Title:Virtual Quality Control using bidirectional LSTM networks and gradient boostingAuthor(s):Amirreza Baghbanpourasl,  Edwin Lughofer,  Pauline Meyer-Heye,  Helmut Zörrer,  Christian EitzingerAbstract:Quality control is usually performed at the end of production lines. This postponed quality control, in case of deviation of the process values from desired quantities can lead to late detection of batches of defective products. A prediction model allows a virtual online quality control. Observation of any unfavourable trend or degradation in the predicted quality can enable preventive measures. There are many cases where quality control is performed manually and less often. This leads to limited data which is challenging for training any machine learning system used for purpose of prediction. In this paper the goal is prediction of quality measure with respect to the past history of all relevant process values such as sensor readings. We present a machine learning model based on bidirectional Long Short-term Memory (LSTM) recurrent neural networks and gradient boosting. This model is applied to a set of real world data from microfluidic chip production plant and we show that despite limited amount of training data compared to the large input space, it is capable of predicting the relevant quality values, and especially their basic trends over time including drifting phases and changing behaviors.Booktitle:Proceedings of the IEEE international conference on industrial informatics (INDIN2019 Industrial Applications of Artificial Intelligence)Publisher:IEEE PressPage Reference:6 page(s)Publishing:2019Series:Proceedings of the IEEE international conference on industrial informatics (INDIN2019 Industrial App

go back