< Back to previous page

Project

AI4IM - Artificial Intelligence for Injection Moulding  (AI4IM)

For the manufacturing industry, injection molding is probably the most important production technique when it comes to producing large series of plastic products. The Flemish injection molding sector, however, has been under pressure for several decades from competitive countries outside the EU such as China and India, where products are produced at lower cost and where in recent years a serious effort was made to catch up in terms of quality of delivered products. Therefore, it is important for Flemish injection molding companies to continue to innovate and reduce costs by focusing on automation (e.g., automatic assembly right after injection molding), faster injection molding lines (more flexible changeover), reducing operator interventions and continuing to guarantee high product quality. This project hopes to contribute to this by adding an artificial intelligence (AI) component to the injection molding process. Goal of this project Specifically, 3 goals are envisioned within this project. - Automatically detect parts with injection molding errors or parts out of specification (e.g. in terms of: dimensions, surface quality, etc.). - Predicting when injection molding errors, parts out of specification or machine problems will occur (predictive maintenance). - Provide new insights into the injection molding process by finding explanations for injection molding errors. How AI works in the injection molding process Within this project, using sensors in the mold or measurements on the injection molding machine: pressure, flow rate and temperature are measured during injection, pressing and cooling of the injected material. To these measured input curves, output data is linked, being: injection errors, dimensions, surface roughness, etc. . Both input and output data are used to train a learning network, so-called: neural network, once this network is trained e.g.: based on data from the first 1000 shots, it can be used to predict output from newly measured pressure, flow or temperature curves. Furthermore, the input data (pressure, flow and temperature) can also be used to predict the product quality in the longer term. This can be used to predict when products will exhibit errors or fall out of specification over a larger number of shots, e.g. over 50 or 100 shots. This allows for intervention in advance to avoid downtime or for injection molding operators to be targeted on lines. This is different from the first case where products out of specification are detected and sorted out, but still cause rejects. Finally, a specific AI technique will be explored within this project, explainable AI. Usually AI is seen as a black-box where a network is trained and will then make decisions without more. With explainable AI an explanation is given why a neural network links specific pieces of input data to the output. This can be linked to what is already known about plastics processing and this in turn can give new insights to injection molding engineers. 
Date:1 Jan 2021 →  31 Dec 2022
Keywords:Artificial Intelligence, Injection Moulding, Machine Learning
Disciplines:Manufacturing processes, methods and technologies, Machine learning and decision making
Results:
For the manufacturing industry, injection moulding is probably the most important production technique when it comes to producing large series of plastic products. However, the Flemish injection moulding sector has been under pressure for several decades now from competing countries outside the EU, such as China and India, where products are produced at lower cost and where there has been some serious catching up in recent years in terms of the quality of delivered products. That is why it is important for Flemish injection moulding companies to continue to innovate and reduce costs by focusing on automation (e.g. automatic assembly immediately after injection moulding), changing injection moulding lines faster (more flexible changeover), reducing operator interventions and continuing to guarantee high product quality.
The AI4IM project contributes to this by introducing and validating machine learning (ML) in the injection moulding process of the Flemish industry using 4 real case studies. The promising results may encourage Flemish injection moulding companies to use these techniques themselves, and the resources on the website effectively give them the tools to get started. 
More information can be found on the website www.ai4im.be