Icici158p - Muntini, Measuring The Quality Of Black Tea

International Conference on Instrumentation, Communication and Information Technology (ICICI) 2005 Proc., August 3 -5 , 2005, Bandung, Indonesia

Measuring Quality of Black Tea From Theaflavins Analysis Using Secondary Measurement 1)

Melania S. Muntini1), Yul Y. Nazaruddin2), The Houw Liong 3), Lienda Handojo4) Department of Physics, Institut Teknologi Sepuluh Nopember (ITS) Surabaya, Indonesia 2) Department of Engineering Physics, Institut Teknologi Bandung, Indonesia 3) Department of Physics, Institut Teknologi Bandung, Indonesia 4) Department of Chemical Engineering, Institut Teknologi Bandung Jl. Ganesa 10 Bandung 40132, Indonesia Phone/Fax: +62-22-2508138 E-mail: [email protected], [email protected]

Abstract – Theaflavins (Tf) is a key compound that significantly contributes in the quality of black tea. It undergoes a series of chemical changes during the fermentation process. Fermentation is one of the most critical processes in black tea processing. There are many parameters that significantly influenced the process including room temperature, thickness of greendhool, and duration of the process. In general, it is difficult to measure theaflavins directly as it involves some chemical analysis and enzymes for pigment. An alternative approach, theaflavins is measured indirectly and inferred from easily made process measurements or secondary measurements. This inferential method of measurements employs a scheme which is called a virtual sensor, which is realized by integrating artificial neural networks with the Extended Kalman Filter algorithm. Secondary variables are several parameters of fermentation process and results of color analysis of tea liquid, whereas primary variable is Theaflavins. The data for implementing this proposed technique were obtained by conducting several real-time experiments at black tea factory in Indonesian Tea and Cinchona Research Institute (PPTK Gambung), West Java. Results show how the quality of black tea can be infered indirectly using the proposed technique.The mean and variance of error between the obtained output of virtual sensor algorithm and the output chemical analysis of theflavins were 1,81 x 10 -4 and 5,07 x 10-6 respectively. Keywords – artificial neural network, black tea, Extended Kalman Filter, indirect measurements, Theaflavins, virtual sensor.

involving withering, rolling, fermentation, drying and sieving. Fermentation is one of the most critical processes in black tea processing. During black tea fermentation, an enzymatic oxidation of tea polyphenols, especially tea chatechins takes place, leading to a formation of a series of coloured chemical compounds, among other, such as Theaflavins (Tfs) that determines the characteristics of the black tea liquors.[1,2,3,6,7,8]. Tfs of the black tea depends on several The parameters that has significantly effects in the fermentation process, i.e. room temperature, thickness of greendhool, and duration of the process. Understanding the relationship of liquors colour to black tea quality would be interesting for development of methods to identify black tea quality, chemically and physically. In general, it is difficult to measure theaflavins directly as it involves some chemical analysis and enzymes for pigment. An alternative approach is proposed in this research, in which theaflavins is measured indirectly and inferred from easily made process measurement or secondary measurement.

I. INTRODUCTION

II. SECONDARY MEASUREMENTS

Black tea is a fermented tea and it is the one of popular beverages in the world. Black tea manufactured is carried out by a series of processes on fresh tea leaves,

In the complex process, some variables may be easy or hard to be measured. A primary variable is a process variable that hard to be measured, while a secondary

117

International Conference on Instrumentation, Communication and Information Technology (ICICI) 2005 Proc., August 3 -5 , 2005, Bandung, Indonesia

network because it were proposed in order to deal with the nonlinear systems.

variable is an easier one. The direct measurement is used to measure secondary variables. The secondary variables must be related to primary variables, because the primary variables are measured indirectly and inferred from them. This inferential method of measurements employs a scheme that is called secondary measurement or virtual sensor[9]. Virtual sensor is realized by integrating Artificial Neural Networks with Extended Kalman Filter (EKF) algorithm. Neural Networks are general tools in modelling nonlinear function since its ability to approximate any nonlinear functions with any desired accuracy.

Figure 2 illustrates the block diagram of virtual sensor algorithm or secondary measurement in measuring of secondary variables.

The structure of neural networks that is used in this research is restricted to Direct Recurrent Neural Network (DRNN). DRNN structure consisting of three layers, input, hidden and output layer respectively, has been applied in modeling of the process. The neural networks structure of DRNN can be seen in Fig. 1. Wj W jk

Fig. 2. Structure of secondary measurement

Suppose θ represents all trainable parameters of the network (i.e. the weights and biases), the model can be rewrite as follows : x 1 ( t + 1) = f 1 ( x1 ( t ), Z ( t + 1), θ 1 ( t ),

11

I ( t + 1)) + ξ 1 t )

10

Wj

x 2 (t + 1) = f 2 ( f1 (.),θ 2 (t )) + ξ 2 (t ) Z ( t + 1) = Z ( t ) + ζ ( t )

21

I n (k)

Y (n)

S j (n) 1b

Fig. 1. A DRNN structure

It gives the following relation N

S j (n) = ∑Wjk Ik (n) +Wj X j (n −1) +Wj 10

11

1b

(1)

[

Y (n) =

∑

j=0

W

21 j

Xj ( n )

X j ( n ) = f ( S j ( n ))

]

T xˆ (t ) = xˆ1 (t ) xˆ 2 (t ) Zˆ (t ) can of be obtained. After linearization, the model can be described using the following state model x(t + 1) = F (t ) x(t ) + G(t )ξ (t ) (8)

k=1

M

(5) (6)

(7) y (t ) = x 2 (t ) + v (t ) where weight and biases at hidden and output layer are θ1, θ2 respectively, {ξ1(t)}and {ξ2(t)} are zero-mean Gaussian White Noise sequence uncorrelated with {v(t)} and with a pre assigned positive definite Var[ζ(t)] = S(t). Based on T x(t ) = [x1 (t ) x 2 (t ) Z (t )] , the estimation

X j (n)

Wj

(4)

(2)

(3)

where I, W, b, and f are input, weight at hidden layer, weight, bias and activation function. Extended Kalman Filter Algorithm is used in training the neural

y (t ) = H (t ) x (t ) + ν (t )

(9)

where {ξ(t)} and {v(t)} are assumed to be uncorrelated zero-mean Gaussian White Noise sequences. F(t), G(t), and H(t) are

118

International Conference on Instrumentation, Communication and Information Technology (ICICI) 2005 Proc., August 3 -5 , 2005, Bandung, Indonesia

assumed to be known matrix-valued. The Extended Kalman Filter procedures to estimate the state are as follows

[

III.3. Fermentation The greendholl were fermented in the 1.5 x 1 x 1.75 m3 mini-plant process with temperature variation between 18 – 260C. The first experiment assumed that the temperature was about 180C, the second about 220C and the third experiment about 260C. Duration of fermentation was 5, 20, 30 minutes and continues at every 10 minutes interval until 150 minutes, each for a greendholl thickness of about 4, 6, 9 and 12 cm. The chemical contents of the black tea were investigated after fermentation process with various operation conditions including duration, room temperature and thickness of greendholl. Fermentation process was stopped by drying greendholl in mini-dryer under inlet and outlet temperature of 1100C and 900C respectively.

]

K(t) = P(t | t − 1)H(t)T R(t) + H(t)P(t | t − 1)H(t)T (10) P (t | t ) = P (t | t − 1) − K (t ) H (t ) P (t | t − 1)

(11)

⎡ xˆ1 (t ) ⎤ ⎡ xˆ1 (t | t − 1) ⎤ ⎢ xˆ (t )⎥ = ⎢ xˆ (t | t − 1)⎥ + K (t )[ y (t ) − H (t ) xˆ (t | t − 1)] ⎥ ⎢ 2 ⎥ ⎢ 2 ⎢⎣ Zˆ (t ) ⎥⎦ ⎢⎣ Zˆ (t | t − 1) ⎥⎦

(12) This paper will show how Extended Kalman Filter can be applied as secondary measurement for theaflavins analysis, which is the one of tea quality indication. III. EXPERIMENTS AND RESULTS III.1. Data collections For the purpose of modelling, a mini-plant of fermentation process has been designed and implemented during the experiment. The research was conducted at Indonesian Tea and Cinchona Research Institute (PPTK Gambung), West Java and the tea powder was collected from the black tea factory of this institute. The tea leaf for processing was obtained from tea plantation of clone Gambung 14-17 at the altitude of 1400 m. The tea plants were allowed to overgrow so that it was possible to pluck up to four leaves and a bud.

III.4. Theaflavin analysis The black tea was analyzed by Robert Smith method [8] to obtain Tf. This method applies maximum difference between maximum length visible wave 380 nm and 460 nm. Metanol or i-butil metil keton solution was used to dissolve extraction of black tea [6] III.5. Results One indicators of black tea quality is colour quality. Usually colour quality measured by tea tasters with visual inspection. Theaflavins is ones of chemical compounds that influenced the black tea quality, particularly in colour quality. Besides, it also influences on liquor and taste. The proposed method that used in the research is to measure Tf using indirect measurement employing an algorithm integrating Artificial Neural Network with Extended Kalman Filter algorithm. Tf content was analyzed using this algorithm, in which Tf is unmeasured input on virtual sensor. Measurable inputs are several parameters that have significantly effects during fermentation process, including room temperature, thickness of greendhool,

III.2. Processing of black tea using orthodox method Tea leaves were heaped in a whitering through to a thickness of about 20 – 35 cm and a constant air flow at hygrometer difference of 1- 20C was maintained for up 14 h. The leaves after whitering were subjected to orthodox rolling for up to 60 min in a three-cranck single action roller. The machine rolled leaves were fired at 900C for 30 min to obtain black tea containing about 3% moisture. Greendholl is tea leaves after whitering and rolling proceed before fermentation.

119

International Conference on Instrumentation, Communication and Information Technology (ICICI) 2005 Proc., August 3 -5 , 2005, Bandung, Indonesia

figure 2. Moreover, Tf contents of black tea are denoted by Z(t) in eq (8) – eq (13). Tf contents will be obtained when the model and system produced the same output or the error between them is acceptable. Using the virtual sensor algorithm, the model of Tf can be developed.

and duration of the process. The output of virtual sensor is model of colour quality. Output system is color quality, which is obtained from the digital photo of color inspection of tea liquid and it was analyzed using the software. Data of output system were obtained from previous research (6). Comparison between the model of color quality and output system is shown in figure 3.

Fig. 3. Comparison between model and system of color quality

Validation of model and output system was done using statistical properties.

Fig. 5. Output of the process and model

Output system that obtained from the first experiment (with temperature 180C) yields the smallest Root Means Square Error (RMSE) which is equal to 0.00031. The validation of secondary measurement result and the system, which is the Tf content of the black tea for fermentation process at temperature 180C and thickness of wetness of tea 9 cm, is shown in Fig. 5.

Fig. 4. Error model and system of colour quality

Furthermore, the proposed scheme was run using on-line system and output of this algorithm will be used to predict unmeasurable input. In the research, the model is assumed to be perfect and work properly. On initial running of the algorithm, there was an error between output model and output system because the model accepts less input than the system, i.e. unmeasurable input. Based on the error, Tf can be analyzed using recursive manner. The above procedure can be viewed as prediction of Tf as shown in

IV. CONCLUSION An alternative method to measure Tf using indirect measurement or secondary measurement has been proposed and presented in this paper. Secondary measurement was accomplished using integration of Neural Networks and Extended Kalman Filter. The results show that the proposed method could be applied to analysis Theaflavins contents in black tea. Based on Theaflavins contents, the quality of black tea, in particular colour quality of infusion, can be measured. 120

International Conference on Instrumentation, Communication and Information Technology (ICICI) 2005 Proc., August 3 -5 , 2005, Bandung, Indonesia

V. REFERENCES [1].

Liang, Y, Lu, J, et al.,”Estimation of Black Tea Quality by Analysis of Chemical Composition and Color Difference of Tea Infusions”, Journal of Elsevier Food Chemistry, vol. 80 (2003), p.283-290 [2]. Su,Y.L, Leung,L.K, et al., “Stability of Tea Theaflavins and Catechins”, Journal of Elsevier Food Chemistry, vol. 83 (2003), p.189-195 [3]. Obanda,M,Owour,P.O,et al.,”Changes in Thearubugin fractions and Theaflavins Lvels due to Variations in Processing Conditions and Their Influence on Black Tea Liquor Brightness and Total Color”, Journal of Elsevier Food Chemistry, vol. 85 (2004), p.163-173 [4]. Temple, Van Boxtel, “Modelling of Fluidized-bed Drying of Black Tea”, Journal of Agricultural Engineering Research, vol 74 (1999), pp.203-212 [5]. Instruksi kerja Pengolahan Teh, PTP. Nusantara VIII Perkebunan Ciater, 1998 [6]. Muntini, M.S, Handojo, L, Nazaruddin,Y.Y, Joni, Santosa, R. ”Kadar Tearubigin dan Tanin pada Proses Fermentasi Teh Hitam”. Proceed. Seminar Nasional dan Pertemuan Patpi, Yogyakarta, 22-23 July 2003 [7]. Owuor, P.O. and Obanda, Martin, , “Comparative Responses in Plain Black Tea Quality Parameters of Different Tea Clones to Fermentation Temperature and Duration”, Journal of Elsevier Food Chemistry, vol. 72 (2001), p.319-327 [8]. Roberts E. H. and R.S. Smith, “Spectrophotometry Measurement of Theaflavins and Thearubigins in Black Tea Liquor”, in Assesments of Quality in Teas, 1961 [9]. Habtom, Ressom, “Dynamik System and Virtual Sensor Modeling Using Neural Network”, Fortschritt-Berichte VDI, Reihe 8, Nr.771, 1999 [10]. Werkhoven, J.,”Tea Processing”, Food and Agricultural Organization of the United Nations, Rome, 1974

121