4.3 Visualization of Frequent Item Sets and Fuzzy Association Rules
5.1.1 Front-end applications
The central element of the information system is the developed data warehouse. It stores all important data about the productions between 12.2005. and 02.2007. Us-ing this DW as a main information source about the production, front-end tools are also needed. An analyzer program is developed in MATLAB, and the communica-tion between the DW and analyzer tool is established via a MySQL ODBC (Open Database Connectivity) connection (Fig. 5.6). The main functions of this tool are as follows:
• General query of technological data (trends)
• Statistical analysis of productions (box-plots)
• Simulation of control system
• Checking of melting index
Figure 5.7: The graphical user interface of the developed analyzer tool, Boxes of the four main functions, listed from the left: 1. Trends (Variables, Start date, Stop date, Trends, Save), 2. Simulator (Outputs, Start date, Stop date, Running, Visualization), 3. Productions (Start date, Stop date, Products, Variables, Trends and box plots, Save), 4. Melting index (Start date, Stop date, Compare, Save)
2500 3000 3500
ppm AC20112BPV
565 570 575
kg/m3
DC2501PV
2 2.05
2.1x 104
kg/h HOMOPV
0 5 10
’2006.07.11. 1:00:00’ Idõ[óra] ’2006.07.11. 13:00:00’
g/10 min
MI
Figure 5.8: Example for trends: changes of variables hydrogen concentration, den-sity of slurry, productivity and melting index in a half day
The use of the information system
Considering the main functions of the analyzer program the graphical user interface (GUI) are divided to four parts as can be seen in Fig. 5.7:
• Trends: In the first frame of GUI simple trends can be generated. The values of the selected technological variables are queried for the given interval with a simple SQL (Structured Query Language) query from the DW then the ob-tained data are visualized by trends (for example in Fig. 5.8). The trends and values of variables can be saved for latter use (e.g. statistical analysis, data mining).
• Productions: The second frame serves an easy but efficient tool for statis-tical analysis of productions. In the plant, the most important technological variables are given on the production sheets. The analyzes of these variables are significant to get information about the productions. During the use of program, in a given interval, all the production dates of the selected product are searched then trends and box-plots for variables of the production sheets and for selected variables are depicted.
A single box plot of a technological variable gives information about the homogeneity (or heterogeneity) of the variable during the given production.
Fig. 5.9 shows an example for a box plot of reactor temperature generated by the developed tool. The rectangle represents the half of data, the green line denotes the median, and the "good" value (appeared in production sheets) of the process variable is presented by a dashed line. In ideal situation the rec-tangle is positioned near to the green line and it has small domain, moreover minimum and maximum values of the variable (the ends of vertical lines) are near to each other. It means that the given production is homogenous consid-ering the selected variable.
69.8
Figure 5.9: Single box-plot of the reactor temperature
69.5
Figure 5.10: Multiple box-plots of reactor temperature, several productions of the same product
540
Figure 5.11: Multiple box-plots of density of slurry, several productions of the same product
Figure 5.12: Trends of variables for box-plot
Alternatively, multiple box plots can be drawn together to compare multi-ple data sets or to compare groups in a single data set. Fig. 5.10 shows an example for multiple box-plots of reactor temperature measured in several productions of the same product. While the productions are homogenous considering the temperature of reactor, the densities of slurry are very differ-ent (see in Fig. 5.11). As box-plots presdiffer-ent information about the productions in a compressed form for the users, to the deeper analysis of such situations, the developed tool generates the trends of variables too. The reference (de-sirable) values of variables (if these information are available in the DW) are depicted with stars on the trends (see an example in Fig. 5.12).
Beside the main function, product comparison by a statistical tool, the tool can be used for reporting about productions. An automatic PDF-creator is developed which converts the generated plots of the given product into a PDF file with the use of Latex word processor.
• Simulations: As the third main function of the developed tool it serves an interface to the simulator of control system. The simulator is based on the first principle model of the technology which has been built in MATLAB Simulink environment based on Honeywell’s APC (Advanced Process Control). The model uses parameters and measured technological data from the process data warehouse. The user can select an interval for simulation and the output variables for visualization (Fig. 5.13). Results are saved by automatic way, therefore after the simulation an arbitrary set of outputs can be visualized. In the following, the structure of the model is discussed.
Figure 5.13: The interval and simulation outputs selection, Labels (top-down) are:
Simulator, Outputs, Start date, Stop date, Running, Visualization
Profit Contr.
RMPCT
PID & Advanced Controllers
PID controllers Process Level
Coordination Level
Figure 5.14: A highly complex, hierarchical distributed control system by Honey-well Inc.
Process Values (PV)
Calculated Values (CV) Set Points(SP)
Operational Values(OP) Process
Level
Coord.
Level
Figure 5.15: An example: the catalyst flow control
The cornerstone of the process control is Honeywell’s Profit Controller, a multi-variable control and optimization application for complex and highly interactive industrial processes. It is based on RMPCT (Robust Multi-variable Predictive Control Technology) of Honeywell Inc., which is a hierarchic spe-cial distributed control system with layered optimization of multi-input-multi-output (MIMO) systems (Fig. 5.14). At process level, HPMs (High Perfor-mance Process Managers) are responsible for basic control through PID con-trollers. The set points of these controllers are given by the Advanced Process Control (APC) System that co-operates with RMPCT model based control system. The calculation blocks of APC have the process values as input and the results are the input set points for basic controllers. The backbone of APC is the heat balance calculation of the bulk polymerization reactors and the gas phase reactor, which is the basis for all the calculated mass flow rates and concentrations in the technology. Other calculations for control are set as well, e.g. the scheme in Fig. 5.15 illustrates the flow rate control of the
Figure 5.16: Implemented APC, calculation blocks
catalyst flow to the first loop reactor. It can be seen, that RMPCT through several calculation steps gives a set point to the PID controller responsible for catalyst flow rate. The process values have to be validated, which is the task of another APC algorithm. The cornerstone of copying the process control model is the accuracy, the implemented model answers the inputs just like the original APC model. Fig. 5.16 shows the complexity and structure of this system in Simulink. On the left the realized model of the pre-polymerization and loop polymerization part of the process are presented where the different colors mean different calculation parts, e.g. the dark gray boxes (in the mid-dle of scheme) calculate the production rate for the three reactors. Under the mask of these colored boxes is the deeper layer of the calculation, e.g. the upper right model calculates the production rate of the ppolymerization re-actor and the bottom right model shows the calculation block of propylene concentration leaving the second loop reactor. The outputs of the model are set point values, operating point values and indicative values, which can be seen on the operator’s display but these last ones were not used in control loops.
In the modelling work, all the APC calculations were implemented in calcu-lation blocks. Therefore, all events of the production process can be simulated
5 5.2 5.4 5.6 5.8 6
’2006.01.01. 07:00:00’ Time[h] ’2006.01.02. 03:00:00’
MFI
PHD Napló
Figure 5.17: Comparison of stored values of the powder melting index (PHD data:
blue line, manually logged data: red line with stars)
1 1.5 2 2.5 3 3.5
’2006.03.10. 05:00:00’ Time[h] ’2006.03.11. 05:00:00’
MFI
PHD Napló
Figure 5.18: Comparison of stored values of the granulate melting index (PHD data:
blue line, manually logged data: red line with stars). It can be seen that while the blue line increases monotone (online measurer of granulate MI breaks down) the red one (real MI values) changes between in other range.
sured and simulated process data can be quickly and easily compared.
• Melting index: The melting indices of powder and granulate are determined by laboratory analysis. On the one hand results of measurements are uploaded in the PHD server by the laboratory staff and on the other hand the results are manually logged by the plant operators. From quality management point of view it is very important to know what the real values of MI were in the productions. Moreover prediction capability of the values of MI during the production process can be really beneficial. However only consistent data can be applied in the identification processes to get adequate MI prediction models. The developed DW consists of MI values from both of the sources
(PHD and paper based) and the analyzer program provides the query func-tion to compare them in a given interval. An example is shown in Fig. 5.17 where the PHD and paper based values of powder MI are compared. It can be seen that two-three records are missing from data of PHD therefore data sets for this interval are very different. Such situations indicate that process of measuring and uploading must be more precise in the plant. The impor-tance of the consistent manually logged data becomes higher when the online measurer of granulate MI breaks down (Fig. 5.18).
The presented process data warehouse contains many process variables for a long period. Beside the classical analysis, interesting relationships of variables and events can be discovered by the new data mining algorithms. The developed meth-ods are applied to mine process data of polypropylene productions. The results are detailed in the following sections.