Transfer learning for enabling quality predictions in small batch production

Multilayer perceptrons (MLP) have found wide application in the prediction of quality features within production processes. This type of model requires many training data points for reasonable accuracy. Production processes with small batches pose a challenge for model training due to a high variance with a relatively small data quantity for individual product batches. The article shows possibilities for making cross-variant quality predictions through the application of transfer learning algorithms. A framework that allows companies to evaluate the feasibility of various transfer learning approaches for predictive quality applications within different product batches is developed and presented.