You are on page 1of 11

Received: 19 December 2016 | Revised: 21 May 2017 | Accepted: 13 June 2017

DOI: 10.1111/jfpe.12601

ORIGINAL ARTICLE

Designing healthy ice creams with linear programming:


An application using traditional Portuguese products

^s Lopes1 | Isabel Martins1,2


Ine | Marta Mesquita1,2 |

Vasco Valença de Sousa3 | Suzana Ferreira-Dias4

1
Universidade de Lisboa, Instituto Superior
de Agronomia, Lisboa, Portugal
Abstract
2
Centro de Matematica, Aplicaço ~es The aim of this study was to propose a user friendly computer-aided tool based on linear program-
Fundamentais e Investigaç~ ao Operacional, ming (LP) models to design ice creams. Using this methodology, healthy ice cream can be obtained
Faculdade de Cie ^ncias, Universidade de by minimizing the calorie content of the formulations subject to nutritional and technological
Lisboa, Lisboa, Portugal
requirements. New requirements, proposed by ice cream developers, are added to the LP models,
3
DuPont Nutrition Biosciences ApS,
in particular, constraints on fiber and sugar alcohol contents, fat to protein ratio, freezing point
Brabrand, Denmark (Actual address: Jelly
Belly Ice Cream, Dubai, United Arab depression factor and relative sweetness. LP models for Water Ice, Milk Ice, Ice Cream, Dairy Ice,
Emirates) Sorbet and Artisanal Ice Cream were formulated. These models were used to produce ice creams
4
Linking Landscape, Environment, including traditional Portuguese products (chick pea, tomato jam, apple jam, honey, carob, Azorean
Agriculture and Food, Instituto Superior de green, and black teas), unusual in ice cream manufacturing. An optimization solver in Excel
Agronomia, Universidade de Lisboa, Lisboa,
environment was applied to solve the models. The optimal blends were tested in an ice-cream
Portugal
pilot-plant.
Correspondence
Isabel Martins, Universidade de Lisboa,
Practical applications
Instituto Superior de Agronomia, Lisboa, The use of linear programming models in ice-cream design allows a rapid examination and evalua-
Portugal. tion of all possible blend combinations and a swift determination of the best mixes in terms of
Email: isabelinha@isa.ulisboa.pt
calorie value. These models can be easily adapted to other objective functions (e.g. cost reduction,
Funding information sensory acceptance), ingredient replacement and changes in requirement constraints. With the
FCT—Fundaç~ao para a Cie^ncia e a Solver tool from Excel, there is no need to acquire specific additional software to perform ice
Tecnologia (Portuguese Foundation for cream optimization. Moreover, this approach can be extended to other fields of food engineering
Science and Technology), Grant/Award
for product development and optimization.
Numbers: UID/MAT/04561/2013 and
UID/AGR/04129/2013

1 | INTRODUCTION sugars, presence of functional ingredients) and its origin. The presence
of natural ingredients and the use of traditional recipes and old-style
One of the challenges of the Food Industry is to create healthier manufacturing practices are sought by the consumer, who is willing to
products, to promote active and longer life. Healthy eating does not pay more for these products.
mean boring, bland food. Several studies show that a diet devoid of In the food industry, to develop a product a formula has to be
pleasure, without the so-called “comforting foods,” does not defined and a recipe has to be created. The formula specifies require-
achieve the desired results. Therefore, pleasant foods like ice cream ments on the composition of the product in terms of nutrients and
(Adam, 2005; Martinell, 2007) may be included in the diet without other characteristics that take into account technological, nutritional,
creating consciousness issues. In fact, considering its basic ingre- ingredient availability and economical features. The recipe gives the
dients, ice cream can be an interesting, healthy snack. When weight or proportion of the ingredients required to meet the formula.
included in a healthy life-style and a balanced diet, ice cream could In the context of ice creams, important formula components are
be an alternative to consuming several nutrients, such as protein, fat, milk solids non-fat, sugars and sweeteners, stabilizers, and
fiber, minerals, and vitamins. emulsifiers.
Nowadays, consumers are more concerned about the calorie value Fat increases the richness of flavors and produces a smooth tex-
of the food they eat, its composition (e.g., content and type of fats, ture by lubricating the palate. It can be either of dairy or vegetable

J Food Process Eng. 2017;e12601. wileyonlinelibrary.com/journal/jfpe V


C 2017 Wiley Periodicals, Inc. | 1 of 11
https://doi.org/10.1111/jfpe.12601
2 of 11 | LOPES ET AL.

origin. Dairy fat sources encompass milk, milk powder, condensed milk, optimize wheat blending to produce bread making flour (Hayta & Çak-
cream, and butter. Vegetable fats are provided by coconut oil, palm makli, 2001); (2) to formulate low cost composite flour mixtures, in
kernel oil and refined palm oil, among others. The fat content provides response to the Recommended Dietary Allowance of the protein,
the largest contribution to the total amount of calories in the product. energy and calcium requirement profile (Balasubramanian, Singh, Patil,
However, there exist legal requirements for each type of ice cream & Onkar, 2012); (3) to develop novel ready-to-eat therapeutic food for-
concerning the fat content, as will be explained later. mulations for Ethiopia, using locally available ingredients, and minimiz-
Milk solids non fat (MSNF) include proteins, such as caseins and ing the ingredients’ cost (Ryan et al., 2014); (4) to identify optimal food
whey proteins, lactose and minerals. The optimum level for MSNF is intake patterns for Japanese populations (Okubo et al., 2015); (5) to
determined by the emulsifying and water binding effect of proteins, formulate nutritious, minimum-cost porridge mix for children aged one
besides lactose crystalization. According to Danisco (Technical Memo- to two, based primarily on local ingredients, in rural Mozambique (De
randum, TM 2510-2e), the optimal level is considered to be 17 parts of Carvalho, Granfeldt, Dejmek, & Håkansson, 2015); and (6) to formulate
MSNF per 100 parts of water and the optimum level for lactose should vegetable pastes maximizing the antioxidant activity (Larrosa, Cadaval,
not exceed 10% water. & Pinto, 2015).
Sugars and sweeteners contribute to a depressed freezing point, Mix calculations in ice creams involve technological requirements
thus allowing it to contain some unfrozen water at very low tempera- of importance to flavor, body, texture, cooling effect, viscosity, whip-
tures, typical of their serving temperatures (215 8C to 218 8C). With- ping ability, and freezing characteristics of the blend. Commonly, these
out this unfrozen water, the ice cream would be too hard to scoop. At requirements encompass specifications as to the contents of milkfat,
218 8C, according to ice cream producers, freezing point depression milk solids non-fat, lactose, and water. When LP is not applied, each

factor (FPDF) values between 10 and 20 give a hard consistency, while requirement is expressed by a single value (see e.g., Kilara & Chandan,

values between 20 and 25 provide normal consistencies and values in 2007). In this case, the sweeteners content of the mix is determined by

the range of 25–40 are characteristic of soft scoopable ice creams. the levels of the FPDF (the sum of the depressive effect on freezing

Sweeteners can be classified as nutritive or non-nutritive. The cat- point of water by the ingredients) and relative sweetness (the sum of

egory of nutritive sweeteners (4 calories per gram) includes sucrose, the sweetness of the ingredients expressed relative to sucrose). Alter-

lactose, dextrose, fructose, corn syrup solids and sugar alcohols. Sugar natively, with LP, each requirement can be expressed by a range of val-
ues. IBM-Data Processing Application (E20-0156-0) proposed a LP
alcohols (e.g., lactitol, sorbitol, and xylitol) are a class of polyols that
model to design minimum cost ice creams which includes lower and
occur in a wide variety of fruits and vegetables.
upper bounds on the contents of fat, milk solids non-fat, carbohy-
Non sugar- added ice creams are mixes relying on the addition of
drates, water and an upper bound on the content of corn syrup solids.
intense non-nutritive sugars, called artificial sugars, which provide no
A software that provides least cost formulations with similar con-
significant calories at utilization levels. They include sucralose, aspar-
straints is available, for sale, at TechWizard ® (Owl Software, Colum-
tame, and acesulfame-K. Non sugar-added ice creams can be made by
bia, MO, www.owlsoft.com). Alternatively, Microsoft Excel
replacing sucrose with a bulking agent such as Litesse® and sugar alco-
spreadsheets to calculate ice cream blends can be purchased at Dairy
hols combined with non-nutritive sweeteners.
Science - Food Technology website (https://www.dairyscience.info/
Stabilizers are a group of compounds, usually polysaccharides,
index.php/technology/180-excel-ice-cream.html). This computer-aided
which are responsible for adding viscosity to the unfrozen portion of
tool performs mix calculations without considering an optimization
water, thus retaining this water so that it cannot migrate within the
criterion.
product. The result is an ice cream that is firmer to the chew.
The aim of this work was to use LP models to design low caloric
The emulsifiers are the group of compounds that assist the devel-
ice creams. Due to the formula flexibility allowed by LP, new require-
opment of an appropriate fat structure and air distribution, necessary
ments can be considered. Taking advantage of this feature, we consid-
for smooth eating and good meltdown characteristics. In general, stabil-
ered, besides the usual requirements, a set of constraints that, to the
izers and emulsifiers account for a maximum of 1% of the ice cream.
best of our knowledge, have not been addressed by LP models for ice
The ratio emulsifier/stabilizer in commercial versions of functional sys-
creams.
tems from DuPont Danisco (e.g., CREMODAN® Naturatex 20, CRE-
The new constraints translate:
MODAN® 250 SORBETLINE) is pre-defined. As we used these
commercial versions, we did not have freedom to manipulate this ratio. 1. Technological requirements related to the FPDF and relative
Linear programming (LP) models have been widely used to perform sweetness (Rel S), and fat to protein ratio.
the calculations that convert a formula into a recipe, the so-called mix For the FPDF and Rel S, lower and upper bounds were defined
calculations. A LP model may be seen as a mathematical representation based on the type of ice cream and consumer preferences Danisco
of all factors relevant to the product manufacture, providing the best (Technical Memorandum, TM 2510-2e). The lower is the FPDF,
solution according to a given optimization criterion. For a review on LP the softer and less icy is the product experience. The lower is the
see for example, Winston and Goldberg (2004). Rel S, usually the lower is the perception of sweetness of the ice
Examples of applications of LP in mix calculations have been pro- cream. These constraints help to obtain recipes with adequate lev-
posed by several authors in different contexts. In particular: (1) to els of carbohydrate content.
LOPES ET AL. | 3 of 11

For the fat to protein ratio, lower and upper bounds were set to upper bound of 10% on the carbohydrate content was set. To obtain a
establish a balance between these ingredients for a proper stabili- high fiber content Water Ice, a minimum of 7% fiber was chosen. Two
zation of the air cells of the ice cream. The higher is the ratio nor- types of Water Ice, Green tea Water Ice and Black tea Water Ice,
mally the better is the stability of the air cells and the stability of based on green and black teas from Azores, respectively, were pro-
the product to meltdown. Although constraints on ratio fat to pro- posed. With the first type, we want to highlight green tea and rasp-
tein are added to address technological concerns, they have also berry concentrate flavors, while with the second type we want to
effects on the nutritional quality of the mix, as they prevent a dis- combine black tea and apple concentrate with cinnamon flavors. To
proportionate rate between these compounds in favor of fat. achieve these goals, lower bounds of 1’% were set for the amounts of
2. Nutritional requirements concerning fiber and sugar alcohol con- raspberry and apple concentrates. The emulsifier/stabilizer used was
tents. CREMODAN® DL.
A lower bound on the content of fiber was set to strengthen the Milk Ice contains at least 2.5% exclusively dairy fat and 15%
healthy image of the product. According to the European Union, MSNF, and excludes any fat and/or protein from other than dairy ori-
concerning nutrition and health claims made on foods (Regulation gin. Hoping to provide a Milk ice as a high fiber content product, we
(EC) No 1924/2006), a product is considered a source of fiber if it set a lower bound on the fiber content of 6%. Strawberry and blue-
contains at least 3% fiber. In particular, a high fiber content prod- berry concentrates were added to create a high fiber content ice with
uct contains at least 6% fiber. a fruity taste of berries. Lower bounds of 10% and 1% were set for the
Sugar alcohols are widely used in food industry as reduced-calorie amounts of strawberry and blueberry concentrates (high fiber prod-
sweeteners (Kilara & Chandan, 2007). To avoid its excessive con- ucts), respectively. The emulsifier/stabilizer used was CREMODAN®
sumption, which may have a laxative effect, an upper bound on 809 CREAMLINE.
the amount of sugar alcohols was added to the LP model. Ice Cream is an emulsion typically made of water and edible fats,
Although constraints on fiber and sugar alcohols were added to proteins and sugars. The objective was to develop an Ice Cream using
address nutritional concerns, they have effects on the FPDF of a vegetable fat combined with a natural sweetener, honey (rich in bio-
the ice creams and thus in the texture of the mix. The higher are active properties and flavor). The fats considered include cocunut oil,
fiber and sugar alcohols contents, the lower is the FPDF. peanut oil and sunflower oil. Hoping to provide an Ice Cream as a good
source of fiber, we set a lower bound on the fiber content of 4.5%.
Moreover, using linear programming methods, a sensitivity analysis
The emulsifier/stabilizer used was CREMODAN® 1501 IcePro.
was performed to show how some changes in the input parameters
Dairy Ice contains at least 5% dairy fat and also excludes fat and/
(contribution of each ingredient to the calorie content of the mix as
or protein of non-dairy origin. The challenge was to make a healthier
well as lower and upper bounds on the requirements) will affect the
version of chocolate Dairy Ice Cream containing carob. Lower bounds
optimal solutions.
of 2% and 3% were set for the amounts of carob and cocoa flours,
The LP models were formulated to obtain recipes of six ice cream
respectively. Carob is a very common product in Portugal, especially in
types, Water Ice, Milk Ice, Ice cream, Dairy Ice, Sorbet and Artisanal Ice
the southern part of the country. The emulsifier/stabilizer used was
Cream, including traditional Portuguese products, uncommon in ice
CREMODAN® 1501 IcePro.
cream manufacturing.
The Sorbet follows closely the same definition of Water Ice, which
The models were built in a free form layout within Excel spread-
does not include fat, but contains at least 25% fruit. This percentage
sheets and solved by the add-in tool Solver. The impact of the new
may however be reduced to 15% if the fruit has a strong acid flavor,
technological and nutritional constraints in the resulting ice creams is
10% in the case of vegetables with a strong flavor and thick consistency
analyzed. A brief comparison with commercial ice creams concerning
or to 7%, should it be nuts. A homemade tomato jam (tomato, sucrose,
nutritional characteristics is also presented.
and cinnamon), raspberry and lemon concentrates were included in the
mix. Lower bounds of 20%, 1%, and 0.5% were set for the amounts of
2 | MATERIALS AND METHODS tomato jam, raspberry and lemon concentrates, respectively. Tomato
jam is one of the most traditional widely appreciated jams in Portugal.
2.1 | Ice cream types
The emulsifier/stabilizer used was CREMODAN® 250 SORBETLINE.
The developed LP models were tested to formulate the following types The Artisanal Ice Cream is usually made by artisan producers using
of ice creams: Water Ice, Milk Ice, Ice Cream, Dairy Ice, Sorbet, and the recipe of a traditional country’s dessert. To test the LP model, the
Artisanal Ice Cream. For each type of ice cream a set of available ingre- dessert chosen was chick pea paste (chick pea, sugar, cinnamon, and
dients is defined. At this point, some considerations about the ingre- lemon). A lower bound of 10% was set for the amount of this ingredi-
dients and requirements for each ice cream type are presented. ent. The emulsifier/stabilizer used was CREMODAN® Naturatex 20.
Water Ice consists mainly of water and sugars. Although this type Based on practical knowledge of ice cream producers, we set the
of ice cream is the category that contains the lowest amount of calo- fat to protein ratio within a range of 2–3 for Milk Ice, Ice Cream, Dairy
ries, as a rule its composition has a lot of sugars and is of low nutri- Ice, and Artisanal Ice Cream. The total amount of sugar alcohols should
tional value. With the expectation of giving Water Ice a new image, an be less than 9% for all types of ice creams of which sugar alcohols are
4 of 11 | LOPES ET AL.

TA BL E 1 Requirements for the different types of ice cream

Type of ice m
Water ice Milk Ice Ice Cream Dairy Ice Sorbet Artisanal Ice Cream

Requirement r Lrm Urm Lrm Urm Lrm Urm Lrm Urm Lrm Urm Lrm Urm

Fat 1 2.5 5 2 5 5 6 3 5 7

Carbohydrates 8 10 15 15 15 15 15

Lactose 10% water 10% water 10% water 10% water

MSNF 15% water 17% water 14% water 17% water 15% water 17% water 15% water 17% water

FPDF 15 30 15 30 15 30 15 30 15 30 15 30

Rel S 14 15 14 17 14 17 14 17 17 19 14 17

Fiber 7 6 4.5 3 2 3

Fat/Protein 2 3 2 3 2 3 2 3

Sugar alcohols 9 9 9 9 9 9

Water 80 85 65 70 65 70 60 65 60 70 60 65

Lrm 5 lower bound; Urm 5 upper bound.


The values for fat, carbohydrates, fiber, and sugar alcohols are given as a % (w/w) of the mix.

ingredients. The maximum lactose content should correspond to 10% amount of calories given by 1g of ingredient i 2 I. Given Z to be the
of the total water weight of the mix. A lower bound of 15 and an upper total amount of calories in the ice cream, the objective function can
bound of 17 were set for the level of MSNF, except for Ice Cream, for be written as
which the lower bound was set to 14. To produce an ice cream with a X
min Z5 ai Xi : (1)
good consistency, not too soft, with good melt down properties, FPDF i2I

was bounded between 15 and 30. Although one goal is to reduce the
amount of sugars, one must ensure that the product is not tasteless, so
2.2.2 | Constraints
the lower bound on Rel S was set to 14 and 17 and the upper bound
The set of constraints that characterize each type of ice cream are for-
to 15, 17, and 19, depending on the ice cream type.
mulated below.
Table 1 summarizes the requirements for all types of ice creams
To easily access the total amount of lactose, MSNF, water, fat and
except those related to the amount of emulsifier/stabilizer. Let M denote
protein in the ice cream mix (in g or % of the mix), a set of auxiliary var-
the set of ice cream types and R the set of requirements. For each type
iables were defined, Ylactose , YMSNF , Ywater , Yfat ; and Yprotein , by the fol-
of ice cream m 2 M, the first and second columns show, respectively, the
lowing constraints:
lower and upper bounds on each requirement r 2 R; Lrm and Urm .
X
bi Xi 2 Ylactose 50; 8i 2 I (2)
i2I
2.2 | Designing ice creams with linear programming X
ci Xi 2 YMSNF 50; 8i 2 I (3)
models i2I
X
According to the set of requirements in Table 1, we propose a LP di Xi 2 Ywater 50; 8i 2 I (4)
model for each ice cream. To formulate the LP models, a set of decision i2I
X
variables has to be defined. Let I denote the set of all available ingre- ei Xi 2Yfat 50; 8i 2 I (5)
dients. Decision variable Xi represents the quantity of ingredient i 2 I i2I
X
to be included in the mix (in g). For the sake of simplicity, we consider fi Xi 2 Yprotein 50; 8i 2 I: (6)
i2I
that the amount of mix to be made is 100 g. Let Sm be the subset of
ingredients i 2 I that can be used in the production of ice cream of Parameters bi , ci , di , ei ; and fi are, respectively, the amount of lac-
type m 2 M: For each m 2 M; variables Xi are defined for all ingredients tose, MSNF, water, fat and protein, in 1g of ingredient i 2 I (in g). Note
in I. The variables corresponding to the ingredients not in Sm are set to that Ywater is the sum of the amount of water provided by the ingre-
zero in the LP model. dients other than the water ingredient itself.
Fat constraints
2.2.1 | Objective function
Yfat  Urm ; r5Fat; m 2 M (7)
In this study, the objective function expresses the goal of minimiz-
Yfat  Lrm ; r5Fat;
ing the total amount of calories of the mix. Let ai denote the
LOPES ET AL. | 5 of 11

m 2 M n fWater Ice; Sorbetg: (8) variables corresponding to these ingredients were set to zero. Taking
into account the quantities of the ingredients provided by the resulting
Carbohydrate constraints
X optimal solution, an amount vm of the emulsifier/stabilizer, adequate
gi Xi  Urm ; r5Carbohydrate; m5Water Ice (9) for type m ice cream, was chosen by the producers. Then a constraint
i2I
X setting this requirement was added to each ice cream type model
gi Xi  Lrm ; r5Carbohydrate; m 2 M: (10)
i2I Xemulsifier=stabilizer 5vm ; m 2 M (25)

where gi is the amount of carbohydrates in 1 g of ingredient i 2 I (in g). and the complete model is re-solved.
Lactose constraint Non-selected ingredients constraint
X
Ylactose 20:1 Ywater  0; m 2 M n fWater Ice; Sorbetg: (11) Xi 50; m 2 M (26)
2Sm
i=
MSNF constraint
This constraint guarantees that the ingredients not in Sm are not
YMSNF 20:17Ywater  0; m 2 M n fWater Ice; Sorbetg (12)
selected.
YMSNF 20:15Ywater  0; m 2 M n fWater Ice; Sorbet; Ice Creamg (13) Amount of mix constraint
X
YMSNF 20:14Ywater  0; m5Ice Cream: (14) Xi 5100 (27)
i2I
FPDF constraint
X This constraint ensures that the sum of weights of all the
hi Xi  Urm ; r5FPDF; m 2 M (15)
i2I
ingredients in the mix is equal to 100 g.
X Non negativity constraints
hi Xi  Lrm ; r5FPDF; m 2 M (16)
i2I Ylactose  0; YMSNF  0; Ywater  0; Yfat  0; Yprotein  0; Xi  0; i 2 I
where hi is the FPDF in 1 g of ingredient i 2 I. (28)

Rel S constraint
X 3 | RESULTS AND DISCUSSION
pi Xi  Urm ; r5Rel S; m 2 M: (17)
i2I
X The package SOLVER add-in from Excel was used to solve the LP mod-
pi Xi  Lrm ; r5Rel S; m 2 M (18)
i2I els. For each ice cream type, the optimal solution obtained is described
in Table 2. Table 3 displays the composition of each optimal solution
where pi is the Rel S in 1g of ingredient i 2 I.
regarding nutritional and technological profiles.
Fiber constraint
X The optimal blends were tested in an ice-cream pilot-plant at Dupont.
qi Xi  Lrm ; r5Fiber; m 2 M (19) The use of an uncommonly used liquid vegetable fat, namely sunflower
i2I
oil, proved to be a great challenge, with regard to freezing and meltdown
where qi is the amount of fiber in 1g of ingredient i 2 I (in g). properties of the mix. To preserve the heat sensitive enzymes of the
Fat/protein constraint honey, part of this ingredient was added after the heat treatments.
Yfat 22 Yprotein  0; m 2 M n fWater Ice; Sorbetg (20)
3.1 | Sensitivity analysis in linear programming
Yfat 23 Yprotein  0; m 2 M n fWater Ice; Sorbetg: (21)
The simplest form of sensitivity analysis studies how the optimal solu-
Notice that these constraints are equivalent to Yfat
Yprotein  2 and
tion responds to a change in a single parameter, while all the others are
Yfat
 3 ðfor Yprotein 6¼ 0Þ, respectively.
Yprotein unchanged.
Sugar alcohols constraint
By choosing Sensitivity in the Solver Results dialog box, the Solver
X
ui Xi  Urm ; r5Sugar alcohols; m 2 M (22) add-in gives a sensitivity report. This report displays information about
i2I
three aspects: forcing each variable which is currently zero to assume a
where ui is the amount of sugar alcohol in 1g of ingredient i 2 I (in g). non-zero value; changing each objective function coefficient; changing
the right-hand side of each constraint. This information is partly pre-
2.2.3 | Water constraint sented in Table 4 for Ice Cream.

Ywater  Urm ; r5Water; m 2 M: (23) Referring to Table 4-Variable Cells, column “Final Value” lists
the optimal values for the variables, that is, the optimal ingredient
Ywater  Lrm ; r5Water; m 2 M: (24)
quantities. The column labeled “Reduced Cost” is of particular inter-
est in the context of ice cream blends. The reduced cost is relevant
2.2.4 | Emulsifier/stabilizer constraint for each non-basic variable (see e.g., Winston & Goldberg, 2004),
In a first step, the model for each ice cream type was solved with no which has zero value in the optimal solution obtained, since it pro-
requirements for the amount of emulsifier/stabilizers. That is, the vides a measure of how much the objective function would change
6 of 11 | LOPES ET AL.

TA BL E 2 Optimal solution for each ice cream type (amounts of ingredients in % or g/100 g)

Green/Black Artisanal
Type of ingredient Ingredient i ‰ I tea water ices milk ice Ice cream Dairy ice Sorbet ice cream

Fats Butter 0 0 0
Coconut oil 0 0
Cream 10.26 12.06 16.31
Peanut oil 0 0
Sunflower oil 0 2.81

Milk solids Whey powder 0 0 0 0


Whole milk powder 0 0 0 0
Skimmed milk 30.98 17.68 29.10 22.51
Skimmed milk powder 0 0 0
Skimmed milk replacer 7.02 8.32 6.64 7.07

Soy protein SUPRO® 0 0 0 0

Sugars Dextrose 0 0 0 0 0
Fructose 7.83 2.52
Glycerol 0 0 0 0 0
Glucose 0 0 0 1.95 0
Honey 0 0 17.97 0 0
Litesse® Two 8.64 7.28 5.39 3.25 2 2.38
Sucrose 0 0 7.11 1.64

Sugar alcohols Lactitol 0 0 9 0


Sorbitol 0 0
Xylitol 0 3.81 6.27 0 6.95

Fruits Apple concentrate 1 Black


Blackberry concentrate 1
Lemon concentrate 0 6.12
Orange 0
Mango 0
Pomegranate 0
Raspberry concentrate 1 White 1
Strawberry concentrate 0 10
Tomato jam 20

Other flavors Chickpea paste 10


Cocoa powder 3
Carob flour 2
Egg yolk 0
Tea 82.37

Emulsifier/stabilizer 0.15 0.85 1 1 0.6 1.5

Water 26.28 46.83 29.56 59.27 31.45

(a penalty amount) if one unit of this variable were forced into the in the amount of energy then they have a different composition and,
solution. For example, if we want to include in the optimal Ice therefore, other parameters of the model change as a result of the
Cream mix 1g of skimmed milk powder the current objective func- replacement. In this case, it is not correct to use the sensitivity analysis
tion value (138.37 kcal) would increase by 0.05208 kcal. Reduced on the objective coefficients provided by the Solver.
costs are valid only over a range of values for the variable in ques- Moreover, as for ice cream formulation, it is often important to
tion, but this range is not displayed. evaluate how a change in a constraint’s right-hand side (the constant
Columns “Allowable Increase” and “Allowable Decrease” give the that makes up the right-hand side of the constraint) will affect the opti-
amount by which each objective function coefficient (in column mal objective value. The shadow price associated with a constraint rep-
“Objective Coefficient”) can be increased or decreased, respectively, resents the change in the objective function value per unit increase in
without changing the optimal activity levels. For example, the coeffi- the right-hand side value of that constraint, assuming that the current
cient of 8.843 kcal per gram of coconut oil can decrease by 0.01766 basis remains optimal (see e.g., Winston & Goldberg, 2004). For some
and increase up to 11 without changing the optimal recipe obtained. constraints of the Ice Cream model this value is shown in column
As each coefficient of the objective function is the amount of energy “Shadow Price” of Table 4-Constraints. The range of feasibility is pro-
per gram of the corresponding ingredient, its change occurs when the vided by columns “Allowable Increase” and “Allowable Decrease.” For
ingredient is replaced by another ingredient. If both ingredients differ example, if we decrease the right-hand side of constraint (11) (the
LOPES ET AL. | 7 of 11

TA BL E 3 Characterization of the optimal solution for each ice cream type regarding nutritional and technological profiles

Ice cream types


Nutrients & others Green/Black tea water ices Milk ice Ice cream Dairy ice Sorbet Artisanal ice cream Units

Water 83.01 65.97 66.63 63.12 69.87 63.47 g

Fat 0.08 4.84 3.75 5.75 0.29 6.74 g


Saturated Fat 0.08 3.14 0.88 3.64 0.25 4.18 g

Protein 0.11 2.41 1.87 2.94 0.20 2.98 g

Carbohydrate 8.83 15.06 21.60 15.75 15.00 15.13 g

Lactose 0.00 6.60 6.66 6.31 0.00 6.35 g

Fiber 7.00 6.00 4.50 3.26 2.00 3.00 g

Minerals 29.62 182.09 90.85 367.29 26.77 139.33 mg


Sodium 0.41 16.74 9.81 17.05 1.14 13.93 mg
Potassium 24.15 71.00 37.10 217.91 12.60 51.27 mg
Calcium 1.05 48.28 22.82 54.67 3.94 38.64 mg
Magnesium 1.83 6.88 2.68 19.56 3.94 3.90 mg
Phosphorus 2.17 39.19 18.44 58.09 5.15 31.57 mg

Energy 52.93 131.26 138.37 148.78 91.16 149.83 kcal

MSNF 0 9.9 9.33 9.47 0 9.52 g

FPDF 20.76/20.78 30 28.45 30 27.84 30

RelS 14 15 15 15 19 15

The values in each column represent the composition of 100 g of the respective mix in terms of nutrients, energy and the correspondent values of
FPDF and relative sweetness (Rel S)

lower bound on the amount of fiber) from 4.5 g to 3.5 g, then the All the produced ice creams were positively evaluated by DuPont
objective function value will decrease by 1.56151 kcal. Or, if we experts and a brief description was made for each type, as follows.
increase from 4.5 g to 5.5 g, the objective function value will increase
by 1.56151 kcal. This shadow price is valid for increases from 4.5 to Water Ice
4.5 plus 1.59448 (6.09448 g) and decreases from 4.5 to 4.5 minus Refreshing water ices with tasteful notes from the green and black
3.30477 (1.19523 g). teas perfectly combined with raspberry and apple/cinnamon touches. The
The Column “Final Value” lists the optimal values for the left-hand black water ice was selected as a monthly concept product for DuPont.
sides of the constraints, whose right-hand sides are displayed in column Possible claims: Reduced sugar; High fiber; No fat; Calorie reduced.
“Constraint R.H. Side.” If both values are equal, the corresponding con- The high fiber content comes from both raspberry and apple.

straint is binding, that is, it is satisfied in the equality by the optimal solu- Also, the ability to reduce the blood sugar by cinnamon and the anti-

tion (e.g., constraints MSNF LB, Rel S LB, Fiber LB, Lactose UB, and oxidant properties of raspberry are important properties that may

Total weight in Table 4-Constraints). The shadow price of a non-binding be considered.

constraint is equal to zero (see e.g., Winston & Goldberg, 2004).


Milk Ice
Regarding Table 4-Variable Cells, we may see that, for Ice Cream,
Smooth berry breeze, with a light fruity taste.
the impact on the optimal energy content is greater with a milk solid
Possible claims: High fiber; Low fat.
ingredient than with a vegetable fat and is substantially greater with
the ingredient SUPRO®. With respect to Table 4-Constraints, increas-
Ice Cream
ing the lower bound on Rel S has a greater impact on the optimal
Smooth, soft and delicate ice cream, flavored with Portuguese
energy content than increasing the lower bound on the fiber content.
honey used as natural sweetener.
Increasing the amount of the emulsifier/stabilizer reduces the mix calo-
Possible claims: Highly polyunsaturated fat.
rie content.
From the nutritional point of view, it is a very interesting product
due to the richness in polyunsaturated fatty acids from sunflower oil and
also to the use of honey, with well-known nutrition and medical benefits.
3.2 | Sensory analysis and nutritional claims
The optimal blends obtained by LP for each ice cream type were sen- Dairy Ice
Luxury, silky chocolate ice cream.
sory evaluated by a group of five expert sensory assessors of DuPont.
8 of 11 | LOPES ET AL.

TA BL E 4 Excel sensitivity report for ice cream

Objective Allowable
Name Final value Reduced cost coefficient Allowable increase decrease

Variable Cells Coconut oil 0 0.017664597 8.843 1E130 0.017664597


Peanut oil 0 0.017664597 8.843 1E130 0.017664597
Sunflower oil 2.812986366 0 8.843 0.01769383 3.071625235
Whey powder 0 0.223003884 3.64 1E130 0.223003884
Whole milk powder 0 0.308990199 5.0072 1E130 0.308990199
Skim milk 17.67922535 0 0.359 0.005017877 0.131012523
Skim milk powder 0 0.052082743 3.65 1E130 0.052082743
Skim milk replacer 8.315993813 0 3.58 0.188322002 1E130
SUPRO® 0 16.94774576 3.8 1E130 16.94774576
Honey 17.97232712 0 3.344 15.08016839 2.653658136
Litesse® Two 5.3875 0 2 2894.259404 1.249210258
Water ingredient 46.83196734 0 0 0.145498796 0.00555962
CREMODAN® 1501 IcePro 1 0 6.5 1E130 1E130
Fat 3.752343919 0 0 8.086238092 8.46062677
Protein 1.876171959 0 0 16.17247618 17.70216355
Lactose 6.662726796 0 0 13.92929738 10.66218809
MSFN 9.327817515 0 0 10.92891367 7.615848639
Water total 66.62726796 0 0 1.530047914 1.066218809

Constraint Allowable
Name Final Value Shadow Price R.H. Side Allowable Increase Decrease

Constraints Fat UB 3.752343919 0 5 1E130 1.247656081


MSNF LB 21.77636 E –15 17.66459693 0 0.626739072 0.482848606
Carbohydrate LB 21.61553504 0 15 6.615535036 1E130
FPDF UB 28.44915387 0 30 1E130 1.550846128
FPDF LB 28.44915387 0 15 13.44915387 1E130
Rel S LB 15 3.556184775 15 1.204121074 3.675549707
Rel S UB 15 0 17 1E130 2
MSNF UB 21.998818039 0 0 1E130 1.998818039
Fiber LB 4.5 1.561512823 4.5 1.594478068 3.30477057
Lactose UB 23.55271 E –15 216.90970923 0 0.360750093 0.653063033
Sugar alcohols UB 0 0 9 1E130 9
Fat/Prot LB 21.876171959 0 0 1E130 1.876171959
Water UB 66.62726796 0 70 1E130 3.372732039
Total weight 100 0.782072648 100 3.965724684 1.913373682
Fat/Prot UB 4.44089 E –16 8.043262755 0 1.312396817 1.317327906
Non available ingredients 0 22134.477296 0 0.006135204 0
Water LB 66.62726796 0 65 1.627267961 1E130
Fat LB 3.752343919 0 2.5 1.252343919 1E130
CREMODAN® 1501 IcePro 1 20.209044013 1 4.006698172 1
Lactose Total 0 216.90970923 0 0.360750093 0.653063033
MSNF Total 27.10543 E –15 17.66459693 0 0.626739072 0.482848606
Water Total 22.84217 E 214 20.782072648 0 1.913373682 3.965724684
Fat Total 5.77316 E –15 8.043262755 0 1.919232914 2.95010135
Protein Total 24.44089 E –16 216.08652551 0 0.658663953 0.656198409

Possible claims: High fiber. with this recipe is the texture, which was not completely homogeneous
Carob, used as ingredient, is known as the “healthy chocolate,” and since the chick pea paste was not completely homogeneous, and thus
has considerable less fat than chocolate. it is possible to feel some chunks.
Possible claims: Natural; source of fiber.
Sorbet
Homemade tomato jam converted into a tasty and refreshing
3.3 | Impact of the new technological and nutritional
sorbet. constraints
Possible claims: Low fat; Calorie reduced. At this point, we evaluate the impact of the new technological and
This sorbet was chosen for monthly concept for DuPont. nutritional constraints in the mix. We removed from the models the
following sets of constraints:
Artisanal Ice Cream
Winning recipe, for those who appreciate the flavors of Portu- 1. one new constraint at a time;
guese traditional desserts, based on chick pea. A possible drawback 2. all the new technological constraints;
LOPES ET AL. | 9 of 11

3. all the new nutritional constraints; respectively, for Dairy Ice and Water Ices. The content of sugar

4. all the new constraints. alcohols only increased for Sorbet.

For each case, the resulting sub-models were solved. Table 5 displays In general, the new constraints seem to be effective for obtaining ice
the values obtained for the technological characteristics or nutrients cream mixes well balanced regarding technological and nutritional
addressed by the removed constraints. The values in each column rep- characteristics.
resent the composition of 100 g of the optimal mix obtained in terms
of fat to protein ratio, Rel S, FPDF, fiber, and sugar alcohols. Whenever
a value is smaller (greater) than the lower bound (upper bound) of the
3.4 | Comparison with commercial ice creams
corresponding constraint, then the relative deviation from this lower In this section, we compare each optimal recipe, given by model (1)–
bound (upper bound) is enclosed in brackets. Concerning the relative (26), with a commercial ice cream of the same type, whose nutritional
deviations, the following results were obtained: characterization is available to consumers on the packaging. For all
types, we selected ice creams of the same type which are market lead-
1. The usual fat and carbohydrate constraints were unable to ensure a
ers in Portugal. We shall refer to the selected ice creams as C-Water
suitable balance between fat and protein and an adequate level of Rel
Ice, C-Milk Ice, C-Ice Cream, C-Dairy Ice, C-Sorbet, and C-Artisanal Ice
S, respectively. For most ice cream LP models, the fiber constraints
Cream. The C-Artisanal Ice Cream is based on a traditional product
are mandatory to guarantee the adequate level of this nutrient. For
from the north of Portugal, the cream of sweet eggs of Aveiro. Table 6
some ice cream types, requirements on the FPDF and sugar alcohols
characterizes these commercial ice creams regarding nutritional pro-
content are achieved only if they are considered in the LP models.
files. Figure 1 graphically compares the optimal recipes with the com-
2. For the ice cream models with constraints on fat to protein ratio,
mercial ice creams with respect to energy (a), saturated fat (b), protein
this ratio decreased (below the lower bound). The FPDP decreased
(c), and carbohydrate (d). Due to lack of information, Water Ice and C-
for Water Ices and Dairy Ice. The Rel S increased for Sorbet (above
Water Ice, as well as Sorbet and C-Sorbet were not compared in terms
the upper bound) and decreased for the other ice creams.
of saturated fat. For the same reason, Water Ice and C-Water Ice were
3. The amount of fiber decreased. Only for Sorbet, the amount of not compared in terms of protein.
sugar alcohols exceeded the upper bound on this ingredient. One can see that the proposed optimal recipes are less caloric
4. The impact in the technological characteristics was similar to (2). than the corresponding commercial ice creams, except for Milk Ice,
Concerning the content of fiber, large and slight decreases occur, which is very similar to C-Milk Ice.

TA BL E 5 Impact of the new technological and nutritional constraints

Nutrients & Green/Black Artisanal


others water ice Milk ice Ice cream Dairy ice Sorbet ice cream Units

(a) Fat/Protein – 1 (–51%) 1.3 (–33%) 1.4 (–31%) – 1.9 (–8%)

FPDF 20.8 30.1 21.8 33.5 (112%) 28.9 17.8

Rel S 7.3 (–48%) 7.6 (–46%) 8.5 (–39%) 3.4 (–76%) 21 (111%) 6.8 (–52%)

Fiber 2.7 (–62%) 1.9 (–68%) 1.2 (–73%) 7.6 1.9 (–6%) 1.1 (–63%) g

Sugar alcohols 0.0 3.8 0.0 6.3 13.4 (148%) 1.1 g

(b) Fat/Protein – 0.8 (–61%) 1.3 (–37%) 1.0 (–48%) – 1.2 (–40%)

FPDF 12. 9 (–14%) 20.0 22.7 14.2 (–6%) 28.9 18.5

Rel S 7.5 (–47%) 7.6 (–53%) 8.5 (–47%) 3.2 (–85%) 21.0 (11%) 6.8 (–51%)

(c) Fiber 2.7 (–62%) 1.9 (–68%) 1.2 (–62%) 1.4 (–53%) 1.7 (–16%) 1.1 (–63%) g

Sugar alcohols 3.2 6.9 0.0 7.2 14.0 (156%) 8.3 g

(d) Fat/Protein – 0.8 (–61%) 1.3 (–37%) 0.9 (–56%) – 1.2 (–40%)

FPDF 12.5 (–17%) 20.0 22.7 11.8 (–21%) 27.1 18.5

Rel S 7.5 (–47%) 7.6 (–53%) 8.5 (–47%) 3.0 (–85%) 16.1 (–6%) 6.8 (–51%)

Fiber 6.4 (–8%) 7 8.3 0.5 (–83%) 2.1 8.5 g

Sugar alcohols 0.0 2.6 0.0 4.6 13.5 (150%) 1.8 g

The subsets of constraints are removed from the LP models: (a) one new constraint at a time; (b) constraints on fat to protein ratio, FPDP and Rel S;
(c) constraints on fiber and sugar alcohols; (d) all the new constraints. The values in brackets represent the relative deviation from the constraint lower
bound (negative value) or upper bound (positive value)
10 of 11 | LOPES ET AL.

F I G U R E 1 Ice cream comparison regarding the amount of energy (a), saturated fat (b), protein (c), and carbohydrate (d) in 100 g of ice
cream (dark gray bar—ice cream obtained; gray bar—commercial ice cream). With respect to saturated fat, C-Water Ice and C-Sorbet have
less than 0.5 g/100 g and less than 0.1 g/100 g, respectively. With respect to protein, C-Water Ice has less than 0.5 g/100 g

With respect to saturated fat (Figure 1b), protein (Figure 1c), and 3. Water Ice/C-Water Ice, Dairy Ice/C-Dairy Ice, Sorbet/C-Sorbe-
carbohydrate (Figure 1d), the optimal recipes have lower amounts than tand Artisanal Ice Cream/C-Artisanal Ice Cream for carbohydrate;
the other ice creams. The largest differences occur with:
The fiber content of Milk Ice and Artisanal Ice Cream is much higher
1. Ice Cream/C-Ice Cream and Dairy Ice/C-Dairy Ice for energy and than that of C-Milk Ice and C-Artisanal Ice Cream, respectively.
saturated fat; The amounts of fat and protein determine the value of fat to pro-

2. Artisanal Ice Cream/C-Artisanal Ice Cream for protein; tein ratio. With respect to this ratio, C-Milk Ice, C-Ice Cream and C-

TA BL E 6 Characterization of popular ice creams in the Portuguese market regarding nutritional profiles

Ice creams
Nutrients & others C-Water Ice C-Milk Ice C-Ice Cream C-Dairy Ice C-Sorbet C-Artisanal Ice Cream

Fat <0.5 3 19 13 <0.5 8.8

Saturated Fat <0.5 2 14 7.8 <0.1 5.14

Protein <0.5 4 3.6 4.2 0.5 6.1

Carbohydrate 21 21 29 29 30 28

Fiber – <0.5 – – – <1

Fat/Protein – 0.(6) 5.2(7) 3.1 – 1.4

Energy 90 130 302 245 126 215

The values in each column represent the composition of 100 g of the respective mix in terms of nutrients(in g) and energy (in kcal) (“-”: there is no
information).
LOPES ET AL. | 11 of 11

Dairy Ice have values outside the range ([2,3]) imposed by constraints RE FE RE NCE S
(12) and (13) for Milk Ice, Ice Cream, Dairy Ice and Artisanal Ice Cream. Adam, D. (2005). How ice cream tackles your brain. Retrieved from The
Guardian- News: http://www.theguardian.com/uk/2005/apr/29/health.
science (2013)
Balasubramanian, S., Singh, K. K., Patil, R. T., & Onkar, K. K. (2012). Quality
4 | CONCLUSIONS
evaluation of millet-soy blended extrudates formulated through linear
programming. Journal of Food Science and Technology, 49(4), 450–458.
The aim of this study was to design healthy ice cream recipes using a
Danisco, Technical Memorandum, TM 2510–2e, Calculation of formulas for ice
friendly computer-aided tool. Linear programming models have been cream mix. Retrieved from http://www.flavours.asia/uploads/7/9/8/9/
developed with the purpose of minimizing the recipes’ calorie con- 7989988/calculation_of_formulas_for_ice_cream_mix_.pdf
tent. Constraints on fiber and sugar alcohol contents, protein to fat De Carvalho, I. S., Granfeldt, Y., Dejmek, P., & Håkansson, A. (2015).
ratio, freezing point depression factor and relative sweetness, besides From diets to foods: Using linear programming to formulate a nutri-
tious, minimum-cost porridge mix for children aged 1 to 2 years.
the usual constraints on fat, carbohydrates, lactose, milk solids non-
Food and Nutrition Bulletin, 36(1), 75–85.
fat and water, were included in the LP models. To evaluate the mod- € (2001). Optimization of wheat blending to produce
Hayta, M., & Çakmakli, U.
els, traditional Portuguese products were considered and the Excel breadmaking flour. Journal of Food Process Engineering, 24(3), 179–192.
Solver optimization tool was used to obtain the optimal solution of IBM -Data Processing Application, E20–0156-0 Linear Programming -
each model. Ice Cream Blending. Retrieved from https://archive.org/details/bitsa-
The impact of the new proposed constraints was analyzed by vers_ibmgeneralrogrammingIceCreamBlending_1450800

removing from the models subsets of these constraints. The solutions of Kilara, A., & Chandan, R. C. (2007). Ice cream and frozen desserts. In Y.
H. Hui (Ed.), Handbook of food products manufacturing (pp. 593–633).
the sub-models showed the effectiveness of the new constrains for bal-
Hoboken, NJ: John Wiley & Sons, Inc.
anced ice creams regarding technological and nutritional characteristics.
Larrosa, A. P. Q., Cadaval, T. R. S., Jr., & Pinto, L. A. A. (2015). Influence
The LP optimal recipes were compared with commercial ice of drying methods on the characteristics of a vegetable paste formu-
creams of the same type, which are market leaders in Portugal. With lated by linear programming maximizing antioxidant activity. LWT -
respect to energy, saturated fat, protein and carbohydrate, the optimal Food Science and Technology, 60, 178–185.

mixes had in general fewer amounts than the other ice creams. When- Martinell, N. (2007). Emotion-recognition software knows what makes you
smile. Retrieved from abcNEWS: http://abcnews.go.com/Technology/
ever it was possible to compare the fiber content, one noticed that the
story?id53381573&page51 (2013)
optimal recipes were much richer in this component. Therefore, in our
Official Journal of the European Union. (2006). Regulation (EC) No 1924/
opinion, the recipes generated by linear programming are healthier 2006 of the European Parliament and of the Council of 20 December
than the tested commercial ice creams of similar types. 2006 on nutrition and health claims made on foods, L 404/9-L 404/25.
The use of linear programming models allows a rapid examina- Okubo, H., Sasaki, S., Murakami, K., Yokoyama, T., Hirota, N., Notsu, A.,
tion and evaluation of all possible blend combinations and a swift . . . Date, C. (2015). Designing optimal food intake patterns to
achieve nutritional goals for Japanese adults through the use of lin-
determination of the best mixes in terms of calorie value. These
ear programming optimization models. Nutrition Journal, 14, 57.
models can be easily adapted to other objective functions (e.g., cost
Ryan, K. N., Adams, K. P., Vosti, S. A., Ordiz, M. I., Cimo, E. D., & Manary, M.
reduction, sensory acceptance), ingredient replacement and changes J. (2014). A comprehensive linear programming tool to optimize formu-
in requirement constraints. With the Solver tool from Excel, there is lations of ready-to-use therapeutic foods: An application to Ethiopia.
no need to acquire specific additional software to perform ice cream The American Journal of Clinical Nutrition, 100(6), 1551–1558.

optimization. Winston, W. L., & Goldberg, J. B. (2004). Operations research: Applications and
algorithms (with CD-ROM and InfoTrac) (4th ed.). Belmont, California:
Thomson Brooks/Cole.

ACKNOWLEDG MENTS
We wish to thank to (1) DuPont Nutrition Biosciences ApS, Ice Cream How to cite this article: Lopes I, Martins I, Mesquita M, Valença
Application – DuPont Nutrition & Health, Brabrand, Denmark for provid- de Sousa V, Ferreira-Dias S. Designing healthy ice creams with
ing technical and financial support; (2) Olivia Pinto for the valuable discus- linear programming: An application using traditional Portuguese
sions. This work was partly funded by FCT—Fundaç~ao para a Cie
^ncia e a products . J Food Process Eng. 2017;e12601. https://doi.org/10.
Tecnologia (Portuguese Foundation for Science and Technology) within 1111/jfpe.12601
the UID/MAT/04561/2013 and UID/AGR/04129/2013 projects.

You might also like