Evaluation of Deep Autoencoders for Prediction of Adjustment Points in the Mass Production of Sensors

In the context of Industry 4.0 the inclusion of additional information from the manufacturing process is a challenging approach. This is demonstrated by an example of calibration process optimization in the mass production of automotive sensor modules. It is investigated to replace a part of a measurement set by prediction. Support-vector regression compared to multiple, linear regression model shows only minor improvements. Feature reduction by deep autoencoders was carried out, but failed to achieve further improvements.