»In my opinion, the results of the Exploring Project, with a focus on a data-based approach, were very promising and have shown another promising way forward. A combination of the Quick Check and the Exploring Project will certainly lead to the desired success of bundling the existing data and expert knowledge and making it usable with the help of AI, so that a very reliable configuration of a dosing machine can be created and possible design errors can be avoided. Thanks to the excellent cooperation with our colleagues at Fraunhofer IPA, we were able to develop the right framework conditions in advance and successfully optimize them during the course of the project.«
Gerd Haag - Senior Manager Dispensing Technology
RAMPF Production Systems GmbH & Co. KG
Contact at the AI Innovation Center
Xinyang Wu
AI-based configurator (KIK)
Initial situation
Rampf Production Systems GmbH & Co. KG supplies fully automated dosing systems for chemical and plastic products. A specialist carries out the configuration of the core technology »dosing technology« for each system individually. This manual work is to be supported by an AI-supported automation solution
Solution idea
A data-based AI system should make it possible to utilize the historical design data collected by Rampf over decades in the system configuration and design. Based on a customer-specified parameter configuration, the system will automatically configure the dosing technology and predict the correct assemblies. In particular, this should also save a significant amount of time.
Implementation of the AI application
Two approaches to assembly prediction were pursued, a similarity analysis and a language model. Both approaches use historical design data as a data basis.
The similarity analysis is based on the distance between a given parameter configuration and the historical parameter configurations and thus provides the most suitable candidates among the historical systems. A statistical analysis of the historical data is also carried out.
The language model uses the historical design data either in the form of a system prompt or as a data basis for fine-tuning, i.e. the targeted retraining of a language model for a specific task.
With both approaches, the user first defines the requirements for the dosing technology, such as the density and viscosity of the raw materials. Based on these inputs, the system automatically suggests suitable components for the dosing system.