idw – Informationsdienst Wissenschaft
Nachrichten, Termine, Experten
Nachrichten, Termine, Experten
idw - Informationsdienst
Wissenschaft
Many everyday objects, from cordless screwdriver casings to toothbrush handles, are produced using the injection moulding process: molten plastic is injected into a mould, where it then cools and solidifies. Before a new product – and therefore a new mould – can be created, a number of factors must be assessed using software: Where is the best place to inject the plastic into the mould? How quickly does the component cool down, and how much does it warp in the process?
At the Centre for Future Production at the University of Augsburg, a team led by Prof. Dr.-Ing. Nils Meyer (Professor of Data-driven Product Engineering and Design) combined classical physics models with AI-based models in a software that performs the necessary calculations in just a few seconds. Making their approach up to 100 times faster than standard software solutions, saving both time and resources.
Accuracy versus speed
To understand the significance of this acceleration, one must take a closer look at the processes involved in the plastics processing industry. If a company wishes to injection-mould a new component, the feasibility must be assessed and the design of the injection mould validated. This is done using software solutions whose calculations are based on physical flow simulations. The results are highly accurate, but the process is computationally intensive and often takes several hours. Particularly in the early stages of product development, a faster and more flexible computational model would be advantageous in order to reach a decision.
This is where Nils Meyer’s idea comes in: “A tool that allows you to assess numerous possibilities in a matter of seconds – such as where it would make the most sense to inject material – can save companies valuable resources such as staff working time, computing time and, consequently, money and energy. We have succeeded in speeding up the process by a factor of 100.” For him, these highly accurate physical methods are the second step when it comes to calculating the final component.
What makes it special: data efficiency & arbitrary geometries
A major challenge is the requirement for the AI-supported tool to be able to predict arbitrary component geometries whilst, at the same time, training efficiently with as little data as possible. Rather than unleashing the ‘AI hammer’ on the problem with vast amounts of data, Meyer’s team has therefore incorporated prior knowledge into the underlying AI model. Meyer: “We already know, for example, that an area a long way from the injection point will be filled later than one directly next to the injection point. We don’t need to learn that from data first.” This meant the model could be trained using just a few hundred examples and is still able to make predictions for any other components.
A highlight: AI-compatible finite element model
Meyer is particularly proud of an integrated feature which, with a twinkle in his eye, he simply describes as “cool”: it is an AI-compatible finite element software programme being developed by his research group. “The finite element method is a computational technique that breaks down complex systems into many small, simple parts so that they can be analysed more effectively,” he explains. This makes it possible, for example, to predict the expected distortion of a component as it cools. Meyer: “What sets our version apart is that we can also use this software to further train our AI model. This allows the model to continue learning from measured deformations in the production process and to keep improving.”
The vision
Whilst the current software assists with the optimisation of components that have already been designed, the aim for the future is to support the entire design process. “We envisage a programme to which you can specify your component requirements, and which then uses AI to generate a design suitable for injection moulding. This helps design staff to quickly find the best solution from the vast range of possible components. It relieves them of repetitive tasks and gives them more time to focus on creative processes and product requirements,” summarises Meyer.
Prof. Dr.-Ing. Nils Meyer
Data-driven Product Engineering and Design
Telefon: +49 821 598-69223
nils.meyer@uni-a.de
Nils Meyer (right) and project team member Julian Greif are developing an AI tool to optimise compon ...
Quelle: Teresa Grunwald
Copyright: University of Augsburg
Merkmale dieser Pressemitteilung:
Journalisten, Wirtschaftsvertreter
Informationstechnik, Werkstoffwissenschaften
überregional
Forschungs- / Wissenstransfer, Forschungsprojekte
Englisch
Sie können Suchbegriffe mit und, oder und / oder nicht verknüpfen, z. B. Philo nicht logie.
Verknüpfungen können Sie mit Klammern voneinander trennen, z. B. (Philo nicht logie) oder (Psycho und logie).
Zusammenhängende Worte werden als Wortgruppe gesucht, wenn Sie sie in Anführungsstriche setzen, z. B. „Bundesrepublik Deutschland“.
Die Erweiterte Suche können Sie auch nutzen, ohne Suchbegriffe einzugeben. Sie orientiert sich dann an den Kriterien, die Sie ausgewählt haben (z. B. nach dem Land oder dem Sachgebiet).
Haben Sie in einer Kategorie kein Kriterium ausgewählt, wird die gesamte Kategorie durchsucht (z.B. alle Sachgebiete oder alle Länder).
