moderator zoa biotek
Sat, 10 Feb 2001 06:06:05 -0800
Yth. Peserta ZOA-BIOTEK-2001, Sekalipun sebentar lagi sesi Bioteknologi Industri ini akan berakhir, dan diganti sesi Bioteknologi Lingkungan, izinkan kami memforwardkan makalah yang mungkin terkait dengan bioteknologi industri. Makalah Poster yang satu ini, bisa dikatakan salah satu aspek penelitian interdisiplin, karena melibatkan ilmu kontrol dan ilmu bioteknologi. Makalah ini berjudul "ADAPTIVE FUZZY-BASED EXPERT SYSTEM ON PALM OIL CRYSTALIZATION PROCESS", yang dibawakan oleh Bapak Son Kuswadi, dkk dari Institut Teknologi Sepuluh Nopember (ITS) Surabaya. Bapak Son Kuswadi, dengan pengalaman publikasi ilmiah yang begitu banyak, kini tengah menjalani studi doktoralnya di Tokyo Institute of Technology (TITech) Japan. Baik, silakan menyimak makalah beliau! Namun, sayang sekali dalam modus teks email, mungkin persamaan matematis dan gambar tidak dapat terlihat jelas. Untuk makalah lengkap disertai gambar-gambarnya, silakan melihat ke http://sinergy-forum.net/zoa/paper/html/postersonkuswadi.htm Moderator Dedy H.B. Wicaksono ================================================= ADAPTIVE FUZZY-BASED EXPERT SYSTEM ON PALM OIL CRYSTALIZATION PROCESS @ 1,2Mohammad NUH, 2,3Son Kuswadi, 2Anang Tjahjono, 2Riyanto Sigit, 1Rusdhianto E.A.K. 1Electrical Engineering Department, Industrial Technology Faculty 2Electronic Engineering Polytechnic Institute of Surabaya Institut Teknologi Sepuluh Nopember, Surabaya Indonesia 3Tokyo Institute of Technology, Tokyo, Japan Abstract: In Crude Palm Oil (CPO), there are two main substances: olein and stearin. Olein as palm oil should be fractionated from stearin through crystallization process, which temperature is the main control variable. To obtain the good quality of olein, the change of temperature of CPO should follow time-varying crystallization curve as references, namely Bernardini protocol [1]. Crystallization process is accomplished by controlling CPO temperature according the above curve. This paper will describe the implementation of AFS-based expert system to replace the manual control done by skilled operator. Operator using his/her experience will control to maintain CPO temperature according to the crystallization curve and then will produce the good crystal. AFS-based expert will elearnf from the operator for some operating condition. After esmartf enough, AFS could automatically control the process. In addition, the modification was done to reduce the using of three kinds of cooling water into two. Therefore, the power consumption is reduced significantly. Laboratory test result shows that olein produced by this process is meet the quality required. Keywords: process control, adaptive-fuzzy-system, palm oil crystallization, human-based expert 1. INTRODUCTION Palm (Elaeis guineensis Jack) is one of the important Indonesiafs non-oil exports not less than US$ 203.5 million/year. However, in palm oil form, Indonesia just produces 2,007,000 tons/year, comparatively small than Malaysia did (6,084 ton/year). In fact, if we consider the land available, that figure could be increased. The constraints are, among others, only few palm oil processing industries available and its dependency to the foreign technology. The using of accurate and robust process automation will increase the palm oil productivity. Most of existing processes (in this case the crystallization process) still employ manual control by well-experienced operators, hence its quality and quantity consistency are questioned. Therefore, we conduct the research on an implementation of crude palm oil process control to improve the process. In this research, the using of computer- based control system of palm oil processing is studied, especially its protocol crystallization process development and its adaptive- fuzzy system (AFS)-based expert system. AFS controller was chosen because of its ability on modeling of operator's expertise, meaning that AFS could ecopyf the operatorfs expertise. In other words, AFS controller at first will learn from operator how to control the process. After they esmartf enough, AFS will take over the controlling process from operator. The research result shows that the developed system could produce the better product thanks to its ability on continuous learning, fuzzy reasoning process and quality consistency. The experimental result shows that the AFS-based learning process of Bernardini protocol by only 100 epochs (every epoch need only 3 minutes), 15 fuzzy rules and 0.5 constant step size could be realized well. Sugeno and Murakami [2] proposed the fuzzy-based human expert modeling, to control model car. Several researchers propose an application of adaptive fuzzy system for the expert modeling, see for example the truck backer-upper control problem [3], flexible arm structure [4], automatic docking of vessel [5] and four-fan helicopter control [6]. In this study we will use a adaptive fuzzy system structure proposed by Wang [3] since it is easy to implement and easy learning process. 2. REVIEW OF ADAPTIVE FUZZY SYSTEM Following Wang [3], the adaptive-fuzzy system (AFS) and its back- propagation learning will be introduced briefly. Suppose that we are given an input-output pair (xp,dp), xpU Rn, dpVR; our task is to determine a fuzzy logic system f(x) as shown in Fig. 1 in the form : such that : is minimized. We assume that ail = 1 and M is given; therefore, the problem becomes training the parameters , , and such that ep of (2) is minimized. In the following we use e, f, and d to denote ep, f(xp), and dp, respectively. To train we use: where l=1,2, ....., M, k=0,1,2,........, and a is a constant step size. From Fig. 1 we see that f(and hence e) depends on only through a, where f=a/b , and Hence using the chain rule we have Substituting (4) into (3) we have where l=1,2, ..., M, and k=0,1,2,....... To train we use where i=1, 2, .... n, l=1,2,....., M, and k=0,1,2,..... We see from Fig. 1 that f(hence e) depend on only through zl; hence using the chain rule, we have Fig. 1 Adaptive-fuzzy system representation Substituting (7) into (6), we obtain training algorithm for: where i=1,2,......n, l=1,2,......, M, and k=0,1,2,........ Using the same method as previously, we obtain the following training algorithm for where i=1,2,....n, l=1,2,...., M, and k=0,1,2,........... 3. HUMAN-BASED EXPERT LEARNING STRATEGY The main idea of learning strategy by employing human-based expert is simply using operatorfs skill as teaching pattern to any adaptive systems (whether ANN, fuzzy or combination of them) [7]. In the learning stage, operator will control the plant meanwhile the AFS is learn how the operatorfs control action based on predetermined input to both AFS and operator (of course, input to operator just only to imitate the input-output relation of operatorfs action), see Fig. 2. Fig. 2 Learning stage: AFS learn from human expert (operator)fs skill to control a plant After considering that the AFS is esmartf enough, meaning that the learning process is long enough (by trial and error method), and then AFS is ready to control the plant by itself. We shall call it as eindependentf stage. See Fig. 3. Fig. 3 Independent stage: AFS control a plant by itself. 4. PALM OIL PROCESS CONTROL The quality of palm oil is depending on fractionation process that consists of crystallization and filtration activities. Crystallization is a cooling process of CPO that makes some of its part changed into crystal form and then this process continues to separate it into two forms. First, known as olein (to be used as cooking oil) in liquid forms. Second, it remains in crystal form known as stearin to be used as foodstuff and chemical industry. Initially CPO is preheated up to 700C and then stored at crystallization tank through oil pipe. There are two ON-OFF valves to control the water inlet to chiller. To make the process to be controlled in almost same state as possible (to avoid the distributed process control that is impractical and expensive), mixer is used, and simple controller controls its speed. The main variable to be controlled in the crystallization is, of course, temperature in cooling tank. Temperature to be controlled should decrease following a curve that known as Bernardini protocol [1], see Fig. 4. In conventional method, operator will control the tank temperature by changing the cooling waterfs valve manually according to the above protocol. It is clear that the said process could not provide any guarantee of product quality; since it is depend on emoodf of operator. Fig. 4 Bernardini protocol, the cooling trend should follow this decreasing pattern. In our proposed method, in other hand, will record the operator control action to be used in learning stage. Fig. 5 shows the experimental set up to implement the proposed idea. During the learning, operator will control the cooling waterfs valve manually by clicking the software process control interface trying to keep as close as possible with Bernardini protocol. The critical stage in this process is after temperature of crystal reaching 320C by quick cooling process (by fully opening the cooling valve). Usually, only experienced operator could control this process to produce palm oil form its crystal form. Our concern is to make AFS could learn from operatorfs skill in this critical stage. @ 5. RESULT AND DISCUSSIONS @ After implementing the control strategy previously described, Fig. 6 shows the experimental result of AFS-based control. It was successfully learn from the operator how to control the process, especially on the critical stage that produce palm oil. Fig. 7 shows the typical palm oil (olein) and its side product (stearin). Laboratory test shows that the product is in acceptable level, considering its melting point (MP) and cloud point (CP) are 520C and 130C respectively. The above result is better than the manual control produced by experienced operator. It seems strange that estudentf (in this case AFS) could outperform the eteacherf (in this case experienced operator). However, several experiments conducted for different system [4,5] show that the said conclusion is always happen. 6. CONCLUSIONS The implementation of AFS as controller of palm oil processing is proposed. The result shows that palm oil produced by the proposed control system is better than the experienced operator. However, this result is still in preliminary stage, meaning that it should follow by careful practical consideration such as flexible human-machine interface, self-calibration of instrument and safety. Moreover, the proposed scheme is depend on the experienced operator. Even though this scheme maintaining the advantage of fuzzy system, in term of expert modeling, but some time more sophisticated scheme is needed to avoid involvement of operator. Our future work is to develop adaptive fuzzy learning control scheme by utilizing Bernardirifs curve as time-varying set point; hence the involvement of operator no longer needed. This proposed scheme is important especially for the newly built plant. Fig 5. Experimental setup Fig 6. Typical experimental result Fig 7. Experimental product, olein (left) and stearin (right) Acknowledgement Part of this work was conducted in Knowledge-based Information Engineering Department of Toyohashi University of Technology when second author on sabbatical leave there in 1997. Thanks due to Prof. Osami Saito (then with Chiba University) and Dr. Li Xu (then with Asahi University). This work also in part supported by a World Bank Project, University Research for Graduate Education (URGE) Batch IV (1998-2000) Reference: [1] Bernardini Ernesto, Vegetable Oils and Fats Processing, B.E. Oil -00128 Rome -VIA Failla 63, Italy, 1983 [2] M. Sugeno and K. Murakami, gAn experimental study on fuzzy parking control using a model carh, in M. Sugeno, eds., Industrial Applications of Fuzzy Control, Amsterdam, North-Holland, 1985. [3] L.X. Wang and J.M. Mendel, gGenerating fuzzy rules from numerical data, with applicationsh, IEEE Trans. On Systems, Man, and Cybern, , vol 22, no 6, pp. 1414-1427, 1992 [4] Son Kuswadi, Wahyu Widodo, Slamet Wahyudi, Mohammad NUH, Osami Saito, gFlexible arm structure control using adaptive fuzzy networkh, Proceeding of 1999 International Symposium on Nonlinear Theory and Its Applications (NOLTAf99), Hilton Waikoloa Village, Hawai, USA, November 28 ? December 2, 1999 [5] Son Kuswadi et.al., gReal-time Vessel Automatic Docking Using Adaptive Fuzzy System-based Controllerh, Proceeding 1999 Industrial Electronic Seminar (IESf99), Graha ITS, Surabaya, Indonesia, Nopember 1999. [6] Yamaguchi T., T. Takagi, T. Mita, gSelf-organizing control using fuzzy neural networks, in C.J. Harris (ed.), gAdvances in Intelligent Controlh, Taylor&Francis, London, 1994 [7] Son Kuswadi, Mohammad NUH, gImplementation of adaptive neuro fuzzy on control system by human-based expert schemeh, Proceeding Calibration, Instrumentation and Metrology, LIPI, Serpong, Indonesia 22-23 September 1998, pp. 317-329, in Indonesian @