Research project Injection moulding machines to be optimised with AI-based tools
With the 'Darwin' project, the Fraunhofer Spinn-off Plus 10 and SKZ want to develop new AI solutions for injection moulding and put them through practical tests. The models will be tested on different machines in order to design a manufacturer-independent solution.
The joint research project of the Fraunhofer spin-off Plus 10 and the SKZ (Süddeutsches Kunststoffzentrum, South German Plastics Centre) aims to develop optimised process parameters for injection moulding machines with the help of AI and machine learning. The parameters are based on the subsequent machine cycle and take into account, for example, the raw material characteristics or environmental conditions. This should make it possible to manufacture without rejects. According to the project team, the behavioural models should be transferable to machines of similar size and technology, regardless of the manufacturer.
Machines learn from each other
Within the framework of the project, detailed behavioural models of injection moulding machines are acquired on the basis of high-frequency machine data. Due to the transferability of pre-trained machine learning models, individual machines can learn from each other. This means that behavioural models of a specific machine do not have to be completely relearned each time, but are only adapted to the machine and the currently running product in a small adaptation phase. As a result, the models will then suggest optimised process parameters for the next cycle.
Demag injection moulding machine trials
The practical relevance of the project is demonstrated by the current test series. The algorithms were recently trained on the Sumitomo (SHI) machines from Demag. These have a control system that enables high-frequency and near-real-time communication in the millisecond range. This type of communication makes the machines practically 'AI-ready'.
Optimised processes are important for sustainable plastics processing
The solutions devised in the project should make it possible to process even sensitive or heterogeneous materials, for example thermoplastic post-consumer recyclates or rapidly cross-linking elastomers, in a process-stable manner, promoting a functioning circular economy in the plastics industry.
The publication of the final results is planned for the end of 2022.