App-based quality prediction and failure cause analysis with artificial neural networks - oraKel



01.10.2019 to 30.09.2021
Organizational Unit:
Chair of Production Metrology and Quality Management, Model-based Systems
German Federation of Industrial Research Associations AiF, Federal Ministry for Economic Affairs and Energy BMWi



work Phone
+49 241 80 28394


The aim of the AiF research project "oraKel" is the development of an automated process in an app for quality prediction and failure cause analysis for production machines. In order to increase their competitiveness, SMEs are pursuing the goal of reducing their inspection effort and optimizing their fault cause analysis. One option is to predict product quality on the basis of process data instead of physically measuring product characteristics and using the necessary model for fault cause analysis.

The solution to enable feature-based quality prediction and automated defect cause analysis is to develop a process based on neural networks - NN. Methods for data preprocessing, process knowledge integration, extraction and sampling are developed and put into a meaningful context. In data preparation, the training data set for the NN is created by converting time series data into data points and balancing the database. Through process knowledge integration, existing process knowledge is introduced into the topology of the NN in the form of mathematical formulas, thus improving the predictions. After training the NN, previously unknown process knowledge about process knowledge extraction is gained from this and can be used for error cause analysis and process improvement. In order to continuously validate the predictions, a sampling inspection plan is developed based on the prediction uncertainty.

The result is a procedure for quality prediction and automated defect cause analysis in which all methods developed are integrated. The procedure is implemented as an app that supports users in its application. The benefits for SMEs result from an app that increases their competitiveness by predicting quality, analysing the causes of defects, reducing inspection costs, uncovering process optimisation potential, increasing resource efficiency and reducing production costs.