Çimento endüstrisinde harmanlama prosesinin sistem tanılaması

Abstract

Bu çalışmada, bugünün teknolojisinde önemini enerji tasarrufu açısından rakamlarla ortaya koyan adaptif kontrol stratejilerine önayak olan sistem tanılama konusu ele alınmıştır. Sistem tanılama konusu tek basma oldukça geniş bir kavramdır. Konu kendi içerisinde değişik yöntemleri barındırmaktadır. Sistem tamlama konusunun farklı iki yöntemi olan en küçük kareler yöntemi ve enstrümantel değişkenler yöntemi, bu çalışmanın ana amacını teşkil etmişlerdir. Çalışma içerisinde bu iki yöntemin neler oldukları, formülasyonların nasıl elde edildikleri, karşılaştırmaları ve değişik özellikleri incelenmiştir. Bu incelemeler ve yorumlamalar Çanakkale Çimento A.Ş. den ve Nuh Çimento A.Ş. den elde edilen verilere uygulanarak ilgili çimento proseslerinin modellenmesinde kullanılmıştır. Bunlara ait bilgisayar programlan ve sonuç grafikleri yine bu çalışmada sunulmaktadır. Bu iki yöntem tek başlarına eldeki datalardan model oluşturmanın yanısıra ardışık algoritmaları kullanıldığında sürekli kontrol uygulanması gereken proseslerde, öz uyarlamak kontrol, kendinden uyarlaman kontrol gibi muhtelif adaptif kontrol stratejilerinin uygulanmalarına da imkan verirler. Yine bu çalışmada yukarıda adları geçen çimento endüstrisi ve bu alanda bilgisayar çahşamalan yardımı ile elde edilen kimyasal analizler ve bu analizlerin nasıl uygulandıktan hakkında da ayrıntılı bilgiler verilmektedir. Bütün bunların yanısıra yine bu çalışma kapsamında sistem tamlamada önemli bir yere sahip olan model geçerlileştirme konusu da incelenmektedir. Bu alanda yardımcı olan kriter ve testler hakkında formülasyonlar ve açıklamalar ilerideki bölümlerde yer almaktadır.

System identification is the field of modeling dynamic systems from experimental data. A dynamic system can be conceptually described as in the following figure. Figure vi.1. A dynamic system with input u(t), output y(t) and disturbance v(t), where t denotes time. The system is driven by input variables u(t) and disturbance v(t). The user can control only the input but not the disturbance. In some signal processing applications the inputs may be missing. The output signals are variables which provide useful information about the system. Many industrial processes, for example for production of chemical compounds like cement, must be controlled in order to run safely and efficiently. To design regulators, some type of model of the process is needed. The models can be various types and degrees of sophistication. Sometimes it is sufficient to know the crossover frequency and phase margin in either a Bode plot or a Nyquist plot. In other cases, such as the design of an optimal controller, the designer need a much more detailed model which also describes the properties of the disturbances acting on the system. In many cases the primary aim of modeling is to aid in design. In general, an identification experiment is performed by exciting the system. It is followed by using some sort of input signal. Then it is necessary to observe its input and output over a time interval. These signals are normally recorded in a computer as the system data. It is then tried to found the system parameters. The first step is to determine an appropriate form of the model. It is typically a linear difference equation of a certain order. As a second step some statistically based method is used to estimate the unknown parameters of the model. These are such as the coefficients of VJ the difference equation. In practice, the estimation of structure and parameters are often done iteratively. This means that a kind of model is chosen andits corresponding parameters are estimated. The model obtained is then tested to see whether it is an appropriate representation of the system. If this is not allright, a more complex model structure is considered and its parameters are estimated and the new model validated etc. The scheme of the procedure is shown in the following figure. Start NJd Design of experiment U- Perfonn experiment Collect data ^LJ£ Determine/ choose model structure N_k Choose method Estimate parameters Apriori knowledge No New data set Yes Figure vii.1. Schematic flowchart of system identification c In this thesis, it is explained that does system identification mean and how to apply system identification to an industrial process. In chapter 1, there is an vu introduction to system identification. It follows in chapter 2, the Least Squares Method (LSM) and the so called Instrumental Variable Method (IVM). In this chapter the both identification methods are explained in details with obtaining their formulations. The IV method is not a well known method as the least squares algorithms. It is also a new alternative to the least squares estimation. The least squares method is a very easy applicated method. But it has a substantial drawback : the parameter estimates are consistent only under restrictive conditions. The LS method could be modified in different ways to overcome this drawback. One modification is developed by REIERSOEL in 1941 and it has been popular in the statistical field for quite a long period. It has been applied to and adapted for dynamic systems both in econometrics and in control engineering. In the latter field pioneering work has been carried out by WONG and POLAK (1967), YOUNG (1965,1970), MAYNE (1967), ROWE (1970) and FINIGAN and ROWE (1974). A well written historical background to the use of IV methods in econometrics and control can be found in YOUNG (1976), who also discussed some refinements of the basic TV method. For a more recent and detailed appraisal of IV methods is given by SOEDERSTROEM and STOICA (1983) and in their papers. The least squares approach is first as known given by GAUSS who wanted to identify the orbital of the asteroid Ceres. In chapter 3, the recursive identification methods are introduced and formulated in details. The given methods are the Recursive Least Squares (RLS) method and the Recursive Instrumental Variables method (RTV). The use of recursive identifications methods is necessary to identify the systems which operate on-line. The need of the recursive methods is often seen by application of the adaptive control systems. Recursive identification methods have the following general features : * They are a central part of adaptive systems where the action is based on the most recent model * Their requirement on primary memory is quite modest, since not all data are stored. * They can be easily modified into real-time algorithms, aimed at tracking time varying parameters. * They can be the first step in a fault detection algorithm, which is used to find out whether the system has changed significantly. Most adaptive control systems are based on recursive identification methods. Then a current estimated model of the process is available at all times. This time varying model is used to determine the parameters of the regulator. In this way the regulator will be dependent on the previous behavior of the process. If an appropriate V1U principle is used to design the regulator then the regulator should adapt to changing characteristics of the process. The chapter 4 contains the explanations about the model validation and formulations about some criterion for the model validation. In system identification both the model structure determination and model validation are important aspects. An overparametrized model structure can lead to unnecessarily complicated computations for finding the parameter estimates and for using the estimated model. An underparametrized model may be inaccurate. The purpose of chapter 4 is to present some basic methods that can be used to find an appropriate model structure. When searching the for the correct model order one can raise different equations, which should be discussed as in chapter 4 : * Is a given model flexible enough ? * Is a given model too complex ? * Which model structure of two or more candidates should be chosen ? All the answers of these three questions are explained in chapter 4. In chapter 5 there is an example from an application in the cement industry. The cement production plant of Çanakkale Çimento A.Ş. in Çanakkale is taken as an application process. The process is explained with details and the data of input and output are used. In cement industry they used four main raw materials. These are the clay, chalkstone, iron and kaolin. The four raw materials are mixed in a predetermined proportion and so it is obtained a clinker of a known quality. These raw materials are taken as the system inputs. For a good quality the proportionality of these raw materials should be in a standard interval. To make an analysis for quality one take samples of the blended material in some points of the process. The samples are analyzed in two ways. One is to make a chemical analysis. Another approach is the X-ray spectral analysis. The rawmaterials are brought into a crusher to crush in small pieces. After then they are conveyed to a stockhall where a reclaimer make a prehomogenization. In the stockhall the samples of the rawmaterial is taken and analyzed. The chemical analysis is made every eight hours. For a confirmation of this the x-ray analysis is made once a day. There are bins at the end of the stockhall. When a misproportion occure the less kind of rawmaterial is given to the mixture. There is also an analysis in the silo after the mill. There are looked for four main oxides. These oxides are calciumoxide (CaO), siliciumoxide (Si02), aluminiumoxide (AI2O3) and ironoxide (Fe2Û3). The sizes of their percentile proportion are criteria for the quality. The chemical analysis is made every hour and the x- ray analysis is made once a day. The four main oxides are thought as the system outputs. These input and output data are used to identify the system. They build the vector of known quantities and the vector of unknown quantities (system parameters) is calculated in two different computer programs. One program is written for computing the system parameters using the least squares method estimation and other IX program is written for the computation of the system parameters using the instrumental variables method estimation. These both computations are made for an ARX (autoregressive with exogenous variables) model. The two computer programs are presented in the last chapter which contains the the explanations of experimental results with plotouts. The results are evaluated and discussed in the same chapter.