Modeling, observer and supervisory controller design for cement mill process

Cement is a finely ground inorganic material, which, when mixed with water, forms a paste that hardens by means of hydration reactions and, after hardening, retains its strength and stability even under water. Quality of cement, mostly, is resembled by mortar compressive strength. Chemical structure, fineness and particle size distribution of finished product have a great influence on mortar compressive strength. Cement grinding process is one of the main processes in a cement plant which has a task of blending clinker with desired amount of gypsum and addition material. At the end of the process, cement is produced and transported to the packaging line. Cement grinding process is a Multi Input Multi Output (MIMO) system. Controlling the blaine value of cement is aimed by adjusting the feed rate and seperator speed parameters. Process begins with feeding clinker, gypsum and addition material depending on the type of cement to the mill. Feed would not mix homogenously at the first pass and due to this fact, the mixture is directed to a seperator. The thin particules that are flung by the centrifugal force of the seperator are delivered to the packaging line and the rest of the mixture is fed to the mill again for another grind. The part of the mixture that is fed back to the mill is called "reject" and rejected mixture is much greater than the fresh feed, which means inputs' effects can only be observed after several grinding cycles. A dynamical model having MIMO random disturbances that are using past inputs/outputs and disturbance characteristics and a controller which can change the feed rates and the seperator speed to keep the actual fineness value close to the reference fineness value should be designed to represent this process. Furthermore, due to the cost of precise automatic fineness measuring systems, an estimator system should be designed to estimate the fineness of the cement. There are many solutions realized by overseas companies in the market. But these solutions are very expensive and not working as efficiently as they are presented. There is a great need of a cost-effective system in the market. In this thesis, it is aimed to make an estimator/observer/soft sensor system that is used to estimate the fineness value of the cement by utilizing the real-time parameters of system such as mill vibration, mill sound, bucket elevator current etc. To do this, a data acquisition system is installed in ADOCIM Tekirdag/Sultankoy branch cement factory. The data acquisition system was accessible remotely and collected data for one month with 1 sample/second setting. Primary data sets collected from the system are :  Raw material feed rates (tons/h)  Rejected material rate (tons/h)  Seperator speed (rpm)  Elevator current (A)  Folaphone (%) Besides the data sets collected via the data acquisition system, fineness level of cement was obtained for the production between 1.00-5.00 am every day. After collecting, data sets were processed to be used in the model. After the process, it was seen that 18 days of CEMIV and 9 days of CEMI type cement production data was collected. Several input output configurations was used to obtain a model. Since there was not enough blaine data, it was assumed that the fineness level of the cement is stabilized when mill output and reject was kept around a certain level. This approach was tested beforehand. After this, blaine model was obtained directly from the data. These configurations can be classified as :  1 Input (Raw material + Reject rate) and 1 Output (Elevator current) model  2 Inputs (Raw material feed rate, Reject rate) and 1 Output (Elevator current) model  3 Inputs (Raw material feed rate, Reject rate, Folaphone (mill level)) and 1 Output (Elevator current) model  4 Inputs (Raw material feed rate, Reject rate, Folaphone (mill level), Seperator Speed) and 1 Output (Elevator Current) model  5 Inputs (Raw material feed rate, Reject rate, Folaphone (mill level), Seperator speed, Elevator current) and 1 Output (Blaine (fineness)) model Also, several different modelling techniques were used to model the system. These methods are :  Auto Regressive eXogenous input (ARX)  Non-linear Auto Regressive eXogenous input (NARX)  Transfer Function Estimation  Principle Component Analysis (PCA)  Artificial Neural Networks (ANN)  Evolving Takagi Sugeno (eTS) The data sets were downsampled for 10 second/sample and 60 seconds/sample settings and models were also obtained with these settings. After obtaining an estimator system having a high fitness level, a supervisory controller system that is used to keep the fineness value around the set value was designed. This supervisory controller system determines the feed values and seperator speed according to the estimated fineness and several other system parameters. Evolving Takagi Sugeno (eTS) technique was used to design the controller model. Controller system will be applied to the real time system, and if successful, will be converted into a market product. This thesis was funded by STZ 1563.2012-2 project code SAN-TEZ (Ministry of Science, Industry and Technology) project with the co-operation of Dal Elektrik ve Otomasyon A.Ş. At this stage, a PID model was tested and this work was published with the title "Çimento Öğütme Sistemi Simülatör Uygulaması" in "Otomasyon Dergisi" December 2015 issue. The articles that are prepared to be published are : "Comparison of models on the fineness estimation for a cement grinding process" "A Novel Cement Fineness Supervisory Control System" Both articles are planned to be published in "Int. Journal of Mineral Processing" journal.