Better injection molding processes with AI plastic inspection

June 2, 2022

The challenge in automated optical inspection (AOI) of injection molded parts is the complex structures and geometries of the parts and corresponding defect patterns. AI inspection systems from Data Spree use deep learning algorithms to reliably detect defects and deviations in the injection molding process, even with highly complex component structures or patterns.

In the production of plastic parts, various influencing factors, such as the process temperature, the raw material (masterbatch) and the production technology, result in a wide variety of defect patterns. Particularly with complex components and geometries, it is enormously challenging for vision systems to reliably map an automated inline inspection.

Reliable detection and classification of complex injection molding defects

The variety ofdefects on injection molded parts ranges from inclusions and overmolding to surface defects. Rule-based image processing systems must be rigidly and inflexibly adapted to each individual defect. This development of classical algorithms is costly and in many cases defects are not reliably detected. AI-based image processing systems from Data Spree do not need to be programmed, but are trained via a simple teach-in process. Based on real process data, the AI can then reliably detect, classify and localize any different defect patterns, regardless of complexity.

Figure 1 shows plastic filter screens and corresponding injection molding defects from the production process. Using an AI-based inspection system (AOI), various defects can be reliably detected here simultaneously in real time, and without a single line of programming code.

Figure 1 AI classification of injection molded parts into OK parts (left) and NOK parts with different defect types (right).

The AI can reliably detect relatively obvious defect patterns, such as the complete absence of the screen structure, but also very complex defect patterns, such as thread pulls in the screen structure, at high process speeds.

Fast teach-in process and intuitive user display

Via Data Spree's own AI platform, the AI logic can be quickly taught, tested and integrated into the production plant. The teach-in process is performed on real process data and thus on authentic error patterns and evaluated automatically. With the help of the simple installation process, the AI can be quickly and easily integrated into the production environment with just one click.

Using Data Spree's proprietary user interface, users can view and analyze current optical inspection results in real time, as well as after the fact.  

Figure 2 Heatmap view for locating detects on the part.

Data Spree's own Heatmap view creates further analysis options for theuser and makes the error classification comprehensible at any time. This means that the artificial intelligence is not a "black box" for the user, but the decision-making basis of the artificial intelligence for the classification of the errors can always be viewed and manually checked at any time.

Use the potential of AI systems, also in your production

AI-based automated optical inspection solutions have the ability to quickly learn and effectively remediate defect patterns of any complexity. The flexibility of AI logic allows the potential of intelligent quality assurance to be harnessed for any type of injection molded part and plastic surface across different industries.

Data Spree offers individually trainable AI inspection solutions for any type of application and task, as well as full integration into the machine environment. The Data Spree proprietary software environment offers all possibilities of data management, AI training and evaluation in a simple user interface. With this optimized AI workflow, reliable and individualized inspection solutions are quickly ready for use and raise production quality to a whole new level.

Here you can find more information about our AI quality assurance solutions:

Learn more about the Data Spree AI platform:

Doyou need advice from our experts? For individual questions for your task, please contact us by e-mail at

Let's talk about your project.
Adopt one of our ready-to-use solutions, start creating your deep-learning-based system, or tell us your requirements to let us build an individual application for you.