WO2000021380A1 - A dry milk product and process for the manufacture of same - Google Patents

Abstract

A process for producing dry milk product containing < 10 % moisture, and preferably < 5 % moisture, having lactose in essentially the crystalline state and containing from 20-40 % fats by weight of which 80-100 % are in the free state by processing dry milk at elevated temperature with shear in a mixing or grinding device. The resulting product is benefical as an ingredient for confectionery, especially chocolate, to reduce viscosity and improve flavor.

Description

TITLE: A DRY MILK PRODUCT AND PROCESS FOR THE MANUFACTURE OF SAME

GRANT REFERENCE The invention here set forth was partially funded by USDA funding provided under the Hatch Act for project #3591.

CROSS REFERENCE TO RELATED APPLICATIONS

This application claims the benefit of U.S. Provisional Application No. 60/103,990, filed October 13, 1998, the disclosure of which is hereby incorporated by reference.

BACKGROUND OF THE INVENTION

Dry milk powder is used for a variety of purposes. One industry that uses significant quantities of dry milk is the chocolate industry, for example, Chocolate Manufacturers' Association members use 3.5 million pounds of whole milk daily. In 1996 and 1997, production of dry whole milk, mostly for the confectionery industry, was approximately 130,000,000 pounds. Dried whole milk powder ( MP) has generally been prepared by two different processes, roller drying and spray drying.

During conventional spray drying of whole milk, milk fat is entrapped in a glassy matrix composed mainly of amorphous lactose and colloidal protein, such that less than 1% of the fat is extractable with organic solvents at room temperature (about 20-25 °C) . This extractable fat is often called " free fat" . For example, extraction of a commercial whole milk powder containing 28.5% total fat with pentane at room temperature for two hours yielded 0.3% free fat. When used in the manufacture of chocolate products, dry milk powder wherein the fat is entrapped leads to undesirably high viscosity in the final product, requiring a greater amount of cocoa butter or emulsifier to standardize the flow properties. This is one reason why roller-dried milk powder has been preferred for chocolate manufacture, i.e., its greater free fat content. However, commercial sources of roller-dried product are few, and roller-dried milk powder is susceptible to oxidative rancidity. For these reasons, spray-dried milk powder is most often used. It has previously been shown that crystallization of the lactose in milk powder results in higher free fat (Aguilar and Ziegler, 1993) , and that lactose from spray- dried milk powder crystallizes in the β form during the process of chocolate conching (Ziegler and Aguilar, 1994) . Conching is a step in chocolate making wherein the initial dried mixture of ingredients is heated, and emulsifiers and more cocoa butter are added to liquify the chocolate and allow the full flavor to develop. The resulting product is smooth and flavorful. Conching is critical to flavor development in ways that are not well understood.

Much has been written about conching. The literature, however, is full of contradictions (Dimick and Hoskin, 1981) and inconsistencies between the published literature and practical observations still abound. For example, it is often stated that a principal outcome of the conching process is dehumidification of the chocolate mass, but the inventors have effectively conched chocolate, reducing the yield value and viscosity, with no noticeable change in moisture content. At a recent workshop, 20 participants, all employed in the confectionery industry, conducted a triangle difference test using conched and unconched (but standardized) milk chocolate and failed to detect a significant difference between the two .

On one point most agree. Conching transforms chocolate mass from a dry powdery aggregate to a homogeneous fluid by dispersing agglomerates. In a 1994 paper presented at Chocolate Technology 1994, the inventors suggested that this deagglomeration and surface wetting was the principal reason for conching dark chocolate and that once accomplished, regardless of the means, the conching process was essentially complete, but that if it were deemed necessary, liquor could be dehumidified and deacidified separately. If Maillard reactions take place, these too could be accomplished by processing the liquor separately. It has since been learned from work on the effect of particle size on sensory properties (Ziegler and Mongia, 1998) , that deagglomeration, through its influence on rheology, can impact flavor and texture. Therefore, it is conceivable that the effective dispersion of particles may be responsible for flavor changes in dark chocolate during conching, even in the absence of any dramatic chemical reactions.

Milk chocolate remains a different story. From previous research (Aguilar and Ziegler, 1993) , it was determined that the physical state of the non-fat milk solids could have a dramatic impact on the flow properties and flavor of chocolate, particularly when those solids are derived from spray-dried powders. Crystallization of amorphous lactose from spray-dried milk powders liberates entrapped milk fat that is then available to become part of the continuous fat phase and reduce viscosity. Furthermore, crystalline lactose is denser than the amorphous powder from which it is obtained and, thereby, reduces the relative particulate phase volume. Niediek (1991) hypothesized that crystallization of amorphous sugar occurs during conching. Since free fat is such an important component in WMP, many researchers have looked into ways to increase it . Two approaches have been taken to increase the free fat in dry milk products. U.S. Patent Nos . 4,871,573 and 5,051,265 describe processes whereby lactose crystallization is seeded during the concentration and drying processes resulting in dried milk with substantially higher free fat. However, only 20-90% of the available fat is freed, and the products are still susceptible to oxidative rancidity. In U.S. Patent Nos. 4,532,146 and 5,672,373, the non-fat portions of the milk are first separated from the milk fat, spray dried, and then recombined with the anhydrous milk fat such that 100% of the fat is free. In fact, for some products, industry practice is to use the skim milk powder (non-fat dry milk) and anhydrous milk fat in the ratio they would be present in whole milk powder. However, anhydrous milk fat is still susceptible to flavor deterioration, and the combination of skim milk powder and anhydrous milk fat does not yield the same quality of flavor as whole milk powder.

For the foregoing reasons, there is a need for a product with the highest free fat content possible (>90%, often 100%) , with the availability and storage of spray-dried milk powder, and retaining the positive flavor attributes of whole milk powder. There is also a need for a process for making dry milk products with such desired characteristics.

SUMMARY OF THE INVENTION An object of the invention is to provide a dry milk product having lactose in essentially crystalline form and containing fat in the free state.

Another object of the invention is to provide a process for producing a dry milk product with the highest free fat content possible.

These and other objects, features, and advantages will become apparent after review of the following description and claims of the invention which follow.

The invention comprises a dry milk product having lactose in essentially the crystalline state and the highest free fat content. Typically the product will contain 20-40% of fat by weight of which 80-100% is in the free state.

The invention also comprises a process for producing dry milk product with the desired characteristics. This product typically contains <10% moisture, and preferably <5% moisture, has lactose in essentially the crystalline state and contains from 20-40% fats by weight of which 80-100% are in the free state. The dry milk is processed at elevated temperature with shear in a mixing or grinding device. The resulting product is beneficial as an ingredient for confectionery, especially chocolate, to reduce viscosity and improve flavor. BRIEF DESCRIPTION OF THE DRAWINGS

Figure 1 shows the X-ray diffraction patterns of feed material to the process (a) and converted product (b) .

Figure 2 shows the particle size distribution of converted milk paste.

Figure 3 shows a response surface plot and three- dimensional view of lactose crystallinity as a function of screw speed and feed rate at process temperature of 71.1 °C (160 °F) . The counter lines with numbers are significantly different at P<0.05. Experimental conditions below the diagonal line were not included in the design.

Figure 4 is a response surface plot of combined lactose crystallinity and free fat as a function of screw speed and feed rate at 71.1 °C (160.0 °F) . The diagonal line shows the design boundary. At these conditions, the expected free fat is 88.4% and lactose crystallinity spike is at 1271 count per second on x-ray diffraction pattern.

Figure 5 is the x-ray diffraction pattern of the unprocessed spray-dried WMP. There is not any characteristic spike. It indicates that the lactose in raw WMP was in amorphous glassy form.

Figure 6 is the x-ray diffraction pattern of the processed spray-dried WMP. The spikes in the pattern indicate that the lactose in WMP is crystallized.

Figure 7 is the particle size distribution of raw WMP. It is a unimodal distribution with volume based average particle size (D_{4 3}) of 137.29 μm.

Figure 8 is the particle size distribution of processed WMP. It is a bimodal distribution with volume based average particle size (D₄₃) value of 50.27 μm. Figure 9 is a system diagram for WMP processing of the present invention.

Figure 10 is a schematic of a fuzzy logic temperature controller for WMP inputs and outputs which could be added to the process of the present invention.

Figure 11 is the fuzzy membership functions for the inputs (a, b, c, d) and output (e) for the controller of Figure 10.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT

The product is a whole milk product with substantially, approaching 100%, free fat and crystalline lactose. Though the form and ratio of crystals (α vs. β) is not critical, the present product has β crystals which is not generally found in existing dry milk products.

The process involves the application of heat and shear forces to a spray-dried milk powder resulting in crystallization of the amorphous lactose and liberation of entrapped fat, which ultimately results in the desired characteristics of higher free fat content and crystalline lactose. The higher free fat content, in turn, results in the desired sensory properties of the end product in which the milk powder is used.

Application of heat, i.e., heating over the glass- transition temperature of the dry milk powder, causes the lactose within the raw material to crystallize. The heat de- vitrifies the glassy matrix and allows the lactose mobility to crystallize. This crystallization liberates entrapped milk fat and reduces the particulate phase volume. The heat should be applied at a level which gives the desired crystallization but does not result in burnt flavors in the product. The heat may be applied extrinsically and/or may be applied intrinsically from the shear (mechanical work input) . Since shear always creates some level of heat, this must be taken into account when applying an extrinsic heat source. The heat may be dry or moist heat . Means of heat control and adjustment of heat depend on the shearing device and/or extrinsic heat source used and are readily known to one of ordinary skill in the art. Examples of extrinsic heat sources include jackets around the shearing device or direct heating of the shearing parts.

Shear is an action or stress caused by an application of forces that causes two parts of a body to slide on each other. Applying shear to the WMP liberates the entrapped fat and disperses it within the product . The shear also is able to maintain the lactose crystals as smaller particle sizes. Heat alone often causes clumping of lactose crystals. Shear should be sufficient to produce the desired crystal size(s) and the desired level of fat liberation.

Suitable shear devices include extruders, high-shear mixers, agitated ball mills, and the like, all of which are readily commercially available and known in the art . For example, Readco Manufacturing in York, PA makes a high-shear, twin-screw mixing device. Shear adjustment is based on the device used and is known to one of ordinary skill in the art. Examples of parameters which affect shear are screw speed, energy input, feed rate, and back pressure.

Shearing and heat can be controlled manually by an operator or by automatic controls.

The introduction of grinding aid can control overheating and improve the grinding and dispersion of lactose crystals. This results in improved process stability and improved handling properties of the product. For example, commercially spray-dried milk powder at 28.5% fat by weight was processed in either a 2" or 5" Readco Manufacturing Continuous Processor. The Continuous Processor is a high-shear, twin-screw mixing device operating at 100-300 rpm. Dry whole milk was fed into the 2" processor at a rate of 4-35 kg/hr using a dry material feeder. The jacket temperature of the processor was maintained at 80-105 °C, i.e., above the glass transition temperature of the milk powder (-70 °C) . There was a noticeable rise in the power requirements of the processor as the lactose crystallized. The mechanical energy input was approximately 75 kW/tonne, once the process stabilized. The product exits the process in a variety of textures, but preferably in a form similar to chocolate mass in the pasty phase of conching. This results because the product is now a dispersion of lactose and milk protein in a continuous milk fat phase, with the approximate fat content of dry- conched chocolate (26.5-28.5%).

When the product was extracted with pentane at room temperature, 90-100% by weight of the fat was in the free state. X-ray diffraction revealed that the lactose had been transformed from the amorphous state to the β crystalline form (Figure 1) .

As the lactose crystallizes, the crystals may grow to sizes greater than 0.5 mm, and at this point, the product looks like wet sand. However, somewhat unexpectedly, the processor is capable of grinding these large crystals to produce a paste with particles less than 100 μm (Figure 2) . This crystallization and grinding generates a significant amount of heat and, consequently, the jacket temperature may be reduced.

The process as described is susceptible to overheating, which may produce burnt off-flavors. However, when controlled, this process can produce caramel -like flavors typical of crumb based chocolates. The overheating can be controlled and the grinding and dispersion of lactose crystals improved by adding a small amount of emulsifier to the process. Adding just 0.3% by weight soy lecithin allowed an increase in throughput from 10 kg/hr to 35 kg/hr while improving the process stability and the handling properties of the product .

Similar improvements were obtained on a 5" processor using a vegetable oil as a grinding aid.

The process can be carried out using the continuous processor with a number of paddle configurations, and even other devices that provide sufficient mixing/grinding action and allow for some means of temperature control, e.g., twin- screw extruders, high-shear mixers, or agitated ball mills.

EXAMPLES

EXAMPLE 1 Grit Formation in Chocolate

To obtain direct evidence of structural collapse and sugar crystallization, the inventors decided to investigate grit formation in over-conched chocolate. Dark and milk chocolate mass and a white coating were dry conched at 95 °C. Initial and final moisture contents are shown in Table 1.

Table 1. Percentage moisture content of chocolate mass.

The dark chocolate actually gained some moisture during the conching process , though probably not enough to be statistically significant. Both the milk chocolate and white coating lost approximately 50% of their initial moisture content .

Only the milk chocolate and white coating became gritty, suggesting some component of the milk powder was responsible. Volume mean particle size, d₄₃ changed little during conching and actually decreased for both the milk chocolate and white coating even though these samples had a gritty mouthfeel. However, a closer look at the particle size distribution revealed increases in particles >90μm for milk powder containing formulas.

Little change was observed in the particle size distribution of dark chocolate as a result of conching. There was a slight decrease in the weight of particles retained in the 20-45 μm and 45-90 μm ranges, with a corresponding increase in the particles below 20 μm, which may indicate mechanical dispersion of agglomerates. For milk chocolate and white coating, there was an increase in very large (>90 μm) and small (<20 μm) particles at the expense of those between 45-90 μm. This explains how a chocolate can become gritty even though the mean particle size decreases.

The sugar composition (ratio of sucrose : lactose) in each fraction was analyzed by HPLC (Table 2) .

Table 2. Sucrose to lactose ratios in milk chocolate fractions .

Prior to conching, particles <45 μm had a higher proportion of sucrose (less lactose) than particles >45 μm, consistent with observations that sucrose undergoes abrasion (surface erosion) during the roll refining process, producing a larger number of particles. After conching, the ratio of sucrose : lactose was nearly the same in all size ranges, i.e., lactose was now distributed evenly through all particle sizes. X-ray diffraction patterns revealed the presence of β-lactose in conched, but not in unconched samples, and scanning electron micrographs showed fine crystalline material adhering to the surface of what appeared to be sucrose particles (Ziegler and Aguilar, 1994) . These observations are consistent with the formation and agglomeration of fine β-lactose crystals that aggregate with sucrose to produce grit. The role of milk proteins is unclear, however, Wursch et al . (1984) proposed that milk proteins play a decisive role in favoring β-lactose crystallization. A glass transition observable in refined milk chocolate and white coating by differential scanning calorimetry was eliminated by conching at 95 °C. Concurrent with this was crystallization of lactose in the β form. No such glass transition was seen in refined dark chocolate, indicating that within the sensitivity of the method no amorphous sugar was present. It was assumed that the changes observed in these excessively conched samples were simply exaggerations of those changes that occur in properly conched chocolate and concluded that for milk powder-based milk chocolate and coating containing amorphous sugar, a glass transition and collapse of the lactose matrix occurs during conching. Furthermore, it was concluded that it was not necessary to bring about these changes in the milk component in the presence of the remaining chocolate ingredients and that the milk powder could be processed separately.

EXAMPLE 2 Process Conditions for 2 " diameter screw continuous processor

Materials and Methods

Spray-dried whole milk powder (WMP) was processed using a continuous processor to increase the functional properties in chocolate and confectionery manufacture. The powder was purchased from Agri-Dairy Products, Inc. (Purchase, NY). Typical composition of the raw WMP was as follows:

Table 3 Typ owder .

A 2" twin-screw co-rotating continuous processor, manufactured by Readco Manufacturing, Inc. (York, PA), was used to process the spray-dried WMP. A control panel displayed the thermocouple sensors' readings and the power consumption on the processor barrel.

An AccuRate" gravimetric single screw powder feeder fed the WMP to the processor. The control of the feeder was accomplished using a microcontroller provided with the feeder.

A circulating water heater, manufactured by Mokon Inc. (Buffalo, NY) , was used to warm the components up and control the process temperature.

Paddle configuration in the processor

The continuous processor is a twin-shaft co-rotating mixer. It provides homogenous mixing, shearing, kneading, and crystallizing of one or more dry materials with one or more liquid materials. The shafts are driven by a 3.728 kW (5 hp) electric motor. The mixing, shearing, kneading, and crystallizing are accomplished with the help of paddles or the processing elements positioned on the shafts in the processor. There are three types of paddles: forward helix, reverse helix, and flat. Forward helix paddles have conveying, reverse helix paddles have mixing and shearing, and flat paddles have shearing and mixing effects on the material being processed. The paddles and feeding screws were configured in such a way that lactose crystallization could take place in a uniform manner while providing stable processing conditions (Table 4) . The discharge gate position is at paddle numbers 1-3, and the powder enters at paddles numbered 25-27. Table 4. Paddle configuration inside the continuous processor barrel

RH= reverse helix, FS feeding screw, F = flat, FH = forward helix Experimental design and sample collection

To determine the effects of the processor screw speed, feed rate and process temperature on WMP properties, an experimental design using a quadratic model was created using ECHIP experimental design software (ECHIP Inc., Hockessin, DE) . There were three control variables in the model : the processor screw speed in the range of 240-290 rpm, barrel temperature in the range of 54.4-71.1 °C (130-160 °F) , and powder feed rate in the range of 11.3-18.1 kg/hr (25-40 lbs/hr) . The design provided 20 experiments, 15 of which were unique combinations of the control variables and 5 of which were replications (Table 5) . The statistical analyses of the data were also accomplished using ECHIP.

Before the experiments were started, the processing elements, shafts and barrel temperatures were warmed up to 85 °C (185 °F) , the screw speed was set to 200 rpm and the feed rate was set to 4.5 kg/hr (10 lbs/hr) . The product was visually and sensory inspected. Once the processed WMP had a dough-like pasty appearance with a color change from pale cream to bright yellow, the ECHIP experiments were conducted in the provided order (Table 5) . For each experiment, samples were collected when the power and barrel temperature stayed constant at the desired feed rate and screw speed. The samples were stored in plastic bags for the analyses of free fat content, lactose crystallinity, particle size distribution, color and moisture content. Power consumption, raw powder temperature, and temperature at the discharge gate were recorded before the samples were collected.

Table 5. Experimental design used to increase the free fat content of the spray-dried WMP. Screw s eed (SS) , feed rate, and temperature were used as the control variables.

^D _,3 ⁼ D_4ι3 values of the particle size

L = CIELAB lightness value, a* = CIELAB redness value, b* = CIELAB yellowness value

Analyses

Free fat : Free fat content was determined suspending 2 g of sample in 15 mL of HPLC grade pentane for 2 hrs . The suspension was stirred until all insoluble particles were well dispersed. After initial suspension, every 30 min. the solution was stirred for 15 sec. by shaking the flask. After two hours, 12 mL solution was placed in Pyrex^® centrifuge tubes and centrifuged until the solid parts were separated from the liquid. Using a pipet, 2 mL of clear supernatant were taken and placed in weighed and labeled aluminum cups and dried in a convection oven at 80 °C (176 °F) for 2 hrs. The cups with fat were weighed at room temperature. From the known total fat content of the WMP and the weight of the extracted fat, the free fat content was determined as a percentage of the total fat content.

Lactose crystallinity: X-ray diffraction patterns of the samples were determined between 10° and 30° (Aguilar and Ziegler, 1994) using an automated x-ray diffractometer (Rigaku Denki , Co. Ltd., Tokyo, Japan) . Particle size distribution: The MasterSizer laser light-scattering particle size analyzer was used to determine the particle size distribution of the products. The analyzer had a MS 15 sample presentation unit and 300 mm lens (Malvern Instruments Ltd. , Malvern, England) . The processed WMP was dispersed in isobutanol (Fisher Scientific, Pittburg, PA) at room temperature until the obscuration value was reached to 0.2, corresponding to a volume concentration of 0.03-0.05% (Aguilar and Ziegler, 1994) . Mechanical stirring and ultrasonic dispersion (50 W at 27 kHz for 2 min.) were applied for dispersing the particles independently.

Color: The CIELAB L* , a*, b* (lightness, redness, and yellowness) values were determined using reflectance spectrometry (Minolta Spectrophotometer, CM-3500) .

Moisture: Moisture contents of the processed and raw WMP were determined drying 2 g samples at 50 °C in an oven for 14 hrs. Similar analyses on raw milk powder samples were also conducted to compare the processed and raw WMP properties (Table 6) .

Results

The effects of screw speed, feed rate, and process temperature on spray dried whole milk properties were determined using a response surface experimental design. The analyses on the processed WMP included the lactose crystallinity, free fat content, particle size distribution, color and moisture content. Results of the experiments are in Table 5. Screw speed in run 3 had to be changed from 240 rpm to 255 rpm because at 240 rpm and 18.1 kg/hr (40 lbs/hr) feed rate the electric motor could not provide enough torque and the safety switch on the electric motor stopped the process. The process conditions providing the desired spikes on x-ray diffraction patterns showing the lactose crystallinity were found to be 13.8 kg/hr (30.5 lbs/hr) feed rate, 266 rpm screw speed, and 71.1 °C (160 °F) process temperature. The temperature had a significant linear effect on the lactose crystallinity (P<0,01). Screw speed and feed rate also showed significant quadratic effects on lactose crystallinity at P<0.05. The response surface plot of lactose crystallinity as a function of screw speed and feed rate at 71.1 °C (160 °F) is shown in Figure 3.

Screw speed and feed rate interaction and quadratic term of screw speed showed significant effects on product moisture content at P<0.05. Linear effects of screw speed and feed rate on power consumption were significant (P<0.001). Feed rate and temperature interaction also showed a linear effect on power consumption at P<0.05.

None of the control variable changes produced statistically significant free fat content differences. Optimal conditions providing the maximum combined responses of free fat content and lactose crystallinity were 14.7 kg/hr (32.5 lbs/hr), 265 rpm, and 71.1 °C (160.0 °F) . The response surface plot of combined lactose crystallinity and free fat responses as a function of screw speed and feed rate at 71.1 °C (160.0 °F) is shown in Figure 4.

The maximum torque and the screw speed that the electric motor could provide were 3.728 kW (5 hp) and 320 rpm, respectively. Taking these limits into account and from preliminary experiments, the maximum feed rate and screw speed were determined to be 18.1 kg/hr (40 lbs/hr) and 290 rpm, respectively. Exceeding these limits resulted in machine failure.

Discussion Preprocessing spray-dried WMP increased the free fat content from less than 10% to over 95%. Spray-dried WMP with high free fat content would reduce cocoa butter addition during conching. This would result in reduction in production costs since the cocoa butter is the most expensive ingredient in chocolate.

Response surface experiments showed that temperature had a significant effect on lactose crystallinity at P_<0.01. Heating resulted in crystallization of the lactose in the WMP in both α- and β-crystalline forms. The x-ray diffraction patterns of the processed WMP shows the characteristic spikes of α- and β-crystalline forms (Figure 5) . On the other hand, the x-ray diffraction pattern of raw WMP did not have any characteristic spikes of the lactose crystalline forms (Figure 6) . This indicates that the lactose in spray-dried WMP was in amorphous glassy form.

Particle size distribution of the WMP was improved with processing. The particle size distribution of the processed and raw WMP are shown in Figure 7 and Figure 8. Particle size distributions of the raw WMP samples show unimodal distributions with volume based average particle size (D₄₃) values of 137.29, 140.28, 139.45 μm (Table 6). On the other hand, processed WMP showed a wider and close to a bimodal particle size distribution, with smaller D_{4 3} values. Preprocessing not only reduced the average particle size of the WMP, it also provided a bimodal particle size distribution which is preferable for chocolate manufacturing (Stauffer, 1998) . Having a bimodal particle size distribution helps to reduce the cocoa butter use. Since the smaller particles fill the voids created by larger particles, packing efficiency is increased with bimodal distribution (Stauffer, 1998) . If the particles are packed well, the surface area that needs to be covered with the continuous fat phase gets smaller. This results in less cocoa butter use and reduction in manufacturing costs.

The color analyses of the raw and processed milk powders were also conducted using reflectance spectroscopy . The three color-reflectance values, CIELAB L* , a*, and b* were obtained for both the processed and raw WMP (Table 5 and 6) . After processing, the color of WMP turned to bright light yellow.

Processing made little change in the moisture content of the WMP, although there was some increase in the moisture content of the processed WMP. This could be due to the structural change that occurred in the particles. Since the spray-dried WMP particles usually have air vacuoles, processing must have released the air and increased the moisture content of the WMP. A better removal of moisture from the oven as the samples dry, would provide more accurate results. The moisture contents of the processed and raw WMP are shown in Tables 5 and 6. EXAMPLE 3

Optimal processing conditions for 2" continuous processor for maximum free fat content

Materials and methods

Spray-dried WMP with minimum 28% total fat content and less than 3% moisture content was processed. A 2" diameter, 5 hp, pilot-scale co-rotating twin-screw continuous processor was used to process the WMP. A gravimetric powder feeder was used to feed the powder into the continuous processor. The maximum capacity of the powder feeder was 36.2 kg/hr (80 lbs/hr) . A circulating water heater was used to provide heating and cooling for the processor barrel. A peristaltic pump was used to inject lecithin into the processor.

Data collection and experimental design

Response surfaces method was used to design the experiments. For the experimental designs, ECHIP (ECHIP Inc., Hockessin, DE) experimental design software was used. The designs provided the minimum number of the combinations of four control variables. There variables were the process temperature, processor screw speed, powder feed rate, and lecithin injection. Two separate experimental designs were conducted. For both designs, the same control variables in different ranges were used. The first experimental design served as a screening design. The ranges of the control variables for the two designs are shown in Table 7.

Table 7. Control variables and ranges that were used in the experiments

The two separate experimental designs provided 54 combinations of the four control variables. For each combination of the control variables, experiments were conducted and samples were collected. Steady state was considered to be reached when the control variables remained constant at the desired conditions. The collected samples were analyzed to determine the product free fat content, average particle size distribution and lactose crystallinity, CIELAB L*, a*, and b* values, and moisture content. Free fat content of the product was determined by suspending 2 g of processed WMP in 15 mL of HPLC grade pentane for 2 hrs. Every 30 min. the solution was stirred for a few seconds. The solutions were centrifuged to separate the solid parts from the solution. Two mL solution from each sample was taken and the pentane was evaporated in a convection oven at 94 °C for an hour. The amount of lecithin added into the process was subtracted from the final weight and free fat content was determined. MasterSizer" laser light-scattering particle size analyzer with 300 mm lens was used to measured the particle size distribution. The WMP was dispersed in isobutanol at room temperature. Ultrasonic dispersion and mechanical stirring were applied for 1.5 min. to ensure the independent particle dispersion. An automated x-ray diffractometer was used to determine the x-ray diffraction pattern of WMP (lactose crystallinity) between 10 and 30°. The CIELAB L* , a*, and b* (lightness, redness, and yellowness) values were determined using reflectance spectrophotometry (Minolta Spectrophotometer, CM- 3500d) . moisture contents of the processed WMP were determined drying 2 g of samples at 50 °C in an oven for 14 hrs .

Neural networks analysis

The neural network analysis was conducted using NeuroShell 2 commercial software (Ward Systems Group, 1993) on an IBM computer with a Pentium processor. The four control variables of the powder feed rate, processor screw speed, process temperature, and amount of lecithin injection were used as the network inputs. The free fat content, lactose crystallinity, average particle size distribution, CIELAB L* , a*, and b*, moisture content, and power consumption were used as the network outputs. A total of 54 patterns of data was obtained from the two experimental designs (Table 8) . Twenty percent of this data was randomly allocated as a test set and 80% was used as training set. A three-layer general regression neural network (GRNN) architecture (Specht, 1991; Caudill, 1993) was used to train the network. There were 4 neurons (four inputs) in the first layer, 44 neurons in the middle layer, and 8 neurons in the output layer (eight outputs) . A linear scaling function was used to scale the data. Unlike backpropagation training, there are no learning rate and momentum parameters in GRNN, but there is a smoothing factor that is determined with test data.

Table 8. WMP processing data set used for network training.

IN_¬

CH FR = feed rate, SS = screw speed, D_{4 3} = avg. particle size, Crys. = crystallinity, CPS count per second

Training of GRNN includes assigning a neuron for each pattern in the pattern layer. The weights between the pattern layer neurons and the input layer neurons are equal to the inputs of the training patterns . The outputs of the training patterns are assigned as the weights between the pattern layer neurons and the summation layer neurons. The number of neurons in hidden layer in GRNN is equal to the number of training patterns. The number of summation neurons is equal to the number of outputs . Whenever new inputs are fed to the trained network, the distances between the new inputs and all the training pattern inputs are calculated using either Euclidean or city block distances. Euclidean distance is the square root of the sum of squares of the new input and the trained input values . City block distance is the absolute difference between the new input and the trained input values . The pattern neurons sum the distances and feed to a nonlinear activation function. The outputs of the pattern neurons are multiplied by the corresponding weights and summed by the summation neurons in the third layer. The summation neuron outputs divided by the sum of the pattern neurons provide values to the output layer. The results of the output neurons are the estimates for the new inputs.

The trained network was saved as a file that could be accessed via Dynamic Link Library (DLL) in a Microsoft Excel 97 spreadsheet. This trained network served as the genetic algorithms' fitness function over which the input variables are evaluated to reach the optimal solution.

Genetic algorithm search A genetic algorithm was used to search for process conditions providing near optimal free fat content, lactose crystallinity, and average particle size under the constraints. The optimization problem can be summarized as follows .

Goal : maximum free fat content of the processed WMP Sub goals: maximum crystallinity and minimum average particle size Constraints: 75 < L* (lightness) < 85 a* (redness) < 3 b* (yellowness) < 35 moisture content < 2% power consumption < 2.6 kW

The criteria for determining the constraints were based on the visual assessments of the samples from each experiment and the observations made during sample collections. The redness (a*) values greater than 3 were not desirable, since the color of the product was close to undesirable brown color with a burnt smell. Even though the electric motor could provide up to 3.7 kW, the machine was not operating properly above 2.6 kW, so the power consumption was included m the problem as another constraint. The ranges for the inputs that the genetic algorithm search applied were as follows:

Powder feed rate: 4.35-27.21 kg/hr (10-60 lbs/hr)

Processor screw speed: 250-310 rpm

Process temperature: 87.8-110 °C (190-230 °F)

Lecithin injection: 0.058-0.45 kg/hr (0.13-1.0 lbs/hr)

The free fat content was represented with a 32 -bit chromosome. Each 8-bit section of the chromosome represented one input variable. An initial population with randomly created 300 chromosomes was assigned. Each individual chromosome m the population provided a solution that may not be the best solution. These randomly assigned chromosomes were decoded to their corresponding decimal numbers and their goodness was evaluated with the fitness function (the trained neural network) . Cross over rate of 0.9, mutation rate of 0.01 and generation gap of 0.98 values were chosen. Generation gap is the percentage of the chromosomes m the current population being m the next generation. Since there were 300 chromosomes in the initial population, only six (300*0.02) of them had a chance to be m the next generation. If the best fitness value was not changed after 50 generations, the search was stopped and the results were displayed. The genetic algorithm search was conducted using GeneHunter™ (Ward Systems Group Inc., Frederick, MD) commercial software .

Results and discussion

The GRNN training was based on a total of 54 patterns. Eighty percent of the patterns were training and 20% of the patterns were the test patterns. The training provided the coefficient of determination values (R²) (Ward Systems Group, 1993) of 0.71 for free fat, 0.82 for crystallinity, 0.64 for average particle size, 0.91 for L* , 0.76 for a*, 0.88 for b*, and 0.86 for moisture content. The R² values for free fat content and average particle size might have been considered to be too low. However, the neural network's success depends on the goodness of the data and the number of patterns. Echip statistical experimental design software provided the minimum number of experiments. In the experimental design for four control variables, five replications were chosen. The differences in the responses of those replications were quite large, which created large disturbances in the network training .

To compare GRNN results, another network using backpropagation (BP) learning was also used for training. A three-layer 4-13-8 BP architecture was chosen. The learning rate, momentum, and initial weights were set as 0.1, 0.1, and 0.3, respectively. The network training was stopped when the minimum average error was not changed after 20000 passes. The coefficient of determination values for free fat content, lactose crystallinity, average particle size, L* , a*, b*, and moisture content 0.53, 0.77, 0.48, 0.84, 0.92, 0.74, 0.66, and 0.83, respectively. The R² values for free fat from the three-layer BP algorithm was about 20% less than the R² values from the GRNN training. The trained GRNN was used as objective function to determine optimal processing conditions providing the highest free fat content, minimum average particle size, and maximum lactose crystallinity while not violating the redness (a*) value. There were little changes in the lightness (L*) and yellowness (b*) values of the processed WMP compared to the raw EMP. Under the constraints discussed in the material and methods, the genetic algorithm resulted in the following optimal processing conditions and the corresponding expected responses . Optimal Process Conditions

Feed rate 24.6 kg/hr (54.4 lbs/hr)

Screw speed 264 rpm

Process temp. 107.7 °C (226 °F)

Lecithin 0.10 kg/hr (0.23 lbs/hr)

Expected responses under the above conditions.

Free fat content 95.7%

Lactose crystallinity 1256 CPS

Average particle size 52.9 μm distribution (D_{4 3})

L* 75.2 a* 3.0 b* 14.8

Moisture content 1.0%

Power consumption 1.5 kW (2.0 hp)

Fuzzy logic controller design Like in the extrusion processes, milk powder processing requires continuous monitoring and control to maintain the operation in these optimal conditions. Variations in the WMP (moisture, density, particle size, etc.) and the interactions between the operating variables make the operation difficult. A human operator usually monitors and makes the decisions of keeping the barrel temperature, screw speed and powder feed rate stable. Among the operating variables, process temperature is the most important variable that is sensitive to the raw material variations. Even small changes in the process temperature affect the product quality and sometimes cause the process to fail.

Using the power consumption and temperatures on the processor as input variables, a fuzzy logic controller is proposed to control the heater and the cooling solenoid valve to maintain the process temperature at allowable ranges. The WMP processing system diagram is shown in Figure 9. The inputs to the fuzzy logic controller will be the temperatures measured at the processor discharge gate and in the middle, optimal temperature and power consumption. The outputs of the controller will be the signals controlling the heater and the solenoid valve for the circulating water.

Power consumption is a good indicator of the system stability. Even if the thermocouple sensors indicate that the product temperature is in the allowable range, due to the variations in the raw material, the temperature of the powder on the paddles may be higher than the sensor readings. When the paddle temperature gets too high, the powder burns on them. This increases the friction between the material and the paddles as well as the barrel wall. Increased friction requires more torque to keep the screw speed in the set level so the power consumption increases.

The inputs and outputs of the fuzzy logic controller for the temperature control system are shown in Figure 10. The inputs of the controller are the error (E) between the optimal desired temperature and the Temperature 1, change in the error (ΔE) and the difference between Temperature 1 and Temperature 2 (ΔT) . Product temperature at the gate is usually about 4.4 °C (10 °F) less than temperature in the middle of the processor. The fuzzy logic controller of the WMP system design objective is to keep the process temperature as close to the optimal processing temperature of 107.7 °C (226 °F) as possible in spite of the variations in the raw WMP properties. The universe of discourse (range) values for the inputs and outputs of the fuzzy logic controller are shown in Table 9. Table 9. Universe of discourse values for the inputs and out uts ,

The inputs and outputs membership functions and corresponding labels are shown in Figure 11.

Since the solenoid valve had two positions (on and off) there is no need to define membership functions for the output " valve" . The fuzzy rule base will be developed using the Combs Method (Combs and Andrews, 1998) . One of the difficulties in fuzzy logic control design is the development of a rule base. As the number of input and output variables and the number of membership functions per variable increases, the number of rules increases exponentially. The Combs method eliminates relating all the inputs to each other before applying them to the output (Andrews, 1997) . One of the requirements of to apply the Combs method is to have an equal number of membership functions per input and output. Whenever the crisp input values are read from the sensors, with fuzzification the corresponding membership values for every input variable are calculated. Then, the Combs method computes the fuzzy outputs relating each input range to its corresponding output range like P (small) to H (off) , E (small) to H (off) , and so on. Then, the center of gravity method will be used to defuzzify the outputs. Finally, a control signal is sent to adjust the heater position. Conclusions

The process conditions providing the highest free fat content (95.7 %) were determined to be 24.5 kg/hr (54.0 lbs/hr) feed rate, 265 rpm screw speed, 108.0 °C (226.5 °F) , and 0.104 kg/hr (0.23 lbs/hr) lecithin addition.

Having described the invention with reference to particular compositions, theories of effectiveness, and the like, it will be apparent to those of skill in the art that it is not intended that the invention be limited by such illustrative embodiments or mechanisms, and that modifications can be made without departing from the scope or spirit of the invention, as defined by the appended claims. It is intended that all such obvious modifications and variations be included within the scope of the present invention as defined in the appended claims. The claims are meant to cover the claimed components and steps in any sequence which is effective to meet the objectives there intended, unless the context specifically indicates to the contrary.

Claims

What is claimed is:

1. A whole milk product comprising whole milk powder wherein milk fat is about 80 to about 100% in a free state and wherein the product contains crystalline lactose.

2. The whole milk product of claim 1 wherein the crystalline lactose contains β-form lactose crystals.

3. The whole milk product of claim 1 wherein the product contains about 20 to about 40% by weight fat and wherein the fat is about 80 to about 100% in the free state.

4. A process for producing a dry whole milk product comprising shearing with heat at a controlled temperature a feedstock comprising dry whole milk wherein the temperature is controlled at a level at or above the glass transition temperature of the dry whole milk.

5. The process of claim 4 wherein the shearing is done using a twin screw mixing device.

6. The process of claim 4 wherein the shearing is done using a high-shear mixer.

7. The process of claim 4 wherein the shearing is done using an agitated ball mill.

8. The process of claim 4 wherein the dry whole milk is spray-dried.

9. The process of claim 4 wherein the feedstock further comprises a processing aid.

10. The process of claim 9 wherein the processing aid is an emulsifier .

11. The process of claim 10 wherein the emulsifier is lecithin.

12. The process of claim 11 wherein the lecithin comprising about 0.3%.

13. The process of claim 9 wherein the processing aid is vegetable oil .

14. The process of claim 4 wherein the process is continuous .

15. The process of claim 4 wherein the process is batch.

16. The process of claim 4 wherein the temperature is controlled at a level at or above the glass transition temperature of the dry whole milk and below that temperature at which the product will have a burnt flavor.

17. The process of claim 4 wherein the dry whole milk is heated to the controlled temperature and the temperature is maintained before shearing.

18. The process of claim 4 wherein the temperature is controlled simultaneously with shearing.

19. The process of claim 4 wherein the heat is applied intrinsically from the shearing.

20. The process of claim 4 wherein the heat is applied extrinsically .

21. The process of claim 4 wherein the shearing is sufficient to maintain a desired size distribution of lactose crystals and to liberate the desired level of fat.

22. The process of claim 5 wherein the temperature is about 80 to about 110°C.

23. The process of claim 5 wherein the mixing device is operated at about 100 to about 300 rpm.

24. The process of claim 5 wherein the mixing device has a mechanical energy input of about 75 kW/ton.

25. The process of claim 5 wherein the mixing device has a 2" diameter screw and the screw operates at about 265 rpm; wherein the temperature is about 108 °C; and wherein the feedstock is fed at about 24.5 kg/hr and further comprises about 0.1 kg/hr lecithin.

26. A dry milk composition produced by the process of claim 4.