Hier finden Sie wissenschaftliche Publikationen aus den Fraunhofer-Instituten.

Automated and Predictive Risk Assessment in Modern Manufacturing Based on Machine Learning

: Müller, Tobias; Kiesel, Raphael; Schmitt, Robert


Schmitt, Robert ; Wissenschaftliche Gesellschaft für Produktionstechnik -WGP-:
Advances in Production Research : Proceedings of the 8th Congress of the German Academic Association for Production Technology (WGP), Aachen, November 19-20, 2018
Cham: Springer Nature, 2019
ISBN: 978-3-030-03450-4 (Print)
ISBN: 978-3-030-03451-1 (Online)
ISBN: 978-3-030-03452-8
German Academic Association for Production Technology (WGP Congress) <8, 2018, Aachen>
Wissenschaftliche Gesellschaft für Produktionstechnik (WGP Congress) <8, 2018, Aachen>
Bundesministerium für Bildung und Forschung BMBF (Deutschland)
01; Quadrika
Quality data based risk assessment for industry 4.0
Fraunhofer IPT ()
algorithm; process control; machine learning

The revision of the ISO 9001 in 2015 and the end of the transition period in 2018 forces companies to integrate risk management into their company structure. Risk assessment as part of risk management poses challenges for many companies, especially SMEs. Existing methods are often complex, subjective and difficult to automate. To address this issue, this paper describes a risk assessment approach that can be fully automated after expert process evaluation. The automated risk assessment is based on a Machine Learning algorithm, which builds a model that predicts the output and allows the use of SPC control charts without measuring components characteristics. Based on the results of the control charts, the risk can be assessed by calculating the distances to critical values and analyzing th e control chart (e.g. run or trend identification). The use of process parameters, which are recorded by sensors, makes it possible to intervene in the process in high risk situations and reduce not only measurements but also the production of scrap. The method was applied to the use case of an injection molding process of a thin-walled thermoplastic. Based on a Design of Experiments the model was built by a Generalized Linear Regression machine learning algorithm. A predictive validation and an event validation test were used to validate the method. A two-sided t-test at a significance level of α = 5% provided equality between predicted and actual mean value. The Event Validation Test provided a 90100% correct classification.