FUZZY CONTROL SYSTEM IN DRYING PROCESS ...

2 downloads 0 Views 121KB Size Report
Aug 25, 2004 - agua sobre la textura del jamón curado, II Congreso Mundial del Jamón Curado, Cáceres, Spain. Stawczyk, J., Comaposada, J., Gou, P. and ...
Drying 2004 – Proceedings of the 14th International Drying Symposium (IDS 2004) São Paulo, Brazil, 22-25 August 2004, vol. B, pp. 919-926

FUZZY CONTROL SYSTEM IN DRYING PROCESS OF FERMENTED DRY-CURED SAUSAGES

Pere Gou1, Josep Comaposada1, Elvira Serra2, Montserrat Corominas3, Manel Poch2 and Jacint Arnau1 1. Centre de Tecnologia de la Carn - IRTA, Monells, Spain, E-mail: [email protected] 2. Facultat de Ciències, Universitat de Girona, Girona, Spain 3. Casademont S.A., Sant Gregori, Girona, Spain Keywords: drying, sausages, fuzzy control ABSTRACT Fermented sausages are dried in natural or artificial driers. Air relative humidity (RH) and temperature (T) are regulated in artificial driers. Nowadays the set points of RH and T are manually readjusted in a staggered form by an expert on the basis of product evaluation and on his own experience. The main drawbacks of this control system are the discontinuity of the product evaluation and the cost of the experts’ training. This work proposes an approach based on the theory of fuzzy sets to readjust the set points of RH and T with the help of on-line instrumental measurements related with the expert evaluation. The input parameters evaluated for the fuzzy control system were: an estimated water activity at the product surface (from heat balance) and the working time of the fan drier. The output parameters were the set points of RH (low and high). The fuzzy control system was applied to the drying process of fermented sausages in an industrial drier at constant temperature. Results show that this system can improve the manual control system and it is helpful for quality assurance during the drying process. It reduces the risk of crust development, one of the most important defects in fermented sausages. INTRODUCTION Drying of cured meat products has been made traditionally by natural systems, taking advantage of a geographic zone or a period of time of the year. In the last decades, these systems have been replaced by Artificial Systems where recirculation, relative humidity and temperature of drying air are regulated. To obtain a high quality dried product, a precise control of many process parameters is required. In most drying installations it is necessary to control inlet and outlet parameters of a drying agent and the material being dried. Nowadays the set points of air relative humidity and temperature are manually

919

readjusted in a staggered form by an expert on the basis of the product evaluation (weight loss and sensory characteristics) and on his own experience. Sensory characteristics used are: aspect (humid or dry aspect), fungi development and surface texture (stickiness or crust development). Hardness or crust development is the most critical parameter. The main drawbacks of this control system are the discontinuity of the sensory evaluation and the cost of the experts’ training. On-line instrumental measurement of parameters related with the expert evaluation could be used to improve this control system. Moisture content and water activity have been related with texture parameters in dry cured ham (Ruiz-Ramírez et al., 2003). Stiebing and Rödel (1992) evaluated the water activity (aw) at the surface in sausages and it was proposed as a criterion for the regulation of the relative humidity of the drying air. However, the equipments used to measure the aw at laboratory are based on equilibrium conditions, and they are not valid for on-line measurement of aw at the surface. Recently, infrared (IR) thermometry has been used to measure the temperature at the surface of different meat samples and an on-line estimate of aw at the surface (aws) has been determined (Gou et al., 2003; Gou et al., 2004). According to this methodology, under some hypothesis, aw at the surface can be estimated from heat balance if relative humidity and temperature of air and temperature at the product surface are known. However, the relationship between texture parameters and aw is not lineal (Ruiz-Ramírez et al., 2003) and, moreover, real drying processes are characterized by complex dynamics. Drying process parameters are full of uncertainties due to complexity of transfer laws. Therefore, the identification of a correct drying process is not a simple task, because parameters cannot be easily found, since they vary during the process (Corrêa et al., 2000). For the last decade fuzzy logic has proven its worth as a practical engineering and problem-solving tool. The fuzzy logic is ideal for modeling and controlling complex, non-linear systems because it systematically handles ambiguity. In complex drying technology it is often impossible to formulate a physico-chemical model of the process, so adaptive fuzzy modeling seems to be an alternative solution in such cases. The Adaptive Neuro-Fuzzy Interface system (Matlab, 2000) allows us to shape membership functions by training them with input data. The learning data should be gained by measurement of drying process (input-output behavior of a process) or generated artificially (calculation by algorithm, relationships from knowledge based on experience, etc.). Stawczyk et al. (2003) have used a fuzzy control system in meat drying experiments. However this system focused only in the air conditions control and not in the product control. The main objective of this study is to evaluate a fuzzy control system for the drying process of dry cured sausages in an industrial dryer. METHODOLOGY Fuzzy control system The fuzzy logic control system needs to specify the number of input and output variables and create the shape of membership functions to define the degrees of truth for each variable in the system. Next, the fuzzy rules and defuzzification technique must be specified. The input variables used in this study were those on-line measurements that have been related with the evaluation criteria used by the expert technicians of the industrial dryer: • aws: the water activity at the surface estimated from heat balance, related with aspect, fungi development (Leistner et al., 1981) and texture at the surface (Ruiz-Ramírez et al., 2003). The aws determination is sensitive to the air RH and T variations and a filter function is needed if this parameter is used in control (Gou et al., 2003). Therefore, a moving average for periods of 6 hours was used.

920



Rt: percentage of time in a period of 6 hours that the fan in the drier is working, which affects the homogeneity of the drying process within the drier and the difference between the air RH and aws (x100). For example, different aws can be achieved with the same air RH by modifying the Rt. Digital inputs were transformed into equivalent fuzzy inputs by using a triangle/trapezoid fuzzy membership function. They were divided into three fuzzy sets for Rt input and into five for aws input. Knowledge from a previous experiment developed in a tunnel drier in a pilot plant (Gou et al., 2004) was used to define the initial membership function for aws. The industrial technicians’ experience was used to define the initial membership function for Rt. Both membership functions were redefined according to the results of the first two drying batches of this study. The manipulated variables (outputs) supplied the increments in the air RH set points (low and high) to the drier control system. Drying experiments Four batches of sausages were dried in an industrial dryer of a Spanish meat product manufacturer (Casademont S.A.) under fuzzy logic control system. A commercial product was used. Meat from shoulders and bellies was ground and mixed with common additives to obtain a fat:protein ratio of about 1.3 in the final product. The stuffed sausages (diameter 32 mm) were hung in a chamber for fermentation. After fermentation they were maintained at 14-16 ºC and at relative humidity of 82-88% until the desired fungi development at the surface was reached. Thereafter, the drying process was performed in a drier with the following dimensions: 5 x 8 m2 of surface and 3.8 m of height. The drying air RH and T were controlled by a drier control system that commanded a central air conditioning unit. The low and high set points of air RH used by the drier control system during the drying process were supplied by the fuzzy logic control system. The set point variations were rounded to integer values. The initial values A were selected by the industrial technicians according to the Figure 1 - Drier section. A: location of product characteristics at that moment and to their experience. monitored samples. The weight loss of a sausage in the same zone of aws determination (Figure 1) was monitored at different time intervals. The aws determination The aws was determined on a sausage located in the lower level of the drier, under the lateral tubes whereby dry air enters the drier (Figure 1). The temperature at the sausage surface during drying was determined by an infrared sensor Raytek (RAYMICO2LT) calibrated at 24.6 ºC using an emissivity coefficient of 0.95. The accuracy of the measurement was + 1 ºC. At the same time, the air RH and temperature at 5 cm from the sausage surface was also determined by humidity and temperature sensors (Testo hygrotest 600, + 2% RH, + 0.3 ºC). Temperature at the sausage surface and air temperature and RH were logged every minute during the drying process. Assuming that, due to low drying rate, accumulation of heat in meat is negligible, the aws could be estimated from the heat balance by equaling the sensible heat flux to the enthalpy flux of evaporating water. For convection dryers, the rate of heat transfer from air to the surface is given by: Q = α ⋅ A ⋅ (Tb − Ts ) (1) It is also assumed that there is no temperature gradient in the solid phase. The transfer rate of heat used to evaporate water from the surface is given by:

921

(

)

Q'= λ ⋅ φ = λ ⋅ k c ⋅ A ⋅ Y * − Y (2) Assuming that all heat is used to evaporate water from the surface and that no heat is gained or lost by the surroundings, equations (1) and (2) would be equal: α (Tb − Ts ) = λ ·k c ·(Y * − Y ) (3) On the other hand, from the Lewis relationship we have: α (4) = C H = C dry −air + C steam ⋅ Y kc By applying (4) in (3) the following equation is obtained: (C dry − air + C steam ·Y )·(Tb − Ts ) = λ ·(Y * − Y ) (5) Y* can be calculated from (5) and aws can be obtained from equation (6): a ws ⋅ P0 (6) Y* = 0.622 ⋅ P − a ws ⋅ P0

RESULTS AND DISCUSSION After the first two drying batches, the final membership functions for aws and Rt were defined. They are shown in Figures 2 and 3 respectively.

Figure 2 - Membership function for aws.

Figure 3 - Membership function for Rt.

According to these figures the optimum aws during the drying process of this kind of sausages is 0.78, with an Rt value of 30%. Low aws tended to develop a crusted surface. High aws tended to produce an

922

excessive fungi development and much too slow the drying process. A low Rt, for high aws, produced an unacceptable heterogeneity within the drying batch, because of the slower drying rate in the ‘humid’ zones of the drier. A high Rt, for low aws, accelerated the crust development in the sausages closest to the entry tubes for dry air. These results together with the technicians’ experience were used to define the final fuzzy rules (Table 1). The output variables indicate what increase or decrease would be necessary to apply to the RH set points (the minimum and the maximum RH). Table 1. Fuzzy rules.

Inputs

Outputs

aws

Rt

Low set point for RH

High set point for RH

Very low Very low Low Low Medium High High Very high Very high

Medium or high Low Medium or high Low All High Low or medium High Low or medium

High increase Low increase Low increase No variation No variation Low decrease Low decrease Low decrease High decrease

High increase High increase Low increase Low increase No variation No variation Low decrease Low decrease High decrease

These outputs were transformed to real variables with triangular fuzzy membership functions by the Center-of-Gravity method, according to the following equivalents: • High increase = + 2 (%) • Low increase = + 1 (%) • No variation = 0 • Low decrease = - 1 (%) • High decrease = - 2 (%) Figure 4 shows an example of fuzzy rules application. In this case for inputs Rt=25% and aws=0.76 the transformed outputs showed an increase of 0.71 points in the high set point of RH (HRmax) and an increase of one point in the low set point of RH (HRmin). After rounding to the integer values, the variation applied to the drier RH set points was an increment of one point in both low and high set points.

Figure 4 - Example of fuzzy rules application.

923

Figures 5 and 6 show the monitored air conditions (temperature, RH and Rt), product characteristics (weight loss and aws) and the RH set points for batches 3 and 4 respectively. Batch 3 began the drying process with higher weight loss (about 9%) and technicians decided to use a higher initial high RH set point (82%) than that used in batch 4 (78%), where the initial weight loss was about 3%. Nevertheless, the drying process applied in both batches, under the fuzzy logic control system, produced fermented drycured sausages with a correct fungi development at the surface and without crusting defect. BATCH 3

90

27

80

24

RH

70 Rt or RH (%)

30

60

21 18

T

50 40

15 12

30

Rt

20

9 6

10

3

0

0 0

50

100

150

Weight loss (%) or T (ºC)

100

200 Hours

RH setpoints (%)

aws (%)

Weight loss

Figure 5 - Drying process for bath 3.

The duration of the drying process was slightly longer than those without fuzzy logic control system (RH set point modifications according to technicians experience). The importance of this difference has to be quantified from an economic point of view, but also in terms of management problems. Therefore, if this extra time required to obtain the same product is important enough, it would be advisable to include a new input in the fuzzy control system related with weight loss rate. The fuzzy control system applied in this study is a simple one and it was helpful in obtaining fermented dry-cured sausages of good quality, but it is possible to improve it by adding new input variables such as the energy cost. It is also important to state that the membership functions are specific for this product (with a determined composition and previous treatments) and this industrial drier (with specific dimensions and product disposition). To apply this control system to a new product or to another industrial drier, new membership functions would have to be defined.

924

BATCH 4

100

30 27

80

24 RH

Rt or RH (%)

70 60

T

50

21 18 15

40

12

30

9

20

Rt

10 0

6

Weight loss (%) or T (ºC)

90

3 0

0

50

100

150

200 Hours

RH setpoints (%)

aws (%)

Weight loss (%)

Figure 6 - Drying process for batch 4.

CONCLUSIONS A fuzzy logic control system, with water activity at the product surface and fan working time as input variables, and modifications in the air relative humidity set points as output variables, was able to control the drying process of fermented dry-cured sausages in an industrial drier correctly. The fermented drycured sausages obtained under this control system have the same quality as those controlled by expert technicians. New input variables related with weight loss rate have to be included as inputs in the control system to achieve a shorter drying process, similar to those applied under the expert technicians’ control. ACKNOWLEDGMENT This work was supported with funds from the Spanish Ministry of Science and Technology (CICYT, Project AGL2000-0517-P4_04). The authors are grateful to Casademont S.A. technicians (Jordi Bernardo, David Bonadona, Josep Roca and Jordi Barris) and Refrica S.A. technician (Jordi Pericay) for their expert comments and management of the industrial drier. NOTATION aw aws A C CH Kc P P0 Q

Water activity Estimated water activity at the product surface Surface area Specific heat Humid heat of drying air Convective mass transfer coefficient Total pressure Saturated vapor pressure of pure water Heat transfer rate

m2 J/kgK J/kgK kg/m²s Pa Pa W

925

RH Rt T Y Greek Symbols

α λ

Subscripts B S

Relative humidity Fan working time in 6 hours Temperature Absolute moisture content of air

% % ºC Kg/kg

Heat transfer coefficient Latent heat of vaporization Rate of evaporated water

W/m2K J/kg Kg/s

Bulk Solid

Superscripts *

At the equilibrium

LITERATURE Corrêa, N.A., Corrêa, R.G. and Freire, J.T. (2000), Adaptive control for drying of paste in spouted bed using the GPC algorithm, Proceedings of the 12th International Drying Symposium, paper Nº 80, Holland Gou, P., Comaposada, J., Arnau, J. and Pakowski, Z. (2003), On-line measurement of water activity at the lean surface of meat products, Proceedings of the 10th Driying Symposium, Lodz, Poland, pp 469476. Gou, P., Comaposada, J., Reichert and Arnau, J. (2004), Relationship between on-line measurements at the surface of meat cured products and drying process, IX International Conference Engineering and Food, 8-11 March 2004, Montpellier, France Leistner, L., Rödel, W.Y. and Krispien, K. (1981), Microbiology of meat products in high and intermediate moisture ranges, In: Water activity: influences on food quality, L.B. Rockland and G.F. Stewart (eds.), Academic Press, New York, pp 855-916 Matlab (2000) Matlab 6. The language of technical computing, The MathWorks Inc. Ruiz-Ramírez, J., Serra, X., Gou, P. and Arnau, J. (2003), Efecto de la actividad de agua y contenido de agua sobre la textura del jamón curado, II Congreso Mundial del Jamón Curado, Cáceres, Spain Stawczyk, J., Comaposada, J., Gou, P. and Arnau, J. (2003), Meat drying experiments under fuzzy control system, Proceedings of the 4th European Congress of Chemical Engineering, Granada, Spain, 21-25 September 2003, Vol 10, topic 11.3 Stiebing, A., Rödel, W. (1992), The continuous measurement of surface water activity in dry sausage. Fleischwirtscharf, 72 (11), 1547-1549

926