ENBIS-16 in Sheffield

11 – 15 September 2016; Sheffield Abstract submission: 20 March – 4 July 2016

Expert Knowledge Systems to Ensure Quality and Reliability in Direct Digital Manufacturing Environment

13 September 2016, 10:10 – 10:30


Submitted by
Eva Scheideler
Eva Scheideler (Hochschule Ostwestfalen-Lippe), Andrea Ahlemeyer (Ahlemeyer-Stubbe Data Mining and More)
Seamless process chains in the Direct Digital Manufacturing in the context of Industry 4.0 shorten the time between offering of individual designed products, production and the delivery of it. To reach that goal further automation is required. Therefore reliable automated and quick evaluation procedures are needed, which ensure the quality of the individual designed products in terms of product safety and product reliability.
Until now human experts check based on their experience whether the consumer desired product parameters a likely to create a product that ensures the required safety level. If necessary the expert will change the product parameter after client consultation to meet the needed safety level. Depending on the specific application the review of the parameter is done so far with special often complex simulation tools. In most of the cases a human expert is need to run simulation tools.
This talk aims to demonstrate, how a meta model generated on simulated data and adapted to the type of product can be used to ensure a reliable, automated and quick evaluation of the specific client preferred product parameters to guarantee the demanded characteristics of the final product with out consultation of human expert knowledge.
A meta model that reach all this characteristics should only run on the client given parameters and information of the demanded characteristics or usages of the future product. In the further description these parameters are called input variables. Parameters based on expert knowledge should be covered by the meta model itself and not be influenced by clients requirements.
In a first step a deterministic complex simulation model is created from the product. Design strategies for computer experiments are used to select data sets. Thereby the complexity and the nonlinearity between the input variables and the response has been taken into account. Deeper domain knowledge is also necessary to create these computer based experiments. The generated data are used to learn with Statistical prediction methods a valuable meta models which have a good forecast ability. The validation of the simplified prediction model is a crucial success factor.
As proof of concept we choose a task from the field of construction. Our example is a vision panel in a facade or door. With sets of typical given parameters and sets of unusual but realistic once, we like to illustrate the power of the meta model. Our goal is it to see whether the safety requirements are based on the individual input variable and without expert interaction are confirmed by the meta model.
Fast reliable prediction models derivative from complex simulation models are indispensable conditions for direct digital manufacturing. Using meta models in automation context is a foundation of manufacturing in future.
View paper

Return to programme