Choosing a machine learning model for predicting the results of a process

Hi guys I have a question regarding Machine Learning. I have a process chain of a production of a material.

At the beginning I have a material mix consisting of three materials. Each material has a defined particle size distribution. The percentage composition of the materials can be varied.

Three sieving processes follow. Depending on the specific energy used, more or less material is screened.

At the end, a product with a specific quality related to the particle size distribution is obtained.

Which model is suitable for the prediction?

Data on different compositions of the input and different amounts of energy used in the different machines as well as the quality of the output were collected in advance.

Can you help me with this?

Another question is if there are models that can predict the input mix and the settings of the machine for a desired output quality.

Best regards

Fabian

Answer 1

[Javier Gazzarri]

Hello Fabian,

Great use case for Machine Learning.

In MATLAB you can actually try several ML algorithms simultaneously and pick the one with the lowest prediction error. Since you already have the data, you can assemble a MATLAB Table with predictors (energy inputs for each step) and response (quality) as columns. Then, you can use classification if quality is non-numeric (low-med-high) or regression if it is (e.g. 0-100%) and train all the models you wish. At the end of the training you'll have a metric for each model's performance. The tools to do this are the Classification Learning and the Regression Learner, which you can find in the APPS section of the toolstrip.

Please let me know when you have the chance to test this.

Javier