he 1995 Ig Nobel Prize in Physics [1], a highly popular spoof of its true counterpart, was awarded to D.M.R. Georget, R. Parker, and A.C. Smith of the Institute of Food Research, Norwich, England, for their "rigorous analysis of soggy breakfast cereal" [2]. Apparently, the prize committee judged that the analysis of processed food properties warranted little sci-entific rigor and bordered on the humorous. The authors accepted the prize with gratitude and offered the following gracious acceptance speech [3]: "In our study of compaction of breakfast cereal flakes, we didnot leave them turned tongue-in-cheek or use any other sensory technique. Rather, we set out to relate macroscale mechanical properties to changes in the scale of constituent food particle molecules. This provides valuable insights into texture. So what does this mean for the manufacturer, and to you, the consumer? Well, it's all about the quest for the ultimate breakfast cereal-eating experience. I

hope that the awarding of this prize will stimulate further research in this area." The above incident epitomizes what is interesting about the technology used in the food industry: On one hand, the finished product is a commodity

rarely associated with scientific or engineering know-how, easily recognizable, and available at an affordable cost. To wit, the English

vernacular, among others, is widely interspersed with metaphorical colloquialisms borrowed from food and cooking, such as "to accept something with a grain of salt" or "to offer a

half-baked product." On the other hand, while there is an abundance of processed-food humor [4], [5]—tasteful or not,

pun intended—food processing is a serious enterprise in every aspect, including science, technology, business, and public

health. The purpose of this article is to focus on technological issues related to snack food manufacturing processes [6].

Specifically, we posit that this technology can be surprisingly sophisticated. While we elaborate on specific examples, we also

POTATO CHIPSAND MICROCHIPS

ARE MORE SIMILARTHAN COMMONLY BELIEVED

MICHAEL NIKOLAOU

stress that it is not our aim to exhaustively cover this diverse and evolving field. In particular, we emphasize process monitoring and control solutions as well as the challenges of salty snack food processes. Undoubtedly, there are other success stories that can be told. We hope that this article stimulates interest in both the snack food manufacturing and control communities to promote creative control ideas.

SNACK FOOD BUSINESSAccording to the U.S. Census Bureau, "the snack food industry comprises establishments primarily engaged in one or more of the following: 1) salting, roasting, drying, cooking, or canning nuts; 2) processing grains or seeds into snacks; 3) manufacturing peanut butter; and 4) manufacturing potato chips, corn chips, popped popcorn, pretzels (except soft), pork rinds, and similar snacks [7]." The snack food industry sector also includes "consumer-ready packaged chocolate and nonchocolate candies, cookies and crackers, unpopped popcorn, and meat snacks [8]." The list of snack food products is growing steadily as competition, new knowledge in nutrition science [9], [10], regulatory mandate, and self-imposed guidelines for healthful public nutrition force companies to introduce snacks with refined features such as new raw material basis, improved texture, shape, color, flavor, and nutritional content [11], [12]. The last factor, in particular, has had a significant effect in recent years as Americans'

dietary habits fluctuate, in response to scientific advances, governmental guidelines, commercial offerings, social fads, and personal preferences. The world snack food market reached an estimated $66 billion in 2003, with the United States accounting for about a third [8]. As lifestyles in other parts of the world become more westernized, the global demand for snack foods continues to increase and evolve. The market is dominated by a few major companies [13] concentrating on efficiencies in their supply chain, manufacturing, and distribution as well as on process and product development, market research, and advertising. However, many small companies are also flourishing by concentrating on specialty snacks for which the consumer is willing to pay a premium.

SNACK FOOD MANUFACTURING TECHNOLOGYAlong with technology for manufacturing snack foods, food science is of paramount importance for this field. By including elements of physical, chemical, and biological sciences, food science is a truly interdisciplinary field of remarkable complexity and considerable breadth and has frequently contributed to, rather than simply benefited from, its constitutive scientific fields. Harold McGee [14], [15] offers a palatable introduction to the chemistry and physics of foods and cooking for the scientifically inclined casual reader and inquisitive engineer. Technology for manufacturing

ANNA MAVROMATIS

snack foods involves a small number of processes. Following [16], we offer a'short description along with more details in "Extrusion Cooking" and "Continuous Immersion Frying."

Extrusion CookingAn extrusion cooker (Figures 1 and 2) is essentially a high-temperature, short-time plug-flow bioreactor that combines into a single unit multiple unit operations: a) mixing of raw materials—such as various ground grains and water, b) shearing/kneading, and c) heating or cooking, using the heat released by friction inside the extruder [17]-[19]. For many extrusion cooking processes, the last function may be followed by puffing (hence the term "puffed snacks"), namely, abrupt expansion of high-temperature plasticized and gelatinized starch from a pressurized chamber into the atmosphere [6]. As is usually the case for processes that combine unit operations in a single unit (such as in reactive distillation), control of extrusion cooking processes is a formidable challenge due to strong coupling among many imprecisely understood phenomena. Additional challenges are posed by variability of raw materials, process modeling difficulties, equipment wear, and external and internal disturbances.

Extrusion Cookingxtrusion is a fabrication process in which a softened material is forced continuously through a die by a screw

rotating in a barrel (Figure 1). Extrusion cookers use heat released in extrusion to form and cook foods of various shapes and textures. These devices offer significant advan-tages over alternative cooking methods, including continuous high-throughput processing, energy efficiency, processing of dry viscous materials, improved textural and flavor character-istics of foods, control of thermal changes of food con-stituents, and use of unconventional ingredients. Single-screw extrusion cookers were developed in the 1940s to make puffed snacks from cereal flours or grits. As size grew to realize economies of scale, extrusion cookers were used for an increasing array of food products in the 1960s and 1970s. Twin-screw extruders [21], which were devel-oped in the 1970s and gained market share in the 1980s, are now widely preferred for their improved conveying and mix-ing capabilities, interchangeable screw profiles, expanded operational capabilities, and extended range of application. Extrusion cooking is used to manufacture food products such as puffed snacks, macaroni, ready-to-eat breakfast cereals, breadings, croutons, soup bases, drink bases, and pregela-tinized starch. Interestingly, pet foods are the largest product group manufactured by extrusion cooking, far exceeding the quantity of extruded human food products [22].

E

' ' '■'■••:."'

AUGUST 2006 « IEEE CONTROL SYSTEMS MAGAZINE 41

Continuous Immersion Frying

mmersion frying, the cooking of foods by immersion in hot edible oil, is widely used commercially to make fried foods such as

potato chips, French fries, doughnuts, extruded snacks, and fish sticks. Immersion frying offers high heat transfer rates and imparts a characteristic flavor, texture, and color to the final product [23]. Of the more than 2.5 million tons of snacks produced in the United States each year, most are immersion fried. The main phenomena of relevance to control of quality-related variables in immersion frying at the fried-food resolution level are the following:

I

» heat transfer (from the hot oil to the surface and interior of

the food) » mass transfer (evaporation of moisture from the food and

transfer of oil into the food) » browning (Maillard reactions between amino acids and reducing sugars).

Additional biochemical reactions that take place constitute a complex system.

At the process scale, a continuous fryer is a plug-flow reactor (Figure 3). A typical continuous frying system involves the following units (Figure 5): a long (-8-15 m) and shallow (~1 m) tank containing the frying oil; a pumping system that pumps and filters the frying oil; a heat exchanger system that transfers heat to the frying oil; a system of conveyors that move the product into, through, and out of the fryer; and a fume exhaust system for removing hot vapors generated during frying.

Invented as "Saratoga" chips in 1853 [24], potato chips (crisps in the United Kingdom) became widely available com-mercially after the invention of continuous slicing and frying in the 1920s and grew in popularity during the Depression era, to eventually become the quintessential American snack.

*s ■■■:,.':'

FryingFrying is a process in which a snack is cooked by floating or immersion in hot oil. Continuous fryers (Figure 3) are used for large-scale operations (~5,000 lb/h throughput), while batch (kettle-style) fryers are experiencing a comeback among small scale producers (<200 lb/h) who concentrate on specialty products. In addition to providing heat for cooking, the frying oil becomes a significant component of the end product, varying from as little as 10% by weight in breaded fish sticks to 40% in potato chips [6]. Responding to consumer preferences, the snack food industry is now emphasizing good-tasting, low-fat

FIGURE 1 Twin-screw extrusion cooking system. The first barrel is a preconditioner, which is used in conjunction with the extruder to increase residence time, reduce mechanical power consumption, and increase capacity. A cooling jacket around the extruder, instru-mented with thermocouples in various places, provides temperature control. (Used with permission of Wenger Manufacturing, Inc.)

snacks. Although conceptually simple, the design and operation of manufacturing processes for fried snacks poses interesting challenges. The key concern is low-cost production of nutritious snacks at consistent quality and minimum waste.

BakingIn baking processes, snacks are cooked by heat transferred through the air by convection, conduction, or radiation. The effectiveness of each of these mechanisms varies with oven and product design and is generally lower than that of heat transfer based on liquids such as oil. Baking does not contribute any added fat to the finished snack, potentially resulting in more healthful products, which /explains the recent surge in the popularity of baked snacks.

DryingDrying is required for several snacks to develop the right crispness. Puffed snacks, for example, must be dried after extrusion to achieve moisture content sufficiently low (to ~4%) to ensure good texture and storage stability. Because a puffed snack with moisture level above or below normal feels stale or spongy, there is a need for accurate control of the drying process.

Additional ProcessesWhile the above processes constitute the heart of most snack food manufacturing lines, where practically all cooking occurs, there are additional processes and equipment that are important for the operation of an entire line.

Oil, powder, and granule applicators, including oil and cheese sprayers, powder dispensers, electrostatic salters, and coating tumblers, are used for snack flavoring after cooking has been completed. By covering the same basic

substrate with different flavorings, this equipment creates multiple products using a single production line.

Storage of raw materials is of high importance given that raw materials are natural products that have limited life and can be seriously affected by storage conditions. For instance, the sugar content of potatoes increases when potatoes are stored at low temperature (2-4 °C) due to slowing of sugars breakdown to carbon dioxide and water. Sugar content can be lowered back to an acceptable level (<0.2-0.4%) by reconditioning through storage above 13 °C for several weeks prior to use. The acceptable sugar content in potatoes ensures that Malliard (browning) reactions and resulting undesirable dark spots on fried potato chips are avoided.

Measuring and weighing equipment is used throughout to provide information about the quality and quantity of the raw materials being processed. Such information (for instance, density or size of raw materials) is often used to adjust process conditions in a feedforward fashion.

Sorting equipment is used at various stages of snack processing, such as in the sorting of potatoes to make chips of uniform size. Various kinds of solids transfer equipment, such as conveyor belts and screw feeders, are

(a)

Water Dry Ingredients

also used between the main processing steps. Depending on the kind of snack being made, additional equipment may be used, such as nut-processing equipment, blanch-ers, roasters, and coolers. Finally, packaging design and related equipment greatly affect the shelf life of the finished product. From a process control viewpoint, ensuring that the weight of packaged snacks is as close to but certainly not below the value specified on the package has obvious economic implications.

FIGURE 2 Simplified schematic of twin-screw extrusion cooker. In food extrusion, the most effective manipulated variables are water and dry-ingredient feed rates, extru-date and barrel temperatures, and screw speed [57].

(b)

FIGURE 3 (a) A high-capacity (4,200 Ib/h) continuous frying system, shown with the oil circulation and heat-exchanger, (b) A compact (400 Ib/h) continuous fryer with the top raised for cleaning and maintenance. The quality of fried chips is quantified by specifications on moisture content, oil content, and color. A common disturbance is product load changes. Quality is controlled mainly by manipulating the residence time and frying temperature of the chips in the fryer. A submerger (electrically operated metallic belt with equidistant paddles perpendicular to the belt; see Figure 5) is used to manipulate residence time. The speed of the submerger's electric motor is adjusted so that submerged chips travel at the desired speed. To manipulate the frying temperature, oil circulates continuously between the fryer and heat exchanger. Multiple injection points of hot oil are used to ensure the desired temperature profile along the fryer according to cooking preferences. Because part of the frying oil leaves with the chips, the oil must be continuously replenished. In addition to creating fast process dynamics, low oil volume and rapid oil turnover ensure fresh product with a long shelf life. A level controller maintains the oil level at its optimum. (Used with permission of Heat and Control, Inc.)

CONTROL ISSUES IN SNACK FOOD PROCESSES

Controllability and Process Design

Snack food manufacturing processes are generally serial (Figure 4) and thus are closer in nature to microelectronics manufacturing processes than to chemical plants. In particular, microelectronics manufacturing entails several tens of elaborate steps to transform silicon and other materials to finished microchips, whereas chemical plants usually rely on reactor/separator schemes that involve partial reaction of raw materials in a reactor, followed by separation of nonreacted raw materials in a separator and recycling to the reactor for further conversion. The latter pose significant control challenges, owing to time lags introduced by recycling streams. Recirculation of heating or cooling

Control Feed Grinding

<> , <>Destoning and Washing Mixing

o <>Peeling Preconditioning

<> <>Inspection Extrusion Cooking

<> <>Slicing Drying/Frying/Toasting

<> <>Slice Washing Coating/Flavoring

~^^~ <>Blanching Cooling

<> <>Dewatering Inspection

<> <>Frying Packaging

^J*~ <>Inspection Distribution

<>Flavoring

<>Cooling

<>Packagina

<>Distribution

(a)

FIGURE 4 Examples of snack-food manufacturing lines, (a) Components of a potato-chip production line. The inspection blocks involve visual inspection of peeled raw potatoes or finished chips. Until recently, inspection was done manually, a strenuous, labor-intensive task that entails removing unwanted raw potato parts and off-spec chips, (b) Components of an extruded snack production line. In both lines, the heart of the system is the reactor, that is, the fryer or extrusion cooker.

media is common in snack food manufacturing, such as recirculation of cooking oil in continuous frying (Figure 5) or cooling water in extrusion cooking (Figure 2), but is in large part decoupled from product throughput. On the other hand, because processed materials are solids, successive processing steps cannot be decoupled by use of intermediate storage tanks (the counterpart of surge tanks for fluids in chemical industry processes) that would absorb throughput fluctuations. As a result, throughput control must be tight and reliability must be high to avoid major process disruptions and product loss [16].

The dynamic behavior of individual processes in snack food manufacturing lines poses interesting control challenges. For example, installation and maintenance costs limit the capability for intermediate measurements related to product quality in many cooking units (such as frying, extrusion cooking, and baking) that are similar in function to plug-flow reactors familiar in chemical plants, thus introducing dead times, typically on the order of 1 min, which create inherent control limitations [25]. Time-varying behavior is frequently unavoidable, due to both raw-material and cooking-medium variability as well as process drifting. Interactions among controlled variables are often inevitable such as when oil replaces moisture content in fried snacks. Nonlinearities frequently arise from the kinet-ics of chemical reactions as well as from mass and energy transport at multiple scales and complex geometries. A typical example is process startup or shutdown, while a more exotic example is multiple steady states for the same values of the primary process variables in extrusion cooking [26]. However, control near a steady state can be effective using multivariable linear control methods [27].

Selection of Measured and Manipulated VariablesSnack food manufacturing processes involve control of process-related and product-related variables. While controlling process-related variables, such as cooking time or temperature, facilitates control of product-related variables in an open-loop fashion, direct feedback control of product-related variables entails additional measurement requirements but is advantageous in controlling product quality.

Controlling product-related variables is essential for ensuring product quality. While consumers can easily determine whether a product is made to their liking, translating this perception to specifications for a small number of product-related variables that can be measured— preferably in real time or, at least, intermittently ir. an analytical laboratory—is difficult. Experience has shown that the connection between product quality and measurable variables is not always dear cut [28]. Snack food producers rely both on sensory-evaluation experts as well extensive statistical analysis of consumer te~: :e?_rs to ider.ur ■ =: iz.es that are correlated with p: . .: quality zr.z :--::--.:-- optimal values for these varii rles, which

(b)

FIGURE 5 Schematic of a multizone continuous fryer. The dual submergers and oil inlets provide improved controllability. Key variables associated with product quality are moisture content, oil content, and color, all of which can be measured in real time at the outlet. Key manipulated variables are oil temperature, submerger speed, and take-out-conveyor speed. Secondary loops ensure that process variables such as oil-bath level, oil temperature, and oil flow are maintained at desired values [16]. Because part of the oil leaves the fryer with the fried chips, the oil is continuously replenished with fresh oil at controlled amounts, thus simultaneously suppressing oil rancidity in the fryer.

can be strictly confidential. For example, it is well established that, for many fried snacks, moisture content, oil content, and color capture qualitative features such as crispness, extent of cooking, and taste fairly well [16]. The availability of inexpensive real-time imaging equipment facilitates quality control by means of multivariate image analysis [29], [30]. Crispness is also routinely tested offline in the laboratory by analyzing the frequency spectrum of the sound made when a snack is crushed. Effectively using a blend of offline laboratory and real-time measurements for process control offers interesting challenges, similar to challenges encountered in the chemical or microelectronics manufacturing industries.

Historically, product-related variables were controlled by open-loop manipulation of process variables, for instance, by ensuring that the food was processed for a certain time. For example, quality control of fried foods was ensured by controlling the temperature of the frying oil and residence time of the food in the fryer. Similarly, studies have connected process variables such as energy release and throughput to the quality of the finished product in extrusion cooking [31]-[34]. Interestingly, this situation is similar to control of product quality variables in crucial steps of semiconductor chip manufacturing, such as plasma etching of silicon wafers [16]. Indeed, the difficulty in taking real-time measurements of product quality during plasma etching has resulted in the dominant approach of the microelectronics industry in which "ecipes" prescribe time profiles of process conditions

lLt need to be precisely controlled (and begging for P1111'. such as the one bv Michael J. Boskin, chair of the

residt-nt's Council of Economic Advisers: "It doesn't e any difference whether a country makes potato chips or computer chips!")

The availability of reliable and inexpensive sensors for direct, real-time measurement of variables related to product quality in the 1980s and 1990s rendered it commercially pos-sible to control these variables directly using computer-aided feedback control [25], [27], prompting modeling and control studies to elucidate both fundamental concepts and practical aspects. For example, the empirical observation that the oil content of fried chips is proportional to the square root of residence time, used in recipe forming, was attributed to heat and mass transfer mechanisms [25] as discussed below.

The availability of manipulated variables (actuators) that render a process easier to control is largely dictated by process design. For example, increased controllability is the main reason that extrusion cookers (Figure 2) use a twin-rather than single-screw configuration. For a similar reason, multizone continuous fryers use multiple submergers (Figure 5) to make it possible for fried snacks to travel at different speeds in different zones of a fryer. As another example, when real-time image processing is used to detect defective fried snacks, an activated chute located under the transport path of fried snacks rapidly opens and closes to ensure that unsatisfactorily fried snacks are removed once they are detected. As an example from the nut-processing industry, an air gun with multiple nozzles is used to blow a suspect nugget out of a stream of cracked nuts if infrared or laser-beam sensors detect poor shape or color [35].

Control StructuresThe control of snack food manufacturing processes is mul-tilevel, not unlike control of chemical processes (Figure 6). Explication of this multilevel structure provides guidance for the rational design of control strategies at various levels, from the manufacturing capacity level down to regulatorv feedback control.

Take-Out Conveyor ■

Capacity Planning (Years)

Production Planning (Years/Months)

Scheduling (Months/Weeks/Hours)

~?*>--------:------------------1

Predictive Control (Minutes/Seconds)

PID Control (Seconds)

Manufacturing Process

FIGURE 6 The multilevel control hierarchy of snack food manufactur-ing processes at multiple time scales is similar to the hierarchy in other process industries. Decisions from upper levels are passed to lower levels, whereas information from lower levels is passed upward. Because the time scale decreases as one moves down-ward, decisions made at each level are assumed to be instantly fol-lowed by the next lower level. Various modeling and decision-making paradigms are used at different levels. For example, explicit stochastic models, first-principles models, and dynamic programming can be used at the planning and scheduling levels, whereas empirical models and online optimization or explicit control algorithms are used at the predictive control and PID levels.

To build manufacturing capacity, one has to consider fluctuations in demand and changes in the appeal to con-sumers for the finished product. Manufacturing multiple products in the same production line offers economies of scale but creates additional operability and control challenges. To ensure controllability of large-scale snack food manufacturing lines, laboratory-scale proof of concept and pilot-plant testing are undertaken before full-scale production.

Given that the raw material is a natural product, production must be carefully planned. Supply-chain optimization can provide significant returns. Production scheduling poses challenges because both the raw material and the finished product have limited storage time, and both supply and demand can fluctuate; for instance, think natural disasters or Super Bowl weekend.

At the predictive control level [36], the availability of real-time sensors for product quality variables facilitates the move away from recipe-based control to real-time feedback control, where process conditions are cascaded as setpoints to underlying controllers. Because snack food manufacturing processes are serial, information related to the quality of intermediate products (such as the sugar

content of stored starchy raw materials, thickness of sliced chips to be fried, moisture content of cooked snacks to be dried, and cooking rate of products to be eventually packaged) is often passed to subsequent steps for use in feedforward control.

Control Objectives, Constraints, and Control LawsThe control of quality characteristics of manufactured snacks offers significant economic incentives. For example, waste minimization in manufacturing lines offers a competitive edge that largely determines the business viability of entire product lines, even though manufacturing costs are in strong competition with advertising and distribution costs, where economies of scale are important. Therefore, a primary objective is to manufacture product at the desired specifications. For example, if the moisture content of a fried snack chip is below specification, the product imparts a scorched flavor, dark color, and excessive oiliness, while above specification the product lacks crispness [37]. Meeting quality-related constraints that are advertised for the finished product (such as low fat) is also important.

Because most raw materials for the snack food industry are natural products, they can exhibit substantial variability. Such variability may be due to the variety of a particular crop, location of cultivation, weather conditions during crop growing and harvesting, and storage conditions. Similar concerns exist for other process industries, such as the petroleum, metal, and pulp and paper industries, which rely on the primary production sector for raw materials, respectively, crude oil, metal ores, and timber. However, in contrast to industries dealing with fluids that can be blended, such as the chemicals and oil-refining industries, the snack food industry deals with products that can be neither blended nor reworked to meet quality specifications, making tight control all the more crucial.

As mentioned above, the availability of real-time sensors, combined with inexpensive real-time computing power in the form of programmable-logic controllers (PLCs) and distributed control systems (DCSs) motivates the use of feedback control approaches. Many of these approaches are model based, raising questions about the level of detail at which a process must be modeled, the kind of model used (for example, first principles or empirical, linear or nonlinear, lumped or distributed parameter), the modeling accuracy required for feedback control (for example, numerical accuracy, fuzzy relationship accuracy) and the requirements posed by the model for controller development and maintenance [16]. Control-related modeling aspects of extrusion cooking are discussed in [38]-[40], while an early case of feedback control of moisture in food extrusion is discussed in [41]. A more recent study of model predictive control is discussed in [27]. Modeling of immersion frying is discussed in [42], while a multiscale model is discussed in [43]. Along the same lines, modeling of a continuous fryer for control is discussed in

46 IEEE CDNTROI SYSTEMS MAGAZINE » Al Ifil 1ST POOR

[25], while modeling and control of continuous frying is discussed in [44]-[47]. Finally, a thorough experimental study on automated controller performance monitoring is discussed in [48].

CASE STUDY: FRYER MODELING AND CONTROLWe now illustrate modeling and control for a snack-food manufacturing process, namely, continuous immersion frying. While the modeling and control approaches are based on well-established fundamentals, this system has interesting features such as multiple scales, from a single potato chip to the production line, interactions among controlled variables, and time delays.

Dynamic Model for a Continuous FryerA typical commercial fryer is around 11 m long and 30 cm deep (Figure 5). Before being fried, potatoes are peeled and cut into thin slices, nominally 1.1 mm. These slices are washed to remove surface starch released in the slicing operation and then immersed into the hot end of the fryer. The motion of the slices in the fryer is controlled by sub-merger paddles, whose speed is adjusted to maintain a desired residence time for the chips, usually around 2 min. During frying, the moisture content in the slices falls from about 80% at the inlet of the fryer to around 2% at the exit of the fryer, while the fat content increases from 0% to 40% by mass. The temperature of the circulating oil drops by 20 °C, primarily due to the energy used to evaporate the moisture trapped inside the potato slices. The cooler oil exiting the fryer is then passed into a heat exchanger, where the oil is reheated to the required temperature and pumped into the inlet of the fryer. The chips leaving the fryer are cooled and transported for packaging.

For control purposes, a continuous fryer can be considered a particulate two-phase flow system [49], where the oil and chips (slices) represent the continuous and dispersed phases, respectively. The model describes how the oil temperature T0fl, chip temperature Tchip, moisture content

Y M»

1+MW+M0

oil content

M0 1 + Mw + M0'

and chip color B change with time and space. Here, Mw and M0 are the moisture (water) and oil mass, respectively, per mass of solids, and B refers to luminosity or light intensity at selected frequencies, usually expressed in normalized, nondimensional units.

We model a continuous fryer as a plug-flow reactor, that is, a tubular chemical reactor with uniform velocity pattern over its cross-sectional area. We assume constant temperature gradient along each fryer cross section. Each potato slice is considered as an infinite slab, which implies that the

heat and mass transfer occur only in the direction perpen-dicular to the slice surface. We assume that moisture from the slice interior is transferred to the surface and then evaporates instantly, so that the rate of moisture evaporation is governed by the diffusion rate of moisture to the surface [50], The oil uptake by the chip is also modeled as a diffusion process following Fick's law [51]. It is assumed that no temperature gradient exists across the chip, due to the small thickness and high heat conductivity of potato solids. For simplicity we also assume that all slices are uniform with respect to thickness, moisture content, oil content, and solids content; shrinkage of a chip during the frying process is negligible; the specific heat and density of oil are constant over the range of operation; the heat required for chemical changes such as starch gelatinization and browning (Maillard reactions) is negligibly small compared to that required for vaporization of moisture [52]; finally, heat losses to the environment are negligible.

Let x and z denote horizontal and vertical coordinates along the fryer, respectively (Figure 7). Energy balance for the oil phase along the fryer length scale (x-direction) yields

Tc) ~ ^Nm,{-poPo

where T0 = T0(t, x) is the oil temperature, assumed to vary in the x direction only; v0 is the oil flow velocity; ht is the heat transfer coefficient between the oil and chip phase surface; Cp0 is the specific heat of the oil; p0 is the density of the oil; Tc = Tc(t, x) is the temperature of the chip phase, assumed to vary in the x direction only; «j is the heat transfer area between the oil and chip surface per unit volume of oil; AH is the evaporation heat of water; and Nm is the moisture flux at the surface of the chip phase (x = Lc/2, where Lc is the chip-phase thickness, see Figure 7), calculated as

FIGURE 7 Coordinates for fryer modeling (fryer not drawn to scale). Oil temperature typically ranges between 150-190 °C. Moisture from inside the chip phase moves to the surface, where it instantly evaporates. At the same time, oil moves into the chip phase to fill the void. The chip and oil phases travel at different velocities, as shown in Figure 5.

X0 =

dT0 dT0---------h Vn------

dt dX(T0 (1)LpoPc

f ...............-............■........----------------------------....._— —...................................... ■—\

TABLE 1 Continuous fryer model parameters used in the case study. All parameters values are typical and were obtained from open-literature references.

Parameter Symbol and Value

Color reactions activation energy (Fitted from data in [55])

Eab = 23.34 kJ/mol

Color reactions kinetic constant (Fitted from data in [55])

/cw = 0.2592 s-'

Moisture diffusivity [50] D.M = D.M exp [^f 1Eaw = 24.2 kJ/mol

0^ = 6.2317 x 10-6m2/s

Oil diffusivity [50] D0 = Do0 exp [^] Eao = 11.89 kJ/molDo0 = 8.8752 x 10-9m2/s

Potato tuber density [52] pc = 1080 kg/m3

Potato tuber moisture content [52] XM = 81 wt%

Oil specific heat [52] Cpo = 2093.0 J/kg-rC

Oil density [52] p0 = 920 kg/m3

Heat transfer coefficient between oil and chip (boiling or nonboiling) surface [52]

ht = 650or250J/s-m2-'C

Moisture heat of evaporation [52] . AH = 2.16892 x 106 J/kg

Moisture specific heat [52] Cpw = 4180 J/kg-°C

Chip specific heat [52]-

Cpc = 3050.0 J/kg-'-C

where Tf-{t) is the temperature of the chip phase at the inlet of the fryer. Assuming that the fryer is initially at steady state, the initial condition for (5) is

Tc(0,x) = Tf(x).

Because of significant moisture gradients in the

chip phase both in the direction of and perpendicular

to the flow, mass balance for moisture content on the

chip phase yields

dMw dMw

Do

where Mw = Mw(t, x, z) is the moisture in the chip

phase, assumed to vary in both the x and z directions.

The boundary conditions are

AW*, 0, z) = M£(Q at the

inlet of the fryer;

Mw t, x

at the surface of the chip phase, because of instant

evaporation; and

Nm = CcDwz=Lc/2

where cs is the solids density and Dw is the apparent diffusivity of

moisture. The boundary condition for (1) at the fryer inlet is

{t, x, 0) = 0

at the center of the chip

phase, because of symmetry. The initial condition for (8) is

Mw(fi,x,z)=M^(x,z).

T0(t,Q) = Tf{t),

where T0n(t) is the temperature of oil at the inlet of the fryer.

Assuming that the fryer is initially at steady state, the initial

condition for (1) is

(4)T0(P, x) = Tf(x),

where the superscript ss denotes steady state. Energy

balance for the chip phase yields

Similarly, mass balance for the chip phase oil content yields

8M0 3M0 d2M0---------h Vr------ = D0--------s-,

dt dX dZ2

where M0 = M0(t, x. z) is the oil content of the chip phase, assumed to

vary in both the x and z directions; and D0 is the apparent diffusivity

of oil. At the inlet of the fryer the chips contain no oil, corresponding

to the boundary condition

BTC , dTc

dt dx

where vc is the velocity of the chip phase flow. The boundary

condition for (5) at the inlet of the fryer is

Mo(t,0,z) = 0.

At the surface of the chip phase, the oil content is assumed to equal

the equilibrium value Ms0 estimated by filling with oil the pore

volume left by the evaporation of water from the solid matrix surface,

yielding the boundary condition

Tc(f,0) = T'n(t), (6) Mo I t,x, 4 1 - Ml (15)

48 IEEE CONTROL SYSTEMS MARA71NF « Al IGj IQT 5nrlK

(7)

d2Mu '

3z2 (8)dxBt

(9)

= 0 (10)

dUn (2)3z

dMu dz

(11)

(12)

(3)

(13)

2ht(T0-Tc), (5)

(-pcPc t̂(14)

Symmetry around the center of the chip phase yields dM„

■(t ,x ,0) = 0.

Assuming that the fryer is initially at steady state, the initial

condition for (13) is

Mo(0, x, z) = 0.

The color of a fried chip depends on the extent of browning

reactions taking place at rate R during frying. The browning

reactions can be lumped into a single color reaction, yielding the

mass balance

(16)

dz

(17)

r 1TABLE 2 Steady-state values for input variables of the fryer model used in the case study. Values that are typical for raw materials and commercial production are used.Parameter Name Symbol and Value

Oil velocity v0 = 0.3667 m/s

Oil volumetric flow rate Fm = 0.1087 m3/sInlet oil temperature r^n = 1603C

Potato slice flow rate Fwc = 2.08 kg/sSlice thickness Lc= 1.09 mmInlet dry basis moisture content Mw = 4.3125 ^«

9B „ -vc— +R, dx

where B = B(t, x) lumps all products of the color reaction,

expressed in normalized, nondimensional units. Assuming all

browning to be the result of frying, we obtain the boundary condition

(19)B(t, 0) = 0

at the inlet of the fryer. Assuming that the fryer is

initially at steady state, the initial condition for (19)

is

at the inlet of the fryer, where Xvc is the volumetric fraction of the

chip phase. Note that (22) is derived assuming that slice thickness is

uniform. However, the thickness of potato slices coming from a

commercial slicer is usually normally distributed with a standard

deviation as high as 10% of the mean value [53]. Nevertheless, the

lack of uniformity is not a problem, as long as the slice thickness

distribution is symmetric. Assuming that the fryer is initially at

steady state, the initial condition for (22) is

B(0,x) = Bss(x).

The reaction rate R is assumed to correspond to

first-order kinetics, namely,

JR sa Bk\,Q expRgTc

where k^o is the reaction rate constant, Ea^ is the

activation energy for the browning reactions, and

Rg is the universal gas constant.

Since in different zones of the fryer, such as the

free-floating and submersion zones shown in Figure

5, the velocities of the chips are different, we need a

continuity equation for the chip phase to calculate

the interfacial area a\ = a\{i, x) between the chip and

oil phase for each zone. Given that the number of

chips in the fryer is conserved, we obtain

dai

' dx '

with the boundary condition

«l(f, 0) = IXvc/Lc

5 10

Length (m)

(d)

FIGURE 8 Steady-state profiles for moisture and oil content, oil temperature, chip temperature, and chip luminosity. The simulation is carried out beyond the fryer length (11 m) to include the length of the take-out conveyor belt (Figure 5). In the take-out conveyor belt segment, the chip continues to lose moisture, but all browning reactions are assumed to drop drastically.

(18)dB

It

100 180(20)

170

160

150

(21)

9flj

~dt (22)

(23)

1

4.5

4

3.5

Mw, 3

2.5

kgH20 2

kg Solid 1.5

1

0.5

0

kg Oil kg Solid

Potato Slice Half-Thickness (mm) (b)

FIGURE 9 (a) Moisture and (b) oil profiles inside a chip at equidistant locations along the fryer (1 = inlet, 5 = exit). The inlet moisture profile indicates that the chip surface is almost instantly dried upon entering the fryer, whereas the entire chip is virtually dry at the exit. However, much of the interior chip space vacated by moisture is not occupied by oil.

-----N Exit % Moisture

1.4 \

1.35 \

1.3 . v_1 PR

;

Steady-State Behavior of a Continuous FryerFigure 8 shows simulated steady-state profiles of output variables along the fryer. Figure 9 shows the moisture and oil content profiles across the

chip phase at various locations in the fryer. Figure 10 shows the dynamic response of the output variables to a step change in one manipulated input. These simulations are used to calibrate real-time measurements from oil sensors that provide surface rather than average oil content of

chips exiting the fryer and to design frying temperature profiles along the fryer to achieve desired product qualities such as low fat content.Open-Loop Control of a Continuous Fryer by Manipulating Residence TimeIt has long been known empirically that the moisture content and oil uptake, which are the main quality-related variables of a fried chip, are correlated with the square root of frying time in batch frying experiments under constant temperature [43], [50], [52], [54]. This relationship has been used to control moisture and oil content of a fried chip by manipulating chip residence time in a continuous fryer. While such a control policy does not use any feedback information related to the controlled variables, it is worth examining the fundamental origin and soundness of the observed correlation.

For a batch fryer, substituting y = (Lc/2) - z into (8) yields

511 IEEE CONTROl SYSTEMS MARA7INF » Al IRJ 1ST pnnfi

«/(0, x) = af(x). (24)

The continuous fryer model comprises (1), (5), (8), (13), (18), and (22). Model parameters used in subsequent simu-lations are summarized in Table 1. Steady-state values for model inputs are shown in Table 2.

36.6

200

400

600

17C

165160155

200 Time (s)

200

400

600

600200 400 Time (s)

400

FIGURE 10 Responses to a 5% increase of inlet oil temperature (manipulated vari-able). Note that the chip temperature at the exit responds much faster than the other variables due to the oil-flow velocity, which is higher than the velocity of the chips.

dMw _ d z M w

with the boundary conditions Mw(t, 0) = 0 and Mw(t, oo) = M^, (for short times) and the initial condition Mw(0, y) = M°. The solution to (25) is

Snack food manufacturing processes

Involve control of process-related

and product-related variables.

Mw

MS,erf y

.V4D^i,(26) Yhe steady-state relative gain array for this model is

from which the mass flux for moisture N^z at the chip surface

isA =

1.6

-0.6

-0.6

1.6(31)

= -<A/^. (27)y=0

Then, the amount of moisture lost over the area A is

t A

= J f NAz(r, y)\y=0 dQdz = -At^^f—,

1.451.42

1,000 2,000 3,000 4,000 5,000

Time (s)

moisture loss Ww , ,„ 1 /D^t- -M?,

volume ALC Lc V 7T

which is the desired result. A similar derivation can be performed for oil

uptake.

The above analysis explains why the residence-time rule developed for batch frying is not directly applicable to open-loop control of continuous fryers for which (8), rather than (25), is applicable. Equation (8) can be put in the form of (25) if one considers variables in deviation from steady state and uses the approximation (BMw/dx) « constant, as Figure 8(a) suggests.

Simplified Model of a Continuous Fryer and Controller DesignUsing the dynamic response of the chip moisture and oil content at the exit of the fryer to a step change in the inlet oil temperature and

submerger speed, we obtain the

approximate linear model0.0192e'

Ayw(s) Ay0(s)■D CD <D Q.0)

cBX!

2,000 3,000

Time (s)

2,000 3,000

Time (s)

2,000 3,000

Time (s)

L 9700s2 + 120s + 1 4300s2 + 115s + 1 AT!"(s)

AS(s)

(25)

3M-̂

(28)WA

N A z ( t , z ) \ u = -D,

o

o which yields

1,000

9J/

4,000 5,000

1.000 4,000 5,000

1,000 4,000 5.000

(29)

-100s-100s

-0.0123e

2005s2 + 80s +1

0.028e-100s

2100s2 + 87s + 1 -

0.115e-100s

(30) FIGURE 11 Closed-loop responses (process outputs and inputs) to a 5% increase of slice thickness (disturbance). Note that the controller forces the potato slices to remain longer in the fryer by reducing the submerger speed while simultaneously lowering the frying tempera-ture by reducing the inlet oil temperature.

4.5

4

3.5

Mw, 3

2.5kgH20 2

kg Solid 1.5

1

0.5

0

Mn

kg Oil

kg Solid

Potato Slice Half-Thickness (mm)

(b)

FIGURE 9 (a) Moisture and (b) oil profiles inside a chip at equidistant locations along the fryer (1 = inlet, 5 = exit). The inlet moisture profile indicates that the chip surface is almost instantly dried upon entering the fryer, whereas the entire chip is virtually dry at the exit. However, much of the interior chip space vacated by moisture is not occupied by oil.

«j(0, x) = (f(x),-----N Exit % Moisture

1.4 • \1.35 ^ \

1.3

1 95

■ v_____•

The continuous fryer model comprises (1), (5), (8), (13), (18), and (22). Model parameters used in subsequent simulations are summarized in Table 1. Steady-state values for model inputs are shown in Table 2.

Exit % Oil

36.4 /^~36.2 /

3635 P.

_^/

200 400Time (s)

2 8 x104

Exit Browning

i.i /"""

2.6■ /

2.5■ /

?4 —'

200 400Time (s)

FIGURE 10 Responses to a 5% increase of inlet oil temperature (manipulated variable). Note that the chip temperature at the exit responds much faster than the other variables due to the oil-flow velocity, which is higher than the velocity of the chips.

Steady-State Behavior of a Continuous FryerFigure 8 shows simulated steady-state profiles of output

variables along the fryer. Figure 9 shows the moisture and oil content profiles across the chip phase at various locations in

the fryer. Figure 10 shows the dynamic response of the output variables to a step change in one manipulated input. These

simulations are used to calibrate real-time measurements from oil sensors that provide surface rather than average oil content

of chips exiting the fryer and to design frying temperature profiles along the fryer to achieve desired product qualities

such as low fat content.

(24)

200

400

600 600

17C

165

160

155

200 Time (s)

600 400

Open-Loop Control of a Continuous Fryer by Manipulating Residence TimeIt has long been known empirically that the moisture content and oil uptake, which are the main quality-related variables of a fried chip, are correlated with the square root of frying time in batch fry-ing experiments under constant temperature [43], [50], [52], [54]. This relationship has been used to control moisture and oil content of a fried chip by manipulating chip residence time in a continuous

fryer. While such a control policy does not use any feedback information related to the controlled variables, it is worth examining the fundamental origin and soundness of the observed correlation.

For a batch fryer, substituting y =s (Lc/2) - z into (8) yields

50 IEEE CONTBOI SYSTFMS MARA7INF » Al irsi 1ST 9nnR

Controlling product-related variables is

essential for ensuring product quality.

indicating that significant interactions arise among individual feedback loops under decentralized control of moisture and oil content. The condition number of the steady-state gain matrix for (30) is 16, suggesting mild ill conditioning [58]. Therefore, robust multivariable control methods are appropriate for this system. Figure 11 shows a simulation of disturbance rejection using (30) with a model predictive controller.

Application to Industrial ProcessThe simulation study shown in Figure 11 provides the basis for designing a model-predictive control system for an industrial continuous fryer. Control practice prior to this project relied on decentralized loops (for instance, for temperature control) as well as open-loop manipulation of residence time. Relative gain array analysis shows that interactions among control loops can pose problems for such a strategy, thus warranting multivariable control.

While variables such as oil and moisture content of fried chips can be controlled individually over only a limited range, independent control is feasible for small deviations. For example, simultaneous manipulation of oil temperature and submerger speed can be used to reduce the oil content of fried chips at a somewhat lower value while maintaining low moisture content, particularly for fryers with multiple submergers and oil-recirculation paths; see Figure 5. All real-time computations for this multivariable control algorithm are performed using off-the-shelf software running on a dedicated PC, which communicates with an existing PLC system.

CONCLUSIONSIn Part Two of the Pulitzer-Prize-winning best-seller Guns, Germs, and Steel [56], the beginning of the rapid ascent of humankind after the latest Ice Age is attributed to the rise and spread of food production, which capitalized on efficiencies in cultivation, storage, and distribution methods. A few thousand years later, Malthus's grim predictions about an impending global famine were refuted by a new wave of efficiencies in food production and distribution, spurred by the Industrial Revolution. Today, extreme efficiency in food production has made it sufficient that only a very small percentage of the industrialized world's population be employed in agriculture. However, the food industry as a whole is sizable, offering both employment and essential products. Efficiencies in all aspects of that industry remain important, both for the benefit of society and for the compet-

itiveness of individual companies and businesses, large or small. In this article we showed that automatic control can greatly contribute toward efficiencies in this multifaceted and interdisciplinary field. In fact, advances at all levels of the hierarchy discussed in Figure 6 can be beneficial, from optimization of supply chains at the top to control of individ-ual units at the bottom. Moreover, because Figure 6 involves feedback loops at all levels, advances can rely on all elements of a feedback loop, including process, sensors, actuators, and information transmission lines.

The examples discussed in this article offer the reader a taste of the challenges and possibilities in the food manufac-turing industry, while illustrating the benefits of crossing dis-cipline borders to benefit from and contribute toward cross-fertilization of ideas. In particular, collaboration between academia and the food manufacturing industry has proven fertile, and, if nurtured, can continue to be fruitful in the future.

ACKNOWLEDGMENTSThis article is based on work partially supported by the National Science Foundation, Shell Development, Frito-Lay, and the Texas Advanced Technology Program. The author gratefully acknowledges that support. Computations for the case study were performed by Dr. Manoj Shouche and Dr. Wang-Yan (Charles) Feng, to whom the author is indebted.

AUTHOR INFORMATIONMichael Nikolaou (Nikolaou@uh.edu) received his diploma from the National Technical University, Athens, in 1984 and his Ph.D. from the University of California, Los Angeles, in 1989, both in chemical engineering. From 1989 to 1997 he was an assistant professor and then associate professor in the Chemical Engineering Department at Texas A&M University. He held a visiting scientist position at the Massachusetts Institute of Technology in 1995, before moving to the University of Houston in 1997, where he is presently associate professor. His research interests are in computer-aided process engineering, with emphasis on process modeling, identification, monitoring, and control, and their applications to the chemical, petroleum, microelectronics, food, and biomedical industries. His industrial experience includes consulting with companies in diverse fields. He can be contacted at the Chemical Engineering Department, University of Houston, Houston, TX 77204-4004 USA.

REFERENCES[1] M. Abrahams, "The 1995 Ig nobel prize winners" [Online]. Available: http://www.irnprobable.eom/ig/ig-pastwinners.htrnl#igl995 [2] D.M.R. Georget, R. Parker, and A.C. Smith, "A study of the effects of water-content in the compaction behavior of breakfast cereal flakes," Powder Technol., vol. 81, no. 2, pp. 189-195,1994.[3] M. Abrahams, The Best of Annals of Improbable Research. San Francisco, CA: Freeman, 1998. [4] L. DeLong, "Engineering bytes and bites" [Online]. Available: