The Plastics Technology Group at the TU Ilmenau (KTI), in cooperation with its partners Steinbeis Qualitätssicherung und Bildverarbeitung GmbH (SQB), Institut für Mikroelektronik- und Mechatronik-Systeme gemeinnützige GmbH (IMMS), eitech Werkzeugbau GmbH (eitech) and Kunststoff- und Holzverarbeitungswerk GmbH (KHW), has launched the joint project ProQuaOpt, which investigates whether the productivity and quality of plastic injection molding processes can be increased by using machine learning methods together with information provided by different sensors.

According to manufacturers, the proportion of rejects in the production process for plastic components is approx. 5%. For the German plastics industry, this amounts to approx. 750kt of plastics that must be recycled. Common solutions to prevent rejects are to monitor the actual values of the injection molding machine, so that the operator has to correct a defined deviation from the target value [1]. Here, the system can only react to defects that occur in the predefined parameter range, even if not-in-order (NiO) parts are already detected before the threshold value.

Fig. 1: Survey of the Product-Process-Quality-Cycle to be developed (PPQRK)

The product-process quality control loop (PPQRK) to be developed, as shown in Figure 1, does not monitor the actual values of the machine, but the quality characteristics of the molded parts by sensor technology. Here, AI methods, such as machine learning [2], are used to develop a learning spectral image processing method for quality inspection in combination with further sensor technology (e.g. IR camera, load cell, temperature sensor). The correlation of different sensor data enables a more precise defect diagnosis, so that in the next process step an AI-based self-learning assistance system varies suitable process parameters to restore the quality of the molded parts and find an optimal operating point for the process to minimize cycle time and energy use [3].

Fig 2: Forced quality defects for data collection (left: photo camera detects incomplete filling, warpage and vacuole, right: thermographic image of the same component to compare the defects)

To train the AI algorithm, machine-generated defective molded parts are produced as shown in Figure 2. In addition, the project intends to develop a method that can generate synthetic data of surface defects with a CAD model of the injection molded part, which reduces the testing effort of the machine-produced defective parts.

The AI-supported PPQRK to be developed is not restricted to specific manufacturers of injection molding machines and is to be upgradable.

The German Federal Ministry of Education and Research (BMBF) supports the research project ProQuaOpt 01IS22019. The project partners are grateful to the BMBF for the financial support of this research topic.

[1] Ogorodnyk, O., & Martinsen, K. (2018). Monitoring and control for thermoplastics injection molding a review. Procedia Cirp, 67, 380-385

[2] Grabmann, M.; Feldhoff, F.; Gläser, G. (2019) IntelligEnt – Künstliche Intelligenz und Machine Learning für den Entwurf und die Verifikation komplexer Systeme. Thüringer Forum Künstliche Intelligenz, 24 May, Erfurt, Germany

[3] Karbasi H, Reiser H, editors. Smart Mold: Real-Time in-Cavity Data Acquisition. First Annual Technical Showcase & Third Annual Workshop, Canada; 2006: Citeseer


Prof. Florian Puch
Technische Universität Ilmenau
Department of Mechanical Engineering/Group
Plastics Technology