Data preparation

The data should be passed to the CutPredictor as pandas data frames. Each experiment (or simulation) with a unique ID 'doe_id' consists of:

• A set of \(d\) process parameters defining the experiment (definition of the material, force applied, etc).
• 1D, 2D or 3D positions in the corresponding space.
• The resulting output variables (such as a deviation) for each position.

The goal is for example to learn to predict the deviation for any position, given a set of (new) process parameters.

Taking the example of the x0 cut in the Cut_x0.ipynb notebook, the process parameters may be:

• Blechdicke
• Niederhalterkraft
• Ziehspalt
• Einlegeposition
• Ziehtiefe
• Stempel_ID

the 1D position is tp and the only value to predict is deviationc.

Two data frames should be prepared:

1. A doe data frame for the \(d\) process parameters, including a unique ID per experiment.
2. A data data frame for the positions and outputs, with a unique experiment ID for each sample.

The doe data frame should look like:

doe_id	Blechdicke	Kraft	Ziehspalt	Position	Ziehtiefe
1	1.01	410.0	2.4	-5	30
2	1.48	400.0	2.4	-4	30
3	1.00	210.0	2.4	-1	30
...	...	...	...	...	...

The data data frame should look like:

doe_id	tp	deviationc
1	0.00	1.43
1	0.01	1.45
1	0.02	1.47
...	...	...

Additional columns are not a problem. The name of the column for the experiment ID (here doe_id) should be the same in both data frames.

Once the data is prepared in that format and saved to disk (csv, hdf5...), the data frames can be loaded:

doe = pd.read_csv('doe.csv')
data = pd.read_csv('data.csv')