Rotary Drying of Olive Stones: Fuzzy Modeling and Control N. C. TSOURVELOUDIS, L. KIRALAKIS Department of Production Engineering and Management Technical University of Crete University Campus, Chania GREECE {nikost, kyralakis}@dpem.tuc.gr Abstract: - A rotary drying process applied to olive stones is described and modeled using fuzzy and neuro fuzzy techniques. Heat and material transfer inside the drying cylinder are rather complicated and therefore it is difficult to be accurately described. A fuzzy controller is designed based on available expertise and knowledge for a given, industrial size, rotary dryer. A second controller is built using the Adaptive Neuro Fuzzy Inference System (ANFIS) based on data taken from an empirical model of the dryer under study. Both controllers tested for various operation conditions and extensive comparative results are presented. Key-Words: - Rotary dryer, olive stones, Fuzzy logic control, Process control 1 Introduction Dryers are used to remove water from solid substances primarily by introducing hot gases into a drying chamber. Among various dryer types, rotary dryers are the most commonly used in minerals and food industry. Rotary dryers consist of a horizontally inclined rotating cylinder. The material, which is fed at one end and discharged at the other end, is dried by contact with heated air, while being transported along the interior of the cylinder. The rotating cylinder acts simultaneously as the conveying device and stirrer, as may be seen in Fig. 1. It is known that in the mathematical modeling of rotary drying procedure is rather complicated and the dynamics involved are non-linear [1], [2]. Further, the control of an industrial size rotary dryer is not an easy task, mainly because of its size and the corresponding long transportation times of the particles, and the delays between control action and observable results due to these actions. In this paper we present a novel approach for the control of the rotary drying process applied to olive stones. Wet mass of olive stones is available in large quantities in olive oil mills after the first extraction of oil. Olive stones still contain oil, which can be chemically subtracted from the dehydrated/dried stones. In order to control the olive stone drying process, we examine and compare two approaches based on fuzzy logic and neuro-fuzzy techniques, respectively. Fuzzy logic is widely used to facilitate problems of controlling rotary dryers and kilns. In [3] a fuzzy model of a pilot plant rotary dryer has been developed. The developed fuzzy model shows a good correlation between the model output and real output but it needs further development. In [2], a fuzzy PI-like controller along with a PI-like neural network have been developed and tested for a laboratory size rotary dryer. Both controllers act as supervisors of the overall system. The fuzzy controller includes three inputs and one output, while the neural net is a multilayered forward network and its training is based on the backpropagation algorithm. The data for training and testing were collected from a pilot plant rotary dryer. In [4] a neuro-fuzzy control approach and fuzzy clustering techniques are presented and their applicability in calcite drying is demonstrated. A comparative study is also presented in [4] based on simulations with a pilot plant dryer and further experimentation of the proposed system is suggested with a real size industrial dryer. The approach we suggest in this work differs from similar works in several design and structural issues, such as, the number of input/output parameters, the material to be dried, and the size of rotary dryer under study. Here we examine and compare fuzzy and neuro fuzzy based techniques for the control of olive stones rotary drying. A fuzzy logic controller is designed based on the knowledge acquired from human experts. A neuro-fuzzy controller is designed based on numerical data of a real system. The paper is organized as follows. Section 2 describes the drying process of olive stones in a real factory. Section 3 describes the mathematical modeling of the drying process. In this section, two different approaches, namely, the fuzzy approach based on Mamdani type controllers and the neuro-fuzzy approach based on Takagi-Sugeno type controllers, are described. In Section 4 experimental