Skip to main content

The Cyberphysical Part Observer

monitoring of part properties in automated intra machine supply

Automated part handling between machines requires precise control over properties such as alignment, orientation and form. Depending on the actual case, criteria need to be defined that ensure proper operation of the adjacent manufacturing step. Check out this application case to learn more.

Key achievements

  • High level of accuracy achieved by the convolutional neural network (CNN) in identifying product alignment
  • Substantial enhancement in the production process reliability, less disruptions, reduced need for human intervention

Handling of parts in automated intra machine handling systems

monitoring key properties of parts

The use case involves applying machine learning to improve the manufacturing process at the Teaching and Learning Factory (TLF), particularly with the SIF-400 system. The TLF is a state-of-the-art cyberphysical factory integrating various Industry 4.0 solutions. This case focuses on using quality-assured data to train a machine learning model to minimize errors in the SIF-400 production process. The initiative demonstrates the potential of digital technologies in enhancing efficiency and reducing errors in manufacturing environments. It also aims to show how innovative solutions can be applied to real-world manufacturing challenges, leading to smarter, more efficient production systems.

The primary challenge identified was the occurrence of errors during the production process in the SIF-400 Teaching and Learning Factory (TLF). Specifically, the issue was with the feeding process of canisters and palettes in the SIF-401 module, the first station of the production line. These errors, such as canisters not being placed correctly on the palettes, were causing slowdowns and even stoppages in the production, necessitating human supervision and intervention. The goal was to find a solution that would reduce these errors and minimize the need for human oversight in the production system.

The solution to the challenge of errors in the production process at the SIF-400 Teaching and Learning Factory (TLF) was the implementation of a machine learning system. This solution was specifically designed to address the issue of canisters not being correctly placed on palettes in the SIF-401 module. The implementation involved several key components:

  • Integration of External Devices and Machine Vision: The system included external devices attached to the SIF-400 system. Machine vision technology was employed to reduce the need for human interaction, intervening only when necessary.
  • Use of an IP-Camera for Motion Detection and Image Capturing: A camera was responsible for detecting motion and capturing images. A custom-designed hook, 3D modelled and attached to a microcontroller, was used to hold the palette with the canister in place for accurate imaging.
  • Development of a Convolutional Neural Network (CNN): To automate the process of identifying errors, a CNN was developed. This neural network was trained to identify tilted products on the assembly line using thousands of images for training and validation. This approach allowed for high accuracy in detecting positioning errors of the products.
  • Data Storage and Processing: Images captured by the IP-camera were stored on an FTP server. A Python script was used to scan the latest uploaded picture for analysis.
  • Binary Image Classification Algorithm: The CNN employed a binary image classification algorithm to determine whether a product was tilted. The system's high accuracy was achieved through extensive training with a large dataset of images.
  • Adaptability and Retraining Capability: The system was designed to be adaptable and could be retrained in response to changes in the production environment, such as alterations in lighting, product shape, etc.

This solution significantly reduced the need for human oversight in the manufacturing process, thereby increasing efficiency and reducing errors in the SIF-400 TLF production line. The implementation of machine learning and the integration of smart technologies showcased an innovative approach to addressing challenges in a cyberphysical factory environment.

The solution indicates a high level of accuracy achieved by the convolutional neural network (CNN) developed for the project. The CNN was reported to have around a 100% accuracy in identifying tilted products on the assembly line, suggesting a significant reduction in errors during the production process.

The near-perfect accuracy of the CNN implies a substantial enhancement in the production process's reliability and consistency. This likely led to fewer disruptions, reduced need for human intervention, and overall improved operational efficiency in the factory.

Potential areas of future development and action:

  • Further Improvements in Accuracy and Efficiency: While the CNN achieved around 100% accuracy, continuous improvement and refinements in the algorithm could be a focus, especially as more data becomes available.
  • Adaptation to Changing Production Environments: The adaptability of the solution is an important perspective, as the system can be retaught if the production environment changes (e.g., lighting, product shape). Future actions could involve retraining the model to accommodate new products or changes in the manufacturing setup.
  • Expansion to Other Production Lines or Factories: The successful implementation of this solution opens the possibility of applying similar machine learning techniques to other production lines within the TLF or even in different manufacturing settings.
  • Further Research and Development: Future perspectives could include an innovative approach in integrating machine learning with manufacturing processes and additional research and development to enhance the capabilities of machine learning in industrial settings.
  • Training and Education: Given the TLF's focus on training and education, future actions might involve using this implementation as training module for students and professionals interested in Industry 4.0 technologies.


discuss the details with the people behind this application

Eliza Toth | PBN

Keen on new use-cases?

Register for our newsletter