Application of Transfer Learning to Chemical Toner Quality Prediction

Abstract

In a chemical toner manufacturing plant, automatic quality control is implemented using the process data collected for long time. On the other hand, the processing method and materials are improved continuously, which requires a lot of man-hours of the operators before the prediction model is reconstructed and the quality control is in service. To reduce the workload, an accurate prediction model needs to be developed from small-size data. This paper proposes a quality prediction method that utilizes transfer learning. Frustratingly Easy Domain Adaptation was implemented for the feature space construction, and Gaussian process regression (GPR) was adopted with Bagging to improve the stability and accuracy of the model. As a result of applying the proposed method to the toner mass production plant, operator’s man-hours was reduced by 75%.