Importing Experimental Data into FormuSense - ProSensus - Multivariate Analysis Consulting for Manufacturing

Typically, an experimental dataset includes information on the available ingredients, the ratios in which these ingredients have been combined in past experiments, the process conditions under which the ingredients were combined, and the resulting quality properties.

Accordingly, these are the four sources of data that can be imported into FormuSense. FormuSense imports data from a single Microsoft Excel workbook file, with a separate worksheet for each source.

Ingredient Properties

Where possible, ingredients should be grouped into classes, according to chemical structure or function. The most important consideration when grouping ingredients into classes is to ensure that the measured properties are relevant to all ingredients within a single class. These organized groupings empower FormuSense to calculate additional input variables that are typically valuable for predictive modeling. (Read more about mixture properties.)

For example, a formulation that contains both solids and liquids should have ingredients split into at least two classes, since viscosity is an important ingredient property but it is only relevant to the liquids and not to the solids. Importantly, the name of an ingredient property worksheet must identically match the name of the ingredient class listed in the formulation ratios worksheet.

A separate Excel sheet should be created for each ingredient class. Ingredient property data should be structured as a matrix with one row for each unique ingredient and one column for each ingredient property (measured, calculated, categorized, or described). Examples of ingredient properties include: molecular weight, density, viscosity, molecular structure description, etc.

As seen in the example, new ingredients, never used in past formulations, but available for consideration in future formulations can be included in this matrix.

Formulation Ratios

Formulation data, meaning the information that explains what ingredients and in what quantities were combined together in each past experiment, should be organized into a single worksheet.

Formulation data for a single family of products is often collected by numerous scientists across many years using various platforms. To account for these complexities, the simplest structure for organizing formulation data has been selected as the FormuSense import format.

FormuSense requires users to include one row for each ingredient used in each formulation. For example, an experiment that uses 4 ingredients (FP-005) will require 4 rows, with the formulation name repeated. Formulations may contain different numbers of ingredients, and the amount used may be entered as an amount (such as mass, moles, or PHR with consistent units), fraction (a value between 0-1, representing the ingredient amount divided by the total amount), or percent (a value between 0-100, representing the fraction x 100).

Importantly, FormuSense requires that the name of any ingredients used in past formulations must identically match the name of an ingredient listed in the related ingredient properties spreadsheet. For any ingredient listed in in the formulation ratio spreadsheet that does not also occur with identical nomenclature in the relevant ingredient class properties worksheet, FormuSense will generate a row of entirely missing ingredient property data for that ingredient.

After imported, FormuSense generates a full matrix of formulation data, with a single row for each past experiment, and a column for each available ingredient in the dataset (across all ingredient classes).

Process Conditions

Any relevant process conditions that may impact final quality properties should be included in a predictive model, and can therefore be imported to FormuSense as one or more separate worksheets.

If desired, process conditions can be grouped into related blocks of variables, with a separate worksheet for each block. This multi-block approach can be especially useful when not all process conditions are relevant to every quality property. In this case, the user may wish to build separate models, including or excluding input variables, depending on the quality property being modeled. Hence, importing the process data in organized blocks can simplify the model-building process.

FormuSense requires each process condition worksheet to be imported with one row per formulation, and one column per process condition. Examples of process conditions include mixing speed, processing temperature, processing time, etc.

Quality Properties

Finally, relevant quality properties can be imported to FormuSense from one or more worksheets.

Similarly to process conditions, it may be useful to group quality properties into related blocks through separate worksheets. Grouping quality variables into multiple blocks can simplify the model-building process if not all input variables are relevant to every quality property. In this case, the user may expect to build separate models for each block of quality properties.

Each quality property worksheet should contain one row per formulation, and one column per quality property.

Need More Help?

Reformatting your data into the structure required for FormuSense import is typically not an onerous task. Contact ProSensus to request assistance or further documentation on how you can readily restructure your experimental data yourself using basic Excel functions.

FormuSense users and trial-users are also encouraged to refer to the FormuSense User Manual for further information.

ProSensus strongly recommends that new users participate in a contextualized training session (CTS) to ensure the successful adoption of FormuSense into your product formulation workflows. Investing in a CTS gives your team the opportunity to learn directly from ProSensus experts as we work collaboratively to evaluate the suitability of your data for advanced modeling and develop effective models, while delivering software training and best-practices guidance.