the
extracted (Fig. 2A, bottom), T1, T2 and T3: the start of pouring of transpose of the matrix. As is represented in Eq. (3), every PC
steam-frothed milk into the glass container is denoted as T1, the contains one score and one loading vector. The scores contain all
end of pouring as T2. Bubbles then rose to the top, disproportion- the relevant information about the samples, whereas the loadings
ation and coalescence occurred and milk drained to the bottom, contain the relevant information about the variables. All the in-
transforming the initial air-filled liquid into actual foam. There was formation that is not covered in the scores and loadings matrices is
a time lag before the steam-frothed milk separated into a collected in the residual matrix, E(I,J), which contains all the in-
measurable milk layer depleted of bubbles at the bottom and a formation not related to the samples and variables, that is, the
distinct foam layer on top. The triple point, T3, indicates when this noise. The scores and the loadings help to establish relationships
initial separation was completed. The height, h, of the foam at the between samples and variables in a general manner and at a glance,
triple point was used as indicator of maximal foam height for a by only looking at the scores and loadings plots for each PC, thus
given sample. extracting valuable information to study and analyse the experi-
The creation of latte art (creating a pattern or design on the mental data obtained.
surface of the coffee drink) by baristas necessitates foam to be in
the initial state, i.e., as an unseparated air-filled liquid in the first
stage of pouring when the espresso crema, that constitutes the 2.7. Data analysis
brown background whereon the latte art is created, is elevated in
the cup. The air-filled milk needs to not float, but instead sink when The image processing and further data analysis was performed
it meets the surface so that no white traces remain at this stage of in MATLAB v. 2013b (The Mathworks, Natick, MA, USA). Analysis of
pouring. In the final stage of creating latte art the barista uses the variance (ANOVA) and principal component analysis (PCA) were
foam as a colour agent and thus it needs to be in the stage after the performed using the PLS_Toolbox implemented in MATLAB
triple point to float and leave a clear white trace for the creation of (Eigenvector Research Inc. Manson, WA, USA).
the desired pattern or design.

2.6. Principal component analysis 3. Results and discussion

Principal component analysis (PCA) was carried out on the ob- 3.1. Principal component analysis of extracted feature curves
tained curves for all the samples. The samples collected were ar-
ranged in rows (I). For each sample, five different curves were The result of the PCA analysis of the feature curves extracted is
obtained. Therefore, each curve was collected, one after the other, shown in Fig. 4. The PCA model with three principal components
forming a row vector, in such a way that there will be as many (PCs) explained more than 80% of variance. The projection of the
variables (J) as the total length of this vector. The final matrix X(I,J) scores between PC1 and PC3 (Fig. 4A) clearly separated the different
had as many rows (I) as samples and as many columns (J) as vari- samples accordingly to their content of fat. Nevertheless, due to the
ables. PCA is a technique widely reported and defined in the liter- high variability (noise) in the data, it was not possible to see any
ature (Bro & Smilde, 2014). Briefly, PCA extracts from the data the clear difference between the samples containing different amounts
uncorrelated information (also called principal components) by of protein. Therefore, we carried out individual PCA analysis for
analysing all the variables at the same time. In this way, X(I,J) is each replicate. As expected, the individual PCA models for each
decomposed into two sub-matrices as follows: replicate were highly consistent, showing the same patterns. Fig. 5
shows the results obtained for one of the replicates and the pattern
X ¼ t1 pT1 þ t2 pT2 þ … þ tf pTf ¼ TPT þ E (3) for the effect of fat content is similar to what was found for all
replicates in Fig. 4. In this case, PC1 and PC2 separate the samples
88 M. Münchow et al. / International Dairy Journal 51 (2015) 84e91

Fig. 4. Principal component (PC) analysis of the feature curves obtained from the images illustrating the time development of the steam-frothed milk foams (n ¼ 3). Panel A: PC1
versus PC3 scores plot. The labels indicate the content of fat as follows: 0.5%, inverted filled triangles (;); 0.9%, stars (✳); 1.5%, filled squares (-); 2.6%, crosses (þ); 3.5%, open
diamonds (◊). Panel B: PC1 versus PC3 loadings plot. Symbols indicate each type of curves extracted from images as follows: lower linear are denoted by inverted filled triangles
(;); lower power by stars (✳), sigmoidal by filled squares (-); upper linear by crosses (þ); upper power by open diamonds (◊). Each set of samples is shown delimited by the
corresponding convex-hull plot.

Fig. 5. Principal component (PC) analysis carried out on a single replicate. Panel A: PC1 versus PC2 scores with different protein content as follows: 3%, inverted filled triangles (;);
3.4%, crosses (þ); 3.8%, filled squares (-). Panel B: PC1 versus PC2 scores plot for samples with different fat content as follows: 0.5%, inverted filled triangles (;); 0.9%, stars (✳);
1.5%, filled squares (-); 2.6%, crosses (þ); 3.5%, open diamonds (◊). Panel C: PC versus PC2 loadings plot. Symbols indicate each type of curves extracted from images as follows:
lower linear are denoted by inverted filled triangles (;); lower power by stars (✳), sigmoidal by filled squares (-); upper linear by crosses (þ); upper power by open diamonds (◊).
Each set of samples is shown delimited by the corresponding convex-hull plot.
 M. Münchow et al. / International Dairy Journal 51 (2015) 84e91 89

according to the protein (top-left of Fig. 5) and fat (top-right of early event in the ageing of the foam and could possibly be reflected
Fig. 5) content, respectively. in a sensory difference; something that we will investigate in
The protein content is mainly differentiated because of the further studies.
different behaviour of the sigmoidal and lower linear curves with No effect was found of the fat content on the difference be-
respect to the upper linear, upper and lower power (see loading tween the singular points T3 and T2. However, as seen from Fig. 7,
plots in the bottom part of Fig. 5). Samples with low concentration a low fat content generally resulted in more stable foam (i.e.,
of protein are more difficult to stabilise when they are put into the reduced foam collapse) than higher fat content, as also indicated
vessel as reflected in changes occurring in the lower linear. The in the PCA. This is in accordance with Kamath, Huppertz,
sigmoidal curve did not shown any interesting reproducible Houlihan, and Deeth (2008a) who noted that skim milk shows a
pattern. This might be due to the inherent inhomogeneity of pronounced foam stability peak at 45
C and that foams formed by
disproportionation and bubble coalescence which can be expected skim milks are generally more stable than foams formed by whole
to induce large variations between the samples. milks. They also found the effect of milk fat on the foaming
The effect of the different fat contents is seen reflected in the properties of whole milks to be largely determined by the physical
lower power, upper power and in the sigmoidal curves (see loading state of milk fat and thus the temperature of foaming, with the
plots in the bottom part of Fig. 5). That is, samples with high fat most pronounced destabilising effect of milk fat on foam forma-
content are more susceptible to changes occurring over time in the tion and stability in the range where fat globules contain both
lower and upper power as well as in the sigmoidal curve. This may solid and liquid fat (i.e., 5e35
C). Similarly, Anderson and Brooker
indicate a lower stability of the formed foam when the fat content is (1988) found that the rate of foam collapse increased with
high, and achieving a more stable and consistent foam with a lower increased fat (0e1.5%) in milk. Borcherding, Hoffmann, Lorenzen,
amount of fat. and Schrader (2008), however, concluded that for homogenised
The information given by PCA is very helpful to describe the milk, a varying fat content (0e3.5%) and a varying fat globule size
different behaviour of samples with different amounts of protein in unpasteurised milk had no decisive effect on the foaming
and fat, as indicated above. The curves extracted from the images properties of milk when foamed at 50e60
C with a low me-
contain valuable information for comparison between samples. chanical input.
Nevertheless, this analysis is merely descriptive and qualitative. Our results also indicate, on the other hand, that lower fat
Therefore, further analysis is needed to quantitatively assess the content resulted in more drained volume (in the lower power,
influence of the content of fat and protein on the stability of the Fig. 7). In line with this, seasonal variations in raw milk (i.e., non-
formed foam. homogenised, as was also used in the present project) have also
shown improved foaming when the fat % is highest (Chen, Lewis, &
3.2. Correlation analysis Grandison, 2014). Recent work by Oetjen, Bilke-Krause, Madani,
and Willers (2014) highlights the effect of temperature on the effect
An important quantitative parameter for assessing foam sta- of fat on foams made from fresh or UHT milk samples (1.5 or 3.5%).
bility over time is the difference between the singular points T3 and At room temperature, there was a negligible effect of fat on the
T2. An increase in protein content causes a delay in the formation of bubble size distribution, but at 40
C, they were significantly
a distinct foam layer, separated from the drained milk, as indicated different, and they relate the stability of the foams to the homo-
by a longer time to reach the triple point, T3 (Fig. 6). This can be geneity of the initial bubble size distribution (better stability with
partly explained by the concomitant increase in viscosity. Using the more homogeneity).
relationship between protein concentration and viscosity for casein We thus assume, since our foams were formed at 65
C, that the
micellar solutions (Beliciu & Moraru, 2011), we calculate a relative decrease in foam drainage with increased fat content observed was
increase in viscosity of approx. 13% when increasing the protein caused by the effect of fat globules on the initial bubble size dis-
content from 3.0 to 3.8%. However, we also assume that the bubble tribution in the foams, with smaller and more homogeneous bub-
size is decreased when more protein is available to stabilise the bles being formed in the foams made from milk with a higher fat
interfacial area, hence contributing to the delay in foam layer for- content (Oetjen et al., 2014), hence decreasing drainage.
mation since small bubbles travel slower than larger bubbles in a Fig. 8 shows boxplots obtained in an ANOVA analysis to deter-
viscous liquid. Even though the magnitude of the observed delay mine if there were significant changes in the initial foam volume (as
was not large (5 s) it still represents an almost 10% delay of a very expressed by the maximum foam height, h) obtained for the

Fig. 6. The time needed for formation of a distinct foam layer (i.e., T3 e T2) as a function of protein content for three replicates processed on different days: A, date 1; B, date 2; C,
date 3.
90 M. Münchow et al. / International Dairy Journal 51 (2015) 84e91

Fig. 7. Percentage of foam lost (calculated according to Eq. (1)) and percentage increased drainage (calculated according to Eq. (2)) referring to fat (A) and protein (B) content. The
labels in A indicate the content of fat as follows: 0.5%, inverted filled triangles (;); 0.9%, stars (✳); 1.5%, filled squares (-); 2.6%, crosses (þ); 3.5%, open diamonds (◊). The labels in
B indicate the content of protein as follows: 3%, inverted filled triangles (;); 3.4%, crosses (þ); 3.8%, filled squares (-). Each set of samples is shown delimited by the corresponding
convex-hull plot.

different levels of fat and protein. The P-values obtained for the espresso machine, as used by baristas. The method is based on
replicates and protein level, at a 95% confidence level, were 0.48 simple and inexpensive equipment and can be applied directly on
and 0.18, respectively. Thus, no significant difference was observed s and other coffee serving venues, but perhaps most
site, e.g., in cafe
between replicates, indicating the methodology was sufficiently importantly, this method can be used in training centres for bar-
robust. Also, no significant influence of protein level was observed. istas to evaluate foam quality much more precisely. At present,
This is in disagreement with the findings from the PCA. Never- digging into the foam with a spoon and visually evaluating bubble
theless, it is fully explained if we consider that PCA is a method that size with the naked eye is the normal practice. The developed
aims at highlighting the differences (variance) between the sam- method provides the possibility for different foams to be compared
ples, whereas ANOVA is a method that compares the within vari- thus aiding in elucidating the effect of various hand craft tech-
ance with the between variance of the samples. In this case the niques or different equipment. In addition, manufacturers of fully
variance between the different levels of protein is not significantly automatic machines for foam creation could use the method to
different. improve equipment for creation of milk foam.
A clear significant influence of fat on foam volume was, how- The investigations of the effect of fat and protein content in milk
ever, observed (a P-value of 2.2  105 at 95% confidence level), on the properties of steam frothed milk indicate that the developed
showing a clear trend for decreased foam volume with increased method, especially when used in conjunction with PCA, is well able
level of fat. This is in accordance with the observation from baristas to discern between samples with fat contents ranging from 0.5% to
that full fat milk creates a smaller amount of foam than skimmed 3.5% and protein contents between 3.0% and 3.8% (the range of
milk. natural variation in protein content for Danish Holstein-Friesian
cows). Increased protein content caused a delay in the formation
4. Conclusions of a distinct foam layer and also slightly increased foam volume
whereas increased fat content was found to decrease foam stability
It was possible to develop a robust method for evaluation of but also resulted in less drainage. These initial results document the
foam properties in steam-frothed milk made on a conventional usefulness of the method for future investigations.

Fig. 8. Box and whisker plot obtained from ANOVA analysis for replicates (A), protein (B) and fat content (C). The top and bottom of the box are the 25th and 75th percentiles
respectively. The whiskers extend to the most extreme data points that are not considered outliers. Dashed line shows the standard deviation, filled small dots inside the box
indicate the mean and the horizontal lines indicate the medians. The open circles represent standard outliers (between 1.5  IQR and 3.0  IQR outside of the interquartile range,
IQR) and closed circles represent extreme outliers (greater than 3.0  IQR outside the IQR).
 M. Münchow et al. / International Dairy Journal 51 (2015) 84e91 91

Acknowledgements Bro, R., & Smilde, A. K. (2014). Principal component analysis. Analytical Methods, 6,
2812e2831.
Chen, B., Lewis, M. J., & Grandison, A. S. (2014). Effect of seasonal variation on the
The financial support of The Danish AgriFish Agency, Ministry of composition and properties of raw milk destined for processing in the UK. Food
Food Agriculture and Fisheries of Denmark (Grant number 3414- Chemistry, 158, 216e223.
08-02343) is gratefully acknowledged. We would like to thank Jens Deeth, H. C., & Lewis, M. J. (2015). Practical consequences of calcium addition to and
removal from milk and milk products. International Journal of Dairy Technology,
Nørgaard from Cafe  Europa 1989 for his contribution to the project 68, 1e10.
in the idea stage, for facilitating the contact with Naturmælk amba Deeth, H. C., & Smith, R. A. D. (1983). Lipolysis and other factors affecting the steam
and for being an invaluable business partner in the project. We also frothing capacity of milk. Australian Journal of Dairy Technology, 38, 14e19.
Goh, J., Kravchuk, O., & Deeth, H. C. (2009). Comparison of mechanical agitation,
like to thank Naturmælk for believing in the project and making all steam injection and air bubbling for foaming milk of different types. Milch-
the effort to make it happen in terms of funding, and for all the wissenschaft, 64, 121e124.
work done by Anne Lund Jensen who prepared all the samples in Huppertz, T. (2010). Foaming properties of milk: a review of the influence of
composition and processing. International Journal of Dairy Technology, 63,
the dairy plant and sent them to the university. 477e488.
Kamath, S., Huppertz, T., Houlihan, A. V., & Deeth, H. C. (2008a). The influence of
References temperature on the foaming of milk. International Dairy Journal, 18, 994e1002.
Kamath, S., Wulandewi, A., & Deeth, H. C. (2008b). Relationship between surface
tension, free fatty acid concentration and foaming properties of milk. Food
Anderson, M., & Brooker, B. E. (1988). Dairy foams. In E. Dickinson, & G. Stainsby
Research International, 41, 623e629.
(Eds.), Advances in food emulsions and foam (pp. 221e255). London, UK: Elsevier
Oetjen, K., Bilke-Krause, C., Madani, M., & Willers, T. (2014). Temperature effect on
Applied Science.
foamability, foam stability, and foam structure of milk. Colloids and Surfaces A,
Augustin, M. A., & Clarke, P. T. (2008). Skim milk powders with enhanced foaming
460, 280e285.
and steam-frothing properties. Dairy Science and Technology, 88, 149e161.
Silva, S., Espiga, A., Niranjan, K., Livings, S., Gumy, J.-C., et al. (2008). Formation and
Beliciu, C. M., & Moraru, C. I. (2011). The effect of protein concentration and heat
stability of milk foams. In G. M. Campbell, M. G. Scanlon, & D. L. Pyle (Eds.),
treatment temperature on micellar casein-soy protein mixtures. Food Hydro-
Bubbles in food 2: Novelty, health and luxury (pp. 153e161). Minnesota, USA:
colloids, 25, 1448e1460.
Eagan Press.
Borcherding, K., Hoffmann, W., Lorenzen, P. C., & Schrader, K. (2008). Effect of milk
Ward, B. R., Goddard, S. J., Augustin, M. A., & McKinnon, I. R. (1997). EDTA-induced
homogenization and foaming temperature on properties and microstructure of
dissociation of casein micelles and its effect on foaming properties of milk.
foams from pasteurized whole milk. Lebensmittel Wissenschaft und Technologie,
Journal of Dairy Research, 64, 495e504.
41, 2036e2043.