What are standard features in G^rafCompounder

previous versions?

• Calculation of the specific gravity, cost per mass and volume cost of each compound in "Input data" window
• Merger: Incorporation of data from different tables into one worksheet, which may contain polymers of different nature. This should be stored in separate worksheets, but can be merged into one worksheet if needed. It is recommended to work with compound data of similar, respective soluble polymers.
• 
  For example, you have a worksheet for NR and another for SBR compounds. You may want to merge both datasets into one worksheet. This are miscible polymers to some extend and a merger increases the number of datasets, which is advantageous for calculation.
• Formulas of rubber with immiscible elastomers should not be in one worksheet due to the influence of morphology on mixing and compound properties. If you decide to work with compounds containing immiscible elastomers in one worksheet, you should be aware of this type of influence and analyze the results of any compound prediction with great caution.
• Column Code: useful to reorganize Ingredients: column after paste of additional data or merger of data tables. If Code: is applied to Properties: they can be organized according to code number as well.
• Cost(per mass): target in criteria window (Density, Cost (per volume) nor total ingredients will be calculated after prediction.)
• Frequency Distribution Diagram: Showing the spread of each ingredient or property in the dataset.

1. What are the new features in G^rafCompounder 5.0

• File name - not the completed file path - is shown in other GC windows (Frequency distribution diagram, 2D and 3D scatter plots)
• Trend-line is added to the 2D scatter plot. In the 2D scatter plot the correlation coefficient is calculated and shown.
• A plus sign is now accepted like the minus sign in the versions before. It may be important if values pasted from table calculation programs
• Paste data from clipboard into criteria window: The paste action works with more flexibility and one can paste tabular data - like a line call out - into the given position in the criteria grid.
• Mixtures created via automated mixing: The criteria displayed within the existing tool tip hovering over the name of the compound. Such criteria snapshots persist into *.gc files.
• An ingredient in compound with missing value [empty cell) does not show up in scatter plot. An explicit zero value makes for this ingredient makes that recipe visible in the diagram
• In scatter plot, the user can disable a recipe by clicking on a button, which removes it from diagram
• In scatter plot 3D, the selected data point is better visible via a shadow on the x/z and the y/z cube side
• Correlation Coefficient and trendline: Tool to analyze a perceived correlation
• 2D (x and y axis) and 3D(x,y and z axis)Diagrams to search for correlation beetween properties or ingredient / properties
• Text in fruequency distribution diagram shows compound ignored due to (a)zero ingredient value, (b) bad ingredient value, (c) bad or unknown property values. Text is only displayed if at least one recipe is ingnored in the diagram.

2. How does G^rafCompounder 5.0 predict a compound with its formula & properties?

• In GrafCompounder, each recipe is treated as a data point, comprising ingredient values and corresponding measured properties. These data points can be represented as n-dimensional vectors, where n denotes the total number of ingredients plus associated properties and their respective values.
• G^rafCompounder uses an advanced engine and algorithm designed for cross-linking of these data points through simple interpolation. By intelligently blending and interpolating the data, G^rafCompounder uncovers recipe possibilities that align with the user's preferences and criteria. Determining the optimal data points to cross-blend with specific factors is facilitated by a state-of-the-art genetic algorithm. This algorithm selectively identifies the most suitable combinations to fulfill the targets of the multi query, acting as a virtual assistant.
• G^rafCompounder employs a so called walking gradient algorithm that seeks the ideal compromise in case of conflicted targets. This process, commonly referred to as a "Fitness function" in KI language, ensures meticulous recipe crafting that aligns with the user's standards.

3. How to improve the accuracy and of compound prediction?

What does influence property variation?



• The largest influence of uncertainty and variation comes from the laboratory mixing machine.
  This machine is often not state of the art. The
  The prove of sufficient repeatability of mixing is the Mooney viscosity; the Mooney scorch Minimum or the Vulcameter Fmin measured at lower temperature for better resolution.
  It is recommended to develop a standard mixing procedure for each polymer and do a couple of mixing trails repeated in another week. Calculate the mean and the standard deviation.

4. How to check and improve the quality of datasets?

Is there an easy procedure to clear a dataset for better prediction?

The production of rubber compounds in the laboratory or in manufacturing, we have batch to batch variation. In addition, we have to endure measurement errors at subsequent testing of the compounds. Both affect the quality of the data. An additional concern in historical databases is the unknown mixing procedures, origin of raw materials, and measurement accuracy. For an AI program such as G^rafCompounder, this particularly affects the calculation of a compound because it is not possible to distinguish between accurate and less accurate values. One way to deal with this problem is to eliminate inaccurate measurement values. This can be done by examining properties of compounds according to the fact, that properties must be correlated because they are dependent on each other via crosslink density. [Literature beside many others: F. R. Eirich (Editor): Science and Technology of Rubber, Academic Press 1978] A simple example is the correlation between hardness and modulus, or between Fmax-Fmin and modulus. More examples can certainly be given. This problem can be solved with the help of the correlation graph in G^rafCompounder. In the first example - this is data consisting of several DoE's (Design of Experiments) - the correlation between Fmax-Fmin and Modulus 100 is shown. The data points can now be clicked individually and the associated compounds deactivated, which worsen the correlation. After clicking "Refresh recipes" the following graph appears: Working with this "cleaned data set" results a more accurate prediction of a compound recipe.
Of course there is a necessity to improve repeatability and accuracy to make such a procedure obsolete.

5. There are difficulties to download and install G^rafcompounder: What can be done?

• I cannot download the G^rafCompounder setup.exe file and my browser just displays a warning message "...exe was blocked because it could harm your device"?

  This could happen with the "Microsoft Edge" browser, which might block a downloaded file it does not recognize yet. To unblock the downloaded file, position your mouse cursor over the warning message. Then a button with three dots "..." should appear right of the warning message. Press that button and then choose "Keep".

• What can I do when running the G^rafCompounder setup exe file just gives me a popup titled "Windows has protected your PC"?

  This might happen when Windows decides to check back with the user because it does not recognise the file yet. To proceed with the installation, click on the "More info" text in the popup. A "Run anyway"-button will appear, which you will need to click.

• What to do when running the G^rafCompounder setup.exe file does nothing and no setup window appears?

  In some cases, this issue has been resolved by running the setup.exe file in compatibility mode (right-click on the file, select "Properties", then select the "Compatibility" tab). In other cases, it might help to run the setup exe file as admin (right-click on the file, select "Run as administrator").

6. The difference of the physical properties between calculated compound properties and confirmation experiment is quite large.

   What can be the reason?

One major concern can be the precision of measurements. According to my own experience batch to batch variation is quite small in laboratory size mixers but larger in plant scale mixers to some extend. Normally the batch to batch variation and its effect on physical properties is well known: either from quality control measurement and analysis or from repeated testing in laboratory scale. But if it is not known, I urge you to evaluate.
The other concern is the measurement error in the testing laboratory. There are a couple of methods / procedures to follow, if there is a problem.

• One method is a Gage R&R study. Advise and templates for table calculation programs can be found on the internet.
• The other method would be the use of Statistic Experimental Design (DoE) in your Laboratory. The analysis of the measurement error is inherent in the software (See for example Design Expert® from Stat Ease Inc.).
• As a third method: Laboratories can participate on Round Robin Tests, which are offered and organized by several institutes and organizations around the world.
One other important concern are mistakes in the database - like typing error or "real" measurement errors.
This is the case very often, if data are transferred by hand into the database but not electronically. You need to verify the data in your database either one by one, which is not possible in larger sets of data. At least the data of the compound used for calculation by the G^rafCompounder software should be reviewed. Compounds with faulty data should be excluded from further calculations. You may decide to repeat measurements and make correction of the datasets as a corrective action alternatively.

7. Is there any proof of linear relations between ingredients in a rubber formula and its properties?

If you analyze a lot of experiments done according the statistic experimental design (DoE) procedure, you have to notice, that almost all relations between ingredients, like fillers and oil, and the physical properties of the compound are in deed linear, made with this ingredients. If Design Expert® is your program of choice, you can visualize these relations with the graphical optimization function. It is easy to create an overlay plot of all physicals included in your research, if their values are put in the "criteria - upper limit" box. The y,x - axis show the factors (filler, oil), the contour lines the value, slope and linearity of the responses (physical properties). The following figure gives an example of an oil / filler experiment based on natural rubber (NR), which was performed recently.



The factors and responses are shown on the left side of the figure. The author has evaluated numerous investigations similar to the one above.
Even if a none linear approach is suggested by the DoE software, it was found, that a linear approach is sufficient for almost all cases included in this evaluation, taken into account measurement and experimental errors.
If it can be said truly, that linear relations between ingredients and physical properties exist, then it seems justified, to use historical or happenstance data sets to perform compound calculations.

8. How to store compound data?

Working with version 1..1 compound data are stored in a calculation program like Excel® or Open Office® Spreadsheet. In Colum 1 header is called Ingredients:, in the same column in a cell further down it is called Properties:. In the second column with the first recipe in a cell above the name of the recipe you put in Recipes:

The data are transferred into the G^rafCompounder with copy and paste.

The following table is shown as an example:



1. How to store compound data with higher Version of G^rafCompounder than 2.004?

In G^rafCompounder Version 2.004 and higher: The compound data can be stored directly in a xx.gc format, which is specific for the software. It can only opened with the G^rafCompounder as long as you have inserted the licence key.
Please see the new tutorial for this and newer versions.

2. Is it possible to open more than one file at the same time?

G^rafCompounder Version 2.004 and higher: A data file can be open directly with double click - right mouse button - on the File in Windows Explorer®. Before double clicking the file make sure, that you have inserted the licence key. If a second ..gc file gets a double click, this file will be open in a new window.


9. Which compound properties should be stored in the Database?

It is recommended, that all measurable compound properties should be stored in the database. Especially the basic physical properties should be completly stored. If one of this properties is a target of the calculation, the compound with missing data will be ignored, means not included in the calculation.

10. How can I compare the compound cost between the newly calculated compound and one of the compounds in my database?

If the material cost of all the compounds in your database is one of the property, the G^rafCompounder will calculate the material cost of your compound like any other property. Of course you can make cost a target and even give cost a weight.

Because cost - as we say: is one of the important physicals - it is recommended, that the data stored in the database consist of basic physicals and cost.

11. The confirmation experiment gives not the expected result! What should I do?

You should control every compound used for the calculation. You may start with the compound, which is used at the highest percentage.

Mostly one of the properties may show a larger difference, the tensile at break for example. If you have identified a compound with a property value, which is suspicious, you may disable this compound and recalculate.

Investigate what has changed between the first and the second calculation. Compare the property of the compound after the second calculation with the compound property, you got through the confirmation experiment.

Most likely this analysis will be the prove, that a compound with a faulty data is detected.

You either disable this compound when running further calculations, or you eliminate this compound from your original database.

12. There are missing data in the properties area. What should I do?

Empty cells in the "property:" portion of the "Input data" table would represent missing data. A compound with an empty cell in "properties:" section will be excluded from calculation automatically. As soon as you put a value in the "criteria" section in the same row, the compound name cell turns yellow to indicate exclusion.
In contrast, an empty cell in the "ingredient:" means, this specific ingredient is not in the compound. The G^rafCompounder will take the empty cell as Ingredient at 0,00 phr, which gives the correct calculation for the final formula, if the resulting compound is composed including this compound.

As an example, ingredient A is set as 1,00 phr in compound 1 and in compound 2 there is an empty cell instead. It indicates, that compound 2 contains no ingredient A. This cell is automatically assigned 0,00 phr for ingredient A. In a blend of 50% compound 1 and 50% compound 2 the value of ingredient A will be 0,50 phr in the calculated formulation, which would be the same value for ingredient A of a physically blended compound 1 with compound 2.

13. In the compound data base there are Polymers with different polarity. How to handle?

What one has to consider, if there are compound formulas in the database with different polarities is the question, whether the polymers are insoluble or soluble. An estimation can be made, if one investigates the polarity scale (see “CRC Handbook of chemistry and physics” or “polymer handbook” - Brandrup, Immergut). If polymers insoluble, the effect on physicals is none linear and the simulation result with the G^rafCompounder will be misleading. Even the author got a solution for an EPDM / SBR Blend (EP1444279A4) with the help of the G^rafCompounder, extra care must be taken in interpreting the results of a simulation. Still the best and safe way to handle such problem, the compounds containing an insoluble polymer should be disabled and the calculation should be done without those recipes.

14. If I want to evaluate new ingredients for my compound, is it possible
    to use the G^rafCompounder?

In case of the evaluation of new ingredients there is nothing in a historic database, which can be calculated. Without any experiments from the past, you can not conclude on unknown effects or correlations. The recommendation would be, that one evaluates the ingredients with the help of statistic experimental design or by any other means. The formulas and results can be used to create a database. Even if the database is small, because of the limited number of experiments done and stored, the G^rafCompounder has no restriction on the size of a database.

15. Is it possible to calculate with negative numbers, like change of Hardness -1°ShA for example?

In case of negative numbers, which you may have as a result of a property change after storage in hot air or media, you need to consider the number format only.
Between the minus (-) sign and the number there should be no space. If there is a space in between, the cell would turn yellow, indicating that this value is not accepted as a number by the G^rafCompounder. As soon as a space is eliminated, the yellow colour will disappear.

16. In the tutorial there are Code Numbers for the ingredients, which we do not use. Is it possible to replace the Code Numbers by others?

You are free to use any Code Numbers you like or given in your database system. It does not matter whether they are simply alphanumeric or mixed letter and number. My preference was to use a letter for a class of ingredients followed by a number for the ingredient in that class:

• A - Polymer
• B - Filler
• C - Process Oil

For example:
• A001 = Polymer 1
• A002 = Polymer 2

and so far for all other ingredients. To avoid number confusion it is recommended to put them in a worksheet in a table calculation program.
The same procedure is required for Properties, because in different worksheets there are not always the same nor have exactly the same name. I have chosen:
• PR001 = Property 1
• PR002 = Property 2

Please use a different worksheet to administer the property code numbers.

17. In case there are Oil extended Rubbers used in the Formulas, will G^rafCompounder calculate correctly?

To allow G^rafCompounder calculating correctly, the oil extended rubber must denoted in the following way:
Rubber 100 phr / Oil 50 phr: Rubber 150 phr
Rubber 100 phr / Oil 75 phr: Rubber 175 phr
You should not denote the oil extended rubber 100 phr and then have specified all ingredients to 100 phr.
In case of a 50 phr oil extended rubber it would mean, that the rubber content in the formula would be 66,67 parts.
It would not be inline with the other formulas in which all ingredients are normalized on 100 phr rubber.

18. In case of Merger of different data files is there a simpler method instead of having a code for all the ingredients and still be able to organize the data in the usual way?

The problem is, that ingredients have a purchase code, but this code is not helping and any change would take too much effort beside its difficulty.
I would suggest, that you group the ingredients in a simple way:
• A - Polymer
• B - Carbon Black
• C - White Filler
• D - Plastizicer
• E - .....
After the merger of two data files for example and "Sort Data" you now have all different Polymers in the first cluster A, followed by the Carbon Black in the cluster B, ...
The A - Polymer portion is shown in the example:

Code:	Ingredients:
A	SBR - 3
A	NR 20
A	SBR - 1
A	NR 60 CV

In the next step you are able to sort the polymers in alphabetic order by mouse click and highlight the ingredients and then "Sort Data".
You need to do this operation in the same way for the
A -, B -, C - , D -, E - ..... Ingredients.
The table in the example above would then look like this for the polymers:

Code:	Ingredients:
A	NR 20
A	NR 60 CV
A	SBR - 1
A	SBR - 3

You would need a bit more time to organize larger data files after a merger, but it avoids a larger effort of preparation.
Please do not forget the code for the Properties. The description above is applicable for the properties in the same way. Please see the tutorial data [Demo data (advanced)] implemented in the program.